1// SPDX-License-Identifier: GPL-2.0 2/* 3 * fs/f2fs/recovery.c 4 * 5 * Copyright (c) 2012 Samsung Electronics Co., Ltd. 6 * http://www.samsung.com/ 7 */ 8#include <linux/fs.h> 9#include <linux/f2fs_fs.h> 10#include "f2fs.h" 11#include "node.h" 12#include "segment.h" 13 14/* 15 * Roll forward recovery scenarios. 16 * 17 * [Term] F: fsync_mark, D: dentry_mark 18 * 19 * 1. inode(x) | CP | inode(x) | dnode(F) 20 * -> Update the latest inode(x). 21 * 22 * 2. inode(x) | CP | inode(F) | dnode(F) 23 * -> No problem. 24 * 25 * 3. inode(x) | CP | dnode(F) | inode(x) 26 * -> Recover to the latest dnode(F), and drop the last inode(x) 27 * 28 * 4. inode(x) | CP | dnode(F) | inode(F) 29 * -> No problem. 30 * 31 * 5. CP | inode(x) | dnode(F) 32 * -> The inode(DF) was missing. Should drop this dnode(F). 33 * 34 * 6. CP | inode(DF) | dnode(F) 35 * -> No problem. 36 * 37 * 7. CP | dnode(F) | inode(DF) 38 * -> If f2fs_iget fails, then goto next to find inode(DF). 39 * 40 * 8. CP | dnode(F) | inode(x) 41 * -> If f2fs_iget fails, then goto next to find inode(DF). 42 * But it will fail due to no inode(DF). 43 */ 44 45static struct kmem_cache *fsync_entry_slab; 46 47bool f2fs_space_for_roll_forward(struct f2fs_sb_info *sbi) 48{ 49 s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count); 50 51 if (sbi->last_valid_block_count + nalloc > sbi->user_block_count) 52 return false; 53 return true; 54} 55 56static struct fsync_inode_entry *get_fsync_inode(struct list_head *head, 57 nid_t ino) 58{ 59 struct fsync_inode_entry *entry; 60 61 list_for_each_entry(entry, head, list) 62 if (entry->inode->i_ino == ino) 63 return entry; 64 65 return NULL; 66} 67 68static struct fsync_inode_entry *add_fsync_inode(struct f2fs_sb_info *sbi, 69 struct list_head *head, nid_t ino, bool quota_inode) 70{ 71 struct inode *inode; 72 struct fsync_inode_entry *entry; 73 int err; 74 75 inode = f2fs_iget_retry(sbi->sb, ino); 76 if (IS_ERR(inode)) 77 return ERR_CAST(inode); 78 79 err = dquot_initialize(inode); 80 if (err) 81 goto err_out; 82 83 if (quota_inode) { 84 err = dquot_alloc_inode(inode); 85 if (err) 86 goto err_out; 87 } 88 89 entry = f2fs_kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO); 90 entry->inode = inode; 91 list_add_tail(&entry->list, head); 92 93 return entry; 94err_out: 95 iput(inode); 96 return ERR_PTR(err); 97} 98 99static void del_fsync_inode(struct fsync_inode_entry *entry, int drop) 100{ 101 if (drop) { 102 /* inode should not be recovered, drop it */ 103 f2fs_inode_synced(entry->inode); 104 } 105 iput(entry->inode); 106 list_del(&entry->list); 107 kmem_cache_free(fsync_entry_slab, entry); 108} 109 110static int init_recovered_filename(const struct inode *dir, 111 struct f2fs_inode *raw_inode, 112 struct f2fs_filename *fname, 113 struct qstr *usr_fname) 114{ 115 int err; 116 117 memset(fname, 0, sizeof(*fname)); 118 fname->disk_name.len = le32_to_cpu(raw_inode->i_namelen); 119 fname->disk_name.name = raw_inode->i_name; 120 121 if (WARN_ON(fname->disk_name.len > F2FS_NAME_LEN)) 122 return -ENAMETOOLONG; 123 124 if (!IS_ENCRYPTED(dir)) { 125 usr_fname->name = fname->disk_name.name; 126 usr_fname->len = fname->disk_name.len; 127 fname->usr_fname = usr_fname; 128 } 129 130 /* Compute the hash of the filename */ 131 if (IS_CASEFOLDED(dir)) { 132 err = f2fs_init_casefolded_name(dir, fname); 133 if (err) 134 return err; 135 f2fs_hash_filename(dir, fname); 