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 <asm/unaligned.h> 9 #include <linux/fs.h> 10 #include <linux/f2fs_fs.h> 11 #include <linux/sched/mm.h> 12 #include "f2fs.h" 13 #include "node.h" 14 #include "segment.h" 15 16 /* 17 * Roll forward recovery scenarios. 18 * 19 * [Term] F: fsync_mark, D: dentry_mark 20 * 21 * 1. inode(x) | CP | inode(x) | dnode(F) 22 * -> Update the latest inode(x). 23 * 24 * 2. inode(x) | CP | inode(F) | dnode(F) 25 * -> No problem. 26 * 27 * 3. inode(x) | CP | dnode(F) | inode(x) 28 * -> Recover to the latest dnode(F), and drop the last inode(x) 29 * 30 * 4. inode(x) | CP | dnode(F) | inode(F) 31 * -> No problem. 32 * 33 * 5. CP | inode(x) | dnode(F) 34 * -> The inode(DF) was missing. Should drop this dnode(F). 35 * 36 * 6. CP | inode(DF) | dnode(F) 37 * -> No problem. 38 * 39 * 7. CP | dnode(F) | inode(DF) 40 * -> If f2fs_iget fails, then goto next to find inode(DF). 41 * 42 * 8. CP | dnode(F) | inode(x) 43 * -> If f2fs_iget fails, then goto next to find inode(DF). 44 * But it will fail due to no inode(DF). 45 */ 46 47 static struct kmem_cache *fsync_entry_slab; 48 49 #if IS_ENABLED(CONFIG_UNICODE) 50 extern struct kmem_cache *f2fs_cf_name_slab; 51 #endif 52 53 bool f2fs_space_for_roll_forward(struct f2fs_sb_info *sbi) 54 { 55 s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count); 56 57 if (sbi->last_valid_block_count + nalloc > sbi->user_block_count) 58 return false; 59 if (NM_I(sbi)->max_rf_node_blocks && 60 percpu_counter_sum_positive(&sbi->rf_node_block_count) >= 61 NM_I(sbi)->max_rf_node_blocks) 62 return false; 63 return true; 64 } 65 66 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head, 67 nid_t ino) 68 { 69 struct fsync_inode_entry *entry; 70 71 list_for_each_entry(entry, head, list) 72 if (entry->inode->i_ino == ino) 73 return entry; 74 75 return NULL; 76 } 77 78 static struct fsync_inode_entry *add_fsync_inode(struct f2fs_sb_info *sbi, 79 struct list_head *head, nid_t ino, bool quota_inode) 80 { 81 struct inode *inode; 82 struct fsync_inode_entry *entry; 83 int err; 84 85 inode = f2fs_iget_retry(sbi->sb, ino); 86 if (IS_ERR(inode)) 87 return ERR_CAST(inode); 88 89 err = f2fs_dquot_initialize(inode); 90 if (err) 91 goto err_out; 92 93 if (quota_inode) { 94 err = dquot_alloc_inode(inode); 95 if (err) 96 goto err_out; 97 } 98 99 entry = f2fs_kmem_cache_alloc(fsync_entry_slab, 100 GFP_F2FS_ZERO, true, NULL); 101 entry->inode = inode; 102 list_add_tail(&entry->list, head); 103 104 return entry; 105 err_out: 106 iput(inode); 107 return ERR_PTR(err); 108 } 109 110 static void del_fsync_inode(struct fsync_inode_entry *entry, int drop) 111 { 112 if (drop) { 113 /* inode should not be recovered, drop it */ 114 f2fs_inode_synced(entry->inode); 115 } 116 iput(entry->inode); 117 list_del(&entry->list); 118 kmem_cache_free(fsync_entry_slab, entry); 119 } 120 121 static int init_recovered_filename(const struct inode *dir, 122 struct f2fs_inode *raw_inode, 123 struct f2fs_filename *fname, 124 struct qstr *usr_fname) 125 { 126 int err; 127 128 memset(fname, 0, sizeof(*fname)); 129 fname->disk_name.len = le32_to_cpu(raw_inode->i_namelen); 130 fname->disk_name.name = raw_inode->i_name; 131 132 if (WARN_ON(fname->disk_name.len > F2FS_NAME_LEN)) 133 return -ENAMETOOLONG; 134 135 if (!IS_ENCRYPTED(dir)) { 136 usr_fname->name = fname->disk_name.name; 137 usr_fname->len = fname->disk_name.len; 138 fname->usr_fname = usr_fname; 139 } 140 141 /* Compute the hash of the filename */ 