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 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)
75 struct fsync_inode_entry *entry;
77 inode = f2fs_iget_retry(sbi->sb, ino);
79 return ERR_CAST(inode);
81 entry = f2fs_kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
83 list_add_tail(&entry->list, head);
88 static void del_fsync_inode(struct fsync_inode_entry *entry)
91 list_del(&entry->list);
92 kmem_cache_free(fsync_entry_slab, entry);
95 static int recover_dentry(struct inode *inode, struct page *ipage,
96 struct list_head *dir_list)
98 struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
99 nid_t pino = le32_to_cpu(raw_inode->i_pino);
100 struct f2fs_dir_entry *de;
101 struct fscrypt_name fname;
103 struct inode *dir, *einode;
104 struct fsync_inode_entry *entry;
108 entry = get_fsync_inode(dir_list, pino);
110 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list, pino);
112 dir = ERR_CAST(entry);
113 err = PTR_ERR(entry);
120 memset(&fname, 0, sizeof(struct fscrypt_name));
121 fname.disk_name.len = le32_to_cpu(raw_inode->i_namelen);
122 fname.disk_name.name = raw_inode->i_name;
124 if (unlikely(fname.disk_name.len > F2FS_NAME_LEN)) {
130 de = __f2fs_find_entry(dir, &fname, &page);
131 if (de && inode->i_ino == le32_to_cpu(de->ino))
135 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
136 if (IS_ERR(einode)) {
138 err = PTR_ERR(einode);
143 err = acquire_orphan_inode(F2FS_I_SB(inode));
148 f2fs_delete_entry(de, page, dir, einode);
151 } else if (IS_ERR(page)) {
154 err = __f2fs_do_add_link(dir, &fname, inode,
155 inode->i_ino, inode->i_mode);
162 f2fs_dentry_kunmap(dir, page);
163 f2fs_put_page(page, 0);
165 if (file_enc_name(inode))
166 name = "<encrypted>";
168 name = raw_inode->i_name;
169 f2fs_msg(inode->i_sb, KERN_NOTICE,
170 "%s: ino = %x, name = %s, dir = %lx, err = %d",
171 __func__, ino_of_node(ipage), name,
172 IS_ERR(dir) ? 0 : dir->i_ino, err);
176 static void recover_inode(struct inode *inode, struct page *page)
178 struct f2fs_inode *raw = F2FS_INODE(page);
181 inode->i_mode = le16_to_cpu(raw->i_mode);
182 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
183 inode->i_atime.tv_sec = le64_to_cpu(raw->i_mtime);
184 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
185 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
186 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
187 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
188 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
190 if (file_enc_name(inode))
191 name = "<encrypted>";
193 name = F2FS_INODE(page)->i_name;
195 f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
196 ino_of_node(page), name);
199 static bool is_same_inode(struct inode *inode, struct page *ipage)
201 struct f2fs_inode *ri = F2FS_INODE(ipage);
202 struct timespec disk;
204 if (!IS_INODE(ipage))
207 disk.tv_sec = le64_to_cpu(ri->i_ctime);
208 disk.tv_nsec = le32_to_cpu(ri->i_ctime_nsec);
209 if (timespec_compare(&inode->i_ctime, &disk) > 0)
212 disk.tv_sec = le64_to_cpu(ri->i_atime);
213 disk.tv_nsec = le32_to_cpu(ri->i_atime_nsec);
214 if (timespec_compare(&inode->i_atime, &disk) > 0)
217 disk.tv_sec = le64_to_cpu(ri->i_mtime);
218 disk.tv_nsec = le32_to_cpu(ri->i_mtime_nsec);
219 if (timespec_compare(&inode->i_mtime, &disk) > 0)
225 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
227 struct curseg_info *curseg;
228 struct page *page = NULL;
232 /* get node pages in the current segment */
233 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
234 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
237 struct fsync_inode_entry *entry;
239 if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
242 page = get_tmp_page(sbi, blkaddr);
244 if (!is_recoverable_dnode(page))
247 if (!is_fsync_dnode(page))
250 entry = get_fsync_inode(head, ino_of_node(page));
252 if (!is_same_inode(entry->inode, page))
255 if (IS_INODE(page) && is_dent_dnode(page)) {
256 err = recover_inode_page(sbi, page);
262 * CP | dnode(F) | inode(DF)
263 * For this case, we should not give up now.