136#ifdef CONFIG_UNICODE 137 /* Case-sensitive match is fine for recovery */ 138 kfree(fname->cf_name.name); 139 fname->cf_name.name = NULL; 140#endif 141 } else { 142 f2fs_hash_filename(dir, fname); 143 } 144 return 0; 145} 146 147static int recover_dentry(struct inode *inode, struct page *ipage, 148 struct list_head *dir_list) 149{ 150 struct f2fs_inode *raw_inode = F2FS_INODE(ipage); 151 nid_t pino = le32_to_cpu(raw_inode->i_pino); 152 struct f2fs_dir_entry *de; 153 struct f2fs_filename fname; 154 struct qstr usr_fname; 155 struct page *page; 156 struct inode *dir, *einode; 157 struct fsync_inode_entry *entry; 158 int err = 0; 159 char *name; 160 161 entry = get_fsync_inode(dir_list, pino); 162 if (!entry) { 163 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list, 164 pino, false); 165 if (IS_ERR(entry)) { 166 dir = ERR_CAST(entry); 167 err = PTR_ERR(entry); 168 goto out; 169 } 170 } 171 172 dir = entry->inode; 173 err = init_recovered_filename(dir, raw_inode, &fname, &usr_fname); 174 if (err) 175 goto out; 176retry: 177 de = __f2fs_find_entry(dir, &fname, &page); 178 if (de && inode->i_ino == le32_to_cpu(de->ino)) 179 goto out_put; 180 181 if (de) { 182 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino)); 183 if (IS_ERR(einode)) { 184 WARN_ON(1); 185 err = PTR_ERR(einode); 186 if (err == -ENOENT) 187 err = -EEXIST; 188 goto out_put; 189 } 190 191 err = dquot_initialize(einode); 192 if (err) { 193 iput(einode); 194 goto out_put; 195 } 196 197 err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode)); 198 if (err) { 199 iput(einode); 200 goto out_put; 201 } 202 f2fs_delete_entry(de, page, dir, einode); 203 iput(einode); 204 goto retry; 205 } else if (IS_ERR(page)) { 206 err = PTR_ERR(page); 207 } else { 208 err = f2fs_add_dentry(dir, &fname, inode, 209 inode->i_ino, inode->i_mode); 210 } 211 if (err == -ENOMEM) 212 goto retry; 213 goto out; 214 215out_put: 216 f2fs_put_page(page, 0); 217out: 218 if (file_enc_name(inode)) 219 name = "<encrypted>"; 220 else 221 name = raw_inode->i_name; 222 f2fs_notice(F2FS_I_SB(inode), "%s: ino = %x, name = %s, dir = %lx, err = %d", 223 __func__, ino_of_node(ipage), name, 224 IS_ERR(dir) ? 0 : dir->i_ino, err); 225 return err; 226} 227 228static int recover_quota_data(struct inode *inode, struct page *page) 229{ 230 struct f2fs_inode *raw = F2FS_INODE(page); 231 struct iattr attr; 232 uid_t i_uid = le32_to_cpu(raw->i_uid); 233 gid_t i_gid = le32_to_cpu(raw->i_gid); 234 int err; 235 236 memset(&attr, 0, sizeof(attr)); 237 238 attr.ia_uid = make_kuid(inode->i_sb->s_user_ns, i_uid); 239 attr.ia_gid = make_kgid(inode->i_sb->s_user_ns, i_gid); 240 241 if (!uid_eq(attr.ia_uid, inode->i_uid)) 242 attr.ia_valid |= ATTR_UID; 243 if (!gid_eq(attr.ia_gid, inode->i_gid)) 244 attr.ia_valid |= ATTR_GID; 245 246 if (!attr.ia_valid) 247 return 0; 248 249 err = dquot_transfer(inode, &attr); 250 if (err) 251 set_sbi_flag(F2FS_I_SB(inode), SBI_QUOTA_NEED_REPAIR); 252 return err; 253} 254 255static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri) 256{ 257 if (ri->i_inline & F2FS_PIN_FILE) 258 set_inode_flag(inode, FI_PIN_FILE); 259 else 260 clear_inode_flag(inode, FI_PIN_FILE); 261 if (ri->i_inline & F2FS_DATA_EXIST) 262 set_inode_flag(inode, FI_DATA_EXIST); 263 else 264 clear_inode_flag(inode, FI_DATA_EXIST); 265} 266 267static int recover_inode(struct inode *inode, struct page *page) 268{ 269 struct f2fs_inode *raw = F2FS_INODE(page); 270 char *name; 271 int err; 272 273 inode->i_mode = le16_to_cpu(raw->i_mode); 274 275 err = recover_quota_data(inode, page); 276 if (err) 277 return err; 278 279 i_uid_write(inode, le32_to_cpu(raw->i_uid)); 280 i_gid_write(inode, le32_to_cpu(raw->i_gid)); 281 282 if (raw->i_inline & F2FS_EXTRA_ATTR) { 283 if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)) && 284 F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize), 285 i_projid)) { 286 projid_t i_projid; 287 kprojid_t kprojid; 288 289 i_projid = (projid_t)le32_to_cpu(raw->i_projid); 290 kprojid = make_kprojid(&init_user_ns, i_projid); 291 292 if (!projid_eq(kprojid, F2FS_I(inode)->i_projid)) { 293 err = f2fs_transfer_project_quota(inode, 294 kprojid); 295 if (err) 296 return err; 297 F2FS_I(inode)->i_projid = kprojid; 298 } 299 } 300 } 301 302 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size)); 303 inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime); 304 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime); 305 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime); 306 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec); 307 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec); 308 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec); 309 310 F2FS_I(inode)->i_advise = raw->i_advise; 311 F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags); 312 f2fs_set_inode_flags(inode); 313 F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] = 314 le16_to_cpu(raw->i_gc_failures); 315 316 recover_inline_flags(inode, raw); 317 318 f2fs_mark_inode_dirty_sync(inode, true); 319 320 if (file_enc_name(inode)) 321 name = "<encrypted>"; 322 else 323 name = F2FS_INODE(page)->i_name; 324 325 f2fs_notice(F2FS_I_SB(inode), "recover_inode: ino = %x, name = %s, inline = %x", 326 ino_of_node(page), name, raw->i_inline); 327 return 0; 328} 329 330static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head, 331 bool check_only) 332{ 333 struct curseg_info *curseg; 334 struct page *page = NULL; 335 block_t blkaddr; 336 unsigned int loop_cnt = 0; 337 unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg - 338 valid_user_blocks(sbi); 339 int err = 0; 340 341 /* get node pages in the current segment */ 342 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 343 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 344 345 while (1) { 346 struct fsync_inode_entry *entry; 347 348 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR)) 349 return 0; 350 351 page = f2fs_get_tmp_page(sbi, blkaddr); 352 if (IS_ERR(page)) { 353 err = PTR_ERR(page); 354 break; 355 } 356 357 if (!is_recoverable_dnode(page)) { 358 f2fs_put_page(page, 1); 359 break; 360 } 361 362 if (!is_fsync_dnode(page)) 363 goto next; 364 365 entry = get_fsync_inode(head, ino_of_node(page)); 366 if (!entry) { 367 bool quota_inode = false; 368 369 if (!check_only && 370 IS_INODE(page) && is_dent_dnode(page)) { 371 err = f2fs_recover_inode_page(sbi, page); 372 if (err) { 373 f2fs_put_page(page, 1); 374 break; 375 } 376 quota_inode = true; 377 } 378 379 /* 380 * CP | dnode(F) | inode(DF) 381 * For this case, we should not give up now. 382 */ 383 entry = add_fsync_inode(sbi, head, ino_of_node(page), 384 quota_inode); 385 if (IS_ERR(entry)) { 386 err = PTR_ERR(entry); 387 if (err == -ENOENT) { 388 err = 