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 f2fs_sb_info *sbi, 72 struct list_head *head, nid_t ino, bool quota_inode) 73 { 74 struct inode *inode; 75 struct fsync_inode_entry *entry; 76 int err; 77 78 inode = f2fs_iget_retry(sbi->sb, ino); 79 if (IS_ERR(inode)) 80 return ERR_CAST(inode); 81 82 err = dquot_initialize(inode); 83 if (err) 84 goto err_out; 85 86 if (quota_inode) { 87 err = dquot_alloc_inode(inode); 88 if (err) 89 goto err_out; 90 } 91 92 entry = f2fs_kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO); 93 entry->inode = inode; 94 list_add_tail(&entry->list, head); 95 96 return entry; 97 err_out: 98 iput(inode); 99 return ERR_PTR(err); 100 } 101 102 static void del_fsync_inode(struct fsync_inode_entry *entry) 103 { 104 iput(entry->inode); 105 list_del(&entry->list); 106 kmem_cache_free(fsync_entry_slab, entry); 107 } 108 109 static int recover_dentry(struct inode *inode, struct page *ipage, 110 struct list_head *dir_list) 111 { 112 struct f2fs_inode *raw_inode = F2FS_INODE(ipage); 113 nid_t pino = le32_to_cpu(raw_inode->i_pino); 114 struct f2fs_dir_entry *de; 115 struct fscrypt_name fname; 116 struct page *page; 117 struct inode *dir, *einode; 118 struct fsync_inode_entry *entry; 119 int err = 0; 120 char *name; 121 122 entry = get_fsync_inode(dir_list, pino); 123 if (!entry) { 124 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list, 125 pino, false); 126 if (IS_ERR(entry)) { 127 dir = ERR_CAST(entry); 128 err = PTR_ERR(entry); 129 goto out; 130 } 131 } 132 133 dir = entry->inode; 134 135 memset(&fname, 0, sizeof(struct fscrypt_name)); 136 fname.disk_name.len = le32_to_cpu(raw_inode->i_namelen); 137 fname.disk_name.name = raw_inode->i_name; 138 139 if (unlikely(fname.disk_name.len > F2FS_NAME_LEN)) { 140 WARN_ON(1); 141 err = -ENAMETOOLONG; 142 goto out; 143 } 144 retry: 145 de = __f2fs_find_entry(dir, &fname, &page); 146 if (de && inode->i_ino == le32_to_cpu(de->ino)) 147 goto out_put; 148 149 if (de) { 150 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino)); 151 if (IS_ERR(einode)) { 152 WARN_ON(1); 153 err = PTR_ERR(einode); 154 if (err == -ENOENT) 155 err = -EEXIST; 156 goto out_put; 157 } 158 159 err = dquot_initialize(einode); 160 if (err) { 161 iput(einode); 162 goto out_put; 163 } 164 165 err = acquire_orphan_inode(F2FS_I_SB(inode)); 166 if (err) { 167 iput(einode); 168 goto out_put; 169 } 170 f2fs_delete_entry(de, page, dir, einode); 171 iput(einode); 172 goto retry; 173 } else if (IS_ERR(page)) { 174 err = PTR_ERR(page); 175 } else { 176 err = __f2fs_do_add_link(dir, &fname, inode, 177 inode->i_ino, inode->i_mode); 178 } 179 if (err == -ENOMEM) 180 goto retry; 181 goto out; 182 183 out_put: 184 f2fs_put_page(page, 0); 185 out: 186 if (file_enc_name(inode)) 187 name = "<encrypted>"; 188 else 189 name = raw_inode->i_name; 190 f2fs_msg(inode->i_sb, KERN_NOTICE, 191 "%s: ino = %x, name = %s, dir = %lx, err = %d", 192 __func__, ino_of_node(ipage), name, 193 IS_ERR(dir) ? 0 : dir->i_ino, err); 194 return err; 195 } 196 197 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri) 198 { 199 if (ri->i_inline & F2FS_PIN_FILE) 200 set_inode_flag(inode, FI_PIN_FILE); 201 else 202 clear_inode_flag(inode, FI_PIN_FILE); 203 if (ri->i_inline & F2FS_DATA_EXIST) 204 set_inode_flag(inode, FI_DATA_EXIST); 205 else 206 clear_inode_flag(inode, FI_DATA_EXIST); 207 if (!