xref: /openbmc/linux/fs/f2fs/recovery.c (revision 239480ab)
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 
50 bool 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 
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 
71 static struct fsync_inode_entry *add_fsync_inode(struct f2fs_sb_info *sbi,
72 					struct list_head *head, nid_t ino)
73 {
74 	struct inode *inode;
75 	struct fsync_inode_entry *entry;
76 
77 	inode = f2fs_iget_retry(sbi->sb, ino);
78 	if (IS_ERR(inode))
79 		return ERR_CAST(inode);
80 
81 	entry = f2fs_kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
82 	entry->inode = inode;
83 	list_add_tail(&entry->list, head);
84 
85 	return entry;
86 }
87 
88 static void del_fsync_inode(struct fsync_inode_entry *entry)
89 {
90 	iput(entry->inode);
91 	list_del(&entry->list);
92 	kmem_cache_free(fsync_entry_slab, entry);
93 }
94 
95 static int recover_dentry(struct inode *inode, struct page *ipage,
96 						struct list_head *dir_list)
97 {
98 	struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
99 	nid_t pino = le32_to_cpu(raw_inode->i_pino);
100 	struct f2fs_dir_entry *de;
101 	struct fscrypt_name fname;
102 	struct page *page;
103 	struct inode *dir, *einode;
104 	struct fsync_inode_entry *entry;
105 	int err = 0;
106 	char *name;
107 
108 	entry = get_fsync_inode(dir_list, pino);
109 	if (!entry) {
110 		entry = add_fsync_inode(F2FS_I_SB(inode), dir_list, pino);
111 		if (IS_ERR(entry)) {
112 			dir = ERR_CAST(entry);
113 			err = PTR_ERR(entry);
114 			goto out;
115 		}
116 	}
117 
118 	dir = entry->inode;
119 
120 	memset(&fname, 0, sizeof(struct fscrypt_name));
121 	fname.disk_name.len = le32_to_cpu(raw_inode->i_namelen);
122 	fname.disk_name.name = raw_inode->i_name;
123 
124 	if (unlikely(fname.disk_name.len > F2FS_NAME_LEN)) {
125 		WARN_ON(1);
126 		err = -ENAMETOOLONG;
127 		goto out;
128 	}
129 retry:
130 	de = __f2fs_find_entry(dir, &fname, &page);
131 	if (de && inode->i_ino == le32_to_cpu(de->ino))
132 		goto out_unmap_put;
133 
134 	if (de) {
135 		einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
136 		if (IS_ERR(einode)) {
137 			WARN_ON(1);
138 			err = PTR_ERR(einode);
139 			if (err == -ENOENT)
140 				err = -EEXIST;
141 			goto out_unmap_put;
142 		}
143 		err = acquire_orphan_inode(F2FS_I_SB(inode));
144 		if (err) {
145 			iput(einode);
146 			goto out_unmap_put;
147 		}
148 		f2fs_delete_entry(de, page, dir, einode);
149 		iput(einode);
150 		goto retry;
151 	} else if (IS_ERR(page)) {
152 		err = PTR_ERR(page);
153 	} else {
154 		err = __f2fs_do_add_link(dir, &fname, inode,
155 					inode->i_ino, inode->i_mode);
156 	}
157 	if (err == -ENOMEM)
158 		goto retry;
159 	goto out;
160 
161 out_unmap_put:
162 	f2fs_dentry_kunmap(dir, page);
163 	f2fs_put_page(page, 0);
164 out:
165 	if (file_enc_name(inode))
166 		name = "<encrypted>";
167 	else
168 		name = raw_inode->i_name;
169 	f2fs_msg(inode->i_sb, KERN_NOTICE,
170 			"%s: ino = %x, name = %s, dir = %lx, err = %d",
171 			__func__, ino_of_node(ipage), name,
172 			IS_ERR(dir) ? 0 : dir->i_ino, err);
173 	return err;
174 }
175 
176 static void recover_inode(struct inode *inode, struct page *page)
177 {
178 	struct f2fs_inode *raw = F2FS_INODE(page);
179 	char *name;
180 
181 	inode->i_mode = le16_to_cpu(raw->i_mode);
182 	f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
183 	inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
184 	inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
185 	inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
186 	inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
187 	inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
188 	inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
189 
190 	F2FS_I(inode)->i_advise = raw->i_advise;
191 
192 	if (file_enc_name(inode))
193 		name = "<encrypted>";
194 	else
195 		name = F2FS_INODE(page)->i_name;
196 
197 	f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
198 			ino_of_node(page), name);
199 }
200 
201 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
202 				bool check_only)
203 {
204 	struct curseg_info *curseg;
205 	struct page *page = NULL;
206 	block_t blkaddr;
207 	int err = 0;
208 
209 	/* get node pages in the current segment */
210 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
211 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
212 
213 	while (1) {
214 		struct fsync_inode_entry *entry;
215 
216 		if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
217 			return 0;
218 
219 		page = get_tmp_page(sbi, blkaddr);
220 
221 		if (!is_recoverable_dnode(page))
222 			break;
223 
224 		if (!is_fsync_dnode(page))
225 			goto next;
226 
227 		entry = get_fsync_inode(head, ino_of_node(page));
228 		if (!entry) {
229 			if (!check_only &&
230 					IS_INODE(page) && is_dent_dnode(page)) {
231 				err = recover_inode_page(sbi, page);
232 				if (err)
233 					break;
234 			}
235 
236 			/*
237 			 * CP | dnode(F) | inode(DF)
238 			 * For this case, we should not give up now.
