xref: /linux/fs/f2fs/recovery.c (revision d195c39052d1da278a00a6744ce59c383b67b191)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/recovery.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include "f2fs.h"
11 #include "node.h"
12 #include "segment.h"
13 
14 /*
15  * Roll forward recovery scenarios.
16  *
17  * [Term] F: fsync_mark, D: dentry_mark
18  *
19  * 1. inode(x) | CP | inode(x) | dnode(F)
20  * -> Update the latest inode(x).
21  *
22  * 2. inode(x) | CP | inode(F) | dnode(F)
23  * -> No problem.
24  *
25  * 3. inode(x) | CP | dnode(F) | inode(x)
26  * -> Recover to the latest dnode(F), and drop the last inode(x)
27  *
28  * 4. inode(x) | CP | dnode(F) | inode(F)
29  * -> No problem.
30  *
31  * 5. CP | inode(x) | dnode(F)
32  * -> The inode(DF) was missing. Should drop this dnode(F).
33  *
34  * 6. CP | inode(DF) | dnode(F)
35  * -> No problem.
36  *
37  * 7. CP | dnode(F) | inode(DF)
38  * -> If f2fs_iget fails, then goto next to find inode(DF).
39  *
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).
43  */
44 
45 static struct kmem_cache *fsync_entry_slab;
46 
47 bool f2fs_space_for_roll_forward(struct f2fs_sb_info *sbi)
48 {
49 	s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count);
50 
51 	if (sbi->last_valid_block_count + nalloc > sbi->user_block_count)
52 		return false;
53 	return true;
54 }
55 
56 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
57 								nid_t ino)
58 {
59 	struct fsync_inode_entry *entry;
60 
61 	list_for_each_entry(entry, head, list)
62 		if (entry->inode->i_ino == ino)
63 			return entry;
64 
65 	return NULL;
66 }
67 
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)
70 {
71 	struct inode *inode;
72 	struct fsync_inode_entry *entry;
73 	int err;
74 
75 	inode = f2fs_iget_retry(sbi->sb, ino);
76 	if (IS_ERR(inode))
77 		return ERR_CAST(inode);
78 
79 	err = dquot_initialize(inode);
80 	if (err)
81 		goto err_out;
82 
83 	if (quota_inode) {
84 		err = dquot_alloc_inode(inode);
85 		if (err)
86 			goto err_out;
87 	}
88 
89 	entry = f2fs_kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
90 	entry->inode = inode;
91 	list_add_tail(&entry->list, head);
92 
93 	return entry;
94 err_out:
95 	iput(inode);
96 	return ERR_PTR(err);
97 }
98 
99 static void del_fsync_inode(struct fsync_inode_entry *entry, int drop)
100 {
101 	if (drop) {
102 		/* inode should not be recovered, drop it */
103 		f2fs_inode_synced(entry->inode);
104 	}
105 	iput(entry->inode);
106 	list_del(&entry->list);
107 	kmem_cache_free(fsync_entry_slab, entry);
108 }
109 
110 static int recover_dentry(struct inode *inode, struct page *ipage,
111 						struct list_head *dir_list)
112 {
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;
117 	struct page *page;
118 	struct inode *dir, *einode;
119 	struct fsync_inode_entry *entry;
120 	int err = 0;
121 	char *name;
122 
123 	entry = get_fsync_inode(dir_list, pino);
124 	if (!entry) {
125 		entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
126 							pino, false);
127 		if (IS_ERR(entry)) {
128 			dir = ERR_CAST(entry);
129 			err = PTR_ERR(entry);
130 			goto out;
131 		}
132 	}
133 
134 	dir = entry->inode;
135 
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;
139 
140 	if (unlikely(fname.disk_name.len > F2FS_NAME_LEN)) {
141 		WARN_ON(1);
142 		err = -ENAMETOOLONG;
143 		goto out;
144 	}
145 retry:
146 	de = __f2fs_find_entry(dir, &fname, &page);
147 	if (de && inode->i_ino == le32_to_cpu(de->ino))
148 		goto out_put;
149 
150 	if (de) {
151 		einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
152 		if (IS_ERR(einode)) {
153 			WARN_ON(1);
154 			err = PTR_ERR(einode);
155 			if (err == -ENOENT)
156 				err = -EEXIST;
157 			goto out_put;
158 		}
159 
160 		err = dquot_initialize(einode);
161 		if (err) {
162 			iput(einode);
163 			goto out_put;
164 		}
165 
166 		err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
167 		if (err) {
168 			iput(einode);
169 			goto out_put;
170 		}
171 		f2fs_delete_entry(de, page, dir, einode);
172 		iput(einode);
