fs/ext4/dir.c

   1 /*
   2  *  linux/fs/ext4/dir.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/dir.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  ext4 directory handling functions
  16  *
  17  *  Big-endian to little-endian byte-swapping/bitmaps by
  18  *        David S. Miller (davem@caip.rutgers.edu), 1995
  19  *
  20  * Hash Tree Directory indexing (c) 2001  Daniel Phillips
  21  *
  22  */
  23
  24 #include <linux/fs.h>
  25 #include <linux/jbd2.h>
  26 #include <linux/buffer_head.h>
  27 #include <linux/slab.h>
  28 #include <linux/rbtree.h>
  29 #include "ext4.h"
  30
  31 static unsigned char ext4_filetype_table[] = {
  32         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
  33 };
  34
  35 static int ext4_readdir(struct file *, void *, filldir_t);
  36 static int ext4_dx_readdir(struct file *filp,
  37                            void *dirent, filldir_t filldir);
  38 static int ext4_release_dir(struct inode *inode,
  39                                 struct file *filp);
  40
  41 const struct file_operations ext4_dir_operations = {
  42         .llseek         = ext4_llseek,
  43         .read           = generic_read_dir,
  44         .readdir        = ext4_readdir,         /* we take BKL. needed?*/
  45         .unlocked_ioctl = ext4_ioctl,
  46 #ifdef CONFIG_COMPAT
  47         .compat_ioctl   = ext4_compat_ioctl,
  48 #endif
  49         .fsync          = ext4_sync_file,
  50         .release        = ext4_release_dir,
  51 };
  52
  53
  54 static unsigned char get_dtype(struct super_block *sb, int filetype)
  55 {
  56         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
  57             (filetype >= EXT4_FT_MAX))
  58                 return DT_UNKNOWN;
  59
  60         return (ext4_filetype_table[filetype]);
  61 }
  62
  63 /*
  64  * Return 0 if the directory entry is OK, and 1 if there is a problem
  65  *
  66  * Note: this is the opposite of what ext2 and ext3 historically returned...
  67  */
  68 int __ext4_check_dir_entry(const char *function, unsigned int line,
  69                            struct inode *dir, struct file *filp,
  70                            struct ext4_dir_entry_2 *de,
  71                            struct buffer_head *bh,
  72                            unsigned int offset)
  73 {
  74         const char *error_msg = NULL;
  75         const int rlen = ext4_rec_len_from_disk(de->rec_len,
  76                                                 dir->i_sb->s_blocksize);
  77
  78         if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
  79                 error_msg = "rec_len is smaller than minimal";
  80         else if (unlikely(rlen % 4 != 0))
  81                 error_msg = "rec_len % 4 != 0";
  82         else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
  83                 error_msg = "rec_len is too small for name_len";
  84         else if (unlikely(((char *) de - bh->b_data) + rlen >
  85                           dir->i_sb->s_blocksize))
  86                 error_msg = "directory entry across blocks";
  87         else if (unlikely(le32_to_cpu(de->inode) >
  88                         le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
  89                 error_msg = "inode out of bounds";
  90         else
  91                 return 0;
  92
  93         if (filp)
  94                 ext4_error_file(filp, function, line, bh ? bh->b_blocknr : 0,
  95                                 "bad entry in directory: %s - offset=%u(%u), "
  96                                 "inode=%u, rec_len=%d, name_len=%d",
  97                                 error_msg, (unsigned) (offset%bh->b_size),
  98                                 offset, le32_to_cpu(de->inode),
  99                                 rlen, de->name_len);
 100         else
 101                 ext4_error_inode(dir, function, line, bh ? bh->b_blocknr : 0,
 102                                 "bad entry in directory: %s - offset=%u(%u), "
 103                                 "inode=%u, rec_len=%d, name_len=%d",
 104                                 error_msg, (unsigned) (offset%bh->b_size),
 105                                 offset, le32_to_cpu(de->inode),
 106                                 rlen, de->name_len);
 107
 108         return 1;
 109 }
 110
 111 static int ext4_readdir(struct file *filp,
 112                          void *dirent, filldir_t filldir)
 113 {
 114         int error = 0;
 115         unsigned int offset;
 116         int i, stored;
 117         struct ext4_dir_entry_2 *de;
 118         struct super_block *sb;
 119         int err;
 120         struct inode *inode = filp->f_path.dentry->d_inode;
 121         int ret = 0;
 122         int dir_has_error = 0;
 123
 124         sb = inode->i_sb;
 125
 126         if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
 127                                     EXT4_FEATURE_COMPAT_DIR_INDEX) &&
 128             ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
 129              ((inode->i_size >> sb->s_blocksize_bits) == 1))) {
 130                 err = ext4_dx_readdir(filp, dirent, filldir);
 131                 if (err != ERR_BAD_DX_DIR) {
 132                         ret = err;
 133                         goto out;
 134                 }
 135                 /*
 136                  * We don't set the inode dirty flag since it's not
 137                  * critical that it get flushed back to the disk.
