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 "ext3.h"
  25
  26 static unsigned char ext3_filetype_table[] = {
  27         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
  28 };
  29
  30 static int ext3_readdir(struct file *, void *, filldir_t);
  31 static int ext3_dx_readdir(struct file * filp,
  32                            void * dirent, filldir_t filldir);
  33 static int ext3_release_dir (struct inode * inode,
  34                                 struct file * filp);
  35
  36 const struct file_operations ext3_dir_operations = {
  37         .llseek         = generic_file_llseek,
  38         .read           = generic_read_dir,
  39         .readdir        = ext3_readdir,         /* we take BKL. needed?*/
  40         .unlocked_ioctl = ext3_ioctl,
  41 #ifdef CONFIG_COMPAT
  42         .compat_ioctl   = ext3_compat_ioctl,
  43 #endif
  44         .fsync          = ext3_sync_file,       /* BKL held */
  45         .release        = ext3_release_dir,
  46 };
  47
  48
  49 static unsigned char get_dtype(struct super_block *sb, int filetype)
  50 {
  51         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE) ||
  52             (filetype >= EXT3_FT_MAX))
  53                 return DT_UNKNOWN;
  54
  55         return (ext3_filetype_table[filetype]);
  56 }
  57
  58
  59 int ext3_check_dir_entry (const char * function, struct inode * dir,
  60                           struct ext3_dir_entry_2 * de,
  61                           struct buffer_head * bh,
  62                           unsigned long offset)
  63 {
  64         const char * error_msg = NULL;
  65         const int rlen = ext3_rec_len_from_disk(de->rec_len);
  66
  67         if (unlikely(rlen < EXT3_DIR_REC_LEN(1)))
  68                 error_msg = "rec_len is smaller than minimal";
  69         else if (unlikely(rlen % 4 != 0))
  70                 error_msg = "rec_len % 4 != 0";
  71         else if (unlikely(rlen < EXT3_DIR_REC_LEN(de->name_len)))
  72                 error_msg = "rec_len is too small for name_len";
  73         else if (unlikely((((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)))
  74                 error_msg = "directory entry across blocks";
  75         else if (unlikely(le32_to_cpu(de->inode) >
  76                         le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count)))
  77                 error_msg = "inode out of bounds";
  78
  79         if (unlikely(error_msg != NULL))
  80                 ext3_error (dir->i_sb, function,
  81                         "bad entry in directory #%lu: %s - "
  82                         "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
  83                         dir->i_ino, error_msg, offset,
  84                         (unsigned long) le32_to_cpu(de->inode),
  85                         rlen, de->name_len);
  86
  87         return error_msg == NULL ? 1 : 0;
  88 }
  89
  90 static int ext3_readdir(struct file * filp,
  91                          void * dirent, filldir_t filldir)
  92 {
  93         int error = 0;
  94         unsigned long offset;
  95         int i, stored;
  96         struct ext3_dir_entry_2 *de;
  97         struct super_block *sb;
  98         int err;
  99         struct inode *inode = filp->f_path.dentry->d_inode;
 100         int ret = 0;
 101         int dir_has_error = 0;
 102
 103         sb = inode->i_sb;
 104
 105         if (EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
 106                                     EXT3_FEATURE_COMPAT_DIR_INDEX) &&
 107             ((EXT3_I(inode)->i_flags & EXT3_INDEX_FL) ||
 108              ((inode->i_size >> sb->s_blocksize_bits) == 1))) {
 109                 err = ext3_dx_readdir(filp, dirent, filldir);
 110                 if (err != ERR_BAD_DX_DIR) {
 111                         ret = err;
 112                         goto out;
 113                 }
 114                 /*
 115                  * We don't set the inode dirty flag since it's not
 116                  * critical that it get flushed back to the disk.