173 		goto retry;
174 	} else if (IS_ERR(page)) {
175 		err = PTR_ERR(page);
176 	} else {
177 		err = f2fs_add_dentry(dir, &fname, inode,
178 					inode->i_ino, inode->i_mode);
179 	}
180 	if (err == -ENOMEM)
181 		goto retry;
182 	goto out;
183 
184 out_put:
185 	f2fs_put_page(page, 0);
186 out:
187 	if (file_enc_name(inode))
188 		name = "<encrypted>";
189 	else
190 		name = raw_inode->i_name;
191 	f2fs_notice(F2FS_I_SB(inode), "%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);
194 	return err;
195 }
196 
197 static int recover_quota_data(struct inode *inode, struct page *page)
198 {
199 	struct f2fs_inode *raw = F2FS_INODE(page);
200 	struct iattr attr;
201 	uid_t i_uid = le32_to_cpu(raw->i_uid);
202 	gid_t i_gid = le32_to_cpu(raw->i_gid);
203 	int err;
204 
205 	memset(&attr, 0, sizeof(attr));
206 
207 	attr.ia_uid = make_kuid(inode->i_sb->s_user_ns, i_uid);
208 	attr.ia_gid = make_kgid(inode->i_sb->s_user_ns, i_gid);
209 
210 	if (!uid_eq(attr.ia_uid, inode->i_uid))
211 		attr.ia_valid |= ATTR_UID;
212 	if (!gid_eq(attr.ia_gid, inode->i_gid))
213 		attr.ia_valid |= ATTR_GID;
214 
215 	if (!attr.ia_valid)
216 		return 0;
217 
218 	err = dquot_transfer(inode, &attr);
219 	if (err)
220 		set_sbi_flag(F2FS_I_SB(inode), SBI_QUOTA_NEED_REPAIR);
221 	return err;
222 }
223 
224 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
225 {
226 	if (ri->i_inline & F2FS_PIN_FILE)
227 		set_inode_flag(inode, FI_PIN_FILE);
228 	else
229 		clear_inode_flag(inode, FI_PIN_FILE);
230 	if (ri->i_inline & F2FS_DATA_EXIST)
231 		set_inode_flag(inode, FI_DATA_EXIST);
232 	else
233 		clear_inode_flag(inode, FI_DATA_EXIST);
234 }
235 
236 static int recover_inode(struct inode *inode, struct page *page)
237 {
238 	struct f2fs_inode *raw = F2FS_INODE(page);
239 	char *name;
240 	int err;
241 
242 	inode->i_mode = le16_to_cpu(raw->i_mode);
243 
244 	err = recover_quota_data(inode, page);
245 	if (err)
246 		return err;
247 
248 	i_uid_write(inode, le32_to_cpu(raw->i_uid));
249 	i_gid_write(inode, le32_to_cpu(raw->i_gid));
250 
251 	if (raw->i_inline & F2FS_EXTRA_ATTR) {
252 		if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)) &&
253 			F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize),
254 								i_projid)) {
255 			projid_t i_projid;
256 			kprojid_t kprojid;
257 
258 			i_projid = (projid_t)le32_to_cpu(raw->i_projid);
259 			kprojid = make_kprojid(&init_user_ns, i_projid);
260 
261 			if (!projid_eq(kprojid, F2FS_I(inode)->i_projid)) {
262 				err = f2fs_transfer_project_quota(inode,
263 								kprojid);
264 				if (err)
265 					return err;
266 				F2FS_I(inode)->i_projid = kprojid;
267 			}
268 		}
269 	}
270 
271 	f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
272 	inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
273 	inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
274 	inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
275 	inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
276 	inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
277 	inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
278 
279 	F2FS_I(inode)->i_advise = raw->i_advise;
280 	F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags);
281 	f2fs_set_inode_flags(inode);
282 	F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] =
283 				le16_to_cpu(raw->i_gc_failures);
284 
285 	recover_inline_flags(inode, raw);
286 
287 	f2fs_mark_inode_dirty_sync(inode, true);
288 
289 	if (file_enc_name(inode))
290 		name = "<encrypted>";
291 	else
292 		name = F2FS_INODE(page)->i_name;
293 
294 	f2fs_notice(F2FS_I_SB(inode), "recover_inode: ino = %x, name = %s, inline = %x",
295 		    ino_of_node(page), name, raw->i_inline);
296 	return 0;
297 }
298 
299 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
300 				bool check_only)
301 {
302 	struct curseg_info *curseg;
303 	struct page *page = NULL;
304 	block_t blkaddr;
305 	unsigned int loop_cnt = 0;
306 	unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg -
307 						valid_user_blocks(sbi);