239 			 */
240 			entry = add_fsync_inode(sbi, head, ino_of_node(page));
241 			if (IS_ERR(entry)) {
242 				err = PTR_ERR(entry);
243 				if (err == -ENOENT) {
244 					err = 0;
245 					goto next;
246 				}
247 				break;
248 			}
249 		}
250 		entry->blkaddr = blkaddr;
251 
252 		if (IS_INODE(page) && is_dent_dnode(page))
253 			entry->last_dentry = blkaddr;
254 next:
255 		/* check next segment */
256 		blkaddr = next_blkaddr_of_node(page);
257 		f2fs_put_page(page, 1);
258 
259 		ra_meta_pages_cond(sbi, blkaddr);
260 	}
261 	f2fs_put_page(page, 1);
262 	return err;
263 }
264 
265 static void destroy_fsync_dnodes(struct list_head *head)
266 {
267 	struct fsync_inode_entry *entry, *tmp;
268 
269 	list_for_each_entry_safe(entry, tmp, head, list)
270 		del_fsync_inode(entry);
271 }
272 
273 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
274 			block_t blkaddr, struct dnode_of_data *dn)
275 {
276 	struct seg_entry *sentry;
277 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
278 	unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
279 	struct f2fs_summary_block *sum_node;
280 	struct f2fs_summary sum;
281 	struct page *sum_page, *node_page;
282 	struct dnode_of_data tdn = *dn;
283 	nid_t ino, nid;
284 	struct inode *inode;
285 	unsigned int offset;
286 	block_t bidx;
287 	int i;
288 
289 	sentry = get_seg_entry(sbi, segno);
290 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
291 		return 0;
292 
293 	/* Get the previous summary */
294 	for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
295 		struct curseg_info *curseg = CURSEG_I(sbi, i);
296 		if (curseg->segno == segno) {
297 			sum = curseg->sum_blk->entries[blkoff];
298 			goto got_it;
299 		}
300 	}
301 
302 	sum_page = get_sum_page(sbi, segno);
303 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
304 	sum = sum_node->entries[blkoff];
305 	f2fs_put_page(sum_page, 1);
306 got_it:
307 	/* Use the locked dnode page and inode */
308 	nid = le32_to_cpu(sum.nid);
309 	if (dn->inode->i_ino == nid) {
310 		tdn.nid = nid;
311 		if (!dn->inode_page_locked)
312 			lock_page(dn->inode_page);
313 		tdn.node_page = dn->inode_page;
314 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
315 		goto truncate_out;
316 	} else if (dn->nid == nid) {
317 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
318 		goto truncate_out;
319 	}
320 
321 	/* Get the node page */
322 	node_page = get_node_page(sbi, nid);
323 	if (IS_ERR(node_page))
324 		return PTR_ERR(node_page);
325 
326 	offset = ofs_of_node(node_page);
327 	ino = ino_of_node(node_page);
328 	f2fs_put_page(node_page, 1);
329 
330 	if (ino != dn->inode->i_ino) {
331 		/* Deallocate previous index in the node page */
332 		inode = f2fs_iget_retry(sbi->sb, ino);
333 		if (IS_ERR(inode))
334 			return PTR_ERR(inode);
335 	} else {
336 		inode = dn->inode;
337 	}
338 
339 	bidx = start_bidx_of_node(offset, inode) + le16_to_cpu(sum.ofs_in_node);
340 
341 	/*
342 	 * if inode page is locked, unlock temporarily, but its reference
343 	 * count keeps alive.
