Commit | Line | Data |
---|---|---|
0a8165d7 | 1 | /* |
d624c96f JK |
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 | { | |
9a55ed65 JK |
43 | void *kaddr = page_address(ipage); |
44 | struct f2fs_node *raw_node = (struct f2fs_node *)kaddr; | |
d624c96f | 45 | struct f2fs_inode *raw_inode = &(raw_node->i); |
74d0b917 | 46 | nid_t pino = le32_to_cpu(raw_inode->i_pino); |
b7f7a5e0 | 47 | struct qstr name; |
d624c96f JK |
48 | struct page *page; |
49 | struct inode *dir; | |
50 | int err = 0; | |
51 | ||
74d0b917 JK |
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); | |
d624c96f JK |
60 | } |
61 | ||
b7f7a5e0 AV |
62 | name.len = le32_to_cpu(raw_inode->i_namelen); |
63 | name.name = raw_inode->i_name; | |
d624c96f | 64 | |
9a55ed65 | 65 | if (f2fs_find_entry(dir, &name, &page)) { |
d624c96f JK |
66 | kunmap(page); |
67 | f2fs_put_page(page, 0); | |
68 | } else { | |
90b2fc64 | 69 | err = __f2fs_add_link(dir, &name, inode); |
d624c96f | 70 | } |
d624c96f | 71 | out: |
f356fe0c JK |
72 | f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode and its dentry: " |
73 | "ino = %x, name = %s, dir = %lx, err = %d", | |
74 | ino_of_node(ipage), raw_inode->i_name, dir->i_ino, err); | |
d624c96f JK |
75 | return err; |
76 | } | |
77 | ||
78 | static int recover_inode(struct inode *inode, struct page *node_page) | |
79 | { | |
80 | void *kaddr = page_address(node_page); | |
81 | struct f2fs_node *raw_node = (struct f2fs_node *)kaddr; | |
82 | struct f2fs_inode *raw_inode = &(raw_node->i); | |
83 | ||
f356fe0c JK |
84 | if (!IS_INODE(node_page)) |
85 | return 0; | |
86 | ||
25ca923b | 87 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
d624c96f JK |
88 | i_size_write(inode, le64_to_cpu(raw_inode->i_size)); |
89 | inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime); | |
90 | inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime); | |
91 | inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime); | |
92 | inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec); | |
93 | inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec); | |
94 | inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec); | |
95 | ||
f356fe0c JK |
96 | if (is_dent_dnode(node_page)) |
97 | return recover_dentry(node_page, inode); | |
98 | ||
99 | f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s", | |
100 | ino_of_node(node_page), raw_inode->i_name); | |
101 | return 0; | |
d624c96f JK |
102 | } |
103 | ||
104 | static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head) | |
105 | { | |
106 | unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver); | |
107 | struct curseg_info *curseg; | |
108 | struct page *page; | |
109 | block_t blkaddr; | |
110 | int err = 0; | |
111 | ||
112 | /* get node pages in the current segment */ | |
113 | curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); | |
114 | blkaddr = START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff; | |
115 | ||
116 | /* read node page */ | |
117 | page = alloc_page(GFP_F2FS_ZERO); | |
118 | if (IS_ERR(page)) | |
119 | return PTR_ERR(page); | |
120 | lock_page(page); | |
121 | ||
122 | while (1) { | |
123 | struct fsync_inode_entry *entry; | |
124 | ||
393ff91f JK |
125 | err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC); |
126 | if (err) | |
d624c96f JK |
127 | goto out; |
128 | ||
393ff91f JK |
129 | lock_page(page); |
130 | ||
6ead1142 | 131 | if (cp_ver != cpver_of_node(page)) |
f356fe0c | 132 | break; |
