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 /* 18 * Roll forward recovery scenarios. 19 * 20 * [Term] F: fsync_mark, D: dentry_mark 21 * 22 * 1. inode(x) | CP | inode(x) | dnode(F) 23 * -> Update the latest inode(x). 24 * 25 * 2. inode(x) | CP | inode(F) | dnode(F) 26 * -> No problem. 27 * 28 * 3. inode(x) | CP | dnode(F) | inode(x) 29 * -> Recover to the latest dnode(F), and drop the last inode(x) 30 * 31 * 4. inode(x) | CP | dnode(F) | inode(F) 32 * -> No problem. 33 * 34 * 5. CP | inode(x) | dnode(F) 35 * -> The inode(DF) was missing. Should drop this dnode(F). 36 * 37 * 6. CP | inode(DF) | dnode(F) 38 * -> No problem. 39 * 40 * 7. CP | dnode(F) | inode(DF) 41 * -> If f2fs_iget fails, then goto next to find inode(DF). 42 * 43 * 8. CP | dnode(F) | inode(x) 44 * -> If f2fs_iget fails, then goto next to find inode(DF). 45 * But it will fail due to no inode(DF). 46 */ 47 48 static struct kmem_cache *fsync_entry_slab; 49 50 bool space_for_roll_forward(struct f2fs_sb_info *sbi) 51 { 52 s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count); 53 54 if (sbi->last_valid_block_count + nalloc > sbi->user_block_count) 55 return false; 56 return true; 57 } 58 59 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head, 60 nid_t ino) 61 { 62 struct fsync_inode_entry *entry; 63 64 list_for_each_entry(entry, head, list) 65 if (entry->inode->i_ino == ino) 66 return entry; 67 68 return NULL; 69 } 70 71 static struct fsync_inode_entry *add_fsync_inode(struct list_head *head, 72 struct inode *inode) 73 { 74 struct fsync_inode_entry *entry; 75 76 entry = kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO); 77 if (!entry) 78 return NULL; 79 80 entry->inode = inode; 81 list_add_tail(&entry->list, head); 82 83 return entry; 84 } 85 86 static void del_fsync_inode(struct fsync_inode_entry *entry) 87 { 88 iput(entry->inode); 89 list_del(&entry->list); 90 kmem_cache_free(fsync_entry_slab, entry); 91 } 92 93 static int recover_dentry(struct inode *inode, struct page *ipage, 94 struct list_head *dir_list) 95 { 96 struct f2fs_inode *raw_inode = F2FS_INODE(ipage); 97 nid_t pino = le32_to_cpu(raw_inode->i_pino); 98 struct f2fs_dir_entry *de; 99 struct qstr name; 100 struct page *page; 101 struct inode *dir, *einode; 102 struct fsync_inode_entry *entry; 103 int err = 0; 104 105 entry = get_fsync_inode(dir_list, pino); 106 if (!entry) { 107 dir = f2fs_iget(inode->i_sb, pino); 108 if (IS_ERR(dir)) { 109 err = PTR_ERR(dir); 110 goto out; 111 } 112 113 entry = add_fsync_inode(dir_list, dir); 114 if (!entry) { 115 err = -ENOMEM; 116 iput(dir); 117 goto out; 118 } 119 } 120 121 dir = entry->inode; 122 123 if (file_enc_name(inode)) 124 return 0; 125 126 name.len = le32_to_cpu(raw_inode->i_namelen); 127 name.name = raw_inode->i_name; 128 129 if (unlikely(name.len > F2FS_NAME_LEN)) { 130 WARN_ON(1); 131 err = -ENAMETOOLONG; 132 goto out; 133 } 134 retry: 135 de = f2fs_find_entry(dir, &name, &page); 136 if (de && inode->i_ino == le32_to_cpu(de->ino)) 137 goto out_unmap_put; 138 139 if (de) { 140 einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino)); 141 if (IS_ERR(einode)) { 142 WARN_ON(1); 143 err = PTR_ERR(einode); 144 if (err == -ENOENT) 145 err = -EEXIST; 146 goto out_unmap_put; 147 } 148 err = acquire_orphan_inode(F2FS_I_SB(inode)); 149 if (err) { 150 iput(einode); 151 goto out_unmap_put; 152 } 153 f2fs_delete_entry(de, page, dir, einode); 154 iput(einode); 155 goto retry; 156 } else if (IS_ERR(page)) { 157 err = PTR_ERR(page); 158 } else { 159 err = __f2fs_add_link(dir, &name, inode, 160 inode->i_ino, inode->i_mode); 161 } 162 goto out; 163 164 out_unmap_put: 165 f2fs_dentry_kunmap(dir, page); 166 f2fs_put_page(page, 0); 167 out: 168 f2fs_msg(inode->i_sb, KERN_NOTICE, 169 "%s: ino = %x, name = %s, dir = %lx, err = %d", 170 __func__, ino_of_node(ipage), raw_inode->i_name, 171 IS_ERR(dir) ? 