xref: /openbmc/linux/fs/f2fs/recovery.c (revision 93d90ad7) 
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 	if (sbi->last_valid_block_count + sbi->alloc_valid_block_count
53 			> sbi->user_block_count)
54 		return false;
55 	return true;
56 }
57 
58 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
59 								nid_t ino)
60 {
61 	struct fsync_inode_entry *entry;
62 
63 	list_for_each_entry(entry, head, list)
64 		if (entry->inode->i_ino == ino)
65 			return entry;
66 
67 	return NULL;
68 }
69 
70 static int recover_dentry(struct inode *inode, struct page *ipage)
71 {
72 	struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
73 	nid_t pino = le32_to_cpu(raw_inode->i_pino);
74 	struct f2fs_dir_entry *de;
75 	struct qstr name;
76 	struct page *page;
77 	struct inode *dir, *einode;
78 	int err = 0;
79 
80 	dir = f2fs_iget(inode->i_sb, pino);
81 	if (IS_ERR(dir)) {
82 		err = PTR_ERR(dir);
83 		goto out;
84 	}
85 
86 	name.len = le32_to_cpu(raw_inode->i_namelen);
87 	name.name = raw_inode->i_name;
88 
89 	if (unlikely(name.len > F2FS_NAME_LEN)) {
90 		WARN_ON(1);
91 		err = -ENAMETOOLONG;
92 		goto out_err;
93 	}
94 retry:
95 	de = f2fs_find_entry(dir, &name, &page);
96 	if (de && inode->i_ino == le32_to_cpu(de->ino)) {
97 		clear_inode_flag(F2FS_I(inode), FI_INC_LINK);
98 		goto out_unmap_put;
99 	}
100 	if (de) {
101 		einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
102 		if (IS_ERR(einode)) {
103 			WARN_ON(1);
104 			err = PTR_ERR(einode);
105 			if (err == -ENOENT)
106 				err = -EEXIST;
107 			goto out_unmap_put;
108 		}
109 		err = acquire_orphan_inode(F2FS_I_SB(inode));
110 		if (err) {
111 			iput(einode);
112 			goto out_unmap_put;
113 		}
114 		f2fs_delete_entry(de, page, dir, einode);
115 		iput(einode);
116 		goto retry;
117 	}
118 	err = __f2fs_add_link(dir, &name, inode);
119 	if (err)
120 		goto out_err;
121 
122 	if (is_inode_flag_set(F2FS_I(dir), FI_DELAY_IPUT)) {
123 		iput(dir);
124 	} else {
125 		add_dirty_dir_inode(dir);
126 		set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
127 	}
128 
129 	goto out;
130 
131 out_unmap_put:
132 	f2fs_dentry_kunmap(dir, page);
133 	f2fs_put_page(page, 0);
134 out_err:
135 	iput(dir);
136 out:
137 	f2fs_msg(inode->i_sb, KERN_NOTICE,
138 			"%s: ino = %x, name = %s, dir = %lx, err = %d",
139 			__func__, ino_of_node(ipage), raw_inode->i_name,
140 			IS_ERR(dir) ? 0 : dir->i_ino, err);
141 	return err;
142 }
143 
144 static void recover_inode(struct inode *inode, struct page *page)
145 {
146 	struct f2fs_inode *raw = F2FS_INODE(page);
147 
148 	inode->i_mode = le16_to_cpu(raw->i_mode);
149 	i_size_write(inode, le64_to_cpu(raw->i_size));
150 	inode->i_atime.tv_sec = le64_to_cpu(raw->i_mtime);
151 	inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
152 	inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
153 	inode->i_atime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
154 	inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
155 	inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
156 
157 	f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
158 			ino_of_node(page), F2FS_INODE(page)->i_name);
159 }
160 
161 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
162 {
163 	unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
164 	struct curseg_info *curseg;
165 	struct page *page = NULL;
166 	block_t blkaddr;
167 	int err = 0;
168 
169 	/* get node pages in the current segment */
170 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
171 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
172 
173 	ra_meta_pages(sbi, blkaddr, 1, META_POR);
174 
175 	while (1) {
176 		struct fsync_inode_entry *entry;
177 
178 		if (blkaddr < MAIN_BLKADDR(sbi) || blkaddr >= MAX_BLKADDR(sbi))
179 			return 0;
180 
181 		page = get_meta_page(sbi, blkaddr);
182 
183 		if (cp_ver != cpver_of_node(page))
184 			break;
185 
186 		if (!is_fsync_dnode(page))
187 			goto next;
188 
189 		entry = get_fsync_inode(head, ino_of_node(page));
190 		if (entry) {
191 			if (IS_INODE(page) && is_dent_dnode(page))
192 				set_inode_flag(F2FS_I(entry->inode),
193 							FI_INC_LINK);
194 		} else {
195 			if (IS_INODE(page) && is_dent_dnode(page)) {
196 				err = recover_inode_page(sbi, page);
197 				if (err)
198 					break;
199 			}
200 
201 			/* add this fsync inode to the list */
202 			entry = kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
203 			if (!entry) {
204 				err = -ENOMEM;
205 				break;
206 			}
207 			/*
208 			 * CP | dnode(F) | inode(DF)
209 			 * For this case, we should not give up now.
