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_unmap_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_unmap_put; 157 } 158 159 err = dquot_initialize(einode); 160 if (err) { 161 iput(einode); 162 goto out_unmap_put; 163 } 164 165 err = acquire_orphan_inode(F2FS_I_SB(inode)); 166 if (err) { 167 iput(einode); 168 goto out_unmap_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_unmap_put: 184 f2fs_dentry_kunmap(dir, page); 185 f2fs_put_page(page, 0); 186 out: 187 if (file_enc_name(inode)) 188 name = "<encrypted>"; 189 else 190 name = raw_inode->i_name; 191 f2fs_msg(inode->i_sb, KERN_NOTICE, 192 "%s: ino = %x, name = %s, dir = %lx, err = %d", 193 __func__, ino_of_node(ipage), name, 194 IS_ERR(dir) ? 0 : dir->i_ino, err); 195 return err; 196 } 197 198 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri) 199 { 200 if (ri->i_inline & F2FS_PIN_FILE) 201 set_inode_flag(inode, FI_PIN_FILE); 202 else 203 clear_inode_flag(inode, FI_PIN_FILE); 204 if (ri->i_inline & F2FS_DATA_EXIST) 205 set_inode_flag(inode, FI_DATA_EXIST); 206 else 207 clear_inode_flag(inode, FI_DATA_EXIST); 208 if (!(ri->i_inline & F2FS_INLINE_DOTS)) 209 clear_inode_flag(inode, FI_INLINE_DOTS); 210 } 211 212 static void recover_inode(struct inode *inode, struct page *page) 213 { 214 struct f2fs_inode *raw = F2FS_INODE(page); 215 char *name; 216 217 inode->i_mode = le16_to_cpu(raw->i_mode); 218 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size)); 219 inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime); 220 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime); 221 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime); 222 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec); 223 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec); 224 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec); 225 226 F2FS_I(inode)->i_advise = raw->i_advise; 227 228 recover_inline_flags(inode, raw); 229 230 if (file_enc_name(inode)) 231 name = "<encrypted>"; 232 else 233 name = F2FS_INODE(page)->i_name; 234 235 f2fs_msg(inode->i_sb, KERN_NOTICE, 236 "recover_inode: ino = %x, name = %s, inline = %x", 237 ino_of_node(page), name, raw->i_inline); 238 } 239 240 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head, 241 bool check_only) 242 { 243 struct curseg_info *curseg; 244 struct page *page = NULL; 245 block_t blkaddr; 246 int err = 0; 247 248 /* get node pages in the current segment */ 249 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 250 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 251 252 while (1) { 253 struct fsync_inode_entry *entry; 254 255 if (!is_valid_blkaddr(sbi, blkaddr, META_POR)) 256 return 0; 257 258 page = get_tmp_page(sbi, blkaddr); 259 260 if (!is_recoverable_dnode(page)) 261 break; 262 263 if (!is_fsync_dnode(page)) 264 goto next; 265 266 entry = get_fsync_inode(head, ino_of_node(page)); 267 if (!entry) { 268 bool quota_inode = false; 269 270 if (!check_only && 271 IS_INODE(page) && is_dent_dnode(page)) { 272 err = recover_inode_page(sbi, page); 273 if (err) 274 break; 275 quota_inode = true; 276 } 277 278 /* 279 * CP | dnode(F) | inode(DF) 280 * For this case, we should not give up now. 281 */ 282 entry = add_fsync_inode(sbi, head, ino_of_node(page), 283 quota_inode); 284 if (IS_ERR(entry)) { 285 err = PTR_ERR(entry); 286 if (err == -ENOENT) { 287 err = 0; 288 goto next; 289 } 290 break; 291 } 292 } 293 entry->blkaddr = blkaddr; 294 295 if (IS_INODE(page) && is_dent_dnode(page)) 296 entry->last_dentry = blkaddr; 