(ri->i_inline & F2FS_INLINE_DOTS)) 208 clear_inode_flag(inode, FI_INLINE_DOTS); 209 } 210 211 static void recover_inode(struct inode *inode, struct page *page) 212 { 213 struct f2fs_inode *raw = F2FS_INODE(page); 214 char *name; 215 216 inode->i_mode = le16_to_cpu(raw->i_mode); 217 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size)); 218 inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime); 219 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime); 220 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime); 221 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec); 222 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec); 223 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec); 224 225 F2FS_I(inode)->i_advise = raw->i_advise; 226 227 recover_inline_flags(inode, raw); 228 229 if (file_enc_name(inode)) 230 name = "<encrypted>"; 231 else 232 name = F2FS_INODE(page)->i_name; 233 234 f2fs_msg(inode->i_sb, KERN_NOTICE, 235 "recover_inode: ino = %x, name = %s, inline = %x", 236 ino_of_node(page), name, raw->i_inline); 237 } 238 239 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head, 240 bool check_only) 241 { 242 struct curseg_info *curseg; 243 struct page *page = NULL; 244 block_t blkaddr; 245 unsigned int loop_cnt = 0; 246 unsigned int free_blocks = sbi->user_block_count - 247 valid_user_blocks(sbi); 248 int err = 0; 249 250 /* get node pages in the current segment */ 251 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 252 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 253 254 while (1) { 255 struct fsync_inode_entry *entry; 256 257 if (!is_valid_blkaddr(sbi, blkaddr, META_POR)) 258 return 0; 259 260 page = get_tmp_page(sbi, blkaddr); 261 262 if (!is_recoverable_dnode(page)) 263 break; 264 265 if (!is_fsync_dnode(page)) 266 goto next; 267 268 entry = get_fsync_inode(head, ino_of_node(page)); 269 if (!entry) { 270 bool quota_inode = false; 271 272 if (!check_only && 273 IS_INODE(page) && is_dent_dnode(page)) { 274 err = recover_inode_page(sbi, page); 275 if (err) 276 break; 277 quota_inode = true; 278 } 279 280 /* 281 * CP | dnode(F) | inode(DF) 282 * For this case, we should not give up now. 283 */ 284 entry = add_fsync_inode(sbi, head, ino_of_node(page), 285 quota_inode); 286 if (IS_ERR(entry)) { 287 err = PTR_ERR(entry); 288 if (err == -ENOENT) { 289 err = 0; 290 goto next; 291 } 292 break; 293 } 294 } 295 entry->blkaddr = blkaddr; 296 297 if (IS_INODE(page) && is_dent_dnode(page)) 298 entry->last_dentry = blkaddr; 299 next: 300 /* sanity check in order to detect looped node chain */ 301 if (++loop_cnt >= free_blocks || 302 blkaddr == next_blkaddr_of_node(page)) { 303 f2fs_msg(sbi->sb, KERN_NOTICE, 304 "%s: detect looped node chain, " 305 "blkaddr:%u, next:%u", 306 __func__, blkaddr, next_blkaddr_of_node(page)); 307 err = -EINVAL; 308 break; 309 } 310 311 /* check next segment */ 312 blkaddr = next_blkaddr_of_node(page); 313 f2fs_put_page(page, 1); 314 315 ra_meta_pages_cond(sbi, blkaddr); 316 } 317 f2fs_put_page(page, 1); 318 return err; 319 } 320 321 static void destroy_fsync_dnodes(struct list_head *head) 322 { 323 struct fsync_inode_entry *entry, *tmp; 324 325 list_for_each_entry_safe(entry, tmp, head, list) 326 del_fsync_inode(entry); 327 } 328 329 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi, 