Commit | Line | Data |
---|---|---|
d624c96f JK |
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 = (struct f2fs_node *)kmap(ipage); | |
44 | struct f2fs_inode *raw_inode = &(raw_node->i); | |
45 | struct dentry dent, parent; | |
46 | struct f2fs_dir_entry *de; | |
47 | struct page *page; | |
48 | struct inode *dir; | |
49 | int err = 0; | |
50 | ||
51 | if (!is_dent_dnode(ipage)) | |
52 | goto out; | |
53 | ||
54 | dir = f2fs_iget(inode->i_sb, le32_to_cpu(raw_inode->i_pino)); | |
55 | if (IS_ERR(dir)) { | |
56 | err = -EINVAL; | |
57 | goto out; | |
58 | } | |
59 | ||
60 | parent.d_inode = dir; | |
61 | dent.d_parent = &parent; | |
62 | dent.d_name.len = le32_to_cpu(raw_inode->i_namelen); | |
63 | dent.d_name.name = raw_inode->i_name; | |
64 | ||
65 | de = f2fs_find_entry(dir, &dent.d_name, &page); | |
66 | if (de) { | |
67 | kunmap(page); | |
68 | f2fs_put_page(page, 0); | |
69 | } else { | |
70 | f2fs_add_link(&dent, inode); | |
71 | } | |
72 | iput(dir); | |
73 | out: | |
74 | kunmap(ipage); | |
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 | ||
84 | inode->i_mode = le32_to_cpu(raw_inode->i_mode); | |
85 | i_size_write(inode, le64_to_cpu(raw_inode->i_size)); | |
86 | inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime); | |
87 | inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime); | |
88 | inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime); | |
89 | inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec); | |
90 | inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec); | |
91 | inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec); | |
92 | ||
93 | return recover_dentry(node_page, inode); | |
94 | } | |
95 | ||
96 | static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head) | |
97 | { | |
98 | unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver); | |
99 | struct curseg_info *curseg; | |
100 | struct page *page; | |
101 | block_t blkaddr; | |
102 | int err = 0; | |
103 | ||
104 | /* get node pages in the current segment */ | |
105 | curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); | |
106 | blkaddr = START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff; | |
107 | ||
108 | /* read node page */ | |
109 | page = alloc_page(GFP_F2FS_ZERO); | |
110 | if (IS_ERR(page)) | |
111 | return PTR_ERR(page); | |
112 | lock_page(page); | |
113 | ||
114 | while (1) { | |
115 | struct fsync_inode_entry *entry; | |
116 | ||
117 | if (f2fs_readpage(sbi, page, blkaddr, READ_SYNC)) | |
118 | goto out; | |
119 | ||
120 | if (cp_ver != cpver_of_node(page)) | |
121 | goto out; | |
122 | ||
123 | if (!is_fsync_dnode(page)) | |
124 | goto next; | |
125 | ||
126 | entry = get_fsync_inode(head, ino_of_node(page)); | |
127 | if (entry) { | |
128 | entry->blkaddr = blkaddr; | |
129 | if (IS_INODE(page) && is_dent_dnode(page)) | |
130 | set_inode_flag(F2FS_I(entry->inode), | |
131 | FI_INC_LINK); | |
132 | } else { | |
133 | if (IS_INODE(page) && is_dent_dnode(page)) { | |
134 | if (recover_inode_page(sbi, page)) { | |
135 | err = -ENOMEM; | |
136 | goto out; | |
137 | } | |
138 | } | |
139 | ||
140 | /* add this fsync inode to the list */ | |
141 | entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS); | |
142 | if (!entry) { | |
143 | err = -ENOMEM; | |
144 | goto out; | |
145 | } | |
146 | ||
147 | INIT_LIST_HEAD(&entry->list); | |
148 | list_add_tail(&entry->list, head); | |
149 | ||
150 | entry->inode = f2fs_iget(sbi->sb, ino_of_node(page)); | |
151 | if (IS_ERR(entry->inode)) { | |
152 | err = PTR_ERR(entry->inode); | |
153 | goto out; | |
154 | } | |
155 | entry->blkaddr = blkaddr; | |
