1 /*
2  * fs/f2fs/recovery.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include "f2fs.h"
14 #include "node.h"
15 #include "segment.h"
16 
17 /*
18  * Roll forward recovery scenarios.
19  *
20  * [Term] F: fsync_mark, D: dentry_mark
21  *
22  * 1. inode(x) | CP | inode(x) | dnode(F)
23  * -> Update the latest inode(x).
24  *
25  * 2. inode(x) | CP | inode(F) | dnode(F)
26  * -> No problem.
27  *
28  * 3. inode(x) | CP | dnode(F) | inode(x)
29  * -> Recover to the latest dnode(F), and drop the last inode(x)
30  *
31  * 4. inode(x) | CP | dnode(F) | inode(F)
32  * -> No problem.
33  *
34  * 5. CP | inode(x) | dnode(F)
35  * -> The inode(DF) was missing. Should drop this dnode(F).
36  *
37  * 6. CP | inode(DF) | dnode(F)
38  * -> No problem.
39  *
40  * 7. CP | dnode(F) | inode(DF)
41  * -> If f2fs_iget fails, then goto next to find inode(DF).
42  *
43  * 8. CP | dnode(F) | inode(x)
44  * -> If f2fs_iget fails, then goto next to find inode(DF).
45  *    But it will fail due to no inode(DF).
46  */
47 
48 static struct kmem_cache *fsync_entry_slab;
49 
f2fs_space_for_roll_forward(struct f2fs_sb_info * sbi)50 bool f2fs_space_for_roll_forward(struct f2fs_sb_info *sbi)
51 {
52 	s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count);
53 
54 	if (sbi->last_valid_block_count + nalloc > sbi->user_block_count)
55 		return false;
56 	return true;
57 }
58 
get_fsync_inode(struct list_head * head,nid_t ino)59 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
60 								nid_t ino)
61 {
62 	struct fsync_inode_entry *entry;
63 
64 	list_for_each_entry(entry, head, list)
65 		if (entry->inode->i_ino == ino)
66 			return entry;
67 
68 	return NULL;
69 }
70 
add_fsync_inode(struct f2fs_sb_info * sbi,struct list_head * head,nid_t ino,bool quota_inode)71 static struct fsync_inode_entry *add_fsync_inode(struct f2fs_sb_info *sbi,
72 			struct list_head *head, nid_t ino, bool quota_inode)
73 {
74 	struct inode *inode;
75 	struct fsync_inode_entry *entry;
76 	int err;
77 
78 	inode = f2fs_iget_retry(sbi->sb, ino);
79 	if (IS_ERR(inode))
80 		return ERR_CAST(inode);
81 
82 	err = dquot_initialize(inode);
83 	if (err)
84 		goto err_out;
85 
86 	if (quota_inode) {
87 		err = dquot_alloc_inode(inode);
88 		if (err)
89 			goto err_out;
90 	}
91 
92 	entry = f2fs_kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
93 	entry->inode = inode;
94 	list_add_tail(&entry->list, head);
95 
96 	return entry;
97 err_out:
98 	iput(inode);
99 	return ERR_PTR(err);
100 }
101 
del_fsync_inode(struct fsync_inode_entry * entry,int drop)102 static void del_fsync_inode(struct fsync_inode_entry *entry, int drop)
103 {
104 	if (drop) {
105 		/* inode should not be recovered, drop it */
106 		f2fs_inode_synced(entry->inode);
107 	}
108 	iput(entry->inode);
109 	list_del(&entry->list);
110 	kmem_cache_free(fsync_entry_slab, entry);
111 }
112 
recover_dentry(struct inode * inode,struct page * ipage,struct list_head * dir_list)113 static int recover_dentry(struct inode *inode, struct page *ipage,
114 						struct list_head *dir_list)
115 {
116 	struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
117 	nid_t pino = le32_to_cpu(raw_inode->i_pino);
118 	struct f2fs_dir_entry *de;
119 	struct fscrypt_name fname;
120 	struct page *page;
121 	struct inode *dir, *einode;
122 	struct fsync_inode_entry *entry;
123 	int err = 0;
124 	char *name;
125 
126 	entry = get_fsync_inode(dir_list, pino);
127 	if (!entry) {
128 		entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
129 							pino, false);
130 		if (IS_ERR(entry)) {
131 			dir = ERR_CAST(entry);
132 			err = PTR_ERR(entry);
133 			goto out;
134 		}
135 	}
136 
137 	dir = entry->inode;
138 
139 	memset(&fname, 0, sizeof(struct fscrypt_name));
140 	fname.disk_name.len = le32_to_cpu(raw_inode->i_namelen);
141 	fname.disk_name.name = raw_inode->i_name;
142 
143 	if (unlikely(fname.disk_name.len > F2FS_NAME_LEN)) {
144 		WARN_ON(1);
145 		err = -ENAMETOOLONG;
146 		goto out;