0 : dir->i_ino, err); 172 return err; 173 } 174 175 static void recover_inode(struct inode *inode, struct page *page) 176 { 177 struct f2fs_inode *raw = F2FS_INODE(page); 178 char *name; 179 180 inode->i_mode = le16_to_cpu(raw->i_mode); 181 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size)); 182 inode->i_atime.tv_sec = le64_to_cpu(raw->i_mtime); 183 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime); 184 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime); 185 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec); 186 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec); 187 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec); 188 189 if (file_enc_name(inode)) 190 name = "<encrypted>"; 191 else 192 name = F2FS_INODE(page)->i_name; 193 194 f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s", 195 ino_of_node(page), name); 196 } 197 198 static bool is_same_inode(struct inode *inode, struct page *ipage) 199 { 200 struct f2fs_inode *ri = F2FS_INODE(ipage); 201 struct timespec disk; 202 203 if (!IS_INODE(ipage)) 204 return true; 205 206 disk.tv_sec = le64_to_cpu(ri->i_ctime); 207 disk.tv_nsec = le32_to_cpu(ri->i_ctime_nsec); 208 if (timespec_compare(&inode->i_ctime, &disk) > 0) 209 return false; 210 211 disk.tv_sec = le64_to_cpu(ri->i_atime); 212 disk.tv_nsec = le32_to_cpu(ri->i_atime_nsec); 213 if (timespec_compare(&inode->i_atime, &disk) > 0) 214 return false; 215 216 disk.tv_sec = le64_to_cpu(ri->i_mtime); 217 disk.tv_nsec = le32_to_cpu(ri->i_mtime_nsec); 218 if (timespec_compare(&inode->i_mtime, &disk) > 0) 219 return false; 220 221 return true; 222 } 223 224 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head) 225 { 226 unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi)); 227 struct curseg_info *curseg; 228 struct inode *inode; 229 struct page *page = NULL; 230 block_t blkaddr; 231 int err = 0; 232 233 /* get node pages in the current segment */ 234 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 235 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 236 237 while (1) { 238 struct fsync_inode_entry *entry; 239 240 if (!is_valid_blkaddr(sbi, blkaddr, META_POR)) 241 return 0; 242 243 page = get_tmp_page(sbi, blkaddr); 244 245 if (cp_ver != cpver_of_node(page)) 246 break; 247 248 if (!is_fsync_dnode(page)) 249 goto next; 250 251 entry = get_fsync_inode(head, ino_of_node(page)); 252 if (entry) { 253 if (!is_same_inode(entry->inode, page)) 254 goto next; 255 } else { 256 if (IS_INODE(page) && is_dent_dnode(page)) { 257 err = recover_inode_page(sbi, page); 258 if (err) 259 break; 260 } 261 262 /* 263 * CP | dnode(F) | inode(DF) 264 * For this case, we should not give up now. 265 */ 266 inode = f2fs_iget(sbi->sb, ino_of_node(page)); 267 if (IS_ERR(inode)) { 268 err = PTR_ERR(inode); 269 if (err == -ENOENT) { 270 err = 0; 271 goto next; 272 } 273 break; 274 } 275 276 /* add this fsync inode to the list */ 277 entry = add_fsync_inode(head, inode); 278 if (!entry) { 279 err = -ENOMEM; 280 iput(inode); 281 break; 282 } 283 } 284 entry->blkaddr = blkaddr; 285 286 if (IS_INODE(page) && is_dent_dnode(page)) 287 entry->last_dentry = blkaddr; 288 next: 289 /* check next segment */ 290 blkaddr = next_blkaddr_of_node(page); 291 f2fs_put_page(page, 1); 292 293 ra_meta_pages_cond(sbi, blkaddr); 294 } 295 f2fs_put_page(page, 1); 296 return err; 297 } 298 299 static void destroy_fsync_dnodes(struct list_head *head) 300 { 301 struct fsync_inode_entry *entry, *tmp; 302 303 list_for_each_entry_safe(entry, tmp, head, list) 304 del_fsync_inode(entry); 305 } 306 307 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi, 308 block_t blkaddr, struct dnode_of_data *dn) 309 { 310 struct seg_entry *sentry; 311 unsigned int segno = GET_SEGNO(sbi, blkaddr); 312 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 313 struct f2fs_summary_block *sum_node; 314 struct f2fs_summary sum; 315 struct page *sum_page, *node_page; 316 struct dnode_of_data tdn = *dn; 317 nid_t ino, nid; 318 struct inode *inode; 319 unsigned int offset; 320 block_t bidx; 321 int i; 322 323 sentry = get_seg_entry(sbi, segno); 324 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map)) 325 return 0; 326 327 /* Get the previous summary */ 328 for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) { 329 struct curseg_info *curseg = CURSEG_I(sbi, i); 330 if (curseg->segno == segno) { 331 sum = curseg->sum_blk->entries[blkoff]; 332 goto got_it; 333 } 334 } 335 336 sum_page = get_sum_page(sbi, segno); 337 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 338 sum = sum_node->entries[blkoff]; 339 f2fs_put_page(sum_page, 1); 340 got_it: 341 /* Use the locked dnode page and inode */ 342 nid = le32_to_cpu(sum.nid); 343 if (dn->inode->i_ino == nid) { 344 tdn.nid = nid; 345 if (!dn->inode_page_locked) 346 lock_page(dn->inode_page); 347 tdn.node_page = dn->inode_page; 348 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 349 goto truncate_out; 350 } else if (dn->nid == nid) { 351 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 352 goto truncate_out; 353 } 354 355 /* Get the node page */ 356 node_page = get_node_page(sbi, nid); 357 if (IS_ERR(node_page)) 358 return PTR_ERR(node_page); 359 360 offset = ofs_of_node(node_page); 361 ino = ino_of_node(node_page); 362 f2fs_put_page(node_page, 1); 363 364 if (ino != dn->inode->i_ino) { 365 /* Deallocate previous index in the node page */ 366 inode = f2fs_iget(sbi->sb, ino); 367 if (IS_ERR(inode)) 368 return PTR_ERR(inode); 369 } else { 370 inode = dn->inode; 371 } 372 373 bidx = start_bidx_of_node(offset, inode) + le16_to_cpu(sum.ofs_in_node); 374 375 /* 376 * if inode page is locked, unlock temporarily, but its reference 377 * count keeps alive. 378 */ 379 if (ino == dn->inode->i_ino && dn->inode_page_locked) 380 unlock_page(dn->inode_page); 381 382 set_new_dnode(&tdn, inode, NULL, NULL, 0); 383 if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE)) 384 goto out; 385 386 if (tdn.data_blkaddr == blkaddr) 387 truncate_data_blocks_range(&tdn, 1); 388 389 f2fs_put_dnode(&tdn); 390 out: 391 if (ino != dn->inode->i_ino) 392 iput(inode); 393 else if (dn->inode_page_locked) 394 lock_page(dn->inode_page); 395 return 0; 396 397 truncate_out: 398 if (datablock_addr(tdn.node_page, tdn.ofs_in_node) == blkaddr) 399 truncate_data_blocks_range(&tdn, 1); 400 if (dn->inode->i_ino == nid && !dn->inode_page_locked) 401 unlock_page(dn->inode_page); 402 return 0; 403 } 404 405 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode, 406 struct page *page, block_t blkaddr) 407 { 408 struct dnode_of_data dn; 409 struct node_info ni; 410 unsigned int start, end; 411 int err = 0, recovered = 0; 412 413 /* step 1: recover xattr */ 414 if (IS_INODE(page)) { 415 recover_inline_xattr(inode, page); 416 } else if (f2fs_has_xattr_block(ofs_of_node(page))) { 417 /* 418 * Deprecated; xattr blocks should be found from cold log. 419 * But, we should remain this for backward compatibility. 