xref: /openbmc/linux/fs/f2fs/recovery.c (revision 5bd8e16d)
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 list_head *this;
31 	struct fsync_inode_entry *entry;
32 
33 	list_for_each(this, head) {
34 		entry = list_entry(this, struct fsync_inode_entry, list);
35 		if (entry->inode->i_ino == ino)
36 			return entry;
37 	}
38 	return NULL;
39 }
40 
41 static int recover_dentry(struct page *ipage, struct inode *inode)
42 {
43 	struct f2fs_node *raw_node = F2FS_NODE(ipage);
44 	struct f2fs_inode *raw_inode = &(raw_node->i);
45 	nid_t pino = le32_to_cpu(raw_inode->i_pino);
46 	struct f2fs_dir_entry *de;
47 	struct qstr name;
48 	struct page *page;
49 	struct inode *dir, *einode;
50 	int err = 0;
51 
52 	dir = check_dirty_dir_inode(F2FS_SB(inode->i_sb), pino);
53 	if (!dir) {
54 		dir = f2fs_iget(inode->i_sb, pino);
55 		if (IS_ERR(dir)) {
56 			err = PTR_ERR(dir);
57 			goto out;
58 		}
59 		set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
60 		add_dirty_dir_inode(dir);
61 	}
62 
63 	name.len = le32_to_cpu(raw_inode->i_namelen);
64 	name.name = raw_inode->i_name;
65 retry:
66 	de = f2fs_find_entry(dir, &name, &page);
67 	if (de && inode->i_ino == le32_to_cpu(de->ino)) {
68 		kunmap(page);
69 		f2fs_put_page(page, 0);
70 		goto out;
71 	}
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;
79 		}
80 		f2fs_delete_entry(de, page, einode);
81 		iput(einode);
82 		goto retry;
83 	}
84 	err = __f2fs_add_link(dir, &name, inode);
85 out:
86 	f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode and its dentry: "
87 			"ino = %x, name = %s, dir = %lx, err = %d",
88 			ino_of_node(ipage), raw_inode->i_name,
89 			IS_ERR(dir) ? 0 : dir->i_ino, err);
90 	return err;
91 }
92 
93 static int recover_inode(struct inode *inode, struct page *node_page)
94 {
95 	struct f2fs_node *raw_node = F2FS_NODE(node_page);
96 	struct f2fs_inode *raw_inode = &(raw_node->i);
97 
98 	if (!IS_INODE(node_page))
99 		return 0;
100 
101 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
102 	i_size_write(inode, le64_to_cpu(raw_inode->i_size));
103 	inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
104 	inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
105 	inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
106 	inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
107 	inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
108 	inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
109 
110 	if (is_dent_dnode(node_page))
111 		return recover_dentry(node_page, inode);
112 
113 	f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
114 			ino_of_node(node_page), raw_inode->i_name);
115 	return 0;
116 }
117 
118 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
119 {
120 	unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
121 	struct curseg_info *curseg;
122 	struct page *page;
123 	block_t blkaddr;
124 	int err = 0;
125 
126 	/* get node pages in the current segment */
127 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
128 	blkaddr = START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff;
129 
130 	/* read node page */
131 	page = alloc_page(GFP_F2FS_ZERO);
132 	if (!page)
133 		return -ENOMEM;
134 	lock_page(page);
135 
136 	while (1) {
137 		struct fsync_inode_entry *entry;
138 
139 		err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC);
140 		if (err)
141 			goto out;
142 
143 		lock_page(page);
144 
145 		if (cp_ver != cpver_of_node(page))
146 			break;
147 
148 		if (!is_fsync_dnode(page))
149 			goto next;
150 
151 		entry = get_fsync_inode(head, ino_of_node(page));
152 		if (entry) {
153 			if (IS_INODE(page) && is_dent_dnode(page))
154 				set_inode_flag(F2FS_I(entry->inode),
155 							FI_INC_LINK);
156 		} else {
157 			if (IS_INODE(page) && is_dent_dnode(page)) {
158 				err = recover_inode_page(sbi, page);
159 				if (err)
160 					break;
161 			}
162 
163 			/* add this fsync inode to the list */
164 			entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS);