0; 389 goto next; 390 } 391 f2fs_put_page(page, 1); 392 break; 393 } 394 } 395 entry->blkaddr = blkaddr; 396 397 if (IS_INODE(page) && is_dent_dnode(page)) 398 entry->last_dentry = blkaddr; 399next: 400 /* sanity check in order to detect looped node chain */ 401 if (++loop_cnt >= free_blocks || 402 blkaddr == next_blkaddr_of_node(page)) { 403 f2fs_notice(sbi, "%s: detect looped node chain, blkaddr:%u, next:%u", 404 __func__, blkaddr, 405 next_blkaddr_of_node(page)); 406 f2fs_put_page(page, 1); 407 err = -EINVAL; 408 break; 409 } 410 411 /* check next segment */ 412 blkaddr = next_blkaddr_of_node(page); 413 f2fs_put_page(page, 1); 414 415 f2fs_ra_meta_pages_cond(sbi, blkaddr); 416 } 417 return err; 418} 419 420static void destroy_fsync_dnodes(struct list_head *head, int drop) 421{ 422 struct fsync_inode_entry *entry, *tmp; 423 424 list_for_each_entry_safe(entry, tmp, head, list) 425 del_fsync_inode(entry, drop); 426} 427 428static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi, 429 block_t blkaddr, struct dnode_of_data *dn) 430{ 431 struct seg_entry *sentry; 432 unsigned int segno = GET_SEGNO(sbi, blkaddr); 433 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 434 struct f2fs_summary_block *sum_node; 435 struct f2fs_summary sum; 436 struct page *sum_page, *node_page; 437 struct dnode_of_data tdn = *dn; 438 nid_t ino, nid; 439 struct inode *inode; 440 unsigned int offset, ofs_in_node, max_addrs; 441 block_t bidx; 442 int i; 443 444 sentry = get_seg_entry(sbi, segno); 445 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map)) 446 return 0; 447 448 /* Get the previous summary */ 449 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 450 struct curseg_info *curseg = CURSEG_I(sbi, i); 451 if (curseg->segno == segno) { 452 sum = curseg->sum_blk->entries[blkoff]; 453 goto got_it; 454 } 455 } 456 457 sum_page = f2fs_get_sum_page(sbi, segno); 458 if (IS_ERR(sum_page)) 459 return PTR_ERR(sum_page); 460 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 461 sum = sum_node->entries[blkoff]; 462 f2fs_put_page(sum_page, 1); 463got_it: 464 /* Use the locked dnode page and inode */ 465 nid = le32_to_cpu(sum.nid); 466 ofs_in_node = le16_to_cpu(sum.ofs_in_node); 467 468 max_addrs = ADDRS_PER_PAGE(dn->node_page, dn->inode); 469 if (ofs_in_node >= max_addrs) { 470 f2fs_err(sbi, "Inconsistent ofs_in_node:%u in summary, ino:%lu, nid:%u, max:%u", 471 ofs_in_node, dn->inode->i_ino, nid, max_addrs); 472 return -EFSCORRUPTED; 473 } 474 475 if (dn->inode->i_ino == nid) { 476 tdn.nid = nid; 477 if (!dn->inode_page_locked) 478 lock_page(dn->inode_page); 479 tdn.node_page = dn->inode_page; 480 tdn.ofs_in_node = ofs_in_node; 481 goto truncate_out; 482 } else if (dn->nid == nid) { 483 tdn.ofs_in_node = ofs_in_node; 484 goto truncate_out; 485 } 486 487 /* Get the node page */ 488 node_page = f2fs_get_node_page(sbi, nid); 489 if (IS_ERR(node_page)) 490 return PTR_ERR(node_page); 491 492 offset = ofs_of_node(node_page); 493 ino = ino_of_node(node_page); 494 f2fs_put_page(node_page, 1); 495 496 if (ino != dn->inode->i_ino) { 497 int ret; 498 499 /* Deallocate previous index in the node page */ 500 inode = f2fs_iget_retry(sbi->sb, ino); 501 if (IS_ERR(inode)) 502 return PTR_ERR(inode); 503 504 ret = dquot_initialize(inode); 505 if (ret) { 506 iput(inode); 507 return ret; 508 } 509 } else { 510 inode = dn->inode; 511 } 512 513 bidx = f2fs_start_bidx_of_node(offset, inode) + 514 le16_to_cpu(sum.ofs_in_node); 515 516 /* 517 * if inode page is locked, unlock temporarily, but its reference 518 * count keeps alive. 