 138                  */
 139                 ext4_clear_inode_flag(filp->f_path.dentry->d_inode,
 140                                       EXT4_INODE_INDEX);
 141         }
 142         stored = 0;
 143         offset = filp->f_pos & (sb->s_blocksize - 1);
 144
 145         while (!error && !stored && filp->f_pos < inode->i_size) {
 146                 struct ext4_map_blocks map;
 147                 struct buffer_head *bh = NULL;
 148
 149                 map.m_lblk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb);
 150                 map.m_len = 1;
 151                 err = ext4_map_blocks(NULL, inode, &map, 0);
 152                 if (err > 0) {
 153                         pgoff_t index = map.m_pblk >>
 154                                         (PAGE_CACHE_SHIFT - inode->i_blkbits);
 155                         if (!ra_has_index(&filp->f_ra, index))
 156                                 page_cache_sync_readahead(
 157                                         sb->s_bdev->bd_inode->i_mapping,
 158                                         &filp->f_ra, filp,
 159                                         index, 1);
 160                         filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
 161                         bh = ext4_bread(NULL, inode, map.m_lblk, 0, &err);
 162                 }
 163
 164                 /*
 165                  * We ignore I/O errors on directories so users have a chance
 166                  * of recovering data when there's a bad sector
 167                  */
 168                 if (!bh) {
 169                         if (!dir_has_error) {
 170                                 EXT4_ERROR_FILE(filp, 0,
 171                                                 "directory contains a "
 172                                                 "hole at offset %llu",
 173                                            (unsigned long long) filp->f_pos);
 174                                 dir_has_error = 1;
 175                         }
 176                         /* corrupt size?  Maybe no more blocks to read */
 177                         if (filp->f_pos > inode->i_blocks << 9)
 178                                 break;
 179                         filp->f_pos += sb->s_blocksize - offset;
 180                         continue;
 181                 }
 182
 183 revalidate:
 184                 /* If the dir block has changed since the last call to
 185                  * readdir(2), then we might be pointing to an invalid
 186                  * dirent right now.  Scan from the start of the block
 187                  * to make sure. */
 188                 if (filp->f_version != inode->i_version) {
 189                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
 190                                 de = (struct ext4_dir_entry_2 *)
 191                                         (bh->b_data + i);
 192                                 /* It's too expensive to do a full
 193                                  * dirent test each time round this
 194                                  * loop, but we do have to test at
 195                                  * least that it is non-zero.  A
 196                                  * failure will be detected in the
 197                                  * dirent test below. */
 198                                 if (ext4_rec_len_from_disk(de->rec_len,
 199                                         sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
 200                                         break;
 201                                 i += ext4_rec_len_from_disk(de->rec_len,
 202                                                             sb->s_blocksize);
 203                         }
 204                         offset = i;
 205                         filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
 206                                 | offset;
 207                         filp->f_version = inode->i_version;
 208                 }
 209
 210                 while (!error && filp->f_pos < inode->i_size
 211                        && offset < sb->s_blocksize) {
 212                         de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
 213                         if (ext4_check_dir_entry(inode, filp, de,
 214                                                  bh, offset)) {
 215                                 /*
 216                                  * On error, skip the f_pos to the next block
 217                                  */
 218                                 filp->f_pos = (filp->f_pos |
 219                                                 (sb->s_blocksize - 1)) + 1;
 220                                 brelse(bh);
 221                                 ret = stored;
 222                                 goto out;
 223                         }
 224                         offset += ext4_rec_len_from_disk(de->rec_len,
 225                                         sb->s_blocksize);
 226                         if (le32_to_cpu(de->inode)) {
 227                                 /* We might block in the next section
 228                                  * if the data destination is
 229                                  * currently swapped out.  So, use a
 230                                  * version stamp to detect whether or
 231                                  * not the directory has been modified
 232                                  * during the copy operation.
