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 "ext4.h" 30 #include "xattr.h" 31 32 static int ext4_dx_readdir(struct file *, struct dir_context *); 33 34 /** 35 * Check if the given dir-inode refers to an htree-indexed directory 36 * (or a directory which could potentially get converted to use htree 37 * indexing). 38 * 39 * Return 1 if it is a dx dir, 0 if not 40 */ 41 static int is_dx_dir(struct inode *inode) 42 { 43 struct super_block *sb = inode->i_sb; 44 45 if (ext4_has_feature_dir_index(inode->i_sb) && 46 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) || 47 ((inode->i_size >> sb->s_blocksize_bits) == 1) || 48 ext4_has_inline_data(inode))) 49 return 1; 50 51 return 0; 52 } 53 54 /* 55 * Return 0 if the directory entry is OK, and 1 if there is a problem 56 * 57 * Note: this is the opposite of what ext2 and ext3 historically returned... 58 * 59 * bh passed here can be an inode block or a dir data block, depending 60 * on the inode inline data flag. 61 */ 62 int __ext4_check_dir_entry(const char *function, unsigned int line, 63 struct inode *dir, struct file *filp, 64 struct ext4_dir_entry_2 *de, 65 struct buffer_head *bh, char *buf, int size, 66 unsigned int offset) 67 { 68 const char *error_msg = NULL; 69 const int rlen = ext4_rec_len_from_disk(de->rec_len, 70 dir->i_sb->s_blocksize); 71 72 if (unlikely(rlen < EXT4_DIR_REC_LEN(1))) 73 error_msg = "rec_len is smaller than minimal"; 74 else if (unlikely(rlen % 4 != 0)) 75 error_msg = "rec_len % 4 != 0"; 76 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len))) 77 error_msg = "rec_len is too small for name_len"; 78 else if (unlikely(((char *) de - buf) + rlen > size)) 79 error_msg = "directory entry overrun"; 80 else if (unlikely(le32_to_cpu(de->inode) > 81 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))) 82 error_msg = "inode out of bounds"; 83 else 84 return 0; 85 86 if (filp) 87 ext4_error_file(filp, function, line, bh->b_blocknr, 88 "bad entry in directory: %s - offset=%u, " 89 "inode=%u, rec_len=%d, name_len=%d, size=%d", 90 error_msg, offset, le32_to_cpu(de->inode), 91 rlen, de->name_len, size); 92 else 93 ext4_error_inode(dir, function, line, bh->b_blocknr, 94 "bad entry in directory: %s - offset=%u, " 95 "inode=%u, rec_len=%d, name_len=%d, size=%d", 96 error_msg, offset, le32_to_cpu(de->inode), 97 rlen, de->name_len, size); 98 99 return 1; 100 } 101 102 static int ext4_readdir(struct file *file, struct dir_context *ctx) 103 { 104 unsigned int offset; 105 int i; 106 struct ext4_dir_entry_2 *de; 107 int err; 108 struct inode *inode = file_inode(file); 109 struct super_block *sb = inode->i_sb; 110 struct buffer_head *bh = NULL; 111 int dir_has_error = 0; 112 struct fscrypt_str fstr = FSTR_INIT(NULL, 0); 113 114 if (ext4_encrypted_inode(inode)) { 115 err = fscrypt_get_encryption_info(inode); 116 if (err && err != -ENOKEY) 117 return err; 118 } 119 120 if (is_dx_dir(inode)) { 121 err = ext4_dx_readdir(file, ctx); 122 if (err != ERR_BAD_DX_DIR) { 123 return err; 124 } 125 /* 126 * We don't set the inode dirty flag since it's not 127 * critical that it get flushed back to the disk. 128 */ 129 ext4_clear_inode_flag(file_inode(file), 130 EXT4_INODE_INDEX); 131 } 132 133 if (ext4_has_inline_data(inode)) { 134 int has_inline_data = 1; 135 err = ext4_read_inline_dir(file, ctx, 136 &has_inline_data); 137 if (has_inline_data) 138 return err; 139 } 140 141 if (ext4_encrypted_inode(inode)) { 142 err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr); 143 if (err < 0) 144 return err; 145 } 146 147 offset = ctx->pos & (sb->s_blocksize - 1); 148 149 while (ctx->pos < inode->i_size) { 150 struct ext4_map_blocks map; 151 152 if (fatal_signal_pending(current)) { 153 err = -ERESTARTSYS; 154 goto errout; 155 } 156 cond_resched(); 157 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb); 158 map.m_len = 1; 159 err = ext4_map_blocks(NULL, inode, &map, 0); 160 if (err > 0) { 161 pgoff_t index = map.m_pblk >> 162 (PAGE_SHIFT - inode->i_blkbits); 163 if (!ra_has_index(&file->f_ra, index)) 164 page_cache_sync_readahead( 165 sb->s_bdev->bd_inode->i_mapping, 166 &file->f_ra, file, 167 index, 1); 168 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT; 169 bh = ext4_bread(NULL, inode, map.m_lblk, 0); 170 if (IS_ERR(bh)) { 171 err = PTR_ERR(bh); 172 bh = NULL; 173 goto errout; 174 } 175 } 176 177 if (!bh) { 178 if (!dir_has_error) { 179 EXT4_ERROR_FILE(file, 0, 180 "directory contains a " 181 "hole at offset %llu", 182 (unsigned long long) ctx->pos); 183 dir_has_error = 1; 184 } 185 /* corrupt size? Maybe no more blocks to read */ 186 if (ctx->pos > inode->i_blocks << 9) 187 break; 188 ctx->pos += sb->s_blocksize - offset; 189 continue; 190 } 191 192 /* Check the checksum */ 193 if (!buffer_verified(bh) && 194 !ext4_dirent_csum_verify(inode, 195 (struct ext4_dir_entry *)bh->b_data)) { 196 EXT4_ERROR_FILE(file, 0, "directory fails checksum " 197 "at offset %llu", 198 (unsigned long long)ctx->pos); 199 ctx->pos += sb->s_blocksize - offset; 200 brelse(bh); 201 bh = NULL; 202 continue; 203 } 204 set_buffer_verified(bh); 205 206 /* If the dir block has changed since the last call to 207 * readdir(2), then we might be pointing to an invalid 208 * dirent right now. Scan from the start of the block 209 * to make sure. */ 210 if (!inode_eq_iversion(inode, file->f_version)) { 211 for (i = 0; i < sb->s_blocksize && i < offset; ) { 212 de = (struct ext4_dir_entry_2 *) 213 (bh->b_data + i); 214 /* It's too expensive to do a full 215 * dirent test each time round this 216 * loop, but we do have to test at 217 * least that it is non-zero. A 218 * failure will be detected in the 219 * dirent test below. */ 220 if (ext4_rec_len_from_disk(de->rec_len, 221 sb->s_blocksize) < EXT4_DIR_REC_LEN(1)) 222 break; 223 i += ext4_rec_len_from_disk(de->rec_len, 224 sb->s_blocksize); 225 } 226 offset = i; 227 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1)) 228 | offset; 229 file->f_version = inode_query_iversion(inode); 230 } 231 232 while (ctx->pos < inode->i_size 233 && offset < sb->s_blocksize) { 234 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset); 235 if (ext4_check_dir_entry(inode, file, de, bh, 236 bh->b_data, bh->b_size, 237 offset)) { 238 /* 239 * On error, skip to the next block 240 */ 241 ctx->pos = (ctx->pos | 242 (sb->s_blocksize - 1)) + 1; 243 break; 244 } 245 offset += ext4_rec_len_from_disk(de->rec_len, 246 sb->s_blocksize); 247 if (le32_to_cpu(de->inode)) { 248 if (!ext4_encrypted_inode(inode)) { 249 if (!dir_emit(ctx, de->name, 250 de->name_len, 251 le32_to_cpu(de->inode), 252 get_dtype(sb, de->file_type))) 253 goto done; 254 } else { 255 int save_len = fstr.len; 256 struct fscrypt_str de_name = 257 FSTR_INIT(de->name, 258 de->name_len); 259 260 /* Directory is encrypted */ 261 err = fscrypt_fname_disk_to_usr(inode, 262 0, 0, &de_name, &fstr); 263 de_name = fstr; 264 fstr.len = save_len; 265 if (err) 266 goto errout; 267 if (!dir_emit(ctx, 268 de_name.name, de_name.len, 269 le32_to_cpu(de->inode), 270 get_dtype(sb, de->file_type))) 271 goto done; 272 } 273 } 274 ctx->pos += ext4_rec_len_from_disk(de->rec_len, 275 