fs/ext3/dir.c

   1 /*
   2  *  linux/fs/ext3/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  *  ext3 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/jbd.h>
  26 #include <linux/ext3_fs.h>
  27 #include <linux/buffer_head.h>
  28 #include <linux/compat.h>
  29 #include <linux/slab.h>
  30 #include <linux/rbtree.h>
  31
  32 static unsigned char ext3_filetype_table[] = {
  33         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
  34 };
  35
  36 static int ext3_dx_readdir(struct file * filp,
  37                            void * dirent, filldir_t filldir);
  38
  39 static unsigned char get_dtype(struct super_block *sb, int filetype)
  40 {
  41         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE) ||
  42             (filetype >= EXT3_FT_MAX))
  43                 return DT_UNKNOWN;
  44
  45         return (ext3_filetype_table[filetype]);
  46 }
  47
  48 /**
  49  * Check if the given dir-inode refers to an htree-indexed directory
  50  * (or a directory which chould potentially get coverted to use htree
  51  * indexing).
  52  *
  53  * Return 1 if it is a dx dir, 0 if not
  54  */
  55 static int is_dx_dir(struct inode *inode)
  56 {
  57         struct super_block *sb = inode->i_sb;
  58
  59         if (EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
  60                      EXT3_FEATURE_COMPAT_DIR_INDEX) &&
  61             ((EXT3_I(inode)->i_flags & EXT3_INDEX_FL) ||
  62              ((inode->i_size >> sb->s_blocksize_bits) == 1)))
  63                 return 1;
  64
  65         return 0;
  66 }
  67
  68 int ext3_check_dir_entry (const char * function, struct inode * dir,
  69                           struct ext3_dir_entry_2 * de,
  70                           struct buffer_head * bh,
  71                           unsigned long offset)
  72 {
  73         const char * error_msg = NULL;
  74         const int rlen = ext3_rec_len_from_disk(de->rec_len);
  75
  76         if (unlikely(rlen < EXT3_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 < EXT3_DIR_REC_LEN(de->name_len)))
  81                 error_msg = "rec_len is too small for name_len";
  82         else if (unlikely((((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)))
  83                 error_msg = "directory entry across blocks";
  84         else if (unlikely(le32_to_cpu(de->inode) >
  85                         le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count)))
  86                 error_msg = "inode out of bounds";
  87
  88         if (unlikely(error_msg != NULL))
  89                 ext3_error (dir->i_sb, function,
  90                         "bad entry in directory #%lu: %s - "
  91                         "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
  92                         dir->i_ino, error_msg, offset,
  93                         (unsigned long) le32_to_cpu(de->inode),
  94                         rlen, de->name_len);
  95
  96         return error_msg == NULL ? 1 : 0;
  97 }
  98
  99 static int ext3_readdir(struct file * filp,
 100                          void * dirent, filldir_t filldir)
 101 {
 102         int error = 0;
 103         unsigned long offset;
 104         int i, stored;
 105         struct ext3_dir_entry_2 *de;
 106         int err;
 107         struct inode *inode = filp->f_path.dentry->d_inode;
 108         struct super_block *sb = inode->i_sb;
 109         int ret = 0;
 110         int dir_has_error = 0;
 111
 112         if (is_dx_dir(inode)) {
 113                 err = ext3_dx_readdir(filp, dirent, filldir);
 114                 if (err != ERR_BAD_DX_DIR) {
 115                         ret = err;
 116                         goto out;
 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                 EXT3_I(filp->f_path.dentry->d_inode)->i_flags &= ~EXT3_INDEX_FL;
 123         }
 124         stored = 0;
 125         offset = filp->f_pos & (sb->s_blocksize - 1);
 126
 127         while (!error && !stored && filp->f_pos < inode->i_size) {
 128                 unsigned long blk = filp->f_pos >> EXT3_BLOCK_SIZE_BITS(sb);
 129                 struct buffer_head map_bh;
 130                 struct buffer_head *bh = NULL;
 131
 132                 map_bh.b_state = 0;
 133                 err = ext3_get_blocks_handle(NULL, inode, blk, 1, &map_bh, 0);
 134                 if (err > 0) {
 135                         pgoff_t index = map_bh.b_blocknr >>
