xref: /openbmc/linux/fs/f2fs/recovery.c (revision e15a5365)
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 init_recovered_filename(const struct inode *dir,
111 				   struct f2fs_inode *raw_inode,
112 				   struct f2fs_filename *fname,
113 				   struct qstr *usr_fname)
114 {
115 	int err;
116 
117 	memset(fname, 0, sizeof(*fname));
118 	fname->disk_name.len = le32_to_cpu(raw_inode->i_namelen);
119 	fname->disk_name.name = raw_inode->i_name;
120 
121 	if (WARN_ON(fname->disk_name.len > F2FS_NAME_LEN))
122 		return -ENAMETOOLONG;
123 
124 	if (!IS_ENCRYPTED(dir)) {
125 		usr_fname->name = fname->disk_name.name;
126 		usr_fname->len = fname->disk_name.len;
127 		fname->usr_fname = usr_fname;
128 	}
129 
130 	/* Compute the hash of the filename */
131 	if (IS_CASEFOLDED(dir)) {
132 		err = f2fs_init_casefolded_name(dir, fname);
133 		if (err)
134 			return err;
135 		f2fs_hash_filename(dir, fname);
136 #ifdef CONFIG_UNICODE
137 		/* Case-sensitive match is fine for recovery */
138 		kfree(fname->cf_name.name);
139 		fname->cf_name.name = NULL;
140 #endif
141 	} else {
142 		f2fs_hash_filename(dir, fname);
143 	}
144 	return 0;
145 }
146 
147 static int recover_dentry(struct inode *inode, struct page *ipage,
148 						struct list_head *dir_list)
149 {
150 	struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
151 	nid_t pino = le32_to_cpu(raw_inode->i_pino);
152 	struct f2fs_dir_entry *de;
153 	struct f2fs_filename fname;
154 	struct qstr usr_fname;
155 	struct page *page;
156 	struct inode *dir, *einode;
157 	struct fsync_inode_entry *entry;
158 	int err = 0;
159 	char *name;
160 
161 	entry = get_fsync_inode(dir_list, pino);
162 	if (!entry) {
163 		entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
164 							pino, false);
165 		if (IS_ERR(entry)) {
166 			dir = ERR_CAST(entry);
167 			err = PTR_ERR(entry);
168 			goto out;
169 		}
170 	}
171 
172 	dir = entry->inode;
173 	err = init_recovered_filename(dir, raw_inode, &fname, &usr_fname);
174 	if (err)
175 		goto out;
176 retry:
177 	de = __f2fs_find_entry(dir, &fname, &page);
178 	if (de && inode->i_ino == le32_to_cpu(de->ino))
179 		goto out_put;
180 
181 	if (de) {
182 		einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
183 		if (IS_ERR(einode)) {
184 			WARN_ON(1);
185 			err = PTR_ERR(einode);
186 			if (err == -ENOENT)
187 				err = -EEXIST;
188 			goto out_put;
189 		}
190 
191 		err = dquot_initialize(einode);
192 		if (err) {
193 			iput(einode);
194 			goto out_put;
195 		}
196 
197 		err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
198 		if (err) {
199 			iput(einode);
200 			goto out_put;
201 		}
202 		f2fs_delete_entry(de, page, dir, einode);
203 		iput(einode);
204 		goto retry;
205 	} else if (IS_ERR(page)) {
206 		err = PTR_ERR(page);
207 	} else {
208 		err = f2fs_add_dentry(dir, &fname, inode,
209 					inode->i_ino, inode->i_mode);
210 	}
211 	if (err == -ENOMEM)
212 		goto retry;
213 	goto out;
214 
215 out_put:
216 	f2fs_put_page(page, 0);
217 out:
218 	if (file_enc_name(inode))
219 		name = "<encrypted>";
220 	else
221 		name = raw_inode->i_name;
222 	f2fs_notice(F2FS_I_SB(inode), "%s: ino = %x, name = %s, dir = %lx, err = %d",
223 		    __func__, ino_of_node(ipage), name,
224 		    IS_ERR(dir) ? 0 : dir->i_ino, err);
225 	return err;
226 }
227 
228 static int recover_quota_data(struct inode *inode, struct page *page)
229 {
230 	struct f2fs_inode *raw = F2FS_INODE(page);
231 	struct iattr attr;
232 	uid_t i_uid = le32_to_cpu(raw->i_uid);
233 	gid_t i_gid = le32_to_cpu(raw->i_gid);
234 	int err;
235 
236 	memset(&attr, 0, sizeof(attr));
237 
238 	attr.ia_uid = make_kuid(inode->i_sb->s_user_ns, i_uid);
239 	attr.ia_gid = make_kgid(inode->i_sb->s_user_ns, i_gid);
240 
241 	if (!uid_eq(attr.ia_uid, inode->i_uid))
242 		attr.ia_valid |= ATTR_UID;
243 	if (!gid_eq(attr.ia_gid, inode->i_gid))
244 		attr.ia_valid |= ATTR_GID;
245 
246 	if (!attr.ia_valid)
247 		return 0;
248 
