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 = F2FS_NODE(ipage); 44 struct f2fs_inode *raw_inode = &(raw_node->i); 45 nid_t pino = le32_to_cpu(raw_inode->i_pino); 46 struct f2fs_dir_entry *de; 47 struct qstr name; 48 struct page *page; 49 struct inode *dir, *einode; 50 int err = 0; 51 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); 60 add_dirty_dir_inode(dir); 61 } 62 63 name.len = le32_to_cpu(raw_inode->i_namelen); 64 name.name = raw_inode->i_name; 65 retry: 66 de = f2fs_find_entry(dir, &name, &page); 67 if (de && inode->i_ino == le32_to_cpu(de->ino)) 68 goto out_unmap_put; 69 if (de) { 70 einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino)); 71 if (IS_ERR(einode)) { 72 WARN_ON(1); 73 if (PTR_ERR(einode) == -ENOENT) 74 err = -EEXIST; 75 goto out_unmap_put; 76 } 77 err = acquire_orphan_inode(F2FS_SB(inode->i_sb)); 78 if (err) { 79 iput(einode); 80 goto out_unmap_put; 81 } 82 f2fs_delete_entry(de, page, einode); 83 iput(einode); 84 goto retry; 85 } 86 err = __f2fs_add_link(dir, &name, inode); 87 goto out; 88 89 out_unmap_put: 90 kunmap(page); 91 f2fs_put_page(page, 0); 92 out: 93 f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode and its dentry: " 94 "ino = %x, name = %s, dir = %lx, err = %d", 95 ino_of_node(ipage), raw_inode->i_name, 96 IS_ERR(dir) ? 0 : dir->i_ino, err); 97 return err; 98 } 99 100 static int recover_inode(struct inode *inode, struct page *node_page) 101 { 102 struct f2fs_node *raw_node = F2FS_NODE(node_page); 103 struct f2fs_inode *raw_inode = &(raw_node->i); 104 105 if (!IS_INODE(node_page)) 106 return 0; 107 108 inode->i_mode = le16_to_cpu(raw_inode->i_mode); 109 i_size_write(inode, le64_to_cpu(raw_inode->i_size)); 110 inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime); 111 inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime); 112 inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime); 113 inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec); 114 inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec); 115 inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec); 116 117 if (is_dent_dnode(node_page)) 118 return recover_dentry(node_page, inode); 119 120 f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s", 121 ino_of_node(node_page), raw_inode->i_name); 122 return 0; 123 } 124 125 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head) 126 { 127 unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi)); 128 struct curseg_info *curseg; 129 struct page *page; 130 block_t blkaddr; 131 int err = 0; 132 133 /* get node pages in the current segment */ 134 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 135 blkaddr = START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff; 136 137 /* read node page */ 138 page = alloc_page(GFP_F2FS_ZERO); 139 if (!page) 140 return -ENOMEM; 141 lock_page(page); 142 143 while (1) { 144 struct fsync_inode_entry *entry; 145 146 err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC); 147 if (err) 148 return err; 149 150 lock_page(page); 151 152 if (cp_ver != cpver_of_node(page)) 153 break; 154 155 if (!is_fsync_dnode(page)) 156 goto next; 157 158 entry = get_fsync_inode(head, ino_of_node(page)); 159 if (entry) { 160 if (IS_INODE(page) && is_dent_dnode(page)) 161 set_inode_flag(F2FS_I(entry->inode), 162 FI_INC_LINK); 163 } else { 164 if (IS_INODE(page) && is_dent_dnode(page)) { 165 err = recover_inode_page(sbi, page); 166 if (err) 167 break; 168 } 169 170 /* add this fsync inode to the list */ 171 entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS); 172 if (!entry) { 173 err = -ENOMEM; 174 break; 175 } 176 177 entry->inode = f2fs_iget(sbi->sb, ino_of_node(page)); 178 if (IS_ERR(entry->inode)) { 179 err = PTR_ERR(entry->inode); 180 kmem_cache_free(fsync_entry_slab, entry); 181 break; 182 } 183 list_add_tail(&entry->list, head); 184 } 185 entry->blkaddr = blkaddr; 186 187 err = recover_inode(entry->inode, page); 188 if (err && err != -ENOENT) 189 break; 190 next: 191 /* check next segment */ 192 blkaddr = next_blkaddr_of_node(page); 193 } 194 195 unlock_page(page); 196 __free_pages(page, 0); 197 198 return err; 199 } 200 201 static void destroy_fsync_dnodes(struct list_head *head) 202 { 203 struct fsync_inode_entry *entry, *tmp; 204 205 list_for_each_entry_safe(entry, tmp, head, list) { 206 iput(entry->inode); 207 list_del(&entry->list); 208 kmem_cache_free(fsync_entry_slab, entry); 209 } 210 } 211 212 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi, 213 block_t blkaddr, struct dnode_of_data *dn) 214 { 215 struct seg_entry *sentry; 216 unsigned int segno = GET_SEGNO(sbi, blkaddr); 217 unsigned short blkoff = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & 218 (sbi->blocks_per_seg - 1); 219 struct f2fs_summary sum; 220 nid_t ino, nid; 221 void *kaddr; 222 struct inode *inode; 223 struct page *node_page; 224 unsigned int offset; 225 block_t bidx; 226 int i; 227 228 sentry = get_seg_entry(sbi, segno); 229 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map)) 230 return 0; 231 232 /* Get the previous summary */ 233 for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) { 234 struct curseg_info *curseg = CURSEG_I(sbi, i); 235 if (curseg->segno == segno) { 236 sum = curseg->sum_blk->entries[blkoff]; 237 break; 238 } 239 } 240 if (i > CURSEG_COLD_DATA) { 241 struct page *sum_page = get_sum_page(sbi, segno); 242 struct f2fs_summary_block *sum_node; 243 kaddr = page_address(sum_page); 244 sum_node = (struct f2fs_summary_block *)kaddr; 245 sum = sum_node->entries[blkoff]; 246 f2fs_put_page(sum_page, 1); 247 } 248 249 /* Use the locked dnode page and inode */ 250 nid = le32_to_cpu(sum.nid); 251 if (dn->inode->i_ino == nid) { 252 struct dnode_of_data tdn = *dn; 253 tdn.nid = nid; 254 tdn.node_page = dn->inode_page; 255 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 256 truncate_data_blocks_range(&tdn, 1); 257 return 0; 258 } else if (dn->nid == nid) { 259 struct dnode_of_data tdn = *dn; 260 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 261 truncate_data_blocks_range(&tdn, 1); 262 return 0; 263 } 264 265 /* Get the node page */ 266 node_page = get_node_page(sbi, nid); 267 if (IS_ERR(node_page)) 268 return PTR_ERR(node_page); 269 270 offset = ofs_of_node(node_page); 271 ino = ino_of_node(node_page); 272 f2fs_put_page(node_page, 1); 273 274 /* Deallocate previous index in the node page */ 275 inode = f2fs_iget(sbi->sb, ino); 276 if (IS_ERR(inode)) 277 return PTR_ERR(inode); 278 279 bidx = start_bidx_of_node(offset, F2FS_I(inode)) + 280 le16_to_cpu(sum.ofs_in_node); 281 282 truncate_hole(inode, bidx, bidx + 1); 283 iput(inode); 284 return 0; 285 } 286 287 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode, 288 struct page *page, block_t blkaddr) 289 { 290 struct f2fs_inode_info *fi = F2FS_I(inode); 291 unsigned int start, end; 292 struct dnode_of_data dn; 293 struct f2fs_summary sum; 294 struct node_info ni; 295 int err = 0, recovered = 0; 296 297 start = start_bidx_of_node(ofs_of_node(page), fi); 298 if (IS_INODE(page)) 299 end = start + ADDRS_PER_INODE(fi); 300 else 301 end = start + ADDRS_PER_BLOCK; 302 303 f2fs_lock_op(sbi); 304 set_new_dnode(&dn, inode, NULL, NULL, 0); 305 306 err = get_dnode_of_data(&dn, start, ALLOC_NODE); 307 if (err) { 308 f2fs_unlock_op(sbi); 309 return err; 310 } 311 312 wait_on_page_writeback(dn.node_page); 313 314 get_node_info(sbi, dn.nid, &ni); 