265 entry = add_fsync_inode(sbi, head, ino_of_node(page));
267 err = PTR_ERR(entry);
268 if (err == -ENOENT) {
275 entry->blkaddr = blkaddr;
277 if (IS_INODE(page) && is_dent_dnode(page))
278 entry->last_dentry = blkaddr;
280 /* check next segment */
281 blkaddr = next_blkaddr_of_node(page);
282 f2fs_put_page(page, 1);
284 ra_meta_pages_cond(sbi, blkaddr);
286 f2fs_put_page(page, 1);
290 static void destroy_fsync_dnodes(struct list_head *head)
292 struct fsync_inode_entry *entry, *tmp;
294 list_for_each_entry_safe(entry, tmp, head, list)
295 del_fsync_inode(entry);
298 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
299 block_t blkaddr, struct dnode_of_data *dn)
301 struct seg_entry *sentry;
302 unsigned int segno = GET_SEGNO(sbi, blkaddr);
303 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
304 struct f2fs_summary_block *sum_node;
305 struct f2fs_summary sum;
306 struct page *sum_page, *node_page;
307 struct dnode_of_data tdn = *dn;
314 sentry = get_seg_entry(sbi, segno);
315 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
318 /* Get the previous summary */
319 for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
320 struct curseg_info *curseg = CURSEG_I(sbi, i);
321 if (curseg->segno == segno) {
322 sum = curseg->sum_blk->entries[blkoff];
327 sum_page = get_sum_page(sbi, segno);
328 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
329 sum = sum_node->entries[blkoff];
330 f2fs_put_page(sum_page, 1);
332 /* Use the locked dnode page and inode */
333 nid = le32_to_cpu(sum.nid);
334 if (dn->inode->i_ino == nid) {
336 if (!dn->inode_page_locked)
337 lock_page(dn->inode_page);
338 tdn.node_page = dn->inode_page;
339 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
341 } else if (dn->nid == nid) {
342 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
346 /* Get the node page */
347 node_page = get_node_page(sbi, nid);
348 if (IS_ERR(node_page))
349 return PTR_ERR(node_page);
351 offset = ofs_of_node(node_page);
352 ino = ino_of_node(node_page);
353 f2fs_put_page(node_page, 1);
355 if (ino != dn->inode->i_ino) {
356 /* Deallocate previous index in the node page */
357 inode = f2fs_iget_retry(sbi->sb, ino);
359 return PTR_ERR(inode);
364 bidx = start_bidx_of_node(offset, inode) + le16_to_cpu(sum.ofs_in_node);
367 * if inode page is locked, unlock temporarily, but its reference
370 if (ino == dn->inode->i_ino && dn->inode_page_locked)
371 unlock_page(dn->inode_page);
373 set_new_dnode(&tdn, inode, NULL, NULL, 0);
374 if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
377 if (tdn.data_blkaddr == blkaddr)
378 truncate_data_blocks_range(&tdn, 1);
380 f2fs_put_dnode(&tdn);
382 if (ino != dn->inode->i_ino)
384 else if (dn->inode_page_locked)
385 lock_page(dn->inode_page);
389 if (datablock_addr(tdn.node_page, tdn.ofs_in_node) == blkaddr)
390 truncate_data_blocks_range(&tdn, 1);
391 if (dn->inode->i_ino == nid && !dn->inode_page_locked)
392 unlock_page(dn->inode_page);
396 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
397 struct page *page, block_t blkaddr)
399 struct dnode_of_data dn;
401 unsigned int start, end;
402 int err = 0, recovered = 0;
404 /* step 1: recover xattr */
405 if (IS_INODE(page)) {
406 recover_inline_xattr(inode, page);
407 } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
409 * Deprecated; xattr blocks should be found from cold log.
410 * But, we should remain this for backward compatibility.
412 recover_xattr_data(inode, page, blkaddr);
416 /* step 2: recover inline data */
417 if (recover_inline_data(inode, page))
420 /* step 3: recover data indices */
421 start = start_bidx_of_node(ofs_of_node(page), inode);
422 end = start + ADDRS_PER_PAGE(page, inode);
424 set_new_dnode(&dn, inode, NULL, NULL, 0);
426 err = get_dnode_of_data(&dn, start, ALLOC_NODE);
428 if (err == -ENOMEM) {
429 congestion_wait(BLK_RW_ASYNC, HZ/50);
435 f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
437 get_node_info(sbi, dn.nid, &ni);
438 f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
439 f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
441 for (; start < end; start++, dn.ofs_in_node++) {
444 src = datablock_addr(dn.node_page, dn.ofs_in_node);
445 dest = datablock_addr(page, dn.ofs_in_node);
447 /* skip recovering if dest is the same as src */
451 /* dest is invalid, just invalidate src block */
452 if (dest == NULL_ADDR) {
453 truncate_data_blocks_range(&dn, 1);
457 if ((start + 1) << PAGE_SHIFT > i_size_read(inode))
458 f2fs_i_size_write(inode, (start + 1) << PAGE_SHIFT);
461 * dest is reserved block, invalidate src block
462 * and then reserve one new block in dnode page.
