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 if (sbi->last_valid_block_count + sbi->alloc_valid_block_count
53 > sbi->user_block_count)
58 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
61 struct fsync_inode_entry *entry;
63 list_for_each_entry(entry, head, list)
64 if (entry->inode->i_ino == ino)
70 static int recover_dentry(struct inode *inode, struct page *ipage)
72 struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
73 nid_t pino = le32_to_cpu(raw_inode->i_pino);
74 struct f2fs_dir_entry *de;
77 struct inode *dir, *einode;
80 dir = f2fs_iget(inode->i_sb, pino);
86 if (file_enc_name(inode)) {
91 name.len = le32_to_cpu(raw_inode->i_namelen);
92 name.name = raw_inode->i_name;
94 if (unlikely(name.len > F2FS_NAME_LEN)) {
100 de = f2fs_find_entry(dir, &name, &page);
101 if (de && inode->i_ino == le32_to_cpu(de->ino))
105 einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
106 if (IS_ERR(einode)) {
108 err = PTR_ERR(einode);
113 err = acquire_orphan_inode(F2FS_I_SB(inode));
118 f2fs_delete_entry(de, page, dir, einode);
122 err = __f2fs_add_link(dir, &name, inode, inode->i_ino, inode->i_mode);
126 if (is_inode_flag_set(F2FS_I(dir), FI_DELAY_IPUT)) {
129 add_dirty_dir_inode(dir);
130 set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
136 f2fs_dentry_kunmap(dir, page);
137 f2fs_put_page(page, 0);
141 f2fs_msg(inode->i_sb, KERN_NOTICE,
142 "%s: ino = %x, name = %s, dir = %lx, err = %d",
143 __func__, ino_of_node(ipage), raw_inode->i_name,
144 IS_ERR(dir) ? 0 : dir->i_ino, err);
148 static void recover_inode(struct inode *inode, struct page *page)
150 struct f2fs_inode *raw = F2FS_INODE(page);
153 inode->i_mode = le16_to_cpu(raw->i_mode);
154 i_size_write(inode, le64_to_cpu(raw->i_size));
155 inode->i_atime.tv_sec = le64_to_cpu(raw->i_mtime);
156 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
157 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
158 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
159 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
160 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
162 if (file_enc_name(inode))
163 name = "<encrypted>";
165 name = F2FS_INODE(page)->i_name;
167 f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
168 ino_of_node(page), name);
171 static bool is_same_inode(struct inode *inode, struct page *ipage)
173 struct f2fs_inode *ri = F2FS_INODE(ipage);
174 struct timespec disk;
176 if (!IS_INODE(ipage))
179 disk.tv_sec = le64_to_cpu(ri->i_ctime);
180 disk.tv_nsec = le32_to_cpu(ri->i_ctime_nsec);
181 if (timespec_compare(&inode->i_ctime, &disk) > 0)
184 disk.tv_sec = le64_to_cpu(ri->i_atime);
185 disk.tv_nsec = le32_to_cpu(ri->i_atime_nsec);
186 if (timespec_compare(&inode->i_atime, &disk) > 0)
189 disk.tv_sec = le64_to_cpu(ri->i_mtime);
190 disk.tv_nsec = le32_to_cpu(ri->i_mtime_nsec);
191 if (timespec_compare(&inode->i_mtime, &disk) > 0)
197 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
199 unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
200 struct curseg_info *curseg;
201 struct page *page = NULL;
205 /* get node pages in the current segment */
206 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
207 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
209 ra_meta_pages(sbi, blkaddr, 1, META_POR, true);
212 struct fsync_inode_entry *entry;
214 if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
217 page = get_tmp_page(sbi, blkaddr);
219 if (cp_ver != cpver_of_node(page))
222 if (!is_fsync_dnode(page))
225 entry = get_fsync_inode(head, ino_of_node(page));
227 if (!is_same_inode(entry->inode, page))
230 if (IS_INODE(page) && is_dent_dnode(page)) {
231 err = recover_inode_page(sbi, page);
236 /* add this fsync inode to the list */
237 entry = kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
243 * CP | dnode(F) | inode(DF)
244 * For this case, we should not give up now.
