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(((char *) de - buf) + rlen > 85 size - EXT4_DIR_REC_LEN(1) && 86 ((char *) de - buf) + rlen != size)) { 87 error_msg = "directory entry too close to block end"; 88 } 89 else if (unlikely(le32_to_cpu(de->inode) > 90 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))) 91 error_msg = "inode out of bounds"; 92 else 93 return 0; 94 95 if (filp) 96 ext4_error_file(filp, function, line, bh->b_blocknr, 97 "bad entry in directory: %s - offset=%u, " 98 "inode=%u, rec_len=%d, name_len=%d, size=%d", 99 error_msg, offset, le32_to_cpu(de->inode), 100 rlen, de->name_len, size); 101 else 102 ext4_error_inode(dir, function, line, bh->b_blocknr, 103 "bad entry in directory: %s - offset=%u, " 104 "inode=%u, rec_len=%d, name_len=%d, size=%d", 105 error_msg, offset, le32_to_cpu(de->inode), 106 rlen, de->name_len, size); 107 108 return 1; 109 } 110 111 static int ext4_readdir(struct file *file, struct dir_context *ctx) 112 { 113 unsigned int offset; 114 int i; 115 struct ext4_dir_entry_2 *de; 116 int err; 117 struct inode *inode = file_inode(file); 118 struct super_block *sb = inode->i_sb; 119 struct buffer_head *bh = NULL; 120 struct fscrypt_str fstr = FSTR_INIT(NULL, 0); 121 122 if (IS_ENCRYPTED(inode)) { 123 err = fscrypt_get_encryption_info(inode); 124 if (err && err != -ENOKEY) 125 return err; 126 } 127 128 if (is_dx_dir(inode)) { 129 err = ext4_dx_readdir(file, ctx); 130 if (err != ERR_BAD_DX_DIR) { 131 return err; 132 } 133 /* 134 * We don't set the inode dirty flag since it's not 135 * critical that it get flushed back to the disk. 136 */ 137 ext4_clear_inode_flag(file_inode(file), 138 EXT4_INODE_INDEX); 139 } 140 141 if (ext4_has_inline_data(inode)) { 142 int has_inline_data = 1; 143 err = ext4_read_inline_dir(file, ctx, 144 &has_inline_data); 145 if (has_inline_data) 146 return err; 147 } 148 149 if (IS_ENCRYPTED(inode)) { 150 err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr); 151 if (err < 0) 152 return err; 153 } 154 155 while (ctx->pos < inode->i_size) { 156 struct ext4_map_blocks map; 157 158 if (fatal_signal_pending(current)) { 159 err = -ERESTARTSYS; 160 goto errout; 161 } 162 cond_resched(); 163 offset = ctx->pos & (sb->s_blocksize - 1); 164 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb); 165 map.m_len = 1; 166 err = ext4_map_blocks(NULL, inode, &map, 0); 167 if (err == 0) { 168 /* m_len should never be zero but let's avoid 169 * an infinite loop if it somehow is */ 170 if (map.m_len == 0) 171 map.m_len = 1; 172 ctx->pos += map.m_len * sb->s_blocksize; 173 continue; 174 } 175 if (err > 0) { 176 pgoff_t index = map.m_pblk >> 177 (PAGE_SHIFT - inode->i_blkbits); 178 if (!ra_has_index(&file->f_ra, index)) 179 page_cache_sync_readahead( 180 sb->s_bdev->bd_inode->i_mapping, 181 &file->f_ra, file, 182 index, 1); 183 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT; 184 bh = ext4_bread(NULL, inode, map.m_lblk, 0); 185 if (IS_ERR(bh)) { 186 err = PTR_ERR(bh); 187 bh = NULL; 188 goto errout; 189 } 190 } 191 192 if (!bh) { 193 /* corrupt size? Maybe no more blocks to read */ 194 if (ctx->pos > inode->i_blocks << 9) 195 break; 196 ctx->pos += sb->s_blocksize - offset; 197 continue; 198 } 199 200 /* Check the checksum */ 201 if (!buffer_verified(bh) && 202 !ext4_dirblock_csum_verify(inode, bh)) { 203 EXT4_ERROR_FILE(file, 0, "directory fails checksum " 204 "at offset %llu", 205 (unsigned long long)ctx->pos); 206 ctx->pos += sb->s_blocksize - offset; 207 brelse(bh); 208 