xref: /linux/fs/f2fs/recovery.c (revision be58f7103700a68d5c7ca60a2bc0b309907599ab)
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 <asm/unaligned.h>
9 #include <linux/fs.h>
10 #include <linux/f2fs_fs.h>
11 #include "f2fs.h"
12 #include "node.h"
13 #include "segment.h"
14 
15 /*
16  * Roll forward recovery scenarios.
17  *
18  * [Term] F: fsync_mark, D: dentry_mark
19  *
20  * 1. inode(x) | CP | inode(x) | dnode(F)
21  * -> Update the latest inode(x).
22  *
23  * 2. inode(x) | CP | inode(F) | dnode(F)
24  * -> No problem.
25  *
26  * 3. inode(x) | CP | dnode(F) | inode(x)
27  * -> Recover to the latest dnode(F), and drop the last inode(x)
28  *
29  * 4. inode(x) | CP | dnode(F) | inode(F)
30  * -> No problem.
31  *
32  * 5. CP | inode(x) | dnode(F)
33  * -> The inode(DF) was missing. Should drop this dnode(F).
34  *
35  * 6. CP | inode(DF) | dnode(F)
36  * -> No problem.
37  *
38  * 7. CP | dnode(F) | inode(DF)
39  * -> If f2fs_iget fails, then goto next to find inode(DF).
40  *
41  * 8. CP | dnode(F) | inode(x)
42  * -> If f2fs_iget fails, then goto next to find inode(DF).
43  *    But it will fail due to no inode(DF).
44  */
45 
46 static struct kmem_cache *fsync_entry_slab;
47 
48 #ifdef CONFIG_UNICODE
49 extern struct kmem_cache *f2fs_cf_name_slab;
50 #endif
51 
52 bool f2fs_space_for_roll_forward(struct f2fs_sb_info *sbi)
53 {
54 	s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count);
55 
56 	if (sbi->last_valid_block_count + nalloc > sbi->user_block_count)
57 		return false;
58 	return true;
59 }
60 
61 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
62 								nid_t ino)
63 {
64 	struct fsync_inode_entry *entry;
65 
66 	list_for_each_entry(entry, head, list)
67 		if (entry->inode->i_ino == ino)
68 			return entry;
69 
70 	return NULL;
71 }
72 
73 static struct fsync_inode_entry *add_fsync_inode(struct f2fs_sb_info *sbi,
74 			struct list_head *head, nid_t ino, bool quota_inode)
75 {
76 	struct inode *inode;
77 	struct fsync_inode_entry *entry;
78 	int err;
79 
80 	inode = f2fs_iget_retry(sbi->sb, ino);
81 	if (IS_ERR(inode))
82 		return ERR_CAST(inode);
83 
84 	err = dquot_initialize(inode);
85 	if (err)
86 		goto err_out;
87 
88 	if (quota_inode) {
89 		err = dquot_alloc_inode(inode);
90 		if (err)
91 			goto err_out;
92 	}
93 
94 	entry = f2fs_kmem_cache_alloc(fsync_entry_slab,
95 					GFP_F2FS_ZERO, true, NULL);
96 	entry->inode = inode;
97 	list_add_tail(&entry->list, head);
98 
99 	return entry;
100 err_out:
101 	iput(inode);
102 	return ERR_PTR(err);
103 }
104 
105 static void del_fsync_inode(struct fsync_inode_entry *entry, int drop)
106 {
107 	if (drop) {
108 		/* inode should not be recovered, drop it */
109 		f2fs_inode_synced(entry->inode);
110 	}
111 	iput(entry->inode);
112 	list_del(&entry->list);
113 	kmem_cache_free(fsync_entry_slab, entry);
114 }
115 
116 static int init_recovered_filename(const struct inode *dir,
117 				   struct f2fs_inode *raw_inode,
118 				   struct f2fs_filename *fname,
119 				   struct qstr *usr_fname)
120 {
121 	int err;
122 
123 	memset(fname, 0, sizeof(*fname));
124 	fname->disk_name.len = le32_to_cpu(raw_inode->i_namelen);
125 	fname->disk_name.name = raw_inode->i_name;
126 
127 	if (WARN_ON(fname->disk_name.len > F2FS_NAME_LEN))
128 		return -ENAMETOOLONG;
129 
130 	if (!IS_ENCRYPTED(dir)) {
131 		usr_fname->name = fname->disk_name.name;
132 		usr_fname->len = fname->disk_name.len;
133 		fname->usr_fname = usr_fname;
134 	}
135 
136 	/* Compute the hash of the filename */
137 	if (IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)) {
138 		/*
139 		 * In this case the hash isn't computable without the key, so it
140 		 * was saved on-disk.
