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 const int next_offset = ((char *) de - buf) + rlen; 76 77 if (unlikely(rlen < EXT4_DIR_REC_LEN(1))) 78 error_msg = "rec_len is smaller than minimal"; 79 else if (unlikely(rlen % 4 != 0)) 80 error_msg = "rec_len % 4 != 0"; 81 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len))) 82 error_msg = "rec_len is too small for name_len"; 83 else if (unlikely(next_offset > size)) 84 error_msg = "directory entry overrun"; 85 else if (unlikely(next_offset > size - EXT4_DIR_REC_LEN(1) && 86 next_offset != size)) 87 error_msg = "directory entry too close to block end"; 88 else if (unlikely(le32_to_cpu(de->inode) > 89 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))) 90 error_msg = "inode out of bounds"; 91 else 92 return 0; 93 94 if (filp) 95 ext4_error_file(filp, function, line, bh->b_blocknr, 96 "bad entry in directory: %s - offset=%u, " 97 "inode=%u, rec_len=%d, name_len=%d, size=%d", 98 error_msg, offset, le32_to_cpu(de->inode), 99 rlen, de->name_len, size); 100 else 101 ext4_error_inode(dir, function, line, bh->b_blocknr, 102 "bad entry in directory: %s - offset=%u, " 103 "inode=%u, rec_len=%d, name_len=%d, size=%d", 104 error_msg, offset, le32_to_cpu(de->inode), 105 rlen, de->name_len, size); 106 107 return 1; 108 } 109 110 static int ext4_readdir(struct file *file, struct dir_context *ctx) 111 { 112 unsigned int offset; 113 int i; 114 struct ext4_dir_entry_2 *de; 115 int err; 116 struct inode *inode = file_inode(file); 117 struct super_block *sb = inode->i_sb; 118 struct buffer_head *bh = NULL; 119 struct fscrypt_str fstr = FSTR_INIT(NULL, 0); 120 121 err = fscrypt_prepare_readdir(inode); 122 if (err) 123 return err; 124 125 if (is_dx_dir(inode)) { 126 err = ext4_dx_readdir(file, ctx); 127 if (err != ERR_BAD_DX_DIR) { 128 return err; 129 } 130 /* Can we just clear INDEX flag to ignore htree information? */ 131 if (!ext4_has_metadata_csum(sb)) { 132 /* 133 * We don't set the inode dirty flag since it's not 134 * critical that it gets flushed back to the disk. 135 */ 136 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX); 137 } 138 } 139 140 if (ext4_has_inline_data(inode)) { 141 int has_inline_data = 1; 142 err = ext4_read_inline_dir(file, ctx, 143 &has_inline_data); 144 if (has_inline_data) 145 return err; 146 } 147 148 if (IS_ENCRYPTED(inode)) { 149 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN, &fstr); 150 if (err < 0) 151 return err; 152 } 153 154 while (ctx->pos < inode->i_size) { 155 struct ext4_map_blocks map; 156 157 if (fatal_signal_pending(current)) { 158 err = -ERESTARTSYS; 159 goto errout; 160 } 161 cond_resched(); 162 offset = ctx->pos & (sb->s_blocksize - 1); 163 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb); 164 map.m_len = 1; 165 err = ext4_map_blocks(NULL, inode, &map, 0); 166 if (err == 0) { 167 /* m_len should never be zero but let's avoid 168 * an infinite loop if it somehow is */ 169 if (map.m_len == 0) 170 map.m_len = 1; 171 ctx->pos += map.m_len * sb->s_blocksize; 172 continue; 173 } 174 if (err > 0) { 175 pgoff_t index = map.m_pblk >> 176 (PAGE_SHIFT - inode->i_blkbits); 177 if (!ra_has_index(&file->f_ra, index)) 178 page_cache_sync_readahead( 179 sb->s_bdev->bd_inode->i_mapping, 180 &file->f_ra, file, 181 index, 1); 182 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT; 183 bh = ext4_bread(NULL, inode, map.m_lblk, 0); 184 if (IS_ERR(bh)) { 185 err = PTR_ERR(bh); 186 bh = NULL; 187 goto errout; 188 } 189 } 190 191 if (!bh) { 192 /* corrupt size? Maybe no more blocks to read */ 193 if (ctx->pos > inode->i_blocks << 9) 194 break; 195 ctx->pos += sb->s_blocksize - offset; 196 continue; 197 } 198 199 /* Check the checksum */ 200 if (!buffer_verified(bh) && 201 !ext4_dirblock_csum_verify(inode, bh)) { 202 EXT4_ERROR_FILE(file, 0, "directory fails checksum " 203 "at offset %llu", 204 (unsigned long long)ctx->pos); 205 ctx->pos += sb->s_blocksize - offset; 206 brelse(bh); 207 bh = NULL; 208 continue; 209 } 210 set_buffer_verified(bh); 211 212 /* If the dir block has changed since the last call to 213 * readdir(2), then we