fs/ext3/dir.c

   1 /*
   2  *  linux/fs/ext3/dir.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   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 ([email protected]), 1995
  19  *
  20  * Hash Tree Directory indexing (c) 2001  Daniel Phillips
  21  *
  22  */
  23
  24 #include <linux/compat.h>
  25 #include "ext3.h"
  26
  27 static unsigned char ext3_filetype_table[] = {
  28         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
  29 };
  30
  31 static int ext3_dx_readdir(struct file * filp,
  32                            void * dirent, filldir_t filldir);
  33
  34 static unsigned char get_dtype(struct super_block *sb, int filetype)
  35 {
  36         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE) ||
  37             (filetype >= EXT3_FT_MAX))
  38                 return DT_UNKNOWN;
  39
  40         return (ext3_filetype_table[filetype]);
  41 }
  42
  43 /**
  44  * Check if the given dir-inode refers to an htree-indexed directory
  45  * (or a directory which chould potentially get coverted to use htree
  46  * indexing).
  47  *
  48  * Return 1 if it is a dx dir, 0 if not
  49  */
  50 static int is_dx_dir(struct inode *inode)
  51 {
  52         struct super_block *sb = inode->i_sb;
  53
  54         if (EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
  55                      EXT3_FEATURE_COMPAT_DIR_INDEX) &&
  56             ((EXT3_I(inode)->i_flags & EXT3_INDEX_FL) ||
  57              ((inode->i_size >> sb->s_blocksize_bits) == 1)))
  58                 return 1;
  59
  60         return 0;
  61 }
  62
  63 int ext3_check_dir_entry (const char * function, struct inode * dir,
  64                           struct ext3_dir_entry_2 * de,
  65                           struct buffer_head * bh,
  66                           unsigned long offset)
  67 {
  68         const char * error_msg = NULL;
  69         const int rlen = ext3_rec_len_from_disk(de->rec_len);
  70
  71         if (unlikely(rlen < EXT3_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 < EXT3_DIR_REC_LEN(de->name_len)))
  76                 error_msg = "rec_len is too small for name_len";
  77         else if (unlikely((((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)))
  78                 error_msg = "directory entry across blocks";
  79         else if (unlikely(le32_to_cpu(de->inode) >
  80                         le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count)))
  81                 error_msg = "inode out of bounds";
  82
  83         if (unlikely(error_msg != NULL))
  84                 ext3_error (dir->i_sb, function,
  85                         "bad entry in directory #%lu: %s - "
  86                         "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
  87                         dir->i_ino, error_msg, offset,
  88                         (unsigned long) le32_to_cpu(de->inode),
  89                         rlen, de->name_len);
  90
  91         return error_msg == NULL ? 1 : 0;
  92 }
  93
  94 static int ext3_readdir(struct file * filp,
  95                          void * dirent, filldir_t filldir)
  96 {
  97         int error = 0;
  98         unsigned long offset;
  99         int i, stored;
 100         struct ext3_dir_entry_2 *de;
 101         int err;
 102         struct inode *inode = filp->f_path.dentry->d_inode;
 103         struct super_block *sb = inode->i_sb;
 104         int ret = 0;
 105         int dir_has_error = 0;
 106
 107         if (is_dx_dir(inode)) {
 108                 err = ext3_dx_readdir(filp, dirent, filldir);
 109                 if (err != ERR_BAD_DX_DIR) {
 110                         ret = err;
 111                         goto out;
 112                 }
 113                 /*
 114                  * We don't set the inode dirty flag since it's not
 115                  * critical that it get flushed back to the disk.
