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) 73 { 74 struct inode *inode; 75 struct fsync_inode_entry *entry; 76 77 inode = f2fs_iget_retry(sbi->sb, ino); 78 if (IS_ERR(inode)) 79 return ERR_CAST(inode); 80 81 entry = f2fs_kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO); 82 entry->inode = inode; 83 list_add_tail(&entry->list, head); 84 85 return entry; 86 } 87 88 static void del_fsync_inode(struct fsync_inode_entry *entry) 89 { 90 iput(entry->inode); 91 list_del(&entry->list); 92 kmem_cache_free(fsync_entry_slab, entry); 93 } 94 95 static int recover_dentry(struct inode *inode, struct page *ipage, 96 struct list_head *dir_list) 97 { 98 struct f2fs_inode *raw_inode = F2FS_INODE(ipage); 99 nid_t pino = le32_to_cpu(raw_inode->i_pino); 100 struct f2fs_dir_entry *de; 101 struct fscrypt_name fname; 102 struct page *page; 103 struct inode *dir, *einode; 104 struct fsync_inode_entry *entry; 105 int err = 0; 106 char *name; 107 108 entry = get_fsync_inode(dir_list, pino); 109 if (!entry) { 110 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list, pino); 111 if (IS_ERR(entry)) { 112 dir = ERR_CAST(entry); 113 err = PTR_ERR(entry); 114 goto out; 115 } 116 } 117 118 dir = entry->inode; 119 120 memset(&fname, 0, sizeof(struct fscrypt_name)); 121 fname.disk_name.len = le32_to_cpu(raw_inode->i_namelen); 122 fname.disk_name.name = raw_inode->i_name; 123 124 if (unlikely(fname.disk_name.len > F2FS_NAME_LEN)) { 125 WARN_ON(1); 126 err = -ENAMETOOLONG; 127 goto out; 128 } 129 retry: 130 de = __f2fs_find_entry(dir, &fname, &page); 131 if (de && inode->i_ino == le32_to_cpu(de->ino)) 132 goto out_unmap_put; 133 134 if (de) { 135 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino)); 136 if (IS_ERR(einode)) { 137 WARN_ON(1); 138 err = PTR_ERR(einode); 139 if (err == -ENOENT) 140 err = -EEXIST; 141 goto out_unmap_put; 142 } 143 err = acquire_orphan_inode(F2FS_I_SB(inode)); 144 if (err) { 145 iput(einode); 146 goto out_unmap_put; 147 } 148 f2fs_delete_entry(de, page, dir, einode); 149 iput(einode); 150 goto retry; 151 } else if (IS_ERR(page)) { 152 err = PTR_ERR(page); 153 } else { 154 err = __f2fs_do_add_link(dir, &fname, inode, 155 inode->i_ino, inode->i_mode); 156 } 157 if (err == -ENOMEM) 158 goto retry; 159 goto out; 160 161 out_unmap_put: 162 f2fs_dentry_kunmap(dir, page); 163 f2fs_put_page(page, 0); 164 out: 165 if (file_enc_name(inode)) 166 name = "<encrypted>"; 167 else 168 name = raw_inode->i_name; 169 f2fs_msg(inode->i_sb, KERN_NOTICE, 170 "%s: ino = %x, name = %s, dir = %lx, err = %d", 171 __func__, ino_of_node(ipage), name, 172 IS_ERR(dir) ? 0 : dir->i_ino, err); 173 return err; 174 } 175 176 static void recover_inode(struct inode *inode, struct page *page) 177 { 178 struct f2fs_inode *raw = F2FS_INODE(page); 179 char *name; 180 181 inode->i_mode = le16_to_cpu(raw->i_mode); 182 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size)); 183 inode->i_atime.tv_sec = le64_to_cpu(raw->i_mtime); 184 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime); 185 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime); 186 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec); 187 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec); 188 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec); 189 190 if (file_enc_name(inode)) 191 name = "<encrypted>"; 192 else 193 name = F2FS_INODE(page)->i_name; 194 195 f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s", 196 ino_of_node(page), name); 197 } 198 199 static bool is_same_inode(struct inode *inode, struct page *ipage) 200 { 201 struct f2fs_inode *ri = F2FS_INODE(ipage); 202 struct timespec disk; 203 204 if (!IS_INODE(ipage)) 205 return true; 206 207 disk.tv_sec = le64_to_cpu(ri->i_ctime); 208 disk.tv_nsec = le32_to_cpu(ri->i_ctime_nsec); 209 if (timespec_compare(&inode->i_ctime, &disk) > 0) 210 return false; 211 212 disk.tv_sec = le64_to_cpu(ri->i_atime); 213 disk.tv_nsec = le32_to_cpu(ri->i_atime_nsec); 214 if (timespec_compare(&inode->i_atime, &disk) > 0) 215 return false; 216 217 disk.tv_sec = le64_to_cpu(ri->i_mtime); 218 disk.tv_nsec = le32_to_cpu(ri->i_mtime_nsec); 219 if (timespec_compare(&inode->i_mtime, &disk) > 0) 220 return false; 221 222 return true; 