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