 116                  */
 117                 EXT3_I(filp->f_path.dentry->d_inode)->i_flags &= ~EXT3_INDEX_FL;
 118         }
 119         stored = 0;
 120         offset = filp->f_pos & (sb->s_blocksize - 1);
 121
 122         while (!error && !stored && filp->f_pos < inode->i_size) {
 123                 unsigned long blk = filp->f_pos >> EXT3_BLOCK_SIZE_BITS(sb);
 124                 struct buffer_head map_bh;
 125                 struct buffer_head *bh = NULL;
 126
 127                 map_bh.b_state = 0;
 128                 err = ext3_get_blocks_handle(NULL, inode, blk, 1, &map_bh, 0);
 129                 if (err > 0) {
 130                         pgoff_t index = map_bh.b_blocknr >>
 131                                         (PAGE_CACHE_SHIFT - inode->i_blkbits);
 132                         if (!ra_has_index(&filp->f_ra, index))
 133                                 page_cache_sync_readahead(
 134                                         sb->s_bdev->bd_inode->i_mapping,
 135                                         &filp->f_ra, filp,
 136                                         index, 1);
 137                         filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
 138                         bh = ext3_bread(NULL, inode, blk, 0, &err);
 139                 }
 140
 141                 /*
 142                  * We ignore I/O errors on directories so users have a chance
 143                  * of recovering data when there's a bad sector
 144                  */
 145                 if (!bh) {
 146                         if (!dir_has_error) {
 147                                 ext3_error(sb, __func__, "directory #%lu "
 148                                         "contains a hole at offset %lld",
 149                                         inode->i_ino, filp->f_pos);
 150                                 dir_has_error = 1;
 151                         }
 152                         /* corrupt size?  Maybe no more blocks to read */
 153                         if (filp->f_pos > inode->i_blocks << 9)
 154                                 break;
 155                         filp->f_pos += sb->s_blocksize - offset;
 156                         continue;
 157                 }
 158
 159 revalidate:
 160                 /* If the dir block has changed since the last call to
 161                  * readdir(2), then we might be pointing to an invalid
 162                  * dirent right now.  Scan from the start of the block
 163                  * to make sure. */
 164                 if (filp->f_version != inode->i_version) {
 165                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
 166                                 de = (struct ext3_dir_entry_2 *)
 167                                         (bh->b_data + i);
 168                                 /* It's too expensive to do a full
 169                                  * dirent test each time round this
 170                                  * loop, but we do have to test at
 171                                  * least that it is non-zero.  A
 172                                  * failure will be detected in the
 173                                  * dirent test below. */
 174                                 if (ext3_rec_len_from_disk(de->rec_len) <
 175                                                 EXT3_DIR_REC_LEN(1))
 176                                         break;
 177                                 i += ext3_rec_len_from_disk(de->rec_len);
 178                         }
 179                         offset = i;
 180                         filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
 181                                 | offset;
 182                         filp->f_version = inode->i_version;
 183                 }
 184
 185                 while (!error && filp->f_pos < inode->i_size
 186                        && offset < sb->s_blocksize) {
 187                         de = (struct ext3_dir_entry_2 *) (bh->b_data + offset);
 188                         if (!ext3_check_dir_entry ("ext3_readdir", inode, de,
 189                                                    bh, offset)) {
 190                                 /* On error, skip the f_pos to the
 191                                    next block. */
 192                                 filp->f_pos = (filp->f_pos |
 193                                                 (sb->s_blocksize - 1)) + 1;
 194                                 brelse (bh);
 195                                 ret = stored;
 196                                 goto out;
 197                         }
 198                         offset += ext3_rec_len_from_disk(de->rec_len);
 199                         if (le32_to_cpu(de->inode)) {
 200                                 /* We might block in the next section
 201                                  * if the data destination is
 202                                  * currently swapped out.  So, use a
 203                                  * version stamp to detect whether or
 204                                  * not the directory has been modified
 205                                  * during the copy operation.