297 next: 298 /* check next segment */ 299 blkaddr = next_blkaddr_of_node(page); 300 f2fs_put_page(page, 1); 301 302 ra_meta_pages_cond(sbi, blkaddr); 303 } 304 f2fs_put_page(page, 1); 305 return err; 306 } 307 308 static void destroy_fsync_dnodes(struct list_head *head) 309 { 310 struct fsync_inode_entry *entry, *tmp; 311 312 list_for_each_entry_safe(entry, tmp, head, list) 313 del_fsync_inode(entry); 314 } 315 316 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi, 317 block_t blkaddr, struct dnode_of_data *dn) 318 { 319 struct seg_entry *sentry; 320 unsigned int segno = GET_SEGNO(sbi, blkaddr); 321 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 322 struct f2fs_summary_block *sum_node; 323 struct f2fs_summary sum; 324 struct page *sum_page, *node_page; 325 struct dnode_of_data tdn = *dn; 326 nid_t ino, nid; 327 struct inode *inode; 328 unsigned int offset; 329 block_t bidx; 330 int i; 331 332 sentry = get_seg_entry(sbi, segno); 333 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map)) 334 return 0; 335 336 /* Get the previous summary */ 337 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 338 struct curseg_info *curseg = CURSEG_I(sbi, i); 339 if (curseg->segno == segno) { 340 sum = curseg->sum_blk->entries[blkoff]; 341 goto got_it; 342 } 343 } 344 345 sum_page = get_sum_page(sbi, segno); 346 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 347 sum = sum_node->entries[blkoff]; 348 f2fs_put_page(sum_page, 1); 349 got_it: 350 /* Use the locked dnode page and inode */ 351 nid = le32_to_cpu(sum.nid); 352 if (dn->inode->i_ino == nid) { 353 tdn.nid = nid; 354 if (!dn->inode_page_locked) 355 lock_page(dn->inode_page); 356 tdn.node_page = dn->inode_page; 357 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 358 goto truncate_out; 359 } else if (dn->nid == nid) { 360 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 361 goto truncate_out; 362 } 363 364 /* Get the node page */ 365 node_page = get_node_page(sbi, nid); 366 if (IS_ERR(node_page)) 367 return PTR_ERR(node_page); 368 369 offset = ofs_of_node(node_page); 370 ino = ino_of_node(node_page); 371 f2fs_put_page(node_page, 1); 372 373 if (ino != dn->inode->i_ino) { 374 int ret; 375 376 /* Deallocate previous index in the node page */ 377 inode = f2fs_iget_retry(sbi->sb, ino); 378 if (IS_ERR(inode)) 379 return PTR_ERR(inode); 380 381 ret = dquot_initialize(inode); 382 if (ret) { 383 iput(inode); 384 return ret; 385 } 386 } else { 387 inode = dn->inode; 388 } 389 390 bidx = start_bidx_of_node(offset, inode) + le16_to_cpu(sum.ofs_in_node); 391 392 /* 393 * if inode page is locked, unlock temporarily, but its reference 394 * count keeps alive. 395 */ 396 if (ino == dn->inode->i_ino && dn->inode_page_locked) 397 unlock_page(dn->inode_page); 398 399 set_new_dnode(&tdn, inode, NULL, NULL, 0); 400 if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE)) 401 goto out; 402 403 if (tdn.data_blkaddr == blkaddr) 404 truncate_data_blocks_range(&tdn, 1); 405 406 f2fs_put_dnode(&tdn); 407 out: 408 if (ino != dn->inode->i_ino) 409 iput(inode); 410 else if (dn->inode_page_locked) 411 lock_page(dn->inode_page); 412 return 0; 413 414 truncate_out: 415 if (datablock_addr(tdn.inode, tdn.node_page, 416 tdn.ofs_in_node) == blkaddr) 417 truncate_data_blocks_range(&tdn, 1); 418 if (dn->inode->i_ino == nid && !dn->inode_page_locked) 419 unlock_page(dn->inode_page); 420 return 0; 421 } 422 423 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode, 424 struct page *page) 425 { 