147 	}
148 retry:
149 	de = __f2fs_find_entry(dir, &fname, &page);
150 	if (de && inode->i_ino == le32_to_cpu(de->ino))
151 		goto out_put;
152 
153 	if (de) {
154 		einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
155 		if (IS_ERR(einode)) {
156 			WARN_ON(1);
157 			err = PTR_ERR(einode);
158 			if (err == -ENOENT)
159 				err = -EEXIST;
160 			goto out_put;
161 		}
162 
163 		err = dquot_initialize(einode);
164 		if (err) {
165 			iput(einode);
166 			goto out_put;
167 		}
168 
169 		err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
170 		if (err) {
171 			iput(einode);
172 			goto out_put;
173 		}
174 		f2fs_delete_entry(de, page, dir, einode);
175 		iput(einode);
176 		goto retry;
177 	} else if (IS_ERR(page)) {
178 		err = PTR_ERR(page);
179 	} else {
180 		err = f2fs_add_dentry(dir, &fname, inode,
181 					inode->i_ino, inode->i_mode);
182 	}
183 	if (err == -ENOMEM)
184 		goto retry;
185 	goto out;
186 
187 out_put:
188 	f2fs_put_page(page, 0);
189 out:
190 	if (file_enc_name(inode))
191 		name = "<encrypted>";
192 	else
193 		name = raw_inode->i_name;
194 	f2fs_msg(inode->i_sb, KERN_NOTICE,
195 			"%s: ino = %x, name = %s, dir = %lx, err = %d",
196 			__func__, ino_of_node(ipage), name,
197 			IS_ERR(dir) ? 0 : dir->i_ino, err);
198 	return err;
199 }
200 
recover_inline_flags(struct inode * inode,struct f2fs_inode * ri)201 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
202 {
203 	if (ri->i_inline & F2FS_PIN_FILE)
204 		set_inode_flag(inode, FI_PIN_FILE);
205 	else
206 		clear_inode_flag(inode, FI_PIN_FILE);
207 	if (ri->i_inline & F2FS_DATA_EXIST)
208 		set_inode_flag(inode, FI_DATA_EXIST);
209 	else
210 		clear_inode_flag(inode, FI_DATA_EXIST);
211 }
212 
recover_inode(struct inode * inode,struct page * page)213 static void recover_inode(struct inode *inode, struct page *page)
214 {
215 	struct f2fs_inode *raw = F2FS_INODE(page);
216 	char *name;
217 
218 	inode->i_mode = le16_to_cpu(raw->i_mode);
219 	i_uid_write(inode, le32_to_cpu(raw->i_uid));
220 	i_gid_write(inode, le32_to_cpu(raw->i_gid));
221 
222 	if (raw->i_inline & F2FS_EXTRA_ATTR) {
223 		if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)->sb) &&
224 			F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize),
225 								i_projid)) {
226 			projid_t i_projid;
227 
228 			i_projid = (projid_t)le32_to_cpu(raw->i_projid);
229 			F2FS_I(inode)->i_projid =
230 				make_kprojid(&init_user_ns, i_projid);
231 		}
232 	}
233 
234 	f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
235 	inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
236 	inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
237 	inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
238 	inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
239 	inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
240 	inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
241 
242 	F2FS_I(inode)->i_advise = raw->i_advise;
243 	F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags);
244 	f2fs_set_inode_flags(inode);
245 	F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] =
246 				le16_to_cpu(raw->i_gc_failures);
247 
248 	recover_inline_flags(inode, raw);
249 
250 	f2fs_mark_inode_dirty_sync(inode, true);
251 
252 	if (file_enc_name(inode))
253 		name = "<encrypted>";
254 	else
255 		name = F2FS_INODE(page)->i_name;
256 
257 	f2fs_msg(inode->i_sb, KERN_NOTICE,
258 		"recover_inode: ino = %x, name = %s, inline = %x",
259 			ino_of_node(page), name, raw->i_inline);
260 }
261 
find_fsync_dnodes(struct f2fs_sb_info * sbi,struct list_head * head,bool check_only)262 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
263 				bool check_only)
264 {
265 	struct curseg_info *curseg;
266 	struct page *page = NULL;
267 	block_t blkaddr;
268 	unsigned int loop_cnt = 0;
269 	unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg -
270 						valid_user_blocks(sbi);
271 	int err = 0;
272 
273 	/* get node pages in the current segment */
