e2fsck/rehash.c

   1 /*
   2  * rehash.c --- rebuild hash tree directories
   3  *
   4  * Copyright (C) 2002 Theodore Ts'o
   5  *
   6  * %Begin-Header%
   7  * This file may be redistributed under the terms of the GNU Public
   8  * License.
   9  * %End-Header%
  10  *
  11  * This algorithm is designed for simplicity of implementation and to
  12  * pack the directory as much as possible.  It however requires twice
  13  * as much memory as the size of the directory.  The maximum size
  14  * directory supported using a 4k blocksize is roughly a gigabyte, and
  15  * so there may very well be problems with machines that don't have
  16  * virtual memory, and obscenely large directories.
  17  *
  18  * An alternate algorithm which is much more disk intensive could be
  19  * written, and probably will need to be written in the future.  The
  20  * design goals of such an algorithm are: (a) use (roughly) constant
  21  * amounts of memory, no matter how large the directory, (b) the
  22  * directory must be safe at all times, even if e2fsck is interrupted
  23  * in the middle, (c) we must use minimal amounts of extra disk
  24  * blocks.  This pretty much requires an incremental approach, where
  25  * we are reading from one part of the directory, and inserting into
  26  * the front half.  So the algorithm will have to keep track of a
  27  * moving block boundary between the new tree and the old tree, and
  28  * files will need to be moved from the old directory and inserted
  29  * into the new tree.  If the new directory requires space which isn't
  30  * yet available, blocks from the beginning part of the old directory
  31  * may need to be moved to the end of the directory to make room for
  32  * the new tree:
  33  *
  34  *    --------------------------------------------------------
  35  *    |  new tree   |        | old tree                      |
  36  *    --------------------------------------------------------
  37  *                  ^ ptr    ^ptr
  38  *                tail new   head old
  39  *
  40  * This is going to be a pain in the tuckus to implement, and will
  41  * require a lot more disk accesses.  So I'm going to skip it for now;
  42  * it's only really going to be an issue for really, really big
  43  * filesystems (when we reach the level of tens of millions of files
  44  * in a single directory).  It will probably be easier to simply
  45  * require that e2fsck use VM first.
  46  */
  47
  48 #include "config.h"
  49 #include <string.h>
  50 #include <ctype.h>
  51 #include <errno.h>
  52 #include "e2fsck.h"
  53 #include "problem.h"
  54
  55 /* Schedule a dir to be rebuilt during pass 3A. */
  56 void e2fsck_rehash_dir_later(e2fsck_t ctx, ext2_ino_t ino)
  57 {
  58         if (!ctx->dirs_to_hash)
  59                 ext2fs_u32_list_create(&ctx->dirs_to_hash, 50);
  60         if (ctx->dirs_to_hash)
  61                 ext2fs_u32_list_add(ctx->dirs_to_hash, ino);
  62 }
  63
  64 /* Ask if a dir will be rebuilt during pass 3A. */
  65 int e2fsck_dir_will_be_rehashed(e2fsck_t ctx, ext2_ino_t ino)
  66 {
  67         if (ctx->options & E2F_OPT_COMPRESS_DIRS)
  68                 return 1;
  69         if (!ctx->dirs_to_hash)
  70                 return 0;
  71         return ext2fs_u32_list_test(ctx->dirs_to_hash, ino);
  72 }
  73
  74 #undef REHASH_DEBUG
  75
  76 struct fill_dir_struct {
  77         char *buf;
  78         struct ext2_inode *inode;
  79         ext2_ino_t ino;
  80         errcode_t err;
  81         e2fsck_t ctx;
  82         struct hash_entry *harray;
  83         blk_t max_array, num_array;
  84         ext2_off64_t dir_size;
  85         int compress;
  86         ext2_ino_t parent;
  87         ext2_ino_t dir;
  88         struct ext2_dir_entry *dot_de;
  89         struct ext2_dir_entry *dotdot_de;
  90 };
  91
  92 struct hash_entry {
  93         ext2_dirhash_t  hash;
  94         ext2_dirhash_t  minor_hash;
  95         ino_t           ino;
  96         struct ext2_dir_entry   *dir;
  97 };
  98
  99 struct out_dir {
 100         blk_t           num;
 101         blk_t           max;
 102         char            *buf;
 103         ext2_dirhash_t  *hashes;
 104 };
 105
 106 static int fill_dir_block(ext2_filsys fs,
