xref: /openbmc/linux/fs/f2fs/recovery.c (revision aac5987a)
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 {
203 	struct curseg_info *curseg;
204 	struct page *page = NULL;
205 	block_t blkaddr;
206 	int err = 0;
207 
208 	/* get node pages in the current segment */
209 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
210 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
211 
212 	while (1) {
213 		struct fsync_inode_entry *entry;
214 
215 		if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
216 			return 0;
217 
218 		page = get_tmp_page(sbi, blkaddr);
219 
220 		if (!is_recoverable_dnode(page))
221 			break;
222 
223 		if (!is_fsync_dnode(page))
224 			goto next;
225 
226 		entry = get_fsync_inode(head, ino_of_node(page));
227 		if (!entry) {
228 			if (IS_INODE(page) && is_dent_dnode(page)) {
229 				err = recover_inode_page(sbi, page);
230 				if (err)
231 					break;
232 			}
233 
234 			/*
235 			 * CP | dnode(F) | inode(DF)
236 			 * For this case, we should not give up now.
237 			 */
238 			entry = add_fsync_inode(sbi, head, ino_of_node(page));
239 			if (IS_ERR(entry)) {
240 				err = PTR_ERR(entry);
241 				if (err == -ENOENT) {
242 					err = 0;
243 					goto next;
244 				}
245 				break;
246 			}
247 		}
248 		entry->blkaddr = blkaddr;
249 
250 		if (IS_INODE(page) && is_dent_dnode(page))
251 			entry->last_dentry = blkaddr;
252 next:
253 		/* check next segment */
254 		blkaddr = next_blkaddr_of_node(page);
255 		f2fs_put_page(page, 1);
256 
257 		ra_meta_pages_cond(sbi, blkaddr);
258 	}
259 	f2fs_put_page(page, 1);
260 	return err;
261 }
262 
263 static void destroy_fsync_dnodes(struct list_head *head)
264 {
265 	struct fsync_inode_entry *entry, *tmp;
266 
267 	list_for_each_entry_safe(entry, tmp, head, list)
268 		del_fsync_inode(entry);
269 }
270 
271 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
272 			block_t blkaddr, struct dnode_of_data *dn)
273 {
274 	struct seg_entry *sentry;
275 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
276 	unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
277 	struct f2fs_summary_block *sum_node;
278 	struct f2fs_summary sum;
279 	struct page *sum_page, *node_page;
280 	struct dnode_of_data tdn = *dn;
281 	nid_t ino, nid;
282 	struct inode *inode;
283 	unsigned int offset;
284 	block_t bidx;
285 	int i;
286 
287 	sentry = get_seg_entry(sbi, segno);
288 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
289 		return 0;
290 
291 	/* Get the previous summary */
292 	for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
293 		struct curseg_info *curseg = CURSEG_I(sbi, i);
294 		if (curseg->segno == segno) {
295 			sum = curseg->sum_blk->entries[blkoff];
296 			goto got_it;
297 		}
298 	}
299 
300 	sum_page = get_sum_page(sbi, segno);
301 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
302 	sum = sum_node->entries[blkoff];
303 	f2fs_put_page(sum_page, 1);
304 got_it:
305 	/* Use the locked dnode page and inode */
306 	nid = le32_to_cpu(sum.nid);
307 	if (dn->inode->i_ino == nid) {
308 		tdn.nid = nid;
309 		if (!dn->inode_page_locked)
310 			lock_page(dn->inode_page);
311 		tdn.node_page = dn->inode_page;
312 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
313 		goto truncate_out;
314 	} else if (dn->nid == nid) {
315 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
316 		goto truncate_out;
317 	}
318 
319 	/* Get the node page */
320 	node_page = get_node_page(sbi, nid);
321 	if (IS_ERR(node_page))
322 		return PTR_ERR(node_page);
323 
324 	offset = ofs_of_node(node_page);
325 	ino = ino_of_node(node_page);
326 	f2fs_put_page(node_page, 1);
327 
328 	if (ino != dn->inode->i_ino) {
329 		/* Deallocate previous index in the node page */
330 		inode = f2fs_iget_retry(sbi->sb, ino);
331 		if (IS_ERR(inode))
332 			return PTR_ERR(inode);
333 	} else {
334 		inode = dn->inode;
335 	}
336 
337 	bidx = start_bidx_of_node(offset, inode) + le16_to_cpu(sum.ofs_in_node);
338 
339 	/*
340 	 * if inode page is locked, unlock temporarily, but its reference
341 	 * count keeps alive.