 206                                  */
 207                                 u64 version = filp->f_version;
 208
 209                                 error = filldir(dirent, de->name,
 210                                                 de->name_len,
 211                                                 filp->f_pos,
 212                                                 le32_to_cpu(de->inode),
 213                                                 get_dtype(sb, de->file_type));
 214                                 if (error)
 215                                         break;
 216                                 if (version != filp->f_version)
 217                                         goto revalidate;
 218                                 stored ++;
 219                         }
 220                         filp->f_pos += ext3_rec_len_from_disk(de->rec_len);
 221                 }
 222                 offset = 0;
 223                 brelse (bh);
 224         }
 225 out:
 226         return ret;
 227 }
 228
 229 static inline int is_32bit_api(void)
 230 {
 231 #ifdef CONFIG_COMPAT
 232         return is_compat_task();
 233 #else
 234         return (BITS_PER_LONG == 32);
 235 #endif
 236 }
 237
 238 /*
 239  * These functions convert from the major/minor hash to an f_pos
 240  * value for dx directories
 241  *
 242  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
 243  * FMODE_64BITHASH explicitly. On the other hand, we allow ext3 to be mounted
 244  * directly on both 32-bit and 64-bit nodes, under such case, neither
 245  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
 246  */
 247 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
 248 {
 249         if ((filp->f_mode & FMODE_32BITHASH) ||
 250             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 251                 return major >> 1;
 252         else
 253                 return ((__u64)(major >> 1) << 32) | (__u64)minor;
 254 }
 255
 256 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
 257 {
 258         if ((filp->f_mode & FMODE_32BITHASH) ||
 259             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 260                 return (pos << 1) & 0xffffffff;
 261         else
 262                 return ((pos >> 32) << 1) & 0xffffffff;
 263 }
 264
 265 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
 266 {
 267         if ((filp->f_mode & FMODE_32BITHASH) ||
 268             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 269                 return 0;
 270         else
 271                 return pos & 0xffffffff;
 272 }
 273
 274 /*
 275  * Return 32- or 64-bit end-of-file for dx directories
 276  */
 277 static inline loff_t ext3_get_htree_eof(struct file *filp)
 278 {
 279         if ((filp->f_mode & FMODE_32BITHASH) ||
 280             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 281                 return EXT3_HTREE_EOF_32BIT;
 282         else
 283                 return EXT3_HTREE_EOF_64BIT;
 284 }
 285
 286
 287 /*
 288  * ext3_dir_llseek() calls generic_file_llseek[_size]() to handle both
 289  * non-htree and htree directories, where the "offset" is in terms
 290  * of the filename hash value instead of the byte offset.
 291  *
 292  * Because we may return a 64-bit hash that is well beyond s_maxbytes,
 293  * we need to pass the max hash as the maximum allowable offset in
 294  * the htree directory case.
 295  *
 296  * NOTE: offsets obtained *before* ext3_set_inode_flag(dir, EXT3_INODE_INDEX)
 297  *       will be invalid once the directory was converted into a dx directory
 298  */
 299 loff_t ext3_dir_llseek(struct file *file, loff_t offset, int origin)
 300 {
 301         struct inode *inode = file->f_mapping->host;
 302         int dx_dir = is_dx_dir(inode);
 303         loff_t htree_max = ext3_get_htree_eof(file);
 304
 305         if (likely(dx_dir))
 306                 return generic_file_llseek_size(file, offset, origin,
 307                                                 htree_max, htree_max);
 308         else
 309                 return generic_file_llseek(file, offset, origin);
 310 }
 311
 312 /*
 313  * This structure holds the nodes of the red-black tree used to store
 314  * the directory entry in hash order.
 315  */
 316 struct fname {
 317         __u32           hash;
 318         __u32           minor_hash;
 319         struct rb_node  rb_hash;
 320         struct fname    *next;
 321         __u32           inode;
 322         __u8            name_len;
 323         __u8            file_type;
 324         char            name[0];
 325 };
 326
 327 /*
 328  * This functoin implements a non-recursive way of freeing all of the
 329  * nodes in the red-black tree.
 330  */
 331 static void free_rb_tree_fname(struct rb_root *root)
 332 {
 333         struct rb_node  *n = root->rb_node;
 334         struct rb_node  *parent;
 335         struct fname    *fname;
 336
 337         while (n) {
 338                 /* Do the node's children first */
 339                 if (n->rb_left) {
 340                         n = n->rb_left;
 341                         continue;
 342                 }
 343                 if (n->rb_right) {
 344                         n = n->rb_right;
 345                         continue;
 346                 }
 347                 /*
 348                  * The node has no children; free it, and then zero
 349                  * out parent's link to it.  Finally go to the
 350                  * beginning of the loop and try to free the parent
 351                  * node.
