xref: /openbmc/linux/fs/f2fs/recovery.c (revision 0199e993) 
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, 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 	kunmap(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 	while (1) {
174 		struct fsync_inode_entry *entry;
175 
176 		if (blkaddr < MAIN_BLKADDR(sbi) || blkaddr >= MAX_BLKADDR(sbi))
177 			return 0;
178 
179 		page = get_meta_page_ra(sbi, blkaddr);
180 
181 		if (cp_ver != cpver_of_node(page))
182 			break;
183 
184 		if (!is_fsync_dnode(page))
185 			goto next;
186 
187 		entry = get_fsync_inode(head, ino_of_node(page));
188 		if (entry) {
189 			if (IS_INODE(page) && is_dent_dnode(page))
190 				set_inode_flag(F2FS_I(entry->inode),
191 							FI_INC_LINK);
192 		} else {
193 			if (IS_INODE(page) && is_dent_dnode(page)) {
194 				err = recover_inode_page(sbi, page);
195 				if (err)
196 					break;
197 			}
198 
199 			/* add this fsync inode to the list */
200 			entry = kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
201 			if (!entry) {
202 				err = -ENOMEM;
203 				break;
204 			}
205 			/*
206 			 * CP | dnode(F) | inode(DF)
207 			 * For this case, we should not give up now.
208 			 */
209 			entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
210 			if (IS_ERR(entry->inode)) {
211 				err = PTR_ERR(entry->inode);
212 				kmem_cache_free(fsync_entry_slab, entry);
213 				if (err == -ENOENT)
214 					goto next;
215 				break;
216 			}
217 			list_add_tail(&entry->list, head);
218 		}
219 		entry->blkaddr = blkaddr;
220 
221 		if (IS_INODE(page)) {
222 			entry->last_inode = blkaddr;
223 			if (is_dent_dnode(page))
224 				entry->last_dentry = blkaddr;
225 		}
226 next:
227 		/* check next segment */
228 		blkaddr = next_blkaddr_of_node(page);
229 		f2fs_put_page(page, 1);
230 	}
231 	f2fs_put_page(page, 1);
232 	return err;
233 }
234 
235 static void destroy_fsync_dnodes(struct list_head *head)
236 {
237 	struct fsync_inode_entry *entry, *tmp;
238 
239 	list_for_each_entry_safe(entry, tmp, head, list) {
240 		iput(entry->inode);
241 		list_del(&entry->list);
242 		kmem_cache_free(fsync_entry_slab, entry);
243 	}
244 }
245 
246 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
247 			block_t blkaddr, struct dnode_of_data *dn)
248 {
249 	struct seg_entry *sentry;
250 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
251 	unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
252 	struct f2fs_summary_block *sum_node;
253 	struct f2fs_summary sum;
254 	struct page *sum_page, *node_page;
255 	nid_t ino, nid;
256 	struct inode *inode;
257 	unsigned int offset;
258 	block_t bidx;
259 	int i;
260 
261 	sentry = get_seg_entry(sbi, segno);
262 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
263 		return 0;
264 
265 	/* Get the previous summary */
266 	for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
267 		struct curseg_info *curseg = CURSEG_I(sbi, i);
268 		if (curseg->segno == segno) {
269 			sum = curseg->sum_blk->entries[blkoff];
270 			goto got_it;
271 		}
272 	}
273 
274 	sum_page = get_sum_page(sbi, segno);
275 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
276 	sum = sum_node->entries[blkoff];
277 	f2fs_put_page(sum_page, 1);
278 got_it:
279 	/* Use the locked dnode page and inode */
280 	nid = le32_to_cpu(sum.nid);
281 	if (dn->inode->i_ino == nid) {
282 		struct dnode_of_data tdn = *dn;
283 		tdn.nid = nid;
284 		tdn.node_page = dn->inode_page;
285 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
286 		truncate_data_blocks_range(&tdn, 1);
287 		return 0;
288 	} else if (dn->nid == nid) {
289 		struct dnode_of_data tdn = *dn;
290 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
291 		truncate_data_blocks_range(&tdn, 1);
292 		return 0;
293 	}
294 
295 	/* Get the node page */
296 	node_page = get_node_page(sbi, nid);
297 	if (IS_ERR(node_page))
298 		return PTR_ERR(node_page);
299 
300 	offset = ofs_of_node(node_page);
301 	ino = ino_of_node(node_page);
302 	f2fs_put_page(node_page, 1);
303 
304 	if (ino != dn->inode->i_ino) {
305 		/* Deallocate previous index in the node page */
306 		inode = f2fs_iget(sbi->sb, ino);
307 		if (IS_ERR(inode))