330 block_t blkaddr, struct dnode_of_data *dn) 331 { 332 struct seg_entry *sentry; 333 unsigned int segno = GET_SEGNO(sbi, blkaddr); 334 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 335 struct f2fs_summary_block *sum_node; 336 struct f2fs_summary sum; 337 struct page *sum_page, *node_page; 338 struct dnode_of_data tdn = *dn; 339 nid_t ino, nid; 340 struct inode *inode; 341 unsigned int offset; 342 block_t bidx; 343 int i; 344 345 sentry = get_seg_entry(sbi, segno); 346 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map)) 347 return 0; 348 349 /* Get the previous summary */ 350 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 351 struct curseg_info *curseg = CURSEG_I(sbi, i); 352 if (curseg->segno == segno) { 353 sum = curseg->sum_blk->entries[blkoff]; 354 goto got_it; 355 } 356 } 357 358 sum_page = get_sum_page(sbi, segno); 359 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 360 sum = sum_node->entries[blkoff]; 361 f2fs_put_page(sum_page, 1); 362 got_it: 363 /* Use the locked dnode page and inode */ 364 nid = le32_to_cpu(sum.nid); 365 if (dn->inode->i_ino == nid) { 366 tdn.nid = nid; 367 if (!dn->inode_page_locked) 368 lock_page(dn->inode_page); 369 tdn.node_page = dn->inode_page; 370 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 371 goto truncate_out; 372 } else if (dn->nid == nid) { 373 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 374 goto truncate_out; 375 } 376 377 /* Get the node page */ 378 node_page = get_node_page(sbi, nid); 379 if (IS_ERR(node_page)) 380 return PTR_ERR(node_page); 381 382 offset = ofs_of_node(node_page); 383 ino = ino_of_node(node_page); 384 f2fs_put_page(node_page, 1); 385 386 if (ino != dn->inode->i_ino) { 387 int ret; 388 389 /* Deallocate previous index in the node page */ 390 inode = f2fs_iget_retry(sbi->sb, ino); 391 if (IS_ERR(inode)) 392 return PTR_ERR(inode); 393 394 ret = dquot_initialize(inode); 395 if (ret) { 396 iput(inode); 397 return ret; 398 } 399 } else { 400 inode = dn->inode; 401 } 402 403 bidx = start_bidx_of_node(offset, inode) + le16_to_cpu(sum.ofs_in_node); 404 405 /* 406 * if inode page is locked, unlock temporarily, but its reference 407 * count keeps alive. 408 */ 409 if (ino == dn->inode->i_ino && dn->inode_page_locked) 410 unlock_page(dn->inode_page); 411 412 set_new_dnode(&tdn, inode, NULL, NULL, 0); 413 if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE)) 414 goto out; 415 416 if (tdn.data_blkaddr == blkaddr) 417 truncate_data_blocks_range(&tdn, 1); 418 419 f2fs_put_dnode(&tdn); 420 out: 421 if (ino != dn->inode->i_ino) 422 iput(inode); 423 else if (dn->inode_page_locked) 424 lock_page(dn->inode_page); 425 return 0; 426 427 truncate_out: 428 if (datablock_addr(tdn.inode, tdn.node_page, 429 tdn.ofs_in_node) == blkaddr) 430 truncate_data_blocks_range(&tdn, 1); 431 if (dn->inode->i_ino == nid && !dn->inode_page_locked) 432 unlock_page(dn->inode_page); 433 return 0; 434 } 435 436 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode, 437 struct page *page) 438 { 439 struct dnode_of_data dn; 440 struct node_info ni; 441 unsigned int start, end; 442 int err = 0, recovered = 0; 443 444 /* step 1: recover xattr */ 445 if (IS_INODE(page)) { 446 recover_inline_xattr(inode, page); 447 } else if (f2fs_has_xattr_block(ofs_of_node(page))) { 448 err = recover_xattr_data(inode, page); 449 