might be pointing to an invalid 214 * dirent right now. Scan from the start of the block 215 * to make sure. */ 216 if (!inode_eq_iversion(inode, file->f_version)) { 217 for (i = 0; i < sb->s_blocksize && i < offset; ) { 218 de = (struct ext4_dir_entry_2 *) 219 (bh->b_data + i); 220 /* It's too expensive to do a full 221 * dirent test each time round this 222 * loop, but we do have to test at 223 * least that it is non-zero. A 224 * failure will be detected in the 225 * dirent test below. */ 226 if (ext4_rec_len_from_disk(de->rec_len, 227 sb->s_blocksize) < EXT4_DIR_REC_LEN(1)) 228 break; 229 i += ext4_rec_len_from_disk(de->rec_len, 230 sb->s_blocksize); 231 } 232 offset = i; 233 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1)) 234 | offset; 235 file->f_version = inode_query_iversion(inode); 236 } 237 238 while (ctx->pos < inode->i_size 239 && offset < sb->s_blocksize) { 240 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset); 241 if (ext4_check_dir_entry(inode, file, de, bh, 242 bh->b_data, bh->b_size, 243 offset)) { 244 /* 245 * On error, skip to the next block 246 */ 247 ctx->pos = (ctx->pos | 248 (sb->s_blocksize - 1)) + 1; 249 break; 250 } 251 offset += ext4_rec_len_from_disk(de->rec_len, 252 sb->s_blocksize); 253 if (le32_to_cpu(de->inode)) { 254 if (!IS_ENCRYPTED(inode)) { 255 if (!dir_emit(ctx, de->name, 256 de->name_len, 257 le32_to_cpu(de->inode), 258 get_dtype(sb, de->file_type))) 259 goto done; 260 } else { 261 int save_len = fstr.len; 262 struct fscrypt_str de_name = 263 FSTR_INIT(de->name, 264 de->name_len); 265 266 /* Directory is encrypted */ 267 err = fscrypt_fname_disk_to_usr(inode, 268 0, 0, &de_name, &fstr); 269 de_name = fstr; 270 fstr.len = save_len; 271 if (err) 272 goto errout; 273 if (!dir_emit(ctx, 274 de_name.name, de_name.len, 275 le32_to_cpu(de->inode), 276 get_dtype(sb, de->file_type))) 277 goto done; 278 } 279 } 280 ctx->pos += ext4_rec_len_from_disk(de->rec_len, 281 sb->s_blocksize); 282 } 283 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode)) 284 goto done; 285 brelse(bh); 286 bh = NULL; 287 offset = 0; 288 } 289 done: 290 err = 0; 291 errout: 292 fscrypt_fname_free_buffer(&fstr); 293 brelse(bh); 294 return err; 295 } 296 297 static inline int is_32bit_api(void) 298 { 299 #ifdef CONFIG_COMPAT 300 return in_compat_syscall(); 301 #else 302 return (BITS_PER_LONG == 32); 303 #endif 304 } 305 306 /* 307 * These functions convert from the major/minor hash to an f_pos 308 * value for dx directories 309 * 310 * Upper layer (for example NFS) should specify FMODE_32BITHASH or 311 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted 312 * directly on both 32-bit and 64-bit nodes, under such case, neither 313 * FMODE_32BITHASH nor FMODE_64BITHASH is specified. 314 */ 315 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor) 316 { 317 if ((filp->f_mode & FMODE_32BITHASH) || 318 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 319 return major >> 1; 320 else 321 return ((__u64)(major >> 1) << 32) | (__u64)minor; 322 } 323 324 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos) 325 { 326 if ((filp->f_mode & FMODE_32BITHASH) || 327 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 328 return (pos << 1) & 0xffffffff; 329 else 330 return ((pos >> 32) << 1) & 0xffffffff; 331 } 332 333 static inline __u32 pos2min_hash(struct file *filp, loff_t pos) 334 { 335 if ((filp->f_mode & FMODE_32BITHASH) || 336 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 337 return 0; 338 else 339 return pos & 0xffffffff; 340 } 341 342 /* 343 * Return 32- or 64-bit end-of-file for dx directories 344 */ 345 static inline loff_t ext4_get_htree_eof(struct file *filp) 346 { 347 if ((filp->f_mode & FMODE_32BITHASH) || 348 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 349 return EXT4_HTREE_EOF_32BIT; 350 else 351 return EXT4_HTREE_EOF_64BIT; 352 } 353 354 355 /* 356 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree 357 * directories, where the "offset" is in terms of the filename hash 358 * value instead of the byte offset. 