344 	 */
345 	if (ino == dn->inode->i_ino && dn->inode_page_locked)
346 		unlock_page(dn->inode_page);
347 
348 	set_new_dnode(&tdn, inode, NULL, NULL, 0);
349 	if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
350 		goto out;
351 
352 	if (tdn.data_blkaddr == blkaddr)
353 		truncate_data_blocks_range(&tdn, 1);
354 
355 	f2fs_put_dnode(&tdn);
356 out:
357 	if (ino != dn->inode->i_ino)
358 		iput(inode);
359 	else if (dn->inode_page_locked)
360 		lock_page(dn->inode_page);
361 	return 0;
362 
363 truncate_out:
364 	if (datablock_addr(tdn.node_page, tdn.ofs_in_node) == blkaddr)
365 		truncate_data_blocks_range(&tdn, 1);
366 	if (dn->inode->i_ino == nid && !dn->inode_page_locked)
367 		unlock_page(dn->inode_page);
368 	return 0;
369 }
370 
371 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
372 					struct page *page, block_t blkaddr)
373 {
374 	struct dnode_of_data dn;
375 	struct node_info ni;
376 	unsigned int start, end;
377 	int err = 0, recovered = 0;
378 
379 	/* step 1: recover xattr */
380 	if (IS_INODE(page)) {
381 		recover_inline_xattr(inode, page);
382 	} else if (f2fs_has_xattr_block(ofs_of_node(page))) {
383 		err = recover_xattr_data(inode, page, blkaddr);
384 		if (!err)
385 			recovered++;
386 		goto out;
387 	}
388 
389 	/* step 2: recover inline data */
390 	if (recover_inline_data(inode, page))
391 		goto out;
392 
393 	/* step 3: recover data indices */
394 	start = start_bidx_of_node(ofs_of_node(page), inode);
395 	end = start + ADDRS_PER_PAGE(page, inode);
396 
397 	set_new_dnode(&dn, inode, NULL, NULL, 0);
398 retry_dn:
399 	err = get_dnode_of_data(&dn, start, ALLOC_NODE);
400 	if (err) {
401 		if (err == -ENOMEM) {
402 			congestion_wait(BLK_RW_ASYNC, HZ/50);
403 			goto retry_dn;
404 		}
405 		goto out;
406 	}
407 
408 	f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
409 
410 	get_node_info(sbi, dn.nid, &ni);
411 	f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
412 	f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
413 
414 	for (; start < end; start++, dn.ofs_in_node++) {
415 		block_t src, dest;
416 
417 		src = datablock_addr(dn.node_page, dn.ofs_in_node);
418 		dest = datablock_addr(page, dn.ofs_in_node);
419 
420 		/* skip recovering if dest is the same as src */
421 		if (src == dest)
422 			continue;
423 
424 		/* dest is invalid, just invalidate src block */
425 		if (dest == NULL_ADDR) {
426 			truncate_data_blocks_range(&dn, 1);
427 			continue;
428 		}
429 
430 		if (!file_keep_isize(inode) &&
431 			(i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
432 			f2fs_i_size_write(inode,
433 				(loff_t)(start + 1) << PAGE_SHIFT);
434 
435 		/*
436 		 * dest is reserved block, invalidate src block
437 		 * and then reserve one new block in dnode page.
