xref: /openbmc/linux/fs/f2fs/recovery.c (revision 5f32c314)
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_inode *raw_inode = F2FS_INODE(ipage);
44 	nid_t pino = le32_to_cpu(raw_inode->i_pino);
45 	struct f2fs_dir_entry *de;
46 	struct qstr name;
47 	struct page *page;
48 	struct inode *dir, *einode;
49 	int err = 0;
50 
51 	dir = check_dirty_dir_inode(F2FS_SB(inode->i_sb), pino);
52 	if (!dir) {
53 		dir = f2fs_iget(inode->i_sb, pino);
54 		if (IS_ERR(dir)) {
55 			err = PTR_ERR(dir);
56 			goto out;
57 		}
58 		set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
59 		add_dirty_dir_inode(dir);
60 	}
61 
62 	name.len = le32_to_cpu(raw_inode->i_namelen);
63 	name.name = raw_inode->i_name;
64 
65 	if (unlikely(name.len > F2FS_NAME_LEN)) {
66 		WARN_ON(1);
67 		err = -ENAMETOOLONG;
68 		goto out;
69 	}
70 retry:
71 	de = f2fs_find_entry(dir, &name, &page);
72 	if (de && inode->i_ino == le32_to_cpu(de->ino))
73 		goto out_unmap_put;
74 	if (de) {
75 		einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
76 		if (IS_ERR(einode)) {
77 			WARN_ON(1);
78 			if (PTR_ERR(einode) == -ENOENT)
79 				err = -EEXIST;
80 			goto out_unmap_put;
81 		}
82 		err = acquire_orphan_inode(F2FS_SB(inode->i_sb));
83 		if (err) {
84 			iput(einode);
85 			goto out_unmap_put;
86 		}
87 		f2fs_delete_entry(de, page, einode);
88 		iput(einode);
89 		goto retry;
90 	}
91 	err = __f2fs_add_link(dir, &name, inode);
92 	goto out;
93 
94 out_unmap_put:
95 	kunmap(page);
96 	f2fs_put_page(page, 0);
97 out:
98 	f2fs_msg(inode->i_sb, KERN_NOTICE,
99 			"%s: ino = %x, name = %s, dir = %lx, err = %d",
100 			__func__, ino_of_node(ipage), raw_inode->i_name,
101 			IS_ERR(dir) ? 0 : dir->i_ino, err);
102 	return err;
103 }
104 
105 static int recover_inode(struct inode *inode, struct page *node_page)
106 {
107 	struct f2fs_inode *raw_inode = F2FS_INODE(node_page);
108 
109 	if (!IS_INODE(node_page))
110 		return 0;
111 
112 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
113 	i_size_write(inode, le64_to_cpu(raw_inode->i_size));
114 	inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
115 	inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
116 	inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
117 	inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
118 	inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
119 	inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
120 
121 	if (is_dent_dnode(node_page))
122 		return recover_dentry(node_page, inode);
123 
124 	f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
125 			ino_of_node(node_page), raw_inode->i_name);
126 	return 0;
127 }
128 
129 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
130 {
131 	unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
132 	struct curseg_info *curseg;
133 	struct page *page;
134 	block_t blkaddr;
135 	int err = 0;
136 
137 	/* get node pages in the current segment */
138 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
139 	blkaddr = START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff;
140 
141 	/* read node page */
142 	page = alloc_page(GFP_F2FS_ZERO);
143 	if (!page)
144 		return -ENOMEM;
145 	lock_page(page);
146 
147 	while (1) {
148 		struct fsync_inode_entry *entry;
149 
150 		err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC);
151 		if (err)
152 			return err;
153 
154 		lock_page(page);
155 
156 		if (cp_ver != cpver_of_node(page))
157 			break;
158 
159 		if (!is_fsync_dnode(page))
160 			goto next;
161 
162 		entry = get_fsync_inode(head, ino_of_node(page));
163 		if (entry) {
164 			if (IS_INODE(page) && is_dent_dnode(page))
165 				set_inode_flag(F2FS_I(entry->inode),
166 							FI_INC_LINK);
167 		} else {
168 			if (IS_INODE(page) && is_dent_dnode(page)) {
169 				err = recover_inode_page(sbi, page);
170 				if (err)
171 					break;
172 			}
173 
174 			/* add this fsync inode to the list */
175 			entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS);
176 			if (!entry) {