 117                  */
 118                 EXT3_I(filp->f_path.dentry->d_inode)->i_flags &= ~EXT3_INDEX_FL;
 119         }
 120         stored = 0;
 121         offset = filp->f_pos & (sb->s_blocksize - 1);
 122
 123         while (!error && !stored && filp->f_pos < inode->i_size) {
 124                 unsigned long blk = filp->f_pos >> EXT3_BLOCK_SIZE_BITS(sb);
 125                 struct buffer_head map_bh;
 126                 struct buffer_head *bh = NULL;
 127
 128                 map_bh.b_state = 0;
 129                 err = ext3_get_blocks_handle(NULL, inode, blk, 1, &map_bh, 0);
 130                 if (err > 0) {
 131                         pgoff_t index = map_bh.b_blocknr >>
 132                                         (PAGE_CACHE_SHIFT - inode->i_blkbits);
 133                         if (!ra_has_index(&filp->f_ra, index))
 134                                 page_cache_sync_readahead(
 135                                         sb->s_bdev->bd_inode->i_mapping,
 136                                         &filp->f_ra, filp,
 137                                         index, 1);
 138                         filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
 139                         bh = ext3_bread(NULL, inode, blk, 0, &err);
 140                 }
 141
 142                 /*
 143                  * We ignore I/O errors on directories so users have a chance
 144                  * of recovering data when there's a bad sector
 145                  */
 146                 if (!bh) {
 147                         if (!dir_has_error) {
 148                                 ext3_error(sb, __func__, "directory #%lu "
 149                                         "contains a hole at offset %lld",
 150                                         inode->i_ino, filp->f_pos);
 151                                 dir_has_error = 1;
 152                         }
 153                         /* corrupt size?  Maybe no more blocks to read */
 154                         if (filp->f_pos > inode->i_blocks << 9)
 155                                 break;
 156                         filp->f_pos += sb->s_blocksize - offset;
 157                         continue;
 158                 }
 159
 160 revalidate:
 161                 /* If the dir block has changed since the last call to
 162                  * readdir(2), then we might be pointing to an invalid
 163                  * dirent right now.  Scan from the start of the block
 164                  * to make sure. */
 165                 if (filp->f_version != inode->i_version) {
 166                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
 167                                 de = (struct ext3_dir_entry_2 *)
 168                                         (bh->b_data + i);
 169                                 /* It's too expensive to do a full
 170                                  * dirent test each time round this
 171                                  * loop, but we do have to test at
 172                                  * least that it is non-zero.  A
 173                                  * failure will be detected in the
 174                                  * dirent test below. */
 175                                 if (ext3_rec_len_from_disk(de->rec_len) <
 176                                                 EXT3_DIR_REC_LEN(1))
 177                                         break;
 178                                 i += ext3_rec_len_from_disk(de->rec_len);
 179                         }
 180                         offset = i;
 181                         filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
 182                                 | offset;
 183                         filp->f_version = inode->i_version;
 184                 }
 185
 186                 while (!error && filp->f_pos < inode->i_size
 187                        && offset < sb->s_blocksize) {
 188                         de = (struct ext3_dir_entry_2 *) (bh->b_data + offset);
 189                         if (!ext3_check_dir_entry ("ext3_readdir", inode, de,
 190                                                    bh, offset)) {
 191                                 /* On error, skip the f_pos to the
 192                                    next block. */
 193                                 filp->f_pos = (filp->f_pos |
 194                                                 (sb->s_blocksize - 1)) + 1;
 195                                 brelse (bh);
 196                                 ret = stored;
 197                                 goto out;
 198                         }
 199                         offset += ext3_rec_len_from_disk(de->rec_len);
 200                         if (le32_to_cpu(de->inode)) {
 201                                 /* We might block in the next section
 202                                  * if the data destination is
 203                                  * currently swapped out.  So, use a
 204                                  * version stamp to detect whether or
 205                                  * not the directory has been modified
 206                                  * during the copy operation.
 207                                  */
 208                                 u64 version = filp->f_version;
 209
 210                                 error = filldir(dirent, de->name,
 211                                                 de->name_len,
 212                                                 filp->f_pos,
 213                                                 le32_to_cpu(de->inode),
 214                                                 get_dtype(sb, de->file_type));
 215                                 if (error)
 216                                         break;
 217                                 if (version != filp->f_version)
 218                                         goto revalidate;
 219                                 stored ++;
 220                         }
 221                         filp->f_pos += ext3_rec_len_from_disk(de->rec_len);
 222                 }
 223                 offset = 0;
 224                 brelse (bh);
 225         }
 226 out:
 227         return ret;
 228 }
 229
 230 /*
 231  * These functions convert from the major/minor hash to an f_pos
 232  * value.
 233  *
 234  * Currently we only use major hash numer.  This is unfortunate, but
 235  * on 32-bit machines, the same VFS interface is used for lseek and
 236  * llseek, so if we use the 64 bit offset, then the 32-bit versions of
 237  * lseek/telldir/seekdir will blow out spectacularly, and from within
 238  * the ext2 low-level routine, we don't know if we're being called by
 239  * a 64-bit version of the system call or the 32-bit version of the
 240  * system call.  Worse yet, NFSv2 only allows for a 32-bit readdir
 241  * cookie.  Sigh.
 242  */
 243 #define hash2pos(major, minor)  (major >> 1)
 244 #define pos2maj_hash(pos)       ((pos << 1) & 0xffffffff)
 245 #define pos2min_hash(pos)       (0)
 246
 247 /*
 248  * This structure holds the nodes of the red-black tree used to store
 249  * the directory entry in hash order.
