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 qstr name; 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 = PTR_ERR(dir); 57 goto out; 58 } 59 60 name.len = le32_to_cpu(raw_inode->i_namelen); 61 name.name = raw_inode->i_name; 62 63 de = f2fs_find_entry(dir, &name, &page); 64 if (de) { 65 kunmap(page); 66 f2fs_put_page(page, 0); 67 } else { 68 err = __f2fs_add_link(dir, &name, inode); 69 } 70 iput(dir); 71 out: 72 kunmap(ipage); 73 return err; 74 } 75 76 static int recover_inode(struct inode *inode, struct page *node_page) 77 { 78 void *kaddr = page_address(node_page); 79 struct f2fs_node *raw_node = (struct f2fs_node *)kaddr; 80 struct f2fs_inode *raw_inode = &(raw_node->i); 81 82 inode->i_mode = le16_to_cpu(raw_inode->i_mode); 83 i_size_write(inode, le64_to_cpu(raw_inode->i_size)); 84 inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime); 85 inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime); 86 inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime); 87 inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec); 88 inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec); 89 inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec); 90 91 return recover_dentry(node_page, inode); 92 } 93 94 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head) 95 { 96 unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver); 97 struct curseg_info *curseg; 98 struct page *page; 99 block_t blkaddr; 100 int err = 0; 101 102 /* get node pages in the current segment */ 103 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 104 blkaddr = START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff; 105 106 /* read node page */ 107 page = alloc_page(GFP_F2FS_ZERO); 108 if (IS_ERR(page)) 109 return PTR_ERR(page); 110 lock_page(page); 111 112 while (1) { 113 struct fsync_inode_entry *entry; 114 115 err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC); 116 if (err) 117 goto out; 118 119 lock_page(page); 120 121 if (cp_ver != cpver_of_node(page)) 122 goto unlock_out; 123 124 if (!is_fsync_dnode(page)) 125 goto next; 126 127 entry = get_fsync_inode(head, ino_of_node(page)); 128 if (entry) { 129 entry->blkaddr = blkaddr; 130 if (IS_INODE(page) && is_dent_dnode(page)) 131 set_inode_flag(F2FS_I(entry->inode), 132 FI_INC_LINK); 133 } else { 134 if (IS_INODE(page) && is_dent_dnode(page)) { 135 err = recover_inode_page(sbi, page); 136 if (err) 137 goto unlock_out; 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 unlock_out; 145 } 146 147 entry->inode = f2fs_iget(sbi->sb, ino_of_node(page)); 148 if (IS_ERR(entry->inode)) { 149 err = PTR_ERR(entry->inode); 150 kmem_cache_free(fsync_entry_slab, entry); 151 goto unlock_out; 152 } 153 154 list_add_tail(&entry->list, head); 155 entry->blkaddr = blkaddr; 156 } 157 if (IS_INODE(page)) { 158 err = recover_inode(entry->inode, page); 159 if (err == -ENOENT) { 160 goto next; 161 } else if (err) { 162 err = -EINVAL; 163 goto unlock_out; 164 } 165 } 166 next: 167 /* check next segment */ 168 blkaddr = next_blkaddr_of_node(page); 169 } 170 unlock_out: 171 unlock_page(page); 172 out: 173 __free_pages(page, 0); 174 return err; 175 } 176 177 static void destroy_fsync_dnodes(struct f2fs_sb_info *sbi, 178 struct list_head *head) 179 { 180 struct fsync_inode_entry *entry, *tmp; 181 182 list_for_each_entry_safe(entry, tmp, head, list) { 183 iput(entry->inode); 184 list_del(&entry->list); 185 kmem_cache_free(fsync_entry_slab, entry); 186 } 187 } 188 189 static void check_index_in_prev_nodes(struct f2fs_sb_info *sbi, 190 block_t blkaddr) 191 { 192 struct seg_entry *sentry; 193 unsigned int segno = GET_SEGNO(sbi, blkaddr); 194 unsigned short blkoff = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & 195 (sbi->blocks_per_seg - 1); 196 struct f2fs_summary sum; 197 nid_t ino; 198 void *kaddr; 199 struct inode *inode; 200 struct page *node_page; 201 block_t bidx; 202 int i; 203 204 sentry = get_seg_entry(sbi, segno); 205 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map)) 206 return; 207 208 /* Get the previous summary */ 209 for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) { 210 struct curseg_info *curseg = CURSEG_I(sbi, i); 211 if (curseg->segno == segno) { 212 sum = curseg->sum_blk->entries[blkoff]; 213 break; 214 } 215 } 216 if (i > CURSEG_COLD_DATA) { 217 struct page *sum_page = get_sum_page(sbi, segno); 218 struct f2fs_summary_block *sum_node; 219 kaddr = page_address(sum_page); 220 sum_node = (struct f2fs_summary_block *)kaddr; 221 sum = sum_node->entries[blkoff]; 222 f2fs_put_page(sum_page, 1); 223 } 