 233                                  */
 234                                 u64 version = filp->f_version;
 235
 236                                 error = filldir(dirent, de->name,
 237                                                 de->name_len,
 238                                                 filp->f_pos,
 239                                                 le32_to_cpu(de->inode),
 240                                                 get_dtype(sb, de->file_type));
 241                                 if (error)
 242                                         break;
 243                                 if (version != filp->f_version)
 244                                         goto revalidate;
 245                                 stored++;
 246                         }
 247                         filp->f_pos += ext4_rec_len_from_disk(de->rec_len,
 248                                                 sb->s_blocksize);
 249                 }
 250                 offset = 0;
 251                 brelse(bh);
 252         }
 253 out:
 254         return ret;
 255 }
 256
 257 /*
 258  * These functions convert from the major/minor hash to an f_pos
 259  * value.
 260  *
 261  * Currently we only use major hash numer.  This is unfortunate, but
 262  * on 32-bit machines, the same VFS interface is used for lseek and
 263  * llseek, so if we use the 64 bit offset, then the 32-bit versions of
 264  * lseek/telldir/seekdir will blow out spectacularly, and from within
 265  * the ext2 low-level routine, we don't know if we're being called by
 266  * a 64-bit version of the system call or the 32-bit version of the
 267  * system call.  Worse yet, NFSv2 only allows for a 32-bit readdir
 268  * cookie.  Sigh.
 269  */
 270 #define hash2pos(major, minor)  (major >> 1)
 271 #define pos2maj_hash(pos)       ((pos << 1) & 0xffffffff)
 272 #define pos2min_hash(pos)       (0)
 273
 274 /*
 275  * This structure holds the nodes of the red-black tree used to store
 276  * the directory entry in hash order.
 277  */
 278 struct fname {
 279         __u32           hash;
 280         __u32           minor_hash;
 281         struct rb_node  rb_hash;
 282         struct fname    *next;
 283         __u32           inode;
 284         __u8            name_len;
 285         __u8            file_type;
 286         char            name[0];
 287 };
 288
 289 /*
 290  * This functoin implements a non-recursive way of freeing all of the
 291  * nodes in the red-black tree.
 292  */
 293 static void free_rb_tree_fname(struct rb_root *root)
 294 {
 295         struct rb_node  *n = root->rb_node;
 296         struct rb_node  *parent;
 297         struct fname    *fname;
 298
 299         while (n) {
 300                 /* Do the node's children first */
 301                 if (n->rb_left) {
 302                         n = n->rb_left;
 303                         continue;
 304                 }
 305                 if (n->rb_right) {
 306                         n = n->rb_right;
 307                         continue;
 308                 }
 309                 /*
 310                  * The node has no children; free it, and then zero
 311                  * out parent's link to it.  Finally go to the
 312                  * beginning of the loop and try to free the parent
 313                  * node.