if (!err) 450 recovered++; 451 goto out; 452 } 453 454 /* step 2: recover inline data */ 455 if (recover_inline_data(inode, page)) 456 goto out; 457 458 /* step 3: recover data indices */ 459 start = start_bidx_of_node(ofs_of_node(page), inode); 460 end = start + ADDRS_PER_PAGE(page, inode); 461 462 set_new_dnode(&dn, inode, NULL, NULL, 0); 463 retry_dn: 464 err = get_dnode_of_data(&dn, start, ALLOC_NODE); 465 if (err) { 466 if (err == -ENOMEM) { 467 congestion_wait(BLK_RW_ASYNC, HZ/50); 468 goto retry_dn; 469 } 470 goto out; 471 } 472 473 f2fs_wait_on_page_writeback(dn.node_page, NODE, true); 474 475 get_node_info(sbi, dn.nid, &ni); 476 f2fs_bug_on(sbi, ni.ino != ino_of_node(page)); 477 f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page)); 478 479 for (; start < end; start++, dn.ofs_in_node++) { 480 block_t src, dest; 481 482 src = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node); 483 dest = datablock_addr(dn.inode, page, dn.ofs_in_node); 484 485 /* skip recovering if dest is the same as src */ 486 if (src == dest) 487 continue; 488 489 /* dest is invalid, just invalidate src block */ 490 if (dest == NULL_ADDR) { 491 truncate_data_blocks_range(&dn, 1); 492 continue; 493 } 494 495 if (!file_keep_isize(inode) && 496 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT))) 497 f2fs_i_size_write(inode, 498 (loff_t)(start + 1) << PAGE_SHIFT); 499 500 /* 501 * dest is reserved block, invalidate src block 502 * and then reserve one new block in dnode page. 503 */ 504 if (dest == NEW_ADDR) { 505 truncate_data_blocks_range(&dn, 1); 506 reserve_new_block(&dn); 507 continue; 508 } 509 510 /* dest is valid block, try to recover from src to dest */ 511 if (is_valid_blkaddr(sbi, dest, META_POR)) { 512 513 if (src == NULL_ADDR) { 514 err = reserve_new_block(&dn); 515 #ifdef CONFIG_F2FS_FAULT_INJECTION 516 while (err) 517 err = reserve_new_block(&dn); 518 #endif 519 /* We should not get -ENOSPC */ 520 f2fs_bug_on(sbi, err); 521 if (err) 522 goto err; 523 } 524 retry_prev: 525 /* Check the previous node page having this index */ 526 err = check_index_in_prev_nodes(sbi, dest, &dn); 527 if (err) { 528 if (err == -ENOMEM) { 529 congestion_wait(BLK_RW_ASYNC, HZ/50); 530 goto retry_prev; 531 } 532 goto err; 533 } 534 535 /* write dummy data page */ 536 f2fs_replace_block(sbi, &dn, src, dest, 537 ni.version, false, false); 538 recovered++; 539 } 540 } 541 542 copy_node_footer(dn.node_page, page); 543 fill_node_footer(dn.node_page, dn.nid, ni.ino, 544 ofs_of_node(page), false); 545 set_page_dirty(dn.node_page); 546 err: 547 f2fs_put_dnode(&dn); 548 out: 549 f2fs_msg(sbi->sb, KERN_NOTICE, 550 "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d", 551 inode->i_ino, 552 file_keep_isize(inode) ? "keep" : "recover", 553 recovered, err); 554 return err; 555 } 556 557 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list, 558 struct list_head *dir_list) 559 { 560 struct curseg_info *curseg; 561 struct page *page = NULL; 562 int err = 0; 563 block_t blkaddr; 564 565 /* get node pages in the current segment */ 566 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 567 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 568 569 while (1) { 570 struct fsync_inode_entry *entry; 571 572 if (!is_valid_blkaddr(sbi, blkaddr, META_POR)) 573 break; 574 575 ra_meta_pages_cond(sbi, blkaddr); 576 577 page = get_tmp_page(sbi, blkaddr); 578 579 if (!is_recoverable_dnode(page)) { 580 f2fs_put_page(page, 1); 581 break; 582 } 583 584 entry = get_fsync_inode(inode_list, ino_of_node(page)); 585 if (!entry) 586 goto next; 587 /* 588 * inode(x) | CP | inode(x) | dnode(F) 589 * In this case, we can lose the latest inode(x). 