1 /* 2 * linux/fs/ext4/dir.c 3 * 4 * Copyright (C) 1992, 1993, 1994, 1995 5 * Remy Card (card@masi.ibp.fr) 6 * Laboratoire MASI - Institut Blaise Pascal 7 * Universite Pierre et Marie Curie (Paris VI) 8 * 9 * from 10 * 11 * linux/fs/minix/dir.c 12 * 13 * Copyright (C) 1991, 1992 Linus Torvalds 14 * 15 * ext4 directory handling functions 16 * 17 * Big-endian to little-endian byte-swapping/bitmaps by 18 * David S. Miller (davem@caip.rutgers.edu), 1995 19 * 20 * Hash Tree Directory indexing (c) 2001 Daniel Phillips 21 * 22 */ 23 24 #include <linux/fs.h> 25 #include <linux/buffer_head.h> 26 #include <linux/slab.h> 27 #include "ext4.h" 28 #include "xattr.h" 29 30 static int ext4_dx_readdir(struct file *, struct dir_context *); 31 32 /** 33 * Check if the given dir-inode refers to an htree-indexed directory 34 * (or a directory which could potentially get converted to use htree 35 * indexing). 36 * 37 * Return 1 if it is a dx dir, 0 if not 38 */ 39 static int is_dx_dir(struct inode *inode) 40 { 41 struct super_block *sb = inode->i_sb; 42 43 if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb, 44 EXT4_FEATURE_COMPAT_DIR_INDEX) && 45 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) || 46 ((inode->i_size >> sb->s_blocksize_bits) == 1) || 47 ext4_has_inline_data(inode))) 48 return 1; 49 50 return 0; 51 } 52 53 /* 54 * Return 0 if the directory entry is OK, and 1 if there is a problem 55 * 56 * Note: this is the opposite of what ext2 and ext3 historically returned... 57 * 58 * bh passed here can be an inode block or a dir data block, depending 59 * on the inode inline data flag. 60 */ 61 int __ext4_check_dir_entry(const char *function, unsigned int line, 62 struct inode *dir, struct file *filp, 63 struct ext4_dir_entry_2 *de, 64 struct buffer_head *bh, char *buf, int size, 65 unsigned int offset) 66 { 67 const char *error_msg = NULL; 68 const int rlen = ext4_rec_len_from_disk(de->rec_len, 69 dir->i_sb->s_blocksize); 70 71 if (unlikely(rlen < EXT4_DIR_REC_LEN(1))) 72 error_msg = "rec_len is smaller than minimal"; 73 else if (unlikely(rlen % 4 != 0)) 74 error_msg = "rec_len % 4 != 0"; 75 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len))) 76 error_msg = "rec_len is too small for name_len"; 77 else if (unlikely(((char *) de - buf) + rlen > size)) 78 error_msg = "directory entry across range"; 79 else if (unlikely(le32_to_cpu(de->inode) > 80 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))) 81 error_msg = "inode out of bounds"; 82 else 83 return 0; 84 85 if (filp) 86 ext4_error_file(filp, function, line, bh->b_blocknr, 87 "bad entry in directory: %s - offset=%u(%u), " 88 "inode=%u, rec_len=%d, name_len=%d", 89 error_msg, (unsigned) (offset % size), 90 offset, le32_to_cpu(de->inode), 91 rlen, de->name_len); 92 else 93 ext4_error_inode(dir, function, line, bh->b_blocknr, 94 "bad entry in directory: %s - offset=%u(%u), " 95 "inode=%u, rec_len=%d, name_len=%d", 96 error_msg, (unsigned) (offset % size), 97 offset, le32_to_cpu(de->inode), 98 rlen, de->name_len); 99 100 return 1; 101 } 102 103 static int ext4_readdir(struct file *file, struct dir_context *ctx) 104 { 105 unsigned int offset; 106 int i; 107 struct ext4_dir_entry_2 *de; 108 int err; 109 struct inode *inode = file_inode(file); 110 struct super_block *sb = inode->i_sb; 111 struct buffer_head *bh = NULL; 112 int dir_has_error = 0; 113 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}; 114 115 if (is_dx_dir(inode)) { 116 err = ext4_dx_readdir(file, ctx); 117 if (err != ERR_BAD_DX_DIR) { 118 return err; 119 } 120 /* 121 * We don't set the inode dirty flag