 352                  */
 353                 parent = rb_parent(n);
 354                 fname = rb_entry(n, struct fname, rb_hash);
 355                 while (fname) {
 356                         struct fname * old = fname;
 357                         fname = fname->next;
 358                         kfree (old);
 359                 }
 360                 if (!parent)
 361                         *root = RB_ROOT;
 362                 else if (parent->rb_left == n)
 363                         parent->rb_left = NULL;
 364                 else if (parent->rb_right == n)
 365                         parent->rb_right = NULL;
 366                 n = parent;
 367         }
 368 }
 369
 370
 371 static struct dir_private_info *ext3_htree_create_dir_info(struct file *filp,
 372                                                            loff_t pos)
 373 {
 374         struct dir_private_info *p;
 375
 376         p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
 377         if (!p)
 378                 return NULL;
 379         p->curr_hash = pos2maj_hash(filp, pos);
 380         p->curr_minor_hash = pos2min_hash(filp, pos);
 381         return p;
 382 }
 383
 384 void ext3_htree_free_dir_info(struct dir_private_info *p)
 385 {
 386         free_rb_tree_fname(&p->root);
 387         kfree(p);
 388 }
 389
 390 /*
 391  * Given a directory entry, enter it into the fname rb tree.
 392  */
 393 int ext3_htree_store_dirent(struct file *dir_file, __u32 hash,
 394                              __u32 minor_hash,
 395                              struct ext3_dir_entry_2 *dirent)
 396 {
 397         struct rb_node **p, *parent = NULL;
 398         struct fname * fname, *new_fn;
 399         struct dir_private_info *info;
 400         int len;
 401
 402         info = (struct dir_private_info *) dir_file->private_data;
 403         p = &info->root.rb_node;
 404
 405         /* Create and allocate the fname structure */
 406         len = sizeof(struct fname) + dirent->name_len + 1;
 407         new_fn = kzalloc(len, GFP_KERNEL);
 408         if (!new_fn)
 409                 return -ENOMEM;
 410         new_fn->hash = hash;
 411         new_fn->minor_hash = minor_hash;
 412         new_fn->inode = le32_to_cpu(dirent->inode);
 413         new_fn->name_len = dirent->name_len;
 414         new_fn->file_type = dirent->file_type;
 415         memcpy(new_fn->name, dirent->name, dirent->name_len);
 416         new_fn->name[dirent->name_len] = 0;
 417
 418         while (*p) {
 419                 parent = *p;
 420                 fname = rb_entry(parent, struct fname, rb_hash);
 421
 422                 /*
 423                  * If the hash and minor hash match up, then we put
 424                  * them on a linked list.  This rarely happens...
 425                  */
 426                 if ((new_fn->hash == fname->hash) &&
 427                     (new_fn->minor_hash == fname->minor_hash)) {
 428                         new_fn->next = fname->next;
 429                         fname->next = new_fn;
 430                         return 0;
 431                 }
 432
 433                 if (new_fn->hash < fname->hash)
 434                         p = &(*p)->rb_left;
 435                 else if (new_fn->hash > fname->hash)
 436                         p = &(*p)->rb_right;
 437                 else if (new_fn->minor_hash < fname->minor_hash)
 438                         p = &(*p)->rb_left;
 439                 else /* if (new_fn->minor_hash > fname->minor_hash) */
 440                         p = &(*p)->rb_right;
 441         }
 442
 443         rb_link_node(&new_fn->rb_hash, parent, p);
 444         rb_insert_color(&new_fn->rb_hash, &info->root);
 445         return 0;
 446 }
 447
 448
 449
 450 /*
 451  * This is a helper function for ext3_dx_readdir.  It calls filldir
 452  * for all entres on the fname linked list.  (Normally there is only
 453  * one entry on the linked list, unless there are 62 bit hash collisions.)