 314                  */
 315                 parent = rb_parent(n);
 316                 fname = rb_entry(n, struct fname, rb_hash);
 317                 while (fname) {
 318                         struct fname *old = fname;
 319                         fname = fname->next;
 320                         kfree(old);
 321                 }
 322                 if (!parent)
 323                         *root = RB_ROOT;
 324                 else if (parent->rb_left == n)
 325                         parent->rb_left = NULL;
 326                 else if (parent->rb_right == n)
 327                         parent->rb_right = NULL;
 328                 n = parent;
 329         }
 330 }
 331
 332
 333 static struct dir_private_info *ext4_htree_create_dir_info(loff_t pos)
 334 {
 335         struct dir_private_info *p;
 336
 337         p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
 338         if (!p)
 339                 return NULL;
 340         p->curr_hash = pos2maj_hash(pos);
 341         p->curr_minor_hash = pos2min_hash(pos);
 342         return p;
 343 }
 344
 345 void ext4_htree_free_dir_info(struct dir_private_info *p)
 346 {
 347         free_rb_tree_fname(&p->root);
 348         kfree(p);
 349 }
 350
 351 /*
 352  * Given a directory entry, enter it into the fname rb tree.
 353  */
 354 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
 355                              __u32 minor_hash,
 356                              struct ext4_dir_entry_2 *dirent)
 357 {
 358         struct rb_node **p, *parent = NULL;
 359         struct fname *fname, *new_fn;
 360         struct dir_private_info *info;
 361         int len;
 362
 363         info = dir_file->private_data;
 364         p = &info->root.rb_node;
 365
 366         /* Create and allocate the fname structure */
 367         len = sizeof(struct fname) + dirent->name_len + 1;
 368         new_fn = kzalloc(len, GFP_KERNEL);
 369         if (!new_fn)
 370                 return -ENOMEM;
 371         new_fn->hash = hash;
 372         new_fn->minor_hash = minor_hash;
 373         new_fn->inode = le32_to_cpu(dirent->inode);
 374         new_fn->name_len = dirent->name_len;
 375         new_fn->file_type = dirent->file_type;
 376         memcpy(new_fn->name, dirent->name, dirent->name_len);
 377         new_fn->name[dirent->name_len] = 0;
 378
 379         while (*p) {
 380                 parent = *p;
 381                 fname = rb_entry(parent, struct fname, rb_hash);
 382
 383                 /*
 384                  * If the hash and minor hash match up, then we put
 385                  * them on a linked list.  This rarely happens...
 386                  */
 387                 if ((new_fn->hash == fname->hash) &&
 388                     (new_fn->minor_hash == fname->minor_hash)) {
 389                         new_fn->next = fname->next;
 390                         fname->next = new_fn;
 391                         return 0;
 392                 }
 393
 394                 if (new_fn->hash < fname->hash)
 395                         p = &(*p)->rb_left;
 396                 else if (new_fn->hash > fname->hash)
 397                         p = &(*p)->rb_right;
 398                 else if (new_fn->minor_hash < fname->minor_hash)
 399                         p = &(*p)->rb_left;
 400                 else /* if (new_fn->minor_hash > fname->minor_hash) */
 401                         p = &(*p)->rb_right;
 402         }
 403
 404         rb_link_node(&new_fn->rb_hash, parent, p);
 405         rb_insert_color(&new_fn->rb_hash, &info->root);
 406         return 0;
 407 }
 408
 409
 410
 411 /*
 412  * This is a helper function for ext4_dx_readdir.  It calls filldir
 413  * for all entres on the fname linked list.  (Normally there is only
 414  * one entry on the linked list, unless there are 62 bit hash collisions.)