308 			return PTR_ERR(inode);
309 	} else {
310 		inode = dn->inode;
311 	}
312 
313 	bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
314 			le16_to_cpu(sum.ofs_in_node);
315 
316 	if (ino != dn->inode->i_ino) {
317 		truncate_hole(inode, bidx, bidx + 1);
318 		iput(inode);
319 	} else {
320 		struct dnode_of_data tdn;
321 		set_new_dnode(&tdn, inode, dn->inode_page, NULL, 0);
322 		if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
323 			return 0;
324 		if (tdn.data_blkaddr != NULL_ADDR)
325 			truncate_data_blocks_range(&tdn, 1);
326 		f2fs_put_page(tdn.node_page, 1);
327 	}
328 	return 0;
329 }
330 
331 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
332 					struct page *page, block_t blkaddr)
333 {
334 	struct f2fs_inode_info *fi = F2FS_I(inode);
335 	unsigned int start, end;
336 	struct dnode_of_data dn;
337 	struct f2fs_summary sum;
338 	struct node_info ni;
339 	int err = 0, recovered = 0;
340 
341 	/* step 1: recover xattr */
342 	if (IS_INODE(page)) {
343 		recover_inline_xattr(inode, page);
344 	} else if (f2fs_has_xattr_block(ofs_of_node(page))) {
345 		recover_xattr_data(inode, page, blkaddr);
346 		goto out;
347 	}
348 
349 	/* step 2: recover inline data */
350 	if (recover_inline_data(inode, page))
351 		goto out;
352 
353 	/* step 3: recover data indices */
354 	start = start_bidx_of_node(ofs_of_node(page), fi);
355 	end = start + ADDRS_PER_PAGE(page, fi);
356 
357 	f2fs_lock_op(sbi);
358 
359 	set_new_dnode(&dn, inode, NULL, NULL, 0);
360 
361 	err = get_dnode_of_data(&dn, start, ALLOC_NODE);
362 	if (err) {
363 		f2fs_unlock_op(sbi);
364 		goto out;
365 	}
366 
367 	f2fs_wait_on_page_writeback(dn.node_page, NODE);
368 
369 	get_node_info(sbi, dn.nid, &ni);
370 	f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
371 	f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
372 
373 	for (; start < end; start++) {
374 		block_t src, dest;
375 
376 		src = datablock_addr(dn.node_page, dn.ofs_in_node);
377 		dest = datablock_addr(page, dn.ofs_in_node);
378 
379 		if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
380 			if (src == NULL_ADDR) {
381 				err = reserve_new_block(&dn);
382 				/* We should not get -ENOSPC */
383 				f2fs_bug_on(sbi, err);
384 			}
385 
386 			/* Check the previous node page having this index */
387 			err = check_index_in_prev_nodes(sbi, dest, &dn);
388 			if (err)
389 				goto err;
390 
391 			set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
392 
393 			/* write dummy data page */
394 			recover_data_page(sbi, NULL, &sum, src, dest);
395 			update_extent_cache(dest, &dn);
396 			recovered++;
397 		}
398 		dn.ofs_in_node++;
399 	}
400 
401 	/* write node page in place */
402 	set_summary(&sum, dn.nid, 0, 0);
403 	if (IS_INODE(dn.node_page))
404 		sync_inode_page(&dn);
405 
406 	copy_node_footer(dn.node_page, page);
407 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
408 					ofs_of_node(page), false);
409 	set_page_dirty(dn.node_page);
410 err:
411 	f2fs_put_dnode(&dn);
412 	f2fs_unlock_op(sbi);
413 out:
414 	f2fs_msg(sbi->sb, KERN_NOTICE,
415 		"recover_data: ino = %lx, recovered = %d blocks, err = %d",
416 		inode->i_ino, recovered, err);
417 	return err;
418 }
419 
420 static int recover_data(struct f2fs_sb_info *sbi,
421 				struct list_head *head, int type)
422 {
423 	unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
424 	struct curseg_info *curseg;
425 	struct page *page = NULL;
426 	int err = 0;
427 	block_t blkaddr;
428 
429 	/* get node pages in the current segment */
430 	curseg = CURSEG_I(sbi, type);
431 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
432 
433 	while (1) {
434 		struct fsync_inode_entry *entry;
435 
436 		if (blkaddr < MAIN_BLKADDR(sbi) || blkaddr >= MAX_BLKADDR(sbi))
437 			break;
438 
439 		page = get_meta_page_ra(sbi, blkaddr);
440 
441 		if (cp_ver != cpver_of_node(page)) {
442 			f2fs_put_page(page, 1);
443 			break;
444 		}
445 
446 		entry = get_fsync_inode(head, ino_of_node(page));
447 		if (!entry)
448 			goto next;
449 		/*
450 		 * inode(x) | CP | inode(x) | dnode(F)
451 		 * In this case, we can lose the latest inode(x).
