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 | { | |
45590710 | 43 | struct f2fs_node *raw_node = F2FS_NODE(ipage); |
d624c96f | 44 | struct f2fs_inode *raw_inode = &(raw_node->i); |
74d0b917 | 45 | nid_t pino = le32_to_cpu(raw_inode->i_pino); |
6b8213d9 | 46 | struct f2fs_dir_entry *de; |
b7f7a5e0 | 47 | struct qstr name; |
d624c96f | 48 | struct page *page; |
6b8213d9 | 49 | struct inode *dir, *einode; |
d624c96f JK |
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); | |
5deb8267 | 60 | add_dirty_dir_inode(dir); |
d624c96f JK |
61 | } |
62 | ||
b7f7a5e0 AV |
63 | name.len = le32_to_cpu(raw_inode->i_namelen); |
64 | name.name = raw_inode->i_name; | |
6b8213d9 JK |
65 | retry: |
66 | de = f2fs_find_entry(dir, &name, &page); | |
67 | if (de && inode->i_ino == le32_to_cpu(de->ino)) { | |
d624c96f JK |
68 | kunmap(page); |
69 | f2fs_put_page(page, 0); | |
6b8213d9 JK |
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; | |
d624c96f | 83 | } |
6b8213d9 | 84 | err = __f2fs_add_link(dir, &name, inode); |
d624c96f | 85 | out: |
f356fe0c JK |
86 | f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode and its dentry: " |
87 | "ino = %x, name = %s, dir = %lx, err = %d", | |
f28c06fa DC |
88 | ino_of_node(ipage), raw_inode->i_name, |
89 | IS_ERR(dir) ? 0 : dir->i_ino, err); | |
d624c96f JK |
90 | return err; |
91 | } | |
92 | ||
93 | static int recover_inode(struct inode *inode, struct page *node_page) | |
94 | { | |
45590710 | 95 | struct f2fs_node *raw_node = F2FS_NODE(node_page); |
d624c96f JK |
96 | struct f2fs_inode *raw_inode = &(raw_node->i); |
97 | ||
f356fe0c JK |
98 | if (!IS_INODE(node_page)) |
99 | return 0; | |
100 | ||
25ca923b | 101 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
d624c96f JK |
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 | ||
f356fe0c JK |
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; | |
d624c96f JK |
116 | } |
117 | ||
118 | static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head) | |
119 | { | |
120 | unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver); | |
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 (IS_ERR(page)) | |
133 | return PTR_ERR(page); | |
134 | lock_page(page); | |
135 | ||
136 | while (1) { | |
137 | struct fsync_inode_entry *entry; | |
138 | ||
393ff91f JK |
139 | err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC); |
140 | if (err) | |
d624c96f JK |
141 | goto out; |
142 | ||
393ff91f JK |
143 | lock_page(page); |
144 | ||
6ead1142 | 145 | if (cp_ver != cpver_of_node(page)) |
f356fe0c | 146 | break; |
d624c96f JK |
147 | |
148 | if (!is_fsync_dnode(page)) | |
149 | goto next; | |
150 | ||
151 | entry = get_fsync_inode(head, ino_of_node(page)); | |
152 | if (entry) { | |
d624c96f JK |
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)) { | |
6ead1142 JK |
158 | err = recover_inode_page(sbi, page); |
159 | if (err) | |
f356fe0c | 160 | break; |
d624c96f JK |
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; | |
f356fe0c | 167 | break; |
d624c96f JK |
168 | } |
169 | ||
d624c96f JK |
170 | entry->inode = f2fs_iget(sbi->sb, ino_of_node(page)); |
171 | if (IS_ERR(entry->inode)) { | |
172 | err = PTR_ERR(entry->inode); | |