210 			 */
211 			entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
212 			if (IS_ERR(entry->inode)) {
213 				err = PTR_ERR(entry->inode);
214 				kmem_cache_free(fsync_entry_slab, entry);
215 				if (err == -ENOENT)
216 					goto next;
217 				break;
218 			}
219 			list_add_tail(&entry->list, head);
220 		}
221 		entry->blkaddr = blkaddr;
222 
223 		if (IS_INODE(page)) {
224 			entry->last_inode = blkaddr;
225 			if (is_dent_dnode(page))
226 				entry->last_dentry = blkaddr;
227 		}
228 next:
229 		/* check next segment */
230 		blkaddr = next_blkaddr_of_node(page);
231 		f2fs_put_page(page, 1);
232 
233 		ra_meta_pages_cond(sbi, blkaddr);
234 	}
235 	f2fs_put_page(page, 1);
236 	return err;
237 }
238 
239 static void destroy_fsync_dnodes(struct list_head *head)
240 {
241 	struct fsync_inode_entry *entry, *tmp;
242 
243 	list_for_each_entry_safe(entry, tmp, head, list) {
244 		iput(entry->inode);
245 		list_del(&entry->list);
246 		kmem_cache_free(fsync_entry_slab, entry);
247 	}
248 }
249 
250 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
251 			block_t blkaddr, struct dnode_of_data *dn)
252 {
253 	struct seg_entry *sentry;
254 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
255 	unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
256 	struct f2fs_summary_block *sum_node;
257 	struct f2fs_summary sum;
258 	struct page *sum_page, *node_page;
259 	nid_t ino, nid;
260 	struct inode *inode;
261 	unsigned int offset;
262 	block_t bidx;
263 	int i;
264 
265 	sentry = get_seg_entry(sbi, segno);
266 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
267 		return 0;
268 
269 	/* Get the previous summary */
270 	for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
271 		struct curseg_info *curseg = CURSEG_I(sbi, i);
272 		if (curseg->segno == segno) {
273 			sum = curseg->sum_blk->entries[blkoff];
274 			goto got_it;
275 		}
276 	}
277 
278 	sum_page = get_sum_page(sbi, segno);
279 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
280 	sum = sum_node->entries[blkoff];
281 	f2fs_put_page(sum_page, 1);
282 got_it:
283 	/* Use the locked dnode page and inode */
284 	nid = le32_to_cpu(sum.nid);
285 	if (dn->inode->i_ino == nid) {
286 		struct dnode_of_data tdn = *dn;
287 		tdn.nid = nid;
288 		tdn.node_page = dn->inode_page;
289 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
290 		truncate_data_blocks_range(&tdn, 1);
291 		return 0;
292 	} else if (dn->nid == nid) {
293 		struct dnode_of_data tdn = *dn;
294 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
295 		truncate_data_blocks_range(&tdn, 1);
296 		return 0;
297 	}
298 
299 	/* Get the node page */
300 	node_page = get_node_page(sbi, nid);
301 	if (IS_ERR(node_page))
302 		return PTR_ERR(node_page);
303 
304 	offset = ofs_of_node(node_page);
305 	ino = ino_of_node(node_page);
306 	f2fs_put_page(node_page, 1);
307 
308 	if (ino != dn->inode->i_ino) {
309 		/* Deallocate previous index in the node page */
310 		inode = f2fs_iget(sbi->sb, ino);
311 		if (IS_ERR(inode))
312 			return PTR_ERR(inode);
313 	} else {
314 		inode = dn->inode;
315 	}
316 
317 	bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
318 			le16_to_cpu(sum.ofs_in_node);
319 
320 	if (ino != dn->inode->i_ino) {
321 		truncate_hole(inode, bidx, bidx + 1);
322 		iput(inode);
323 	} else {
324 		struct dnode_of_data tdn;
325 		set_new_dnode(&tdn, inode, dn->inode_page, NULL, 0);
326 		if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
327 			return 0;
328 		if (tdn.data_blkaddr != NULL_ADDR)
329 			truncate_data_blocks_range(&tdn, 1);
330 		f2fs_put_page(tdn.node_page, 1);
