fs/ext4/dir.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/dir.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/dir.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  ext4 directory handling functions
  17  *
  18  *  Big-endian to little-endian byte-swapping/bitmaps by
  19  *        David S. Miller (davem@caip.rutgers.edu), 1995
  20  *
  21  * Hash Tree Directory indexing (c) 2001  Daniel Phillips
  22  *
  23  */
  24
  25 #include <linux/fs.h>
  26 #include <linux/buffer_head.h>
  27 #include <linux/slab.h>
  28 #include <linux/iversion.h>
  29 #include <linux/unicode.h>
  30 #include "ext4.h"
  31 #include "xattr.h"
  32
  33 static int ext4_dx_readdir(struct file *, struct dir_context *);
  34
  35 /**
  36  * is_dx_dir() - check if a directory is using htree indexing
  37  * @inode: directory inode
  38  *
  39  * Check if the given dir-inode refers to an htree-indexed directory
  40  * (or a directory which could potentially get converted to use htree
  41  * indexing).
  42  *
  43  * Return 1 if it is a dx dir, 0 if not
  44  */
  45 static int is_dx_dir(struct inode *inode)
  46 {
  47         struct super_block *sb = inode->i_sb;
  48
  49         if (ext4_has_feature_dir_index(inode->i_sb) &&
  50             ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
  51              ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
  52              ext4_has_inline_data(inode)))
  53                 return 1;
  54
  55         return 0;
  56 }
  57
  58 /*
  59  * Return 0 if the directory entry is OK, and 1 if there is a problem
  60  *
  61  * Note: this is the opposite of what ext2 and ext3 historically returned...
  62  *
  63  * bh passed here can be an inode block or a dir data block, depending
  64  * on the inode inline data flag.
  65  */
  66 int __ext4_check_dir_entry(const char *function, unsigned int line,
  67                            struct inode *dir, struct file *filp,
  68                            struct ext4_dir_entry_2 *de,
  69                            struct buffer_head *bh, char *buf, int size,
  70                            unsigned int offset)
  71 {
  72         const char *error_msg = NULL;
  73         const int rlen = ext4_rec_len_from_disk(de->rec_len,
  74                                                 dir->i_sb->s_blocksize);
  75
  76         if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
  77                 error_msg = "rec_len is smaller than minimal";
  78         else if (unlikely(rlen % 4 != 0))
  79                 error_msg = "rec_len % 4 != 0";
  80         else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
  81                 error_msg = "rec_len is too small for name_len";
  82         else if (unlikely(((char *) de - buf) + rlen > size))
  83                 error_msg = "directory entry overrun";
  84         else if (unlikely(((char *) de - buf) + rlen >
  85                           size - EXT4_DIR_REC_LEN(1) &&
  86                           ((char *) de - buf) + rlen != size)) {
  87                 error_msg = "directory entry too close to block end";
  88         }
  89         else if (unlikely(le32_to_cpu(de->inode) >
  90                         le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
  91                 error_msg = "inode out of bounds";
  92         else
  93                 return 0;
  94
  95         if (filp)
  96                 ext4_error_file(filp, function, line, bh->b_blocknr,
  97                                 "bad entry in directory: %s - offset=%u, "
  98                                 "inode=%u, rec_len=%d, name_len=%d, size=%d",
  99                                 error_msg, offset, le32_to_cpu(de->inode),
 100                                 rlen, de->name_len, size);
 101         else
 102                 ext4_error_inode(dir, function, line, bh->b_blocknr,
 103                                 "bad entry in directory: %s - offset=%u, "
 104                                 "inode=%u, rec_len=%d, name_len=%d, size=%d",
 105                                  error_msg, offset, le32_to_cpu(de->inode),
 106                                  rlen, de->name_len, size);
 107
 108         return 1;
 109 }
 110
 111 static int ext4_readdir(struct file *file, struct dir_context *ctx)
 112 {
 113         unsigned int offset;
 114         int i;
 115         struct ext4_dir_entry_2 *de;
 116         int err;
 117         struct inode *inode = file_inode(file);
 118         struct super_block *sb = inode->i_sb;
 119         struct buffer_head *bh = NULL;
 120         struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
 121
 122         if (IS_ENCRYPTED(inode)) {
 123                 err = fscrypt_get_encryption_info(inode);
 124                 if (err && err != -ENOKEY)
 125                         return err;
 126         }
 127
 128         if (is_dx_dir(inode)) {
 129                 err = ext4_dx_readdir(file, ctx);
 130                 if (err != ERR_BAD_DX_DIR) {
 131                         return err;
 132                 }
 133                 /* Can we just clear INDEX flag to ignore htree information? */
 134                 if (!ext4_has_metadata_csum(sb)) {
 135                         /*
 136                          * We don't set the inode dirty flag since it's not
 137                          * critical that it gets flushed back to the disk.