 136                                         (PAGE_CACHE_SHIFT - inode->i_blkbits);
 137                         if (!ra_has_index(&filp->f_ra, index))
 138                                 page_cache_sync_readahead(
 139                                         sb->s_bdev->bd_inode->i_mapping,
 140                                         &filp->f_ra, filp,
 141                                         index, 1);
 142                         filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
 143                         bh = ext3_bread(NULL, inode, blk, 0, &err);
 144                 }
 145
 146                 /*
 147                  * We ignore I/O errors on directories so users have a chance
 148                  * of recovering data when there's a bad sector
 149                  */
 150                 if (!bh) {
 151                         if (!dir_has_error) {
 152                                 ext3_error(sb, __func__, "directory #%lu "
 153                                         "contains a hole at offset %lld",
 154                                         inode->i_ino, filp->f_pos);
 155                                 dir_has_error = 1;
 156                         }
 157                         /* corrupt size?  Maybe no more blocks to read */
 158                         if (filp->f_pos > inode->i_blocks << 9)
 159                                 break;
 160                         filp->f_pos += sb->s_blocksize - offset;
 161                         continue;
 162                 }
 163
 164 revalidate:
 165                 /* If the dir block has changed since the last call to
 166                  * readdir(2), then we might be pointing to an invalid
 167                  * dirent right now.  Scan from the start of the block
 168                  * to make sure. */
 169                 if (filp->f_version != inode->i_version) {
 170                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
 171                                 de = (struct ext3_dir_entry_2 *)
 172                                         (bh->b_data + i);
 173                                 /* It's too expensive to do a full
 174                                  * dirent test each time round this
 175                                  * loop, but we do have to test at
 176                                  * least that it is non-zero.  A
 177                                  * failure will be detected in the
 178                                  * dirent test below. */
 179                                 if (ext3_rec_len_from_disk(de->rec_len) <
 180                                                 EXT3_DIR_REC_LEN(1))
 181                                         break;
 182                                 i += ext3_rec_len_from_disk(de->rec_len);
 183                         }
 184                         offset = i;
 185                         filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
 186                                 | offset;
 187                         filp->f_version = inode->i_version;
 188                 }
 189
 190                 while (!error && filp->f_pos < inode->i_size
 191                        && offset < sb->s_blocksize) {
 192                         de = (struct ext3_dir_entry_2 *) (bh->b_data + offset);
 193                         if (!ext3_check_dir_entry ("ext3_readdir", inode, de,
 194                                                    bh, offset)) {
 195                                 /* On error, skip the f_pos to the
 196                                    next block. */
 197                                 filp->f_pos = (filp->f_pos |
 198                                                 (sb->s_blocksize - 1)) + 1;
 199                                 brelse (bh);
 200                                 ret = stored;
 201                                 goto out;
 202                         }
 203                         offset += ext3_rec_len_from_disk(de->rec_len);
 204                         if (le32_to_cpu(de->inode)) {
 205                                 /* We might block in the next section
 206                                  * if the data destination is
 207                                  * currently swapped out.  So, use a
 208                                  * version stamp to detect whether or
 209                                  * not the directory has been modified
 210                                  * during the copy operation.