bh = NULL; 209 continue; 210 } 211 set_buffer_verified(bh); 212 213 /* If the dir block has changed since the last call to 214 * readdir(2), then we might be pointing to an invalid 215 * dirent right now. Scan from the start of the block 216 * to make sure. */ 217 if (!inode_eq_iversion(inode, file->f_version)) { 218 for (i = 0; i < sb->s_blocksize && i < offset; ) { 219 de = (struct ext4_dir_entry_2 *) 220 (bh->b_data + i); 221 /* It's too expensive to do a full 222 * dirent test each time round this 223 * loop, but we do have to test at 224 * least that it is non-zero. A 225 * failure will be detected in the 226 * dirent test below. */ 227 if (ext4_rec_len_from_disk(de->rec_len, 228 sb->s_blocksize) < EXT4_DIR_REC_LEN(1)) 229 break; 230 i += ext4_rec_len_from_disk(de->rec_len, 231 sb->s_blocksize); 232 } 233 offset = i; 234 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1)) 235 | offset; 236 file->f_version = inode_query_iversion(inode); 237 } 238 239 while (ctx->pos < inode->i_size 240 && offset < sb->s_blocksize) { 241 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset); 242 if (ext4_check_dir_entry(inode, file, de, bh, 243 bh->b_data, bh->b_size, 244 offset)) { 245 /* 246 * On error, skip to the next block 247 */ 248 ctx->pos = (ctx->pos | 249 (sb->s_blocksize - 1)) + 1; 250 break; 251 } 252 offset += ext4_rec_len_from_disk(de->rec_len, 253 sb->s_blocksize); 254 if (le32_to_cpu(de->inode)) { 255 if (!IS_ENCRYPTED(inode)) { 256 if (!dir_emit(ctx, de->name, 257 de->name_len, 258 le32_to_cpu(de->inode), 259 get_dtype(sb, de->file_type))) 260 goto done; 261 } else { 262 int save_len = fstr.len; 263 struct fscrypt_str de_name = 264 FSTR_INIT(de->name, 265 de->name_len); 266 267 /* Directory is encrypted */ 268 err = fscrypt_fname_disk_to_usr(inode, 269 0, 0, &de_name, &fstr); 270 de_name = fstr; 271 fstr.len = save_len; 272 if (err) 273 goto errout; 274 if (!dir_emit(ctx, 275 de_name.name, de_name.len, 276 le32_to_cpu(de->inode), 277 get_dtype(sb, de->file_type))) 278 goto done; 279 } 280 } 281 ctx->pos += ext4_rec_len_from_disk(de->rec_len, 282 sb->s_blocksize); 283 } 284 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode)) 285 goto done; 286 brelse(bh); 287 bh = NULL; 288 offset = 0; 289 } 290 done: 291 err = 0; 292 errout: 293 fscrypt_fname_free_buffer(&fstr); 294 brelse(bh); 295 return err; 296 } 297 298 static inline int is_32bit_api(void) 299 { 300 #ifdef CONFIG_COMPAT 301 return in_compat_syscall(); 302 #else 303 return (BITS_PER_LONG == 32); 304 #endif 305 } 306 307 /* 308 * These functions convert from the major/minor hash to an f_pos 309 * value for dx directories 310 * 311 * Upper layer (for example NFS) should specify FMODE_32BITHASH or 312 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted 313 * directly on both 32-bit and 64-bit nodes, under such case, neither 314 * FMODE_32BITHASH nor FMODE_64BITHASH is specified. 315 */ 316 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor) 317 { 318 if ((filp->f_mode & FMODE_32BITHASH) || 319 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 320 return major >> 1; 321 else 322 return ((__u64)(major >> 1) << 32) | (__u64)minor; 323 } 324 325 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos) 326 { 327 if ((filp->f_mode & FMODE_32BITHASH) || 328 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 329 return (pos << 1) & 0xffffffff; 330 else 331 return ((pos >> 32) << 1) & 0xffffffff; 332 } 333 334 static inline __u32 pos2min_hash(struct file *filp, loff_t pos) 335 { 336 if ((filp->f_mode & FMODE_32BITHASH) || 337 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 338 return 0; 339 else 340 return pos & 0xffffffff; 341 } 342 343 /* 344 * Return 32- or 64-bit end-of-file for dx directories 345 */ 346 static inline loff_t ext4_get_htree_eof(struct file *filp) 347 { 348 if ((filp->f_mode & FMODE_32BITHASH) || 349 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 350 return EXT4_HTREE_EOF_32BIT; 351 else 352 return EXT4_HTREE_EOF_64BIT; 353 } 354 355 356 /* 357 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree 358 * directories, where the "offset" is in terms of the filename hash 359 * value instead of the byte offset. 360 * 361 * Because we may return a 64-bit hash that is well beyond offset limits, 362 * we need to pass the max hash as the maximum allowable offset in 363 * the htree directory case. 364 * 365 * For non-htree, ext4_llseek already chooses the proper max offset. 366 */ 367 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence) 368 { 369 struct inode *inode = file->f_mapping->host; 370 int dx_dir = is_dx_dir(inode); 371 loff_t ret, htree_max = ext4_get_htree_eof(file); 372 373 if (likely(dx_dir)) 374 ret = generic_file_llseek_size(file, offset, whence, 375 htree_max, htree_max); 376 else 377 ret = ext4_llseek(file, offset, whence); 378 file->f_version = inode_peek_iversion(inode) - 1; 379 return ret; 380 } 381 382 /* 383 * This structure holds the nodes of the red-black tree used to store 384 * the directory entry in hash order. 385 */ 386 struct fname { 387 __u32 hash; 388 __u32 minor_hash; 389 struct rb_node rb_hash; 390 struct fname *next; 391 __u32 inode; 392 __u8 name_len; 393 __u8 file_type; 394 char name[0]; 395 }; 396 397 /* 398 * This functoin implements a non-recursive way of freeing all of the 399 * nodes in the red-black tree. 400 */ 401 static void free_rb_tree_fname(struct rb_root *root) 402 { 403 struct fname *fname, *next; 404 405 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash) 406 while (fname) { 407 struct fname *old = fname; 408 fname = fname->next; 409 kfree(old); 410 } 411 412 *root = RB_ROOT; 413 } 414 415 416 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp, 417 loff_t pos) 418 { 419 struct dir_private_info *p; 420 421 p = kzalloc(sizeof(*p), GFP_KERNEL); 422 if (!p) 423 return NULL; 424 p->curr_hash = pos2maj_hash(filp, pos); 425 p->curr_minor_hash = pos2min_hash(filp, pos); 426 return p; 427 } 428 429 void ext4_htree_free_dir_info(struct dir_private_info *p) 430 { 431 free_rb_tree_fname(&p->root); 432 kfree(p); 433 } 434 435 /* 436 * Given a directory entry, enter it into the fname rb tree. 437 * 438 * When filename encryption is enabled, the dirent will hold the 439 * encrypted filename, while the htree will hold decrypted filename. 440 * The decrypted filename is passed in via ent_name. parameter. 441 */ 442 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash, 443 __u32 minor_hash, 444 struct ext4_dir_entry_2 *dirent, 445 struct fscrypt_str *ent_name) 446 { 447 struct rb_node **p, *parent = NULL; 448 struct fname *fname, *new_fn; 449 struct dir_private_info *info; 450 int len; 451 452 info = dir_file->private_data; 453 p = &info->root.rb_node; 454 455 /* Create and allocate the fname structure */ 456 len = sizeof(struct fname) + ent_name->len + 1; 457 new_fn = kzalloc(len, GFP_KERNEL); 458 if (!new_fn) 459 return -ENOMEM; 460 new_fn->hash = hash; 461 new_fn->minor_hash = minor_hash; 462 new_fn->inode = le32_to_cpu(dirent->inode); 463 new_fn->name_len = ent_name->len; 464 new_fn->file_type = dirent->file_type; 465 memcpy(new_fn->name, ent_name->name, ent_name->len); 466 new_fn->name[ent_name->len] = 0; 467 468 while (*p) { 469 parent = *p; 470 fname = rb_entry(parent, struct fname, rb_hash); 471 472 /* 473 * If the hash and minor hash match up, then we put 474 * them on a linked list. This rarely happens... 