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 /* 59 * Return 0 if the directory entry is OK, and 1 if there is a problem 60 * 61 * Note: this is the opposite of what ext2 and ext3 historically returned... 62 * 63 * bh passed here can be an inode block or a dir data block, depending 64 * on the inode inline data flag. 65 */ 66 int __ext4_check_dir_entry(const char *function, unsigned int line, 67 struct inode *dir, struct file *filp, 68 struct ext4_dir_entry_2 *de, 69 struct buffer_head *bh, char *buf, int size, 70 unsigned int offset) 71 { 72 const char *error_msg = NULL; 73 const int rlen = ext4_rec_len_from_disk(de->rec_len, 74 dir->i_sb->s_blocksize); 75 76 if (unlikely(rlen < EXT4_DIR_REC_LEN(1))) 77 error_msg = "rec_len is smaller than minimal"; 78 else if (unlikely(rlen % 4 != 0)) 79 error_msg = "rec_len % 4 != 0"; 80 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len))) 81 error_msg = "rec_len is too small for name_len"; 82 else if (unlikely(((char *) de - buf) + rlen > size)) 83 error_msg = "directory entry overrun"; 84 else if (unlikely(le32_to_cpu(de->inode) > 85 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))) 86 error_msg = "inode out of bounds"; 87 else 88 return 0; 89 90 if (filp) 91 ext4_error_file(filp, function, line, bh->b_blocknr, 92 "bad entry in directory: %s - offset=%u, " 93 "inode=%u, rec_len=%d, name_len=%d, size=%d", 94 error_msg, offset, le32_to_cpu(de->inode), 95 rlen, de->name_len, size); 96 else 97 ext4_error_inode(dir, function, line, bh->b_blocknr, 98 "bad entry in directory: %s - offset=%u, " 99 "inode=%u, rec_len=%d, name_len=%d, size=%d", 100 error_msg, offset, le32_to_cpu(de->inode), 101 rlen, de->name_len, size); 102 103 return 1; 104 } 105 106 static int ext4_readdir(struct file *file, struct dir_context *ctx) 107 { 108 unsigned int offset; 109 int i; 110 struct ext4_dir_entry_2 *de; 111 int err; 112 struct inode *inode = file_inode(file); 113 struct super_block *sb = inode->i_sb; 114 struct buffer_head *bh = NULL; 115 struct fscrypt_str fstr = FSTR_INIT(NULL, 0); 116 117 if (IS_ENCRYPTED(inode)) { 118 err = fscrypt_get_encryption_info(inode); 119 if (err && err != -ENOKEY) 120 return err; 121 } 122 123 if (is_dx_dir(inode)) { 124 err = ext4_dx_readdir(file, ctx); 125 if (err != ERR_BAD_DX_DIR) { 126 return err; 127 } 128 /* 129 * We don't set the inode dirty flag since it's not 130 * critical that it get flushed back to the disk. 131 */ 132 ext4_clear_inode_flag(file_inode(file), 133 EXT4_INODE_INDEX); 134 } 135 136 if (ext4_has_inline_data(inode)) { 137 int has_inline_data = 1; 138 err = ext4_read_inline_dir(file, ctx, 139 &has_inline_data); 140 if (has_inline_data) 141 return err; 142 } 143 144 if (IS_ENCRYPTED(inode)) { 145 err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr); 146 if (err < 0) 147 return err; 148 } 149 150 while (ctx->pos < inode->i_size) { 151 struct ext4_map_blocks map; 152 153 if (fatal_signal_pending(current)) { 154 err = -ERESTARTSYS; 155 goto errout; 156 } 157 cond_resched(); 158 offset = ctx->pos & (sb->s_blocksize - 1); 159 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb); 160 map.m_len = 1; 161 err = ext4_map_blocks(NULL, inode, &map, 0); 162 if (err == 0) { 163 /* m_len should never be zero but let's avoid 164 * an infinite loop if it somehow is */ 165 if (map.m_len == 0) 166 map.m_len = 1; 167 ctx->pos += map.m_len * sb->s_blocksize; 168 continue; 169 } 170 if (err > 0) { 171 pgoff_t index = map.m_pblk >> 172 (PAGE_SHIFT - inode->i_blkbits); 173 if (!ra_has_index(&file->f_ra, index)) 174 page_cache_sync_readahead( 175 sb->s_bdev->bd_inode->i_mapping, 176 &file->f_ra, file, 177 index, 1); 