165 			if (!entry) {
166 				err = -ENOMEM;
167 				break;
168 			}
169 
170 			entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
171 			if (IS_ERR(entry->inode)) {
172 				err = PTR_ERR(entry->inode);
173 				kmem_cache_free(fsync_entry_slab, entry);
174 				break;
175 			}
176 			list_add_tail(&entry->list, head);
177 		}
178 		entry->blkaddr = blkaddr;
179 
180 		err = recover_inode(entry->inode, page);
181 		if (err && err != -ENOENT)
182 			break;
183 next:
184 		/* check next segment */
185 		blkaddr = next_blkaddr_of_node(page);
186 	}
187 	unlock_page(page);
188 out:
189 	__free_pages(page, 0);
190 	return err;
191 }
192 
193 static void destroy_fsync_dnodes(struct list_head *head)
194 {
195 	struct fsync_inode_entry *entry, *tmp;
196 
197 	list_for_each_entry_safe(entry, tmp, head, list) {
198 		iput(entry->inode);
199 		list_del(&entry->list);
200 		kmem_cache_free(fsync_entry_slab, entry);
201 	}
202 }
203 
204 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
205 			block_t blkaddr, struct dnode_of_data *dn)
206 {
207 	struct seg_entry *sentry;
208 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
209 	unsigned short blkoff = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) &
210 					(sbi->blocks_per_seg - 1);
211 	struct f2fs_summary sum;
212 	nid_t ino, nid;
213 	void *kaddr;
214 	struct inode *inode;
215 	struct page *node_page;
216 	unsigned int offset;
217 	block_t bidx;
218 	int i;
219 
220 	sentry = get_seg_entry(sbi, segno);
221 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
222 		return 0;
223 
224 	/* Get the previous summary */
225 	for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
226 		struct curseg_info *curseg = CURSEG_I(sbi, i);
227 		if (curseg->segno == segno) {
228 			sum = curseg->sum_blk->entries[blkoff];
229 			break;
230 		}
231 	}
232 	if (i > CURSEG_COLD_DATA) {
233 		struct page *sum_page = get_sum_page(sbi, segno);
234 		struct f2fs_summary_block *sum_node;
235 		kaddr = page_address(sum_page);
236 		sum_node = (struct f2fs_summary_block *)kaddr;
237 		sum = sum_node->entries[blkoff];
238 		f2fs_put_page(sum_page, 1);
239 	}
240 
241 	/* Use the locked dnode page and inode */
242 	nid = le32_to_cpu(sum.nid);
243 	if (dn->inode->i_ino == nid) {
244 		struct dnode_of_data tdn = *dn;
245 		tdn.nid = nid;
246 		tdn.node_page = dn->inode_page;
247 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
248 		truncate_data_blocks_range(&tdn, 1);
249 		return 0;
250 	} else if (dn->nid == nid) {
251 		struct dnode_of_data tdn = *dn;
252 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
253 		truncate_data_blocks_range(&tdn, 1);
254 		return 0;
255 	}
256 
257 	/* Get the node page */
258 	node_page = get_node_page(sbi, nid);
259 	if (IS_ERR(node_page))
260 		return PTR_ERR(node_page);
261 
262 	offset = ofs_of_node(node_page);
263 	ino = ino_of_node(node_page);
264 	f2fs_put_page(node_page, 1);
265 
266 	/* Deallocate previous index in the node page */
267 	inode = f2fs_iget(sbi->sb, ino);
268 	if (IS_ERR(inode))
269 		return PTR_ERR(inode);
270 
271 	bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
272 					le16_to_cpu(sum.ofs_in_node);
273 
274 	truncate_hole(inode, bidx, bidx + 1);
275 	iput(inode);
276 	return 0;
277 }
278 
279 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
280 					struct page *page, block_t blkaddr)
281 {
282 	struct f2fs_inode_info *fi = F2FS_I(inode);
283 	unsigned int start, end;
284 	struct dnode_of_data dn;
285 	struct f2fs_summary sum;
286 	struct node_info ni;
287 	int err = 0, recovered = 0;
288 	int ilock;
289 
290 	start = start_bidx_of_node(ofs_of_node(page), fi);
291 	if (IS_INODE(page))
292 		end = start + ADDRS_PER_INODE(fi);
293 	else
294 		end = start + ADDRS_PER_BLOCK;
295 
296 	ilock = mutex_lock_op(sbi);
297 	set_new_dnode(&dn, inode, NULL, NULL, 0);
298 
299 	err = get_dnode_of_data(&dn, start, ALLOC_NODE);
300 	if (err) {