223 } 224 225 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head) 226 { 227 struct curseg_info *curseg; 228 struct page *page = NULL; 229 block_t blkaddr; 230 int err = 0; 231 232 /* get node pages in the current segment */ 233 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 234 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 235 236 while (1) { 237 struct fsync_inode_entry *entry; 238 239 if (!is_valid_blkaddr(sbi, blkaddr, META_POR)) 240 return 0; 241 242 page = get_tmp_page(sbi, blkaddr); 243 244 if (!is_recoverable_dnode(page)) 245 break; 246 247 if (!is_fsync_dnode(page)) 248 goto next; 249 250 entry = get_fsync_inode(head, ino_of_node(page)); 251 if (entry) { 252 if (!is_same_inode(entry->inode, page)) 253 goto next; 254 } else { 255 if (IS_INODE(page) && is_dent_dnode(page)) { 256 err = recover_inode_page(sbi, page); 257 if (err) 258 break; 259 } 260 261 /* 262 * CP | dnode(F) | inode(DF) 263 * For this case, we should not give up now. 264 */ 265 entry = add_fsync_inode(sbi, head, ino_of_node(page)); 266 if (IS_ERR(entry)) { 267 err = PTR_ERR(entry); 268 if (err == -ENOENT) { 269 err = 0; 270 goto next; 271 } 272 break; 273 } 274 } 275 entry->blkaddr = blkaddr; 276 277 if (IS_INODE(page) && is_dent_dnode(page)) 278 entry->last_dentry = blkaddr; 279 next: 280 /* check next segment */ 281 blkaddr = next_blkaddr_of_node(page); 282 f2fs_put_page(page, 1); 283 284 ra_meta_pages_cond(sbi, blkaddr); 285 } 286 f2fs_put_page(page, 1); 287 return err; 288 } 289 290 static void destroy_fsync_dnodes(struct list_head *head) 291 { 292 struct fsync_inode_entry *entry, *tmp; 293 294 list_for_each_entry_safe(entry, tmp, head, list) 295 del_fsync_inode(entry); 296 } 297 298 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi, 299 block_t blkaddr, struct dnode_of_data *dn) 300 { 301 struct seg_entry *sentry; 302 unsigned int segno = GET_SEGNO(sbi, blkaddr); 303 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 304 struct f2fs_summary_block *sum_node; 305 struct f2fs_summary sum; 306 struct page *sum_page, *node_page; 307 struct dnode_of_data tdn = *dn; 308 nid_t ino, nid; 309 struct inode *inode; 310 unsigned int offset; 311 block_t bidx; 312 int i; 313 314 sentry = get_seg_entry(sbi, segno); 315 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map)) 316 return 0; 317 318 /* Get the previous summary */ 319 for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) { 320 struct curseg_info *curseg = CURSEG_I(sbi, i); 321 if (curseg->segno == segno) { 322 sum = curseg->sum_blk->entries[blkoff]; 323 goto got_it; 324 } 325 } 326 327 sum_page = get_sum_page(sbi, segno); 328 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 329 sum = sum_node->entries[blkoff]; 330 f2fs_put_page(sum_page, 1); 331 got_it: 332 /* Use the locked dnode page and inode */ 333 nid = le32_to_cpu(sum.nid); 334 if (dn->inode->i_ino == nid) { 335 tdn.nid = nid; 336 if (!dn->inode_page_locked) 337 lock_page(dn->inode_page); 338 tdn.node_page = dn->inode_page; 339 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 340 goto truncate_out; 341 } else if (dn->nid == nid) { 342 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 343 goto truncate_out; 344 } 345 346 /* Get the node page */ 347 node_page = get_node_page(sbi, nid); 348 if (IS_ERR(node_page)) 349 return PTR_ERR(node_page); 350 351 offset = ofs_of_node(node_page); 352 ino = ino_of_node(node_page); 353 f2fs_put_page(node_page, 1); 354 355 if (ino != dn->inode->i_ino) { 356 /* Deallocate previous index in the node page */ 357 inode = f2fs_iget_retry(sbi->sb, ino); 358 if (IS_ERR(inode)) 359 return PTR_ERR(inode); 360 } else { 361 inode = dn->inode; 362 } 363 364 bidx = start_bidx_of_node(offset, inode) + le16_to_cpu(sum.ofs_in_node); 365 366 /* 367 * if inode page is locked, unlock temporarily, but its reference 368 * count keeps alive. 