178 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT; 179 bh = ext4_bread(NULL, inode, map.m_lblk, 0); 180 if (IS_ERR(bh)) { 181 err = PTR_ERR(bh); 182 bh = NULL; 183 goto errout; 184 } 185 } 186 187 if (!bh) { 188 /* corrupt size? Maybe no more blocks to read */ 189 if (ctx->pos > inode->i_blocks << 9) 190 break; 191 ctx->pos += sb->s_blocksize - offset; 192 continue; 193 } 194 195 /* Check the checksum */ 196 if (!buffer_verified(bh) && 197 !ext4_dirent_csum_verify(inode, 198 (struct ext4_dir_entry *)bh->b_data)) { 199 EXT4_ERROR_FILE(file, 0, "directory fails checksum " 200 "at offset %llu", 201 (unsigned long long)ctx->pos); 202 ctx->pos += sb->s_blocksize - offset; 203 brelse(bh); 204 bh = NULL; 205 continue; 206 } 207 set_buffer_verified(bh); 208 209 /* If the dir block has changed since the last call to 210 * readdir(2), then we might be pointing to an invalid 211 * dirent right now. Scan from the start of the block 212 * to make sure. */ 213 if (!inode_eq_iversion(inode, file->f_version)) { 214 for (i = 0; i < sb->s_blocksize && i < offset; ) { 215 de = (struct ext4_dir_entry_2 *) 216 (bh->b_data + i); 217 /* It's too expensive to do a full 218 * dirent test each time round this 219 * loop, but we do have to test at 220 * least that it is non-zero. A 221 * failure will be detected in the 222 * dirent test below. */ 223 if (ext4_rec_len_from_disk(de->rec_len, 224 sb->s_blocksize) < EXT4_DIR_REC_LEN(1)) 225 break; 226 i += ext4_rec_len_from_disk(de->rec_len, 227 sb->s_blocksize); 228 } 229 offset = i; 230 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1)) 231 | offset; 232 file->f_version = inode_query_iversion(inode); 233 } 234 235 while (ctx->pos < inode->i_size 236 && offset < sb->s_blocksize) { 237 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset); 238 if (ext4_check_dir_entry(inode, file, de, bh, 239 bh->b_data, bh->b_size, 240 offset)) { 241 /* 242 * On error, skip to the next block 243 */ 244 ctx->pos = (ctx->pos | 245 (sb->s_blocksize - 1)) + 1; 246 break; 247 } 248 offset += ext4_rec_len_from_disk(de->rec_len, 249 sb->s_blocksize); 250 if (le32_to_cpu(de->inode)) { 251 if (!IS_ENCRYPTED(inode)) { 252 if (!dir_emit(ctx, de->name, 253 de->name_len, 254 le32_to_cpu(de->inode), 255 get_dtype(sb, de->file_type))) 256 goto done; 257 } else { 258 int save_len = fstr.len; 259 struct fscrypt_str de_name = 260 FSTR_INIT(de->name, 261 de->name_len); 262 263 /* Directory is encrypted */ 264 err = fscrypt_fname_disk_to_usr(inode, 265 0, 0, &de_name, &fstr); 266 de_name = fstr; 267 fstr.len = save_len; 268 if (err) 269 goto errout; 270 if (!dir_emit(ctx, 271 de_name.name, de_name.len, 272 le32_to_cpu(de->inode), 273 get_dtype(sb, de->file_type))) 274 goto done; 275 } 276 } 277 ctx->pos += ext4_rec_len_from_disk(de->rec_len, 278 sb->s_blocksize); 279 } 280 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode)) 281 goto done; 282 brelse(bh); 283 bh = NULL; 284 offset = 0; 285 } 286 done: 287 err = 0; 288 errout: 289 fscrypt_fname_free_buffer(&fstr); 290 brelse(bh); 291 return err; 292 } 293 294 static inline int is_32bit_api(void) 295 { 296 #ifdef CONFIG_COMPAT 297 return in_compat_syscall(); 298 #else 299 return (BITS_PER_LONG == 32); 300 #endif 301 } 302 303 /* 304 * These functions convert from the major/minor hash to an f_pos 305 * value for dx directories 306 * 307 * Upper layer (for example NFS) should specify FMODE_32BITHASH or 308 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted 309 * directly on both 32-bit and 64-bit nodes, under such case, neither 310 * FMODE_32BITHASH nor FMODE_64BITHASH is specified. 