142 if (IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)) { 143 /* 144 * In this case the hash isn't computable without the key, so it 145 * was saved on-disk. 146 */ 147 if (fname->disk_name.len + sizeof(f2fs_hash_t) > F2FS_NAME_LEN) 148 return -EINVAL; 149 fname->hash = get_unaligned((f2fs_hash_t *) 150 &raw_inode->i_name[fname->disk_name.len]); 151 } else if (IS_CASEFOLDED(dir)) { 152 err = f2fs_init_casefolded_name(dir, fname); 153 if (err) 154 return err; 155 f2fs_hash_filename(dir, fname); 156 #if IS_ENABLED(CONFIG_UNICODE) 157 /* Case-sensitive match is fine for recovery */ 158 kmem_cache_free(f2fs_cf_name_slab, fname->cf_name.name); 159 fname->cf_name.name = NULL; 160 #endif 161 } else { 162 f2fs_hash_filename(dir, fname); 163 } 164 return 0; 165 } 166 167 static int recover_dentry(struct inode *inode, struct page *ipage, 168 struct list_head *dir_list) 169 { 170 struct f2fs_inode *raw_inode = F2FS_INODE(ipage); 171 nid_t pino = le32_to_cpu(raw_inode->i_pino); 172 struct f2fs_dir_entry *de; 173 struct f2fs_filename fname; 174 struct qstr usr_fname; 175 struct page *page; 176 struct inode *dir, *einode; 177 struct fsync_inode_entry *entry; 178 int err = 0; 179 char *name; 180 181 entry = get_fsync_inode(dir_list, pino); 182 if (!entry) { 183 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list, 184 pino, false); 185 if (IS_ERR(entry)) { 186 dir = ERR_CAST(entry); 187 err = PTR_ERR(entry); 188 goto out; 189 } 190 } 191 192 dir = entry->inode; 193 err = init_recovered_filename(dir, raw_inode, &fname, &usr_fname); 194 if (err) 195 goto out; 196 retry: 197 de = __f2fs_find_entry(dir, &fname, &page); 198 if (de && inode->i_ino == le32_to_cpu(de->ino)) 199 goto out_put; 200 201 if (de) { 202 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino)); 203 if (IS_ERR(einode)) { 204 WARN_ON(1); 205 err = PTR_ERR(einode); 206 if (err == -ENOENT) 207 err = -EEXIST; 208 goto out_put; 209 } 210 211 err = f2fs_dquot_initialize(einode); 212 if (err) { 213 iput(einode); 214 goto out_put; 215 } 216 217 err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode)); 218 if (err) { 219 iput(einode); 220 goto out_put; 221 } 222 f2fs_delete_entry(de, page, dir, einode); 223 iput(einode); 224 goto retry; 225 } else if (IS_ERR(page)) { 226 err = PTR_ERR(page); 227 } else { 228 err = f2fs_add_dentry(dir, &fname, inode, 229 inode->i_ino, inode->i_mode); 230 } 231 if (err == -ENOMEM) 232 goto retry; 233 goto out; 234 235 out_put: 236 f2fs_put_page(page, 0); 237 out: 238 if (file_enc_name(inode)) 239 name = "<encrypted>"; 240 else 241 name = raw_inode->i_name; 242 f2fs_notice(F2FS_I_SB(inode), "%s: ino = %x, name = %s, dir = %lx, err = %d", 243 __func__, ino_of_node(ipage), name, 244 IS_ERR(dir) ? 0 : dir->i_ino, err); 245 return err; 246 } 247 248 static int recover_quota_data(struct inode *inode, struct page *page) 249 { 250 struct f2fs_inode *raw = F2FS_INODE(page); 251 struct iattr attr; 252 uid_t i_uid = le32_to_cpu(raw->i_uid); 253 gid_t i_gid = le32_to_cpu(raw->i_gid); 254 int err; 255 256 memset(&attr, 0, sizeof(attr)); 257 258 attr.ia_vfsuid = VFSUIDT_INIT(make_kuid(inode->i_sb->s_user_ns, i_uid)); 259 attr.ia_vfsgid = VFSGIDT_INIT(make_kgid(inode->i_sb->s_user_ns, i_gid)); 260 261 if (!vfsuid_eq(attr.ia_vfsuid, i_uid_into_vfsuid(&init_user_ns, inode))) 262 attr.ia_valid |= ATTR_UID; 263 if (!vfsgid_eq(attr.ia_vfsgid, i_gid_into_vfsgid(&init_user_ns, inode))) 264 attr.ia_valid |= ATTR_GID; 265 266 if (!attr.ia_valid) 