d624c96f JK |
133 | |
134 | if (!is_fsync_dnode(page)) | |
135 | goto next; | |
136 | ||
137 | entry = get_fsync_inode(head, ino_of_node(page)); | |
138 | if (entry) { | |
d624c96f JK |
139 | if (IS_INODE(page) && is_dent_dnode(page)) |
140 | set_inode_flag(F2FS_I(entry->inode), | |
141 | FI_INC_LINK); | |
142 | } else { | |
143 | if (IS_INODE(page) && is_dent_dnode(page)) { | |
6ead1142 JK |
144 | err = recover_inode_page(sbi, page); |
145 | if (err) | |
f356fe0c | 146 | break; |
d624c96f JK |
147 | } |
148 | ||
149 | /* add this fsync inode to the list */ | |
150 | entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS); | |
151 | if (!entry) { | |
152 | err = -ENOMEM; | |
f356fe0c | 153 | break; |
d624c96f JK |
154 | } |
155 | ||
d624c96f JK |
156 | entry->inode = f2fs_iget(sbi->sb, ino_of_node(page)); |
157 | if (IS_ERR(entry->inode)) { | |
158 | err = PTR_ERR(entry->inode); | |
fd8bb65f | 159 | kmem_cache_free(fsync_entry_slab, entry); |
f356fe0c | 160 | break; |
d624c96f | 161 | } |
fd8bb65f | 162 | list_add_tail(&entry->list, head); |
d624c96f | 163 | } |
addbe45b JK |
164 | entry->blkaddr = blkaddr; |
165 | ||
f356fe0c JK |
166 | err = recover_inode(entry->inode, page); |
167 | if (err && err != -ENOENT) | |
168 | break; | |
d624c96f JK |
169 | next: |
170 | /* check next segment */ | |
171 | blkaddr = next_blkaddr_of_node(page); | |
d624c96f | 172 | } |
d624c96f | 173 | unlock_page(page); |
393ff91f | 174 | out: |
d624c96f JK |
175 | __free_pages(page, 0); |
176 | return err; | |
177 | } | |
178 | ||
179 | static void destroy_fsync_dnodes(struct f2fs_sb_info *sbi, | |
180 | struct list_head *head) | |
181 | { | |
d8b79b2f DC |
182 | struct fsync_inode_entry *entry, *tmp; |
183 | ||
184 | list_for_each_entry_safe(entry, tmp, head, list) { | |
d624c96f JK |
185 | iput(entry->inode); |
186 | list_del(&entry->list); | |
187 | kmem_cache_free(fsync_entry_slab, entry); | |
188 | } | |
189 | } | |
190 | ||
191 | static void check_index_in_prev_nodes(struct f2fs_sb_info *sbi, | |
b292dcab | 192 | block_t blkaddr, struct dnode_of_data *dn) |
d624c96f JK |
193 | { |
194 | struct seg_entry *sentry; | |
195 | unsigned int segno = GET_SEGNO(sbi, blkaddr); | |
196 | unsigned short blkoff = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & | |
197 | (sbi->blocks_per_seg - 1); | |
198 | struct f2fs_summary sum; | |
b292dcab | 199 | nid_t ino, nid; |
d624c96f JK |
200 | void *kaddr; |
201 | struct inode *inode; | |
202 | struct page *node_page; | |
203 | block_t bidx; | |
204 | int i; | |
205 | ||
206 | sentry = get_seg_entry(sbi, segno); | |
207 | if (!f2fs_test_bit(blkoff, sentry->cur_valid_map)) | |
208 | return; | |
209 | ||
210 | /* Get the previous summary */ | |
211 | for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) { | |
212 | struct curseg_info *curseg = CURSEG_I(sbi, i); | |
213 | if (curseg->segno == segno) { | |
214 | sum = curseg->sum_blk->entries[blkoff]; | |
215 | break; | |
216 | } | |
217 | } | |
218 | if (i > CURSEG_COLD_DATA) { | |
219 | struct page *sum_page = get_sum_page(sbi, segno); | |
220 | struct f2fs_summary_block *sum_node; | |
221 | kaddr = page_address(sum_page); | |
222 | sum_node = (struct f2fs_summary_block *)kaddr; | |
223 | sum = sum_node->entries[blkoff]; | |
224 | f2fs_put_page(sum_page, 1); | |
225 | } | |
226 | ||
b292dcab JK |
227 | /* Use the locked dnode page and inode */ |
228 | nid = le32_to_cpu(sum.nid); | |