308 	int err = 0;
309 
310 	/* get node pages in the current segment */
311 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
312 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
313 
314 	while (1) {
315 		struct fsync_inode_entry *entry;
316 
317 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
318 			return 0;
319 
320 		page = f2fs_get_tmp_page(sbi, blkaddr);
321 		if (IS_ERR(page)) {
322 			err = PTR_ERR(page);
323 			break;
324 		}
325 
326 		if (!is_recoverable_dnode(page)) {
327 			f2fs_put_page(page, 1);
328 			break;
329 		}
330 
331 		if (!is_fsync_dnode(page))
332 			goto next;
333 
334 		entry = get_fsync_inode(head, ino_of_node(page));
335 		if (!entry) {
336 			bool quota_inode = false;
337 
338 			if (!check_only &&
339 					IS_INODE(page) && is_dent_dnode(page)) {
340 				err = f2fs_recover_inode_page(sbi, page);
341 				if (err) {
342 					f2fs_put_page(page, 1);
343 					break;
344 				}
345 				quota_inode = true;
346 			}
347 
348 			/*
349 			 * CP | dnode(F) | inode(DF)
350 			 * For this case, we should not give up now.
351 			 */
352 			entry = add_fsync_inode(sbi, head, ino_of_node(page),
353 								quota_inode);
354 			if (IS_ERR(entry)) {
355 				err = PTR_ERR(entry);
356 				if (err == -ENOENT) {
357 					err = 0;
358 					goto next;
359 				}
360 				f2fs_put_page(page, 1);
361 				break;
362 			}
363 		}
364 		entry->blkaddr = blkaddr;
365 
366 		if (IS_INODE(page) && is_dent_dnode(page))
367 			entry->last_dentry = blkaddr;
368 next:
369 		/* sanity check in order to detect looped node chain */
370 		if (++loop_cnt >= free_blocks ||
371 			blkaddr == next_blkaddr_of_node(page)) {
372 			f2fs_notice(sbi, "%s: detect looped node chain, blkaddr:%u, next:%u",
373 				    __func__, blkaddr,
374 				    next_blkaddr_of_node(page));
375 			f2fs_put_page(page, 1);
376 			err = -EINVAL;
377 			break;
378 		}
379 
380 		/* check next segment */
381 		blkaddr = next_blkaddr_of_node(page);
382 		f2fs_put_page(page, 1);
383 
384 		f2fs_ra_meta_pages_cond(sbi, blkaddr);
385 	}
386 	return err;
387 }
388 
389 static void destroy_fsync_dnodes(struct list_head *head, int drop)
390 {
391 	struct fsync_inode_entry *entry, *tmp;
392 
393 	list_for_each_entry_safe(entry, tmp, head, list)
394 		del_fsync_inode(entry, drop);
395 }
396 
397 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
398 			block_t blkaddr, struct dnode_of_data *dn)
399 {
400 	struct seg_entry *sentry;
401 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
402 	unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
403 	struct f2fs_summary_block *sum_node;
404 	struct f2fs_summary sum;
405 	struct page *sum_page, *node_page;
406 	struct dnode_of_data tdn = *dn;
407 	nid_t ino, nid;
408 	struct inode *inode;
409 	unsigned int offset;
410 	block_t bidx;
411 	int i;
412 
413 	sentry = get_seg_entry(sbi, segno);
414 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
415 		return 0;
416 
417 	/* Get the previous summary */
418 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
419 		struct curseg_info *curseg = CURSEG_I(sbi, i);
420 		if (curseg->segno == segno) {
421 			sum = curseg->sum_blk->entries[blkoff];
422 			goto got_it;
423 		}
424 	}
425 
426 	sum_page = f2fs_get_sum_page(sbi, segno);
427 	if (IS_ERR(sum_page))
428 		return PTR_ERR(sum_page);
429 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
430 	sum = sum_node->entries[blkoff];
431 	f2fs_put_page(sum_page, 1);
432 got_it:
433 	/* Use the locked dnode page and inode */
434 	nid = le32_to_cpu(sum.nid);
435 	if (dn->inode->i_ino == nid) {
436 		tdn.nid = nid;
437 		if (!dn->inode_page_locked)
438 			lock_page(dn->inode_page);
439 		tdn.node_page = dn->inode_page;
440 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
441 		goto truncate_out;
442 	} else if (dn->nid == nid) {
443 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
444 		goto truncate_out;
445 	}
446 
447 	/* Get the node page */
448 	node_page = f2fs_get_node_page(sbi, nid);
449 	if (IS_ERR(node_page))
450 		return PTR_ERR(node_page);
