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 int dir_has_error = 0; 112 113 if (is_dx_dir(inode)) { 114 err = ext4_dx_readdir(file, ctx); 115 if (err != ERR_BAD_DX_DIR) { 116 return err; 117 } 118 /* 119 * We don't set the inode dirty flag since it's not 120 * critical that it get flushed back to the disk. 121 */ 122 ext4_clear_inode_flag(file_inode(file), 123 EXT4_INODE_INDEX); 124 } 125 126 if (ext4_has_inline_data(inode)) { 127 int has_inline_data = 1; 128 int ret = ext4_read_inline_dir(file, ctx, 129 &has_inline_data); 130 if (has_inline_data) 131 return ret; 132 } 133 134 offset = ctx->pos & (sb->s_blocksize - 1); 135 136 while (ctx->pos < inode->i_size) { 137 struct ext4_map_blocks map; 138 struct buffer_head *bh = NULL; 139 140 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb); 141 map.m_len = 1; 142 err = ext4_map_blocks(NULL, inode, &map, 0); 143 if (err > 0) { 144 pgoff_t index = map.m_pblk >> 145 (PAGE_CACHE_SHIFT - inode->i_blkbits); 146 if (!ra_has_index(&file->f_ra, index)) 147 page_cache_sync_readahead( 148 sb->s_bdev->bd_inode->i_mapping, 149 &file->f_ra, file, 150 index, 1); 151 file->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT; 152 bh = ext4_bread(NULL, inode, map.m_lblk, 0); 153 if (IS_ERR(bh)) 154 return PTR_ERR(bh); 155 } 156 157 if (!bh) { 158 if (!dir_has_error) { 159 EXT4_ERROR_FILE(file, 0, 160 "directory contains a " 161 "hole at offset %llu", 162 (unsigned long long) ctx->pos); 163 dir_has_error = 1; 164 } 165 /* corrupt size? Maybe no more blocks to read */ 166 if (ctx->pos > inode->i_blocks << 9) 167 break; 168 ctx->pos += sb->s_blocksize - offset; 169 continue; 170 } 171 172 /* Check the checksum */ 173 if (!buffer_verified(bh) && 174 !ext4_dirent_csum_verify(inode, 175 (struct ext4_dir_entry *)bh->b_data)) { 176 EXT4_ERROR_FILE(file, 0, "directory fails checksum " 177 "at offset %llu", 178 (unsigned long long)ctx->pos); 179 ctx->pos += sb->s_blocksize - offset; 180 brelse(bh); 181 continue; 182 } 183 set_buffer_verified(bh); 184 185 /* If the dir block has changed since the last call to 186 * readdir(2), then we might be pointing to an invalid 187 * dirent right now. Scan from the start of the block 188 * to make sure. */ 189 if (file->f_version != inode->i_version) { 190 for (i = 0; i < sb->s_blocksize && i < offset; ) { 191 de = (struct ext4_dir_entry_2 *) 192 (bh->b_data + i); 193 /* It's too expensive to do a full 194 * dirent test each time round this 195 * loop, but we do have to test at 196 * least that it is non-zero. A 197 * failure will be detected in the 198 * dirent test below. */ 199 if (ext4_rec_len_from_disk(de->rec_len, 200 sb->s_blocksize) < EXT4_DIR_REC_LEN(1)) 201 break; 202 i += ext4_rec_len_from_disk(de->rec_len, 203 sb->s_blocksize); 204 } 205 offset = i; 206 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1)) 207 | offset; 208 file->f_version = inode->i_version; 209 } 210 211 while (ctx->pos < inode->i_size 212 && offset < sb->s_blocksize) { 213 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset); 214 if (ext4_check_dir_entry(inode, file, de, bh, 215 bh->b_data, bh->b_size, 216 offset)) { 217 /* 218 * On error, skip to the next block 219 */ 220 ctx->pos = (ctx->pos | 221 (sb->s_blocksize - 1)) + 1; 222 break; 223 } 224 offset += ext4_rec_len_from_disk(de->rec_len, 225 sb->s_blocksize); 226 if (le32_to_cpu(de->inode)) { 227 if (!dir_emit(ctx, de->name, 