fd8bb65f | 173 | kmem_cache_free(fsync_entry_slab, entry); |
f356fe0c | 174 | break; |
d624c96f | 175 | } |
fd8bb65f | 176 | list_add_tail(&entry->list, head); |
d624c96f | 177 | } |
addbe45b JK |
178 | entry->blkaddr = blkaddr; |
179 | ||
f356fe0c JK |
180 | err = recover_inode(entry->inode, page); |
181 | if (err && err != -ENOENT) | |
182 | break; | |
d624c96f JK |
183 | next: |
184 | /* check next segment */ | |
185 | blkaddr = next_blkaddr_of_node(page); | |
d624c96f | 186 | } |
d624c96f | 187 | unlock_page(page); |
393ff91f | 188 | out: |
d624c96f JK |
189 | __free_pages(page, 0); |
190 | return err; | |
191 | } | |
192 | ||
5ebefc5b | 193 | static void destroy_fsync_dnodes(struct list_head *head) |
d624c96f | 194 | { |
d8b79b2f DC |
195 | struct fsync_inode_entry *entry, *tmp; |
196 | ||
197 | list_for_each_entry_safe(entry, tmp, head, list) { | |
d624c96f JK |
198 | iput(entry->inode); |
199 | list_del(&entry->list); | |
200 | kmem_cache_free(fsync_entry_slab, entry); | |
201 | } | |
202 | } | |
203 | ||
39cf72cf | 204 | static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi, |
b292dcab | 205 | block_t blkaddr, struct dnode_of_data *dn) |
d624c96f JK |
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; | |
b292dcab | 212 | nid_t ino, nid; |
d624c96f JK |
213 | void *kaddr; |
214 | struct inode *inode; | |
215 | struct page *node_page; | |
216 | block_t bidx; | |
217 | int i; | |
218 | ||
219 | sentry = get_seg_entry(sbi, segno); | |
220 | if (!f2fs_test_bit(blkoff, sentry->cur_valid_map)) | |
39cf72cf | 221 | return 0; |
d624c96f JK |
222 | |
223 | /* Get the previous summary */ | |
224 | for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) { | |
225 | struct curseg_info *curseg = CURSEG_I(sbi, i); | |
226 | if (curseg->segno == segno) { | |
227 | sum = curseg->sum_blk->entries[blkoff]; | |
228 | break; | |
229 | } | |
230 | } | |
231 | if (i > CURSEG_COLD_DATA) { | |
232 | struct page *sum_page = get_sum_page(sbi, segno); | |
233 | struct f2fs_summary_block *sum_node; | |
234 | kaddr = page_address(sum_page); | |
235 | sum_node = (struct f2fs_summary_block *)kaddr; | |
236 | sum = sum_node->entries[blkoff]; | |
237 | f2fs_put_page(sum_page, 1); | |
238 | } | |
239 | ||
b292dcab JK |
240 | /* Use the locked dnode page and inode */ |
241 | nid = le32_to_cpu(sum.nid); | |
242 | if (dn->inode->i_ino == nid) { | |
243 | struct dnode_of_data tdn = *dn; | |
244 | tdn.nid = nid; | |
245 | tdn.node_page = dn->inode_page; | |
060dd67b | 246 | tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); |
b292dcab | 247 | truncate_data_blocks_range(&tdn, 1); |
39cf72cf | 248 | return 0; |
b292dcab JK |
249 | } else if (dn->nid == nid) { |
250 | struct dnode_of_data tdn = *dn; | |
060dd67b | 251 | tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); |
b292dcab | 252 | truncate_data_blocks_range(&tdn, 1); |
39cf72cf | 253 | return 0; |
b292dcab JK |
254 | } |
255 | ||
d624c96f | 256 | /* Get the node page */ |
b292dcab | 257 | node_page = get_node_page(sbi, nid); |
39cf72cf JK |
258 | if (IS_ERR(node_page)) |
259 | return PTR_ERR(node_page); | |
d624c96f | 260 | bidx = start_bidx_of_node(ofs_of_node(node_page)) + |
b292dcab | 261 | le16_to_cpu(sum.ofs_in_node); |