420 */ 421 recover_xattr_data(inode, page, blkaddr); 422 goto out; 423 } 424 425 /* step 2: recover inline data */ 426 if (recover_inline_data(inode, page)) 427 goto out; 428 429 /* step 3: recover data indices */ 430 start = start_bidx_of_node(ofs_of_node(page), inode); 431 end = start + ADDRS_PER_PAGE(page, inode); 432 433 set_new_dnode(&dn, inode, NULL, NULL, 0); 434 435 err = get_dnode_of_data(&dn, start, ALLOC_NODE); 436 if (err) 437 goto out; 438 439 f2fs_wait_on_page_writeback(dn.node_page, NODE, true); 440 441 get_node_info(sbi, dn.nid, &ni); 442 f2fs_bug_on(sbi, ni.ino != ino_of_node(page)); 443 f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page)); 444 445 for (; start < end; start++, dn.ofs_in_node++) { 446 block_t src, dest; 447 448 src = datablock_addr(dn.node_page, dn.ofs_in_node); 449 dest = datablock_addr(page, dn.ofs_in_node); 450 451 /* skip recovering if dest is the same as src */ 452 if (src == dest) 453 continue; 454 455 /* dest is invalid, just invalidate src block */ 456 if (dest == NULL_ADDR) { 457 truncate_data_blocks_range(&dn, 1); 458 continue; 459 } 460 461 if ((start + 1) << PAGE_SHIFT > i_size_read(inode)) 462 f2fs_i_size_write(inode, (start + 1) << PAGE_SHIFT); 463 464 /* 465 * dest is reserved block, invalidate src block 466 * and then reserve one new block in dnode page. 467 */ 468 if (dest == NEW_ADDR) { 469 truncate_data_blocks_range(&dn, 1); 470 reserve_new_block(&dn); 471 continue; 472 } 473 474 /* dest is valid block, try to recover from src to dest */ 475 if (is_valid_blkaddr(sbi, dest, META_POR)) { 476 477 if (src == NULL_ADDR) { 478 err = reserve_new_block(&dn); 479 #ifdef CONFIG_F2FS_FAULT_INJECTION 480 while (err) 481 err = reserve_new_block(&dn); 482 #endif 483 /* We should not get -ENOSPC */ 484 f2fs_bug_on(sbi, err); 485 if (err) 486 goto err; 487 } 488 489 /* Check the previous node page having this index */ 490 err = check_index_in_prev_nodes(sbi, dest, &dn); 491 if (err) 492 goto err; 493 494 /* write dummy data page */ 495 f2fs_replace_block(sbi, &dn, src, dest, 496 ni.version, false, false); 497 recovered++; 498 } 499 } 500 501 copy_node_footer(dn.node_page, page); 502 fill_node_footer(dn.node_page, dn.nid, ni.ino, 503 ofs_of_node(page), false); 504 set_page_dirty(dn.node_page); 505 err: 506 f2fs_put_dnode(&dn); 507 out: 508 f2fs_msg(sbi->sb, KERN_NOTICE, 509 "recover_data: ino = %lx, recovered = %d blocks, err = %d", 510 inode->i_ino, recovered, err); 511 return err; 512 } 513 514 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list, 515 struct list_head *dir_list) 516 { 517 unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi)); 518 struct curseg_info *curseg; 519 struct page *page = NULL; 520 int err = 0; 521 block_t blkaddr; 522 523 /* get node pages in the current segment */ 524 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 525 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 526 527 while (1) { 528 struct fsync_inode_entry *entry; 529 530 if (!is_valid_blkaddr(sbi, blkaddr, META_POR)) 531 break; 532 533 ra_meta_pages_cond(sbi, blkaddr); 534 535 page = get_tmp_page(sbi, blkaddr); 536 537 if (cp_ver != cpver_of_node(page)) { 538 f2fs_put_page(page, 1); 539 break; 540 } 541 542 entry = get_fsync_inode(inode_list, ino_of_node(page)); 543 if (!entry) 544 goto next; 545 /* 546 * inode(x) | CP | inode(x) | dnode(F) 547 * In this case, we can lose the latest inode(x). 