xref: /openbmc/linux/fs/f2fs/recovery.c (revision 1c2f87c2)
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 static struct kmem_cache *fsync_entry_slab;
18 
19 bool space_for_roll_forward(struct f2fs_sb_info *sbi)
20 {
21 	if (sbi->last_valid_block_count + sbi->alloc_valid_block_count
22 			> sbi->user_block_count)
23 		return false;
24 	return true;
25 }
26 
27 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
28 								nid_t ino)
29 {
30 	struct fsync_inode_entry *entry;
31 
32 	list_for_each_entry(entry, head, list)
33 		if (entry->inode->i_ino == ino)
34 			return entry;
35 
36 	return NULL;
37 }
38 
39 static int recover_dentry(struct page *ipage, struct inode *inode)
40 {
41 	struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
42 	nid_t pino = le32_to_cpu(raw_inode->i_pino);
43 	struct f2fs_dir_entry *de;
44 	struct qstr name;
45 	struct page *page;
46 	struct inode *dir, *einode;
47 	int err = 0;
48 
49 	dir = check_dirty_dir_inode(F2FS_SB(inode->i_sb), pino);
50 	if (!dir) {
51 		dir = f2fs_iget(inode->i_sb, pino);
52 		if (IS_ERR(dir)) {
53 			err = PTR_ERR(dir);
54 			goto out;
55 		}
56 		set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
57 		add_dirty_dir_inode(dir);
58 	}
59 
60 	name.len = le32_to_cpu(raw_inode->i_namelen);
61 	name.name = raw_inode->i_name;
62 
63 	if (unlikely(name.len > F2FS_NAME_LEN)) {
64 		WARN_ON(1);
65 		err = -ENAMETOOLONG;
66 		goto out;
67 	}
68 retry:
69 	de = f2fs_find_entry(dir, &name, &page);
70 	if (de && inode->i_ino == le32_to_cpu(de->ino))
71 		goto out_unmap_put;
72 	if (de) {
73 		einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
74 		if (IS_ERR(einode)) {
75 			WARN_ON(1);
76 			if (PTR_ERR(einode) == -ENOENT)
77 				err = -EEXIST;
78 			goto out_unmap_put;
79 		}
80 		err = acquire_orphan_inode(F2FS_SB(inode->i_sb));
81 		if (err) {
82 			iput(einode);
83 			goto out_unmap_put;
84 		}
85 		f2fs_delete_entry(de, page, einode);
86 		iput(einode);
87 		goto retry;
88 	}
89 	err = __f2fs_add_link(dir, &name, inode);
90 	goto out;
91 
92 out_unmap_put:
93 	kunmap(page);
94 	f2fs_put_page(page, 0);
95 out:
96 	f2fs_msg(inode->i_sb, KERN_NOTICE,
97 			"%s: ino = %x, name = %s, dir = %lx, err = %d",
98 			__func__, ino_of_node(ipage), raw_inode->i_name,
99 			IS_ERR(dir) ? 0 : dir->i_ino, err);
100 	return err;
101 }
102 
103 static int recover_inode(struct inode *inode, struct page *node_page)
104 {
105 	struct f2fs_inode *raw_inode = F2FS_INODE(node_page);
106 
107 	if (!IS_INODE(node_page))
108 		return 0;
109 
110 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
111 	i_size_write(inode, le64_to_cpu(raw_inode->i_size));
112 	inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
113 	inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
114 	inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
115 	inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
116 	inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
117 	inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
118 
119 	if (is_dent_dnode(node_page))
120 		return recover_dentry(node_page, inode);
121 
122 	f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
123 			ino_of_node(node_page), raw_inode->i_name);
124 	return 0;
125 }
126 
127 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
128 {
129 	unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
130 	struct curseg_info *curseg;
131 	struct page *page;
132 	block_t blkaddr;
133 	int err = 0;
134 
135 	/* get node pages in the current segment */
136 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
137 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
138 
139 	/* read node page */
140 	page = alloc_page(GFP_F2FS_ZERO);
141 	if (!page)
142 		return -ENOMEM;
143 	lock_page(page);
144 
145 	while (1) {
146 		struct fsync_inode_entry *entry;
147 
148 		err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC);
149 		if (err)
150 			return err;
151 
152 		lock_page(page);
153 
154 		if (cp_ver != cpver_of_node(page))
155 			break;
156 
157 		if (!is_fsync_dnode(page))
158 			goto next;
159 
160 		entry = get_fsync_inode(head, ino_of_node(page));
161 		if (entry) {