267 return 0; 268 269 err = dquot_transfer(&init_user_ns, inode, &attr); 270 if (err) 271 set_sbi_flag(F2FS_I_SB(inode), SBI_QUOTA_NEED_REPAIR); 272 return err; 273 } 274 275 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri) 276 { 277 if (ri->i_inline & F2FS_PIN_FILE) 278 set_inode_flag(inode, FI_PIN_FILE); 279 else 280 clear_inode_flag(inode, FI_PIN_FILE); 281 if (ri->i_inline & F2FS_DATA_EXIST) 282 set_inode_flag(inode, FI_DATA_EXIST); 283 else 284 clear_inode_flag(inode, FI_DATA_EXIST); 285 } 286 287 static int recover_inode(struct inode *inode, struct page *page) 288 { 289 struct f2fs_inode *raw = F2FS_INODE(page); 290 char *name; 291 int err; 292 293 inode->i_mode = le16_to_cpu(raw->i_mode); 294 295 err = recover_quota_data(inode, page); 296 if (err) 297 return err; 298 299 i_uid_write(inode, le32_to_cpu(raw->i_uid)); 300 i_gid_write(inode, le32_to_cpu(raw->i_gid)); 301 302 if (raw->i_inline & F2FS_EXTRA_ATTR) { 303 if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)) && 304 F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize), 305 i_projid)) { 306 projid_t i_projid; 307 kprojid_t kprojid; 308 309 i_projid = (projid_t)le32_to_cpu(raw->i_projid); 310 kprojid = make_kprojid(&init_user_ns, i_projid); 311 312 if (!projid_eq(kprojid, F2FS_I(inode)->i_projid)) { 313 err = f2fs_transfer_project_quota(inode, 314 kprojid); 315 if (err) 316 return err; 317 F2FS_I(inode)->i_projid = kprojid; 318 } 319 } 320 } 321 322 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size)); 323 inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime); 324 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime); 325 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime); 326 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec); 327 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec); 328 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec); 329 330 F2FS_I(inode)->i_advise = raw->i_advise; 331 F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags); 332 f2fs_set_inode_flags(inode); 333 F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] = 334 le16_to_cpu(raw->i_gc_failures); 335 336 recover_inline_flags(inode, raw); 337 338 f2fs_mark_inode_dirty_sync(inode, true); 339 340 if (file_enc_name(inode)) 341 name = "<encrypted>"; 342 else 343 name = F2FS_INODE(page)->i_name; 344 345 f2fs_notice(F2FS_I_SB(inode), "recover_inode: ino = %x, name = %s, inline = %x", 346 ino_of_node(page), name, raw->i_inline); 347 return 0; 348 } 349 350 static unsigned int adjust_por_ra_blocks(struct f2fs_sb_info *sbi, 351 unsigned int ra_blocks, unsigned int blkaddr, 352 unsigned int next_blkaddr) 353 { 354 if (blkaddr + 1 == next_blkaddr) 355 ra_blocks = min_t(unsigned int, RECOVERY_MAX_RA_BLOCKS, 356 ra_blocks * 2); 357 else if (next_blkaddr % sbi->blocks_per_seg) 358 ra_blocks = max_t(unsigned int, RECOVERY_MIN_RA_BLOCKS, 359 ra_blocks / 2); 360 return ra_blocks; 361 } 362 363 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head, 364 bool check_only) 365 { 366 struct curseg_info *curseg; 367 struct page *page = NULL; 368 block_t blkaddr; 369 unsigned int loop_cnt = 0; 370 unsigned int ra_blocks = RECOVERY_MAX_RA_BLOCKS; 371 unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg - 372 valid_user_blocks(sbi); 373 int err = 0; 374 375 /* get node pages in the current segment */ 376 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 377 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 378 379 while (1) { 