464 if (dest == NEW_ADDR) {
465 truncate_data_blocks_range(&dn, 1);
466 reserve_new_block(&dn);
470 /* dest is valid block, try to recover from src to dest */
471 if (is_valid_blkaddr(sbi, dest, META_POR)) {
473 if (src == NULL_ADDR) {
474 err = reserve_new_block(&dn);
475 #ifdef CONFIG_F2FS_FAULT_INJECTION
477 err = reserve_new_block(&dn);
479 /* We should not get -ENOSPC */
480 f2fs_bug_on(sbi, err);
485 /* Check the previous node page having this index */
486 err = check_index_in_prev_nodes(sbi, dest, &dn);
488 if (err == -ENOMEM) {
489 congestion_wait(BLK_RW_ASYNC, HZ/50);
495 /* write dummy data page */
496 f2fs_replace_block(sbi, &dn, src, dest,
497 ni.version, false, false);
502 copy_node_footer(dn.node_page, page);
503 fill_node_footer(dn.node_page, dn.nid, ni.ino,
504 ofs_of_node(page), false);
505 set_page_dirty(dn.node_page);
509 f2fs_msg(sbi->sb, KERN_NOTICE,
510 "recover_data: ino = %lx, recovered = %d blocks, err = %d",
511 inode->i_ino, recovered, err);
515 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
516 struct list_head *dir_list)
518 struct curseg_info *curseg;
519 struct page *page = NULL;
523 /* get node pages in the current segment */
524 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
525 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
528 struct fsync_inode_entry *entry;
530 if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
533 ra_meta_pages_cond(sbi, blkaddr);
535 page = get_tmp_page(sbi, blkaddr);
537 if (!is_recoverable_dnode(page)) {
538 f2fs_put_page(page, 1);
542 entry = get_fsync_inode(inode_list, ino_of_node(page));
546 * inode(x) | CP | inode(x) | dnode(F)
547 * In this case, we can lose the latest inode(x).
548 * So, call recover_inode for the inode update.
551 recover_inode(entry->inode, page);
552 if (entry->last_dentry == blkaddr) {
553 err = recover_dentry(entry->inode, page, dir_list);
555 f2fs_put_page(page, 1);
559 err = do_recover_data(sbi, entry->inode, page, blkaddr);
561 f2fs_put_page(page, 1);
565 if (entry->blkaddr == blkaddr)
566 del_fsync_inode(entry);
568 /* check next segment */
569 blkaddr = next_blkaddr_of_node(page);
570 f2fs_put_page(page, 1);
573 allocate_new_segments(sbi);
577 int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
579 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
580 struct list_head inode_list;
581 struct list_head dir_list;
585 bool need_writecp = false;
587 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
588 sizeof(struct fsync_inode_entry));
589 if (!fsync_entry_slab)
592 INIT_LIST_HEAD(&inode_list);
593 INIT_LIST_HEAD(&dir_list);
595 /* prevent checkpoint */
596 mutex_lock(&sbi->cp_mutex);
598 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
600 /* step #1: find fsynced inode numbers */
601 err = find_fsync_dnodes(sbi, &inode_list);
602 if (err || list_empty(&inode_list))
612 /* step #2: recover data */
613 err = recover_data(sbi, &inode_list, &dir_list);
615 f2fs_bug_on(sbi, !list_empty(&inode_list));
617 destroy_fsync_dnodes(&inode_list);
619 /* truncate meta pages to be used by the recovery */
620 truncate_inode_pages_range(META_MAPPING(sbi),
621 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
624 truncate_inode_pages_final(NODE_MAPPING(sbi));
625 truncate_inode_pages_final(META_MAPPING(sbi));
628 clear_sbi_flag(sbi, SBI_POR_DOING);
630 set_ckpt_flags(sbi, CP_ERROR_FLAG);
631 mutex_unlock(&sbi->cp_mutex);
633 /* let's drop all the directory inodes for clean checkpoint */
634 destroy_fsync_dnodes(&dir_list);
636 if (!err && need_writecp) {
637 struct cp_control cpc = {
638 .reason = CP_RECOVERY,
640 err = write_checkpoint(sbi, &cpc);
643 kmem_cache_destroy(fsync_entry_slab);
644 return ret ? ret: err;