451 
452 	offset = ofs_of_node(node_page);
453 	ino = ino_of_node(node_page);
454 	f2fs_put_page(node_page, 1);
455 
456 	if (ino != dn->inode->i_ino) {
457 		int ret;
458 
459 		/* Deallocate previous index in the node page */
460 		inode = f2fs_iget_retry(sbi->sb, ino);
461 		if (IS_ERR(inode))
462 			return PTR_ERR(inode);
463 
464 		ret = dquot_initialize(inode);
465 		if (ret) {
466 			iput(inode);
467 			return ret;
468 		}
469 	} else {
470 		inode = dn->inode;
471 	}
472 
473 	bidx = f2fs_start_bidx_of_node(offset, inode) +
474 				le16_to_cpu(sum.ofs_in_node);
475 
476 	/*
477 	 * if inode page is locked, unlock temporarily, but its reference
478 	 * count keeps alive.
479 	 */
480 	if (ino == dn->inode->i_ino && dn->inode_page_locked)
481 		unlock_page(dn->inode_page);
482 
483 	set_new_dnode(&tdn, inode, NULL, NULL, 0);
484 	if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
485 		goto out;
486 
487 	if (tdn.data_blkaddr == blkaddr)
488 		f2fs_truncate_data_blocks_range(&tdn, 1);
489 
490 	f2fs_put_dnode(&tdn);
491 out:
492 	if (ino != dn->inode->i_ino)
493 		iput(inode);
494 	else if (dn->inode_page_locked)
495 		lock_page(dn->inode_page);
496 	return 0;
497 
498 truncate_out:
499 	if (f2fs_data_blkaddr(&tdn) == blkaddr)
500 		f2fs_truncate_data_blocks_range(&tdn, 1);
501 	if (dn->inode->i_ino == nid && !dn->inode_page_locked)
502 		unlock_page(dn->inode_page);
503 	return 0;
504 }
505 
506 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
507 					struct page *page)
508 {
509 	struct dnode_of_data dn;
510 	struct node_info ni;
511 	unsigned int start, end;
512 	int err = 0, recovered = 0;
513 
514 	/* step 1: recover xattr */
515 	if (IS_INODE(page)) {
516 		f2fs_recover_inline_xattr(inode, page);
517 	} else if (f2fs_has_xattr_block(ofs_of_node(page))) {
518 		err = f2fs_recover_xattr_data(inode, page);
519 		if (!err)
520 			recovered++;
521 		goto out;
522 	}
523 
524 	/* step 2: recover inline data */
525 	if (f2fs_recover_inline_data(inode, page))
526 		goto out;
527 
528 	/* step 3: recover data indices */
529 	start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
530 	end = start + ADDRS_PER_PAGE(page, inode);
531 
532 	set_new_dnode(&dn, inode, NULL, NULL, 0);
533 retry_dn:
534 	err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
535 	if (err) {
536 		if (err == -ENOMEM) {
537 			congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
538 			goto retry_dn;
539 		}
540 		goto out;
541 	}
542 
543 	f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
544 
545 	err = f2fs_get_node_info(sbi, dn.nid, &ni);
546 	if (err)
547 		goto err;
548 
549 	f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
550 
551 	if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
552 		f2fs_warn(sbi, "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
553 			  inode->i_ino, ofs_of_node(dn.node_page),
554 			  ofs_of_node(page));
555 		err = -EFSCORRUPTED;
556 		goto err;
557 	}
558 
559 	for (; start < end; start++, dn.ofs_in_node++) {
560 		block_t src, dest;
561 
562 		src = f2fs_data_blkaddr(&dn);
563 		dest = data_blkaddr(dn.inode, page, dn.ofs_in_node);
564 
565 		if (__is_valid_data_blkaddr(src) &&
566 			!f2fs_is_valid_blkaddr(sbi, src, META_POR)) {
567 			err = -EFSCORRUPTED;
568 			goto err;
569 		}
570 
571 		if (__is_valid_data_blkaddr(dest) &&
572 			!f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
573 			err = -EFSCORRUPTED;
574 			goto err;
575 		}
576 
577 		/* skip recovering if dest is the same as src */
578 		if (src == dest)
579 			continue;
580 
581 		/* dest is invalid, just invalidate src block */
582 		if (dest == NULL_ADDR) {
583 			f2fs_truncate_data_blocks_range(&dn, 1);
584 			continue;
585 		}
586 
587 		if (!file_keep_isize(inode) &&
588 			(i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
589 			f2fs_i_size_write(inode,
590 				(loff_t)(start + 1) << PAGE_SHIFT);
591 
592 		/*
593 		 * dest is reserved block, invalidate src block
594 		 * and then reserve one new block in dnode page.