438 		 */
439 		if (dest == NEW_ADDR) {
440 			truncate_data_blocks_range(&dn, 1);
441 			reserve_new_block(&dn);
442 			continue;
443 		}
444 
445 		/* dest is valid block, try to recover from src to dest */
446 		if (is_valid_blkaddr(sbi, dest, META_POR)) {
447 
448 			if (src == NULL_ADDR) {
449 				err = reserve_new_block(&dn);
450 #ifdef CONFIG_F2FS_FAULT_INJECTION
451 				while (err)
452 					err = reserve_new_block(&dn);
453 #endif
454 				/* We should not get -ENOSPC */
455 				f2fs_bug_on(sbi, err);
456 				if (err)
457 					goto err;
458 			}
459 retry_prev:
460 			/* Check the previous node page having this index */
461 			err = check_index_in_prev_nodes(sbi, dest, &dn);
462 			if (err) {
463 				if (err == -ENOMEM) {
464 					congestion_wait(BLK_RW_ASYNC, HZ/50);
465 					goto retry_prev;
466 				}
467 				goto err;
468 			}
469 
470 			/* write dummy data page */
471 			f2fs_replace_block(sbi, &dn, src, dest,
472 						ni.version, false, false);
473 			recovered++;
474 		}
475 	}
476 
477 	copy_node_footer(dn.node_page, page);
478 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
479 					ofs_of_node(page), false);
480 	set_page_dirty(dn.node_page);
481 err:
482 	f2fs_put_dnode(&dn);
483 out:
484 	f2fs_msg(sbi->sb, KERN_NOTICE,
485 		"recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
486 		inode->i_ino,
487 		file_keep_isize(inode) ? "keep" : "recover",
488 		recovered, err);
489 	return err;
490 }
491 
492 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
493 						struct list_head *dir_list)
494 {
495 	struct curseg_info *curseg;
496 	struct page *page = NULL;
497 	int err = 0;
498 	block_t blkaddr;
499 
500 	/* get node pages in the current segment */
501 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
502 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
503 
504 	while (1) {
505 		struct fsync_inode_entry *entry;
506 
507 		if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
508 			break;
509 
510 		ra_meta_pages_cond(sbi, blkaddr);
511 
512 		page = get_tmp_page(sbi, blkaddr);
513 
514 		if (!is_recoverable_dnode(page)) {
515 			f2fs_put_page(page, 1);
516 			break;
517 		}
518 
519 		entry = get_fsync_inode(inode_list, ino_of_node(page));
520 		if (!entry)
521 			goto next;
522 		/*
523 		 * inode(x) | CP | inode(x) | dnode(F)
524 		 * In this case, we can lose the latest inode(x).
525 		 * So, call recover_inode for the inode update.
526 		 */
527 		if (IS_INODE(page))
528 			recover_inode(entry->inode, page);
529 		if (entry->last_dentry == blkaddr) {
530 			err = recover_dentry(entry->inode, page, dir_list);
531 			if (err) {
532 				f2fs_put_page(page, 1);
533 				break;
534 			}
535 		}
536 		err = do_recover_data(sbi, entry->inode, page, blkaddr);
537 		if (err) {
538 			f2fs_put_page(page, 1);
539 			break;
540 		}
541 
542 		if (entry->blkaddr == blkaddr)
543 			del_fsync_inode(entry);
544 next:
545 		/* check next segment */
546 		blkaddr = next_blkaddr_of_node(page);
547 		f2fs_put_page(page, 1);
548 	}
549 	if (!err)
550 		allocate_new_segments(sbi);
551 	return err;
552 }
553 
554 int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
555 {
556 	struct list_head inode_list;
557 	struct list_head dir_list;
558 	int err;
559 	int ret = 0;
560 	bool need_writecp = false;
561 
562 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
563 			sizeof(struct fsync_inode_entry));
564 	if (!fsync_entry_slab)
565 		return -ENOMEM;
566 
567 	INIT_LIST_HEAD(&inode_list);
568 	INIT_LIST_HEAD(&dir_list);
569 
570 	/* prevent checkpoint */
571 	mutex_lock(&sbi->cp_mutex);
572 
573 	/* step #1: find fsynced inode numbers */
574 	err = find_fsync_dnodes(sbi, &inode_list, check_only);
575 	if (err || list_empty(&inode_list))
576 		goto out;
577 
578 	if (check_only) {
579 		ret = 1;
580 		goto out;
581 	}
582 
583 	need_writecp = true;
584 
585 	/* step #2: recover data */
586 	err = recover_data(sbi, &inode_list, &dir_list);
587 	if (!err)
588 		f2fs_bug_on(sbi, !list_empty(&inode_list));
589 out:
590 	destroy_fsync_dnodes(&inode_list);
591 
592 	/* truncate meta pages to be used by the recovery */
593 	truncate_inode_pages_range(META_MAPPING(sbi),
594 			(loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
595 
596 	if (err) {
597 		truncate_inode_pages_final(NODE_MAPPING(sbi));
598 		truncate_inode_pages_final(META_MAPPING(sbi));
599 	}
600 
601 	clear_sbi_flag(sbi, SBI_POR_DOING);
602 	if (err)
603 		set_ckpt_flags(sbi, CP_ERROR_FLAG);
604 	mutex_unlock(&sbi->cp_mutex);
605 
606 	/* let's drop all the directory inodes for clean checkpoint */
607 	destroy_fsync_dnodes(&dir_list);
608 
609 	if (!err && need_writecp) {
610 		struct cp_control cpc = {
611 			.reason = CP_RECOVERY,
612 		};
613 		err = write_checkpoint(sbi, &cpc);
614 	}
615 
616 	kmem_cache_destroy(fsync_entry_slab);
617 	return ret ? ret: err;
618 }
619