141 		 */
142 		if (fname->disk_name.len + sizeof(f2fs_hash_t) > F2FS_NAME_LEN)
143 			return -EINVAL;
144 		fname->hash = get_unaligned((f2fs_hash_t *)
145 				&raw_inode->i_name[fname->disk_name.len]);
146 	} else if (IS_CASEFOLDED(dir)) {
147 		err = f2fs_init_casefolded_name(dir, fname);
148 		if (err)
149 			return err;
150 		f2fs_hash_filename(dir, fname);
151 #ifdef CONFIG_UNICODE
152 		/* Case-sensitive match is fine for recovery */
153 		kmem_cache_free(f2fs_cf_name_slab, fname->cf_name.name);
154 		fname->cf_name.name = NULL;
155 #endif
156 	} else {
157 		f2fs_hash_filename(dir, fname);
158 	}
159 	return 0;
160 }
161 
162 static int recover_dentry(struct inode *inode, struct page *ipage,
163 						struct list_head *dir_list)
164 {
165 	struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
166 	nid_t pino = le32_to_cpu(raw_inode->i_pino);
167 	struct f2fs_dir_entry *de;
168 	struct f2fs_filename fname;
169 	struct qstr usr_fname;
170 	struct page *page;
171 	struct inode *dir, *einode;
172 	struct fsync_inode_entry *entry;
173 	int err = 0;
174 	char *name;
175 
176 	entry = get_fsync_inode(dir_list, pino);
177 	if (!entry) {
178 		entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
179 							pino, false);
180 		if (IS_ERR(entry)) {
181 			dir = ERR_CAST(entry);
182 			err = PTR_ERR(entry);
183 			goto out;
184 		}
185 	}
186 
187 	dir = entry->inode;
188 	err = init_recovered_filename(dir, raw_inode, &fname, &usr_fname);
189 	if (err)
190 		goto out;
191 retry:
192 	de = __f2fs_find_entry(dir, &fname, &page);
193 	if (de && inode->i_ino == le32_to_cpu(de->ino))
194 		goto out_put;
195 
196 	if (de) {
197 		einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
198 		if (IS_ERR(einode)) {
199 			WARN_ON(1);
200 			err = PTR_ERR(einode);
201 			if (err == -ENOENT)
202 				err = -EEXIST;
203 			goto out_put;
204 		}
205 
206 		err = dquot_initialize(einode);
207 		if (err) {
208 			iput(einode);
209 			goto out_put;
210 		}
211 
212 		err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
213 		if (err) {
214 			iput(einode);
215 			goto out_put;
216 		}
217 		f2fs_delete_entry(de, page, dir, einode);
218 		iput(einode);
219 		goto retry;
220 	} else if (IS_ERR(page)) {
221 		err = PTR_ERR(page);
222 	} else {
223 		err = f2fs_add_dentry(dir, &fname, inode,
224 					inode->i_ino, inode->i_mode);
225 	}
226 	if (err == -ENOMEM)
227 		goto retry;
228 	goto out;
229 
230 out_put:
231 	f2fs_put_page(page, 0);
232 out:
233 	if (file_enc_name(inode))
234 		name = "<encrypted>";
235 	else
236 		name = raw_inode->i_name;
237 	f2fs_notice(F2FS_I_SB(inode), "%s: ino = %x, name = %s, dir = %lx, err = %d",
238 		    __func__, ino_of_node(ipage), name,
239 		    IS_ERR(dir) ? 0 : dir->i_ino, err);
240 	return err;
241 }
242 
243 static int recover_quota_data(struct inode *inode, struct page *page)
244 {
245 	struct f2fs_inode *raw = F2FS_INODE(page);
246 	struct iattr attr;
247 	uid_t i_uid = le32_to_cpu(raw->i_uid);
248 	gid_t i_gid = le32_to_cpu(raw->i_gid);
249 	int err;
250 
251 	memset(&attr, 0, sizeof(attr));
252 
253 	attr.ia_uid = make_kuid(inode->i_sb->s_user_ns, i_uid);
254 	attr.ia_gid = make_kgid(inode->i_sb->s_user_ns, i_gid);
255 
256 	if (!uid_eq(attr.ia_uid, inode->i_uid))
257 		attr.ia_valid |= ATTR_UID;
258 	if (!gid_eq(attr.ia_gid, inode->i_gid))
259 		attr.ia_valid |= ATTR_GID;
260 
261 	if (!attr.ia_valid)
262 		return 0;
263 
264 	err = dquot_transfer(inode, &attr);
265 	if (err)
266 		set_sbi_flag(F2FS_I_SB(inode), SBI_QUOTA_NEED_REPAIR);
267 	return err;
268 }
269 
270 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
271 {
272 	if (ri->i_inline & F2FS_PIN_FILE)