274 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
275 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
276 
277 	while (1) {
278 		struct fsync_inode_entry *entry;
279 
280 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
281 			return 0;
282 
283 		page = f2fs_get_tmp_page(sbi, blkaddr);
284 		if (IS_ERR(page)) {
285 			err = PTR_ERR(page);
286 			break;
287 		}
288 
289 		if (!is_recoverable_dnode(page))
290 			break;
291 
292 		if (!is_fsync_dnode(page))
293 			goto next;
294 
295 		entry = get_fsync_inode(head, ino_of_node(page));
296 		if (!entry) {
297 			bool quota_inode = false;
298 
299 			if (!check_only &&
300 					IS_INODE(page) && is_dent_dnode(page)) {
301 				err = f2fs_recover_inode_page(sbi, page);
302 				if (err)
303 					break;
304 				quota_inode = true;
305 			}
306 
307 			/*
308 			 * CP | dnode(F) | inode(DF)
309 			 * For this case, we should not give up now.
310 			 */
311 			entry = add_fsync_inode(sbi, head, ino_of_node(page),
312 								quota_inode);
313 			if (IS_ERR(entry)) {
314 				err = PTR_ERR(entry);
315 				if (err == -ENOENT) {
316 					err = 0;
317 					goto next;
318 				}
319 				break;
320 			}
321 		}
322 		entry->blkaddr = blkaddr;
323 
324 		if (IS_INODE(page) && is_dent_dnode(page))
325 			entry->last_dentry = blkaddr;
326 next:
327 		/* sanity check in order to detect looped node chain */
328 		if (++loop_cnt >= free_blocks ||
329 			blkaddr == next_blkaddr_of_node(page)) {
330 			f2fs_msg(sbi->sb, KERN_NOTICE,
331 				"%s: detect looped node chain, "
332 				"blkaddr:%u, next:%u",
333 				__func__, blkaddr, next_blkaddr_of_node(page));
334 			err = -EINVAL;
335 			break;
336 		}
337 
338 		/* check next segment */
339 		blkaddr = next_blkaddr_of_node(page);
340 		f2fs_put_page(page, 1);
341 
342 		f2fs_ra_meta_pages_cond(sbi, blkaddr);
343 	}
344 	f2fs_put_page(page, 1);
345 	return err;
346 }
347 
destroy_fsync_dnodes(struct list_head * head,int drop)348 static void destroy_fsync_dnodes(struct list_head *head, int drop)
349 {
350 	struct fsync_inode_entry *entry, *tmp;
351 
352 	list_for_each_entry_safe(entry, tmp, head, list)
353 		del_fsync_inode(entry, drop);
354 }
355 
check_index_in_prev_nodes(struct f2fs_sb_info * sbi,block_t blkaddr,struct dnode_of_data * dn)356 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
357 			block_t blkaddr, struct dnode_of_data *dn)
358 {
359 	struct seg_entry *sentry;
360 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
361 	unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
362 	struct f2fs_summary_block *sum_node;
363 	struct f2fs_summary sum;
364 	struct page *sum_page, *node_page;
365 	struct dnode_of_data tdn = *dn;
366 	nid_t ino, nid;
367 	struct inode *inode;
368 	unsigned int offset;
369 	block_t bidx;
370 	int i;
371 
372 	sentry = get_seg_entry(sbi, segno);
373 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
374 		return 0;
375 
376 	/* Get the previous summary */
377 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
378 		struct curseg_info *curseg = CURSEG_I(sbi, i);
379 		if (curseg->segno == segno) {
380 			sum = curseg->sum_blk->entries[blkoff];
381 			goto got_it;
382 		}
383 	}
384 
385 	sum_page = f2fs_get_sum_page(sbi, segno);
386 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
387 	sum = sum_node->entries[blkoff];
388 	f2fs_put_page(sum_page, 1);
389 got_it:
390 	/* Use the locked dnode page and inode */
391 	nid = le32_to_cpu(sum.nid);
392 	if (dn->inode->i_ino == nid) {
393 		tdn.nid = nid;
394 		if (!dn->inode_page_locked)
395 			lock_page(dn->inode_page);
396 		tdn.node_page = dn->inode_page;
397 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
398 		goto truncate_out;
399 	} else if (dn->nid == nid) {
400 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
401 		goto truncate_out;
402 	}
403 
404 	/* Get the node page */
405 	node_page = f2fs_get_node_page(sbi, nid);
406 	if (IS_ERR(node_page))
407 		return PTR_ERR(node_page);