sb->s_blocksize); 276 } 277 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode)) 278 goto done; 279 brelse(bh); 280 bh = NULL; 281 offset = 0; 282 } 283 done: 284 err = 0; 285 errout: 286 #ifdef CONFIG_EXT4_FS_ENCRYPTION 287 fscrypt_fname_free_buffer(&fstr); 288 #endif 289 brelse(bh); 290 return err; 291 } 292 293 static inline int is_32bit_api(void) 294 { 295 #ifdef CONFIG_COMPAT 296 return in_compat_syscall(); 297 #else 298 return (BITS_PER_LONG == 32); 299 #endif 300 } 301 302 /* 303 * These functions convert from the major/minor hash to an f_pos 304 * value for dx directories 305 * 306 * Upper layer (for example NFS) should specify FMODE_32BITHASH or 307 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted 308 * directly on both 32-bit and 64-bit nodes, under such case, neither 309 * FMODE_32BITHASH nor FMODE_64BITHASH is specified. 310 */ 311 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor) 312 { 313 if ((filp->f_mode & FMODE_32BITHASH) || 314 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 315 return major >> 1; 316 else 317 return ((__u64)(major >> 1) << 32) | (__u64)minor; 318 } 319 320 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos) 321 { 322 if ((filp->f_mode & FMODE_32BITHASH) || 323 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 324 return (pos << 1) & 0xffffffff; 325 else 326 return ((pos >> 32) << 1) & 0xffffffff; 327 } 328 329 static inline __u32 pos2min_hash(struct file *filp, loff_t pos) 330 { 331 if ((filp->f_mode & FMODE_32BITHASH) || 332 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 333 return 0; 334 else 335 return pos & 0xffffffff; 336 } 337 338 /* 339 * Return 32- or 64-bit end-of-file for dx directories 340 */ 341 static inline loff_t ext4_get_htree_eof(struct file *filp) 342 { 343 if ((filp->f_mode & FMODE_32BITHASH) || 344 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 345 return EXT4_HTREE_EOF_32BIT; 346 else 347 return EXT4_HTREE_EOF_64BIT; 348 } 349 350 351 /* 352 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree 353 * directories, where the "offset" is in terms of the filename hash 354 * value instead of the byte offset. 355 * 356 * Because we may return a 64-bit hash that is well beyond offset limits, 357 * we need to pass the max hash as the maximum allowable offset in 358 * the htree directory case. 359 * 360 * For non-htree, ext4_llseek already chooses the proper max offset. 361 */ 362 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence) 363 { 364 struct inode *inode = file->f_mapping->host; 365 int dx_dir = is_dx_dir(inode); 366 loff_t ret, htree_max = ext4_get_htree_eof(file); 367 368 if (likely(dx_dir)) 369 ret = generic_file_llseek_size(file, offset, whence, 370 htree_max, htree_max); 371 else 372 ret = ext4_llseek(file, offset, whence); 373 file->f_version = inode_peek_iversion(inode) - 1; 374 return ret; 375 } 376 377 /* 378 * This structure holds the nodes of the red-black tree used to store 379 * the directory entry in hash order. 380 */ 381 struct fname { 382 __u32 hash; 383 __u32 minor_hash; 384 struct rb_node rb_hash; 385 struct fname *next; 386 __u32 inode; 387 __u8 name_len; 388 __u8 file_type; 389 char name[0]; 390 }; 391 392 /* 393 * This functoin implements a non-recursive way of freeing all of the 394 * nodes in the red-black tree. 395 */ 396 static void free_rb_tree_fname(struct rb_root *root) 397 { 398 struct fname *fname, *next; 399 400 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash) 401 while (fname) { 402 struct fname *old = fname; 403 fname = fname->next; 404 