273 		set_inode_flag(inode, FI_PIN_FILE);
274 	else
275 		clear_inode_flag(inode, FI_PIN_FILE);
276 	if (ri->i_inline & F2FS_DATA_EXIST)
277 		set_inode_flag(inode, FI_DATA_EXIST);
278 	else
279 		clear_inode_flag(inode, FI_DATA_EXIST);
280 }
281 
282 static int recover_inode(struct inode *inode, struct page *page)
283 {
284 	struct f2fs_inode *raw = F2FS_INODE(page);
285 	char *name;
286 	int err;
287 
288 	inode->i_mode = le16_to_cpu(raw->i_mode);
289 
290 	err = recover_quota_data(inode, page);
291 	if (err)
292 		return err;
293 
294 	i_uid_write(inode, le32_to_cpu(raw->i_uid));
295 	i_gid_write(inode, le32_to_cpu(raw->i_gid));
296 
297 	if (raw->i_inline & F2FS_EXTRA_ATTR) {
298 		if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)) &&
299 			F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize),
300 								i_projid)) {
301 			projid_t i_projid;
302 			kprojid_t kprojid;
303 
304 			i_projid = (projid_t)le32_to_cpu(raw->i_projid);
305 			kprojid = make_kprojid(&init_user_ns, i_projid);
306 
307 			if (!projid_eq(kprojid, F2FS_I(inode)->i_projid)) {
308 				err = f2fs_transfer_project_quota(inode,
309 								kprojid);
310 				if (err)
311 					return err;
312 				F2FS_I(inode)->i_projid = kprojid;
313 			}
314 		}
315 	}
316 
317 	f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
318 	inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
319 	inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
320 	inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
321 	inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
322 	inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
323 	inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
324 
325 	F2FS_I(inode)->i_advise = raw->i_advise;
326 	F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags);
327 	f2fs_set_inode_flags(inode);
328 	F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] =
329 				le16_to_cpu(raw->i_gc_failures);
330 
331 	recover_inline_flags(inode, raw);
332 
333 	f2fs_mark_inode_dirty_sync(inode, true);
334 
335 	if (file_enc_name(inode))
336 		name = "<encrypted>";
337 	else
338 		name = F2FS_INODE(page)->i_name;
339 
340 	f2fs_notice(F2FS_I_SB(inode), "recover_inode: ino = %x, name = %s, inline = %x",
341 		    ino_of_node(page), name, raw->i_inline);
342 	return 0;
343 }
344 
345 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
346 				bool check_only)
347 {
348 	struct curseg_info *curseg;
349 	struct page *page = NULL;
350 	block_t blkaddr;
351 	unsigned int loop_cnt = 0;
352 	unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg -
353 						valid_user_blocks(sbi);
354 	int err = 0;
355 
356 	/* get node pages in the current segment */
357 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
358 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
359 
360 	while (1) {
361 		struct fsync_inode_entry *entry;
362 
363 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
364 			return 0;
365 
366 		page = f2fs_get_tmp_page(sbi, blkaddr);
367 		if (IS_ERR(page)) {
368 			err = PTR_ERR(page);
369 			break;
370 		}
371 
372 		if (!is_recoverable_dnode(page)) {
373 			f2fs_put_page(page, 1);
374 			break;
375 		}
376 
377 		if (!is_fsync_dnode(page))
378 			goto next;
379 
380 		entry = get_fsync_inode(head, ino_of_node(page));
381 		if (!entry) {
382 			bool quota_inode = false;
383 
384 			if (!check_only &&
385 					IS_INODE(page) && is_dent_dnode(page)) {
386 				err = f2fs_recover_inode_page(sbi, page);
387 				if (err) {
388 					f2fs_put_page(page, 1);
389 					break;
390 				}
391 				quota_inode = true;
392 			}
393 
394 			/*
395 			 * CP | dnode(F) | inode(DF)
396 			 * For this case, we should not give up now.