228 de->name_len, 229 le32_to_cpu(de->inode), 230 get_dtype(sb, de->file_type))) { 231 brelse(bh); 232 return 0; 233 } 234 } 235 ctx->pos += ext4_rec_len_from_disk(de->rec_len, 236 sb->s_blocksize); 237 } 238 offset = 0; 239 brelse(bh); 240 if (ctx->pos < inode->i_size) { 241 if (!dir_relax(inode)) 242 return 0; 243 } 244 } 245 return 0; 246 } 247 248 static inline int is_32bit_api(void) 249 { 250 #ifdef CONFIG_COMPAT 251 return is_compat_task(); 252 #else 253 return (BITS_PER_LONG == 32); 254 #endif 255 } 256 257 /* 258 * These functions convert from the major/minor hash to an f_pos 259 * value for dx directories 260 * 261 * Upper layer (for example NFS) should specify FMODE_32BITHASH or 262 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted 263 * directly on both 32-bit and 64-bit nodes, under such case, neither 264 * FMODE_32BITHASH nor FMODE_64BITHASH is specified. 265 */ 266 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor) 267 { 268 if ((filp->f_mode & FMODE_32BITHASH) || 269 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 270 return major >> 1; 271 else 272 return ((__u64)(major >> 1) << 32) | (__u64)minor; 273 } 274 275 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos) 276 { 277 if ((filp->f_mode & FMODE_32BITHASH) || 278 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 279 return (pos << 1) & 0xffffffff; 280 else 281 return ((pos >> 32) << 1) & 0xffffffff; 282 } 283 284 static inline __u32 pos2min_hash(struct file *filp, loff_t pos) 285 { 286 if ((filp->f_mode & FMODE_32BITHASH) || 287 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 288 return 0; 289 else 290 return pos & 0xffffffff; 291 } 292 293 /* 294 * Return 32- or 64-bit end-of-file for dx directories 295 */ 296 static inline loff_t ext4_get_htree_eof(struct file *filp) 297 { 298 if ((filp->f_mode & FMODE_32BITHASH) || 299 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 300 return EXT4_HTREE_EOF_32BIT; 301 else 302 return EXT4_HTREE_EOF_64BIT; 303 } 304 305 306 /* 307 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree 308 * directories, where the "offset" is in terms of the filename hash 309 * value instead of the byte offset. 310 * 311 * Because we may return a 64-bit hash that is well beyond offset limits, 312 * we need to pass the max hash as the maximum allowable offset in 313 * the htree directory case. 314 * 315 * For non-htree, ext4_llseek already chooses the proper max offset. 316 */ 317 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence) 318 { 319 struct inode *inode = file->f_mapping->host; 320 int dx_dir = is_dx_dir(inode); 321 loff_t htree_max = ext4_get_htree_eof(file); 322 323 if (likely(dx_dir)) 324 return generic_file_llseek_size(file, offset, whence, 325 htree_max, htree_max); 326 else 327 return ext4_llseek(file, offset, whence); 328 } 329 330 /* 331 * This structure holds the nodes of the red-black tree used to store 332 * the directory entry in hash order. 