311 */ 312 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor) 313 { 314 if ((filp->f_mode & FMODE_32BITHASH) || 315 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 316 return major >> 1; 317 else 318 return ((__u64)(major >> 1) << 32) | (__u64)minor; 319 } 320 321 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos) 322 { 323 if ((filp->f_mode & FMODE_32BITHASH) || 324 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 325 return (pos << 1) & 0xffffffff; 326 else 327 return ((pos >> 32) << 1) & 0xffffffff; 328 } 329 330 static inline __u32 pos2min_hash(struct file *filp, loff_t pos) 331 { 332 if ((filp->f_mode & FMODE_32BITHASH) || 333 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 334 return 0; 335 else 336 return pos & 0xffffffff; 337 } 338 339 /* 340 * Return 32- or 64-bit end-of-file for dx directories 341 */ 342 static inline loff_t ext4_get_htree_eof(struct file *filp) 343 { 344 if ((filp->f_mode & FMODE_32BITHASH) || 345 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 346 return EXT4_HTREE_EOF_32BIT; 347 else 348 return EXT4_HTREE_EOF_64BIT; 349 } 350 351 352 /* 353 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree 354 * directories, where the "offset" is in terms of the filename hash 355 * value instead of the byte offset. 356 * 357 * Because we may return a 64-bit hash that is well beyond offset limits, 358 * we need to pass the max hash as the maximum allowable offset in 359 * the htree directory case. 360 * 361 * For non-htree, ext4_llseek already chooses the proper max offset. 362 */ 363 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence) 364 { 365 struct inode *inode = file->f_mapping->host; 366 int dx_dir = is_dx_dir(inode); 367 loff_t ret, htree_max = ext4_get_htree_eof(file); 368 369 if (likely(dx_dir)) 370 ret = generic_file_llseek_size(file, offset, whence, 371 htree_max, htree_max); 372 else 373 ret = ext4_llseek(file, offset, whence); 374 file->f_version = inode_peek_iversion(inode) - 1; 375 return ret; 376 } 377 378 /* 379 * This structure holds the nodes of the red-black tree used to store 380 * the directory entry in hash order. 381 */ 382 struct fname { 383 __u32 hash; 384 __u32 minor_hash; 385 struct rb_node rb_hash; 386 struct fname *next; 387 __u32 inode; 388 __u8 name_len; 389 __u8 file_type; 390 char name[0]; 391 }; 392 393 /* 394 * This functoin implements a non-recursive way of freeing all of the 395 * nodes in the red-black tree. 396 */ 397 static void free_rb_tree_fname(struct rb_root *root) 398 { 399 struct fname *fname, *next; 400 401 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash) 402 while (fname) { 403 struct fname *old = fname; 404 fname = fname->next; 405 kfree(old); 406 } 407 408 *root = RB_ROOT; 409 } 410 411 412 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp, 413 loff_t pos) 414 { 415 struct dir_private_info *p; 416 417 p = kzalloc(sizeof(*p), GFP_KERNEL); 418 if (!p) 419 return NULL; 420 p->curr_hash = pos2maj_hash(filp, pos); 421 p->curr_minor_hash = pos2min_hash(filp, pos); 422 return p; 423 } 424 425 void ext4_htree_free_dir_info(struct dir_private_info *p) 426 { 427 free_rb_tree_fname(&p->root); 428 kfree(p); 429 } 430 431 /* 432 * Given a directory entry, enter it into the fname rb tree. 433 * 434 * When filename encryption is enabled, the dirent will hold the 435 * encrypted filename, while the htree will hold decrypted filename. 436 * The decrypted filename is passed in via ent_name. parameter. 