342 	 */
343 	if (ino == dn->inode->i_ino && dn->inode_page_locked)
344 		unlock_page(dn->inode_page);
345 
346 	set_new_dnode(&tdn, inode, NULL, NULL, 0);
347 	if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
348 		goto out;
349 
350 	if (tdn.data_blkaddr == blkaddr)
351 		truncate_data_blocks_range(&tdn, 1);
352 
353 	f2fs_put_dnode(&tdn);
354 out:
355 	if (ino != dn->inode->i_ino)
356 		iput(inode);
357 	else if (dn->inode_page_locked)
358 		lock_page(dn->inode_page);
359 	return 0;
360 
361 truncate_out:
362 	if (datablock_addr(tdn.node_page, tdn.ofs_in_node) == blkaddr)
363 		truncate_data_blocks_range(&tdn, 1);
364 	if (dn->inode->i_ino == nid && !dn->inode_page_locked)
365 		unlock_page(dn->inode_page);
366 	return 0;
367 }
368 
369 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
370 					struct page *page, block_t blkaddr)
371 {
372 	struct dnode_of_data dn;
373 	struct node_info ni;
374 	unsigned int start, end;
375 	int err = 0, recovered = 0;
376 
377 	/* step 1: recover xattr */
378 	if (IS_INODE(page)) {
379 		recover_inline_xattr(inode, page);
380 	} else if (f2fs_has_xattr_block(ofs_of_node(page))) {
381 		err = recover_xattr_data(inode, page, blkaddr);
382 		if (!err)
383 			recovered++;
384 		goto out;
385 	}
386 
387 	/* step 2: recover inline data */
388 	if (recover_inline_data(inode, page))
389 		goto out;
390 
391 	/* step 3: recover data indices */
392 	start = start_bidx_of_node(ofs_of_node(page), inode);
393 	end = start + ADDRS_PER_PAGE(page, inode);
394 
395 	set_new_dnode(&dn, inode, NULL, NULL, 0);
396 retry_dn:
397 	err = get_dnode_of_data(&dn, start, ALLOC_NODE);
398 	if (err) {
399 		if (err == -ENOMEM) {
400 			congestion_wait(BLK_RW_ASYNC, HZ/50);
401 			goto retry_dn;
402 		}
403 		goto out;
404 	}
405 
406 	f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
407 
408 	get_node_info(sbi, dn.nid, &ni);
409 	f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
410 	f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
411 
412 	for (; start < end; start++, dn.ofs_in_node++) {
413 		block_t src, dest;
414 
415 		src = datablock_addr(dn.node_page, dn.ofs_in_node);
416 		dest = datablock_addr(page, dn.ofs_in_node);
417 
418 		/* skip recovering if dest is the same as src */
419 		if (src == dest)
420 			continue;
421 
422 		/* dest is invalid, just invalidate src block */
423 		if (dest == NULL_ADDR) {
424 			truncate_data_blocks_range(&dn, 1);
425 			continue;
426 		}
427 
428 		if (!file_keep_isize(inode) &&
429 			(i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
430 			f2fs_i_size_write(inode,
431 				(loff_t)(start + 1) << PAGE_SHIFT);
432 
433 		/*
434 		 * dest is reserved block, invalidate src block
435 		 * and then reserve one new block in dnode page.