 211                                  */
 212                                 u64 version = filp->f_version;
 213
 214                                 error = filldir(dirent, de->name,
 215                                                 de->name_len,
 216                                                 filp->f_pos,
 217                                                 le32_to_cpu(de->inode),
 218                                                 get_dtype(sb, de->file_type));
 219                                 if (error)
 220                                         break;
 221                                 if (version != filp->f_version)
 222                                         goto revalidate;
 223                                 stored ++;
 224                         }
 225                         filp->f_pos += ext3_rec_len_from_disk(de->rec_len);
 226                 }
 227                 offset = 0;
 228                 brelse (bh);
 229         }
 230 out:
 231         return ret;
 232 }
 233
 234 static inline int is_32bit_api(void)
 235 {
 236 #ifdef CONFIG_COMPAT
 237         return is_compat_task();
 238 #else
 239         return (BITS_PER_LONG == 32);
 240 #endif
 241 }
 242
 243 /*
 244  * These functions convert from the major/minor hash to an f_pos
 245  * value for dx directories
 246  *
 247  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
 248  * FMODE_64BITHASH explicitly. On the other hand, we allow ext3 to be mounted
 249  * directly on both 32-bit and 64-bit nodes, under such case, neither
 250  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
 251  */
 252 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
 253 {
 254         if ((filp->f_mode & FMODE_32BITHASH) ||
 255             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 256                 return major >> 1;
 257         else
 258                 return ((__u64)(major >> 1) << 32) | (__u64)minor;
 259 }
 260
 261 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
 262 {
 263         if ((filp->f_mode & FMODE_32BITHASH) ||
 264             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 265                 return (pos << 1) & 0xffffffff;
 266         else
 267                 return ((pos >> 32) << 1) & 0xffffffff;
 268 }
 269
 270 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
 271 {
 272         if ((filp->f_mode & FMODE_32BITHASH) ||
 273             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 274                 return 0;
 275         else
 276                 return pos & 0xffffffff;
 277 }
 278
 279 /*
 280  * Return 32- or 64-bit end-of-file for dx directories
 281  */
 282 static inline loff_t ext3_get_htree_eof(struct file *filp)
 283 {
 284         if ((filp->f_mode & FMODE_32BITHASH) ||
 285             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 286                 return EXT3_HTREE_EOF_32BIT;
 287         else
 288                 return EXT3_HTREE_EOF_64BIT;
 289 }
 290
 291
 292 /*
 293  * ext3_dir_llseek() calls generic_file_llseek[_size]() to handle both
 294  * non-htree and htree directories, where the "offset" is in terms
 295  * of the filename hash value instead of the byte offset.
 296  *
 297  * Because we may return a 64-bit hash that is well beyond s_maxbytes,
 298  * we need to pass the max hash as the maximum allowable offset in
 299  * the htree directory case.
 300  *
 301  * NOTE: offsets obtained *before* ext3_set_inode_flag(dir, EXT3_INODE_INDEX)
 302  *       will be invalid once the directory was converted into a dx directory
 303  */
 304 loff_t ext3_dir_llseek(struct file *file, loff_t offset, int origin)
 305 {
 306         struct inode *inode = file->f_mapping->host;
 307         int dx_dir = is_dx_dir(inode);
 308
 309         if (likely(dx_dir))
 310                 return generic_file_llseek_size(file, offset, origin,
 311                                                 ext3_get_htree_eof(file));
 312         else
 313                 return generic_file_llseek(file, offset, origin);
 314 }
 315
 316 /*
 317  * This structure holds the nodes of the red-black tree used to store
 318  * the directory entry in hash order.
 319  */
 320 struct fname {
 321         __u32           hash;
 322         __u32           minor_hash;
 323         struct rb_node  rb_hash;
 324         struct fname    *next;
 325         __u32           inode;
 326         __u8            name_len;
 327         __u8            file_type;
 328         char            name[0];
 329 };
 330
 331 /*
 332  * This functoin implements a non-recursive way of freeing all of the
 333  * nodes in the red-black tree.
 334  */
 335 static void free_rb_tree_fname(struct rb_root *root)
 336 {
 337         struct rb_node  *n = root->rb_node;
 338         struct rb_node  *parent;
 339         struct fname    *fname;
 340
 341         while (n) {
 342                 /* Do the node's children first */
 343                 if (n->rb_left) {
 344                         n = n->rb_left;
 345                         continue;
 346                 }
 347                 if (n->rb_right) {
 348                         n = n->rb_right;
 349                         continue;
 350                 }
 351                 /*
 352                  * The node has no children; free it, and then zero
 353                  * out parent's link to it.  Finally go to the
 354                  * beginning of the loop and try to free the parent
 355                  * node.