 250  */
 251 struct fname {
 252         __u32           hash;
 253         __u32           minor_hash;
 254         struct rb_node  rb_hash;
 255         struct fname    *next;
 256         __u32           inode;
 257         __u8            name_len;
 258         __u8            file_type;
 259         char            name[0];
 260 };
 261
 262 /*
 263  * This functoin implements a non-recursive way of freeing all of the
 264  * nodes in the red-black tree.
 265  */
 266 static void free_rb_tree_fname(struct rb_root *root)
 267 {
 268         struct rb_node  *n = root->rb_node;
 269         struct rb_node  *parent;
 270         struct fname    *fname;
 271
 272         while (n) {
 273                 /* Do the node's children first */
 274                 if (n->rb_left) {
 275                         n = n->rb_left;
 276                         continue;
 277                 }
 278                 if (n->rb_right) {
 279                         n = n->rb_right;
 280                         continue;
 281                 }
 282                 /*
 283                  * The node has no children; free it, and then zero
 284                  * out parent's link to it.  Finally go to the
 285                  * beginning of the loop and try to free the parent
 286                  * node.
 287                  */
 288                 parent = rb_parent(n);
 289                 fname = rb_entry(n, struct fname, rb_hash);
 290                 while (fname) {
 291                         struct fname * old = fname;
 292                         fname = fname->next;
 293                         kfree (old);
 294                 }
 295                 if (!parent)
 296                         *root = RB_ROOT;
 297                 else if (parent->rb_left == n)
 298                         parent->rb_left = NULL;
 299                 else if (parent->rb_right == n)
 300                         parent->rb_right = NULL;
 301                 n = parent;
 302         }
 303 }
 304
 305
 306 static struct dir_private_info *ext3_htree_create_dir_info(loff_t pos)
 307 {
 308         struct dir_private_info *p;
 309
 310         p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
 311         if (!p)
 312                 return NULL;
 313         p->curr_hash = pos2maj_hash(pos);
 314         p->curr_minor_hash = pos2min_hash(pos);
 315         return p;
 316 }
 317
 318 void ext3_htree_free_dir_info(struct dir_private_info *p)
 319 {
 320         free_rb_tree_fname(&p->root);
 321         kfree(p);
 322 }
 323
 324 /*
 325  * Given a directory entry, enter it into the fname rb tree.
 326  */
 327 int ext3_htree_store_dirent(struct file *dir_file, __u32 hash,
 328                              __u32 minor_hash,
 329                              struct ext3_dir_entry_2 *dirent)
 330 {
 331         struct rb_node **p, *parent = NULL;
 332         struct fname * fname, *new_fn;
 333         struct dir_private_info *info;
 334         int len;
 335
 336         info = (struct dir_private_info *) dir_file->private_data;
 337         p = &info->root.rb_node;
 338
 339         /* Create and allocate the fname structure */
 340         len = sizeof(struct fname) + dirent->name_len + 1;
 341         new_fn = kzalloc(len, GFP_KERNEL);
 342         if (!new_fn)
 343                 return -ENOMEM;
 344         new_fn->hash = hash;
 345         new_fn->minor_hash = minor_hash;
 346         new_fn->inode = le32_to_cpu(dirent->inode);
 347         new_fn->name_len = dirent->name_len;
 348         new_fn->file_type = dirent->file_type;
 349         memcpy(new_fn->name, dirent->name, dirent->name_len);
 350         new_fn->name[dirent->name_len] = 0;
 351
 352         while (*p) {
 353                 parent = *p;
 354                 fname = rb_entry(parent, struct fname, rb_hash);
 355
 356                 /*
 357                  * If the hash and minor hash match up, then we put
 358                  * them on a linked list.  This rarely happens...
 359                  */
 360                 if ((new_fn->hash == fname->hash) &&
 361                     (new_fn->minor_hash == fname->minor_hash)) {
 362                         new_fn->next = fname->next;
 363                         fname->next = new_fn;
 364                         return 0;
 365                 }
 366
 367                 if (new_fn->hash < fname->hash)
 368                         p = &(*p)->rb_left;
 369                 else if (new_fn->hash > fname->hash)
 370                         p = &(*p)->rb_right;
 371                 else if (new_fn->minor_hash < fname->minor_hash)
 372                         p = &(*p)->rb_left;
 373                 else /* if (new_fn->minor_hash > fname->minor_hash) */
 374                         p = &(*p)->rb_right;
 375         }
 376
 377         rb_link_node(&new_fn->rb_hash, parent, p);
 378         rb_insert_color(&new_fn->rb_hash, &info->root);
 379         return 0;
 380 }
 381
 382
 383
 384 /*
 385  * This is a helper function for ext3_dx_readdir.  It calls filldir
 386  * for all entres on the fname linked list.  (Normally there is only
 387  * one entry on the linked list, unless there are 62 bit hash collisions.)