436 		 */
437 		if (dest == NEW_ADDR) {
438 			truncate_data_blocks_range(&dn, 1);
439 			reserve_new_block(&dn);
440 			continue;
441 		}
442 
443 		/* dest is valid block, try to recover from src to dest */
444 		if (is_valid_blkaddr(sbi, dest, META_POR)) {
445 
446 			if (src == NULL_ADDR) {
447 				err = reserve_new_block(&dn);
448 #ifdef CONFIG_F2FS_FAULT_INJECTION
449 				while (err)
450 					err = reserve_new_block(&dn);
451 #endif
452 				/* We should not get -ENOSPC */
453 				f2fs_bug_on(sbi, err);
454 				if (err)
455 					goto err;
456 			}
457 retry_prev:
458 			/* Check the previous node page having this index */
459 			err = check_index_in_prev_nodes(sbi, dest, &dn);
460 			if (err) {
461 				if (err == -ENOMEM) {
462 					congestion_wait(BLK_RW_ASYNC, HZ/50);
463 					goto retry_prev;
464 				}
465 				goto err;
466 			}
467 
468 			/* write dummy data page */
469 			f2fs_replace_block(sbi, &dn, src, dest,
470 						ni.version, false, false);
471 			recovered++;
472 		}
473 	}
474 
475 	copy_node_footer(dn.node_page, page);
476 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
477 					ofs_of_node(page), false);
478 	set_page_dirty(dn.node_page);
479 err:
480 	f2fs_put_dnode(&dn);
481 out:
482 	f2fs_msg(sbi->sb, KERN_NOTICE,
483 		"recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
484 		inode->i_ino,
485 		file_keep_isize(inode) ? "keep" : "recover",
486 		recovered, err);
487 	return err;
488 }
489 
490 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
491 						struct list_head *dir_list)
492 {
493 	struct curseg_info *curseg;
494 	struct page *page = NULL;
495 	int err = 0;
496 	block_t blkaddr;
497 
498 	/* get node pages in the current segment */
499 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
500 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
501 
502 	while (1) {
503 		struct fsync_inode_entry *entry;
504 
505 		if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
506 			break;
507 
508 		ra_meta_pages_cond(sbi, blkaddr);
509 
510 		page = get_tmp_page(sbi, blkaddr);
511 
512 		if (!is_recoverable_dnode(page)) {
513 			f2fs_put_page(page, 1);
514 			break;
515 		}
516 
517 		entry = get_fsync_inode(inode_list, ino_of_node(page));
518 		if (!entry)
519 			goto next;
520 		/*
521 		 * inode(x) | CP | inode(x) | dnode(F)
522 		 * In this case, we can lose the latest inode(x).
523 		 * So, call recover_inode for the inode update.
524 		 */
525 		if (IS_INODE(page))
526 			recover_inode(entry->inode, page);
527 		if (entry->last_dentry == blkaddr) {
528 			err = recover_dentry(entry->inode, page, dir_list);
529 			if (err) {
530 				f2fs_put_page(page, 1);
531 				break;
532 			}
533 		}
534 		err = do_recover_data(sbi, entry->inode, page, blkaddr);
535 		if (err) {
536 			f2fs_put_page(page, 1);
537 			break;
538 		}
539 
540 		if (entry->blkaddr == blkaddr)
541 			del_fsync_inode(entry);
542 next:
543 		/* check next segment */
544 		blkaddr = next_blkaddr_of_node(page);
545 		f2fs_put_page(page, 1);
546 	}
547 	if (!err)
548 		allocate_new_segments(sbi);
549 	return err;
550 }
551 
552 int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
553 {
554 	struct list_head inode_list;
555 	struct list_head dir_list;
556 	int err;
557 	int ret = 0;
558 	bool need_writecp = false;
559 
560 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
561 			sizeof(struct fsync_inode_entry));
562 	if (!fsync_entry_slab)
563 		return -ENOMEM;
564 
565 	INIT_LIST_HEAD(&inode_list);
566 	INIT_LIST_HEAD(&dir_list);
567 
568 	/* prevent checkpoint */
569 	mutex_lock(&sbi->cp_mutex);
570 
571 	/* step #1: find fsynced inode numbers */
572 	err = find_fsync_dnodes(sbi, &inode_list);
573 	if (err || list_empty(&inode_list))
574 		goto out;
575 
576 	if (check_only) {
577 		ret = 1;
578 		goto out;
579 	}
580 
581 	need_writecp = true;
582 
583 	/* step #2: recover data */
584 	err = recover_data(sbi, &inode_list, &dir_list);
585 	if (!err)
586 		f2fs_bug_on(sbi, !list_empty(&inode_list));
587 out:
588 	destroy_fsync_dnodes(&inode_list);
589 
590 	/* truncate meta pages to be used by the recovery */
591 	truncate_inode_pages_range(META_MAPPING(sbi),
592 			(loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
593 
594 	if (err) {
595 		truncate_inode_pages_final(NODE_MAPPING(sbi));
596 		truncate_inode_pages_final(META_MAPPING(sbi));
597 	}
598 
599 	clear_sbi_flag(sbi, SBI_POR_DOING);
600 	if (err)
601 		set_ckpt_flags(sbi, CP_ERROR_FLAG);
602 	mutex_unlock(&sbi->cp_mutex);
603 
604 	/* let's drop all the directory inodes for clean checkpoint */
605 	destroy_fsync_dnodes(&dir_list);
606 
607 	if (!err && need_writecp) {
608 		struct cp_control cpc = {
609 			.reason = CP_RECOVERY,
610 		};
611 		err = write_checkpoint(sbi, &cpc);
612 	}
613 
614 	kmem_cache_destroy(fsync_entry_slab);
615 	return ret ? ret: err;
616 }
617