249 	err = dquot_transfer(inode, &attr);
250 	if (err)
251 		set_sbi_flag(F2FS_I_SB(inode), SBI_QUOTA_NEED_REPAIR);
252 	return err;
253 }
254 
255 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
256 {
257 	if (ri->i_inline & F2FS_PIN_FILE)
258 		set_inode_flag(inode, FI_PIN_FILE);
259 	else
260 		clear_inode_flag(inode, FI_PIN_FILE);
261 	if (ri->i_inline & F2FS_DATA_EXIST)
262 		set_inode_flag(inode, FI_DATA_EXIST);
263 	else
264 		clear_inode_flag(inode, FI_DATA_EXIST);
265 }
266 
267 static int recover_inode(struct inode *inode, struct page *page)
268 {
269 	struct f2fs_inode *raw = F2FS_INODE(page);
270 	char *name;
271 	int err;
272 
273 	inode->i_mode = le16_to_cpu(raw->i_mode);
274 
275 	err = recover_quota_data(inode, page);
276 	if (err)
277 		return err;
278 
279 	i_uid_write(inode, le32_to_cpu(raw->i_uid));
280 	i_gid_write(inode, le32_to_cpu(raw->i_gid));
281 
282 	if (raw->i_inline & F2FS_EXTRA_ATTR) {
283 		if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)) &&
284 			F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize),
285 								i_projid)) {
286 			projid_t i_projid;
287 			kprojid_t kprojid;
288 
289 			i_projid = (projid_t)le32_to_cpu(raw->i_projid);
290 			kprojid = make_kprojid(&init_user_ns, i_projid);
291 
292 			if (!projid_eq(kprojid, F2FS_I(inode)->i_projid)) {
293 				err = f2fs_transfer_project_quota(inode,
294 								kprojid);
295 				if (err)
296 					return err;
297 				F2FS_I(inode)->i_projid = kprojid;
298 			}
299 		}
300 	}
301 
302 	f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
303 	inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
304 	inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
305 	inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
306 	inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
307 	inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
308 	inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
309 
310 	F2FS_I(inode)->i_advise = raw->i_advise;
311 	F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags);
312 	f2fs_set_inode_flags(inode);
313 	F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] =
314 				le16_to_cpu(raw->i_gc_failures);
315 
316 	recover_inline_flags(inode, raw);
317 
318 	f2fs_mark_inode_dirty_sync(inode, true);
319 
320 	if (file_enc_name(inode))
321 		name = "<encrypted>";
322 	else
323 		name = F2FS_INODE(page)->i_name;
324 
325 	f2fs_notice(F2FS_I_SB(inode), "recover_inode: ino = %x, name = %s, inline = %x",
326 		    ino_of_node(page), name, raw->i_inline);
327 	return 0;
328 }
329 
330 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
331 				bool check_only)
332 {
333 	struct curseg_info *curseg;
334 	struct page *page = NULL;
335 	block_t blkaddr;
336 	unsigned int loop_cnt = 0;
337 	unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg -
338 						valid_user_blocks(sbi);
339 	int err = 0;
340 
341 	/* get node pages in the current segment */
342 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
343 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
344 
345 	while (1) {
346 		struct fsync_inode_entry *entry;
347 
348 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
349 			return 0;
350 
351 		page = f2fs_get_tmp_page(sbi, blkaddr);
352 		if (IS_ERR(page)) {
353 			err = PTR_ERR(page);
354 			break;
355 		}
356 
357 		if (!is_recoverable_dnode(page)) {
358 			f2fs_put_page(page, 1);
359 			break;
360 		}
361 
362 		if (!is_fsync_dnode(page))
363 			goto next;
364 
365 		entry = get_fsync_inode(head, ino_of_node(page));
366 		if (!entry) {
367 			bool quota_inode = false;
368 
369 			if (!check_only &&
370 					IS_INODE(page) && is_dent_dnode(page)) {
371 				err = f2fs_recover_inode_page(sbi, page);
372 				if (err) {
373 					f2fs_put_page(page, 1);
374 					break;
375 				}
376 				quota_inode = true;
377 			}
378 
379 			/*
380 			 * CP | dnode(F) | inode(DF)
381 			 * For this case, we should not give up now.