246 entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
247 if (IS_ERR(entry->inode)) {
248 err = PTR_ERR(entry->inode);
249 kmem_cache_free(fsync_entry_slab, entry);
250 if (err == -ENOENT) {
256 list_add_tail(&entry->list, head);
258 entry->blkaddr = blkaddr;
260 if (IS_INODE(page)) {
261 entry->last_inode = blkaddr;
262 if (is_dent_dnode(page))
263 entry->last_dentry = blkaddr;
266 /* check next segment */
267 blkaddr = next_blkaddr_of_node(page);
268 f2fs_put_page(page, 1);
270 ra_meta_pages_cond(sbi, blkaddr);
272 f2fs_put_page(page, 1);
276 static void destroy_fsync_dnodes(struct list_head *head)
278 struct fsync_inode_entry *entry, *tmp;
280 list_for_each_entry_safe(entry, tmp, head, list) {
282 list_del(&entry->list);
283 kmem_cache_free(fsync_entry_slab, entry);
287 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
288 block_t blkaddr, struct dnode_of_data *dn)
290 struct seg_entry *sentry;
291 unsigned int segno = GET_SEGNO(sbi, blkaddr);
292 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
293 struct f2fs_summary_block *sum_node;
294 struct f2fs_summary sum;
295 struct page *sum_page, *node_page;
296 struct dnode_of_data tdn = *dn;
303 sentry = get_seg_entry(sbi, segno);
304 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
307 /* Get the previous summary */
308 for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
309 struct curseg_info *curseg = CURSEG_I(sbi, i);
310 if (curseg->segno == segno) {
311 sum = curseg->sum_blk->entries[blkoff];
316 sum_page = get_sum_page(sbi, segno);
317 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
318 sum = sum_node->entries[blkoff];
319 f2fs_put_page(sum_page, 1);
321 /* Use the locked dnode page and inode */
322 nid = le32_to_cpu(sum.nid);
323 if (dn->inode->i_ino == nid) {
325 if (!dn->inode_page_locked)
326 lock_page(dn->inode_page);
327 tdn.node_page = dn->inode_page;
328 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
330 } else if (dn->nid == nid) {
331 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
335 /* Get the node page */
336 node_page = get_node_page(sbi, nid);
337 if (IS_ERR(node_page))
338 return PTR_ERR(node_page);
340 offset = ofs_of_node(node_page);
341 ino = ino_of_node(node_page);
342 f2fs_put_page(node_page, 1);
344 if (ino != dn->inode->i_ino) {
345 /* Deallocate previous index in the node page */
346 inode = f2fs_iget(sbi->sb, ino);
348 return PTR_ERR(inode);
353 bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
354 le16_to_cpu(sum.ofs_in_node);
357 * if inode page is locked, unlock temporarily, but its reference
360 if (ino == dn->inode->i_ino && dn->inode_page_locked)
361 unlock_page(dn->inode_page);
363 set_new_dnode(&tdn, inode, NULL, NULL, 0);
364 if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
367 if (tdn.data_blkaddr == blkaddr)
368 truncate_data_blocks_range(&tdn, 1);
370 f2fs_put_dnode(&tdn);
372 if (ino != dn->inode->i_ino)
374 else if (dn->inode_page_locked)
375 lock_page(dn->inode_page);
379 if (datablock_addr(tdn.node_page, tdn.ofs_in_node) == blkaddr)
380 truncate_data_blocks_range(&tdn, 1);
381 if (dn->inode->i_ino == nid && !dn->inode_page_locked)
382 unlock_page(dn->inode_page);
386 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
387 struct page *page, block_t blkaddr)
389 struct f2fs_inode_info *fi = F2FS_I(inode);
390 unsigned int start, end;
391 struct dnode_of_data dn;
393 int err = 0, recovered = 0;
395 /* step 1: recover xattr */
396 if (IS_INODE(page)) {
397 recover_inline_xattr(inode, page);
398 } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
400 * Deprecated; xattr blocks should be found from cold log.
401 * But, we should remain this for backward compatibility.
403 recover_xattr_data(inode, page, blkaddr);
407 /* step 2: recover inline data */
408 if (recover_inline_data(inode, page))
411 /* step 3: recover data indices */
412 start = start_bidx_of_node(ofs_of_node(page), fi);
413 end = start + ADDRS_PER_PAGE(page, fi);
415 set_new_dnode(&dn, inode, NULL, NULL, 0);
417 err = get_dnode_of_data(&dn, start, ALLOC_NODE);
421 f2fs_wait_on_page_writeback(dn.node_page, NODE);
423 get_node_info(sbi, dn.nid, &ni);
424 f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
425 f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
427 for (; start < end; start++, dn.ofs_in_node++) {
430 src = datablock_addr(dn.node_page, dn.ofs_in_node);
431 dest = datablock_addr(page, dn.ofs_in_node);
433 /* skip recovering if dest is the same as src */
437 /* dest is invalid, just invalidate src block */
438 if (dest == NULL_ADDR) {
439 truncate_data_blocks_range(&dn, 1);
444 * dest is reserved block, invalidate src block
445 * and then reserve one new block in dnode page.