162 			if (IS_INODE(page) && is_dent_dnode(page))
163 				set_inode_flag(F2FS_I(entry->inode),
164 							FI_INC_LINK);
165 		} else {
166 			if (IS_INODE(page) && is_dent_dnode(page)) {
167 				err = recover_inode_page(sbi, page);
168 				if (err)
169 					break;
170 			}
171 
172 			/* add this fsync inode to the list */
173 			entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS);
174 			if (!entry) {
175 				err = -ENOMEM;
176 				break;
177 			}
178 
179 			entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
180 			if (IS_ERR(entry->inode)) {
181 				err = PTR_ERR(entry->inode);
182 				kmem_cache_free(fsync_entry_slab, entry);
183 				break;
184 			}
185 			list_add_tail(&entry->list, head);
186 		}
187 		entry->blkaddr = blkaddr;
188 
189 		err = recover_inode(entry->inode, page);
190 		if (err && err != -ENOENT)
191 			break;
192 next:
193 		/* check next segment */
194 		blkaddr = next_blkaddr_of_node(page);
195 	}
196 
197 	unlock_page(page);
198 	__free_pages(page, 0);
199 
200 	return err;
201 }
202 
203 static void destroy_fsync_dnodes(struct list_head *head)
204 {
205 	struct fsync_inode_entry *entry, *tmp;
206 
207 	list_for_each_entry_safe(entry, tmp, head, list) {
208 		iput(entry->inode);
209 		list_del(&entry->list);
210 		kmem_cache_free(fsync_entry_slab, entry);
211 	}
212 }
213 
214 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
215 			block_t blkaddr, struct dnode_of_data *dn)
216 {
217 	struct seg_entry *sentry;
218 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
219 	unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
220 	struct f2fs_summary_block *sum_node;
221 	struct f2fs_summary sum;
222 	struct page *sum_page, *node_page;
223 	nid_t ino, nid;
224 	struct inode *inode;
225 	unsigned int offset;
226 	block_t bidx;
227 	int i;
228 
229 	sentry = get_seg_entry(sbi, segno);
230 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
231 		return 0;
232 
233 	/* Get the previous summary */
234 	for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
235 		struct curseg_info *curseg = CURSEG_I(sbi, i);
236 		if (curseg->segno == segno) {
237 			sum = curseg->sum_blk->entries[blkoff];
238 			goto got_it;
239 		}
240 	}
241 
242 	sum_page = get_sum_page(sbi, segno);
243 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
244 	sum = sum_node->entries[blkoff];
245 	f2fs_put_page(sum_page, 1);
246 got_it:
247 	/* Use the locked dnode page and inode */
248 	nid = le32_to_cpu(sum.nid);
249 	if (dn->inode->i_ino == nid) {
250 		struct dnode_of_data tdn = *dn;
251 		tdn.nid = nid;
252 		tdn.node_page = dn->inode_page;
253 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
254 		truncate_data_blocks_range(&tdn, 1);
255 		return 0;
256 	} else if (dn->nid == nid) {
257 		struct dnode_of_data tdn = *dn;
258 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
259 		truncate_data_blocks_range(&tdn, 1);
260 		return 0;
261 	}
262 
263 	/* Get the node page */
264 	node_page = get_node_page(sbi, nid);
265 	if (IS_ERR(node_page))
266 		return PTR_ERR(node_page);
267 
268 	offset = ofs_of_node(node_page);
269 	ino = ino_of_node(node_page);
270 	f2fs_put_page(node_page, 1);
271 
272 	/* Deallocate previous index in the node page */
273 	inode = f2fs_iget(sbi->sb, ino);
274 	if (IS_ERR(inode))
275 		return PTR_ERR(inode);
276 
277 	bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
278 					le16_to_cpu(sum.ofs_in_node);
279 
280 	truncate_hole(inode, bidx, bidx + 1);
281 	iput(inode);
282 	return 0;
283 }
284 
285 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
286 					struct page *page, block_t blkaddr)
287 {
288 	struct f2fs_inode_info *fi = F2FS_I(inode);
289 	unsigned int start, end;
290 	struct dnode_of_data dn;
291 	struct f2fs_summary sum;
292 	struct node_info ni;
293 	int err = 0, recovered = 0;
294 
295 	if (recover_inline_data(inode, page))
296 		goto out;
297 
298 	if (recover_xattr_data(inode, page, blkaddr))
299 		goto out;
300 
301 	start = start_bidx_of_node(ofs_of_node(page), fi);
302 	if (IS_INODE(page))
303 		end = start + ADDRS_PER_INODE(fi);
304 	else
305 		end = start + ADDRS_PER_BLOCK;
306 