475 */ 476 if ((new_fn->hash == fname->hash) && 477 (new_fn->minor_hash == fname->minor_hash)) { 478 new_fn->next = fname->next; 479 fname->next = new_fn; 480 return 0; 481 } 482 483 if (new_fn->hash < fname->hash) 484 p = &(*p)->rb_left; 485 else if (new_fn->hash > fname->hash) 486 p = &(*p)->rb_right; 487 else if (new_fn->minor_hash < fname->minor_hash) 488 p = &(*p)->rb_left; 489 else /* if (new_fn->minor_hash > fname->minor_hash) */ 490 p = &(*p)->rb_right; 491 } 492 493 rb_link_node(&new_fn->rb_hash, parent, p); 494 rb_insert_color(&new_fn->rb_hash, &info->root); 495 return 0; 496 } 497 498 499 500 /* 501 * This is a helper function for ext4_dx_readdir. It calls filldir 502 * for all entres on the fname linked list. (Normally there is only 503 * one entry on the linked list, unless there are 62 bit hash collisions.) 504 */ 505 static int call_filldir(struct file *file, struct dir_context *ctx, 506 struct fname *fname) 507 { 508 struct dir_private_info *info = file->private_data; 509 struct inode *inode = file_inode(file); 510 struct super_block *sb = inode->i_sb; 511 512 if (!fname) { 513 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: " 514 "called with null fname?!?", __func__, __LINE__, 515 inode->i_ino, current->comm); 516 return 0; 517 } 518 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash); 519 while (fname) { 520 if (!dir_emit(ctx, fname->name, 521 fname->name_len, 522 fname->inode, 523 get_dtype(sb, fname->file_type))) { 524 info->extra_fname = fname; 525 return 1; 526 } 527 fname = fname->next; 528 } 529 return 0; 530 } 531 532 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx) 533 { 534 struct dir_private_info *info = file->private_data; 535 struct inode *inode = file_inode(file); 536 struct fname *fname; 537 int ret; 538 539 if (!info) { 540 info = ext4_htree_create_dir_info(file, ctx->pos); 541 if (!info) 542 return -ENOMEM; 543 file->private_data = info; 544 } 545 546 if (ctx->pos == ext4_get_htree_eof(file)) 547 return 0; /* EOF */ 548 549 /* Some one has messed with f_pos; reset the world */ 550 if (info->last_pos != ctx->pos) { 551 free_rb_tree_fname(&info->root); 552 info->curr_node = NULL; 553 info->extra_fname = NULL; 554 info->curr_hash = pos2maj_hash(file, ctx->pos); 555 info->curr_minor_hash = pos2min_hash(file, ctx->pos); 556 } 557 558 /* 559 * If there are any leftover names on the hash collision 560 * chain, return them first. 561 */ 562 if (info->extra_fname) { 563 if (call_filldir(file, ctx, info->extra_fname)) 564 goto finished; 565 info->extra_fname = NULL; 566 goto next_node; 567 } else if (!info->curr_node) 568 info->curr_node = rb_first(&info->root); 569 570 while (1) { 571 /* 572 * Fill the rbtree if we have no more entries, 573 * or the inode has changed since we last read in the 574 * cached entries. 