447 if (dest == NEW_ADDR) {
448 truncate_data_blocks_range(&dn, 1);
449 err = reserve_new_block(&dn);
450 f2fs_bug_on(sbi, err);
454 /* dest is valid block, try to recover from src to dest */
455 if (is_valid_blkaddr(sbi, dest, META_POR)) {
457 if (src == NULL_ADDR) {
458 err = reserve_new_block(&dn);
459 /* We should not get -ENOSPC */
460 f2fs_bug_on(sbi, err);
463 /* Check the previous node page having this index */
464 err = check_index_in_prev_nodes(sbi, dest, &dn);
468 /* write dummy data page */
469 f2fs_replace_block(sbi, &dn, src, dest,
475 if (IS_INODE(dn.node_page))
476 sync_inode_page(&dn);
478 copy_node_footer(dn.node_page, page);
479 fill_node_footer(dn.node_page, dn.nid, ni.ino,
480 ofs_of_node(page), false);
481 set_page_dirty(dn.node_page);
485 f2fs_msg(sbi->sb, KERN_NOTICE,
486 "recover_data: ino = %lx, recovered = %d blocks, err = %d",
487 inode->i_ino, recovered, err);
491 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *head)
493 unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
494 struct curseg_info *curseg;
495 struct page *page = NULL;
499 /* get node pages in the current segment */
500 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
501 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
504 struct fsync_inode_entry *entry;
506 if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
509 ra_meta_pages_cond(sbi, blkaddr);
511 page = get_tmp_page(sbi, blkaddr);
513 if (cp_ver != cpver_of_node(page)) {
514 f2fs_put_page(page, 1);
518 entry = get_fsync_inode(head, ino_of_node(page));
522 * inode(x) | CP | inode(x) | dnode(F)
523 * In this case, we can lose the latest inode(x).
524 * So, call recover_inode for the inode update.
526 if (entry->last_inode == blkaddr)
527 recover_inode(entry->inode, page);
528 if (entry->last_dentry == blkaddr) {
529 err = recover_dentry(entry->inode, page);
531 f2fs_put_page(page, 1);
535 err = do_recover_data(sbi, entry->inode, page, blkaddr);
537 f2fs_put_page(page, 1);
541 if (entry->blkaddr == blkaddr) {
543 list_del(&entry->list);
544 kmem_cache_free(fsync_entry_slab, entry);
547 /* check next segment */
548 blkaddr = next_blkaddr_of_node(page);
549 f2fs_put_page(page, 1);
552 allocate_new_segments(sbi);
556 int recover_fsync_data(struct f2fs_sb_info *sbi)
558 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
559 struct list_head inode_list;
562 bool need_writecp = false;
564 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
565 sizeof(struct fsync_inode_entry));
566 if (!fsync_entry_slab)
569 INIT_LIST_HEAD(&inode_list);
571 /* prevent checkpoint */
572 mutex_lock(&sbi->cp_mutex);
574 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
576 /* step #1: find fsynced inode numbers */
577 err = find_fsync_dnodes(sbi, &inode_list);
581 if (list_empty(&inode_list))
586 /* step #2: recover data */
587 err = recover_data(sbi, &inode_list);
589 f2fs_bug_on(sbi, !list_empty(&inode_list));
591 destroy_fsync_dnodes(&inode_list);
592 kmem_cache_destroy(fsync_entry_slab);
594 /* truncate meta pages to be used by the recovery */
595 truncate_inode_pages_range(META_MAPPING(sbi),
596 (loff_t)MAIN_BLKADDR(sbi) << PAGE_CACHE_SHIFT, -1);
599 truncate_inode_pages_final(NODE_MAPPING(sbi));
600 truncate_inode_pages_final(META_MAPPING(sbi));
603 clear_sbi_flag(sbi, SBI_POR_DOING);
605 bool invalidate = false;
607 if (discard_next_dnode(sbi, blkaddr))
610 /* Flush all the NAT/SIT pages */
611 while (get_pages(sbi, F2FS_DIRTY_META))
612 sync_meta_pages(sbi, META, LONG_MAX);
614 /* invalidate temporary meta page */
616 invalidate_mapping_pages(META_MAPPING(sbi),
619 set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
620 mutex_unlock(&sbi->cp_mutex);
621 } else if (need_writecp) {
622 struct cp_control cpc = {
623 .reason = CP_RECOVERY,
625 mutex_unlock(&sbi->cp_mutex);
626 err = write_checkpoint(sbi, &cpc);
628 mutex_unlock(&sbi->cp_mutex);