 138                          */
 139                         ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
 140                 }
 141         }
 142
 143         if (ext4_has_inline_data(inode)) {
 144                 int has_inline_data = 1;
 145                 err = ext4_read_inline_dir(file, ctx,
 146                                            &has_inline_data);
 147                 if (has_inline_data)
 148                         return err;
 149         }
 150
 151         if (IS_ENCRYPTED(inode)) {
 152                 err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr);
 153                 if (err < 0)
 154                         return err;
 155         }
 156
 157         while (ctx->pos < inode->i_size) {
 158                 struct ext4_map_blocks map;
 159
 160                 if (fatal_signal_pending(current)) {
 161                         err = -ERESTARTSYS;
 162                         goto errout;
 163                 }
 164                 cond_resched();
 165                 offset = ctx->pos & (sb->s_blocksize - 1);
 166                 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
 167                 map.m_len = 1;
 168                 err = ext4_map_blocks(NULL, inode, &map, 0);
 169                 if (err == 0) {
 170                         /* m_len should never be zero but let's avoid
 171                          * an infinite loop if it somehow is */
 172                         if (map.m_len == 0)
 173                                 map.m_len = 1;
 174                         ctx->pos += map.m_len * sb->s_blocksize;
 175                         continue;
 176                 }
 177                 if (err > 0) {
 178                         pgoff_t index = map.m_pblk >>
 179                                         (PAGE_SHIFT - inode->i_blkbits);
 180                         if (!ra_has_index(&file->f_ra, index))
 181                                 page_cache_sync_readahead(
 182                                         sb->s_bdev->bd_inode->i_mapping,
 183                                         &file->f_ra, file,
 184                                         index, 1);
 185                         file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
 186                         bh = ext4_bread(NULL, inode, map.m_lblk, 0);
 187                         if (IS_ERR(bh)) {
 188                                 err = PTR_ERR(bh);
 189                                 bh = NULL;
 190                                 goto errout;
 191                         }
 192                 }
 193
 194                 if (!bh) {
 195                         /* corrupt size?  Maybe no more blocks to read */
 196                         if (ctx->pos > inode->i_blocks << 9)
 197                                 break;
 198                         ctx->pos += sb->s_blocksize - offset;
 199                         continue;
 200                 }
 201
 202                 /* Check the checksum */
 203                 if (!buffer_verified(bh) &&
 204                     !ext4_dirblock_csum_verify(inode, bh)) {
 205                         EXT4_ERROR_FILE(file, 0, "directory fails checksum "
 206                                         "at offset %llu",
 207                                         (unsigned long long)ctx->pos);
 208                         ctx->pos += sb->s_blocksize - offset;
 209                         brelse(bh);
 210                         bh = NULL;
 211                         continue;
 212                 }
 213                 set_buffer_verified(bh);
 214
 215                 /* If the dir block has changed since the last call to
 216                  * readdir(2), then we might be pointing to an invalid
 217                  * dirent right now.  Scan from the start of the block
 218                  * to make sure. */
 219                 if (!inode_eq_iversion(inode, file->f_version)) {
 220                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
 221                                 de = (struct ext4_dir_entry_2 *)
 222                                         (bh->b_data + i);
 223                                 /* It's too expensive to do a full
 224                                  * dirent test each time round this
 225                                  * loop, but we do have to test at
 226                                  * least that it is non-zero.  A
 227                                  * failure will be detected in the
 228                                  * dirent test below. */
 229                                 if (ext4_rec_len_from_disk(de->rec_len,
 230                                         sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
 231                                         break;
 232                                 i += ext4_rec_len_from_disk(de->rec_len,
 233                                                             sb->s_blocksize);
 234                         }
 235                         offset = i;
 236                         ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
 237                                 | offset;
 238                         file->f_version = inode_query_iversion(inode);
 239                 }
 240
 241                 while (ctx->pos < inode->i_size
 242                        && offset < sb->s_blocksize) {
 243                         de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
