1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/dir.c 4 * 5 * Copyright (C) 1992, 1993, 1994, 1995 6 * Remy Card (card@masi.ibp.fr) 7 * Laboratoire MASI - Institut Blaise Pascal 8 * Universite Pierre et Marie Curie (Paris VI) 9 * 10 * from 11 * 12 * linux/fs/minix/dir.c 13 * 14 * Copyright (C) 1991, 1992 Linus Torvalds 15 * 16 * ext4 directory handling functions 17 * 18 * Big-endian to little-endian byte-swapping/bitmaps by 19 * David S. Miller (davem@caip.rutgers.edu), 1995 20 * 21 * Hash Tree Directory indexing (c) 2001 Daniel Phillips 22 * 23 */ 24 25 #include <linux/fs.h> 26 #include <linux/buffer_head.h> 27 #include <linux/slab.h> 28 #include <linux/iversion.h> 29 #include <linux/unicode.h> 30 #include "ext4.h" 31 #include "xattr.h" 32 33 static int ext4_dx_readdir(struct file *, struct dir_context *); 34 35 /** 36 * is_dx_dir() - check if a directory is using htree indexing 37 * @inode: directory inode 38 * 39 * Check if the given dir-inode refers to an htree-indexed directory 40 * (or a directory which could potentially get converted to use htree 41 * indexing). 42 * 43 * Return 1 if it is a dx dir, 0 if not 44 */ 45 static int is_dx_dir(struct inode *inode) 46 { 47 struct super_block *sb = inode->i_sb; 48 49 if (ext4_has_feature_dir_index(inode->i_sb) && 50 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) || 51 ((inode->i_size >> sb->s_blocksize_bits) == 1) || 52 ext4_has_inline_data(inode))) 53 return 1; 54 55 return 0; 56 } 57 58 static bool is_fake_dir_entry(struct ext4_dir_entry_2 *de) 59 { 60 /* Check if . or .. , or skip if namelen is 0 */ 61 if ((de->name_len > 0) && (de->name_len <= 2) && (de->name[0] == '.') && 62 (de->name[1] == '.' || de->name[1] == '\0')) 63 return true; 64 /* Check if this is a csum entry */ 65 if (de->file_type == EXT4_FT_DIR_CSUM) 66 return true; 67 return false; 68 } 69 70 /* 71 * Return 0 if the directory entry is OK, and 1 if there is a problem 72 * 73 * Note: this is the opposite of what ext2 and ext3 historically returned... 74 * 75 * bh passed here can be an inode block or a dir data block, depending 76 * on the inode inline data flag. 77 */ 78 int __ext4_check_dir_entry(const char *function, unsigned int line, 79 struct inode *dir, struct file *filp, 80 struct ext4_dir_entry_2 *de, 81 struct buffer_head *bh, char *buf, int size, 82 unsigned int offset) 83 { 84 const char *error_msg = NULL; 85 const int rlen = ext4_rec_len_from_disk(de->rec_len, 86 dir->i_sb->s_blocksize); 87 const int next_offset = ((char *) de - buf) + rlen; 88 bool fake = is_fake_dir_entry(de); 89 bool has_csum = ext4_has_metadata_csum(dir->i_sb); 90 91 if (unlikely(rlen < ext4_dir_rec_len(1, fake ? NULL : dir))) 92 error_msg = "rec_len is smaller than minimal"; 93 else if (unlikely(rlen % 4 != 0)) 94 error_msg = "rec_len % 4 != 0"; 95 else if (unlikely(rlen < ext4_dir_rec_len(de->name_len, 96 fake ? NULL : dir))) 97 error_msg = "rec_len is too small for name_len"; 98 else if (unlikely(next_offset > size)) 99 error_msg = "directory entry overrun"; 100 else if (unlikely(next_offset > size - ext4_dir_rec_len(1, 101 has_csum ? NULL : dir) && 102 next_offset != size)) 103 error_msg = "directory entry too close to block end"; 104 else if (unlikely(le32_to_cpu(de->inode) > 105 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))) 106 error_msg = "inode out of bounds"; 107 else if (unlikely(next_offset == size && de->name_len == 1 && 108 de->name[0] == '.')) 