452 		 * So, call recover_inode for the inode update.
453 		 */
454 		if (entry->last_inode == blkaddr)
455 			recover_inode(entry->inode, page);
456 		if (entry->last_dentry == blkaddr) {
457 			err = recover_dentry(entry->inode, page);
458 			if (err) {
459 				f2fs_put_page(page, 1);
460 				break;
461 			}
462 		}
463 		err = do_recover_data(sbi, entry->inode, page, blkaddr);
464 		if (err) {
465 			f2fs_put_page(page, 1);
466 			break;
467 		}
468 
469 		if (entry->blkaddr == blkaddr) {
470 			iput(entry->inode);
471 			list_del(&entry->list);
472 			kmem_cache_free(fsync_entry_slab, entry);
473 		}
474 next:
475 		/* check next segment */
476 		blkaddr = next_blkaddr_of_node(page);
477 		f2fs_put_page(page, 1);
478 	}
479 	if (!err)
480 		allocate_new_segments(sbi);
481 	return err;
482 }
483 
484 int recover_fsync_data(struct f2fs_sb_info *sbi)
485 {
486 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
487 	struct list_head inode_list;
488 	block_t blkaddr;
489 	int err;
490 	bool need_writecp = false;
491 
492 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
493 			sizeof(struct fsync_inode_entry));
494 	if (!fsync_entry_slab)
495 		return -ENOMEM;
496 
497 	INIT_LIST_HEAD(&inode_list);
498 
499 	/* step #1: find fsynced inode numbers */
500 	sbi->por_doing = true;
501 
502 	/* prevent checkpoint */
503 	mutex_lock(&sbi->cp_mutex);
504 
505 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
506 
507 	err = find_fsync_dnodes(sbi, &inode_list);
508 	if (err)
509 		goto out;
510 
511 	if (list_empty(&inode_list))
512 		goto out;
513 
514 	need_writecp = true;
515 
516 	/* step #2: recover data */
517 	err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
518 	if (!err)
519 		f2fs_bug_on(sbi, !list_empty(&inode_list));
520 out:
521 	destroy_fsync_dnodes(&inode_list);
522 	kmem_cache_destroy(fsync_entry_slab);
523 
524 	/* truncate meta pages to be used by the recovery */
525 	truncate_inode_pages_range(META_MAPPING(sbi),
526 			MAIN_BLKADDR(sbi) << PAGE_CACHE_SHIFT, -1);
527 
528 	if (err) {
529 		truncate_inode_pages_final(NODE_MAPPING(sbi));
530 		truncate_inode_pages_final(META_MAPPING(sbi));
531 	}
532 
533 	sbi->por_doing = false;
534 	if (err) {
535 		discard_next_dnode(sbi, blkaddr);
536 
537 		/* Flush all the NAT/SIT pages */
538 		while (get_pages(sbi, F2FS_DIRTY_META))
539 			sync_meta_pages(sbi, META, LONG_MAX);
540 		set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
541 		mutex_unlock(&sbi->cp_mutex);
542 	} else if (need_writecp) {
543 		struct cp_control cpc = {
544 			.reason = CP_SYNC,
545 		};
546 		mutex_unlock(&sbi->cp_mutex);
547 		write_checkpoint(sbi, &cpc);
548 	} else {
549 		mutex_unlock(&sbi->cp_mutex);
550 	}
551 	return err;
552 }
553