 244                         if (ext4_check_dir_entry(inode, file, de, bh,
 245                                                  bh->b_data, bh->b_size,
 246                                                  offset)) {
 247                                 /*
 248                                  * On error, skip to the next block
 249                                  */
 250                                 ctx->pos = (ctx->pos |
 251                                                 (sb->s_blocksize - 1)) + 1;
 252                                 break;
 253                         }
 254                         offset += ext4_rec_len_from_disk(de->rec_len,
 255                                         sb->s_blocksize);
 256                         if (le32_to_cpu(de->inode)) {
 257                                 if (!IS_ENCRYPTED(inode)) {
 258                                         if (!dir_emit(ctx, de->name,
 259                                             de->name_len,
 260                                             le32_to_cpu(de->inode),
 261                                             get_dtype(sb, de->file_type)))
 262                                                 goto done;
 263                                 } else {
 264                                         int save_len = fstr.len;
 265                                         struct fscrypt_str de_name =
 266                                                         FSTR_INIT(de->name,
 267                                                                 de->name_len);
 268
 269                                         /* Directory is encrypted */
 270                                         err = fscrypt_fname_disk_to_usr(inode,
 271                                                 0, 0, &de_name, &fstr);
 272                                         de_name = fstr;
 273                                         fstr.len = save_len;
 274                                         if (err)
 275                                                 goto errout;
 276                                         if (!dir_emit(ctx,
 277                                             de_name.name, de_name.len,
 278                                             le32_to_cpu(de->inode),
 279                                             get_dtype(sb, de->file_type)))
 280                                                 goto done;
 281                                 }
 282                         }
 283                         ctx->pos += ext4_rec_len_from_disk(de->rec_len,
 284                                                 sb->s_blocksize);
 285                 }
 286                 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
 287                         goto done;
 288                 brelse(bh);
 289                 bh = NULL;
 290                 offset = 0;
 291         }
 292 done:
 293         err = 0;
 294 errout:
 295         fscrypt_fname_free_buffer(&fstr);
 296         brelse(bh);
 297         return err;
 298 }
 299
 300 static inline int is_32bit_api(void)
 301 {
 302 #ifdef CONFIG_COMPAT
 303         return in_compat_syscall();
 304 #else
 305         return (BITS_PER_LONG == 32);
 306 #endif
 307 }
 308
 309 /*
 310  * These functions convert from the major/minor hash to an f_pos
 311  * value for dx directories
 312  *
 313  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
 314  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
 315  * directly on both 32-bit and 64-bit nodes, under such case, neither
 316  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
 317  */
 318 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
 319 {
 320         if ((filp->f_mode & FMODE_32BITHASH) ||
 321             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 322                 return major >> 1;
 323         else
 324                 return ((__u64)(major >> 1) << 32) | (__u64)minor;
 325 }
 326
 327 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
 328 {
 329         if ((filp->f_mode & FMODE_32BITHASH) ||
 330             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 331                 return (pos << 1) & 0xffffffff;
 332         else
 333                 return ((pos >> 32) << 1) & 0xffffffff;
 334 }
 335
 336 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
 337 {
 338         if ((filp->f_mode & FMODE_32BITHASH) ||
 339             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 340                 return 0;
 341         else
 342                 return pos & 0xffffffff;
 343 }
 344
 345 /*
 346  * Return 32- or 64-bit end-of-file for dx directories
 347  */
 348 static inline loff_t ext4_get_htree_eof(struct file *filp)
 349 {
 350         if ((filp->f_mode & FMODE_32BITHASH) ||
 351             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 352                 return EXT4_HTREE_EOF_32BIT;
 353         else
 354                 return EXT4_HTREE_EOF_64BIT;
 355 }
 356
 357
 358 /*
 359  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
 360  * directories, where the "offset" is in terms of the filename hash
 361  * value instead of the byte offset.
 362  *
 363  * Because we may return a 64-bit hash that is well beyond offset limits,
 364  * we need to pass the max hash as the maximum allowable offset in
 365  * the htree directory case.
 366  *
 367  * For non-htree, ext4_llseek already chooses the proper max offset.