d624c96f JK |
262 | ino = ino_of_node(node_page); |
263 | f2fs_put_page(node_page, 1); | |
264 | ||
265 | /* Deallocate previous index in the node page */ | |
d4686d56 | 266 | inode = f2fs_iget(sbi->sb, ino); |
06025f4d | 267 | if (IS_ERR(inode)) |
39cf72cf | 268 | return PTR_ERR(inode); |
06025f4d | 269 | |
d624c96f JK |
270 | truncate_hole(inode, bidx, bidx + 1); |
271 | iput(inode); | |
39cf72cf | 272 | return 0; |
d624c96f JK |
273 | } |
274 | ||
6ead1142 | 275 | static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode, |
d624c96f JK |
276 | struct page *page, block_t blkaddr) |
277 | { | |
278 | unsigned int start, end; | |
279 | struct dnode_of_data dn; | |
280 | struct f2fs_summary sum; | |
281 | struct node_info ni; | |
f356fe0c | 282 | int err = 0, recovered = 0; |
39936837 | 283 | int ilock; |
d624c96f JK |
284 | |
285 | start = start_bidx_of_node(ofs_of_node(page)); | |
286 | if (IS_INODE(page)) | |
287 | end = start + ADDRS_PER_INODE; | |
288 | else | |
289 | end = start + ADDRS_PER_BLOCK; | |
290 | ||
39936837 | 291 | ilock = mutex_lock_op(sbi); |
d624c96f | 292 | set_new_dnode(&dn, inode, NULL, NULL, 0); |
39936837 | 293 | |
6ead1142 | 294 | err = get_dnode_of_data(&dn, start, ALLOC_NODE); |
39936837 JK |
295 | if (err) { |
296 | mutex_unlock_op(sbi, ilock); | |
6ead1142 | 297 | return err; |
39936837 | 298 | } |
d624c96f JK |
299 | |
300 | wait_on_page_writeback(dn.node_page); | |
301 | ||
302 | get_node_info(sbi, dn.nid, &ni); | |
303 | BUG_ON(ni.ino != ino_of_node(page)); | |
304 | BUG_ON(ofs_of_node(dn.node_page) != ofs_of_node(page)); | |
305 | ||
306 | for (; start < end; start++) { | |
307 | block_t src, dest; | |
308 | ||
309 | src = datablock_addr(dn.node_page, dn.ofs_in_node); | |
310 | dest = datablock_addr(page, dn.ofs_in_node); | |
311 | ||
312 | if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) { | |
313 | if (src == NULL_ADDR) { | |
314 | int err = reserve_new_block(&dn); | |
315 | /* We should not get -ENOSPC */ | |
316 | BUG_ON(err); | |
317 | } | |
318 | ||
319 | /* Check the previous node page having this index */ | |
39cf72cf JK |
320 | err = check_index_in_prev_nodes(sbi, dest, &dn); |
321 | if (err) | |
322 | goto err; | |
d624c96f JK |
323 | |
324 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); | |
325 | ||
326 | /* write dummy data page */ | |
327 | recover_data_page(sbi, NULL, &sum, src, dest); | |
328 | update_extent_cache(dest, &dn); | |
f356fe0c | 329 | recovered++; |
d624c96f JK |
330 | } |
331 | dn.ofs_in_node++; | |
332 | } | |
333 | ||
334 | /* write node page in place */ | |
335 | set_summary(&sum, dn.nid, 0, 0); | |
336 | if (IS_INODE(dn.node_page)) | |
337 | sync_inode_page(&dn); | |
338 | ||
339 | copy_node_footer(dn.node_page, page); | |
340 | fill_node_footer(dn.node_page, dn.nid, ni.ino, | |
341 | ofs_of_node(page), false); | |
342 | set_page_dirty(dn.node_page); | |
343 | ||
344 | recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr); | |
39cf72cf | 345 | err: |
d624c96f | 346 | f2fs_put_dnode(&dn); |
39936837 | 347 | mutex_unlock_op(sbi, ilock); |
f356fe0c JK |
348 | |
349 | f2fs_msg(sbi->sb, KERN_NOTICE, "recover_data: ino = %lx, " | |
39cf72cf JK |