156 | } | |
157 | if (IS_INODE(page)) { | |
158 | err = recover_inode(entry->inode, page); | |
159 | if (err) | |
160 | goto out; | |
161 | } | |
162 | next: | |
163 | /* check next segment */ | |
164 | blkaddr = next_blkaddr_of_node(page); | |
165 | ClearPageUptodate(page); | |
166 | } | |
167 | out: | |
168 | unlock_page(page); | |
169 | __free_pages(page, 0); | |
170 | return err; | |
171 | } | |
172 | ||
173 | static void destroy_fsync_dnodes(struct f2fs_sb_info *sbi, | |
174 | struct list_head *head) | |
175 | { | |
176 | struct list_head *this; | |
177 | struct fsync_inode_entry *entry; | |
178 | list_for_each(this, head) { | |
179 | entry = list_entry(this, struct fsync_inode_entry, list); | |
180 | iput(entry->inode); | |
181 | list_del(&entry->list); | |
182 | kmem_cache_free(fsync_entry_slab, entry); | |
183 | } | |
184 | } | |
185 | ||
186 | static void check_index_in_prev_nodes(struct f2fs_sb_info *sbi, | |
187 | block_t blkaddr) | |
188 | { | |
189 | struct seg_entry *sentry; | |
190 | unsigned int segno = GET_SEGNO(sbi, blkaddr); | |
191 | unsigned short blkoff = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & | |
192 | (sbi->blocks_per_seg - 1); | |
193 | struct f2fs_summary sum; | |
194 | nid_t ino; | |
195 | void *kaddr; | |
196 | struct inode *inode; | |
197 | struct page *node_page; | |
198 | block_t bidx; | |
199 | int i; | |
200 | ||
201 | sentry = get_seg_entry(sbi, segno); | |
202 | if (!f2fs_test_bit(blkoff, sentry->cur_valid_map)) | |
203 | return; | |
204 | ||
205 | /* Get the previous summary */ | |
206 | for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) { | |
207 | struct curseg_info *curseg = CURSEG_I(sbi, i); | |
208 | if (curseg->segno == segno) { | |
209 | sum = curseg->sum_blk->entries[blkoff]; | |
210 | break; | |
211 | } | |
212 | } | |
213 | if (i > CURSEG_COLD_DATA) { | |
214 | struct page *sum_page = get_sum_page(sbi, segno); | |
215 | struct f2fs_summary_block *sum_node; | |
216 | kaddr = page_address(sum_page); | |
217 | sum_node = (struct f2fs_summary_block *)kaddr; | |
218 | sum = sum_node->entries[blkoff]; | |
219 | f2fs_put_page(sum_page, 1); | |
220 | } | |
221 | ||
222 | /* Get the node page */ | |
223 | node_page = get_node_page(sbi, le32_to_cpu(sum.nid)); | |
224 | bidx = start_bidx_of_node(ofs_of_node(node_page)) + | |
225 | le16_to_cpu(sum.ofs_in_node); | |
226 | ino = ino_of_node(node_page); | |
227 | f2fs_put_page(node_page, 1); | |
228 | ||
229 | /* Deallocate previous index in the node page */ | |
230 | inode = f2fs_iget_nowait(sbi->sb, ino); | |
231 | truncate_hole(inode, bidx, bidx + 1); | |
232 | iput(inode); | |
233 | } | |
234 | ||
235 | static void do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode, | |
236 | struct page *page, block_t blkaddr) | |
237 | { | |
238 | unsigned int start, end; | |
239 | struct dnode_of_data dn; | |
240 | struct f2fs_summary sum; | |
241 | struct node_info ni; | |
242 | ||
243 | start = start_bidx_of_node(ofs_of_node(page)); | |
244 | if (IS_INODE(page)) | |
245 | end = start + ADDRS_PER_INODE; | |
246 | else | |
247 | end = start + ADDRS_PER_BLOCK; | |
248 | ||
249 | set_new_dnode(&dn, inode, NULL, NULL, 0); | |
250 | if (get_dnode_of_data(&dn, start, 0)) | |
251 | return; | |
252 | ||
253 | wait_on_page_writeback(dn.node_page); | |
254 | ||
255 | get_node_info(sbi, dn.nid, &ni); | |
256 | BUG_ON(ni.ino != ino_of_node(page)); | |
257 | BUG_ON(ofs_of_node(dn.node_page) != ofs_of_node(page)); | |
258 | ||
259 | for (; start < end; start++) { | |
260 | block_t src, dest; | |
261 | ||
262 | src = datablock_addr(dn.node_page, dn.ofs_in_node); | |