109 error_msg = "'.' directory cannot be the last in data block"; 110 else 111 return 0; 112 113 if (filp) 114 ext4_error_file(filp, function, line, bh->b_blocknr, 115 "bad entry in directory: %s - offset=%u, " 116 "inode=%u, rec_len=%d, size=%d fake=%d", 117 error_msg, offset, le32_to_cpu(de->inode), 118 rlen, size, fake); 119 else 120 ext4_error_inode(dir, function, line, bh->b_blocknr, 121 "bad entry in directory: %s - offset=%u, " 122 "inode=%u, rec_len=%d, size=%d fake=%d", 123 error_msg, offset, le32_to_cpu(de->inode), 124 rlen, size, fake); 125 126 return 1; 127 } 128 129 static int ext4_readdir(struct file *file, struct dir_context *ctx) 130 { 131 unsigned int offset; 132 int i; 133 struct ext4_dir_entry_2 *de; 134 int err; 135 struct inode *inode = file_inode(file); 136 struct super_block *sb = inode->i_sb; 137 struct buffer_head *bh = NULL; 138 struct fscrypt_str fstr = FSTR_INIT(NULL, 0); 139 140 err = fscrypt_prepare_readdir(inode); 141 if (err) 142 return err; 143 144 if (is_dx_dir(inode)) { 145 err = ext4_dx_readdir(file, ctx); 146 if (err != ERR_BAD_DX_DIR) 147 return err; 148 149 /* Can we just clear INDEX flag to ignore htree information? */ 150 if (!ext4_has_metadata_csum(sb)) { 151 /* 152 * We don't set the inode dirty flag since it's not 153 * critical that it gets flushed back to the disk. 154 */ 155 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX); 156 } 157 } 158 159 if (ext4_has_inline_data(inode)) { 160 int has_inline_data = 1; 161 err = ext4_read_inline_dir(file, ctx, 162 &has_inline_data); 163 if (has_inline_data) 164 return err; 165 } 166 167 if (IS_ENCRYPTED(inode)) { 168 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN, &fstr); 169 if (err < 0) 170 return err; 171 } 172 173 while (ctx->pos < inode->i_size) { 174 struct ext4_map_blocks map; 175 176 if (fatal_signal_pending(current)) { 177 err = -ERESTARTSYS; 178 goto errout; 179 } 180 cond_resched(); 181 offset = ctx->pos & (sb->s_blocksize - 1); 182 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb); 183 map.m_len = 1; 184 err = ext4_map_blocks(NULL, inode, &map, 0); 185 if (err == 0) { 186 /* m_len should never be zero but let's avoid 187 * an infinite loop if it somehow is */ 188 if (map.m_len == 0) 189 map.m_len = 1; 190 ctx->pos += map.m_len * sb->s_blocksize; 191 continue; 192 } 193 if (err > 0) { 194 pgoff_t index = map.m_pblk >> 195 (PAGE_SHIFT - inode->i_blkbits); 196 if (!ra_has_index(&file->f_ra, index)) 197 page_cache_sync_readahead( 198 sb->s_bdev->bd_inode->i_mapping, 199 &file->f_ra, file, 200 index, 1); 201 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT; 202 bh = ext4_bread(NULL, inode, map.m_lblk, 0); 203 if (IS_ERR(bh)) { 204 err = PTR_ERR(bh); 205 bh = NULL; 206 goto errout; 207 } 208 } 209 210 if (!bh) { 211 /* corrupt size? Maybe no more blocks to read */ 212 if (ctx->pos > inode->i_blocks << 9) 213 break; 214 ctx->pos += sb->s_blocksize - offset; 215 continue; 216 } 217 218 /* Check the checksum */ 219 if (!buffer_verified(bh) && 220 !ext4_dirblock_csum_verify(inode, bh)) { 221 EXT4_ERROR_FILE(file, 0, "directory fails checksum " 222 "at offset %llu", 223 (unsigned long long)ctx->pos); 224 ctx->pos += sb->s_blocksize - offset; 225 brelse(bh); 226 bh = NULL; 227 continue; 228 } 229 set_buffer_verified(bh); 230 231 /* If the dir block has changed since the last call to 232 * readdir(2), then we might be pointing to an invalid 233 * dirent right now. Scan from the start of the block 234 * to make sure. */ 235 if (!inode_eq_iversion(inode, file->f_version)) { 236 for (i = 0; i < sb->s_blocksize && i < offset; ) { 237 de = (struct ext4_dir_entry_2 *) 238 (bh->b_data + i); 239 /* It's too expensive to do a full 240 * dirent test each time round this 241 * loop, but we do have to test at 242 * least that it is non-zero. A 243 * failure will be detected in the 244 * dirent test below. */ 245 if (ext4_rec_len_from_disk(de->rec_len, 246 sb->s_blocksize) < ext4_dir_rec_len(1, 247 inode)) 248 break; 249 i += ext4_rec_len_from_disk(de->rec_len, 250 sb->s_blocksize); 251 } 252 offset = i; 253 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1)) 254 | offset; 255 file->f_version = inode_query_iversion(inode); 256 } 257 258 while (ctx->pos < inode->i_size 259 && offset < sb->s_blocksize) { 260 