380 struct fsync_inode_entry *entry; 381 382 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR)) 383 return 0; 384 385 page = f2fs_get_tmp_page(sbi, blkaddr); 386 if (IS_ERR(page)) { 387 err = PTR_ERR(page); 388 break; 389 } 390 391 if (!is_recoverable_dnode(page)) { 392 f2fs_put_page(page, 1); 393 break; 394 } 395 396 if (!is_fsync_dnode(page)) 397 goto next; 398 399 entry = get_fsync_inode(head, ino_of_node(page)); 400 if (!entry) { 401 bool quota_inode = false; 402 403 if (!check_only && 404 IS_INODE(page) && is_dent_dnode(page)) { 405 err = f2fs_recover_inode_page(sbi, page); 406 if (err) { 407 f2fs_put_page(page, 1); 408 break; 409 } 410 quota_inode = true; 411 } 412 413 /* 414 * CP | dnode(F) | inode(DF) 415 * For this case, we should not give up now. 416 */ 417 entry = add_fsync_inode(sbi, head, ino_of_node(page), 418 quota_inode); 419 if (IS_ERR(entry)) { 420 err = PTR_ERR(entry); 421 if (err == -ENOENT) { 422 err = 0; 423 goto next; 424 } 425 f2fs_put_page(page, 1); 426 break; 427 } 428 } 429 entry->blkaddr = blkaddr; 430 431 if (IS_INODE(page) && is_dent_dnode(page)) 432 entry->last_dentry = blkaddr; 433 next: 434 /* sanity check in order to detect looped node chain */ 435 if (++loop_cnt >= free_blocks || 436 blkaddr == next_blkaddr_of_node(page)) { 437 f2fs_notice(sbi, "%s: detect looped node chain, blkaddr:%u, next:%u", 438 __func__, blkaddr, 439 next_blkaddr_of_node(page)); 440 f2fs_put_page(page, 1); 441 err = -EINVAL; 442 break; 443 } 444 445 ra_blocks = adjust_por_ra_blocks(sbi, ra_blocks, blkaddr, 446 next_blkaddr_of_node(page)); 447 448 /* check next segment */ 449 blkaddr = next_blkaddr_of_node(page); 450 f2fs_put_page(page, 1); 451 452 f2fs_ra_meta_pages_cond(sbi, blkaddr, ra_blocks); 453 } 454 return err; 455 } 456 457 static void destroy_fsync_dnodes(struct list_head *head, int drop) 458 { 459 struct fsync_inode_entry *entry, *tmp; 460 461 list_for_each_entry_safe(entry, tmp, head, list) 462 del_fsync_inode(entry, drop); 463 } 464 465 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi, 466 block_t blkaddr, struct dnode_of_data *dn) 467 { 468 struct seg_entry *sentry; 469 unsigned int segno = GET_SEGNO(sbi, blkaddr); 470 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr); 471 struct f2fs_summary_block *sum_node; 472 struct f2fs_summary sum; 473 struct page *sum_page, *node_page; 474 struct dnode_of_data tdn = *dn; 475 nid_t ino, nid; 476 struct inode *inode; 477 unsigned int offset; 478 block_t bidx; 479 int i; 480 481 sentry = get_seg_entry(sbi, segno); 482 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map)) 483 return 0; 484 485 /* Get the previous summary */ 486 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) { 487 struct curseg_info *curseg = CURSEG_I(sbi, i); 488 489 if (curseg->segno == segno) { 490 sum = curseg->sum_blk->entries[blkoff]; 491 goto got_it; 492 } 493 } 494 495 sum_page = f2fs_get_sum_page(sbi, segno); 496 if (IS_ERR(sum_page)) 497 return PTR_ERR(sum_page); 498 sum_node = (struct f2fs_summary_block *)page_address(sum_page); 499 sum = sum_node->entries[blkoff]; 500 f2fs_put_page(sum_page, 1); 501 got_it: 502 /* Use the locked dnode page and inode */ 503 nid = le32_to_cpu(sum.nid); 504 if (dn->inode->i_ino == nid) { 505 tdn.nid = nid; 506 if (!dn->inode_page_locked) 507 lock_page(dn->inode_page); 508 tdn.node_page = dn->inode_page; 509 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 510 goto truncate_out; 511 } else if (dn->nid == nid) { 512 