426 struct dnode_of_data dn; 427 struct node_info ni; 428 unsigned int start, end; 429 int err = 0, recovered = 0; 430 431 /* step 1: recover xattr */ 432 if (IS_INODE(page)) { 433 recover_inline_xattr(inode, page); 434 } else if (f2fs_has_xattr_block(ofs_of_node(page))) { 435 err = recover_xattr_data(inode, page); 436 if (!err) 437 recovered++; 438 goto out; 439 } 440 441 /* step 2: recover inline data */ 442 if (recover_inline_data(inode, page)) 443 goto out; 444 445 /* step 3: recover data indices */ 446 start = start_bidx_of_node(ofs_of_node(page), inode); 447 end = start + ADDRS_PER_PAGE(page, inode); 448 449 set_new_dnode(&dn, inode, NULL, NULL, 0); 450 retry_dn: 451 err = get_dnode_of_data(&dn, start, ALLOC_NODE); 452 if (err) { 453 if (err == -ENOMEM) { 454 congestion_wait(BLK_RW_ASYNC, HZ/50); 455 goto retry_dn; 456 } 457 goto out; 458 } 459 460 f2fs_wait_on_page_writeback(dn.node_page, NODE, true); 461 462 get_node_info(sbi, dn.nid, &ni); 463 f2fs_bug_on(sbi, ni.ino != ino_of_node(page)); 464 f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page)); 465 466 for (; start < end; start++, dn.ofs_in_node++) { 467 block_t src, dest; 468 469 src = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node); 470 dest = datablock_addr(dn.inode, page, dn.ofs_in_node); 471 472 /* skip recovering if dest is the same as src */ 473 if (src == dest) 474 continue; 475 476 /* dest is invalid, just invalidate src block */ 477 if (dest == NULL_ADDR) { 478 truncate_data_blocks_range(&dn, 1); 479 continue; 480 } 481 482 if (!file_keep_isize(inode) && 483 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT))) 484 f2fs_i_size_write(inode, 485 (loff_t)(start + 1) << PAGE_SHIFT); 486 487 /* 488 * dest is reserved block, invalidate src block 489 * and then reserve one new block in dnode page. 490 */ 491 if (dest == NEW_ADDR) { 492 truncate_data_blocks_range(&dn, 1); 493 reserve_new_block(&dn); 494 continue; 495 } 496 497 /* dest is valid block, try to recover from src to dest */ 498 if (is_valid_blkaddr(sbi, dest, META_POR)) { 499 500 if (src == NULL_ADDR) { 501 err = reserve_new_block(&dn); 502 #ifdef CONFIG_F2FS_FAULT_INJECTION 503 while (err) 504 err = reserve_new_block(&dn); 505 #endif 506 /* We should not get -ENOSPC */ 507 f2fs_bug_on(sbi, err); 508 if (err) 509 goto err; 510 } 511 retry_prev: 512 /* Check the previous node page having this index */ 513 err = check_index_in_prev_nodes(sbi, dest, &dn); 514 if (err) { 515 if (err == -ENOMEM) { 516 congestion_wait(BLK_RW_ASYNC, HZ/50); 517 goto retry_prev; 518 } 519 goto err; 520 } 521 522 /* write dummy data page */ 523 f2fs_replace_block(sbi, &dn, src, dest, 524 ni.version, false, false); 525 recovered++; 526 } 527 } 528 529 copy_node_footer(dn.node_page, page); 530 fill_node_footer(dn.node_page, dn.nid, ni.ino, 531 ofs_of_node(page), false); 532 set_page_dirty(dn.node_page); 533 err: 534 f2fs_put_dnode(&dn); 535 out: 536 f2fs_msg(sbi->sb, KERN_NOTICE, 537 "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d", 538 inode->i_ino, 539 file_keep_isize(inode) ? "keep" : "recover", 540 recovered, err); 541 return err; 542 } 543 544 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list, 545 struct list_head *dir_list) 546 { 547 struct curseg_info *curseg; 548 struct page *page = NULL; 549 int err = 0; 550 block_t blkaddr; 551 552 /* get node pages in the current segment */ 553 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 554 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 555 556 while (1) { 557 struct fsync_inode_entry *entry; 558 559 if (!is_valid_blkaddr(sbi, blkaddr, META_POR)) 560 break; 561 562 ra_meta_pages_cond(sbi, blkaddr); 563 564 page = get_tmp_page(sbi, blkaddr); 565 566 if (!is_recoverable_dnode(page)) { 567 f2fs_put_page(page, 1); 568 break; 569 } 570 571 entry = get_fsync_inode(inode_list, ino_of_node(page)); 572 if (!entry) 573 goto next; 574 /* 575 * inode(x) | CP | inode(x) | dnode(F) 576 * In this case, we can lose the latest inode(x). 