519 */ 520 if (ino == dn->inode->i_ino && dn->inode_page_locked) 521 unlock_page(dn->inode_page); 522 523 set_new_dnode(&tdn, inode, NULL, NULL, 0); 524 if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE)) 525 goto out; 526 527 if (tdn.data_blkaddr == blkaddr) 528 f2fs_truncate_data_blocks_range(&tdn, 1); 529 530 f2fs_put_dnode(&tdn); 531out: 532 if (ino != dn->inode->i_ino) 533 iput(inode); 534 else if (dn->inode_page_locked) 535 lock_page(dn->inode_page); 536 return 0; 537 538truncate_out: 539 if (f2fs_data_blkaddr(&tdn) == blkaddr) 540 f2fs_truncate_data_blocks_range(&tdn, 1); 541 if (dn->inode->i_ino == nid && !dn->inode_page_locked) 542 unlock_page(dn->inode_page); 543 return 0; 544} 545 546static int f2fs_reserve_new_block_retry(struct dnode_of_data *dn) 547{ 548 int i, err = 0; 549 550 for (i = DEFAULT_FAILURE_RETRY_COUNT; i > 0; i--) { 551 err = f2fs_reserve_new_block(dn); 552 if (!err) 553 break; 554 } 555 556 return err; 557} 558 559static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode, 560 struct page *page) 561{ 562 struct dnode_of_data dn; 563 struct node_info ni; 564 unsigned int start, end; 565 int err = 0, recovered = 0; 566 567 /* step 1: recover xattr */ 568 if (IS_INODE(page)) { 569 err = f2fs_recover_inline_xattr(inode, page); 570 if (err) 571 goto out; 572 } else if (f2fs_has_xattr_block(ofs_of_node(page))) { 573 err = f2fs_recover_xattr_data(inode, page); 574 if (!err) 575 recovered++; 576 goto out; 577 } 578 579 /* step 2: recover inline data */ 580 err = f2fs_recover_inline_data(inode, page); 581 if (err) { 582 if (err == 1) 583 err = 0; 584 goto out; 585 } 586 587 /* step 3: recover data indices */ 588 start = f2fs_start_bidx_of_node(ofs_of_node(page), inode); 589 end = start + ADDRS_PER_PAGE(page, inode); 590 591 set_new_dnode(&dn, inode, NULL, NULL, 0); 592retry_dn: 593 err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE); 594 if (err) { 595 if (err == -ENOMEM) { 596 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT); 597 goto retry_dn; 598 } 599 goto out; 600 } 601 602 f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true); 603 604 err = f2fs_get_node_info(sbi, dn.nid, &ni); 605 if (err) 606 goto err; 607 608 f2fs_bug_on(sbi, ni.ino != ino_of_node(page)); 609 610 if (ofs_of_node(dn.node_page) != ofs_of_node(page)) { 611 f2fs_warn(sbi, "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u", 612 inode->i_ino, ofs_of_node(dn.node_page), 613 ofs_of_node(page)); 614 err = -EFSCORRUPTED; 615 goto err; 616 } 617 618 for (; start < end; start++, dn.ofs_in_node++) { 619 block_t src, dest; 620 621 src = f2fs_data_blkaddr(&dn); 622 dest = data_blkaddr(dn.inode, page, dn.ofs_in_node); 623 624 if (__is_valid_data_blkaddr(src) && 625 !f2fs_is_valid_blkaddr(sbi, src, META_POR)) { 626 err = -EFSCORRUPTED; 627 goto err; 628 } 629 630 if (__is_valid_data_blkaddr(dest) && 631 !f2fs_is_valid_blkaddr(sbi, dest, META_POR)) { 632 err = -EFSCORRUPTED; 633 goto err; 634 } 635 636 /* skip recovering if dest is the same as src */ 637 if (src == dest) 638 continue; 639 640 /* dest is invalid, just invalidate src block */ 641 if (dest == NULL_ADDR) { 642 f2fs_truncate_data_blocks_range(&dn, 1); 643 continue; 