408 
409 	offset = ofs_of_node(node_page);
410 	ino = ino_of_node(node_page);
411 	f2fs_put_page(node_page, 1);
412 
413 	if (ino != dn->inode->i_ino) {
414 		int ret;
415 
416 		/* Deallocate previous index in the node page */
417 		inode = f2fs_iget_retry(sbi->sb, ino);
418 		if (IS_ERR(inode))
419 			return PTR_ERR(inode);
420 
421 		ret = dquot_initialize(inode);
422 		if (ret) {
423 			iput(inode);
424 			return ret;
425 		}
426 	} else {
427 		inode = dn->inode;
428 	}
429 
430 	bidx = f2fs_start_bidx_of_node(offset, inode) +
431 				le16_to_cpu(sum.ofs_in_node);
432 
433 	/*
434 	 * if inode page is locked, unlock temporarily, but its reference
435 	 * count keeps alive.
436 	 */
437 	if (ino == dn->inode->i_ino && dn->inode_page_locked)
438 		unlock_page(dn->inode_page);
439 
440 	set_new_dnode(&tdn, inode, NULL, NULL, 0);
441 	if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
442 		goto out;
443 
444 	if (tdn.data_blkaddr == blkaddr)
445 		f2fs_truncate_data_blocks_range(&tdn, 1);
446 
447 	f2fs_put_dnode(&tdn);
448 out:
449 	if (ino != dn->inode->i_ino)
450 		iput(inode);
451 	else if (dn->inode_page_locked)
452 		lock_page(dn->inode_page);
453 	return 0;
454 
455 truncate_out:
456 	if (datablock_addr(tdn.inode, tdn.node_page,
457 					tdn.ofs_in_node) == blkaddr)
458 		f2fs_truncate_data_blocks_range(&tdn, 1);
459 	if (dn->inode->i_ino == nid && !dn->inode_page_locked)
460 		unlock_page(dn->inode_page);
461 	return 0;
462 }
463 
do_recover_data(struct f2fs_sb_info * sbi,struct inode * inode,struct page * page)464 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
465 					struct page *page)
466 {
467 	struct dnode_of_data dn;
468 	struct node_info ni;
469 	unsigned int start, end;
470 	int err = 0, recovered = 0;
471 
472 	/* step 1: recover xattr */
473 	if (IS_INODE(page)) {
474 		err = f2fs_recover_inline_xattr(inode, page);
475 		if (err)
476 			goto out;
477 	} else if (f2fs_has_xattr_block(ofs_of_node(page))) {
478 		err = f2fs_recover_xattr_data(inode, page);
479 		if (!err)
480 			recovered++;
481 		goto out;
482 	}
483 
484 	/* step 2: recover inline data */
485 	err = f2fs_recover_inline_data(inode, page);
486 	if (err) {
487 		if (err == 1)
488 			err = 0;
489 		goto out;
490 	}
491 
492 	/* step 3: recover data indices */
493 	start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
494 	end = start + ADDRS_PER_PAGE(page, inode);
495 
496 	set_new_dnode(&dn, inode, NULL, NULL, 0);
497 retry_dn:
498 	err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
499 	if (err) {
500 		if (err == -ENOMEM) {
501 			congestion_wait(BLK_RW_ASYNC, HZ/50);
502 			goto retry_dn;
503 		}
504 		goto out;
505 	}
506 
507 	f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
508 
509 	err = f2fs_get_node_info(sbi, dn.nid, &ni);
510 	if (err)
511 		goto err;
512 
513 	f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
514 
515 	if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
516 		f2fs_msg(sbi->sb, KERN_WARNING,
517 			"Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
518 			inode->i_ino, ofs_of_node(dn.node_page),
519 			ofs_of_node(page));
520 		err = -EFSCORRUPTED;
521 		goto err;
522 	}
523 
524 	for (; start < end; start++, dn.ofs_in_node++) {
525 		block_t src, dest;
526 
527 		src = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
528 		dest = datablock_addr(dn.inode, page, dn.ofs_in_node);
529 
530 		/* skip recovering if dest is the same as src */
531 		if (src == dest)
532 			continue;
533 
534 		/* dest is invalid, just invalidate src block */
535 		if (dest == NULL_ADDR) {
536 			f2fs_truncate_data_blocks_range(&dn, 1);
537 			continue;
538 		}
539 
540 		if (!file_keep_isize(inode) &&
541 			(i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
542 			f2fs_i_size_write(inode,
543 				(loff_t)(start + 1) << PAGE_SHIFT);
544 
545 		/*
546 		 * dest is reserved block, invalidate src block
547 		 * and then reserve one new block in dnode page.