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node); 513 goto truncate_out; 514 } 515 516 /* Get the node page */ 517 node_page = f2fs_get_node_page(sbi, nid); 518 if (IS_ERR(node_page)) 519 return PTR_ERR(node_page); 520 521 offset = ofs_of_node(node_page); 522 ino = ino_of_node(node_page); 523 f2fs_put_page(node_page, 1); 524 525 if (ino != dn->inode->i_ino) { 526 int ret; 527 528 /* Deallocate previous index in the node page */ 529 inode = f2fs_iget_retry(sbi->sb, ino); 530 if (IS_ERR(inode)) 531 return PTR_ERR(inode); 532 533 ret = f2fs_dquot_initialize(inode); 534 if (ret) { 535 iput(inode); 536 return ret; 537 } 538 } else { 539 inode = dn->inode; 540 } 541 542 bidx = f2fs_start_bidx_of_node(offset, inode) + 543 le16_to_cpu(sum.ofs_in_node); 544 545 /* 546 * if inode page is locked, unlock temporarily, but its reference 547 * count keeps alive. 548 */ 549 if (ino == dn->inode->i_ino && dn->inode_page_locked) 550 unlock_page(dn->inode_page); 551 552 set_new_dnode(&tdn, inode, NULL, NULL, 0); 553 if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE)) 554 goto out; 555 556 if (tdn.data_blkaddr == blkaddr) 557 f2fs_truncate_data_blocks_range(&tdn, 1); 558 559 f2fs_put_dnode(&tdn); 560 out: 561 if (ino != dn->inode->i_ino) 562 iput(inode); 563 else if (dn->inode_page_locked) 564 lock_page(dn->inode_page); 565 return 0; 566 567 truncate_out: 568 if (f2fs_data_blkaddr(&tdn) == blkaddr) 569 f2fs_truncate_data_blocks_range(&tdn, 1); 570 if (dn->inode->i_ino == nid && !dn->inode_page_locked) 571 unlock_page(dn->inode_page); 572 return 0; 573 } 574 575 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode, 576 struct page *page) 577 { 578 struct dnode_of_data dn; 579 struct node_info ni; 580 unsigned int start, end; 581 int err = 0, recovered = 0; 582 583 /* step 1: recover xattr */ 584 if (IS_INODE(page)) { 585 err = f2fs_recover_inline_xattr(inode, page); 586 if (err) 587 goto out; 588 } else if (f2fs_has_xattr_block(ofs_of_node(page))) { 589 err = f2fs_recover_xattr_data(inode, page); 590 if (!err) 591 recovered++; 592 goto out; 593 } 594 595 /* step 2: recover inline data */ 596 err = f2fs_recover_inline_data(inode, page); 597 if (err) { 598 if (err == 1) 599 err = 0; 600 goto out; 601 } 602 603 /* step 3: recover data indices */ 604 start = f2fs_start_bidx_of_node(ofs_of_node(page), inode); 605 end = start + ADDRS_PER_PAGE(page, inode); 606 607 set_new_dnode(&dn, inode, NULL, NULL, 0); 608 retry_dn: 609 err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE); 610 if (err) { 611 if (err == -ENOMEM) { 612 memalloc_retry_wait(GFP_NOFS); 613 goto retry_dn; 614 } 615 goto out; 616 } 617 618 f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true); 619 620 err = f2fs_get_node_info(sbi, dn.nid, &ni, false); 621 if (err) 622 goto err; 623 624 f2fs_bug_on(sbi, ni.ino != ino_of_node(page)); 625 626 if (ofs_of_node(dn.node_page) != ofs_of_node(page)) { 627 f2fs_warn(sbi, "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u", 628 inode->i_ino, ofs_of_node(dn.node_page), 629 ofs_of_node(page)); 630 err = -EFSCORRUPTED; 631 goto err; 632 } 633 634 for (; start < end; start++, dn.ofs_in_node++) { 635 block_t src, dest; 636 637 src = f2fs_data_blkaddr(&dn); 638 dest = data_blkaddr(dn.inode, page, dn.ofs_in_node); 639 640 if (__is_valid_data_blkaddr(src) && 641 !f2fs_is_valid_blkaddr(sbi, src, META_POR)) { 642 err = -EFSCORRUPTED; 643 goto err; 644 } 645 646 if (__is_valid_data_blkaddr(dest) && 647 !f2fs_is_valid_blkaddr(sbi, dest, META_POR)) { 648 err = -EFSCORRUPTED; 649 goto err; 650 } 651 652 /* skip recovering if dest is the same as src */ 653 if (src == dest) 654 continue; 655 656 /* dest is invalid, just invalidate src block */ 657 if (dest == NULL_ADDR) { 658 f2fs_truncate_data_blocks_range(&dn, 1); 659 continue; 660 } 661 662 if (!file_keep_isize(inode) && 663 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT))) 664 f2fs_i_size_write(inode, 665 (loff_t)(start + 1) << PAGE_SHIFT); 666 667 /* 668 * dest is reserved block, invalidate src block 669 * and then reserve one new block in dnode page. 670 */ 671 if (dest == NEW_ADDR) { 672 f2fs_truncate_data_blocks_range(&dn, 1); 673 f2fs_reserve_new_block(&dn); 674 continue; 675 } 676 677 /* dest is valid block, try to recover from src to dest */ 678 if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) { 679 680 if (src == NULL_ADDR) { 681 err = f2fs_reserve_new_block(&dn); 682 while (err && 683 IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION)) 684 err = f2fs_reserve_new_block(&dn); 685 /* We should not get -ENOSPC */ 686 f2fs_bug_on(sbi, err); 687 if (err) 688 goto err; 689 } 690 retry_prev: 691 /* Check the previous node page having this index */ 692 err = check_index_in_prev_nodes(sbi, dest, &dn); 693 if (err) { 694 if (err == -ENOMEM) { 695 memalloc_retry_wait(GFP_NOFS); 696 goto retry_prev; 697 } 698 goto err; 699 } 700 701 /* write dummy data page */ 702 f2fs_replace_block(sbi, &dn, src, dest, 703 ni.version, false, false); 704 recovered++; 705 } 706 } 707 708 copy_node_footer(dn.node_page, page); 709 fill_node_footer(dn.node_page, dn.nid, ni.ino, 710 ofs_of_node(page), false); 711 set_page_dirty(dn.node_page); 712 err: 713 f2fs_put_dnode(&dn); 714 out: 715 f2fs_notice(sbi, "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d", 716 inode->i_ino, file_keep_isize(inode) ? "keep" : "recover", 717 recovered, err); 718 return err; 719 } 720 721 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list, 722 struct list_head *tmp_inode_list, struct list_head *dir_list) 723 { 724 struct curseg_info *curseg; 725 struct page *page = NULL; 726 int err = 0; 727 block_t blkaddr; 728 unsigned int ra_blocks = RECOVERY_MAX_RA_BLOCKS; 729 730 /* get node pages in the current segment */ 731 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE); 732 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg); 733 734 while (1) { 735 struct fsync_inode_entry *entry; 736 737 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR)) 738 break; 739 740 page = f2fs_get_tmp_page(sbi, blkaddr); 741 if (IS_ERR(page)) { 742 err = PTR_ERR(page); 743 break; 744 } 745 746 if (!is_recoverable_dnode(page)) { 747 f2fs_put_page(page, 1); 748 break; 749 } 750 751 entry = get_fsync_inode(inode_list, ino_of_node(page)); 752 if (!entry) 753 goto next; 754 /* 755 * inode(x) | CP | inode(x) | dnode(F) 756 * In this case, we can lose the latest inode(x). 757 * So, call recover_inode for the inode update. 