397 			 */
398 			entry = add_fsync_inode(sbi, head, ino_of_node(page),
399 								quota_inode);
400 			if (IS_ERR(entry)) {
401 				err = PTR_ERR(entry);
402 				if (err == -ENOENT) {
403 					err = 0;
404 					goto next;
405 				}
406 				f2fs_put_page(page, 1);
407 				break;
408 			}
409 		}
410 		entry->blkaddr = blkaddr;
411 
412 		if (IS_INODE(page) && is_dent_dnode(page))
413 			entry->last_dentry = blkaddr;
414 next:
415 		/* sanity check in order to detect looped node chain */
416 		if (++loop_cnt >= free_blocks ||
417 			blkaddr == next_blkaddr_of_node(page)) {
418 			f2fs_notice(sbi, "%s: detect looped node chain, blkaddr:%u, next:%u",
419 				    __func__, blkaddr,
420 				    next_blkaddr_of_node(page));
421 			f2fs_put_page(page, 1);
422 			err = -EINVAL;
423 			break;
424 		}
425 
426 		/* check next segment */
427 		blkaddr = next_blkaddr_of_node(page);
428 		f2fs_put_page(page, 1);
429 
430 		f2fs_ra_meta_pages_cond(sbi, blkaddr);
431 	}
432 	return err;
433 }
434 
435 static void destroy_fsync_dnodes(struct list_head *head, int drop)
436 {
437 	struct fsync_inode_entry *entry, *tmp;
438 
439 	list_for_each_entry_safe(entry, tmp, head, list)
440 		del_fsync_inode(entry, drop);
441 }
442 
443 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
444 			block_t blkaddr, struct dnode_of_data *dn)
445 {
446 	struct seg_entry *sentry;
447 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
448 	unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
449 	struct f2fs_summary_block *sum_node;
450 	struct f2fs_summary sum;
451 	struct page *sum_page, *node_page;
452 	struct dnode_of_data tdn = *dn;
453 	nid_t ino, nid;
454 	struct inode *inode;
455 	unsigned int offset;
456 	block_t bidx;
457 	int i;
458 
459 	sentry = get_seg_entry(sbi, segno);
460 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
461 		return 0;
462 
463 	/* Get the previous summary */
464 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
465 		struct curseg_info *curseg = CURSEG_I(sbi, i);
466 
467 		if (curseg->segno == segno) {
468 			sum = curseg->sum_blk->entries[blkoff];
469 			goto got_it;
470 		}
471 	}
472 
473 	sum_page = f2fs_get_sum_page(sbi, segno);
474 	if (IS_ERR(sum_page))
475 		return PTR_ERR(sum_page);
476 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
477 	sum = sum_node->entries[blkoff];
478 	f2fs_put_page(sum_page, 1);
479 got_it:
480 	/* Use the locked dnode page and inode */
481 	nid = le32_to_cpu(sum.nid);
482 	if (dn->inode->i_ino == nid) {
483 		tdn.nid = nid;
484 		if (!dn->inode_page_locked)
485 			lock_page(dn->inode_page);
486 		tdn.node_page = dn->inode_page;
487 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
488 		goto truncate_out;
489 	} else if (dn->nid == nid) {
490 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
491 		goto truncate_out;
492 	}
493 
494 	/* Get the node page */
495 	node_page = f2fs_get_node_page(sbi, nid);
496 	if (IS_ERR(node_page))
497 		return PTR_ERR(node_page);
498 
499 	offset = ofs_of_node(node_page);
500 	ino = ino_of_node(node_page);
501 	f2fs_put_page(node_page, 1);
502 
503 	if (ino != dn->inode->i_ino) {
504 		int ret;
505 
506 		/* Deallocate previous index in the node page */
507 		inode = f2fs_iget_retry(sbi->sb, ino);
508 		if (IS_ERR(inode))
509 			return PTR_ERR(inode);
510 
511 		ret = dquot_initialize(inode);
512 		if (ret) {
513 			iput(inode);
514 			return ret;
515 		}
516 	} else {
517 		inode = dn->inode;
518 	}
519 
520 	bidx = f2fs_start_bidx_of_node(offset, inode) +
521 				le16_to_cpu(sum.ofs_in_node);
522 
523 	/*
524 	 * if inode page is locked, unlock temporarily, but its reference
525 	 * count keeps alive.