331 	}
332 	return 0;
333 }
334 
335 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
336 					struct page *page, block_t blkaddr)
337 {
338 	struct f2fs_inode_info *fi = F2FS_I(inode);
339 	unsigned int start, end;
340 	struct dnode_of_data dn;
341 	struct f2fs_summary sum;
342 	struct node_info ni;
343 	int err = 0, recovered = 0;
344 
345 	/* step 1: recover xattr */
346 	if (IS_INODE(page)) {
347 		recover_inline_xattr(inode, page);
348 	} else if (f2fs_has_xattr_block(ofs_of_node(page))) {
349 		recover_xattr_data(inode, page, blkaddr);
350 		goto out;
351 	}
352 
353 	/* step 2: recover inline data */
354 	if (recover_inline_data(inode, page))
355 		goto out;
356 
357 	/* step 3: recover data indices */
358 	start = start_bidx_of_node(ofs_of_node(page), fi);
359 	end = start + ADDRS_PER_PAGE(page, fi);
360 
361 	f2fs_lock_op(sbi);
362 
363 	set_new_dnode(&dn, inode, NULL, NULL, 0);
364 
365 	err = get_dnode_of_data(&dn, start, ALLOC_NODE);
366 	if (err) {
367 		f2fs_unlock_op(sbi);
368 		goto out;
369 	}
370 
371 	f2fs_wait_on_page_writeback(dn.node_page, NODE);
372 
373 	get_node_info(sbi, dn.nid, &ni);
374 	f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
375 	f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
376 
377 	for (; start < end; start++) {
378 		block_t src, dest;
379 
380 		src = datablock_addr(dn.node_page, dn.ofs_in_node);
381 		dest = datablock_addr(page, dn.ofs_in_node);
382 
383 		if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
384 			if (src == NULL_ADDR) {
385 				err = reserve_new_block(&dn);
386 				/* We should not get -ENOSPC */
387 				f2fs_bug_on(sbi, err);
388 			}
389 
390 			/* Check the previous node page having this index */
391 			err = check_index_in_prev_nodes(sbi, dest, &dn);
392 			if (err)
393 				goto err;
394 
395 			set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
396 
397 			/* write dummy data page */
398 			recover_data_page(sbi, NULL, &sum, src, dest);
399 			update_extent_cache(dest, &dn);
400 			recovered++;
401 		}
402 		dn.ofs_in_node++;
403 	}
404 
405 	/* write node page in place */
406 	set_summary(&sum, dn.nid, 0, 0);
407 	if (IS_INODE(dn.node_page))
408 		sync_inode_page(&dn);
409 
410 	copy_node_footer(dn.node_page, page);
411 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
412 					ofs_of_node(page), false);
413 	set_page_dirty(dn.node_page);
414 err:
415 	f2fs_put_dnode(&dn);
416 	f2fs_unlock_op(sbi);
417 out:
418 	f2fs_msg(sbi->sb, KERN_NOTICE,
419 		"recover_data: ino = %lx, recovered = %d blocks, err = %d",
420 		inode->i_ino, recovered, err);
421 	return err;
422 }
423 
424 static int recover_data(struct f2fs_sb_info *sbi,
425 				struct list_head *head, int type)
426 {
427 	unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
428 	struct curseg_info *curseg;
429 	struct page *page = NULL;
430 	int err = 0;
431 	block_t blkaddr;
432 
433 	/* get node pages in the current segment */
434 	curseg = CURSEG_I(sbi, type);
435 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
436 
437 	while (1) {
438 		struct fsync_inode_entry *entry;
439 
440 		if (blkaddr < MAIN_BLKADDR(sbi) || blkaddr >= MAX_BLKADDR(sbi))
441 			break;
442 
443 		ra_meta_pages_cond(sbi, blkaddr);
444 
445 		page = get_meta_page(sbi, blkaddr);
446 
447 		if (cp_ver != cpver_of_node(page)) {
448 			f2fs_put_page(page, 1);
449 			break;
450 		}
451 
452 		entry = get_fsync_inode(head, ino_of_node(page));
453 		if (!entry)
454 			goto next;
455 		/*
456 		 * inode(x) | CP | inode(x) | dnode(F)
457 		 * In this case, we can lose the latest inode(x).