382 			 */
383 			entry = add_fsync_inode(sbi, head, ino_of_node(page),
384 								quota_inode);
385 			if (IS_ERR(entry)) {
386 				err = PTR_ERR(entry);
387 				if (err == -ENOENT) {
388 					err = 0;
389 					goto next;
390 				}
391 				f2fs_put_page(page, 1);
392 				break;
393 			}
394 		}
395 		entry->blkaddr = blkaddr;
396 
397 		if (IS_INODE(page) && is_dent_dnode(page))
398 			entry->last_dentry = blkaddr;
399 next:
400 		/* sanity check in order to detect looped node chain */
401 		if (++loop_cnt >= free_blocks ||
402 			blkaddr == next_blkaddr_of_node(page)) {
403 			f2fs_notice(sbi, "%s: detect looped node chain, blkaddr:%u, next:%u",
404 				    __func__, blkaddr,
405 				    next_blkaddr_of_node(page));
406 			f2fs_put_page(page, 1);
407 			err = -EINVAL;
408 			break;
409 		}
410 
411 		/* check next segment */
412 		blkaddr = next_blkaddr_of_node(page);
413 		f2fs_put_page(page, 1);
414 
415 		f2fs_ra_meta_pages_cond(sbi, blkaddr);
416 	}
417 	return err;
418 }
419 
420 static void destroy_fsync_dnodes(struct list_head *head, int drop)
421 {
422 	struct fsync_inode_entry *entry, *tmp;
423 
424 	list_for_each_entry_safe(entry, tmp, head, list)
425 		del_fsync_inode(entry, drop);
426 }
427 
428 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
429 			block_t blkaddr, struct dnode_of_data *dn)
430 {
431 	struct seg_entry *sentry;
432 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
433 	unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
434 	struct f2fs_summary_block *sum_node;
435 	struct f2fs_summary sum;
436 	struct page *sum_page, *node_page;
437 	struct dnode_of_data tdn = *dn;
438 	nid_t ino, nid;
439 	struct inode *inode;
440 	unsigned int offset;
441 	block_t bidx;
442 	int i;
443 
444 	sentry = get_seg_entry(sbi, segno);
445 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
446 		return 0;
447 
448 	/* Get the previous summary */
449 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
450 		struct curseg_info *curseg = CURSEG_I(sbi, i);
451 		if (curseg->segno == segno) {
452 			sum = curseg->sum_blk->entries[blkoff];
453 			goto got_it;
454 		}
455 	}
456 
457 	sum_page = f2fs_get_sum_page(sbi, segno);
458 	if (IS_ERR(sum_page))
459 		return PTR_ERR(sum_page);
460 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
461 	sum = sum_node->entries[blkoff];
462 	f2fs_put_page(sum_page, 1);
463 got_it:
464 	/* Use the locked dnode page and inode */
465 	nid = le32_to_cpu(sum.nid);
466 	if (dn->inode->i_ino == nid) {
467 		tdn.nid = nid;
468 		if (!dn->inode_page_locked)
469 			lock_page(dn->inode_page);
470 		tdn.node_page = dn->inode_page;
471 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
472 		goto truncate_out;
473 	} else if (dn->nid == nid) {
474 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
475 		goto truncate_out;
476 	}
477 
478 	/* Get the node page */
479 	node_page = f2fs_get_node_page(sbi, nid);
480 	if (IS_ERR(node_page))
481 		return PTR_ERR(node_page);
482 
483 	offset = ofs_of_node(node_page);
484 	ino = ino_of_node(node_page);
485 	f2fs_put_page(node_page, 1);
486 
487 	if (ino != dn->inode->i_ino) {
488 		int ret;
489 
490 		/* Deallocate previous index in the node page */
491 		inode = f2fs_iget_retry(sbi->sb, ino);
492 		if (IS_ERR(inode))
493 			return PTR_ERR(inode);
494 
495 		ret = dquot_initialize(inode);
496 		if (ret) {
497 			iput(inode);
498 			return ret;
499 		}
500 	} else {
501 		inode = dn->inode;
502 	}
503 
504 	bidx = f2fs_start_bidx_of_node(offset, inode) +
505 				le16_to_cpu(sum.ofs_in_node);
506 
507 	/*
508 	 * if inode page is locked, unlock temporarily, but its reference
509 	 * count keeps alive.
