xref: /openbmc/linux/fs/f2fs/recovery.c (revision e481ff3f)
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, GFP_F2FS_ZERO);
95 	entry->inode = inode;
96 	list_add_tail(&entry->list, head);
97 
98 	return entry;
99 err_out:
100 	iput(inode);
101 	return ERR_PTR(err);
102 }
103 
104 static void del_fsync_inode(struct fsync_inode_entry *entry, int drop)
105 {
106 	if (drop) {
107 		/* inode should not be recovered, drop it */
108 		f2fs_inode_synced(entry->inode);
109 	}
110 	iput(entry->inode);
111 	list_del(&entry->list);
112 	kmem_cache_free(fsync_entry_slab, entry);
113 }
114 
115 static int init_recovered_filename(const struct inode *dir,
116 				   struct f2fs_inode *raw_inode,
117 				   struct f2fs_filename *fname,
118 				   struct qstr *usr_fname)
119 {
120 	int err;
121 
122 	memset(fname, 0, sizeof(*fname));
123 	fname->disk_name.len = le32_to_cpu(raw_inode->i_namelen);
124 	fname->disk_name.name = raw_inode->i_name;
125 
126 	if (WARN_ON(fname->disk_name.len > F2FS_NAME_LEN))
127 		return -ENAMETOOLONG;
128 
129 	if (!IS_ENCRYPTED(dir)) {
130 		usr_fname->name = fname->disk_name.name;
131 		usr_fname->len = fname->disk_name.len;
132 		fname->usr_fname = usr_fname;
133 	}
134 
135 	/* Compute the hash of the filename */
136 	if (IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)) {
137 		/*
138 		 * In this case the hash isn't computable without the key, so it
139 		 * was saved on-disk.
140 		 */
141 		if (fname->disk_name.len + sizeof(f2fs_hash_t) > F2FS_NAME_LEN)
142 			return -EINVAL;
143 		fname->hash = get_unaligned((f2fs_hash_t *)
144 				&raw_inode->i_name[fname->disk_name.len]);
145 	} else if (IS_CASEFOLDED(dir)) {
146 		err = f2fs_init_casefolded_name(dir, fname);
147 		if (err)
148 			return err;
149 		f2fs_hash_filename(dir, fname);
150 #ifdef CONFIG_UNICODE
151 		/* Case-sensitive match is fine for recovery */
152 		kmem_cache_free(f2fs_cf_name_slab, fname->cf_name.name);
153 		fname->cf_name.name = NULL;
154 #endif
155 	} else {
156 		f2fs_hash_filename(dir, fname);
157 	}
158 	return 0;
159 }
160 
161 static int recover_dentry(struct inode *inode, struct page *ipage,
162 						struct list_head *dir_list)
163 {
164 	struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
165 	nid_t pino = le32_to_cpu(raw_inode->i_pino);
166 	struct f2fs_dir_entry *de;
167 	struct f2fs_filename fname;
168 	struct qstr usr_fname;
169 	struct page *page;
170 	struct inode *dir, *einode;
171 	struct fsync_inode_entry *entry;
172 	int err = 0;
173 	char *name;
174 
175 	entry = get_fsync_inode(dir_list, pino);
176 	if (!entry) {
177 		entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
178 							pino, false);
179 		if (IS_ERR(entry)) {
180 			dir = ERR_CAST(entry);
181 			err = PTR_ERR(entry);
182 			goto out;
183 		}
184 	}
185 
186 	dir = entry->inode;
187 	err = init_recovered_filename(dir, raw_inode, &fname, &usr_fname);
188 	if (err)
189 		goto out;
190 retry:
191 	de = __f2fs_find_entry(dir, &fname, &page);
192 	if (de && inode->i_ino == le32_to_cpu(de->ino))
193 		goto out_put;
194 
195 	if (de) {
196 		einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
197 		if (IS_ERR(einode)) {
198 			WARN_ON(1);
199 			err = PTR_ERR(einode);
200 			if (err == -ENOENT)
201 				err = -EEXIST;
202 			goto out_put;
203 		}
204 
205 		err = dquot_initialize(einode);
206 		if (err) {
207 			iput(einode);
208 			goto out_put;
209 		}
210 
211 		err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
212 		if (err) {
213 			iput(einode);
214 			goto out_put;
215 		}
216 		f2fs_delete_entry(de, page, dir, einode);
217 		iput(einode);
218 		goto retry;
219 	} else if (IS_ERR(page)) {
220 		err = PTR_ERR(page);
221 	} else {
222 		err = f2fs_add_dentry(dir, &fname, inode,
223 					inode->i_ino, inode->i_mode);
224 	}
225 	if (err == -ENOMEM)
226 		goto retry;
227 	goto out;
228 
229 out_put:
230 	f2fs_put_page(page, 0);
231 out:
232 	if (file_enc_name(inode))
233 		name = "<encrypted>";