 368  */
 369 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
 370 {
 371         struct inode *inode = file->f_mapping->host;
 372         int dx_dir = is_dx_dir(inode);
 373         loff_t ret, htree_max = ext4_get_htree_eof(file);
 374
 375         if (likely(dx_dir))
 376                 ret = generic_file_llseek_size(file, offset, whence,
 377                                                     htree_max, htree_max);
 378         else
 379                 ret = ext4_llseek(file, offset, whence);
 380         file->f_version = inode_peek_iversion(inode) - 1;
 381         return ret;
 382 }
 383
 384 /*
 385  * This structure holds the nodes of the red-black tree used to store
 386  * the directory entry in hash order.
 387  */
 388 struct fname {
 389         __u32           hash;
 390         __u32           minor_hash;
 391         struct rb_node  rb_hash;
 392         struct fname    *next;
 393         __u32           inode;
 394         __u8            name_len;
 395         __u8            file_type;
 396         char            name[0];
 397 };
 398
 399 /*
 400  * This functoin implements a non-recursive way of freeing all of the
 401  * nodes in the red-black tree.
 402  */
 403 static void free_rb_tree_fname(struct rb_root *root)
 404 {
 405         struct fname *fname, *next;
 406
 407         rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
 408                 while (fname) {
 409                         struct fname *old = fname;
 410                         fname = fname->next;
 411                         kfree(old);
 412                 }
 413
 414         *root = RB_ROOT;
 415 }
 416
 417
 418 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
 419                                                            loff_t pos)
 420 {
 421         struct dir_private_info *p;
 422
 423         p = kzalloc(sizeof(*p), GFP_KERNEL);
 424         if (!p)
 425                 return NULL;
 426         p->curr_hash = pos2maj_hash(filp, pos);
 427         p->curr_minor_hash = pos2min_hash(filp, pos);
 428         return p;
 429 }
 430
 431 void ext4_htree_free_dir_info(struct dir_private_info *p)
 432 {
 433         free_rb_tree_fname(&p->root);
 434         kfree(p);
 435 }
 436
 437 /*
 438  * Given a directory entry, enter it into the fname rb tree.
 439  *
 440  * When filename encryption is enabled, the dirent will hold the
 441  * encrypted filename, while the htree will hold decrypted filename.
 442  * The decrypted filename is passed in via ent_name.  parameter.
 443  */
 444 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
 445                              __u32 minor_hash,
 446                             struct ext4_dir_entry_2 *dirent,
 447                             struct fscrypt_str *ent_name)
 448 {
 449         struct rb_node **p, *parent = NULL;
 450         struct fname *fname, *new_fn;
 451         struct dir_private_info *info;
 452         int len;
 453
 454         info = dir_file->private_data;
 455         p = &info->root.rb_node;
 456
 457         /* Create and allocate the fname structure */
 458         len = sizeof(struct fname) + ent_name->len + 1;
 459         new_fn = kzalloc(len, GFP_KERNEL);
 460         if (!new_fn)
 461                 return -ENOMEM;
 462         new_fn->hash = hash;
 463         new_fn->minor_hash = minor_hash;
 464         new_fn->inode = le32_to_cpu(dirent->inode);
 465         new_fn->name_len = ent_name->len;
 466         new_fn->file_type = dirent->file_type;
 467         memcpy(new_fn->name, ent_name->name, ent_name->len);
 468         new_fn->name[ent_name->len] = 0;
 469
 470         while (*p) {
 471                 parent = *p;
 472                 fname = rb_entry(parent, struct fname, rb_hash);
 473
 474                 /*
 475                  * If the hash and minor hash match up, then we put
 476                  * them on a linked list.  This rarely happens...
 477                  */
 478                 if ((new_fn->hash == fname->hash) &&
 479                     (new_fn->minor_hash == fname->minor_hash)) {
 480                         new_fn->next = fname->next;
 481                         fname->next = new_fn;
 482                         return 0;
 483                 }
 484
 485                 if (new_fn->hash < fname->hash)
 486                         p = &(*p)->rb_left;
 487                 else if (new_fn->hash > fname->hash)
 488                         p = &(*p)->rb_right;
 489                 else if (new_fn->minor_hash < fname->minor_hash)
 490                         p = &(*p)->rb_left;
 491                 else /* if (new_fn->minor_hash > fname->minor_hash) */
 492                         p = &(*p)->rb_right;
 493         }
 494
 495         rb_link_node(&new_fn->rb_hash, parent, p);
 496         rb_insert_color(&new_fn->rb_hash, &info->root);
 497         return 0;
 498 }
 499
 500
 501
 502 /*
 503  * This is a helper function for ext4_dx_readdir.  It calls filldir
 504  * for all entres on the fname linked list.  (Normally there is only
 505  * one entry on the linked list, unless there are 62 bit hash collisions.)