333 */ 334 struct fname { 335 __u32 hash; 336 __u32 minor_hash; 337 struct rb_node rb_hash; 338 struct fname *next; 339 __u32 inode; 340 __u8 name_len; 341 __u8 file_type; 342 char name[0]; 343 }; 344 345 /* 346 * This functoin implements a non-recursive way of freeing all of the 347 * nodes in the red-black tree. 348 */ 349 static void free_rb_tree_fname(struct rb_root *root) 350 { 351 struct fname *fname, *next; 352 353 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash) 354 while (fname) { 355 struct fname *old = fname; 356 fname = fname->next; 357 kfree(old); 358 } 359 360 *root = RB_ROOT; 361 } 362 363 364 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp, 365 loff_t pos) 366 { 367 struct dir_private_info *p; 368 369 p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL); 370 if (!p) 371 return NULL; 372 p->curr_hash = pos2maj_hash(filp, pos); 373 p->curr_minor_hash = pos2min_hash(filp, pos); 374 return p; 375 } 376 377 void ext4_htree_free_dir_info(struct dir_private_info *p) 378 { 379 free_rb_tree_fname(&p->root); 380 kfree(p); 381 } 382 383 /* 384 * Given a directory entry, enter it into the fname rb tree. 385 */ 386 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash, 387 __u32 minor_hash, 388 struct ext4_dir_entry_2 *dirent) 389 { 390 struct rb_node **p, *parent = NULL; 391 struct fname *fname, *new_fn; 392 struct dir_private_info *info; 393 int len; 394 395 info = dir_file->private_data; 396 p = &info->root.rb_node; 397 398 /* Create and allocate the fname structure */ 399 len = sizeof(struct fname) + dirent->name_len + 1; 400 new_fn = kzalloc(len, GFP_KERNEL); 401 if (!new_fn) 402 return -ENOMEM; 403 new_fn->hash = hash; 404 new_fn->minor_hash = minor_hash; 405 new_fn->inode = le32_to_cpu(dirent->inode); 406 new_fn->name_len = dirent->name_len; 407 new_fn->file_type = dirent->file_type; 408 memcpy(new_fn->name, dirent->name, dirent->name_len); 409 new_fn->name[dirent->name_len] = 0; 410 411 while (*p) { 412 parent = *p; 413 fname = rb_entry(parent, struct fname, rb_hash); 414 415 /* 416 * If the hash and minor hash match up, then we put 417 * them on a linked list. This rarely happens... 418 */ 419 if ((new_fn->hash == fname->hash) && 420 (new_fn->minor_hash == fname->minor_hash)) { 421 new_fn->next = fname->next; 422 fname->next = new_fn; 423 return 0; 424 } 425 426 if (new_fn->hash < fname->hash) 427 p = &(*p)->rb_left; 428 else if (new_fn->hash > fname->hash) 429 p = &(*p)->rb_right; 430 else if (new_fn->minor_hash < fname->minor_hash) 431 p = &(*p)->rb_left; 432 else /* if (new_fn->minor_hash > fname->minor_hash) */ 433 p = &(*p)->rb_right; 434 } 435 436 rb_link_node(&new_fn->rb_hash, parent, p); 437 rb_insert_color(&new_fn->rb_hash, &info->root); 438 return 0; 439 } 440 441 442 443 /* 444 * This is a helper function for ext4_dx_readdir. It calls filldir 445 * for all entres on the fname linked list. (Normally there is only 446 * one entry on the linked list, unless there are 62 bit hash collisions.) 447 */ 448 static int call_filldir(struct file *file, struct dir_context *ctx, 449 struct fname *fname) 450 { 451 struct dir_private_info *info = file->private_data; 452 struct inode *inode = file_inode(file); 453 struct super_block *sb = inode->i_sb; 454 455 if (!fname) { 456 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: " 457 "called with null fname?!?", __func__, __LINE__, 458 inode->i_ino, current->comm); 459 return 0; 460 } 461 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash); 462 while (fname) { 463 if (!dir_emit(ctx, fname->name, 464 fname->name_len, 465 fname->inode, 466 get_dtype(sb, fname->file_type))) { 467 info->extra_fname = fname; 468 return 1; 469 } 470 fname = fname->next; 471 } 472 return 0; 473 } 474 475 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx) 476 { 477 struct dir_private_info *info = file->private_data; 