526 	 */
527 	if (ino == dn->inode->i_ino && dn->inode_page_locked)
528 		unlock_page(dn->inode_page);
529 
530 	set_new_dnode(&tdn, inode, NULL, NULL, 0);
531 	if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
532 		goto out;
533 
534 	if (tdn.data_blkaddr == blkaddr)
535 		f2fs_truncate_data_blocks_range(&tdn, 1);
536 
537 	f2fs_put_dnode(&tdn);
538 out:
539 	if (ino != dn->inode->i_ino)
540 		iput(inode);
541 	else if (dn->inode_page_locked)
542 		lock_page(dn->inode_page);
543 	return 0;
544 
545 truncate_out:
546 	if (f2fs_data_blkaddr(&tdn) == blkaddr)
547 		f2fs_truncate_data_blocks_range(&tdn, 1);
548 	if (dn->inode->i_ino == nid && !dn->inode_page_locked)
549 		unlock_page(dn->inode_page);
550 	return 0;
551 }
552 
553 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
554 					struct page *page)
555 {
556 	struct dnode_of_data dn;
557 	struct node_info ni;
558 	unsigned int start, end;
559 	int err = 0, recovered = 0;
560 
561 	/* step 1: recover xattr */
562 	if (IS_INODE(page)) {
563 		err = f2fs_recover_inline_xattr(inode, page);
564 		if (err)
565 			goto out;
566 	} else if (f2fs_has_xattr_block(ofs_of_node(page))) {
567 		err = f2fs_recover_xattr_data(inode, page);
568 		if (!err)
569 			recovered++;
570 		goto out;
571 	}
572 
573 	/* step 2: recover inline data */
574 	err = f2fs_recover_inline_data(inode, page);
575 	if (err) {
576 		if (err == 1)
577 			err = 0;
578 		goto out;
579 	}
580 
581 	/* step 3: recover data indices */
582 	start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
583 	end = start + ADDRS_PER_PAGE(page, inode);
584 
585 	set_new_dnode(&dn, inode, NULL, NULL, 0);
586 retry_dn:
587 	err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
588 	if (err) {
589 		if (err == -ENOMEM) {
590 			congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
591 			goto retry_dn;
592 		}
593 		goto out;
594 	}
595 
596 	f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
597 
598 	err = f2fs_get_node_info(sbi, dn.nid, &ni);
599 	if (err)
600 		goto err;
601 
602 	f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
603 
604 	if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
605 		f2fs_warn(sbi, "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
606 			  inode->i_ino, ofs_of_node(dn.node_page),
607 			  ofs_of_node(page));
608 		err = -EFSCORRUPTED;
609 		goto err;
610 	}
611 
612 	for (; start < end; start++, dn.ofs_in_node++) {
613 		block_t src, dest;
614 
615 		src = f2fs_data_blkaddr(&dn);
616 		dest = data_blkaddr(dn.inode, page, dn.ofs_in_node);
617 
618 		if (__is_valid_data_blkaddr(src) &&
619 			!f2fs_is_valid_blkaddr(sbi, src, META_POR)) {
620 			err = -EFSCORRUPTED;
621 			goto err;
622 		}
623 
624 		if (__is_valid_data_blkaddr(dest) &&
625 			!f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
626 			err = -EFSCORRUPTED;
627 			goto err;
628 		}
629 
630 		/* skip recovering if dest is the same as src */
631 		if (src == dest)
632 			continue;
633 
634 		/* dest is invalid, just invalidate src block */
635 		if (dest == NULL_ADDR) {
636 			f2fs_truncate_data_blocks_range(&dn, 1);
637 			continue;
638 		}
639 
640 		if (!file_keep_isize(inode) &&
641 			(i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
642 			f2fs_i_size_write(inode,
643 				(loff_t)(start + 1) << PAGE_SHIFT);
644 
645 		/*
646 		 * dest is reserved block, invalidate src block
647 		 * and then reserve one new block in dnode page.