644 } 645 646 if (!file_keep_isize(inode) && 647 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT))) 648 f2fs_i_size_write(inode, 649 (loff_t)(start + 1) << PAGE_SHIFT); 650 651 /* 652 * dest is reserved block, invalidate src block 653 * and then reserve one new block in dnode page. 654 */ 655 if (dest == NEW_ADDR) { 656 f2fs_truncate_data_blocks_range(&dn, 1); 657 658 err = f2fs_reserve_new_block_retry(&dn); 659 if (err) 660 goto err; 661 continue; 662 } 663 664 /* dest is valid block, try to recover from src to dest */ 665 if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) { 666 if (src == NULL_ADDR) { 667 err = f2fs_reserve_new_block_retry(&dn); 668 if (err) 669 goto err; 670 } 671retry_prev: 672 /* Check the previous node page having this index */ 673 err = check_index_in_prev_nodes(sbi, dest, &dn); 674 if (err) { 675 if (err == -ENOMEM) { 676 congestion_wait(BLK_RW_ASYNC, 677 DEFAULT_IO_TIMEOUT); 678 goto retry_prev; 679 } 680 goto err; 681 } 682 683 if (f2fs_is_valid_blkaddr(sbi, dest, 684 DATA_GENERIC_ENHANCE_UPDATE)) { 685 f2fs_err(sbi, "Inconsistent dest blkaddr:%u, ino:%lu, ofs:%u", 686 dest, inode->i_ino, dn.ofs_in_node); 687 err = -EFSCORRUPTED; 688 goto err; 689 } 690 691 /* write dummy data page */ 692 f2fs_replace_block(sbi, &dn, src, dest, 693 ni.version, false, false); 694 recovered++; 695 } 696 } 697 698 copy_node_footer(dn.node_page, page); 699 fill_node_footer(dn.node_page, dn.nid, ni.ino, 700 ofs_of_node(page), false); 701 set_page_dirty(dn.node_page); 702err: 703 f2fs_put_dnode(&dn); 704out: 705 f2fs_notice(sbi, "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d", 706 inode->i_ino, file_keep_isize(inode) ? "keep" : "recover", 707 recovered, err); 708 return err; 709} 710 711static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list, 712 struct list_head *tmp_inode_list, struct list_head *dir_list) 713{ 714 struct curseg_info *curseg; 715 struct page *page = NULL; 716 int err = 0; 717 block_t blkaddr; 718 719 /* get node pages in the current segment */ 720 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 721 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 722 723 while (1) { 724 struct fsync_inode_entry *entry; 725 726 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR)) 727 break; 728 729 f2fs_ra_meta_pages_cond(sbi, blkaddr); 730 731 page = f2fs_get_tmp_page(sbi, blkaddr); 732 if (IS_ERR(page)) { 733 err = PTR_ERR(page); 734 break; 735 } 736 737 if (!is_recoverable_dnode(page)) { 738 f2fs_put_page(page, 1); 739 break; 740 } 741 742 entry = get_fsync_inode(inode_list, ino_of_node(page)); 743 if (!entry) 744 goto next; 745 /* 746 * inode(x) | CP | inode(x) | dnode(F) 747 * In this case, we can lose the latest inode(x). 748 * So, call recover_inode for the inode update. 749 */ 750 if (IS_INODE(page)) { 751 err = recover_inode(entry->inode, page); 752 if (err) { 753 f2fs_put_page(page, 1); 754 break; 755 } 756 } 757 if (entry->last_dentry == blkaddr) { 758 err = recover_dentry(entry->inode, page, dir_list); 759 if (err) { 760 f2fs_put_page(page, 1); 761 break; 762 } 763 } 764 err = do_recover_data(sbi, entry->inode, page); 765 if (err) { 766 f2fs_put_page(page, 1); 767 break; 768 } 769 770 if (entry->blkaddr == blkaddr) 771 list_move_tail(&entry->list, tmp_inode_list); 772next: 773 /* check next segment */ 774 blkaddr = next_blkaddr_of_node(page); 775 f2fs_put_page(page, 