 107                           blk64_t *block_nr,
 108                           e2_blkcnt_t blockcnt,
 109                           blk64_t ref_block EXT2FS_ATTR((unused)),
 110                           int ref_offset EXT2FS_ATTR((unused)),
 111                           void *priv_data)
 112 {
 113         struct fill_dir_struct  *fd = (struct fill_dir_struct *) priv_data;
 114         struct hash_entry       *new_array, *ent;
 115         struct ext2_dir_entry   *dirent;
 116         char                    *dir;
 117         unsigned int            offset, dir_offset, rec_len, name_len;
 118         int                     hash_alg, hash_flags;
 119
 120         if (blockcnt < 0)
 121                 return 0;
 122
 123         offset = blockcnt * fs->blocksize;
 124         if (offset + fs->blocksize > fd->inode->i_size) {
 125                 fd->err = EXT2_ET_DIR_CORRUPTED;
 126                 return BLOCK_ABORT;
 127         }
 128
 129         dir = (fd->buf+offset);
 130         if (*block_nr == 0) {
 131                 memset(dir, 0, fs->blocksize);
 132                 dirent = (struct ext2_dir_entry *) dir;
 133                 (void) ext2fs_set_rec_len(fs, fs->blocksize, dirent);
 134         } else {
 135                 int flags = fs->flags;
 136                 fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
 137                 fd->err = ext2fs_read_dir_block4(fs, *block_nr, dir, 0,
 138                                                  fd->dir);
 139                 fs->flags = (flags & EXT2_FLAG_IGNORE_CSUM_ERRORS) |
 140                             (fs->flags & ~EXT2_FLAG_IGNORE_CSUM_ERRORS);
 141                 if (fd->err)
 142                         return BLOCK_ABORT;
 143         }
 144         hash_flags = fd->inode->i_flags & EXT4_CASEFOLD_FL;
 145         hash_alg = fs->super->s_def_hash_version;
 146         if ((hash_alg <= EXT2_HASH_TEA) &&
 147             (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
 148                 hash_alg += 3;
 149         /* While the directory block is "hot", index it. */
 150         dir_offset = 0;
 151         while (dir_offset < fs->blocksize) {
 152                 dirent = (struct ext2_dir_entry *) (dir + dir_offset);
 153                 (void) ext2fs_get_rec_len(fs, dirent, &rec_len);
 154                 name_len = ext2fs_dirent_name_len(dirent);
 155                 if (((dir_offset + rec_len) > fs->blocksize) ||
 156                     (rec_len < 8) ||
 157                     ((rec_len % 4) != 0) ||
 158                     (name_len + 8 > rec_len)) {
 159                         fd->err = EXT2_ET_DIR_CORRUPTED;
 160                         return BLOCK_ABORT;
 161                 }
 162                 dir_offset += rec_len;
 163                 if (dirent->inode == 0)
 164                         continue;
 165                 if ((name_len) == 0) {
 166                         fd->err = EXT2_ET_DIR_CORRUPTED;
 167                         return BLOCK_ABORT;
 168                 }
 169                 if (!fd->compress && (name_len == 1) &&
 170                     (dirent->name[0] == '.')) {
 171                         fd->dot_de = dirent;
 172                         continue;
 173                 }
 174                 if (!fd->compress && (name_len == 2) &&
 175                     (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
 176                         fd->parent = dirent->inode;
 177                         fd->dotdot_de = dirent;
 178                         continue;
 179                 }
 180                 if (fd->num_array >= fd->max_array) {
 181                         errcode_t retval;
 182
 183                         retval = ext2fs_resize_array(sizeof(struct hash_entry),
 184                                                      fd->max_array,
 185                                                      fd->max_array + 500,
 186                                                      &fd->harray);
 187                         if (retval) {
 188                                 fd->err = retval;
 189                                 return BLOCK_ABORT;
 190                         }
 191                         fd->max_array += 500;
 192                 }