 388  */
 389 static int call_filldir(struct file * filp, void * dirent,
 390                         filldir_t filldir, struct fname *fname)
 391 {
 392         struct dir_private_info *info = filp->private_data;
 393         loff_t  curr_pos;
 394         struct inode *inode = filp->f_path.dentry->d_inode;
 395         struct super_block * sb;
 396         int error;
 397
 398         sb = inode->i_sb;
 399
 400         if (!fname) {
 401                 printk("call_filldir: called with null fname?!?\n");
 402                 return 0;
 403         }
 404         curr_pos = hash2pos(fname->hash, fname->minor_hash);
 405         while (fname) {
 406                 error = filldir(dirent, fname->name,
 407                                 fname->name_len, curr_pos,
 408                                 fname->inode,
 409                                 get_dtype(sb, fname->file_type));
 410                 if (error) {
 411                         filp->f_pos = curr_pos;
 412                         info->extra_fname = fname;
 413                         return error;
 414                 }
 415                 fname = fname->next;
 416         }
 417         return 0;
 418 }
 419
 420 static int ext3_dx_readdir(struct file * filp,
 421                          void * dirent, filldir_t filldir)
 422 {
 423         struct dir_private_info *info = filp->private_data;
 424         struct inode *inode = filp->f_path.dentry->d_inode;
 425         struct fname *fname;
 426         int     ret;
 427
 428         if (!info) {
 429                 info = ext3_htree_create_dir_info(filp->f_pos);
 430                 if (!info)
 431                         return -ENOMEM;
 432                 filp->private_data = info;
 433         }
 434
 435         if (filp->f_pos == EXT3_HTREE_EOF)
 436                 return 0;       /* EOF */
 437
 438         /* Some one has messed with f_pos; reset the world */
 439         if (info->last_pos != filp->f_pos) {
 440                 free_rb_tree_fname(&info->root);
 441                 info->curr_node = NULL;
 442                 info->extra_fname = NULL;
 443                 info->curr_hash = pos2maj_hash(filp->f_pos);
 444                 info->curr_minor_hash = pos2min_hash(filp->f_pos);
 445         }
 446
 447         /*
 448          * If there are any leftover names on the hash collision
 449          * chain, return them first.
 450          */
 451         if (info->extra_fname) {
 452                 if (call_filldir(filp, dirent, filldir, info->extra_fname))
 453                         goto finished;
 454                 info->extra_fname = NULL;
 455                 goto next_node;
 456         } else if (!info->curr_node)
 457                 info->curr_node = rb_first(&info->root);
 458
 459         while (1) {
 460                 /*
 461                  * Fill the rbtree if we have no more entries,
 462                  * or the inode has changed since we last read in the
 463                  * cached entries.
 464                  */
 465                 if ((!info->curr_node) ||
 466                     (filp->f_version != inode->i_version)) {
 467                         info->curr_node = NULL;
 468                         free_rb_tree_fname(&info->root);
 469                         filp->f_version = inode->i_version;
 470                         ret = ext3_htree_fill_tree(filp, info->curr_hash,
 471                                                    info->curr_minor_hash,
 472                                                    &info->next_hash);
 473                         if (ret < 0)
 474                                 return ret;
 475                         if (ret == 0) {
 476                                 filp->f_pos = EXT3_HTREE_EOF;
 477                                 break;
 478                         }
 479                         info->curr_node = rb_first(&info->root);
 480                 }
 481
 482                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
 483                 info->curr_hash = fname->hash;
 484                 info->curr_minor_hash = fname->minor_hash;
 485                 if (call_filldir(filp, dirent, filldir, fname))
 486                         break;
 487         next_node:
 488                 info->curr_node = rb_next(info->curr_node);
 489                 if (info->curr_node) {
 490                         fname = rb_entry(info->curr_node, struct fname,
 491                                          rb_hash);
 492                         info->curr_hash = fname->hash;
 493                         info->curr_minor_hash = fname->minor_hash;
 494                 } else {
 495                         if (info->next_hash == ~0) {
 496                                 filp->f_pos = EXT3_HTREE_EOF;
 497                                 break;
 498                         }
 499                         info->curr_hash = info->next_hash;
 500                         info->curr_minor_hash = 0;
 501                 }
 502         }
 503 finished:
 504         info->last_pos = filp->f_pos;
 505         return 0;
 506 }
 507
 508 static int ext3_release_dir (struct inode * inode, struct file * filp)
 509 {
 510        if (filp->private_data)
 511                 ext3_htree_free_dir_info(filp->private_data);
 512
 513         return 0;
 514 }