648 		 */
649 		if (dest == NEW_ADDR) {
650 			f2fs_truncate_data_blocks_range(&dn, 1);
651 			f2fs_reserve_new_block(&dn);
652 			continue;
653 		}
654 
655 		/* dest is valid block, try to recover from src to dest */
656 		if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
657 
658 			if (src == NULL_ADDR) {
659 				err = f2fs_reserve_new_block(&dn);
660 				while (err &&
661 				       IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION))
662 					err = f2fs_reserve_new_block(&dn);
663 				/* We should not get -ENOSPC */
664 				f2fs_bug_on(sbi, err);
665 				if (err)
666 					goto err;
667 			}
668 retry_prev:
669 			/* Check the previous node page having this index */
670 			err = check_index_in_prev_nodes(sbi, dest, &dn);
671 			if (err) {
672 				if (err == -ENOMEM) {
673 					congestion_wait(BLK_RW_ASYNC,
674 							DEFAULT_IO_TIMEOUT);
675 					goto retry_prev;
676 				}
677 				goto err;
678 			}
679 
680 			/* write dummy data page */
681 			f2fs_replace_block(sbi, &dn, src, dest,
682 						ni.version, false, false);
683 			recovered++;
684 		}
685 	}
686 
687 	copy_node_footer(dn.node_page, page);
688 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
689 					ofs_of_node(page), false);
690 	set_page_dirty(dn.node_page);
691 err:
692 	f2fs_put_dnode(&dn);
693 out:
694 	f2fs_notice(sbi, "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
695 		    inode->i_ino, file_keep_isize(inode) ? "keep" : "recover",
696 		    recovered, err);
697 	return err;
698 }
699 
700 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
701 		struct list_head *tmp_inode_list, struct list_head *dir_list)
702 {
703 	struct curseg_info *curseg;
704 	struct page *page = NULL;
705 	int err = 0;
706 	block_t blkaddr;
707 
708 	/* get node pages in the current segment */
709 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
710 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
711 
712 	while (1) {
713 		struct fsync_inode_entry *entry;
714 
715 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
716 			break;
717 
718 		f2fs_ra_meta_pages_cond(sbi, blkaddr);
719 
720 		page = f2fs_get_tmp_page(sbi, blkaddr);
721 		if (IS_ERR(page)) {
722 			err = PTR_ERR(page);
723 			break;
724 		}
725 
726 		if (!is_recoverable_dnode(page)) {
727 			f2fs_put_page(page, 1);
728 			break;
729 		}
730 
731 		entry = get_fsync_inode(inode_list, ino_of_node(page));
732 		if (!entry)
733 			goto next;
734 		/*
735 		 * inode(x) | CP | inode(x) | dnode(F)
736 		 * In this case, we can lose the latest inode(x).
737 		 * So, call recover_inode for the inode update.