 193                 ent = fd->harray + fd->num_array++;
 194                 ent->dir = dirent;
 195                 fd->dir_size += EXT2_DIR_REC_LEN(dirent);
 196                 ent->ino = dirent->inode;
 197                 if (fd->compress)
 198                         ent->hash = ent->minor_hash = 0;
 199                 else {
 200                         fd->err = ext2fs_dirhash2(hash_alg,
 201                                                   dirent->name, name_len,
 202                                                   fs->encoding, hash_flags,
 203                                                   fs->super->s_hash_seed,
 204                                                   &ent->hash, &ent->minor_hash);
 205                         if (fd->err)
 206                                 return BLOCK_ABORT;
 207                 }
 208         }
 209
 210         return 0;
 211 }
 212
 213 /* Used for sorting the hash entry */
 214 static EXT2_QSORT_TYPE ino_cmp(const void *a, const void *b)
 215 {
 216         const struct hash_entry *he_a = (const struct hash_entry *) a;
 217         const struct hash_entry *he_b = (const struct hash_entry *) b;
 218
 219         return (he_a->ino - he_b->ino);
 220 }
 221
 222 /* Used for sorting the hash entry */
 223 static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
 224 {
 225         const struct hash_entry *he_a = (const struct hash_entry *) a;
 226         const struct hash_entry *he_b = (const struct hash_entry *) b;
 227         unsigned int he_a_len, he_b_len, min_len;
 228         int     ret;
 229
 230         he_a_len = ext2fs_dirent_name_len(he_a->dir);
 231         he_b_len = ext2fs_dirent_name_len(he_b->dir);
 232         min_len = he_a_len;
 233         if (min_len > he_b_len)
 234                 min_len = he_b_len;
 235
 236         ret = memcmp(he_a->dir->name, he_b->dir->name, min_len);
 237         if (ret == 0) {
 238                 if (he_a_len > he_b_len)
 239                         ret = 1;
 240                 else if (he_a_len < he_b_len)
 241                         ret = -1;
 242                 else
 243                         ret = he_b->dir->inode - he_a->dir->inode;
 244         }
 245         return ret;
 246 }
 247
 248 /* Used for sorting the hash entry */
 249 static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
 250 {
 251         const struct hash_entry *he_a = (const struct hash_entry *) a;
 252         const struct hash_entry *he_b = (const struct hash_entry *) b;
 253         int     ret;
 254
 255         if (he_a->hash > he_b->hash)
 256                 ret = 1;
 257         else if (he_a->hash < he_b->hash)
 258                 ret = -1;
 259         else {
 260                 if (he_a->minor_hash > he_b->minor_hash)
 261                         ret = 1;
 262                 else if (he_a->minor_hash < he_b->minor_hash)
 263                         ret = -1;
 264                 else
 265                         ret = name_cmp(a, b);
 266         }
 267         return ret;
 268 }
 269
 270 static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
 271                                 blk_t blocks)
 272 {
 273         errcode_t retval;
 274
 275         if (outdir->max) {
 276                 retval = ext2fs_resize_array(fs->blocksize, outdir->max, blocks,
 277                                              &outdir->buf);
 278                 if (retval)
 279                         return retval;
 280                 retval = ext2fs_resize_array(sizeof(ext2_dirhash_t),
 281                                              outdir->max, blocks,
 282                                              &outdir->hashes);
 283                 if (retval)
 284                         return retval;
 285         } else {
 286                 retval = ext2fs_get_array(fs->blocksize, blocks, &outdir->buf);
 287                 if (retval)
 288                         return retval;
 289                 retval = ext2fs_get_array(sizeof(ext2_dirhash_t), blocks,
 290                                           &outdir->hashes);
 291                 if (retval)
 292                         return retval;
 293                 outdir->num = 0;
 294         }
 295         outdir->max = blocks;