577 * So, call recover_inode for the inode update. 578 */ 579 if (IS_INODE(page)) 580 recover_inode(entry->inode, page); 581 if (entry->last_dentry == blkaddr) { 582 err = recover_dentry(entry->inode, page, dir_list); 583 if (err) { 584 f2fs_put_page(page, 1); 585 break; 586 } 587 } 588 err = do_recover_data(sbi, entry->inode, page); 589 if (err) { 590 f2fs_put_page(page, 1); 591 break; 592 } 593 594 if (entry->blkaddr == blkaddr) 595 del_fsync_inode(entry); 596 next: 597 /* check next segment */ 598 blkaddr = next_blkaddr_of_node(page); 599 f2fs_put_page(page, 1); 600 } 601 if (!err) 602 allocate_new_segments(sbi); 603 return err; 604 } 605 606 int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only) 607 { 608 struct list_head inode_list; 609 struct list_head dir_list; 610 int err; 611 int ret = 0; 612 unsigned long s_flags = sbi->sb->s_flags; 613 bool need_writecp = false; 614 #ifdef CONFIG_QUOTA 615 int quota_enabled; 616 #endif 617 618 if (s_flags & SB_RDONLY) { 619 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs"); 620 sbi->sb->s_flags &= ~SB_RDONLY; 621 } 622 623 #ifdef CONFIG_QUOTA 624 /* Needed for iput() to work correctly and not trash data */ 625 sbi->sb->s_flags |= SB_ACTIVE; 626 /* Turn on quotas so that they are updated correctly */ 627 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY); 628 #endif 629 630 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry", 631 sizeof(struct fsync_inode_entry)); 632 if (!fsync_entry_slab) { 633 err = -ENOMEM; 634 goto out; 635 } 636 637 INIT_LIST_HEAD(&inode_list); 638 INIT_LIST_HEAD(&dir_list); 639 640 /* prevent checkpoint */ 641 mutex_lock(&sbi->cp_mutex); 642 643 /* step #1: find fsynced inode numbers */ 644 err = find_fsync_dnodes(sbi, &inode_list, check_only); 645 if (err || list_empty(&inode_list)) 646 goto skip; 647 648 if (check_only) { 649 ret = 1; 650 goto skip; 651 } 652 653 need_writecp = true; 654 655 /* step #2: recover data */ 656 err = recover_data(sbi, &inode_list, &dir_list); 657 if (!err) 658 f2fs_bug_on(sbi, !list_empty(&inode_list)); 659 skip: 660 destroy_fsync_dnodes(&inode_list); 661 662 /* truncate meta pages to be used by the recovery */ 663 truncate_inode_pages_range(META_MAPPING(sbi), 664 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1); 665 666 if (err) { 667 truncate_inode_pages_final(NODE_MAPPING(sbi)); 668 truncate_inode_pages_final(META_MAPPING(sbi)); 669 } 670 671 clear_sbi_flag(sbi, SBI_POR_DOING); 672 mutex_unlock(&sbi->cp_mutex); 673 674 /* let's drop all the directory inodes for clean checkpoint */ 675 destroy_fsync_dnodes(&dir_list); 676 677 if (!err && need_writecp) { 678 struct cp_control cpc = { 679 .reason = CP_RECOVERY, 680 }; 681 err = write_checkpoint(sbi, &cpc); 682 } 683 684 kmem_cache_destroy(fsync_entry_slab); 685 out: 686 #ifdef CONFIG_QUOTA 687 /* Turn quotas off */ 688 if (quota_enabled) 689 f2fs_quota_off_umount(sbi->sb); 690 #endif 691 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 692 693 return ret ? ret: err; 694 } 695