458 		 * So, call recover_inode for the inode update.
459 		 */
460 		if (entry->last_inode == blkaddr)
461 			recover_inode(entry->inode, page);
462 		if (entry->last_dentry == blkaddr) {
463 			err = recover_dentry(entry->inode, page);
464 			if (err) {
465 				f2fs_put_page(page, 1);
466 				break;
467 			}
468 		}
469 		err = do_recover_data(sbi, entry->inode, page, blkaddr);
470 		if (err) {
471 			f2fs_put_page(page, 1);
472 			break;
473 		}
474 
475 		if (entry->blkaddr == blkaddr) {
476 			iput(entry->inode);
477 			list_del(&entry->list);
478 			kmem_cache_free(fsync_entry_slab, entry);
479 		}
480 next:
481 		/* check next segment */
482 		blkaddr = next_blkaddr_of_node(page);
483 		f2fs_put_page(page, 1);
484 	}
485 	if (!err)
486 		allocate_new_segments(sbi);
487 	return err;
488 }
489 
490 int recover_fsync_data(struct f2fs_sb_info *sbi)
491 {
492 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
493 	struct list_head inode_list;
494 	block_t blkaddr;
495 	int err;
496 	bool need_writecp = false;
497 
498 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
499 			sizeof(struct fsync_inode_entry));
500 	if (!fsync_entry_slab)
501 		return -ENOMEM;
502 
503 	INIT_LIST_HEAD(&inode_list);
504 
505 	/* step #1: find fsynced inode numbers */
506 	sbi->por_doing = true;
507 
508 	/* prevent checkpoint */
509 	mutex_lock(&sbi->cp_mutex);
510 
511 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
512 
513 	err = find_fsync_dnodes(sbi, &inode_list);
514 	if (err)
515 		goto out;
516 
517 	if (list_empty(&inode_list))
518 		goto out;
519 
520 	need_writecp = true;
521 
522 	/* step #2: recover data */
523 	err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
524 	if (!err)
525 		f2fs_bug_on(sbi, !list_empty(&inode_list));
526 out:
527 	destroy_fsync_dnodes(&inode_list);
528 	kmem_cache_destroy(fsync_entry_slab);
529 
530 	/* truncate meta pages to be used by the recovery */
531 	truncate_inode_pages_range(META_MAPPING(sbi),
532 			MAIN_BLKADDR(sbi) << PAGE_CACHE_SHIFT, -1);
533 
534 	if (err) {
535 		truncate_inode_pages_final(NODE_MAPPING(sbi));
536 		truncate_inode_pages_final(META_MAPPING(sbi));
537 	}
538 
539 	sbi->por_doing = false;
540 	if (err) {
541 		discard_next_dnode(sbi, blkaddr);
542 
543 		/* Flush all the NAT/SIT pages */
544 		while (get_pages(sbi, F2FS_DIRTY_META))
545 			sync_meta_pages(sbi, META, LONG_MAX);
546 		set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
547 		mutex_unlock(&sbi->cp_mutex);
548 	} else if (need_writecp) {
549 		struct cp_control cpc = {
550 			.reason = CP_SYNC,
551 		};
552 		mutex_unlock(&sbi->cp_mutex);
553 		write_checkpoint(sbi, &cpc);
554 	} else {
555 		mutex_unlock(&sbi->cp_mutex);
556 	}
557 	return err;
558 }
559