 296         return 0;
 297 }
 298
 299 static void free_out_dir(struct out_dir *outdir)
 300 {
 301         free(outdir->buf);
 302         free(outdir->hashes);
 303         outdir->max = 0;
 304         outdir->num =0;
 305 }
 306
 307 static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
 308                          char ** ret)
 309 {
 310         errcode_t       retval;
 311
 312         if (outdir->num >= outdir->max) {
 313                 int increment = outdir->max / 10;
 314
 315                 if (increment < 50)
 316                         increment = 50;
 317                 retval = alloc_size_dir(fs, outdir, outdir->max + increment);
 318                 if (retval)
 319                         return retval;
 320         }
 321         *ret = outdir->buf + (size_t)outdir->num++ * fs->blocksize;
 322         memset(*ret, 0, fs->blocksize);
 323         return 0;
 324 }
 325
 326 /*
 327  * This function is used to make a unique filename.  We do this by
 328  * appending ~0, and then incrementing the number.  However, we cannot
 329  * expand the length of the filename beyond the padding available in
 330  * the directory entry.
 331  */
 332 static void mutate_name(char *str, unsigned int *len)
 333 {
 334         int i;
 335         unsigned int l = *len;
 336
 337         /*
 338          * First check to see if it looks the name has been mutated
 339          * already
 340          */
 341         for (i = l-1; i > 0; i--) {
 342                 if (!isdigit(str[i]))
 343                         break;
 344         }
 345         if ((i == (int)l - 1) || (str[i] != '~')) {
 346                 if (((l-1) & 3) < 2)
 347                         l += 2;
 348                 else
 349                         l = (l+3) & ~3;
 350                 str[l-2] = '~';
 351                 str[l-1] = '0';
 352                 *len = l;
 353                 return;
 354         }
 355         for (i = l-1; i >= 0; i--) {
 356                 if (isdigit(str[i])) {
 357                         if (str[i] == '9')
 358                                 str[i] = '0';
 359                         else {
 360                                 str[i]++;
 361                                 return;
 362                         }
 363                         continue;
 364                 }
 365                 if (i == 1) {
 366                         if (str[0] == 'z')
 367                                 str[0] = 'A';
 368                         else if (str[0] == 'Z') {
 369                                 str[0] = '~';
 370                                 str[1] = '0';
 371                         } else
 372                                 str[0]++;
 373                 } else if (i > 0) {
 374                         str[i] = '1';
 375                         str[i-1] = '~';
 376                 } else {
 377                         if (str[0] == '~')
 378                                 str[0] = 'a';
 379                         else
 380                                 str[0]++;
 381                 }
 382                 break;
 383         }
 384 }
 385
 386 static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
 387                                     ext2_ino_t ino,
 388                                     struct fill_dir_struct *fd)
 389 {
 390         struct problem_context  pctx;
 391         struct hash_entry       *ent, *prev;
 392         blk_t                   i, j;
 393         int                     fixed = 0;
 394         char                    new_name[256];
 395         unsigned int            new_len;
 396         int                     hash_alg;
 397         int hash_flags = fd->inode->i_flags & EXT4_CASEFOLD_FL;
 398
 399         clear_problem_context(&pctx);
 400         pctx.ino = ino;
 401
 402         hash_alg = fs->super->s_def_hash_version;
 403         if ((hash_alg <= EXT2_HASH_TEA) &&
 404             (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
 405                 hash_alg += 3;
 406
 407         for (i=1; i < fd->num_array; i++) {
 408                 ent = fd->harray + i;
 409                 prev = ent - 1;
 410                 if (!ent->dir->inode ||
 411                     (ext2fs_dirent_name_len(ent->dir) !=