301 		mutex_unlock_op(sbi, ilock);
302 		return err;
303 	}
304 
305 	wait_on_page_writeback(dn.node_page);
306 
307 	get_node_info(sbi, dn.nid, &ni);
308 	BUG_ON(ni.ino != ino_of_node(page));
309 	BUG_ON(ofs_of_node(dn.node_page) != ofs_of_node(page));
310 
311 	for (; start < end; start++) {
312 		block_t src, dest;
313 
314 		src = datablock_addr(dn.node_page, dn.ofs_in_node);
315 		dest = datablock_addr(page, dn.ofs_in_node);
316 
317 		if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
318 			if (src == NULL_ADDR) {
319 				int err = reserve_new_block(&dn);
320 				/* We should not get -ENOSPC */
321 				BUG_ON(err);
322 			}
323 
324 			/* Check the previous node page having this index */
325 			err = check_index_in_prev_nodes(sbi, dest, &dn);
326 			if (err)
327 				goto err;
328 
329 			set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
330 
331 			/* write dummy data page */
332 			recover_data_page(sbi, NULL, &sum, src, dest);
333 			update_extent_cache(dest, &dn);
334 			recovered++;
335 		}
336 		dn.ofs_in_node++;
337 	}
338 
339 	/* write node page in place */
340 	set_summary(&sum, dn.nid, 0, 0);
341 	if (IS_INODE(dn.node_page))
342 		sync_inode_page(&dn);
343 
344 	copy_node_footer(dn.node_page, page);
345 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
346 					ofs_of_node(page), false);
347 	set_page_dirty(dn.node_page);
348 
349 	recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr);
350 err:
351 	f2fs_put_dnode(&dn);
352 	mutex_unlock_op(sbi, ilock);
353 
354 	f2fs_msg(sbi->sb, KERN_NOTICE, "recover_data: ino = %lx, "
355 			"recovered_data = %d blocks, err = %d",
356 			inode->i_ino, recovered, err);
357 	return err;
358 }
359 
360 static int recover_data(struct f2fs_sb_info *sbi,
361 				struct list_head *head, int type)
362 {
363 	unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
364 	struct curseg_info *curseg;
365 	struct page *page;
366 	int err = 0;
367 	block_t blkaddr;
368 
369 	/* get node pages in the current segment */
370 	curseg = CURSEG_I(sbi, type);
371 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
372 
373 	/* read node page */
374 	page = alloc_page(GFP_NOFS | __GFP_ZERO);
375 	if (!page)
376 		return -ENOMEM;
377 
378 	lock_page(page);
379 
380 	while (1) {
381 		struct fsync_inode_entry *entry;
382 
383 		err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC);
384 		if (err)
385 			goto out;
386 
387 		lock_page(page);
388 
389 		if (cp_ver != cpver_of_node(page))
390 			break;
391 
392 		entry = get_fsync_inode(head, ino_of_node(page));
393 		if (!entry)
394 			goto next;
395 
396 		err = do_recover_data(sbi, entry->inode, page, blkaddr);
397 		if (err)
398 			break;
399 
400 		if (entry->blkaddr == blkaddr) {
401 			iput(entry->inode);
402 			list_del(&entry->list);
403 			kmem_cache_free(fsync_entry_slab, entry);
404 		}
405 next:
406 		/* check next segment */
407 		blkaddr = next_blkaddr_of_node(page);
408 	}
409 	unlock_page(page);
410 out:
411 	__free_pages(page, 0);
412 
413 	if (!err)
414 		allocate_new_segments(sbi);
415 	return err;
416 }
417 
418 int recover_fsync_data(struct f2fs_sb_info *sbi)
419 {
420 	struct list_head inode_list;
421 	int err;
422 
423 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
424 			sizeof(struct fsync_inode_entry), NULL);
425 	if (unlikely(!fsync_entry_slab))
426 		return -ENOMEM;
427 
428 	INIT_LIST_HEAD(&inode_list);
429 
430 	/* step #1: find fsynced inode numbers */
431 	sbi->por_doing = 1;
432 	err = find_fsync_dnodes(sbi, &inode_list);
433 	if (err)
434 		goto out;
435 
436 	if (list_empty(&inode_list))
437 		goto out;
438 
439 	/* step #2: recover data */
440 	err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
441 	BUG_ON(!list_empty(&inode_list));
442 out:
443 	destroy_fsync_dnodes(&inode_list);
444 	kmem_cache_destroy(fsync_entry_slab);
445 	sbi->por_doing = 0;
446 	if (!err)
447 		write_checkpoint(sbi, false);
448 	return err;
449 }
450