575 */ 576 if ((!info->curr_node) || 577 !inode_eq_iversion(inode, file->f_version)) { 578 info->curr_node = NULL; 579 free_rb_tree_fname(&info->root); 580 file->f_version = inode_query_iversion(inode); 581 ret = ext4_htree_fill_tree(file, info->curr_hash, 582 info->curr_minor_hash, 583 &info->next_hash); 584 if (ret < 0) 585 return ret; 586 if (ret == 0) { 587 ctx->pos = ext4_get_htree_eof(file); 588 break; 589 } 590 info->curr_node = rb_first(&info->root); 591 } 592 593 fname = rb_entry(info->curr_node, struct fname, rb_hash); 594 info->curr_hash = fname->hash; 595 info->curr_minor_hash = fname->minor_hash; 596 if (call_filldir(file, ctx, fname)) 597 break; 598 next_node: 599 info->curr_node = rb_next(info->curr_node); 600 if (info->curr_node) { 601 fname = rb_entry(info->curr_node, struct fname, 602 rb_hash); 603 info->curr_hash = fname->hash; 604 info->curr_minor_hash = fname->minor_hash; 605 } else { 606 if (info->next_hash == ~0) { 607 ctx->pos = ext4_get_htree_eof(file); 608 break; 609 } 610 info->curr_hash = info->next_hash; 611 info->curr_minor_hash = 0; 612 } 613 } 614 finished: 615 info->last_pos = ctx->pos; 616 return 0; 617 } 618 619 static int ext4_dir_open(struct inode * inode, struct file * filp) 620 { 621 if (IS_ENCRYPTED(inode)) 622 return fscrypt_get_encryption_info(inode) ? -EACCES : 0; 623 return 0; 624 } 625 626 static int ext4_release_dir(struct inode *inode, struct file *filp) 627 { 628 if (filp->private_data) 629 ext4_htree_free_dir_info(filp->private_data); 630 631 return 0; 632 } 633 634 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf, 635 int buf_size) 636 { 637 struct ext4_dir_entry_2 *de; 638 int rlen; 639 unsigned int offset = 0; 640 char *top; 641 642 de = (struct ext4_dir_entry_2 *)buf; 643 top = buf + buf_size; 644 while ((char *) de < top) { 645 if (ext4_check_dir_entry(dir, NULL, de, bh, 646 buf, buf_size, offset)) 647 return -EFSCORRUPTED; 648 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 649 de = (struct ext4_dir_entry_2 *)((char *)de + rlen); 650 offset += rlen; 651 } 652 if ((char *) de > top) 653 return -EFSCORRUPTED; 654 655 return 0; 656 } 657 658 const struct file_operations ext4_dir_operations = { 659 .llseek = ext4_dir_llseek, 660 .read = generic_read_dir, 661 .iterate_shared = ext4_readdir, 662 .unlocked_ioctl = ext4_ioctl, 663 #ifdef CONFIG_COMPAT 664 .compat_ioctl = ext4_compat_ioctl, 665 #endif 666 .fsync = ext4_sync_file, 667 .open = ext4_dir_open, 668 .release = ext4_release_dir, 669 }; 670 671 #ifdef CONFIG_UNICODE 672 static int ext4_d_compare(const struct dentry *dentry, unsigned int len, 673 const char *str, const struct qstr *name) 674 { 675 struct qstr qstr = {.name = str, .len = len }; 676 struct inode *inode = dentry->d_parent->d_inode; 677 678 if (!IS_CASEFOLDED(inode) || !EXT4_SB(inode->i_sb)->s_encoding) { 679 if (len != name->len) 680 return -1; 681 return memcmp(str, name->name, len); 682 } 683 684 return ext4_ci_compare(inode, name, &qstr, false); 685 } 686 687 static int ext4_d_hash(const struct dentry *dentry, struct qstr *str) 688 { 689 const struct ext4_sb_info *sbi = EXT4_SB(dentry->d_sb); 690 const struct unicode_map *um = sbi->s_encoding; 691 unsigned char *norm; 692 int len, ret = 0; 693 694 if (!IS_CASEFOLDED(dentry->d_inode) || !um) 695 return 0; 696 697 norm = kmalloc(PATH_MAX, GFP_ATOMIC); 698 if (!norm) 699 return -ENOMEM; 700 701 len = utf8_casefold(um, str, norm, PATH_MAX); 702 if (len < 0) { 703 if (ext4_has_strict_mode(sbi)) 704 ret = -EINVAL; 705 goto out; 706 } 707 str->hash = full_name_hash(dentry, norm, len); 708 out: 709 kfree(norm); 710 return ret; 711 } 712 713 const struct dentry_operations ext4_dentry_ops = { 714 .d_hash = ext4_d_hash, 715 .d_compare = ext4_d_compare, 716 }; 717 #endif 718