548 		 */
549 		if (dest == NEW_ADDR) {
550 			f2fs_truncate_data_blocks_range(&dn, 1);
551 			f2fs_reserve_new_block(&dn);
552 			continue;
553 		}
554 
555 		/* dest is valid block, try to recover from src to dest */
556 		if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
557 
558 			if (src == NULL_ADDR) {
559 				err = f2fs_reserve_new_block(&dn);
560 				while (err &&
561 				       IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION))
562 					err = f2fs_reserve_new_block(&dn);
563 				/* We should not get -ENOSPC */
564 				f2fs_bug_on(sbi, err);
565 				if (err)
566 					goto err;
567 			}
568 retry_prev:
569 			/* Check the previous node page having this index */
570 			err = check_index_in_prev_nodes(sbi, dest, &dn);
571 			if (err) {
572 				if (err == -ENOMEM) {
573 					congestion_wait(BLK_RW_ASYNC, HZ/50);
574 					goto retry_prev;
575 				}
576 				goto err;
577 			}
578 
579 			/* write dummy data page */
580 			f2fs_replace_block(sbi, &dn, src, dest,
581 						ni.version, false, false);
582 			recovered++;
583 		}
584 	}
585 
586 	copy_node_footer(dn.node_page, page);
587 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
588 					ofs_of_node(page), false);
589 	set_page_dirty(dn.node_page);
590 err:
591 	f2fs_put_dnode(&dn);
592 out:
593 	f2fs_msg(sbi->sb, KERN_NOTICE,
594 		"recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
595 		inode->i_ino,
596 		file_keep_isize(inode) ? "keep" : "recover",
597 		recovered, err);
598 	return err;
599 }
600 
recover_data(struct f2fs_sb_info * sbi,struct list_head * inode_list,struct list_head * tmp_inode_list,struct list_head * dir_list)601 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
602 		struct list_head *tmp_inode_list, struct list_head *dir_list)
603 {
604 	struct curseg_info *curseg;
605 	struct page *page = NULL;
606 	int err = 0;
607 	block_t blkaddr;
608 
609 	/* get node pages in the current segment */
610 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
611 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
612 
613 	while (1) {
614 		struct fsync_inode_entry *entry;
615 
616 		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
617 			break;
618 
619 		f2fs_ra_meta_pages_cond(sbi, blkaddr);
620 
621 		page = f2fs_get_tmp_page(sbi, blkaddr);
622 		if (IS_ERR(page)) {
623 			err = PTR_ERR(page);
624 			break;
625 		}
626 
627 		if (!is_recoverable_dnode(page)) {
628 			f2fs_put_page(page, 1);
629 			break;
630 		}
631 
632 		entry = get_fsync_inode(inode_list, ino_of_node(page));
633 		if (!entry)
634 			goto next;
635 		/*
636 		 * inode(x) | CP | inode(x) | dnode(F)
637 		 * In this case, we can lose the latest inode(x).
638 		 * So, call recover_inode for the inode update.