437 */ 438 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash, 439 __u32 minor_hash, 440 struct ext4_dir_entry_2 *dirent, 441 struct fscrypt_str *ent_name) 442 { 443 struct rb_node **p, *parent = NULL; 444 struct fname *fname, *new_fn; 445 struct dir_private_info *info; 446 int len; 447 448 info = dir_file->private_data; 449 p = &info->root.rb_node; 450 451 /* Create and allocate the fname structure */ 452 len = sizeof(struct fname) + ent_name->len + 1; 453 new_fn = kzalloc(len, GFP_KERNEL); 454 if (!new_fn) 455 return -ENOMEM; 456 new_fn->hash = hash; 457 new_fn->minor_hash = minor_hash; 458 new_fn->inode = le32_to_cpu(dirent->inode); 459 new_fn->name_len = ent_name->len; 460 new_fn->file_type = dirent->file_type; 461 memcpy(new_fn->name, ent_name->name, ent_name->len); 462 new_fn->name[ent_name->len] = 0; 463 464 while (*p) { 465 parent = *p; 466 fname = rb_entry(parent, struct fname, rb_hash); 467 468 /* 469 * If the hash and minor hash match up, then we put 470 * them on a linked list. This rarely happens... 471 */ 472 if ((new_fn->hash == fname->hash) && 473 (new_fn->minor_hash == fname->minor_hash)) { 474 new_fn->next = fname->next; 475 fname->next = new_fn; 476 return 0; 477 } 478 479 if (new_fn->hash < fname->hash) 480 p = &(*p)->rb_left; 481 else if (new_fn->hash > fname->hash) 482 p = &(*p)->rb_right; 483 else if (new_fn->minor_hash < fname->minor_hash) 484 p = &(*p)->rb_left; 485 else /* if (new_fn->minor_hash > fname->minor_hash) */ 486 p = &(*p)->rb_right; 487 } 488 489 rb_link_node(&new_fn->rb_hash, parent, p); 490 rb_insert_color(&new_fn->rb_hash, &info->root); 491 return 0; 492 } 493 494 495 496 /* 497 * This is a helper function for ext4_dx_readdir. It calls filldir 498 * for all entres on the fname linked list. (Normally there is only 499 * one entry on the linked list, unless there are 62 bit hash collisions.) 500 */ 501 static int call_filldir(struct file *file, struct dir_context *ctx, 502 struct fname *fname) 503 { 504 struct dir_private_info *info = file->private_data; 505 struct inode *inode = file_inode(file); 506 struct super_block *sb = inode->i_sb; 507 508 if (!fname) { 509 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: " 510 "called with null fname?!?", __func__, __LINE__, 511 inode->i_ino, current->comm); 512 return 0; 513 } 514 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash); 515 while (fname) { 516 if (!dir_emit(ctx, fname->name, 517 fname->name_len, 518 fname->inode, 519 get_dtype(sb, fname->file_type))) { 520 info->extra_fname = fname; 521 return 1; 522 } 523 fname = fname->next; 524 } 525 return 0; 526 } 527 528 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx) 529 { 530 struct dir_private_info *info = file->private_data; 531 struct inode *inode = file_inode(file); 532 struct fname *fname; 533 int ret; 534 535 if (!info) { 536 info = ext4_htree_create_dir_info(file, ctx->pos); 537 if (!info) 538 return -ENOMEM; 539 file->private_data = info; 540 } 541 542 if (ctx->pos == ext4_get_htree_eof(file)) 543 return 0; /* EOF */ 544 545 /* Some one has messed with f_pos; reset the world */ 546 if (info->last_pos != ctx->pos) { 547 free_rb_tree_fname(&info->root); 548 info->curr_node = NULL; 549 info->extra_fname = NULL; 550 info->curr_hash = pos2maj_hash(file, ctx->pos); 551 info->curr_minor_hash = pos2min_hash(file, ctx->pos); 552 } 553 554 /* 555 * If there are any leftover names on the hash collision 556 * chain, return them first. 557 */ 558 if (info->extra_fname) { 559 if (call_filldir(file, ctx, info->extra_fname)) 560 goto finished; 561 info->extra_fname = NULL; 562 goto next_node; 563 } else if (!info->curr_node) 564 info->curr_node = rb_first(&info->root); 565 566 while (1) { 567 /* 568 * Fill the rbtree if we have no more entries, 569 * or the inode has changed since we last read in the 570 * cached entries. 