177 				err = -ENOMEM;
178 				break;
179 			}
180 
181 			entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
182 			if (IS_ERR(entry->inode)) {
183 				err = PTR_ERR(entry->inode);
184 				kmem_cache_free(fsync_entry_slab, entry);
185 				break;
186 			}
187 			list_add_tail(&entry->list, head);
188 		}
189 		entry->blkaddr = blkaddr;
190 
191 		err = recover_inode(entry->inode, page);
192 		if (err && err != -ENOENT)
193 			break;
194 next:
195 		/* check next segment */
196 		blkaddr = next_blkaddr_of_node(page);
197 	}
198 
199 	unlock_page(page);
200 	__free_pages(page, 0);
201 
202 	return err;
203 }
204 
205 static void destroy_fsync_dnodes(struct list_head *head)
206 {
207 	struct fsync_inode_entry *entry, *tmp;
208 
209 	list_for_each_entry_safe(entry, tmp, head, list) {
210 		iput(entry->inode);
211 		list_del(&entry->list);
212 		kmem_cache_free(fsync_entry_slab, entry);
213 	}
214 }
215 
216 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
217 			block_t blkaddr, struct dnode_of_data *dn)
218 {
219 	struct seg_entry *sentry;
220 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
221 	unsigned short blkoff = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) &
222 					(sbi->blocks_per_seg - 1);
223 	struct f2fs_summary sum;
224 	nid_t ino, nid;
225 	void *kaddr;
226 	struct inode *inode;
227 	struct page *node_page;
228 	unsigned int offset;
229 	block_t bidx;
230 	int i;
231 
232 	sentry = get_seg_entry(sbi, segno);
233 	if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
234 		return 0;
235 
236 	/* Get the previous summary */
237 	for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
238 		struct curseg_info *curseg = CURSEG_I(sbi, i);
239 		if (curseg->segno == segno) {
240 			sum = curseg->sum_blk->entries[blkoff];
241 			break;
242 		}
243 	}
244 	if (i > CURSEG_COLD_DATA) {
245 		struct page *sum_page = get_sum_page(sbi, segno);
246 		struct f2fs_summary_block *sum_node;
247 		kaddr = page_address(sum_page);
248 		sum_node = (struct f2fs_summary_block *)kaddr;
249 		sum = sum_node->entries[blkoff];
250 		f2fs_put_page(sum_page, 1);
251 	}
252 
253 	/* Use the locked dnode page and inode */
254 	nid = le32_to_cpu(sum.nid);
255 	if (dn->inode->i_ino == nid) {
256 		struct dnode_of_data tdn = *dn;
257 		tdn.nid = nid;
258 		tdn.node_page = dn->inode_page;
259 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
260 		truncate_data_blocks_range(&tdn, 1);
261 		return 0;
262 	} else if (dn->nid == nid) {
263 		struct dnode_of_data tdn = *dn;
264 		tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
265 		truncate_data_blocks_range(&tdn, 1);
266 		return 0;
267 	}
268 
269 	/* Get the node page */
270 	node_page = get_node_page(sbi, nid);
271 	if (IS_ERR(node_page))
272 		return PTR_ERR(node_page);
273 
274 	offset = ofs_of_node(node_page);
275 	ino = ino_of_node(node_page);
276 	f2fs_put_page(node_page, 1);
277 
278 	/* Deallocate previous index in the node page */
279 	inode = f2fs_iget(sbi->sb, ino);
280 	if (IS_ERR(inode))
281 		return PTR_ERR(inode);
282 
283 	bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
284 					le16_to_cpu(sum.ofs_in_node);
285 
286 	truncate_hole(inode, bidx, bidx + 1);
287 	iput(inode);
288 	return 0;
289 }
290 
291 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
292 					struct page *page, block_t blkaddr)
293 {
294 	struct f2fs_inode_info *fi = F2FS_I(inode);
295 	unsigned int start, end;
296 	struct dnode_of_data dn;
297 	struct f2fs_summary sum;
298 	struct node_info ni;
299 	int err = 0, recovered = 0;
300 
301 	if (recover_inline_data(inode, page))
302 		goto out;
303 
304 	start = start_bidx_of_node(ofs_of_node(page), fi);
305 	if (IS_INODE(page))
306 		end = start + ADDRS_PER_INODE(fi);
307 	else
308 		end = start + ADDRS_PER_BLOCK;
309 
310 	f2fs_lock_op(sbi);
311 
312 	set_new_dnode(&dn, inode, NULL, NULL, 0);
313 
314 	err = get_dnode_of_data(&dn, start, ALLOC_NODE);
315 	if (err) {
316 		f2fs_unlock_op(sbi);