 454  */
 455 static int call_filldir(struct file * filp, void * dirent,
 456                         filldir_t filldir, struct fname *fname)
 457 {
 458         struct dir_private_info *info = filp->private_data;
 459         loff_t  curr_pos;
 460         struct inode *inode = filp->f_path.dentry->d_inode;
 461         struct super_block * sb;
 462         int error;
 463
 464         sb = inode->i_sb;
 465
 466         if (!fname) {
 467                 printk("call_filldir: called with null fname?!?\n");
 468                 return 0;
 469         }
 470         curr_pos = hash2pos(filp, fname->hash, fname->minor_hash);
 471         while (fname) {
 472                 error = filldir(dirent, fname->name,
 473                                 fname->name_len, curr_pos,
 474                                 fname->inode,
 475                                 get_dtype(sb, fname->file_type));
 476                 if (error) {
 477                         filp->f_pos = curr_pos;
 478                         info->extra_fname = fname;
 479                         return error;
 480                 }
 481                 fname = fname->next;
 482         }
 483         return 0;
 484 }
 485
 486 static int ext3_dx_readdir(struct file * filp,
 487                          void * dirent, filldir_t filldir)
 488 {
 489         struct dir_private_info *info = filp->private_data;
 490         struct inode *inode = filp->f_path.dentry->d_inode;
 491         struct fname *fname;
 492         int     ret;
 493
 494         if (!info) {
 495                 info = ext3_htree_create_dir_info(filp, filp->f_pos);
 496                 if (!info)
 497                         return -ENOMEM;
 498                 filp->private_data = info;
 499         }
 500
 501         if (filp->f_pos == ext3_get_htree_eof(filp))
 502                 return 0;       /* EOF */
 503
 504         /* Some one has messed with f_pos; reset the world */
 505         if (info->last_pos != filp->f_pos) {
 506                 free_rb_tree_fname(&info->root);
 507                 info->curr_node = NULL;
 508                 info->extra_fname = NULL;
 509                 info->curr_hash = pos2maj_hash(filp, filp->f_pos);
 510                 info->curr_minor_hash = pos2min_hash(filp, filp->f_pos);
 511         }
 512
 513         /*
 514          * If there are any leftover names on the hash collision
 515          * chain, return them first.
 516          */
 517         if (info->extra_fname) {
 518                 if (call_filldir(filp, dirent, filldir, info->extra_fname))
 519                         goto finished;
 520                 info->extra_fname = NULL;
 521                 goto next_node;
 522         } else if (!info->curr_node)
 523                 info->curr_node = rb_first(&info->root);
 524
 525         while (1) {
 526                 /*
 527                  * Fill the rbtree if we have no more entries,
 528                  * or the inode has changed since we last read in the
 529                  * cached entries.
 530                  */
 531                 if ((!info->curr_node) ||
 532                     (filp->f_version != inode->i_version)) {
 533                         info->curr_node = NULL;
 534                         free_rb_tree_fname(&info->root);
 535                         filp->f_version = inode->i_version;
 536                         ret = ext3_htree_fill_tree(filp, info->curr_hash,
 537                                                    info->curr_minor_hash,
 538                                                    &info->next_hash);
 539                         if (ret < 0)
 540                                 return ret;
 541                         if (ret == 0) {
 542                                 filp->f_pos = ext3_get_htree_eof(filp);
 543                                 break;
 544                         }
 545                         info->curr_node = rb_first(&info->root);
 546                 }
 547
 548                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
 549                 info->curr_hash = fname->hash;
 550                 info->curr_minor_hash = fname->minor_hash;
 551                 if (call_filldir(filp, dirent, filldir, fname))
 552                         break;
 553         next_node:
 554                 info->curr_node = rb_next(info->curr_node);
 555                 if (info->curr_node) {
 556                         fname = rb_entry(info->curr_node, struct fname,
 557                                          rb_hash);
 558                         info->curr_hash = fname->hash;
 559                         info->curr_minor_hash = fname->minor_hash;
 560                 } else {
 561                         if (info->next_hash == ~0) {
 562                                 filp->f_pos = ext3_get_htree_eof(filp);
 563                                 break;
 564                         }
 565                         info->curr_hash = info->next_hash;
 566                         info->curr_minor_hash = 0;
 567                 }
 568         }
 569 finished:
 570         info->last_pos = filp->f_pos;
 571         return 0;
 572 }
 573
 574 static int ext3_release_dir (struct inode * inode, struct file * filp)
 575 {
 576        if (filp->private_data)
 577                 ext3_htree_free_dir_info(filp->private_data);
 578
 579         return 0;
 580 }
 581
 582 const struct file_operations ext3_dir_operations = {
 583         .llseek         = ext3_dir_llseek,
 584         .read           = generic_read_dir,
 585         .readdir        = ext3_readdir,
 586         .unlocked_ioctl = ext3_ioctl,
 587 #ifdef CONFIG_COMPAT
 588         .compat_ioctl   = ext3_compat_ioctl,
 589 #endif
 590         .fsync          = ext3_sync_file,
 591         .release        = ext3_release_dir,
 592 };