234 	else
235 		name = raw_inode->i_name;
236 	f2fs_notice(F2FS_I_SB(inode), "%s: ino = %x, name = %s, dir = %lx, err = %d",
237 		    __func__, ino_of_node(ipage), name,
238 		    IS_ERR(dir) ? 0 : dir->i_ino, err);
239 	return err;
240 }
241 
242 static int recover_quota_data(struct inode *inode, struct page *page)
243 {
244 	struct f2fs_inode *raw = F2FS_INODE(page);
245 	struct iattr attr;
246 	uid_t i_uid = le32_to_cpu(raw->i_uid);
247 	gid_t i_gid = le32_to_cpu(raw->i_gid);
248 	int err;
249 
250 	memset(&attr, 0, sizeof(attr));
251 
252 	attr.ia_uid = make_kuid(inode->i_sb->s_user_ns, i_uid);
253 	attr.ia_gid = make_kgid(inode->i_sb->s_user_ns, i_gid);
254 
255 	if (!uid_eq(attr.ia_uid, inode->i_uid))
256 		attr.ia_valid |= ATTR_UID;
257 	if (!gid_eq(attr.ia_gid, inode->i_gid))
258 		attr.ia_valid |= ATTR_GID;
259 
260 	if (!attr.ia_valid)
261 		return 0;
262 
263 	err = dquot_transfer(inode, &attr);
264 	if (err)
265 		set_sbi_flag(F2FS_I_SB(inode), SBI_QUOTA_NEED_REPAIR);
266 	return err;
267 }
268 
269 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
270 {
271 	if (ri->i_inline & F2FS_PIN_FILE)
272 		set_inode_flag(inode, FI_PIN_FILE);
273 	else
274 		clear_inode_flag(inode, FI_PIN_FILE);
275 	if (ri->i_inline & F2FS_DATA_EXIST)
276 		set_inode_flag(inode, FI_DATA_EXIST);
277 	else
278 		clear_inode_flag(inode, FI_DATA_EXIST);
279 }
280 
281 static int recover_inode(struct inode *inode, struct page *page)
282 {
283 	struct f2fs_inode *raw = F2FS_INODE(page);
284 	char *name;
285 	int err;
286 
287 	inode->i_mode = le16_to_cpu(raw->i_mode);
288 
289 	err = recover_quota_data(inode, page);
290 	if (err)
291 		return err;
292 
293 	i_uid_write(inode, le32_to_cpu(raw->i_uid));
294 	i_gid_write(inode, le32_to_cpu(raw->i_gid));
295 
296 	if (raw->i_inline & F2FS_EXTRA_ATTR) {
297 		if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)) &&
298 			F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize),
299 								i_projid)) {
300 			projid_t i_projid;
301 			kprojid_t kprojid;
302 
303 			i_projid = (projid_t)le32_to_cpu(raw->i_projid);
304 			kprojid = make_kprojid(&init_user_ns, i_projid);
305 
306 			if (!projid_eq(kprojid, F2FS_I(inode)->i_projid)) {
307 				err = f2fs_transfer_project_quota(inode,
308 								kprojid);
309 				if (err)
310 					return err;
311 				F2FS_I(inode)->i_projid = kprojid;
312 			}
313 		}
314 	}
315 
316 	f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
317 	inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
318 	inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
319 	inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
320 	inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
321 	inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
322 	inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
323 
324 	F2FS_I(inode)->i_advise = raw->i_advise;
325 	F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags);
326 	f2fs_set_inode_flags(inode);
327 	F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] =
328 				le16_to_cpu(raw->i_gc_failures);
329 
330 	recover_inline_flags(inode, raw);
331 
332 	f2fs_mark_inode_dirty_sync(inode, true);
333 
334 	if (file_enc_name(inode))
335 		name = "<encrypted>";
336 	else
337 		name = F2FS_INODE(page)->i_name;
338 
339 	f2fs_notice(F2FS_I_SB(inode), "recover_inode: ino = %x, name = %s, inline = %x",
340 		    ino_of_node(page), name, raw->i_inline);
341 	return 0;
342 }
343 
344 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
345 				bool check_only)
346 {
347 	struct curseg_info *curseg;
348 	struct page *page = NULL;
349 	block_t blkaddr;
350 	unsigned int loop_cnt = 0;
351 	unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg -
352 						valid_user_blocks(sbi);
353 	int err = 0;
354 
355 	/* get node pages in the current segment */
356 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
357 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
358 
359 	while (1) {