kfree(old); 405 } 406 407 *root = RB_ROOT; 408 } 409 410 411 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp, 412 loff_t pos) 413 { 414 struct dir_private_info *p; 415 416 p = kzalloc(sizeof(*p), GFP_KERNEL); 417 if (!p) 418 return NULL; 419 p->curr_hash = pos2maj_hash(filp, pos); 420 p->curr_minor_hash = pos2min_hash(filp, pos); 421 return p; 422 } 423 424 void ext4_htree_free_dir_info(struct dir_private_info *p) 425 { 426 free_rb_tree_fname(&p->root); 427 kfree(p); 428 } 429 430 /* 431 * Given a directory entry, enter it into the fname rb tree. 432 * 433 * When filename encryption is enabled, the dirent will hold the 434 * encrypted filename, while the htree will hold decrypted filename. 435 * The decrypted filename is passed in via ent_name. parameter. 436 */ 437 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash, 438 __u32 minor_hash, 439 struct ext4_dir_entry_2 *dirent, 440 struct fscrypt_str *ent_name) 441 { 442 struct rb_node **p, *parent = NULL; 443 struct fname *fname, *new_fn; 444 struct dir_private_info *info; 445 int len; 446 447 info = dir_file->private_data; 448 p = &info->root.rb_node; 449 450 /* Create and allocate the fname structure */ 451 len = sizeof(struct fname) + ent_name->len + 1; 452 new_fn = kzalloc(len, GFP_KERNEL); 453 if (!new_fn) 454 return -ENOMEM; 455 new_fn->hash = hash; 456 new_fn->minor_hash = minor_hash; 457 new_fn->inode = le32_to_cpu(dirent->inode); 458 new_fn->name_len = ent_name->len; 459 new_fn->file_type = dirent->file_type; 460 memcpy(new_fn->name, ent_name->name, ent_name->len); 461 new_fn->name[ent_name->len] = 0; 462 463 while (*p) { 464 parent = *p; 465 fname = rb_entry(parent, struct fname, rb_hash); 466 467 /* 468 * If the hash and minor hash match up, then we put 469 * them on a linked list. This rarely happens... 470 */ 471 if ((new_fn->hash == fname->hash) && 472 (new_fn->minor_hash == fname->minor_hash)) { 473 new_fn->next = fname->next; 474 fname->next = new_fn; 475 return 0; 476 } 477 478 if (new_fn->hash < fname->hash) 479 p = &(*p)->rb_left; 480 else if (new_fn->hash > fname->hash) 481 p = &(*p)->rb_right; 482 else if (new_fn->minor_hash < fname->minor_hash) 483 p = &(*p)->rb_left; 484 else /* if (new_fn->minor_hash > fname->minor_hash) */ 485 p = &(*p)->rb_right; 486 } 487 488 rb_link_node(&new_fn->rb_hash, parent, p); 489 rb_insert_color(&new_fn->rb_hash, &info->root); 490 return 0; 491 } 492 493 494 495 /* 496 * This is a helper function for ext4_dx_readdir. It calls filldir 497 * for all entres on the fname linked list. (Normally there is only 498 * one entry on the linked list, unless there are 62 bit hash collisions.) 499 */ 500 static int call_filldir(struct file *file, struct dir_context *ctx, 501 struct fname *fname) 502 { 503 struct dir_private_info *info = file->private_data; 504 struct inode *inode = file_inode(file); 505 struct super_block *sb = inode->i_sb; 506 507 if (!fname) { 508 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: " 509 "called with null fname?!?", __func__, __LINE__, 510 inode->i_ino, current->comm); 511 return 0; 512 } 513 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash); 514 while (fname) { 515 if (!dir_emit(ctx, fname->name, 516 fname->name_len, 517 fname->inode, 518 get_dtype(sb, fname->file_type))) { 519 info->extra_fname = fname; 520 return 1; 521 } 522 fname = fname->next; 523 } 524 return 0; 525 } 526 527 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx) 528 { 529 struct dir_private_info *info = file->private_data; 530 struct inode *inode = file_inode(file); 531 struct