since it's not 122 * critical that it get flushed back to the disk. 123 */ 124 ext4_clear_inode_flag(file_inode(file), 125 EXT4_INODE_INDEX); 126 } 127 128 if (ext4_has_inline_data(inode)) { 129 int has_inline_data = 1; 130 err = ext4_read_inline_dir(file, ctx, 131 &has_inline_data); 132 if (has_inline_data) 133 return err; 134 } 135 136 err = ext4_setup_fname_crypto(inode); 137 if (err) 138 return err; 139 if (ext4_encrypted_inode(inode)) { 140 err = ext4_fname_crypto_alloc_buffer(inode, EXT4_NAME_LEN, 141 &fname_crypto_str); 142 if (err < 0) 143 return err; 144 } 145 146 offset = ctx->pos & (sb->s_blocksize - 1); 147 148 while (ctx->pos < inode->i_size) { 149 struct ext4_map_blocks map; 150 151 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb); 152 map.m_len = 1; 153 err = ext4_map_blocks(NULL, inode, &map, 0); 154 if (err > 0) { 155 pgoff_t index = map.m_pblk >> 156 (PAGE_CACHE_SHIFT - inode->i_blkbits); 157 if (!ra_has_index(&file->f_ra, index)) 158 page_cache_sync_readahead( 159 sb->s_bdev->bd_inode->i_mapping, 160 &file->f_ra, file, 161 index, 1); 162 file->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT; 163 bh = ext4_bread(NULL, inode, map.m_lblk, 0); 164 if (IS_ERR(bh)) 165 return PTR_ERR(bh); 166 } 167 168 if (!bh) { 169 if (!dir_has_error) { 170 EXT4_ERROR_FILE(file, 0, 171 "directory contains a " 172 "hole at offset %llu", 173 (unsigned long long) ctx->pos); 174 dir_has_error = 1; 175 } 176 /* corrupt size? Maybe no more blocks to read */ 177 if (ctx->pos > inode->i_blocks << 9) 178 break; 179 ctx->pos += sb->s_blocksize - offset; 180 continue; 181 } 182 183 /* Check the checksum */ 184 if (!buffer_verified(bh) && 185 !ext4_dirent_csum_verify(inode, 186 (struct ext4_dir_entry *)bh->b_data)) { 187 EXT4_ERROR_FILE(file, 0, "directory fails checksum " 188 "at offset %llu", 189 (unsigned long long)ctx->pos); 190 ctx->pos += sb->s_blocksize - offset; 191 brelse(bh); 192 bh = NULL; 193 continue; 194 } 195 set_buffer_verified(bh); 196 197 /* If the dir block has changed since the last call to 198 * readdir(2), then we might be pointing to an invalid 199 * dirent right now. Scan from the start of the block 200 * to make sure. */ 201 if (file->f_version != inode->i_version) { 202 for (i = 0; i < sb->s_blocksize && i < offset; ) { 203 de = (struct ext4_dir_entry_2 *) 204 (bh->b_data + i); 205 /* It's too expensive to do a full 206 * dirent test each time round this 207 * loop, but we do have to test at 208 * least that it is non-zero. A 209 * failure will be detected in the 210 * dirent test below. */ 211 if (ext4_rec_len_from_disk(de->rec_len, 212 sb->s_blocksize) < EXT4_DIR_REC_LEN(1)) 213 break; 214 i += ext4_rec_len_from_disk(de->rec_len, 215 sb->s_blocksize); 216 } 217 offset = i; 218 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1)) 219 | offset; 220 file->f_version = inode->i_version; 221 } 222 223 while (ctx->pos < inode->i_size 224 && offset < sb->s_blocksize) { 225 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset); 226 if (ext4_check_dir_entry(inode, file, de, bh, 227 bh->b_data, bh->b_size, 228 offset)) { 229 /* 230 * On error, skip to the next block 231 */ 232 ctx->pos = (ctx->pos | 233 (sb->s_blocksize - 1)) + 1; 234 break; 235 } 236 offset += ext4_rec_len_from_disk(de->rec_len, 237 sb->s_blocksize); 238 if (le32_to_cpu(de->inode)) { 239 if (!ext4_encrypted_inode(inode)) { 240 if (!dir_emit(ctx, de->name, 241 de->name_len, 242 le32_to_cpu(de->inode), 243 get_dtype(sb, de->file_type))) 244 goto done; 245 } else { 246 int save_len = fname_crypto_str.len; 247 248 /* Directory is encrypted */ 249 err = ext4_fname_disk_to_usr(inode, 250 NULL, de, &fname_crypto_str); 251 fname_crypto_str.len = save_len; 252 if (err < 0) 253 goto errout; 254 if (!dir_emit(ctx, 255 fname_crypto_str.name, err, 256 le32_to_cpu(de->inode), 257 get_dtype(sb, de->file_type))) 258 goto done; 259 } 260 } 261 ctx->pos += ext4_rec_len_from_disk(de->rec_len, 262 sb->s_blocksize); 263 } 264 if ((ctx->pos < inode->i_size) && !dir_relax(inode)) 265 goto done; 266 brelse(bh); 267 bh = NULL; 268 offset = 0; 269 } 270 done: 271 err = 0; 272 errout: 273 #ifdef CONFIG_EXT4_FS_ENCRYPTION 274 ext4_fname_crypto_free_buffer(&fname_crypto_str); 275 #endif 276 brelse(bh); 277 return err; 278 } 279 280 static inline int is_32bit_api(void) 281 { 282 #ifdef CONFIG_COMPAT 283 return is_compat_task(); 284 #else 285 return (BITS_PER_LONG == 32); 286 #endif 287 } 288 289 /* 290 * These functions convert from the major/minor hash to an f_pos 291 * value for dx directories 292 * 293 * Upper layer (for example NFS) should specify FMODE_32BITHASH or 294 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted 295 * directly on both 32-bit and 64-bit nodes, under such case, neither 296 * FMODE_32BITHASH nor FMODE_64BITHASH is specified. 297 */ 298 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor) 299 { 300 if ((filp->f_mode & FMODE_32BITHASH) || 301 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 302 return major >> 1; 303 else 304 return ((__u64)(major >> 1) << 32) | (__u64)minor; 305 } 306 307 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos) 308 { 309 if ((filp->f_mode & FMODE_32BITHASH) || 310 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 311 return (pos << 1) & 0xffffffff; 312 else 313 return ((pos >> 32) << 1) & 0xffffffff; 314 } 315 316 static inline __u32 pos2min_hash(struct file *filp, loff_t pos) 317 { 318 if ((filp->f_mode & FMODE_32BITHASH) || 319 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 320 return 0; 321 else 322 return pos & 0xffffffff; 323 } 324 325 /* 326 * Return 32- or 64-bit end-of-file for dx directories 327 */ 328 static inline loff_t ext4_get_htree_eof(struct file *filp) 329 { 330 if ((filp->f_mode & FMODE_32BITHASH) || 331 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 332 return EXT4_HTREE_EOF_32BIT; 333 else 334 return EXT4_HTREE_EOF_64BIT; 335 } 336 337 338 /* 339 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree 340 * directories, where the "offset" is in terms of the filename hash 341 * value instead of the byte offset. 342 * 343 * Because we may return a 64-bit hash that is well beyond offset limits, 344 * we need to pass the max hash as the maximum allowable offset in 345 * the htree directory case. 346 * 347 * For non-htree, ext4_llseek already chooses the proper max offset. 348 */ 349 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence) 350 { 351 struct inode *inode = file->f_mapping->host; 352 int dx_dir = is_dx_dir(inode); 353 loff_t htree_max = ext4_get_htree_eof(file); 354 355 if (likely(dx_dir)) 356 return generic_file_llseek_size(file, offset, whence, 357 htree_max, htree_max); 358 else 359 return ext4_llseek(file, offset, whence); 360 } 361 362 /* 363 * This structure holds the nodes of the red-black tree used to store 364 * the directory entry in hash order. 