360 		struct fsync_inode_entry *entry;
361 
362 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
363 			return 0;
364 
365 		page = f2fs_get_tmp_page(sbi, blkaddr);
366 		if (IS_ERR(page)) {
367 			err = PTR_ERR(page);
368 			break;
369 		}
370 
371 		if (!is_recoverable_dnode(page)) {
372 			f2fs_put_page(page, 1);
373 			break;
374 		}
375 
376 		if (!is_fsync_dnode(page))
377 			goto next;
378 
379 		entry = get_fsync_inode(head, ino_of_node(page));
380 		if (!entry) {
381 			bool quota_inode = false;
382 
383 			if (!check_only &&
384 					IS_INODE(page) && is_dent_dnode(page)) {
385 				err = f2fs_recover_inode_page(sbi, page);
386 				if (err) {
387 					f2fs_put_page(page, 1);
388 					break;
389 				}
390 				quota_inode = true;
391 			}
392 
393 			/*
394 			 * CP | dnode(F) | inode(DF)
395 			 * For this case, we should not give up now.
396 			 */
397 			entry = add_fsync_inode(sbi, head, ino_of_node(page),
398 								quota_inode);
399 			if (IS_ERR(entry)) {
400 				err = PTR_ERR(entry);
401 				if (err == -ENOENT) {
402 					err = 0;
403 					goto next;
404 				}
405 				f2fs_put_page(page, 1);
406 				break;
407 			}
408 		}
409 		entry->blkaddr = blkaddr;
410 
411 		if (IS_INODE(page) && is_dent_dnode(page))
412 			entry->last_dentry = blkaddr;
413 next:
414 		/* sanity check in order to detect looped node chain */
415 		if (++loop_cnt >= free_blocks ||
416 			blkaddr == next_blkaddr_of_node(page)) {
417 			f2fs_notice(sbi, "%s: detect looped node chain, blkaddr:%u, next:%u",
418 				    __func__, blkaddr,
419 				    next_blkaddr_of_node(page));
420 			f2fs_put_page(page, 1);
421 			err = -EINVAL;
422 			break;
423 		}
424 
425 		/* check next segment */
426 		blkaddr = next_blkaddr_of_node(page);
427 		f2fs_put_page(page, 1);
428 
429 		f2fs_ra_meta_pages_cond(sbi, blkaddr);
430 	}
431 	return err;
432 }
433 
434 static void destroy_fsync_dnodes(struct list_head *head, int drop)
435 {
436 	struct fsync_inode_entry *entry, *tmp;
437 
438 	list_for_each_entry_safe(entry, tmp, head, list)
439 		del_fsync_inode(entry, drop);
440 }
441 
442 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
443 			block_t blkaddr, struct dnode_of_data *dn)
444 {
445 	struct seg_entry *sentry;
446 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
447 	unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
448 	struct f2fs_summary_block *sum_node;
449 	struct f2fs_summary sum;
450 	struct page *sum_page, *node_page;
451 	struct dnode_of_data tdn = *dn;
452 	nid_t ino, nid;
453 	struct inode *inode;
454 	unsigned int offset;
455 	block_t bidx;
456 	int i;
457 
458 	sentry = get_seg_entry(sbi, segno);
459 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
460 		return 0;
461 
462 	/* Get the previous summary */
463 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
464 		struct curseg_info *curseg = CURSEG_I(sbi, i);
465 
466 		if (curseg->segno == segno) {
467 			sum = curseg->sum_blk->entries[blkoff];
468 			goto got_it;
469 		}
470 	}
471 
472 	sum_page = f2fs_get_sum_page(sbi, segno);
473 	if (IS_ERR(sum_page))
474 		return PTR_ERR(sum_page);
475 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
476 	sum = sum_node->entries[blkoff];
477 	f2fs_put_page(sum_page, 1);
478 got_it:
479 	/* Use the locked dnode page and inode */
480 	nid = le32_to_cpu(sum.nid);
481 	if (dn->inode->i_ino == nid) {
482 		tdn.nid = nid;
483 		if (!dn->inode_page_locked)
484 			lock_page(dn->inode_page);
485 		tdn.node_page = dn->inode_page;
486 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
487 		goto truncate_out;
488 	} else if (dn->nid == nid) {
489 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
490 		goto truncate_out;
491 	}
492 
493 	/* Get the node page */
494 	node_page = f2fs_get_node_page(sbi, nid);
495 	if (IS_ERR(node_page))
496 		return PTR_ERR(node_page);
497 
498 	offset = ofs_of_node(node_page);
499 	ino = ino_of_node(node_page);
500 	f2fs_put_page(node_page, 1);
501 
502 	if (ino != dn->inode->i_ino) {