595 		 */
596 		if (dest == NEW_ADDR) {
597 			f2fs_truncate_data_blocks_range(&dn, 1);
598 			f2fs_reserve_new_block(&dn);
599 			continue;
600 		}
601 
602 		/* dest is valid block, try to recover from src to dest */
603 		if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
604 
605 			if (src == NULL_ADDR) {
606 				err = f2fs_reserve_new_block(&dn);
607 				while (err &&
608 				       IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION))
609 					err = f2fs_reserve_new_block(&dn);
610 				/* We should not get -ENOSPC */
611 				f2fs_bug_on(sbi, err);
612 				if (err)
613 					goto err;
614 			}
615 retry_prev:
616 			/* Check the previous node page having this index */
617 			err = check_index_in_prev_nodes(sbi, dest, &dn);
618 			if (err) {
619 				if (err == -ENOMEM) {
620 					congestion_wait(BLK_RW_ASYNC,
621 							DEFAULT_IO_TIMEOUT);
622 					goto retry_prev;
623 				}
624 				goto err;
625 			}
626 
627 			/* write dummy data page */
628 			f2fs_replace_block(sbi, &dn, src, dest,
629 						ni.version, false, false);
630 			recovered++;
631 		}
632 	}
633 
634 	copy_node_footer(dn.node_page, page);
635 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
636 					ofs_of_node(page), false);
637 	set_page_dirty(dn.node_page);
638 err:
639 	f2fs_put_dnode(&dn);
640 out:
641 	f2fs_notice(sbi, "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
642 		    inode->i_ino, file_keep_isize(inode) ? "keep" : "recover",
643 		    recovered, err);
644 	return err;
645 }
646 
647 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
648 		struct list_head *tmp_inode_list, struct list_head *dir_list)
649 {
650 	struct curseg_info *curseg;
651 	struct page *page = NULL;
652 	int err = 0;
653 	block_t blkaddr;
654 
655 	/* get node pages in the current segment */
656 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
657 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
658 
659 	while (1) {
660 		struct fsync_inode_entry *entry;
661 
662 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
663 			break;
664 
665 		f2fs_ra_meta_pages_cond(sbi, blkaddr);
666 
667 		page = f2fs_get_tmp_page(sbi, blkaddr);
668 		if (IS_ERR(page)) {
669 			err = PTR_ERR(page);
670 			break;
671 		}
672 
673 		if (!is_recoverable_dnode(page)) {
674 			f2fs_put_page(page, 1);
675 			break;
676 		}
677 
678 		entry = get_fsync_inode(inode_list, ino_of_node(page));
679 		if (!entry)
680 			goto next;
681 		/*
682 		 * inode(x) | CP | inode(x) | dnode(F)
683 		 * In this case, we can lose the latest inode(x).
684 		 * So, call recover_inode for the inode update.