229 | if (dn->inode->i_ino == nid) { | |
230 | struct dnode_of_data tdn = *dn; | |
231 | tdn.nid = nid; | |
232 | tdn.node_page = dn->inode_page; | |
233 | tdn.ofs_in_node = sum.ofs_in_node; | |
234 | truncate_data_blocks_range(&tdn, 1); | |
235 | return; | |
236 | } else if (dn->nid == nid) { | |
237 | struct dnode_of_data tdn = *dn; | |
238 | tdn.ofs_in_node = sum.ofs_in_node; | |
239 | truncate_data_blocks_range(&tdn, 1); | |
240 | return; | |
241 | } | |
242 | ||
d624c96f | 243 | /* Get the node page */ |
b292dcab | 244 | node_page = get_node_page(sbi, nid); |
d624c96f | 245 | bidx = start_bidx_of_node(ofs_of_node(node_page)) + |
b292dcab | 246 | le16_to_cpu(sum.ofs_in_node); |
d624c96f JK |
247 | ino = ino_of_node(node_page); |
248 | f2fs_put_page(node_page, 1); | |
249 | ||
250 | /* Deallocate previous index in the node page */ | |
d4686d56 | 251 | inode = f2fs_iget(sbi->sb, ino); |
06025f4d NJ |
252 | if (IS_ERR(inode)) |
253 | return; | |
254 | ||
d624c96f JK |
255 | truncate_hole(inode, bidx, bidx + 1); |
256 | iput(inode); | |
257 | } | |
258 | ||
6ead1142 | 259 | static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode, |
d624c96f JK |
260 | struct page *page, block_t blkaddr) |
261 | { | |
262 | unsigned int start, end; | |
263 | struct dnode_of_data dn; | |
264 | struct f2fs_summary sum; | |
265 | struct node_info ni; | |
f356fe0c | 266 | int err = 0, recovered = 0; |
39936837 | 267 | int ilock; |
d624c96f JK |
268 | |
269 | start = start_bidx_of_node(ofs_of_node(page)); | |
270 | if (IS_INODE(page)) | |
271 | end = start + ADDRS_PER_INODE; | |
272 | else | |
273 | end = start + ADDRS_PER_BLOCK; | |
274 | ||
39936837 | 275 | ilock = mutex_lock_op(sbi); |
d624c96f | 276 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
39936837 | 277 | |
6ead1142 | 278 | err = get_dnode_of_data(&dn, start, ALLOC_NODE); |
39936837 JK |
279 | if (err) { |
280 | mutex_unlock_op(sbi, ilock); | |
6ead1142 | 281 | return err; |
39936837 | 282 | } |
d624c96f JK |
283 | |
284 | wait_on_page_writeback(dn.node_page); | |
285 | ||
286 | get_node_info(sbi, dn.nid, &ni); | |
287 | BUG_ON(ni.ino != ino_of_node(page)); | |
288 | BUG_ON(ofs_of_node(dn.node_page) != ofs_of_node(page)); | |
289 | ||
290 | for (; start < end; start++) { | |
291 | block_t src, dest; | |
292 | ||
293 | src = datablock_addr(dn.node_page, dn.ofs_in_node); | |
294 | dest = datablock_addr(page, dn.ofs_in_node); | |
295 | ||
296 | if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) { | |
297 | if (src == NULL_ADDR) { | |
298 | int err = reserve_new_block(&dn); | |
299 | /* We should not get -ENOSPC */ | |
300 | BUG_ON(err); | |
301 | } | |
302 | ||
303 | /* Check the previous node page having this index */ | |
b292dcab | 304 | check_index_in_prev_nodes(sbi, dest, &dn); |
d624c96f JK |
305 | |
306 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); | |
307 | ||
308 | /* write dummy data page */ | |
309 | recover_data_page(sbi, NULL, &sum, src, dest); | |
310 | update_extent_cache(dest, &dn); | |
f356fe0c | 311 | recovered++; |
d624c96f JK |
312 | } |
313 | dn.ofs_in_node++; | |
314 | } | |
315 | ||
316 | /* write node page in place */ | |
317 | set_summary(&sum, dn.nid, 0, 0); | |
318 | if (IS_INODE(dn.node_page)) | |
319 | sync_inode_page(&dn); | |
320 | ||
321 | copy_node_footer(dn.node_page, page); | |
322 | fill_node_footer(dn.node_page, dn.nid, ni.ino, | |
323 | ofs_of_node(page), false); | |
324 | set_page_dirty(dn.node_page); | |