503 		int ret;
504 
505 		/* Deallocate previous index in the node page */
506 		inode = f2fs_iget_retry(sbi->sb, ino);
507 		if (IS_ERR(inode))
508 			return PTR_ERR(inode);
509 
510 		ret = dquot_initialize(inode);
511 		if (ret) {
512 			iput(inode);
513 			return ret;
514 		}
515 	} else {
516 		inode = dn->inode;
517 	}
518 
519 	bidx = f2fs_start_bidx_of_node(offset, inode) +
520 				le16_to_cpu(sum.ofs_in_node);
521 
522 	/*
523 	 * if inode page is locked, unlock temporarily, but its reference
524 	 * count keeps alive.
525 	 */
526 	if (ino == dn->inode->i_ino && dn->inode_page_locked)
527 		unlock_page(dn->inode_page);
528 
529 	set_new_dnode(&tdn, inode, NULL, NULL, 0);
530 	if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
531 		goto out;
532 
533 	if (tdn.data_blkaddr == blkaddr)
534 		f2fs_truncate_data_blocks_range(&tdn, 1);
535 
536 	f2fs_put_dnode(&tdn);
537 out:
538 	if (ino != dn->inode->i_ino)
539 		iput(inode);
540 	else if (dn->inode_page_locked)
541 		lock_page(dn->inode_page);
542 	return 0;
543 
544 truncate_out:
545 	if (f2fs_data_blkaddr(&tdn) == blkaddr)
546 		f2fs_truncate_data_blocks_range(&tdn, 1);
547 	if (dn->inode->i_ino == nid && !dn->inode_page_locked)
548 		unlock_page(dn->inode_page);
549 	return 0;
550 }
551 
552 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
553 					struct page *page)
554 {
555 	struct dnode_of_data dn;
556 	struct node_info ni;
557 	unsigned int start, end;
558 	int err = 0, recovered = 0;
559 
560 	/* step 1: recover xattr */
561 	if (IS_INODE(page)) {
562 		err = f2fs_recover_inline_xattr(inode, page);
563 		if (err)
564 			goto out;
565 	} else if (f2fs_has_xattr_block(ofs_of_node(page))) {
566 		err = f2fs_recover_xattr_data(inode, page);
567 		if (!err)
568 			recovered++;
569 		goto out;
570 	}
571 
572 	/* step 2: recover inline data */
573 	err = f2fs_recover_inline_data(inode, page);
574 	if (err) {
575 		if (err == 1)
576 			err = 0;
577 		goto out;
578 	}
579 
580 	/* step 3: recover data indices */
581 	start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
582 	end = start + ADDRS_PER_PAGE(page, inode);
583 
584 	set_new_dnode(&dn, inode, NULL, NULL, 0);
585 retry_dn:
586 	err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
587 	if (err) {
588 		if (err == -ENOMEM) {
589 			congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
590 			goto retry_dn;
591 		}
592 		goto out;
593 	}
594 
595 	f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
596 
597 	err = f2fs_get_node_info(sbi, dn.nid, &ni);
598 	if (err)
599 		goto err;
600 
601 	f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
602 
603 	if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
604 		f2fs_warn(sbi, "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
605 			  inode->i_ino, ofs_of_node(dn.node_page),
606 			  ofs_of_node(page));
607 		err = -EFSCORRUPTED;
608 		goto err;
609 	}
610 
611 	for (; start < end; start++, dn.ofs_in_node++) {
612 		block_t src, dest;
613 
614 		src = f2fs_data_blkaddr(&dn);
615 		dest = data_blkaddr(dn.inode, page, dn.ofs_in_node);
616 
617 		if (__is_valid_data_blkaddr(src) &&
618 			!f2fs_is_valid_blkaddr(sbi, src, META_POR)) {
619 			err = -EFSCORRUPTED;
620 			goto err;
621 		}
622 
623 		if (__is_valid_data_blkaddr(dest) &&
624 			!f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
625 			err = -EFSCORRUPTED;
626 			goto err;
627 		}
628 
629 		/* skip recovering if dest is the same as src */
630 		if (src == dest)
631 			continue;
632 
633 		/* dest is invalid, just invalidate src block */
634 		if (dest == NULL_ADDR) {
635 			f2fs_truncate_data_blocks_range(&dn, 1);
636 			continue;
637 		}
638 
639 		if (!file_keep_isize(inode) &&
640 			(i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
641 			f2fs_i_size_write(inode,
642 				(loff_t)(start + 1) << PAGE_SHIFT);
643 
644 		/*
645 		 * dest is reserved block, invalidate src block
646 		 * and then reserve one new block in dnode page.