 506  */
 507 static int call_filldir(struct file *file, struct dir_context *ctx,
 508                         struct fname *fname)
 509 {
 510         struct dir_private_info *info = file->private_data;
 511         struct inode *inode = file_inode(file);
 512         struct super_block *sb = inode->i_sb;
 513
 514         if (!fname) {
 515                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
 516                          "called with null fname?!?", __func__, __LINE__,
 517                          inode->i_ino, current->comm);
 518                 return 0;
 519         }
 520         ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
 521         while (fname) {
 522                 if (!dir_emit(ctx, fname->name,
 523                                 fname->name_len,
 524                                 fname->inode,
 525                                 get_dtype(sb, fname->file_type))) {
 526                         info->extra_fname = fname;
 527                         return 1;
 528                 }
 529                 fname = fname->next;
 530         }
 531         return 0;
 532 }
 533
 534 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
 535 {
 536         struct dir_private_info *info = file->private_data;
 537         struct inode *inode = file_inode(file);
 538         struct fname *fname;
 539         int ret = 0;
 540
 541         if (!info) {
 542                 info = ext4_htree_create_dir_info(file, ctx->pos);
 543                 if (!info)
 544                         return -ENOMEM;
 545                 file->private_data = info;
 546         }
 547
 548         if (ctx->pos == ext4_get_htree_eof(file))
 549                 return 0;       /* EOF */
 550
 551         /* Some one has messed with f_pos; reset the world */
 552         if (info->last_pos != ctx->pos) {
 553                 free_rb_tree_fname(&info->root);
 554                 info->curr_node = NULL;
 555                 info->extra_fname = NULL;
 556                 info->curr_hash = pos2maj_hash(file, ctx->pos);
 557                 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
 558         }
 559
 560         /*
 561          * If there are any leftover names on the hash collision
 562          * chain, return them first.
 563          */
 564         if (info->extra_fname) {
 565                 if (call_filldir(file, ctx, info->extra_fname))
 566                         goto finished;
 567                 info->extra_fname = NULL;
 568                 goto next_node;
 569         } else if (!info->curr_node)
 570                 info->curr_node = rb_first(&info->root);
 571
 572         while (1) {
 573                 /*
 574                  * Fill the rbtree if we have no more entries,
 575                  * or the inode has changed since we last read in the
 576                  * cached entries.
 577                  */
 578                 if ((!info->curr_node) ||
 579                     !inode_eq_iversion(inode, file->f_version)) {
 580                         info->curr_node = NULL;
 581                         free_rb_tree_fname(&info->root);
 582                         file->f_version = inode_query_iversion(inode);
 583                         ret = ext4_htree_fill_tree(file, info->curr_hash,
 584                                                    info->curr_minor_hash,
 585                                                    &info->next_hash);
 586                         if (ret < 0)
 587                                 goto finished;
 588                         if (ret == 0) {
 589                                 ctx->pos = ext4_get_htree_eof(file);
 590                                 break;
 591                         }
 592                         info->curr_node = rb_first(&info->root);
 593                 }
 594
 595                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
 596                 info->curr_hash = fname->hash;
 597                 info->curr_minor_hash = fname->minor_hash;
 598                 if (call_filldir(file, ctx, fname))
 599                         break;
 600         next_node:
 601                 info->curr_node = rb_next(info->curr_node);
 602                 if (info->curr_node) {
 603                         fname = rb_entry(info->curr_node, struct fname,
 604                                          rb_hash);
 605                         info->curr_hash = fname->hash;
 606                         info->curr_minor_hash = fname->minor_hash;
 607                 } else {
 608                         if (info->next_hash == ~0) {
 609                                 ctx->pos = ext4_get_htree_eof(file);
 610                                 break;