758 */ 759 if (IS_INODE(page)) { 760 err = recover_inode(entry->inode, page); 761 if (err) { 762 f2fs_put_page(page, 1); 763 break; 764 } 765 } 766 if (entry->last_dentry == blkaddr) { 767 err = recover_dentry(entry->inode, page, dir_list); 768 if (err) { 769 f2fs_put_page(page, 1); 770 break; 771 } 772 } 773 err = do_recover_data(sbi, entry->inode, page); 774 if (err) { 775 f2fs_put_page(page, 1); 776 break; 777 } 778 779 if (entry->blkaddr == blkaddr) 780 list_move_tail(&entry->list, tmp_inode_list); 781 next: 782 ra_blocks = adjust_por_ra_blocks(sbi, ra_blocks, blkaddr, 783 next_blkaddr_of_node(page)); 784 785 /* check next segment */ 786 blkaddr = next_blkaddr_of_node(page); 787 f2fs_put_page(page, 1); 788 789 f2fs_ra_meta_pages_cond(sbi, blkaddr, ra_blocks); 790 } 791 if (!err) 792 f2fs_allocate_new_segments(sbi); 793 return err; 794 } 795 796 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only) 797 { 798 struct list_head inode_list, tmp_inode_list; 799 struct list_head dir_list; 800 int err; 801 int ret = 0; 802 unsigned long s_flags = sbi->sb->s_flags; 803 bool need_writecp = false; 804 bool fix_curseg_write_pointer = false; 805 #ifdef CONFIG_QUOTA 806 int quota_enabled; 807 #endif 808 809 if (s_flags & SB_RDONLY) { 810 f2fs_info(sbi, "recover fsync data on readonly fs"); 811 sbi->sb->s_flags &= ~SB_RDONLY; 812 } 813 814 #ifdef CONFIG_QUOTA 815 /* Turn on quotas so that they are updated correctly */ 816 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY); 817 #endif 818 819 INIT_LIST_HEAD(&inode_list); 820 INIT_LIST_HEAD(&tmp_inode_list); 821 INIT_LIST_HEAD(&dir_list); 822 823 /* prevent checkpoint */ 824 f2fs_down_write(&sbi->cp_global_sem); 825 826 /* step #1: find fsynced inode numbers */ 827 err = find_fsync_dnodes(sbi, &inode_list, check_only); 828 if (err || list_empty(&inode_list)) 829 goto skip; 830 831 if (check_only) { 832 ret = 1; 833 goto skip; 834 } 835 836 need_writecp = true; 837 838 /* step #2: recover data */ 839 err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list); 840 if (!err) 841 f2fs_bug_on(sbi, !list_empty(&inode_list)); 842 else 843 f2fs_bug_on(sbi, sbi->sb->s_flags & SB_ACTIVE); 844 skip: 845 fix_curseg_write_pointer = !check_only || list_empty(&inode_list); 846 847 destroy_fsync_dnodes(&inode_list, err); 848 destroy_fsync_dnodes(&tmp_inode_list, err); 849 850 /* truncate meta pages to be used by the recovery */ 851 truncate_inode_pages_range(META_MAPPING(sbi), 852 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1); 853 854 if (err) { 855 truncate_inode_pages_final(NODE_MAPPING(sbi)); 856 truncate_inode_pages_final(META_MAPPING(sbi)); 857 } 858 859 /* 860 * If fsync data succeeds or there is no fsync data to recover, 861 * and the f2fs is not read only, check and fix zoned block devices' 862 * write pointer consistency. 863 */ 864 if (!err && fix_curseg_write_pointer && !f2fs_readonly(sbi->sb) && 865 f2fs_sb_has_blkzoned(sbi)) { 866 err = f2fs_fix_curseg_write_pointer(sbi); 867 ret = err; 868 } 869 870 if (!err) 871 clear_sbi_flag(sbi, SBI_POR_DOING); 872 873 f2fs_up_write(&sbi->cp_global_sem); 874 875 /* let's drop all the directory inodes for clean checkpoint */ 876 destroy_fsync_dnodes(&dir_list, err); 877 878 if (need_writecp) { 879 set_sbi_flag(sbi, SBI_IS_RECOVERED); 880 881 if (!err) { 882 struct cp_control cpc = { 883 .reason = CP_RECOVERY, 884 }; 885 err = f2fs_write_checkpoint(sbi, &cpc); 886 } 887 } 888 889 #ifdef CONFIG_QUOTA 890 /* Turn quotas off */ 891 if (quota_enabled) 892 f2fs_quota_off_umount(sbi->sb); 893 #endif 894 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */ 895 896 return ret ? ret : err; 897 } 898 899 int __init f2fs_create_recovery_cache(void) 900 { 901 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry", 902 sizeof(struct fsync_inode_entry)); 903 if (!fsync_entry_slab) 904 return -ENOMEM; 905 return 0; 906 } 907 908 void f2fs_destroy_recovery_cache(void) 909 { 910 kmem_cache_destroy(fsync_entry_slab); 911 } 912