647 		 */
648 		if (dest == NEW_ADDR) {
649 			f2fs_truncate_data_blocks_range(&dn, 1);
650 			f2fs_reserve_new_block(&dn);
651 			continue;
652 		}
653 
654 		/* dest is valid block, try to recover from src to dest */
655 		if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
656 
657 			if (src == NULL_ADDR) {
658 				err = f2fs_reserve_new_block(&dn);
659 				while (err &&
660 				       IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION))
661 					err = f2fs_reserve_new_block(&dn);
662 				/* We should not get -ENOSPC */
663 				f2fs_bug_on(sbi, err);
664 				if (err)
665 					goto err;
666 			}
667 retry_prev:
668 			/* Check the previous node page having this index */
669 			err = check_index_in_prev_nodes(sbi, dest, &dn);
670 			if (err) {
671 				if (err == -ENOMEM) {
672 					congestion_wait(BLK_RW_ASYNC,
673 							DEFAULT_IO_TIMEOUT);
674 					goto retry_prev;
675 				}
676 				goto err;
677 			}
678 
679 			/* write dummy data page */
680 			f2fs_replace_block(sbi, &dn, src, dest,
681 						ni.version, false, false);
682 			recovered++;
683 		}
684 	}
685 
686 	copy_node_footer(dn.node_page, page);
687 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
688 					ofs_of_node(page), false);
689 	set_page_dirty(dn.node_page);
690 err:
691 	f2fs_put_dnode(&dn);
692 out:
693 	f2fs_notice(sbi, "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
694 		    inode->i_ino, file_keep_isize(inode) ? "keep" : "recover",
695 		    recovered, err);
696 	return err;
697 }
698 
699 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
700 		struct list_head *tmp_inode_list, struct list_head *dir_list)
701 {
702 	struct curseg_info *curseg;
703 	struct page *page = NULL;
704 	int err = 0;
705 	block_t blkaddr;
706 
707 	/* get node pages in the current segment */
708 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
709 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
710 
711 	while (1) {
712 		struct fsync_inode_entry *entry;
713 
714 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
715 			break;
716 
717 		f2fs_ra_meta_pages_cond(sbi, blkaddr);
718 
719 		page = f2fs_get_tmp_page(sbi, blkaddr);
720 		if (IS_ERR(page)) {
721 			err = PTR_ERR(page);
722 			break;
723 		}
724 
725 		if (!is_recoverable_dnode(page)) {
726 			f2fs_put_page(page, 1);
727 			break;
728 		}
729 
730 		entry = get_fsync_inode(inode_list, ino_of_node(page));
731 		if (!entry)
732 			goto next;
733 		/*
734 		 * inode(x) | CP | inode(x) | dnode(F)
735 		 * In this case, we can lose the latest inode(x).
736 		 * So, call recover_inode for the inode update.