369 */ 370 if (ino == dn->inode->i_ino && dn->inode_page_locked) 371 unlock_page(dn->inode_page); 372 373 set_new_dnode(&tdn, inode, NULL, NULL, 0); 374 if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE)) 375 goto out; 376 377 if (tdn.data_blkaddr == blkaddr) 378 truncate_data_blocks_range(&tdn, 1); 379 380 f2fs_put_dnode(&tdn); 381 out: 382 if (ino != dn->inode->i_ino) 383 iput(inode); 384 else if (dn->inode_page_locked) 385 lock_page(dn->inode_page); 386 return 0; 387 388 truncate_out: 389 if (datablock_addr(tdn.node_page, tdn.ofs_in_node) == blkaddr) 390 truncate_data_blocks_range(&tdn, 1); 391 if (dn->inode->i_ino == nid && !dn->inode_page_locked) 392 unlock_page(dn->inode_page); 393 return 0; 394 } 395 396 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode, 397 struct page *page, block_t blkaddr) 398 { 399 struct dnode_of_data dn; 400 struct node_info ni; 401 unsigned int start, end; 402 int err = 0, recovered = 0; 403 404 /* step 1: recover xattr */ 405 if (IS_INODE(page)) { 406 recover_inline_xattr(inode, page); 407 } else if (f2fs_has_xattr_block(ofs_of_node(page))) { 408 /* 409 * Deprecated; xattr blocks should be found from cold log. 410 * But, we should remain this for backward compatibility. 411 */ 412 recover_xattr_data(inode, page, blkaddr); 413 goto out; 414 } 415 416 /* step 2: recover inline data */ 417 if (recover_inline_data(inode, page)) 418 goto out; 419 420 /* step 3: recover data indices */ 421 start = start_bidx_of_node(ofs_of_node(page), inode); 422 end = start + ADDRS_PER_PAGE(page, inode); 423 424 set_new_dnode(&dn, inode, NULL, NULL, 0); 425 retry_dn: 426 err = get_dnode_of_data(&dn, start, ALLOC_NODE); 427 if (err) { 428 if (err == -ENOMEM) { 429 congestion_wait(BLK_RW_ASYNC, HZ/50); 430 goto retry_dn; 431 } 432 goto out; 433 } 434 435 f2fs_wait_on_page_writeback(dn.node_page, NODE, true); 436 437 get_node_info(sbi, dn.nid, &ni); 438 f2fs_bug_on(sbi, ni.ino != ino_of_node(page)); 439 f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page)); 440 441 for (; start < end; start++, dn.ofs_in_node++) { 442 block_t src, dest; 443 444 src = datablock_addr(dn.node_page, dn.ofs_in_node); 445 dest = datablock_addr(page, dn.ofs_in_node); 446 447 /* skip recovering if dest is the same as src */ 448 if (src == dest) 449 continue; 450 451 /* dest is invalid, just invalidate src block */ 452 if (dest == NULL_ADDR) { 453 truncate_data_blocks_range(&dn, 1); 454 continue; 455 } 456 457 if ((start + 1) << PAGE_SHIFT > i_size_read(inode)) 458 f2fs_i_size_write(inode, (start + 1) << PAGE_SHIFT); 459 460 /* 461 * dest is reserved block, invalidate src block 462 * and then reserve one new block in dnode page. 463 */ 464 if (dest == NEW_ADDR) { 465 truncate_data_blocks_range(&dn, 1); 466 reserve_new_block(&dn); 467 continue; 468 } 469 470 /* dest is valid block, try to recover from src to dest */ 471 if (is_valid_blkaddr(sbi, dest, META_POR)) { 472 473 if (src == NULL_ADDR) { 474 err = reserve_new_block(&dn); 475 #ifdef CONFIG_F2FS_FAULT_INJECTION 476 while (err) 477 err = reserve_new_block(&dn); 478 #endif 479 /* We should not get -ENOSPC */ 480 f2fs_bug_on(sbi, err); 481 if (err) 482 goto err; 483 } 484 retry_prev: 485 /* Check the previous node page having this index */ 486 err = check_index_in_prev_nodes(sbi, dest, &dn); 487 if (err) { 488 if (err == -ENOMEM) { 489 congestion_wait(BLK_RW_ASYNC, HZ/50); 490 goto retry_prev; 491 } 492 goto err; 493 } 494 495 /* write dummy data page */ 496 f2fs_replace_block(sbi, &dn, src, dest, 497 ni.version, false, false); 498 recovered++; 499 } 500 } 501 502 copy_node_footer(dn.node_page, page); 503 fill_node_footer(dn.node_page, dn.nid, ni.ino, 504 ofs_of_node(page), false); 505 set_page_dirty(dn.node_page); 506 err: 507 f2fs_put_dnode(&dn); 508 out: 509 f2fs_msg(sbi->sb, KERN_NOTICE, 510 "recover_data: ino = %lx, recovered = %d blocks, err = %d", 511 inode->i_ino, recovered, err); 512 return err; 513 } 514 515 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list, 516 struct list_head *dir_list) 517 { 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 (!is_recoverable_dnode(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 set_ckpt_flags(sbi, CP_ERROR_FLAG); 631 mutex_unlock(&sbi->cp_mutex); 632 633 /* let's drop all the directory inodes for clean checkpoint */ 634 destroy_fsync_dnodes(&dir_list); 635 636 if (!err && need_writecp) { 637 struct cp_control cpc = { 638 .reason = CP_RECOVERY, 639 }; 640 err = write_checkpoint(sbi, &cpc); 641 } 642 643 kmem_cache_destroy(fsync_entry_slab); 644 return ret ? ret: err; 645 } 646