350 | "recovered_data = %d blocks, err = %d", |
351 | inode->i_ino, recovered, err); | |
352 | return err; | |
d624c96f JK |
353 | } |
354 | ||
6ead1142 | 355 | static int recover_data(struct f2fs_sb_info *sbi, |
d624c96f JK |
356 | struct list_head *head, int type) |
357 | { | |
358 | unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver); | |
359 | struct curseg_info *curseg; | |
360 | struct page *page; | |
6ead1142 | 361 | int err = 0; |
d624c96f JK |
362 | block_t blkaddr; |
363 | ||
364 | /* get node pages in the current segment */ | |
365 | curseg = CURSEG_I(sbi, type); | |
366 | blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); | |
367 | ||
368 | /* read node page */ | |
369 | page = alloc_page(GFP_NOFS | __GFP_ZERO); | |
370 | if (IS_ERR(page)) | |
6ead1142 JK |
371 | return -ENOMEM; |
372 | ||
d624c96f JK |
373 | lock_page(page); |
374 | ||
375 | while (1) { | |
376 | struct fsync_inode_entry *entry; | |
377 | ||
6ead1142 JK |
378 | err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC); |
379 | if (err) | |
d624c96f JK |
380 | goto out; |
381 | ||
393ff91f JK |
382 | lock_page(page); |
383 | ||
d624c96f | 384 | if (cp_ver != cpver_of_node(page)) |
45856aff | 385 | break; |
d624c96f JK |
386 | |
387 | entry = get_fsync_inode(head, ino_of_node(page)); | |
388 | if (!entry) | |
389 | goto next; | |
390 | ||
6ead1142 JK |
391 | err = do_recover_data(sbi, entry->inode, page, blkaddr); |
392 | if (err) | |
45856aff | 393 | break; |
d624c96f JK |
394 | |
395 | if (entry->blkaddr == blkaddr) { | |
396 | iput(entry->inode); | |
397 | list_del(&entry->list); | |
398 | kmem_cache_free(fsync_entry_slab, entry); | |
399 | } | |
400 | next: | |
401 | /* check next segment */ | |
402 | blkaddr = next_blkaddr_of_node(page); | |
d624c96f | 403 | } |
d624c96f | 404 | unlock_page(page); |
393ff91f | 405 | out: |
d624c96f JK |
406 | __free_pages(page, 0); |
407 | ||
6ead1142 JK |
408 | if (!err) |
409 | allocate_new_segments(sbi); | |
410 | return err; | |
d624c96f JK |
411 | } |
412 | ||
6ead1142 | 413 | int recover_fsync_data(struct f2fs_sb_info *sbi) |
d624c96f JK |
414 | { |
415 | struct list_head inode_list; | |
6ead1142 | 416 | int err; |
d624c96f JK |
417 | |
418 | fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry", | |
419 | sizeof(struct fsync_inode_entry), NULL); | |
420 | if (unlikely(!fsync_entry_slab)) | |
6ead1142 | 421 | return -ENOMEM; |
d624c96f JK |
422 | |
423 | INIT_LIST_HEAD(&inode_list); | |
424 | ||
425 | /* step #1: find fsynced inode numbers */ | |
8c26d7d5 | 426 | sbi->por_doing = 1; |
6ead1142 JK |
427 | err = find_fsync_dnodes(sbi, &inode_list); |
428 | if (err) | |
d624c96f JK |
429 | goto out; |
430 | ||
431 | if (list_empty(&inode_list)) | |
432 | goto out; | |
433 | ||
434 | /* step #2: recover data */ | |
6ead1142 | 435 | err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE); |
d624c96f JK |
436 | BUG_ON(!list_empty(&inode_list)); |
437 | out: | |
5ebefc5b | 438 | destroy_fsync_dnodes(&inode_list); |
d624c96f | 439 | kmem_cache_destroy(fsync_entry_slab); |
8c26d7d5 | 440 | sbi->por_doing = 0; |
2c2c149f JK |
441 | if (!err) |
442 | write_checkpoint(sbi, false); | |
6ead1142 | 443 | return err; |
d624c96f | 444 | } |