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset); 261 if (ext4_check_dir_entry(inode, file, de, bh, 262 bh->b_data, bh->b_size, 263 offset)) { 264 /* 265 * On error, skip to the next block 266 */ 267 ctx->pos = (ctx->pos | 268 (sb->s_blocksize - 1)) + 1; 269 break; 270 } 271 offset += ext4_rec_len_from_disk(de->rec_len, 272 sb->s_blocksize); 273 if (le32_to_cpu(de->inode)) { 274 if (!IS_ENCRYPTED(inode)) { 275 if (!dir_emit(ctx, de->name, 276 de->name_len, 277 le32_to_cpu(de->inode), 278 get_dtype(sb, de->file_type))) 279 goto done; 280 } else { 281 int save_len = fstr.len; 282 struct fscrypt_str de_name = 283 FSTR_INIT(de->name, 284 de->name_len); 285 u32 hash; 286 u32 minor_hash; 287 288 if (IS_CASEFOLDED(inode)) { 289 hash = EXT4_DIRENT_HASH(de); 290 minor_hash = EXT4_DIRENT_MINOR_HASH(de); 291 } else { 292 hash = 0; 293 minor_hash = 0; 294 } 295 296 /* Directory is encrypted */ 297 err = fscrypt_fname_disk_to_usr(inode, 298 hash, minor_hash, &de_name, &fstr); 299 de_name = fstr; 300 fstr.len = save_len; 301 if (err) 302 goto errout; 303 if (!dir_emit(ctx, 304 de_name.name, de_name.len, 305 le32_to_cpu(de->inode), 306 get_dtype(sb, de->file_type))) 307 goto done; 308 } 309 } 310 ctx->pos += ext4_rec_len_from_disk(de->rec_len, 311 sb->s_blocksize); 312 } 313 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode)) 314 goto done; 315 brelse(bh); 316 bh = NULL; 317 } 318 done: 319 err = 0; 320 errout: 321 fscrypt_fname_free_buffer(&fstr); 322 brelse(bh); 323 return err; 324 } 325 326 static inline int is_32bit_api(void) 327 { 328 #ifdef CONFIG_COMPAT 329 return in_compat_syscall(); 330 #else 331 return (BITS_PER_LONG == 32); 332 #endif 333 } 334 335 /* 336 * These functions convert from the major/minor hash to an f_pos 337 * value for dx directories 338 * 339 * Upper layer (for example NFS) should specify FMODE_32BITHASH or 340 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted 341 * directly on both 32-bit and 64-bit nodes, under such case, neither 342 * FMODE_32BITHASH nor FMODE_64BITHASH is specified. 343 */ 344 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor) 345 { 346 if ((filp->f_mode & FMODE_32BITHASH) || 347 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 348 return major >> 1; 349 else 350 return ((__u64)(major >> 1) << 32) | (__u64)minor; 351 } 352 353 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos) 354 { 355 if ((filp->f_mode & FMODE_32BITHASH) || 356 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 357 return (pos << 1) & 0xffffffff; 358 else 359 return ((pos >> 32) << 1) & 0xffffffff; 360 } 361 362 static inline __u32 pos2min_hash(struct file *filp, loff_t pos) 363 { 364 if ((filp->f_mode & FMODE_32BITHASH) || 365 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 366 return 0; 367 else 368 return pos & 0xffffffff; 369 } 370 371 /* 372 * Return 32- or 64-bit end-of-file for dx directories 373 */ 374 static inline loff_t ext4_get_htree_eof(struct file *filp) 375 { 376 if ((filp->f_mode & FMODE_32BITHASH) || 377 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 378 return EXT4_HTREE_EOF_32BIT; 379 else 380 return EXT4_HTREE_EOF_64BIT; 381 } 382 383 384 /* 385 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree 386 * directories, where the "offset" is in terms of the filename hash 387 * value instead of the byte offset. 388 * 389 * Because we may return a 64-bit hash that is well beyond offset limits, 390 * we need to pass the max hash as the maximum allowable offset in 391 * the htree directory case. 392 * 393 * For non-htree, ext4_llseek already chooses the proper max offset. 