548 * So, call recover_inode for the inode update. 549 */ 550 if (IS_INODE(page)) 551 recover_inode(entry->inode, page); 552 if (entry->last_dentry == blkaddr) { 553 err = recover_dentry(entry->inode, page, dir_list); 554 if (err) { 555 f2fs_put_page(page, 1); 556 break; 557 } 558 } 559 err = do_recover_data(sbi, entry->inode, page, blkaddr); 560 if (err) { 561 f2fs_put_page(page, 1); 562 break; 563 } 564 565 if (entry->blkaddr == blkaddr) 566 del_fsync_inode(entry); 567 next: 568 /* check next segment */ 569 blkaddr = next_blkaddr_of_node(page); 570 f2fs_put_page(page, 1); 571 } 572 if (!err) 573 allocate_new_segments(sbi); 574 return err; 575 } 576 577 int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only) 578 { 579 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 580 struct list_head inode_list; 581 struct list_head dir_list; 582 block_t blkaddr; 583 int err; 584 int ret = 0; 585 bool need_writecp = false; 586 587 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry", 588 sizeof(struct fsync_inode_entry)); 589 if (!fsync_entry_slab) 590 return -ENOMEM; 591 592 INIT_LIST_HEAD(&inode_list); 593 INIT_LIST_HEAD(&dir_list); 594 595 /* prevent checkpoint */ 596 mutex_lock(&sbi->cp_mutex); 597 598 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 599 600 /* step #1: find fsynced inode numbers */ 601 err = find_fsync_dnodes(sbi, &inode_list); 602 if (err || list_empty(&inode_list)) 603 goto out; 604 605 if (check_only) { 606 ret = 1; 607 goto out; 608 } 609 610 need_writecp = true; 611 612 /* step #2: recover data */ 613 err = recover_data(sbi, &inode_list, &dir_list); 614 if (!err) 615 f2fs_bug_on(sbi, !list_empty(&inode_list)); 616 out: 617 destroy_fsync_dnodes(&inode_list); 618 619 /* truncate meta pages to be used by the recovery */ 620 truncate_inode_pages_range(META_MAPPING(sbi), 621 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1); 622 623 if (err) { 624 truncate_inode_pages_final(NODE_MAPPING(sbi)); 625 truncate_inode_pages_final(META_MAPPING(sbi)); 626 } 627 628 clear_sbi_flag(sbi, SBI_POR_DOING); 629 if (err) { 630 bool invalidate = false; 631 632 if (test_opt(sbi, LFS)) { 633 update_meta_page(sbi, NULL, blkaddr); 634 invalidate = true; 635 } else if (discard_next_dnode(sbi, blkaddr)) { 636 invalidate = true; 637 } 638 639 /* Flush all the NAT/SIT pages */ 640 while (get_pages(sbi, F2FS_DIRTY_META)) 641 sync_meta_pages(sbi, META, LONG_MAX); 642 643 /* invalidate temporary meta page */ 644 if (invalidate) 645 invalidate_mapping_pages(META_MAPPING(sbi), 646 blkaddr, blkaddr); 647 648 set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG); 649 mutex_unlock(&sbi->cp_mutex); 650 } else if (need_writecp) { 651 struct cp_control cpc = { 652 .reason = CP_RECOVERY, 653 }; 654 mutex_unlock(&sbi->cp_mutex); 655 err = write_checkpoint(sbi, &cpc); 656 } else { 657 mutex_unlock(&sbi->cp_mutex); 658 } 659 660 destroy_fsync_dnodes(&dir_list); 661 kmem_cache_destroy(fsync_entry_slab); 662 return ret ? ret: err; 663 } 664