365 */ 366 struct fname { 367 __u32 hash; 368 __u32 minor_hash; 369 struct rb_node rb_hash; 370 struct fname *next; 371 __u32 inode; 372 __u8 name_len; 373 __u8 file_type; 374 char name[0]; 375 }; 376 377 /* 378 * This functoin implements a non-recursive way of freeing all of the 379 * nodes in the red-black tree. 380 */ 381 static void free_rb_tree_fname(struct rb_root *root) 382 { 383 struct fname *fname, *next; 384 385 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash) 386 while (fname) { 387 struct fname *old = fname; 388 fname = fname->next; 389 kfree(old); 390 } 391 392 *root = RB_ROOT; 393 } 394 395 396 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp, 397 loff_t pos) 398 { 399 struct dir_private_info *p; 400 401 p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL); 402 if (!p) 403 return NULL; 404 p->curr_hash = pos2maj_hash(filp, pos); 405 p->curr_minor_hash = pos2min_hash(filp, pos); 406 return p; 407 } 408 409 void ext4_htree_free_dir_info(struct dir_private_info *p) 410 { 411 free_rb_tree_fname(&p->root); 412 kfree(p); 413 } 414 415 /* 416 * Given a directory entry, enter it into the fname rb tree. 417 * 418 * When filename encryption is enabled, the dirent will hold the 419 * encrypted filename, while the htree will hold decrypted filename. 420 * The decrypted filename is passed in via ent_name. parameter. 421 */ 422 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash, 423 __u32 minor_hash, 424 struct ext4_dir_entry_2 *dirent, 425 struct ext4_str *ent_name) 426 { 427 struct rb_node **p, *parent = NULL; 428 struct fname *fname, *new_fn; 429 struct dir_private_info *info; 430 int len; 431 432 info = dir_file->private_data; 433 p = &info->root.rb_node; 434 435 /* Create and allocate the fname structure */ 436 len = sizeof(struct fname) + ent_name->len + 1; 437 new_fn = kzalloc(len, GFP_KERNEL); 438 if (!new_fn) 439 return -ENOMEM; 440 new_fn->hash = hash; 441 new_fn->minor_hash = minor_hash; 442 new_fn->inode = le32_to_cpu(dirent->inode); 443 new_fn->name_len = ent_name->len; 444 new_fn->file_type = dirent->file_type; 445 memcpy(new_fn->name, ent_name->name, ent_name->len); 446 new_fn->name[ent_name->len] = 0; 447 448 while (*p) { 449 parent = *p; 450 fname = rb_entry(parent, struct fname, rb_hash); 451 452 /* 453 * If the hash and minor hash match up, then we put 454 * them on a linked list. This rarely happens... 455 */ 456 if ((new_fn->hash == fname->hash) && 457 (new_fn->minor_hash == fname->minor_hash)) { 458 new_fn->next = fname->next; 459 fname->next = new_fn; 460 return 0; 461 } 462 463 if (new_fn->hash < fname->hash) 464 p = &(*p)->rb_left; 465 else if (new_fn->hash > fname->hash) 466 p = &(*p)->rb_right; 467 else if (new_fn->minor_hash < fname->minor_hash) 468 p = &(*p)->rb_left; 469 else /* if (new_fn->minor_hash > fname->minor_hash) */ 470 p = &(*p)->rb_right; 471 } 472 473 rb_link_node(&new_fn->rb_hash, parent, p); 474 rb_insert_color(&new_fn->rb_hash, &info->root); 475 return 0; 476 } 477 478 479 480 /* 481 * This is a helper function for ext4_dx_readdir. It calls filldir 482 * for all entres on the fname linked list. (Normally there is only 483 * one entry on the linked list, unless there are 62 bit hash collisions.) 484 */ 485 static int call_filldir(struct file *file, struct dir_context *ctx, 486 struct fname *fname) 487 { 488 struct dir_private_info *info = file->private_data; 489 struct inode *inode = file_inode(file); 490 struct super_block *sb = inode->i_sb; 491 492 if (!fname) { 493 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: " 494 "called with null fname?!?", __func__, __LINE__, 495 inode->i_ino, current->comm); 496 return 0; 497 } 498 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash); 499 while (fname) { 500 if (!dir_emit(ctx, fname->name, 501 fname->name_len, 502 fname->inode, 503 get_dtype(sb, fname->file_type))) { 504 info->extra_fname = fname; 505 