685 		 */
686 		if (IS_INODE(page)) {
687 			err = recover_inode(entry->inode, page);
688 			if (err) {
689 				f2fs_put_page(page, 1);
690 				break;
691 			}
692 		}
693 		if (entry->last_dentry == blkaddr) {
694 			err = recover_dentry(entry->inode, page, dir_list);
695 			if (err) {
696 				f2fs_put_page(page, 1);
697 				break;
698 			}
699 		}
700 		err = do_recover_data(sbi, entry->inode, page);
701 		if (err) {
702 			f2fs_put_page(page, 1);
703 			break;
704 		}
705 
706 		if (entry->blkaddr == blkaddr)
707 			list_move_tail(&entry->list, tmp_inode_list);
708 next:
709 		/* check next segment */
710 		blkaddr = next_blkaddr_of_node(page);
711 		f2fs_put_page(page, 1);
712 	}
713 	if (!err)
714 		f2fs_allocate_new_segments(sbi, NO_CHECK_TYPE);
715 	return err;
716 }
717 
718 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
719 {
720 	struct list_head inode_list, tmp_inode_list;
721 	struct list_head dir_list;
722 	int err;
723 	int ret = 0;
724 	unsigned long s_flags = sbi->sb->s_flags;
725 	bool need_writecp = false;
726 	bool fix_curseg_write_pointer = false;
727 #ifdef CONFIG_QUOTA
728 	int quota_enabled;
729 #endif
730 
731 	if (s_flags & SB_RDONLY) {
732 		f2fs_info(sbi, "recover fsync data on readonly fs");
733 		sbi->sb->s_flags &= ~SB_RDONLY;
734 	}
735 
736 #ifdef CONFIG_QUOTA
737 	/* Needed for iput() to work correctly and not trash data */
738 	sbi->sb->s_flags |= SB_ACTIVE;
739 	/* Turn on quotas so that they are updated correctly */
740 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
741 #endif
742 
743 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
744 			sizeof(struct fsync_inode_entry));
745 	if (!fsync_entry_slab) {
746 		err = -ENOMEM;
747 		goto out;
748 	}
749 
750 	INIT_LIST_HEAD(&inode_list);
751 	INIT_LIST_HEAD(&tmp_inode_list);
752 	INIT_LIST_HEAD(&dir_list);
753 
754 	/* prevent checkpoint */
755 	mutex_lock(&sbi->cp_mutex);
756 
757 	/* step #1: find fsynced inode numbers */
758 	err = find_fsync_dnodes(sbi, &inode_list, check_only);
759 	if (err || list_empty(&inode_list))
760 		goto skip;
761 
762 	if (check_only) {
763 		ret = 1;
764 		goto skip;
765 	}
766 
767 	need_writecp = true;
768 
769 	/* step #2: recover data */
770 	err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
771 	if (!err)
772 		f2fs_bug_on(sbi, !list_empty(&inode_list));
773 	else {
774 		/* restore s_flags to let iput() trash data */
775 		sbi->sb->s_flags = s_flags;
776 	}
777 skip:
778 	fix_curseg_write_pointer = !check_only || list_empty(&inode_list);
779 
780 	destroy_fsync_dnodes(&inode_list, err);
781 	destroy_fsync_dnodes(&tmp_inode_list, err);
782 
783 	/* truncate meta pages to be used by the recovery */
784 	truncate_inode_pages_range(META_MAPPING(sbi),
785 			(loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
786 
787 	if (err) {
788 		truncate_inode_pages_final(NODE_MAPPING(sbi));
789 		truncate_inode_pages_final(META_MAPPING(sbi));
790 	}
791 
792 	/*
793 	 * If fsync data succeeds or there is no fsync data to recover,
794 	 * and the f2fs is not read only, check and fix zoned block devices'
795 	 * write pointer consistency.
796 	 */
797 	if (!err && fix_curseg_write_pointer && !f2fs_readonly(sbi->sb) &&
798 			f2fs_sb_has_blkzoned(sbi)) {
799 		err = f2fs_fix_curseg_write_pointer(sbi);
800 		ret = err;
801 	}
802 
803 	if (!err)
804 		clear_sbi_flag(sbi, SBI_POR_DOING);
805 
806 	mutex_unlock(&sbi->cp_mutex);
807 
808 	/* let's drop all the directory inodes for clean checkpoint */
809 	destroy_fsync_dnodes(&dir_list, err);
810 
811 	if (need_writecp) {
812 		set_sbi_flag(sbi, SBI_IS_RECOVERED);
813 
814 		if (!err) {
815 			struct cp_control cpc = {
816 				.reason = CP_RECOVERY,
817 			};
818 			err = f2fs_write_checkpoint(sbi, &cpc);
819 		}
820 	}
821 
822 	kmem_cache_destroy(fsync_entry_slab);
823 out:
824 #ifdef CONFIG_QUOTA
825 	/* Turn quotas off */
826 	if (quota_enabled)
827 		f2fs_quota_off_umount(sbi->sb);
828 #endif
829 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
830 
831 	return ret ? ret: err;
832 }
833