571 */ 572 if ((!info->curr_node) || 573 !inode_eq_iversion(inode, file->f_version)) { 574 info->curr_node = NULL; 575 free_rb_tree_fname(&info->root); 576 file->f_version = inode_query_iversion(inode); 577 ret = ext4_htree_fill_tree(file, info->curr_hash, 578 info->curr_minor_hash, 579 &info->next_hash); 580 if (ret < 0) 581 return ret; 582 if (ret == 0) { 583 ctx->pos = ext4_get_htree_eof(file); 584 break; 585 } 586 info->curr_node = rb_first(&info->root); 587 } 588 589 fname = rb_entry(info->curr_node, struct fname, rb_hash); 590 info->curr_hash = fname->hash; 591 info->curr_minor_hash = fname->minor_hash; 592 if (call_filldir(file, ctx, fname)) 593 break; 594 next_node: 595 info->curr_node = rb_next(info->curr_node); 596 if (info->curr_node) { 597 fname = rb_entry(info->curr_node, struct fname, 598 rb_hash); 599 info->curr_hash = fname->hash; 600 info->curr_minor_hash = fname->minor_hash; 601 } else { 602 if (info->next_hash == ~0) { 603 ctx->pos = ext4_get_htree_eof(file); 604 break; 605 } 606 info->curr_hash = info->next_hash; 607 info->curr_minor_hash = 0; 608 } 609 } 610 finished: 611 info->last_pos = ctx->pos; 612 return 0; 613 } 614 615 static int ext4_dir_open(struct inode * inode, struct file * filp) 616 { 617 if (IS_ENCRYPTED(inode)) 618 return fscrypt_get_encryption_info(inode) ? -EACCES : 0; 619 return 0; 620 } 621 622 static int ext4_release_dir(struct inode *inode, struct file *filp) 623 { 624 if (filp->private_data) 625 ext4_htree_free_dir_info(filp->private_data); 626 627 return 0; 628 } 629 630 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf, 631 int buf_size) 632 { 633 struct ext4_dir_entry_2 *de; 634 int rlen; 635 unsigned int offset = 0; 636 char *top; 637 638 de = (struct ext4_dir_entry_2 *)buf; 639 top = buf + buf_size; 640 while ((char *) de < top) { 641 if (ext4_check_dir_entry(dir, NULL, de, bh, 642 buf, buf_size, offset)) 643 return -EFSCORRUPTED; 644 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 645 de = (struct ext4_dir_entry_2 *)((char *)de + rlen); 646 offset += rlen; 647 } 648 if ((char *) de > top) 649 return -EFSCORRUPTED; 650 651 return 0; 652 } 653 654 const struct file_operations ext4_dir_operations = { 655 .llseek = ext4_dir_llseek, 656 .read = generic_read_dir, 657 .iterate_shared = ext4_readdir, 658 .unlocked_ioctl = ext4_ioctl, 659 #ifdef CONFIG_COMPAT 660 .compat_ioctl = ext4_compat_ioctl, 661 #endif 662 .fsync = ext4_sync_file, 663 .open = ext4_dir_open, 664 .release = ext4_release_dir, 665 }; 666 667 #ifdef CONFIG_UNICODE 668 static int ext4_d_compare(const struct dentry *dentry, unsigned int len, 669 const char *str, const struct qstr *name) 670 { 671 struct qstr qstr = {.name = str, .len = len }; 672 673 if (!IS_CASEFOLDED(dentry->d_parent->d_inode)) { 674 if (len != name->len) 675 return -1; 676 return memcmp(str, name->name, len); 677 } 678 679 return ext4_ci_compare(dentry->d_parent->d_inode, name, &qstr, false); 680 } 681 682 static int ext4_d_hash(const struct dentry *dentry, struct qstr *str) 683 { 684 const struct ext4_sb_info *sbi = EXT4_SB(dentry->d_sb); 685 const struct unicode_map *um = sbi->s_encoding; 686 unsigned char *norm; 687 int len, ret = 0; 688 689 if (!IS_CASEFOLDED(dentry->d_inode)) 690 return 0; 691 692 norm = kmalloc(PATH_MAX, GFP_ATOMIC); 693 if (!norm) 694 return -ENOMEM; 695 696 len = utf8_casefold(um, str, norm, PATH_MAX); 697 if (len < 0) { 698 if (ext4_has_strict_mode(sbi)) 699 ret = -EINVAL; 700 goto out; 701 } 702 str->hash = full_name_hash(dentry, norm, len); 703 out: 704 kfree(norm); 705 return ret; 706 } 707 708 const struct dentry_operations ext4_dentry_ops = { 709 .d_hash = ext4_d_hash, 710 .d_compare = ext4_d_compare, 711 }; 712 #endif 713