263 | dest = datablock_addr(page, dn.ofs_in_node); | |
264 | ||
265 | if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) { | |
266 | if (src == NULL_ADDR) { | |
267 | int err = reserve_new_block(&dn); | |
268 | /* We should not get -ENOSPC */ | |
269 | BUG_ON(err); | |
270 | } | |
271 | ||
272 | /* Check the previous node page having this index */ | |
273 | check_index_in_prev_nodes(sbi, dest); | |
274 | ||
275 | set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); | |
276 | ||
277 | /* write dummy data page */ | |
278 | recover_data_page(sbi, NULL, &sum, src, dest); | |
279 | update_extent_cache(dest, &dn); | |
280 | } | |
281 | dn.ofs_in_node++; | |
282 | } | |
283 | ||
284 | /* write node page in place */ | |
285 | set_summary(&sum, dn.nid, 0, 0); | |
286 | if (IS_INODE(dn.node_page)) | |
287 | sync_inode_page(&dn); | |
288 | ||
289 | copy_node_footer(dn.node_page, page); | |
290 | fill_node_footer(dn.node_page, dn.nid, ni.ino, | |
291 | ofs_of_node(page), false); | |
292 | set_page_dirty(dn.node_page); | |
293 | ||
294 | recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr); | |
295 | f2fs_put_dnode(&dn); | |
296 | } | |
297 | ||
298 | static void recover_data(struct f2fs_sb_info *sbi, | |
299 | struct list_head *head, int type) | |
300 | { | |
301 | unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver); | |
302 | struct curseg_info *curseg; | |
303 | struct page *page; | |
304 | block_t blkaddr; | |
305 | ||
306 | /* get node pages in the current segment */ | |
307 | curseg = CURSEG_I(sbi, type); | |
308 | blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); | |
309 | ||
310 | /* read node page */ | |
311 | page = alloc_page(GFP_NOFS | __GFP_ZERO); | |
312 | if (IS_ERR(page)) | |
313 | return; | |
314 | lock_page(page); | |
315 | ||
316 | while (1) { | |
317 | struct fsync_inode_entry *entry; | |
318 | ||
319 | if (f2fs_readpage(sbi, page, blkaddr, READ_SYNC)) | |
320 | goto out; | |
321 | ||
322 | if (cp_ver != cpver_of_node(page)) | |
323 | goto out; | |
324 | ||
325 | entry = get_fsync_inode(head, ino_of_node(page)); | |
326 | if (!entry) | |
327 | goto next; | |
328 | ||
329 | do_recover_data(sbi, entry->inode, page, blkaddr); | |
330 | ||
331 | if (entry->blkaddr == blkaddr) { | |
332 | iput(entry->inode); | |
333 | list_del(&entry->list); | |
334 | kmem_cache_free(fsync_entry_slab, entry); | |
335 | } | |
336 | next: | |
337 | /* check next segment */ | |
338 | blkaddr = next_blkaddr_of_node(page); | |
339 | ClearPageUptodate(page); | |
340 | } | |
341 | out: | |
342 | unlock_page(page); | |
343 | __free_pages(page, 0); | |
344 | ||
345 | allocate_new_segments(sbi); | |
346 | } | |
347 | ||
348 | void recover_fsync_data(struct f2fs_sb_info *sbi) | |
349 | { | |
350 | struct list_head inode_list; | |
351 | ||
352 | fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry", | |
353 | sizeof(struct fsync_inode_entry), NULL); | |
354 | if (unlikely(!fsync_entry_slab)) | |
355 | return; | |
356 | ||
357 | INIT_LIST_HEAD(&inode_list); | |
358 | ||
359 | /* step #1: find fsynced inode numbers */ | |
360 | if (find_fsync_dnodes(sbi, &inode_list)) | |
361 | goto out; | |
362 | ||
363 | if (list_empty(&inode_list)) | |
364 | goto out; | |
365 | ||
366 | /* step #2: recover data */ | |
367 | sbi->por_doing = 1; | |
368 | recover_data(sbi, &inode_list, CURSEG_WARM_NODE); | |
369 | sbi->por_doing = 0; | |
370 | BUG_ON(!list_empty(&inode_list)); | |
371 | out: | |
372 | destroy_fsync_dnodes(sbi, &inode_list); | |
373 | kmem_cache_destroy(fsync_entry_slab); | |
374 | write_checkpoint(sbi, false, false); | |
375 | } |