307 	f2fs_lock_op(sbi);
308 
309 	set_new_dnode(&dn, inode, NULL, NULL, 0);
310 
311 	err = get_dnode_of_data(&dn, start, ALLOC_NODE);
312 	if (err) {
313 		f2fs_unlock_op(sbi);
314 		goto out;
315 	}
316 
317 	f2fs_wait_on_page_writeback(dn.node_page, NODE);
318 
319 	get_node_info(sbi, dn.nid, &ni);
320 	f2fs_bug_on(ni.ino != ino_of_node(page));
321 	f2fs_bug_on(ofs_of_node(dn.node_page) != ofs_of_node(page));
322 
323 	for (; start < end; start++) {
324 		block_t src, dest;
325 
326 		src = datablock_addr(dn.node_page, dn.ofs_in_node);
327 		dest = datablock_addr(page, dn.ofs_in_node);
328 
329 		if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
330 			if (src == NULL_ADDR) {
331 				err = reserve_new_block(&dn);
332 				/* We should not get -ENOSPC */
333 				f2fs_bug_on(err);
334 			}
335 
336 			/* Check the previous node page having this index */
337 			err = check_index_in_prev_nodes(sbi, dest, &dn);
338 			if (err)
339 				goto err;
340 
341 			set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
342 
343 			/* write dummy data page */
344 			recover_data_page(sbi, NULL, &sum, src, dest);
345 			update_extent_cache(dest, &dn);
346 			recovered++;
347 		}
348 		dn.ofs_in_node++;
349 	}
350 
351 	/* write node page in place */
352 	set_summary(&sum, dn.nid, 0, 0);
353 	if (IS_INODE(dn.node_page))
354 		sync_inode_page(&dn);
355 
356 	copy_node_footer(dn.node_page, page);
357 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
358 					ofs_of_node(page), false);
359 	set_page_dirty(dn.node_page);
360 
361 	recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr);
362 err:
363 	f2fs_put_dnode(&dn);
364 	f2fs_unlock_op(sbi);
365 out:
366 	f2fs_msg(sbi->sb, KERN_NOTICE,
367 		"recover_data: ino = %lx, recovered = %d blocks, err = %d",
368 		inode->i_ino, recovered, err);
369 	return err;
370 }
371 
372 static int recover_data(struct f2fs_sb_info *sbi,
373 				struct list_head *head, int type)
374 {
375 	unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
376 	struct curseg_info *curseg;
377 	struct page *page;
378 	int err = 0;
379 	block_t blkaddr;
380 
381 	/* get node pages in the current segment */
382 	curseg = CURSEG_I(sbi, type);
383 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
384 
385 	/* read node page */
386 	page = alloc_page(GFP_F2FS_ZERO);
387 	if (!page)
388 		return -ENOMEM;
389 
390 	lock_page(page);
391 
392 	while (1) {
393 		struct fsync_inode_entry *entry;
394 
395 		err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC);
396 		if (err)
397 			return err;
398 
399 		lock_page(page);
400 
401 		if (cp_ver != cpver_of_node(page))
402 			break;
403 
404 		entry = get_fsync_inode(head, ino_of_node(page));
405 		if (!entry)
406 			goto next;
407 
408 		err = do_recover_data(sbi, entry->inode, page, blkaddr);
409 		if (err)
410 			break;
411 
412 		if (entry->blkaddr == blkaddr) {
413 			iput(entry->inode);
414 			list_del(&entry->list);
415 			kmem_cache_free(fsync_entry_slab, entry);
416 		}
417 next:
418 		/* check next segment */
419 		blkaddr = next_blkaddr_of_node(page);
420 	}
421 
422 	unlock_page(page);
423 	__free_pages(page, 0);
424 
425 	if (!err)
426 		allocate_new_segments(sbi);
427 	return err;
428 }
429 
430 int recover_fsync_data(struct f2fs_sb_info *sbi)
431 {
432 	struct list_head inode_list;
433 	int err;
434 	bool need_writecp = false;
435 
436 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
437 			sizeof(struct fsync_inode_entry));
438 	if (!fsync_entry_slab)
439 		return -ENOMEM;
440 
441 	INIT_LIST_HEAD(&inode_list);
442 
443 	/* step #1: find fsynced inode numbers */
444 	sbi->por_doing = true;
445 	err = find_fsync_dnodes(sbi, &inode_list);
446 	if (err)
447 		goto out;
448 
449 	if (list_empty(&inode_list))
450 		goto out;
451 
452 	need_writecp = true;
453 
454 	/* step #2: recover data */
455 	err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
456 	f2fs_bug_on(!list_empty(&inode_list));
457 out:
458 	destroy_fsync_dnodes(&inode_list);
459 	kmem_cache_destroy(fsync_entry_slab);
460 	sbi->por_doing = false;
461 	if (!err && need_writecp)
462 		write_checkpoint(sbi, false);
463 	return err;
464 }
465