1); 776 } 777 if (!err) 778 f2fs_allocate_new_segments(sbi); 779 return err; 780} 781 782int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only) 783{ 784 struct list_head inode_list, tmp_inode_list; 785 struct list_head dir_list; 786 int err; 787 int ret = 0; 788 unsigned long s_flags = sbi->sb->s_flags; 789 bool need_writecp = false; 790 bool fix_curseg_write_pointer = false; 791#ifdef CONFIG_QUOTA 792 int quota_enabled; 793#endif 794 795 if (s_flags & SB_RDONLY) { 796 f2fs_info(sbi, "recover fsync data on readonly fs"); 797 sbi->sb->s_flags &= ~SB_RDONLY; 798 } 799 800#ifdef CONFIG_QUOTA 801 /* Needed for iput() to work correctly and not trash data */ 802 sbi->sb->s_flags |= SB_ACTIVE; 803 /* Turn on quotas so that they are updated correctly */ 804 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY); 805#endif 806 807 INIT_LIST_HEAD(&inode_list); 808 INIT_LIST_HEAD(&tmp_inode_list); 809 INIT_LIST_HEAD(&dir_list); 810 811 /* prevent checkpoint */ 812 mutex_lock(&sbi->cp_mutex); 813 814 /* step #1: find fsynced inode numbers */ 815 err = find_fsync_dnodes(sbi, &inode_list, check_only); 816 if (err || list_empty(&inode_list)) 817 goto skip; 818 819 if (check_only) { 820 ret = 1; 821 goto skip; 822 } 823 824 need_writecp = true; 825 826 /* step #2: recover data */ 827 err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list); 828 if (!err) 829 f2fs_bug_on(sbi, !list_empty(&inode_list)); 830 else { 831 /* restore s_flags to let iput() trash data */ 832 sbi->sb->s_flags = s_flags; 833 } 834skip: 835 fix_curseg_write_pointer = !check_only || list_empty(&inode_list); 836 837 destroy_fsync_dnodes(&inode_list, err); 838 destroy_fsync_dnodes(&tmp_inode_list, err); 839 840 /* truncate meta pages to be used by the recovery */ 841 truncate_inode_pages_range(META_MAPPING(sbi), 842 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1); 843 844 if (err) { 845 truncate_inode_pages_final(NODE_MAPPING(sbi)); 846 truncate_inode_pages_final(META_MAPPING(sbi)); 847 } 848 849 /* 850 * If fsync data succeeds or there is no fsync data to recover, 851 * and the f2fs is not read only, check and fix zoned block devices' 852 * write pointer consistency. 853 */ 854 if (!err && fix_curseg_write_pointer && !f2fs_readonly(sbi->sb) && 855 f2fs_sb_has_blkzoned(sbi)) { 856 err = f2fs_fix_curseg_write_pointer(sbi); 857 if (!err) 858 err = f2fs_check_write_pointer(sbi); 859 ret = err; 860 } 861 862 if (!err) 863 clear_sbi_flag(sbi, SBI_POR_DOING); 864 865 mutex_unlock(&sbi->cp_mutex); 866 867 /* let's drop all the directory inodes for clean checkpoint */ 868 destroy_fsync_dnodes(&dir_list, err); 869 870 if (need_writecp) { 871 set_sbi_flag(sbi, SBI_IS_RECOVERED); 872 873 if (!err) { 874 struct cp_control cpc = { 875 .reason = CP_RECOVERY, 876 }; 877 err = f2fs_write_checkpoint(sbi, &cpc); 878 } 879 } 880 881#ifdef CONFIG_QUOTA 882 /* Turn quotas off */ 883 if (quota_enabled) 884 f2fs_quota_off_umount(sbi->sb); 885#endif 886 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 887 888 return ret ? ret: err; 889} 890 891int __init f2fs_create_recovery_cache(void) 892{ 893 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry", 894 sizeof(struct fsync_inode_entry)); 895 if (!fsync_entry_slab) 896 return -ENOMEM; 897 return 0; 898} 899 900void f2fs_destroy_recovery_cache(void) 901{ 902 kmem_cache_destroy(fsync_entry_slab); 903} 904