359 * 360 * Because we may return a 64-bit hash that is well beyond offset limits, 361 * we need to pass the max hash as the maximum allowable offset in 362 * the htree directory case. 363 * 364 * For non-htree, ext4_llseek already chooses the proper max offset. 365 */ 366 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence) 367 { 368 struct inode *inode = file->f_mapping->host; 369 int dx_dir = is_dx_dir(inode); 370 loff_t ret, htree_max = ext4_get_htree_eof(file); 371 372 if (likely(dx_dir)) 373 ret = generic_file_llseek_size(file, offset, whence, 374 htree_max, htree_max); 375 else 376 ret = ext4_llseek(file, offset, whence); 377 file->f_version = inode_peek_iversion(inode) - 1; 378 return ret; 379 } 380 381 /* 382 * This structure holds the nodes of the red-black tree used to store 383 * the directory entry in hash order. 384 */ 385 struct fname { 386 __u32 hash; 387 __u32 minor_hash; 388 struct rb_node rb_hash; 389 struct fname *next; 390 __u32 inode; 391 __u8 name_len; 392 __u8 file_type; 393 char name[]; 394 }; 395 396 /* 397 * This functoin implements a non-recursive way of freeing all of the 398 * nodes in the red-black tree. 399 */ 400 static void free_rb_tree_fname(struct rb_root *root) 401 { 402 struct fname *fname, *next; 403 404 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash) 405 while (fname) { 406 struct fname *old = fname; 407 fname = fname->next; 408 kfree(old); 409 } 410 411 *root = RB_ROOT; 412 } 413 414 415 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp, 416 loff_t pos) 417 { 418 struct dir_private_info *p; 419 420 p = kzalloc(sizeof(*p), GFP_KERNEL); 421 if (!p) 422 return NULL; 423 p->curr_hash = pos2maj_hash(filp, pos); 424 p->curr_minor_hash = pos2min_hash(filp, pos); 425 return p; 426 } 427 428 void ext4_htree_free_dir_info(struct dir_private_info *p) 429 { 430 free_rb_tree_fname(&p->root); 431 kfree(p); 432 } 433 434 /* 435 * Given a directory entry, enter it into the fname rb tree. 436 * 437 * When filename encryption is enabled, the dirent will hold the 438 * encrypted filename, while the htree will hold decrypted filename. 439 * The decrypted filename is passed in via ent_name. parameter. 440 */ 441 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash, 442 __u32 minor_hash, 443 struct ext4_dir_entry_2 *dirent, 444 struct fscrypt_str *ent_name) 445 { 446 struct rb_node **p, *parent = NULL; 447 struct fname *fname, *new_fn; 448 struct dir_private_info *info; 449 int len; 450 451 info = dir_file->private_data; 452 p = &info->root.rb_node; 453 454 /* Create and allocate the fname structure */ 455 len = sizeof(struct fname) + ent_name->len + 1; 456 new_fn = kzalloc(len, GFP_KERNEL); 457 if (!new_fn) 458 return -ENOMEM; 459 new_fn->hash = hash; 460 new_fn->minor_hash = minor_hash; 461 new_fn->inode = le32_to_cpu(dirent->inode); 462 new_fn->name_len = ent_name->len; 463 new_fn->file_type = dirent->file_type; 464 memcpy(new_fn->name, ent_name->name, ent_name->len); 465 466 while (*p) { 467 parent = *p; 468 fname = rb_entry(parent, struct fname, rb_hash); 469 470 /* 471 * If the hash and minor hash match up, then we put 472 * them on a linked list. This rarely happens... 473 */ 474 if ((new_fn->hash == fname->hash) && 475 (new_fn->minor_hash == fname->minor_hash)) { 476 new_fn->next = fname->next; 477 fname->next = new_fn; 478 return 0; 479 } 480 481 if (new_fn->hash < fname->hash) 482 p = &(*p)->rb_left; 483 else if (new_fn->hash > fname->hash) 484 p = &(*p)->rb_right; 485 else if (new_fn->minor_hash < fname->minor_hash) 486 p = &(*p)->rb_left; 487 else /* if (new_fn->minor_hash > fname->minor_hash) */ 488 p = &(*p)->rb_right; 489 } 490 491 rb_link_node(&new_fn->rb_hash, parent, p); 492 rb_insert_color(&new_fn->rb_hash, &info->root); 493 return 0; 494 } 495 496 497 498 /* 499 * This is a helper function for ext4_dx_readdir. It calls filldir 500 * for all entres on the fname linked list. (Normally there is only 501 * one entry on the linked list, unless there are 62 bit hash collisions.) 