478 struct inode *inode = file_inode(file); 479 struct fname *fname; 480 int ret; 481 482 if (!info) { 483 info = ext4_htree_create_dir_info(file, ctx->pos); 484 if (!info) 485 return -ENOMEM; 486 file->private_data = info; 487 } 488 489 if (ctx->pos == ext4_get_htree_eof(file)) 490 return 0; /* EOF */ 491 492 /* Some one has messed with f_pos; reset the world */ 493 if (info->last_pos != ctx->pos) { 494 free_rb_tree_fname(&info->root); 495 info->curr_node = NULL; 496 info->extra_fname = NULL; 497 info->curr_hash = pos2maj_hash(file, ctx->pos); 498 info->curr_minor_hash = pos2min_hash(file, ctx->pos); 499 } 500 501 /* 502 * If there are any leftover names on the hash collision 503 * chain, return them first. 504 */ 505 if (info->extra_fname) { 506 if (call_filldir(file, ctx, info->extra_fname)) 507 goto finished; 508 info->extra_fname = NULL; 509 goto next_node; 510 } else if (!info->curr_node) 511 info->curr_node = rb_first(&info->root); 512 513 while (1) { 514 /* 515 * Fill the rbtree if we have no more entries, 516 * or the inode has changed since we last read in the 517 * cached entries. 518 */ 519 if ((!info->curr_node) || 520 (file->f_version != inode->i_version)) { 521 info->curr_node = NULL; 522 free_rb_tree_fname(&info->root); 523 file->f_version = inode->i_version; 524 ret = ext4_htree_fill_tree(file, info->curr_hash, 525 info->curr_minor_hash, 526 &info->next_hash); 527 if (ret < 0) 528 return ret; 529 if (ret == 0) { 530 ctx->pos = ext4_get_htree_eof(file); 531 break; 532 } 533 info->curr_node = rb_first(&info->root); 534 } 535 536 fname = rb_entry(info->curr_node, struct fname, rb_hash); 537 info->curr_hash = fname->hash; 538 info->curr_minor_hash = fname->minor_hash; 539 if (call_filldir(file, ctx, fname)) 540 break; 541 next_node: 542 info->curr_node = rb_next(info->curr_node); 543 if (info->curr_node) { 544 fname = rb_entry(info->curr_node, struct fname, 545 rb_hash); 546 info->curr_hash = fname->hash; 547 info->curr_minor_hash = fname->minor_hash; 548 } else { 549 if (info->next_hash == ~0) { 550 ctx->pos = ext4_get_htree_eof(file); 551 break; 552 } 553 info->curr_hash = info->next_hash; 554 info->curr_minor_hash = 0; 555 } 556 } 557 finished: 558 info->last_pos = ctx->pos; 559 return 0; 560 } 561 562 static int ext4_release_dir(struct inode *inode, struct file *filp) 563 { 564 if (filp->private_data) 565 ext4_htree_free_dir_info(filp->private_data); 566 567 return 0; 568 } 569 570 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf, 571 int buf_size) 572 { 573 struct ext4_dir_entry_2 *de; 574 int nlen, rlen; 575 unsigned int offset = 0; 576 char *top; 577 578 de = (struct ext4_dir_entry_2 *)buf; 579 top = buf + buf_size; 580 while ((char *) de < top) { 581 if (ext4_check_dir_entry(dir, NULL, de, bh, 582 buf, buf_size, offset)) 583 return -EIO; 584 nlen = EXT4_DIR_REC_LEN(de->name_len); 585 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); 586 de = (struct ext4_dir_entry_2 *)((char *)de + rlen); 587 offset += rlen; 588 } 589 if ((char *) de > top) 590 return -EIO; 591 592 return 0; 593 } 594 595 const struct file_operations ext4_dir_operations = { 596 .llseek = ext4_dir_llseek, 597 .read = generic_read_dir, 598 .iterate = ext4_readdir, 599 .unlocked_ioctl = ext4_ioctl, 600 #ifdef CONFIG_COMPAT 601 .compat_ioctl = ext4_compat_ioctl, 602 #endif 603 .fsync = ext4_sync_file, 604 .release = ext4_release_dir, 605 }; 606