315 f2fs_bug_on(ni.ino != ino_of_node(page)); 316 f2fs_bug_on(ofs_of_node(dn.node_page) != ofs_of_node(page)); 317 318 for (; start < end; start++) { 319 block_t src, dest; 320 321 src = datablock_addr(dn.node_page, dn.ofs_in_node); 322 dest = datablock_addr(page, dn.ofs_in_node); 323 324 if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) { 325 if (src == NULL_ADDR) { 326 err = reserve_new_block(&dn); 327 /* We should not get -ENOSPC */ 328 f2fs_bug_on(err); 329 } 330 331 /* Check the previous node page having this index */ 332 err = check_index_in_prev_nodes(sbi, dest, &dn); 333 if (err) 334 goto err; 335 336 set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version); 337 338 /* write dummy data page */ 339 recover_data_page(sbi, NULL, &sum, src, dest); 340 update_extent_cache(dest, &dn); 341 recovered++; 342 } 343 dn.ofs_in_node++; 344 } 345 346 /* write node page in place */ 347 set_summary(&sum, dn.nid, 0, 0); 348 if (IS_INODE(dn.node_page)) 349 sync_inode_page(&dn); 350 351 copy_node_footer(dn.node_page, page); 352 fill_node_footer(dn.node_page, dn.nid, ni.ino, 353 ofs_of_node(page), false); 354 set_page_dirty(dn.node_page); 355 356 recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr); 357 err: 358 f2fs_put_dnode(&dn); 359 f2fs_unlock_op(sbi); 360 361 f2fs_msg(sbi->sb, KERN_NOTICE, "recover_data: ino = %lx, " 362 "recovered_data = %d blocks, err = %d", 363 inode->i_ino, recovered, err); 364 return err; 365 } 366 367 static int recover_data(struct f2fs_sb_info *sbi, 368 struct list_head *head, int type) 369 { 370 unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi)); 371 struct curseg_info *curseg; 372 struct page *page; 373 int err = 0; 374 block_t blkaddr; 375 376 /* get node pages in the current segment */ 377 curseg = CURSEG_I(sbi, type); 378 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 379 380 /* read node page */ 381 page = alloc_page(GFP_F2FS_ZERO); 382 if (!page) 383 return -ENOMEM; 384 385 lock_page(page); 386 387 while (1) { 388 struct fsync_inode_entry *entry; 389 390 err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC); 391 if (err) 392 return err; 393 394 lock_page(page); 395 396 if (cp_ver != cpver_of_node(page)) 397 break; 398 399 entry = get_fsync_inode(head, ino_of_node(page)); 400 if (!entry) 401 goto next; 402 403 err = do_recover_data(sbi, entry->inode, page, blkaddr); 404 if (err) 405 break; 406 407 if (entry->blkaddr == blkaddr) { 408 iput(entry->inode); 409 list_del(&entry->list); 410 kmem_cache_free(fsync_entry_slab, entry); 411 } 412 next: 413 /* check next segment */ 414 blkaddr = next_blkaddr_of_node(page); 415 } 416 417 unlock_page(page); 418 __free_pages(page, 0); 419 420 if (!err) 421 allocate_new_segments(sbi); 422 return err; 423 } 424 425 int recover_fsync_data(struct f2fs_sb_info *sbi) 426 { 427 struct list_head inode_list; 428 int err; 429 bool need_writecp = false; 430 431 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry", 432 sizeof(struct fsync_inode_entry), NULL); 433 if (!fsync_entry_slab) 434 return -ENOMEM; 435 436 INIT_LIST_HEAD(&inode_list); 437 438 /* step #1: find fsynced inode numbers */ 439 sbi->por_doing = true; 440 err = find_fsync_dnodes(sbi, &inode_list); 441 if (err) 442 goto out; 443 444 if (list_empty(&inode_list)) 445 goto out; 446 447 need_writecp = true; 448 449 /* step #2: recover data */ 450 err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE); 451 f2fs_bug_on(!list_empty(&inode_list)); 452 out: 453 destroy_fsync_dnodes(&inode_list); 454 kmem_cache_destroy(fsync_entry_slab); 455 sbi->por_doing = false; 456 if (!err && need_writecp) 457 write_checkpoint(sbi, false); 458 return err; 459 } 460