502 */ 503 static int call_filldir(struct file *file, struct dir_context *ctx, 504 struct fname *fname) 505 { 506 struct dir_private_info *info = file->private_data; 507 struct inode *inode = file_inode(file); 508 struct super_block *sb = inode->i_sb; 509 510 if (!fname) { 511 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: " 512 "called with null fname?!?", __func__, __LINE__, 513 inode->i_ino, current->comm); 514 return 0; 515 } 516 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash); 517 while (fname) { 518 if (!dir_emit(ctx, fname->name, 519 fname->name_len, 520 fname->inode, 521 get_dtype(sb, fname->file_type))) { 522 info->extra_fname = fname; 523 return 1; 524 } 525 fname = fname->next; 526 } 527 return 0; 528 } 529 530 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx) 531 { 532 struct dir_private_info *info = file->private_data; 533 struct inode *inode = file_inode(file); 534 struct fname *fname; 535 int ret; 536 537 if (!info) { 538 info = ext4_htree_create_dir_info(file, ctx->pos); 539 if (!info) 540 return -ENOMEM; 541 file->private_data = info; 542 } 543 544 if (ctx->pos == ext4_get_htree_eof(file)) 545 return 0; /* EOF */ 546 547 /* Some one has messed with f_pos; reset the world */ 548 if (info->last_pos != ctx->pos) { 549 free_rb_tree_fname(&info->root); 550 info->curr_node = NULL; 551 info->extra_fname = NULL; 552 info->curr_hash = pos2maj_hash(file, ctx->pos); 553 info->curr_minor_hash = pos2min_hash(file, ctx->pos); 554 } 555 556 /* 557 * If there are any leftover names on the hash collision 558 * chain, return them first. 559 */ 560 if (info->extra_fname) { 561 if (call_filldir(file, ctx, info->extra_fname)) 562 goto finished; 563 info->extra_fname = NULL; 564 goto next_node; 565 } else if (!info->curr_node) 566 info->curr_node = rb_first(&info->root); 567 568 while (1) { 569 /* 570 * Fill the rbtree if we have no more entries, 571 * or the inode has changed since we last read in the 572 * cached entries. 573 */ 574 if ((!info->curr_node) || 575 !inode_eq_iversion(inode, file->f_version)) { 576 info->curr_node = NULL; 577 free_rb_tree_fname(&info->root); 578 file->f_version = inode_query_iversion(inode); 579 ret = ext4_htree_fill_tree(file, info->curr_hash, 580 info->curr_minor_hash, 581 &info->next_hash); 582 if (ret < 0) 583 return ret; 584 if (ret == 0) { 585 ctx->pos = ext4_get_htree_eof(file); 586 break; 587 } 588 info->curr_node = rb_first(&info->root); 589 } 590 591 fname = rb_entry(info->curr_node, struct fname, rb_hash); 592 info->curr_hash = fname->hash; 593 info->curr_minor_hash = fname->minor_hash; 594 if (call_filldir(file, ctx, fname)) 595 break; 596 next_node: 597 info->curr_node = rb_next(info->curr_node); 598 if (info->curr_node) { 599 fname = rb_entry(info->curr_node, struct fname, 600 rb_hash); 601 info->curr_hash = fname->hash; 602 info->curr_minor_hash = fname->minor_hash; 603 } else { 604 if (info->next_hash == ~0) { 605 ctx->pos = ext4_get_htree_eof(file); 606 break; 607 } 608 info->curr_hash = info->next_hash; 609 info->curr_minor_hash = 0; 610 } 611 } 612 finished: 613 info->last_pos = ctx->pos; 614 return 0; 615 } 616 617 static int ext4_release_dir(struct inode *inode, struct file *filp) 618 { 619 if (filp->private_data) 620 ext4_htree_free_dir_info(filp->private_data); 621 622 return 0; 623 } 624 625 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf, 626 int buf_size) 627 { 628 struct ext4_dir_entry_2 *de; 629 int rlen; 630 unsigned int offset = 0; 631 char *top; 632 633 de = (struct ext4_dir_entry_2 *)buf; 634 top = buf + buf_size; 635 while ((char *) de < top) { 636 if (ext4_check_dir_entry(dir, NULL, de, bh, 637 buf, buf_size, offset)) 638 return -EFSCORRUPTED; 639 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 640 de = (struct ext4_dir_entry_2 *)((char *)de + rlen); 641 offset += rlen; 642 } 643 if ((char *) de > top) 644 return -EFSCORRUPTED; 645 646 return 0; 647 } 648 649 const struct file_operations ext4_dir_operations = { 650 .llseek = ext4_dir_llseek, 651 .read = generic_read_dir, 652 .iterate_shared = ext4_readdir, 653 .unlocked_ioctl = ext4_ioctl, 654 #ifdef CONFIG_COMPAT 655 .compat_ioctl = ext4_compat_ioctl, 656 #endif 657 .fsync = ext4_sync_file, 658 .release = ext4_release_dir, 659 }; 660