 356                  */
 357                 parent = rb_parent(n);
 358                 fname = rb_entry(n, struct fname, rb_hash);
 359                 while (fname) {
 360                         struct fname * old = fname;
 361                         fname = fname->next;
 362                         kfree (old);
 363                 }
 364                 if (!parent)
 365                         *root = RB_ROOT;
 366                 else if (parent->rb_left == n)
 367                         parent->rb_left = NULL;
 368                 else if (parent->rb_right == n)
 369                         parent->rb_right = NULL;
 370                 n = parent;
 371         }
 372 }
 373
 374
 375 static struct dir_private_info *ext3_htree_create_dir_info(struct file *filp,
 376                                                            loff_t pos)
 377 {
 378         struct dir_private_info *p;
 379
 380         p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
 381         if (!p)
 382                 return NULL;
 383         p->curr_hash = pos2maj_hash(filp, pos);
 384         p->curr_minor_hash = pos2min_hash(filp, pos);
 385         return p;
 386 }
 387
 388 void ext3_htree_free_dir_info(struct dir_private_info *p)
 389 {
 390         free_rb_tree_fname(&p->root);
 391         kfree(p);
 392 }
 393
 394 /*
 395  * Given a directory entry, enter it into the fname rb tree.
 396  */
 397 int ext3_htree_store_dirent(struct file *dir_file, __u32 hash,
 398                              __u32 minor_hash,
 399                              struct ext3_dir_entry_2 *dirent)
 400 {
 401         struct rb_node **p, *parent = NULL;
 402         struct fname * fname, *new_fn;
 403         struct dir_private_info *info;
 404         int len;
 405
 406         info = (struct dir_private_info *) dir_file->private_data;
 407         p = &info->root.rb_node;
 408
 409         /* Create and allocate the fname structure */
 410         len = sizeof(struct fname) + dirent->name_len + 1;
 411         new_fn = kzalloc(len, GFP_KERNEL);
 412         if (!new_fn)
 413                 return -ENOMEM;
 414         new_fn->hash = hash;
 415         new_fn->minor_hash = minor_hash;
 416         new_fn->inode = le32_to_cpu(dirent->inode);
 417         new_fn->name_len = dirent->name_len;
 418         new_fn->file_type = dirent->file_type;
 419         memcpy(new_fn->name, dirent->name, dirent->name_len);
 420         new_fn->name[dirent->name_len] = 0;
 421
 422         while (*p) {
 423                 parent = *p;
 424                 fname = rb_entry(parent, struct fname, rb_hash);
 425
 426                 /*
 427                  * If the hash and minor hash match up, then we put
 428                  * them on a linked list.  This rarely happens...
 429                  */
 430                 if ((new_fn->hash == fname->hash) &&
 431                     (new_fn->minor_hash == fname->minor_hash)) {
 432                         new_fn->next = fname->next;
 433                         fname->next = new_fn;
 434                         return 0;
 435                 }
 436
 437                 if (new_fn->hash < fname->hash)
 438                         p = &(*p)->rb_left;
 439                 else if (new_fn->hash > fname->hash)
 440                         p = &(*p)->rb_right;
 441                 else if (new_fn->minor_hash < fname->minor_hash)
 442                         p = &(*p)->rb_left;
 443                 else /* if (new_fn->minor_hash > fname->minor_hash) */
 444                         p = &(*p)->rb_right;
 445         }
 446
 447         rb_link_node(&new_fn->rb_hash, parent, p);
 448         rb_insert_color(&new_fn->rb_hash, &info->root);
 449         return 0;
 450 }
 451
 452
 453
 454 /*
 455  * This is a helper function for ext3_dx_readdir.  It calls filldir
 456  * for all entres on the fname linked list.  (Normally there is only
 457  * one entry on the linked list, unless there are 62 bit hash collisions.)