 611                         }
 612                         info->curr_hash = info->next_hash;
 613                         info->curr_minor_hash = 0;
 614                 }
 615         }
 616 finished:
 617         info->last_pos = ctx->pos;
 618         return ret < 0 ? ret : 0;
 619 }
 620
 621 static int ext4_dir_open(struct inode * inode, struct file * filp)
 622 {
 623         if (IS_ENCRYPTED(inode))
 624                 return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
 625         return 0;
 626 }
 627
 628 static int ext4_release_dir(struct inode *inode, struct file *filp)
 629 {
 630         if (filp->private_data)
 631                 ext4_htree_free_dir_info(filp->private_data);
 632
 633         return 0;
 634 }
 635
 636 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
 637                       int buf_size)
 638 {
 639         struct ext4_dir_entry_2 *de;
 640         int rlen;
 641         unsigned int offset = 0;
 642         char *top;
 643
 644         de = (struct ext4_dir_entry_2 *)buf;
 645         top = buf + buf_size;
 646         while ((char *) de < top) {
 647                 if (ext4_check_dir_entry(dir, NULL, de, bh,
 648                                          buf, buf_size, offset))
 649                         return -EFSCORRUPTED;
 650                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 651                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
 652                 offset += rlen;
 653         }
 654         if ((char *) de > top)
 655                 return -EFSCORRUPTED;
 656
 657         return 0;
 658 }
 659
 660 const struct file_operations ext4_dir_operations = {
 661         .llseek         = ext4_dir_llseek,
 662         .read           = generic_read_dir,
 663         .iterate_shared = ext4_readdir,
 664         .unlocked_ioctl = ext4_ioctl,
 665 #ifdef CONFIG_COMPAT
 666         .compat_ioctl   = ext4_compat_ioctl,
 667 #endif
 668         .fsync          = ext4_sync_file,
 669         .open           = ext4_dir_open,
 670         .release        = ext4_release_dir,
 671 };
 672
 673 #ifdef CONFIG_UNICODE
 674 static int ext4_d_compare(const struct dentry *dentry, unsigned int len,
 675                           const char *str, const struct qstr *name)
 676 {
 677         struct qstr qstr = {.name = str, .len = len };
 678         const struct dentry *parent = READ_ONCE(dentry->d_parent);
 679         const struct inode *inode = READ_ONCE(parent->d_inode);
 680         char strbuf[DNAME_INLINE_LEN];
 681
 682         if (!inode || !IS_CASEFOLDED(inode) ||
 683             !EXT4_SB(inode->i_sb)->s_encoding) {
 684                 if (len != name->len)
 685                         return -1;
 686                 return memcmp(str, name->name, len);
 687         }
 688
 689         /*
 690          * If the dentry name is stored in-line, then it may be concurrently
 691          * modified by a rename.  If this happens, the VFS will eventually retry
 692          * the lookup, so it doesn't matter what ->d_compare() returns.
 693          * However, it's unsafe to call utf8_strncasecmp() with an unstable
 694          * string.  Therefore, we have to copy the name into a temporary buffer.
 695          */
 696         if (len <= DNAME_INLINE_LEN - 1) {
 697                 memcpy(strbuf, str, len);
 698                 strbuf[len] = 0;
 699                 qstr.name = strbuf;
 700                 /* prevent compiler from optimizing out the temporary buffer */
 701                 barrier();
 702         }
 703
 704         return ext4_ci_compare(inode, name, &qstr, false);
 705 }
 706
 707 static int ext4_d_hash(const struct dentry *dentry, struct qstr *str)
 708 {
 709         const struct ext4_sb_info *sbi = EXT4_SB(dentry->d_sb);
 710         const struct unicode_map *um = sbi->s_encoding;
 711         const struct inode *inode = READ_ONCE(dentry->d_inode);
 712         unsigned char *norm;
 713         int len, ret = 0;
 714
 715         if (!inode || !IS_CASEFOLDED(inode) || !um)
 716                 return 0;
 717
 718         norm = kmalloc(PATH_MAX, GFP_ATOMIC);
 719         if (!norm)
 720                 return -ENOMEM;
 721
 722         len = utf8_casefold(um, str, norm, PATH_MAX);
 723         if (len < 0) {
 724                 if (ext4_has_strict_mode(sbi))
 725                         ret = -EINVAL;
 726                 goto out;
 727         }
 728         str->hash = full_name_hash(dentry, norm, len);
 729 out:
 730         kfree(norm);
 731         return ret;
 732 }
 733
 734 const struct dentry_operations ext4_dentry_ops = {
 735         .d_hash = ext4_d_hash,
 736         .d_compare = ext4_d_compare,
 737 };
 738 #endif