639 		 */
640 		if (IS_INODE(page))
641 			recover_inode(entry->inode, page);
642 		if (entry->last_dentry == blkaddr) {
643 			err = recover_dentry(entry->inode, page, dir_list);
644 			if (err) {
645 				f2fs_put_page(page, 1);
646 				break;
647 			}
648 		}
649 		err = do_recover_data(sbi, entry->inode, page);
650 		if (err) {
651 			f2fs_put_page(page, 1);
652 			break;
653 		}
654 
655 		if (entry->blkaddr == blkaddr)
656 			list_move_tail(&entry->list, tmp_inode_list);
657 next:
658 		/* check next segment */
659 		blkaddr = next_blkaddr_of_node(page);
660 		f2fs_put_page(page, 1);
661 	}
662 	if (!err)
663 		f2fs_allocate_new_segments(sbi);
664 	return err;
665 }
666 
f2fs_recover_fsync_data(struct f2fs_sb_info * sbi,bool check_only)667 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
668 {
669 	struct list_head inode_list, tmp_inode_list;
670 	struct list_head dir_list;
671 	int err;
672 	int ret = 0;
673 	unsigned long s_flags = sbi->sb->s_flags;
674 	bool need_writecp = false;
675 #ifdef CONFIG_QUOTA
676 	int quota_enabled;
677 #endif
678 
679 	if (s_flags & SB_RDONLY) {
680 		f2fs_msg(sbi->sb, KERN_INFO,
681 				"recover fsync data on readonly fs");
682 		sbi->sb->s_flags &= ~SB_RDONLY;
683 	}
684 
685 #ifdef CONFIG_QUOTA
686 	/* Needed for iput() to work correctly and not trash data */
687 	sbi->sb->s_flags |= SB_ACTIVE;
688 	/* Turn on quotas so that they are updated correctly */
689 	quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
690 #endif
691 
692 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
693 			sizeof(struct fsync_inode_entry));
694 	if (!fsync_entry_slab) {
695 		err = -ENOMEM;
696 		goto out;
697 	}
698 
699 	INIT_LIST_HEAD(&inode_list);
700 	INIT_LIST_HEAD(&tmp_inode_list);
701 	INIT_LIST_HEAD(&dir_list);
702 
703 	/* prevent checkpoint */
704 	mutex_lock(&sbi->cp_mutex);
705 
706 	/* step #1: find fsynced inode numbers */
707 	err = find_fsync_dnodes(sbi, &inode_list, check_only);
708 	if (err || list_empty(&inode_list))
709 		goto skip;
710 
711 	if (check_only) {
712 		ret = 1;
713 		goto skip;
714 	}
715 
716 	need_writecp = true;
717 
718 	/* step #2: recover data */
719 	err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
720 	if (!err)
721 		f2fs_bug_on(sbi, !list_empty(&inode_list));
722 	else {
723 		/* restore s_flags to let iput() trash data */
724 		sbi->sb->s_flags = s_flags;
725 	}
726 skip:
727 	destroy_fsync_dnodes(&inode_list, err);
728 	destroy_fsync_dnodes(&tmp_inode_list, err);
729 
730 	/* truncate meta pages to be used by the recovery */
731 	truncate_inode_pages_range(META_MAPPING(sbi),
732 			(loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
733 
734 	if (err) {
735 		truncate_inode_pages_final(NODE_MAPPING(sbi));
736 		truncate_inode_pages_final(META_MAPPING(sbi));
737 	} else {
738 		clear_sbi_flag(sbi, SBI_POR_DOING);
739 	}
740 	mutex_unlock(&sbi->cp_mutex);
741 
742 	/* let's drop all the directory inodes for clean checkpoint */
743 	destroy_fsync_dnodes(&dir_list, err);
744 
745 	if (need_writecp) {
746 		set_sbi_flag(sbi, SBI_IS_RECOVERED);
747 
748 		if (!err) {
749 			struct cp_control cpc = {
750 				.reason = CP_RECOVERY,
751 			};
752 			err = f2fs_write_checkpoint(sbi, &cpc);
753 		}
754 	}
755 
756 	kmem_cache_destroy(fsync_entry_slab);
757 out:
758 #ifdef CONFIG_QUOTA
759 	/* Turn quotas off */
760 	if (quota_enabled)
761 		f2fs_quota_off_umount(sbi->sb);
762 #endif
763 	sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
764 
765 	return ret ? ret: err;
766 }
767