394 */ 395 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence) 396 { 397 struct inode *inode = file->f_mapping->host; 398 int dx_dir = is_dx_dir(inode); 399 loff_t ret, htree_max = ext4_get_htree_eof(file); 400 401 if (likely(dx_dir)) 402 ret = generic_file_llseek_size(file, offset, whence, 403 htree_max, htree_max); 404 else 405 ret = ext4_llseek(file, offset, whence); 406 file->f_version = inode_peek_iversion(inode) - 1; 407 return ret; 408 } 409 410 /* 411 * This structure holds the nodes of the red-black tree used to store 412 * the directory entry in hash order. 413 */ 414 struct fname { 415 __u32 hash; 416 __u32 minor_hash; 417 struct rb_node rb_hash; 418 struct fname *next; 419 __u32 inode; 420 __u8 name_len; 421 __u8 file_type; 422 char name[]; 423 }; 424 425 /* 426 * This function implements a non-recursive way of freeing all of the 427 * nodes in the red-black tree. 428 */ 429 static void free_rb_tree_fname(struct rb_root *root) 430 { 431 struct fname *fname, *next; 432 433 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash) 434 while (fname) { 435 struct fname *old = fname; 436 fname = fname->next; 437 kfree(old); 438 } 439 440 *root = RB_ROOT; 441 } 442 443 444 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp, 445 loff_t pos) 446 { 447 struct dir_private_info *p; 448 449 p = kzalloc(sizeof(*p), GFP_KERNEL); 450 if (!p) 451 return NULL; 452 p->curr_hash = pos2maj_hash(filp, pos); 453 p->curr_minor_hash = pos2min_hash(filp, pos); 454 return p; 455 } 456 457 void ext4_htree_free_dir_info(struct dir_private_info *p) 458 { 459 free_rb_tree_fname(&p->root); 460 kfree(p); 461 } 462 463 /* 464 * Given a directory entry, enter it into the fname rb tree. 465 * 466 * When filename encryption is enabled, the dirent will hold the 467 * encrypted filename, while the htree will hold decrypted filename. 468 * The decrypted filename is passed in via ent_name. parameter. 469 */ 470 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash, 471 __u32 minor_hash, 472 struct ext4_dir_entry_2 *dirent, 473 struct fscrypt_str *ent_name) 474 { 475 struct rb_node **p, *parent = NULL; 476 struct fname *fname, *new_fn; 477 struct dir_private_info *info; 478 int len; 479 480 info = dir_file->private_data; 481 p = &info->root.rb_node; 482 483 /* Create and allocate the fname structure */ 484 len = sizeof(struct fname) + ent_name->len + 1; 485 new_fn = kzalloc(len, GFP_KERNEL); 486 if (!new_fn) 487 return -ENOMEM; 488 new_fn->hash = hash; 489 new_fn->minor_hash = minor_hash; 490 new_fn->inode = le32_to_cpu(dirent->inode); 491 new_fn->name_len = ent_name->len; 492 new_fn->file_type = dirent->file_type; 493 memcpy(new_fn->name, ent_name->name, ent_name->len); 494 495 while (*p) { 496 parent = *p; 497 fname = rb_entry(parent, struct fname, rb_hash); 498 499 /* 500 * If the hash and minor hash match up, then we put 501 * them on a linked list. This rarely happens... 502 */ 503 if ((new_fn->hash == fname->hash) && 504 (new_fn->minor_hash == fname->minor_hash)) { 505 new_fn->next = fname->next; 506 fname->next = new_fn; 507 return 0; 508 } 509 510 if (new_fn->hash < fname->hash) 511 p = &(*p)->rb_left; 512 else if (new_fn->hash > fname->hash) 513 p = &(*p)->rb_right; 514 else if (new_fn->minor_hash < fname->minor_hash) 515 p = &(*p)->rb_left; 516 else /* if (new_fn->minor_hash > fname->minor_hash) */ 517 p = &(*p)->rb_right; 518 } 519 520 rb_link_node(&new_fn->rb_hash, parent, p); 521 rb_insert_color(&new_fn->rb_hash, &info->root); 522 return 0; 523 } 524 525 526 527 /* 528 * This is a helper function for ext4_dx_readdir. It calls filldir 529 * for all entries on the fname linked list. (Normally there is only 530 * one entry on the linked list, unless there are 62 bit hash collisions.) 