224 225 /* Get the node page */ 226 node_page = get_node_page(sbi, le32_to_cpu(sum.nid)); 227 bidx = start_bidx_of_node(ofs_of_node(node_page)) + 228 le16_to_cpu(sum.ofs_in_node); 229 ino = ino_of_node(node_page); 230 f2fs_put_page(node_page, 1); 231 232 /* Deallocate previous index in the node page */ 233 inode = f2fs_iget(sbi->sb, ino); 234 if (IS_ERR(inode)) 235 return; 236 237 truncate_hole(inode, bidx, bidx + 1); 238 iput(inode); 239 } 240 241 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode, 242 struct page *page, block_t blkaddr) 243 { 244 unsigned int start, end; 245 struct dnode_of_data dn; 246 struct f2fs_summary sum; 247 struct node_info ni; 248 int err = 0; 249 int ilock; 250 251 start = start_bidx_of_node(ofs_of_node(page)); 252 if (IS_INODE(page)) 253 end = start + ADDRS_PER_INODE; 254 else 255 end = start + ADDRS_PER_BLOCK; 256 257 ilock = mutex_lock_op(sbi); 258 set_new_dnode(&dn, inode, NULL, NULL, 0); 259 260 err = get_dnode_of_data(&dn, start, ALLOC_NODE); 261 if (err) { 262 mutex_unlock_op(sbi, ilock); 263 return err; 264 } 265 266 wait_on_page_writeback(dn.node_page); 267 268 get_node_info(sbi, dn.nid, &ni); 269 BUG_ON(ni.ino != ino_of_node(page)); 270 BUG_ON(ofs_of_node(dn.node_page) != ofs_of_node(page)); 271 272 for (; start < end; start++) { 273 block_t src, dest; 274 275 src = datablock_addr(dn.node_page, dn.ofs_in_node); 276 dest = datablock_addr(page, dn.ofs_in_node); 277 278 if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) { 279 if (src == NULL_ADDR) { 280 int err = reserve_new_block(&dn); 281 /* We should not get -ENOSPC */ 282 BUG_ON(err); 283 } 284 285 /* Check the previous node page having this index */ 286 check_index_in_prev_nodes(sbi, dest); 287 288 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); 289 290 /* write dummy data page */ 291 recover_data_page(sbi, NULL, &sum, src, dest); 292 update_extent_cache(dest, &dn); 293 } 294 dn.ofs_in_node++; 295 } 296 297 /* write node page in place */ 298 set_summary(&sum, dn.nid, 0, 0); 299 if (IS_INODE(dn.node_page)) 300 sync_inode_page(&dn); 301 302 copy_node_footer(dn.node_page, page); 303 fill_node_footer(dn.node_page, dn.nid, ni.ino, 304 ofs_of_node(page), false); 305 set_page_dirty(dn.node_page); 306 307 recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr); 308 f2fs_put_dnode(&dn); 309 mutex_unlock_op(sbi, ilock); 310 return 0; 311 } 312 313 static int recover_data(struct f2fs_sb_info *sbi, 314 struct list_head *head, int type) 315 { 316 unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver); 317 struct curseg_info *curseg; 318 struct page *page; 319 int err = 0; 320 block_t blkaddr; 321 322 /* get node pages in the current segment */ 323 curseg = CURSEG_I(sbi, type); 324 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 325 326 /* read node page */ 327 page = alloc_page(GFP_NOFS | __GFP_ZERO); 328 if (IS_ERR(page)) 329 return -ENOMEM; 330 331 lock_page(page); 332 333 while (1) { 334 struct fsync_inode_entry *entry; 335 336 err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC); 337 if (err) 338 goto out; 339 340 lock_page(page); 341 342 if (cp_ver != cpver_of_node(page)) 343 goto unlock_out; 344 345 entry = get_fsync_inode(head, ino_of_node(page)); 346 if (!entry) 347 goto next; 348 349 err = do_recover_data(sbi, entry->inode, page, blkaddr); 350 if (err) 351 goto out; 352 353 if (entry->blkaddr == blkaddr) { 354 iput(entry->inode); 355 list_del(&entry->list); 356 kmem_cache_free(fsync_entry_slab, entry); 357 } 358 next: 359 /* check next segment */ 360 blkaddr = next_blkaddr_of_node(page); 361 } 362 unlock_out: 363 unlock_page(page); 364 out: 365 __free_pages(page, 0); 366 367 if (!err) 368 allocate_new_segments(sbi); 369 return err; 370 } 371 372 int recover_fsync_data(struct f2fs_sb_info *sbi) 373 { 374 struct list_head inode_list; 375 int err; 376 377 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry", 378 sizeof(struct fsync_inode_entry), NULL); 379 if (unlikely(!fsync_entry_slab)) 380 return -ENOMEM; 381 382 INIT_LIST_HEAD(&inode_list); 383 384 /* step #1: find fsynced inode numbers */ 385 err = find_fsync_dnodes(sbi, &inode_list); 386 if (err) 387 goto out; 388 389 if (list_empty(&inode_list)) 390 goto out; 391 392 /* step #2: recover data */ 393 sbi->por_doing = 1; 394 err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE); 395 sbi->por_doing = 0; 396 BUG_ON(!list_empty(&inode_list)); 397 out: 398 destroy_fsync_dnodes(sbi, &inode_list); 399 kmem_cache_destroy(fsync_entry_slab); 400 write_checkpoint(sbi, false); 401 return err; 402 } 403