325 | ||
326 | recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr); | |
327 | f2fs_put_dnode(&dn); | |
39936837 | 328 | mutex_unlock_op(sbi, ilock); |
f356fe0c JK |
329 | |
330 | f2fs_msg(sbi->sb, KERN_NOTICE, "recover_data: ino = %lx, " | |
331 | "recovered_data = %d blocks", | |
332 | inode->i_ino, recovered); | |
6ead1142 | 333 | return 0; |
d624c96f JK |
334 | } |
335 | ||
6ead1142 | 336 | static int recover_data(struct f2fs_sb_info *sbi, |
d624c96f JK |
337 | struct list_head *head, int type) |
338 | { | |
339 | unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver); | |
340 | struct curseg_info *curseg; | |
341 | struct page *page; | |
6ead1142 | 342 | int err = 0; |
d624c96f JK |
343 | block_t blkaddr; |
344 | ||
345 | /* get node pages in the current segment */ | |
346 | curseg = CURSEG_I(sbi, type); | |
347 | blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); | |
348 | ||
349 | /* read node page */ | |
350 | page = alloc_page(GFP_NOFS | __GFP_ZERO); | |
351 | if (IS_ERR(page)) | |
6ead1142 JK |
352 | return -ENOMEM; |
353 | ||
d624c96f JK |
354 | lock_page(page); |
355 | ||
356 | while (1) { | |
357 | struct fsync_inode_entry *entry; | |
358 | ||
6ead1142 JK |
359 | err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC); |
360 | if (err) | |
d624c96f JK |
361 | goto out; |
362 | ||
393ff91f JK |
363 | lock_page(page); |
364 | ||
d624c96f | 365 | if (cp_ver != cpver_of_node(page)) |
45856aff | 366 | break; |
d624c96f JK |
367 | |
368 | entry = get_fsync_inode(head, ino_of_node(page)); | |
369 | if (!entry) | |
370 | goto next; | |
371 | ||
6ead1142 JK |
372 | err = do_recover_data(sbi, entry->inode, page, blkaddr); |
373 | if (err) | |
45856aff | 374 | break; |
d624c96f JK |
375 | |
376 | if (entry->blkaddr == blkaddr) { | |
377 | iput(entry->inode); | |
378 | list_del(&entry->list); | |
379 | kmem_cache_free(fsync_entry_slab, entry); | |
380 | } | |
381 | next: | |
382 | /* check next segment */ | |
383 | blkaddr = next_blkaddr_of_node(page); | |
d624c96f | 384 | } |
d624c96f | 385 | unlock_page(page); |
393ff91f | 386 | out: |
d624c96f JK |
387 | __free_pages(page, 0); |
388 | ||
6ead1142 JK |
389 | if (!err) |
390 | allocate_new_segments(sbi); | |
391 | return err; | |
d624c96f JK |
392 | } |
393 | ||
6ead1142 | 394 | int recover_fsync_data(struct f2fs_sb_info *sbi) |
d624c96f JK |
395 | { |
396 | struct list_head inode_list; | |
6ead1142 | 397 | int err; |
d624c96f JK |
398 | |
399 | fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry", | |
400 | sizeof(struct fsync_inode_entry), NULL); | |
401 | if (unlikely(!fsync_entry_slab)) | |
6ead1142 | 402 | return -ENOMEM; |
d624c96f JK |
403 | |
404 | INIT_LIST_HEAD(&inode_list); | |
405 | ||
406 | /* step #1: find fsynced inode numbers */ | |
8c26d7d5 | 407 | sbi->por_doing = 1; |
6ead1142 JK |
408 | err = find_fsync_dnodes(sbi, &inode_list); |
409 | if (err) | |
d624c96f JK |
410 | goto out; |
411 | ||
412 | if (list_empty(&inode_list)) | |
413 | goto out; | |
414 | ||
415 | /* step #2: recover data */ | |
6ead1142 | 416 | err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE); |
d624c96f JK |
417 | BUG_ON(!list_empty(&inode_list)); |
418 | out: | |
419 | destroy_fsync_dnodes(sbi, &inode_list); | |
420 | kmem_cache_destroy(fsync_entry_slab); | |
8c26d7d5 | 421 | sbi->por_doing = 0; |
2c2c149f JK |
422 | if (!err) |
423 | write_checkpoint(sbi, false); | |
6ead1142 | 424 | return err; |
d624c96f | 425 | } |