510 	 */
511 	if (ino == dn->inode->i_ino && dn->inode_page_locked)
512 		unlock_page(dn->inode_page);
513 
514 	set_new_dnode(&tdn, inode, NULL, NULL, 0);
515 	if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
516 		goto out;
517 
518 	if (tdn.data_blkaddr == blkaddr)
519 		f2fs_truncate_data_blocks_range(&tdn, 1);
520 
521 	f2fs_put_dnode(&tdn);
522 out:
523 	if (ino != dn->inode->i_ino)
524 		iput(inode);
525 	else if (dn->inode_page_locked)
526 		lock_page(dn->inode_page);
527 	return 0;
528 
529 truncate_out:
530 	if (f2fs_data_blkaddr(&tdn) == blkaddr)
531 		f2fs_truncate_data_blocks_range(&tdn, 1);
532 	if (dn->inode->i_ino == nid && !dn->inode_page_locked)
533 		unlock_page(dn->inode_page);
534 	return 0;
535 }
536 
537 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
538 					struct page *page)
539 {
540 	struct dnode_of_data dn;
541 	struct node_info ni;
542 	unsigned int start, end;
543 	int err = 0, recovered = 0;
544 
545 	/* step 1: recover xattr */
546 	if (IS_INODE(page)) {
547 		err = f2fs_recover_inline_xattr(inode, page);
548 		if (err)
549 			goto out;
550 	} else if (f2fs_has_xattr_block(ofs_of_node(page))) {
551 		err = f2fs_recover_xattr_data(inode, page);
552 		if (!err)
553 			recovered++;
554 		goto out;
555 	}
556 
557 	/* step 2: recover inline data */
558 	err = f2fs_recover_inline_data(inode, page);
559 	if (err) {
560 		if (err == 1)
561 			err = 0;
562 		goto out;
563 	}
564 
565 	/* step 3: recover data indices */
566 	start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
567 	end = start + ADDRS_PER_PAGE(page, inode);
568 
569 	set_new_dnode(&dn, inode, NULL, NULL, 0);
570 retry_dn:
571 	err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
572 	if (err) {
573 		if (err == -ENOMEM) {
574 			congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
575 			goto retry_dn;
576 		}
577 		goto out;
578 	}
579 
580 	f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
581 
582 	err = f2fs_get_node_info(sbi, dn.nid, &ni);
583 	if (err)
584 		goto err;
585 
586 	f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
587 
588 	if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
589 		f2fs_warn(sbi, "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
590 			  inode->i_ino, ofs_of_node(dn.node_page),
591 			  ofs_of_node(page));
592 		err = -EFSCORRUPTED;
593 		goto err;
594 	}
595 
596 	for (; start < end; start++, dn.ofs_in_node++) {
597 		block_t src, dest;
598 
599 		src = f2fs_data_blkaddr(&dn);
600 		dest = data_blkaddr(dn.inode, page, dn.ofs_in_node);
601 
602 		if (__is_valid_data_blkaddr(src) &&
603 			!f2fs_is_valid_blkaddr(sbi, src, META_POR)) {
604 			err = -EFSCORRUPTED;
605 			goto err;
606 		}
607 
608 		if (__is_valid_data_blkaddr(dest) &&
609 			!f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
610 			err = -EFSCORRUPTED;
611 			goto err;
612 		}
613 
614 		/* skip recovering if dest is the same as src */
615 		if (src == dest)
616 			continue;
617 
618 		/* dest is invalid, just invalidate src block */
619 		if (dest == NULL_ADDR) {
620 			f2fs_truncate_data_blocks_range(&dn, 1);
621 			continue;
622 		}
623 
624 		if (!file_keep_isize(inode) &&
625 			(i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
626 			f2fs_i_size_write(inode,
627 				(loff_t)(start + 1) << PAGE_SHIFT);
628 
629 		/*
630 		 * dest is reserved block, invalidate src block
631 		 * and then reserve one new block in dnode page.