return 1; 506 } 507 fname = fname->next; 508 } 509 return 0; 510 } 511 512 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx) 513 { 514 struct dir_private_info *info = file->private_data; 515 struct inode *inode = file_inode(file); 516 struct fname *fname; 517 int ret; 518 519 if (!info) { 520 info = ext4_htree_create_dir_info(file, ctx->pos); 521 if (!info) 522 return -ENOMEM; 523 file->private_data = info; 524 } 525 526 if (ctx->pos == ext4_get_htree_eof(file)) 527 return 0; /* EOF */ 528 529 /* Some one has messed with f_pos; reset the world */ 530 if (info->last_pos != ctx->pos) { 531 free_rb_tree_fname(&info->root); 532 info->curr_node = NULL; 533 info->extra_fname = NULL; 534 info->curr_hash = pos2maj_hash(file, ctx->pos); 535 info->curr_minor_hash = pos2min_hash(file, ctx->pos); 536 } 537 538 /* 539 * If there are any leftover names on the hash collision 540 * chain, return them first. 541 */ 542 if (info->extra_fname) { 543 if (call_filldir(file, ctx, info->extra_fname)) 544 goto finished; 545 info->extra_fname = NULL; 546 goto next_node; 547 } else if (!info->curr_node) 548 info->curr_node = rb_first(&info->root); 549 550 while (1) { 551 /* 552 * Fill the rbtree if we have no more entries, 553 * or the inode has changed since we last read in the 554 * cached entries. 555 */ 556 if ((!info->curr_node) || 557 (file->f_version != inode->i_version)) { 558 info->curr_node = NULL; 559 free_rb_tree_fname(&info->root); 560 file->f_version = inode->i_version; 561 ret = ext4_htree_fill_tree(file, info->curr_hash, 562 info->curr_minor_hash, 563 &info->next_hash); 564 if (ret < 0) 565 return ret; 566 if (ret == 0) { 567 ctx->pos = ext4_get_htree_eof(file); 568 break; 569 } 570 info->curr_node = rb_first(&info->root); 571 } 572 573 fname = rb_entry(info->curr_node, struct fname, rb_hash); 574 info->curr_hash = fname->hash; 575 info->curr_minor_hash = fname->minor_hash; 576 if (call_filldir(file, ctx, fname)) 577 break; 578 next_node: 579 info->curr_node = rb_next(info->curr_node); 580 if (info->curr_node) { 581 fname = rb_entry(info->curr_node, struct fname, 582 rb_hash); 583 info->curr_hash = fname->hash; 584 info->curr_minor_hash = fname->minor_hash; 585 } else { 586 if (info->next_hash == ~0) { 587 ctx->pos = ext4_get_htree_eof(file); 588 break; 589 } 590 info->curr_hash = info->next_hash; 591 info->curr_minor_hash = 0; 592 } 593 } 594 finished: 595 info->last_pos = ctx->pos; 596 return 0; 597 } 598 599 static int ext4_release_dir(struct inode *inode, struct file *filp) 600 { 601 if (filp->private_data) 602 ext4_htree_free_dir_info(filp->private_data); 603 604 return 0; 605 } 606 607 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf, 608 int buf_size) 609 { 610 struct ext4_dir_entry_2 *de; 611 int nlen, rlen; 612 unsigned int offset = 0; 613 char *top; 614 615 de = (struct ext4_dir_entry_2 *)buf; 616 top = buf + buf_size; 617 while ((char *) de < top) { 618 if (ext4_check_dir_entry(dir, NULL, de, bh, 619 buf, buf_size, offset)) 620 return -EIO; 621 nlen = EXT4_DIR_REC_LEN(de->name_len); 622 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 623 de = (struct ext4_dir_entry_2 *)((char *)de + rlen); 624 offset += rlen; 625 } 626 if ((char *) de > top) 627 return -EIO; 628 629 return 0; 630 } 631 632 const struct file_operations ext4_dir_operations = { 633 .llseek = ext4_dir_llseek, 634 .read = generic_read_dir, 635 .iterate = ext4_readdir, 636 .unlocked_ioctl = ext4_ioctl, 637 #ifdef CONFIG_COMPAT 638 .compat_ioctl = ext4_compat_ioctl, 639 #endif 640 .fsync = ext4_sync_file, 641 .release = ext4_release_dir, 642 }; 643