738 		 */
739 		if (IS_INODE(page)) {
740 			err = recover_inode(entry->inode, page);
741 			if (err) {
742 				f2fs_put_page(page, 1);
743 				break;
744 			}
745 		}
746 		if (entry->last_dentry == blkaddr) {
747 			err = recover_dentry(entry->inode, page, dir_list);
748 			if (err) {
749 				f2fs_put_page(page, 1);
750 				break;
751 			}
752 		}
753 		err = do_recover_data(sbi, entry->inode, page);
754 		if (err) {
755 			f2fs_put_page(page, 1);
756 			break;
757 		}
758 
759 		if (entry->blkaddr == blkaddr)
760 			list_move_tail(&entry->list, tmp_inode_list);
761 next:
762 		/* check next segment */
763 		blkaddr = next_blkaddr_of_node(page);
764 		f2fs_put_page(page, 1);
765 	}
766 	if (!err)
767 		f2fs_allocate_new_segments(sbi);
768 	return err;
769 }
770 
771 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
772 {
773 	struct list_head inode_list, tmp_inode_list;
774 	struct list_head dir_list;
775 	int err;
776 	int ret = 0;
777 	unsigned long s_flags = sbi->sb->s_flags;
778 	bool need_writecp = false;
779 	bool fix_curseg_write_pointer = false;
780 #ifdef CONFIG_QUOTA
781 	int quota_enabled;
782 #endif
783 
784 	if (s_flags & SB_RDONLY) {
785 		f2fs_info(sbi, "recover fsync data on readonly fs");
786 		sbi->sb->s_flags &= ~SB_RDONLY;
787 	}
788 
789 #ifdef CONFIG_QUOTA
790 	/* Needed for iput() to work correctly and not trash data */
791 	sbi->sb->s_flags |= SB_ACTIVE;
792 	/* Turn on quotas so that they are updated correctly */
793 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
794 #endif
795 
796 	INIT_LIST_HEAD(&inode_list);
797 	INIT_LIST_HEAD(&tmp_inode_list);
798 	INIT_LIST_HEAD(&dir_list);
799 
800 	/* prevent checkpoint */
801 	down_write(&sbi->cp_global_sem);
802 
803 	/* step #1: find fsynced inode numbers */
804 	err = find_fsync_dnodes(sbi, &inode_list, check_only);
805 	if (err || list_empty(&inode_list))
806 		goto skip;
807 
808 	if (check_only) {
809 		ret = 1;
810 		goto skip;
811 	}
812 
813 	need_writecp = true;
814 
815 	/* step #2: recover data */
816 	err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
817 	if (!err)
818 		f2fs_bug_on(sbi, !list_empty(&inode_list));
819 	else {
820 		/* restore s_flags to let iput() trash data */
821 		sbi->sb->s_flags = s_flags;
822 	}
823 skip:
824 	fix_curseg_write_pointer = !check_only || list_empty(&inode_list);
825 
826 	destroy_fsync_dnodes(&inode_list, err);
827 	destroy_fsync_dnodes(&tmp_inode_list, err);
828 
829 	/* truncate meta pages to be used by the recovery */
830 	truncate_inode_pages_range(META_MAPPING(sbi),
831 			(loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
832 
833 	if (err) {
834 		truncate_inode_pages_final(NODE_MAPPING(sbi));
835 		truncate_inode_pages_final(META_MAPPING(sbi));
836 	}
837 
838 	/*
839 	 * If fsync data succeeds or there is no fsync data to recover,
840 	 * and the f2fs is not read only, check and fix zoned block devices'
841 	 * write pointer consistency.
842 	 */
843 	if (!err && fix_curseg_write_pointer && !f2fs_readonly(sbi->sb) &&
844 			f2fs_sb_has_blkzoned(sbi)) {
845 		err = f2fs_fix_curseg_write_pointer(sbi);
846 		ret = err;
847 	}
848 
849 	if (!err)
850 		clear_sbi_flag(sbi, SBI_POR_DOING);
851 
852 	up_write(&sbi->cp_global_sem);
853 
854 	/* let's drop all the directory inodes for clean checkpoint */
855 	destroy_fsync_dnodes(&dir_list, err);
856 
857 	if (need_writecp) {
858 		set_sbi_flag(sbi, SBI_IS_RECOVERED);
859 
860 		if (!err) {
861 			struct cp_control cpc = {
862 				.reason = CP_RECOVERY,
863 			};
864 			err = f2fs_write_checkpoint(sbi, &cpc);
865 		}
866 	}
867 
868 #ifdef CONFIG_QUOTA
869 	/* Turn quotas off */
870 	if (quota_enabled)
871 		f2fs_quota_off_umount(sbi->sb);
872 #endif
873 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
874 
875 	return ret ? ret : err;
876 }
877 
878 int __init f2fs_create_recovery_cache(void)
879 {
880 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
881 					sizeof(struct fsync_inode_entry));
882 	if (!fsync_entry_slab)
883 		return -ENOMEM;
884 	return 0;
885 }
886 
887 void f2fs_destroy_recovery_cache(void)
888 {
889 	kmem_cache_destroy(fsync_entry_slab);
890 }
891