632 		 */
633 		if (dest == NEW_ADDR) {
634 			f2fs_truncate_data_blocks_range(&dn, 1);
635 			f2fs_reserve_new_block(&dn);
636 			continue;
637 		}
638 
639 		/* dest is valid block, try to recover from src to dest */
640 		if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
641 
642 			if (src == NULL_ADDR) {
643 				err = f2fs_reserve_new_block(&dn);
644 				while (err &&
645 				       IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION))
646 					err = f2fs_reserve_new_block(&dn);
647 				/* We should not get -ENOSPC */
648 				f2fs_bug_on(sbi, err);
649 				if (err)
650 					goto err;
651 			}
652 retry_prev:
653 			/* Check the previous node page having this index */
654 			err = check_index_in_prev_nodes(sbi, dest, &dn);
655 			if (err) {
656 				if (err == -ENOMEM) {
657 					congestion_wait(BLK_RW_ASYNC,
658 							DEFAULT_IO_TIMEOUT);
659 					goto retry_prev;
660 				}
661 				goto err;
662 			}
663 
664 			/* write dummy data page */
665 			f2fs_replace_block(sbi, &dn, src, dest,
666 						ni.version, false, false);
667 			recovered++;
668 		}
669 	}
670 
671 	copy_node_footer(dn.node_page, page);
672 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
673 					ofs_of_node(page), false);
674 	set_page_dirty(dn.node_page);
675 err:
676 	f2fs_put_dnode(&dn);
677 out:
678 	f2fs_notice(sbi, "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
679 		    inode->i_ino, file_keep_isize(inode) ? "keep" : "recover",
680 		    recovered, err);
681 	return err;
682 }
683 
684 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
685 		struct list_head *tmp_inode_list, struct list_head *dir_list)
686 {
687 	struct curseg_info *curseg;
688 	struct page *page = NULL;
689 	int err = 0;
690 	block_t blkaddr;
691 
692 	/* get node pages in the current segment */
693 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
694 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
695 
696 	while (1) {
697 		struct fsync_inode_entry *entry;
698 
699 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
700 			break;
701 
702 		f2fs_ra_meta_pages_cond(sbi, blkaddr);
703 
704 		page = f2fs_get_tmp_page(sbi, blkaddr);
705 		if (IS_ERR(page)) {
706 			err = PTR_ERR(page);
707 			break;
708 		}
709 
710 		if (!is_recoverable_dnode(page)) {
711 			f2fs_put_page(page, 1);
712 			break;
713 		}
714 
715 		entry = get_fsync_inode(inode_list, ino_of_node(page));
716 		if (!entry)
717 			goto next;
718 		/*
719 		 * inode(x) | CP | inode(x) | dnode(F)
720 		 * In this case, we can lose the latest inode(x).
721 		 * So, call recover_inode for the inode update.