590 * So, call recover_inode for the inode update. 591 */ 592 if (IS_INODE(page)) 593 recover_inode(entry->inode, page); 594 if (entry->last_dentry == blkaddr) { 595 err = recover_dentry(entry->inode, page, dir_list); 596 if (err) { 597 f2fs_put_page(page, 1); 598 break; 599 } 600 } 601 err = do_recover_data(sbi, entry->inode, page); 602 if (err) { 603 f2fs_put_page(page, 1); 604 break; 605 } 606 607 if (entry->blkaddr == blkaddr) 608 del_fsync_inode(entry); 609 next: 610 /* check next segment */ 611 blkaddr = next_blkaddr_of_node(page); 612 f2fs_put_page(page, 1); 613 } 614 if (!err) 615 allocate_new_segments(sbi); 616 return err; 617 } 618 619 int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only) 620 { 621 struct list_head inode_list; 622 struct list_head dir_list; 623 int err; 624 int ret = 0; 625 unsigned long s_flags = sbi->sb->s_flags; 626 bool need_writecp = false; 627 #ifdef CONFIG_QUOTA 628 int quota_enabled; 629 #endif 630 631 if (s_flags & SB_RDONLY) { 632 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs"); 633 sbi->sb->s_flags &= ~SB_RDONLY; 634 } 635 636 #ifdef CONFIG_QUOTA 637 /* Needed for iput() to work correctly and not trash data */ 638 sbi->sb->s_flags |= SB_ACTIVE; 639 /* Turn on quotas so that they are updated correctly */ 640 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY); 641 #endif 642 643 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry", 644 sizeof(struct fsync_inode_entry)); 645 if (!fsync_entry_slab) { 646 err = -ENOMEM; 647 goto out; 648 } 649 650 INIT_LIST_HEAD(&inode_list); 651 INIT_LIST_HEAD(&dir_list); 652 653 /* prevent checkpoint */ 654 mutex_lock(&sbi->cp_mutex); 655 656 /* step #1: find fsynced inode numbers */ 657 err = find_fsync_dnodes(sbi, &inode_list, check_only); 658 if (err || list_empty(&inode_list)) 659 goto skip; 660 661 if (check_only) { 662 ret = 1; 663 goto skip; 664 } 665 666 need_writecp = true; 667 668 /* step #2: recover data */ 669 err = recover_data(sbi, &inode_list, &dir_list); 670 if (!err) 671 f2fs_bug_on(sbi, !list_empty(&inode_list)); 672 skip: 673 destroy_fsync_dnodes(&inode_list); 674 675 /* truncate meta pages to be used by the recovery */ 676 truncate_inode_pages_range(META_MAPPING(sbi), 677 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1); 678 679 if (err) { 680 truncate_inode_pages_final(NODE_MAPPING(sbi)); 681 truncate_inode_pages_final(META_MAPPING(sbi)); 682 } 683 684 clear_sbi_flag(sbi, SBI_POR_DOING); 685 mutex_unlock(&sbi->cp_mutex); 686 687 /* let's drop all the directory inodes for clean checkpoint */ 688 destroy_fsync_dnodes(&dir_list); 689 690 if (!err && need_writecp) { 691 struct cp_control cpc = { 692 .reason = CP_RECOVERY, 693 }; 694 err = write_checkpoint(sbi, &cpc); 695 } 696 697 kmem_cache_destroy(fsync_entry_slab); 698 out: 699 #ifdef CONFIG_QUOTA 700 /* Turn quotas off */ 701 if (quota_enabled) 702 f2fs_quota_off_umount(sbi->sb); 703 #endif 704 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 705 706 return ret ? ret: err; 707 } 708