531 */ 532 static int call_filldir(struct file *file, struct dir_context *ctx, 533 struct fname *fname) 534 { 535 struct dir_private_info *info = file->private_data; 536 struct inode *inode = file_inode(file); 537 struct super_block *sb = inode->i_sb; 538 539 if (!fname) { 540 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: " 541 "called with null fname?!?", __func__, __LINE__, 542 inode->i_ino, current->comm); 543 return 0; 544 } 545 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash); 546 while (fname) { 547 if (!dir_emit(ctx, fname->name, 548 fname->name_len, 549 fname->inode, 550 get_dtype(sb, fname->file_type))) { 551 info->extra_fname = fname; 552 return 1; 553 } 554 fname = fname->next; 555 } 556 return 0; 557 } 558 559 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx) 560 { 561 struct dir_private_info *info = file->private_data; 562 struct inode *inode = file_inode(file); 563 struct fname *fname; 564 int ret = 0; 565 566 if (!info) { 567 info = ext4_htree_create_dir_info(file, ctx->pos); 568 if (!info) 569 return -ENOMEM; 570 file->private_data = info; 571 } 572 573 if (ctx->pos == ext4_get_htree_eof(file)) 574 return 0; /* EOF */ 575 576 /* Some one has messed with f_pos; reset the world */ 577 if (info->last_pos != ctx->pos) { 578 free_rb_tree_fname(&info->root); 579 info->curr_node = NULL; 580 info->extra_fname = NULL; 581 info->curr_hash = pos2maj_hash(file, ctx->pos); 582 info->curr_minor_hash = pos2min_hash(file, ctx->pos); 583 } 584 585 /* 586 * If there are any leftover names on the hash collision 587 * chain, return them first. 588 */ 589 if (info->extra_fname) { 590 if (call_filldir(file, ctx, info->extra_fname)) 591 goto finished; 592 info->extra_fname = NULL; 593 goto next_node; 594 } else if (!info->curr_node) 595 info->curr_node = rb_first(&info->root); 596 597 while (1) { 598 /* 599 * Fill the rbtree if we have no more entries, 600 * or the inode has changed since we last read in the 601 * cached entries. 602 */ 603 if ((!info->curr_node) || 604 !inode_eq_iversion(inode, file->f_version)) { 605 info->curr_node = NULL; 606 free_rb_tree_fname(&info->root); 607 file->f_version = inode_query_iversion(inode); 608 ret = ext4_htree_fill_tree(file, info->curr_hash, 609 info->curr_minor_hash, 610 &info->next_hash); 611 if (ret < 0) 612 goto finished; 613 if (ret == 0) { 614 ctx->pos = ext4_get_htree_eof(file); 615 break; 616 } 617 info->curr_node = rb_first(&info->root); 618 } 619 620 fname = rb_entry(info->curr_node, struct fname, rb_hash); 621 info->curr_hash = fname->hash; 622 info->curr_minor_hash = fname->minor_hash; 623 if (call_filldir(file, ctx, fname)) 624 break; 625 next_node: 626 info->curr_node = rb_next(info->curr_node); 627 if (info->curr_node) { 628 fname = rb_entry(info->curr_node, struct fname, 629 rb_hash); 630 info->curr_hash = fname->hash; 631 info->curr_minor_hash = fname->minor_hash; 632 } else { 633 if (info->next_hash == ~0) { 634 ctx->pos = ext4_get_htree_eof(file); 635 break; 636 } 637 info->curr_hash = info->next_hash; 638 info->curr_minor_hash = 0; 639 } 640 } 641 finished: 642 info->last_pos = ctx->pos; 643 return ret < 0 ? ret : 0; 644 } 645 646 static int ext4_release_dir(struct inode *inode, struct file *filp) 647 { 648 if (filp->private_data) 649 ext4_htree_free_dir_info(filp->private_data); 650 651 return 0; 652 } 653 654 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf, 655 int buf_size) 656 { 657 struct ext4_dir_entry_2 *de; 658 int rlen; 659 unsigned int offset = 0; 660 char *top; 661 662 de = buf; 663 top = buf + buf_size; 664 while ((char *) de < top) { 665 if (ext4_check_dir_entry(dir, NULL, de, bh, 666 buf, buf_size, offset)) 667 return -EFSCORRUPTED; 668 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 669 de = (struct ext4_dir_entry_2 *)((char *)de + rlen); 670 offset += rlen; 671 } 672 if ((char *) de > top) 673 return -EFSCORRUPTED; 674 675 return 0; 676 } 677 678 const struct file_operations ext4_dir_operations = { 679 .llseek = ext4_dir_llseek, 680 .read = generic_read_dir, 681 .iterate_shared = ext4_readdir, 682 .unlocked_ioctl = ext4_ioctl, 683 #ifdef CONFIG_COMPAT 684 .compat_ioctl = ext4_compat_ioctl, 685 #endif 686 .fsync = ext4_sync_file, 687 .release = ext4_release_dir, 688 }; 689