722 		 */
723 		if (IS_INODE(page)) {
724 			err = recover_inode(entry->inode, page);
725 			if (err) {
726 				f2fs_put_page(page, 1);
727 				break;
728 			}
729 		}
730 		if (entry->last_dentry == blkaddr) {
731 			err = recover_dentry(entry->inode, page, dir_list);
732 			if (err) {
733 				f2fs_put_page(page, 1);
734 				break;
735 			}
736 		}
737 		err = do_recover_data(sbi, entry->inode, page);
738 		if (err) {
739 			f2fs_put_page(page, 1);
740 			break;
741 		}
742 
743 		if (entry->blkaddr == blkaddr)
744 			list_move_tail(&entry->list, tmp_inode_list);
745 next:
746 		/* check next segment */
747 		blkaddr = next_blkaddr_of_node(page);
748 		f2fs_put_page(page, 1);
749 	}
750 	if (!err)
751 		f2fs_allocate_new_segments(sbi);
752 	return err;
753 }
754 
755 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
756 {
757 	struct list_head inode_list, tmp_inode_list;
758 	struct list_head dir_list;
759 	int err;
760 	int ret = 0;
761 	unsigned long s_flags = sbi->sb->s_flags;
762 	bool need_writecp = false;
763 	bool fix_curseg_write_pointer = false;
764 #ifdef CONFIG_QUOTA
765 	int quota_enabled;
766 #endif
767 
768 	if (s_flags & SB_RDONLY) {
769 		f2fs_info(sbi, "recover fsync data on readonly fs");
770 		sbi->sb->s_flags &= ~SB_RDONLY;
771 	}
772 
773 #ifdef CONFIG_QUOTA
774 	/* Needed for iput() to work correctly and not trash data */
775 	sbi->sb->s_flags |= SB_ACTIVE;
776 	/* Turn on quotas so that they are updated correctly */
777 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
778 #endif
779 
780 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
781 			sizeof(struct fsync_inode_entry));
782 	if (!fsync_entry_slab) {
783 		err = -ENOMEM;
784 		goto out;
785 	}
786 
787 	INIT_LIST_HEAD(&inode_list);
788 	INIT_LIST_HEAD(&tmp_inode_list);
789 	INIT_LIST_HEAD(&dir_list);
790 
791 	/* prevent checkpoint */
792 	mutex_lock(&sbi->cp_mutex);
793 
794 	/* step #1: find fsynced inode numbers */
795 	err = find_fsync_dnodes(sbi, &inode_list, check_only);
796 	if (err || list_empty(&inode_list))
797 		goto skip;
798 
799 	if (check_only) {
800 		ret = 1;
801 		goto skip;
802 	}
803 
804 	need_writecp = true;
805 
806 	/* step #2: recover data */
807 	err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
808 	if (!err)
809 		f2fs_bug_on(sbi, !list_empty(&inode_list));
810 	else {
811 		/* restore s_flags to let iput() trash data */
812 		sbi->sb->s_flags = s_flags;
813 	}
814 skip:
815 	fix_curseg_write_pointer = !check_only || list_empty(&inode_list);
816 
817 	destroy_fsync_dnodes(&inode_list, err);
818 	destroy_fsync_dnodes(&tmp_inode_list, err);
819 
820 	/* truncate meta pages to be used by the recovery */
821 	truncate_inode_pages_range(META_MAPPING(sbi),
822 			(loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
823 
824 	if (err) {
825 		truncate_inode_pages_final(NODE_MAPPING(sbi));
826 		truncate_inode_pages_final(META_MAPPING(sbi));
827 	}
828 
829 	/*
830 	 * If fsync data succeeds or there is no fsync data to recover,
831 	 * and the f2fs is not read only, check and fix zoned block devices'
832 	 * write pointer consistency.
833 	 */
834 	if (!err && fix_curseg_write_pointer && !f2fs_readonly(sbi->sb) &&
835 			f2fs_sb_has_blkzoned(sbi)) {
836 		err = f2fs_fix_curseg_write_pointer(sbi);
837 		ret = err;
838 	}
839 
840 	if (!err)
841 		clear_sbi_flag(sbi, SBI_POR_DOING);
842 
843 	mutex_unlock(&sbi->cp_mutex);
844 
845 	/* let's drop all the directory inodes for clean checkpoint */
846 	destroy_fsync_dnodes(&dir_list, err);
847 
848 	if (need_writecp) {
849 		set_sbi_flag(sbi, SBI_IS_RECOVERED);
850 
851 		if (!err) {
852 			struct cp_control cpc = {
853 				.reason = CP_RECOVERY,
854 			};
855 			err = f2fs_write_checkpoint(sbi, &cpc);
856 		}
857 	}
858 
859 	kmem_cache_destroy(fsync_entry_slab);
860 out:
861 #ifdef CONFIG_QUOTA
862 	/* Turn quotas off */
863 	if (quota_enabled)
864 		f2fs_quota_off_umount(sbi->sb);
865 #endif
866 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
867 
868 	return ret ? ret: err;
869 }
870