fname *fname; 532 int ret; 533 534 if (!info) { 535 info = ext4_htree_create_dir_info(file, ctx->pos); 536 if (!info) 537 return -ENOMEM; 538 file->private_data = info; 539 } 540 541 if (ctx->pos == ext4_get_htree_eof(file)) 542 return 0; /* EOF */ 543 544 /* Some one has messed with f_pos; reset the world */ 545 if (info->last_pos != ctx->pos) { 546 free_rb_tree_fname(&info->root); 547 info->curr_node = NULL; 548 info->extra_fname = NULL; 549 info->curr_hash = pos2maj_hash(file, ctx->pos); 550 info->curr_minor_hash = pos2min_hash(file, ctx->pos); 551 } 552 553 /* 554 * If there are any leftover names on the hash collision 555 * chain, return them first. 556 */ 557 if (info->extra_fname) { 558 if (call_filldir(file, ctx, info->extra_fname)) 559 goto finished; 560 info->extra_fname = NULL; 561 goto next_node; 562 } else if (!info->curr_node) 563 info->curr_node = rb_first(&info->root); 564 565 while (1) { 566 /* 567 * Fill the rbtree if we have no more entries, 568 * or the inode has changed since we last read in the 569 * cached entries. 570 */ 571 if ((!info->curr_node) || 572 !inode_eq_iversion(inode, file->f_version)) { 573 info->curr_node = NULL; 574 free_rb_tree_fname(&info->root); 575 file->f_version = inode_query_iversion(inode); 576 ret = ext4_htree_fill_tree(file, info->curr_hash, 577 info->curr_minor_hash, 578 &info->next_hash); 579 if (ret < 0) 580 return ret; 581 if (ret == 0) { 582 ctx->pos = ext4_get_htree_eof(file); 583 break; 584 } 585 info->curr_node = rb_first(&info->root); 586 } 587 588 fname = rb_entry(info->curr_node, struct fname, rb_hash); 589 info->curr_hash = fname->hash; 590 info->curr_minor_hash = fname->minor_hash; 591 if (call_filldir(file, ctx, fname)) 592 break; 593 next_node: 594 info->curr_node = rb_next(info->curr_node); 595 if (info->curr_node) { 596 fname = rb_entry(info->curr_node, struct fname, 597 rb_hash); 598 info->curr_hash = fname->hash; 599 info->curr_minor_hash = fname->minor_hash; 600 } else { 601 if (info->next_hash == ~0) { 602 ctx->pos = ext4_get_htree_eof(file); 603 break; 604 } 605 info->curr_hash = info->next_hash; 606 info->curr_minor_hash = 0; 607 } 608 } 609 finished: 610 info->last_pos = ctx->pos; 611 return 0; 612 } 613 614 static int ext4_dir_open(struct inode * inode, struct file * filp) 615 { 616 if (ext4_encrypted_inode(inode)) 617 return fscrypt_get_encryption_info(inode) ? -EACCES : 0; 618 return 0; 619 } 620 621 static int ext4_release_dir(struct inode *inode, struct file *filp) 622 { 623 if (filp->private_data) 624 ext4_htree_free_dir_info(filp->private_data); 625 626 return 0; 627 } 628 629 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf, 630 int buf_size) 631 { 632 struct ext4_dir_entry_2 *de; 633 int rlen; 634 unsigned int offset = 0; 635 char *top; 636 637 de = (struct ext4_dir_entry_2 *)buf; 638 top = buf + buf_size; 639 while ((char *) de < top) { 640 if (ext4_check_dir_entry(dir, NULL, de, bh, 641 buf, buf_size, offset)) 642 return -EFSCORRUPTED; 643 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 644 de = (struct ext4_dir_entry_2 *)((char *)de + rlen); 645 offset += rlen; 646 } 647 if ((char *) de > top) 648 return -EFSCORRUPTED; 649 650 return 0; 651 } 652 653 const struct file_operations ext4_dir_operations = { 654 .llseek = ext4_dir_llseek, 655 .read = generic_read_dir, 656 .iterate_shared = ext4_readdir, 657 .unlocked_ioctl = ext4_ioctl, 658 #ifdef CONFIG_COMPAT 659 .compat_ioctl = ext4_compat_ioctl, 660 #endif 661 .fsync = ext4_sync_file, 662 .open = ext4_dir_open, 663 .release = ext4_release_dir, 664 }; 665