737 		 */
738 		if (IS_INODE(page)) {
739 			err = recover_inode(entry->inode, page);
740 			if (err) {
741 				f2fs_put_page(page, 1);
742 				break;
743 			}
744 		}
745 		if (entry->last_dentry == blkaddr) {
746 			err = recover_dentry(entry->inode, page, dir_list);
747 			if (err) {
748 				f2fs_put_page(page, 1);
749 				break;
750 			}
751 		}
752 		err = do_recover_data(sbi, entry->inode, page);
753 		if (err) {
754 			f2fs_put_page(page, 1);
755 			break;
756 		}
757 
758 		if (entry->blkaddr == blkaddr)
759 			list_move_tail(&entry->list, tmp_inode_list);
760 next:
761 		/* check next segment */
762 		blkaddr = next_blkaddr_of_node(page);
763 		f2fs_put_page(page, 1);
764 	}
765 	if (!err)
766 		f2fs_allocate_new_segments(sbi);
767 	return err;
768 }
769 
770 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
771 {
772 	struct list_head inode_list, tmp_inode_list;
773 	struct list_head dir_list;
774 	int err;
775 	int ret = 0;
776 	unsigned long s_flags = sbi->sb->s_flags;
777 	bool need_writecp = false;
778 	bool fix_curseg_write_pointer = false;
779 #ifdef CONFIG_QUOTA
780 	int quota_enabled;
781 #endif
782 
783 	if (s_flags & SB_RDONLY) {
784 		f2fs_info(sbi, "recover fsync data on readonly fs");
785 		sbi->sb->s_flags &= ~SB_RDONLY;
786 	}
787 
788 #ifdef CONFIG_QUOTA
789 	/* Needed for iput() to work correctly and not trash data */
790 	sbi->sb->s_flags |= SB_ACTIVE;
791 	/* Turn on quotas so that they are updated correctly */
792 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
793 #endif
794 
795 	INIT_LIST_HEAD(&inode_list);
796 	INIT_LIST_HEAD(&tmp_inode_list);
797 	INIT_LIST_HEAD(&dir_list);
798 
799 	/* prevent checkpoint */
800 	down_write(&sbi->cp_global_sem);
801 
802 	/* step #1: find fsynced inode numbers */
803 	err = find_fsync_dnodes(sbi, &inode_list, check_only);
804 	if (err || list_empty(&inode_list))
805 		goto skip;
806 
807 	if (check_only) {
808 		ret = 1;
809 		goto skip;
810 	}
811 
812 	need_writecp = true;
813 
814 	/* step #2: recover data */
815 	err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
816 	if (!err)
817 		f2fs_bug_on(sbi, !list_empty(&inode_list));
818 	else {
819 		/* restore s_flags to let iput() trash data */
820 		sbi->sb->s_flags = s_flags;
821 	}
822 skip:
823 	fix_curseg_write_pointer = !check_only || list_empty(&inode_list);
824 
825 	destroy_fsync_dnodes(&inode_list, err);
826 	destroy_fsync_dnodes(&tmp_inode_list, err);
827 
828 	/* truncate meta pages to be used by the recovery */
829 	truncate_inode_pages_range(META_MAPPING(sbi),
830 			(loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
831 
832 	if (err) {
833 		truncate_inode_pages_final(NODE_MAPPING(sbi));
834 		truncate_inode_pages_final(META_MAPPING(sbi));
835 	}
836 
837 	/*
838 	 * If fsync data succeeds or there is no fsync data to recover,
839 	 * and the f2fs is not read only, check and fix zoned block devices'
840 	 * write pointer consistency.
841 	 */
842 	if (!err && fix_curseg_write_pointer && !f2fs_readonly(sbi->sb) &&
843 			f2fs_sb_has_blkzoned(sbi)) {
844 		err = f2fs_fix_curseg_write_pointer(sbi);
845 		ret = err;
846 	}
847 
848 	if (!err)
849 		clear_sbi_flag(sbi, SBI_POR_DOING);
850 
851 	up_write(&sbi->cp_global_sem);
852 
853 	/* let's drop all the directory inodes for clean checkpoint */
854 	destroy_fsync_dnodes(&dir_list, err);
855 
856 	if (need_writecp) {
857 		set_sbi_flag(sbi, SBI_IS_RECOVERED);
858 
859 		if (!err) {
860 			struct cp_control cpc = {
861 				.reason = CP_RECOVERY,
862 			};
863 			err = f2fs_write_checkpoint(sbi, &cpc);
864 		}
865 	}
866 
867 #ifdef CONFIG_QUOTA
868 	/* Turn quotas off */
869 	if (quota_enabled)
870 		f2fs_quota_off_umount(sbi->sb);
871 #endif
872 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
873 
874 	return ret ? ret : err;
875 }
876 
877 int __init f2fs_create_recovery_cache(void)
878 {
879 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
880 					sizeof(struct fsync_inode_entry));
881 	if (!fsync_entry_slab)
882 		return -ENOMEM;
883 	return 0;
884 }
885 
886 void f2fs_destroy_recovery_cache(void)
887 {
888 	kmem_cache_destroy(fsync_entry_slab);
889 }
890