 415  */
 416 static int call_filldir(struct file *filp, void *dirent,
 417                         filldir_t filldir, struct fname *fname)
 418 {
 419         struct dir_private_info *info = filp->private_data;
 420         loff_t  curr_pos;
 421         struct inode *inode = filp->f_path.dentry->d_inode;
 422         struct super_block *sb;
 423         int error;
 424
 425         sb = inode->i_sb;
 426
 427         if (!fname) {
 428                 printk(KERN_ERR "EXT4-fs: call_filldir: called with "
 429                        "null fname?!?\n");
 430                 return 0;
 431         }
 432         curr_pos = hash2pos(fname->hash, fname->minor_hash);
 433         while (fname) {
 434                 error = filldir(dirent, fname->name,
 435                                 fname->name_len, curr_pos,
 436                                 fname->inode,
 437                                 get_dtype(sb, fname->file_type));
 438                 if (error) {
 439                         filp->f_pos = curr_pos;
 440                         info->extra_fname = fname;
 441                         return error;
 442                 }
 443                 fname = fname->next;
 444         }
 445         return 0;
 446 }
 447
 448 static int ext4_dx_readdir(struct file *filp,
 449                          void *dirent, filldir_t filldir)
 450 {
 451         struct dir_private_info *info = filp->private_data;
 452         struct inode *inode = filp->f_path.dentry->d_inode;
 453         struct fname *fname;
 454         int     ret;
 455
 456         if (!info) {
 457                 info = ext4_htree_create_dir_info(filp->f_pos);
 458                 if (!info)
 459                         return -ENOMEM;
 460                 filp->private_data = info;
 461         }
 462
 463         if (filp->f_pos == EXT4_HTREE_EOF)
 464                 return 0;       /* EOF */
 465
 466         /* Some one has messed with f_pos; reset the world */
 467         if (info->last_pos != filp->f_pos) {
 468                 free_rb_tree_fname(&info->root);
 469                 info->curr_node = NULL;
 470                 info->extra_fname = NULL;
 471                 info->curr_hash = pos2maj_hash(filp->f_pos);
 472                 info->curr_minor_hash = pos2min_hash(filp->f_pos);
 473         }
 474
 475         /*
 476          * If there are any leftover names on the hash collision
 477          * chain, return them first.
 478          */
 479         if (info->extra_fname) {
 480                 if (call_filldir(filp, dirent, filldir, info->extra_fname))
 481                         goto finished;
 482                 info->extra_fname = NULL;
 483                 goto next_node;
 484         } else if (!info->curr_node)
 485                 info->curr_node = rb_first(&info->root);
 486
 487         while (1) {
 488                 /*
 489                  * Fill the rbtree if we have no more entries,
 490                  * or the inode has changed since we last read in the
 491                  * cached entries.
 492                  */
 493                 if ((!info->curr_node) ||
 494                     (filp->f_version != inode->i_version)) {
 495                         info->curr_node = NULL;
 496                         free_rb_tree_fname(&info->root);
 497                         filp->f_version = inode->i_version;
 498                         ret = ext4_htree_fill_tree(filp, info->curr_hash,
 499                                                    info->curr_minor_hash,
 500                                                    &info->next_hash);
 501                         if (ret < 0)
 502                                 return ret;
 503                         if (ret == 0) {
 504                                 filp->f_pos = EXT4_HTREE_EOF;
 505                                 break;
 506                         }
 507                         info->curr_node = rb_first(&info->root);
 508                 }
 509
 510                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
 511                 info->curr_hash = fname->hash;
 512                 info->curr_minor_hash = fname->minor_hash;
 513                 if (call_filldir(filp, dirent, filldir, fname))
 514                         break;
 515         next_node:
 516                 info->curr_node = rb_next(info->curr_node);
 517                 if (info->curr_node) {
 518                         fname = rb_entry(info->curr_node, struct fname,
 519                                          rb_hash);
 520                         info->curr_hash = fname->hash;
 521                         info->curr_minor_hash = fname->minor_hash;
 522                 } else {
 523                         if (info->next_hash == ~0) {
 524                                 filp->f_pos = EXT4_HTREE_EOF;
 525                                 break;
 526                         }
 527                         info->curr_hash = info->next_hash;
 528                         info->curr_minor_hash = 0;
 529                 }
 530         }
 531 finished:
 532         info->last_pos = filp->f_pos;
 533         return 0;
 534 }
 535
 536 static int ext4_release_dir(struct inode *inode, struct file *filp)
 537 {
 538         if (filp->private_data)
 539                 ext4_htree_free_dir_info(filp->private_data);
 540
 541         return 0;
 542 }