317 		goto out;
318 	}
319 
320 	wait_on_page_writeback(dn.node_page);
321 
322 	get_node_info(sbi, dn.nid, &ni);
323 	f2fs_bug_on(ni.ino != ino_of_node(page));
324 	f2fs_bug_on(ofs_of_node(dn.node_page) != ofs_of_node(page));
325 
326 	for (; start < end; start++) {
327 		block_t src, dest;
328 
329 		src = datablock_addr(dn.node_page, dn.ofs_in_node);
330 		dest = datablock_addr(page, dn.ofs_in_node);
331 
332 		if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
333 			if (src == NULL_ADDR) {
334 				err = reserve_new_block(&dn);
335 				/* We should not get -ENOSPC */
336 				f2fs_bug_on(err);
337 			}
338 
339 			/* Check the previous node page having this index */
340 			err = check_index_in_prev_nodes(sbi, dest, &dn);
341 			if (err)
342 				goto err;
343 
344 			set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
345 
346 			/* write dummy data page */
347 			recover_data_page(sbi, NULL, &sum, src, dest);
348 			update_extent_cache(dest, &dn);
349 			recovered++;
350 		}
351 		dn.ofs_in_node++;
352 	}
353 
354 	/* write node page in place */
355 	set_summary(&sum, dn.nid, 0, 0);
356 	if (IS_INODE(dn.node_page))
357 		sync_inode_page(&dn);
358 
359 	copy_node_footer(dn.node_page, page);
360 	fill_node_footer(dn.node_page, dn.nid, ni.ino,
361 					ofs_of_node(page), false);
362 	set_page_dirty(dn.node_page);
363 
364 	recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr);
365 err:
366 	f2fs_put_dnode(&dn);
367 	f2fs_unlock_op(sbi);
368 out:
369 	f2fs_msg(sbi->sb, KERN_NOTICE,
370 		"recover_data: ino = %lx, recovered = %d blocks, err = %d",
371 		inode->i_ino, recovered, err);
372 	return err;
373 }
374 
375 static int recover_data(struct f2fs_sb_info *sbi,
376 				struct list_head *head, int type)
377 {
378 	unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
379 	struct curseg_info *curseg;
380 	struct page *page;
381 	int err = 0;
382 	block_t blkaddr;
383 
384 	/* get node pages in the current segment */
385 	curseg = CURSEG_I(sbi, type);
386 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
387 
388 	/* read node page */
389 	page = alloc_page(GFP_F2FS_ZERO);
390 	if (!page)
391 		return -ENOMEM;
392 
393 	lock_page(page);
394 
395 	while (1) {
396 		struct fsync_inode_entry *entry;
397 
398 		err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC);
399 		if (err)
400 			return err;
401 
402 		lock_page(page);
403 
404 		if (cp_ver != cpver_of_node(page))
405 			break;
406 
407 		entry = get_fsync_inode(head, ino_of_node(page));
408 		if (!entry)
409 			goto next;
410 
411 		err = do_recover_data(sbi, entry->inode, page, blkaddr);
412 		if (err)
413 			break;
414 
415 		if (entry->blkaddr == blkaddr) {
416 			iput(entry->inode);
417 			list_del(&entry->list);
418 			kmem_cache_free(fsync_entry_slab, entry);
419 		}
420 next:
421 		/* check next segment */
422 		blkaddr = next_blkaddr_of_node(page);
423 	}
424 
425 	unlock_page(page);
426 	__free_pages(page, 0);
427 
428 	if (!err)
429 		allocate_new_segments(sbi);
430 	return err;
431 }
432 
433 int recover_fsync_data(struct f2fs_sb_info *sbi)
434 {
435 	struct list_head inode_list;
436 	int err;
437 	bool need_writecp = false;
438 
439 	fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
440 			sizeof(struct fsync_inode_entry), NULL);
441 	if (!fsync_entry_slab)
442 		return -ENOMEM;
443 
444 	INIT_LIST_HEAD(&inode_list);
445 
446 	/* step #1: find fsynced inode numbers */
447 	sbi->por_doing = true;
448 	err = find_fsync_dnodes(sbi, &inode_list);
449 	if (err)
450 		goto out;
451 
452 	if (list_empty(&inode_list))
453 		goto out;
454 
455 	need_writecp = true;
456 
457 	/* step #2: recover data */
458 	err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
459 	f2fs_bug_on(!list_empty(&inode_list));
460 out:
461 	destroy_fsync_dnodes(&inode_list);
462 	kmem_cache_destroy(fsync_entry_slab);
463 	sbi->por_doing = false;
464 	if (!err && need_writecp)
465 		write_checkpoint(sbi, false);
466 	return err;
467 }
468