 458  */
 459 static int call_filldir(struct file * filp, void * dirent,
 460                         filldir_t filldir, struct fname *fname)
 461 {
 462         struct dir_private_info *info = filp->private_data;
 463         loff_t  curr_pos;
 464         struct inode *inode = filp->f_path.dentry->d_inode;
 465         struct super_block * sb;
 466         int error;
 467
 468         sb = inode->i_sb;
 469
 470         if (!fname) {
 471                 printk("call_filldir: called with null fname?!?\n");
 472                 return 0;
 473         }
 474         curr_pos = hash2pos(filp, fname->hash, fname->minor_hash);
 475         while (fname) {
 476                 error = filldir(dirent, fname->name,
 477                                 fname->name_len, curr_pos,
 478                                 fname->inode,
 479                                 get_dtype(sb, fname->file_type));
 480                 if (error) {
 481                         filp->f_pos = curr_pos;
 482                         info->extra_fname = fname;
 483                         return error;
 484                 }
 485                 fname = fname->next;
 486         }
 487         return 0;
 488 }
 489
 490 static int ext3_dx_readdir(struct file * filp,
 491                          void * dirent, filldir_t filldir)
 492 {
 493         struct dir_private_info *info = filp->private_data;
 494         struct inode *inode = filp->f_path.dentry->d_inode;
 495         struct fname *fname;
 496         int     ret;
 497
 498         if (!info) {
 499                 info = ext3_htree_create_dir_info(filp, filp->f_pos);
 500                 if (!info)
 501                         return -ENOMEM;
 502                 filp->private_data = info;
 503         }
 504
 505         if (filp->f_pos == ext3_get_htree_eof(filp))
 506                 return 0;       /* EOF */
 507
 508         /* Some one has messed with f_pos; reset the world */
 509         if (info->last_pos != filp->f_pos) {
 510                 free_rb_tree_fname(&info->root);
 511                 info->curr_node = NULL;
 512                 info->extra_fname = NULL;
 513                 info->curr_hash = pos2maj_hash(filp, filp->f_pos);
 514                 info->curr_minor_hash = pos2min_hash(filp, filp->f_pos);
 515         }
 516
 517         /*
 518          * If there are any leftover names on the hash collision
 519          * chain, return them first.
 520          */
 521         if (info->extra_fname) {
 522                 if (call_filldir(filp, dirent, filldir, info->extra_fname))
 523                         goto finished;
 524                 info->extra_fname = NULL;
 525                 goto next_node;
 526         } else if (!info->curr_node)
 527                 info->curr_node = rb_first(&info->root);
 528
 529         while (1) {
 530                 /*
 531                  * Fill the rbtree if we have no more entries,
 532                  * or the inode has changed since we last read in the
 533                  * cached entries.
 534                  */
 535                 if ((!info->curr_node) ||
 536                     (filp->f_version != inode->i_version)) {
 537                         info->curr_node = NULL;
 538                         free_rb_tree_fname(&info->root);
 539                         filp->f_version = inode->i_version;
 540                         ret = ext3_htree_fill_tree(filp, info->curr_hash,
 541                                                    info->curr_minor_hash,
 542                                                    &info->next_hash);
 543                         if (ret < 0)
 544                                 return ret;
 545                         if (ret == 0) {
 546                                 filp->f_pos = ext3_get_htree_eof(filp);
 547                                 break;
 548                         }
 549                         info->curr_node = rb_first(&info->root);
 550                 }
 551
 552                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
 553                 info->curr_hash = fname->hash;
 554                 info->curr_minor_hash = fname->minor_hash;
 555                 if (call_filldir(filp, dirent, filldir, fname))
 556                         break;
 557         next_node:
 558                 info->curr_node = rb_next(info->curr_node);
 559                 if (info->curr_node) {
 560                         fname = rb_entry(info->curr_node, struct fname,
 561                                          rb_hash);
 562                         info->curr_hash = fname->hash;
 563                         info->curr_minor_hash = fname->minor_hash;
 564                 } else {
 565                         if (info->next_hash == ~0) {
 566                                 filp->f_pos = ext3_get_htree_eof(filp);
 567                                 break;
 568                         }
 569                         info->curr_hash = info->next_hash;
 570                         info->curr_minor_hash = 0;
 571                 }
 572         }
 573 finished:
 574         info->last_pos = filp->f_pos;
 575         return 0;
 576 }
 577
 578 static int ext3_release_dir (struct inode * inode, struct file * filp)
 579 {
 580        if (filp->private_data)
 581                 ext3_htree_free_dir_info(filp->private_data);
 582
 583         return 0;
 584 }
 585
 586 const struct file_operations ext3_dir_operations = {
 587         .llseek         = ext3_dir_llseek,
 588         .read           = generic_read_dir,
 589         .readdir        = ext3_readdir,
 590         .unlocked_ioctl = ext3_ioctl,
 591 #ifdef CONFIG_COMPAT
 592         .compat_ioctl   = ext3_compat_ioctl,
 593 #endif
 594         .fsync          = ext3_sync_file,
 595         .release        = ext3_release_dir,
 596 };