 412                      ext2fs_dirent_name_len(prev->dir)) ||
 413                     memcmp(ent->dir->name, prev->dir->name,
 414                              ext2fs_dirent_name_len(ent->dir)))
 415                         continue;
 416                 pctx.dirent = ent->dir;
 417                 if ((ent->dir->inode == prev->dir->inode) &&
 418                     fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
 419                         e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
 420                         ent->dir->inode = 0;
 421                         fixed++;
 422                         continue;
 423                 }
 424                 new_len = ext2fs_dirent_name_len(ent->dir);
 425                 if (new_len == 0) {
 426                          /* should never happen */
 427                         ext2fs_unmark_valid(fs);
 428                         continue;
 429                 }
 430                 memcpy(new_name, ent->dir->name, new_len);
 431                 mutate_name(new_name, &new_len);
 432                 for (j=0; j < fd->num_array; j++) {
 433                         if ((i==j) ||
 434                             (new_len !=
 435                              (unsigned) ext2fs_dirent_name_len(fd->harray[j].dir)) ||
 436                             memcmp(new_name, fd->harray[j].dir->name, new_len))
 437                                 continue;
 438                         mutate_name(new_name, &new_len);
 439
 440                         j = -1;
 441                 }
 442                 new_name[new_len] = 0;
 443                 pctx.str = new_name;
 444                 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
 445                         memcpy(ent->dir->name, new_name, new_len);
 446                         ext2fs_dirent_set_name_len(ent->dir, new_len);
 447                         ext2fs_dirhash2(hash_alg, new_name, new_len,
 448                                         fs->encoding, hash_flags,
 449                                         fs->super->s_hash_seed,
 450                                         &ent->hash, &ent->minor_hash);
 451                         fixed++;
 452                 }
 453         }
 454         return fixed;
 455 }
 456
 457
 458 static errcode_t copy_dir_entries(e2fsck_t ctx,
 459                                   struct fill_dir_struct *fd,
 460                                   struct out_dir *outdir)
 461 {
 462         ext2_filsys             fs = ctx->fs;
 463         errcode_t               retval;
 464         char                    *block_start;
 465         struct hash_entry       *ent;
 466         struct ext2_dir_entry   *dirent;
 467         unsigned int            rec_len, prev_rec_len, left, slack, offset;
 468         int                     i;
 469         ext2_dirhash_t          prev_hash;
 470         int                     csum_size = 0;
 471         struct                  ext2_dir_entry_tail *t;
 472
 473         if (ctx->htree_slack_percentage == 255) {
 474                 profile_get_uint(ctx->profile, "options",
 475                                  "indexed_dir_slack_percentage",
 476                                  0, 20,
 477                                  &ctx->htree_slack_percentage);
 478                 if (ctx->htree_slack_percentage > 100)
 479                         ctx->htree_slack_percentage = 20;
 480         }
 481
 482         if (ext2fs_has_feature_metadata_csum(fs->super))
 483                 csum_size = sizeof(struct ext2_dir_entry_tail);
 484
 485         outdir->max = 0;
 486         retval = alloc_size_dir(fs, outdir,
 487                                 (fd->dir_size / fs->blocksize) + 2);
 488         if (retval)
 489                 return retval;
 490         outdir->num = fd->compress ? 0 : 1;
 491         offset = 0;
 492         outdir->hashes[0] = 0;
 493         prev_hash = 1;
 494         if ((retval = get_next_block(fs, outdir, &block_start)))
 495                 return retval;
 496         dirent = (struct ext2_dir_entry *) block_start;
 497         prev_rec_len = 0;
 498         rec_len = 0;
 499         left = fs->blocksize - csum_size;
 500         slack = fd->compress ? 12 :
 501                 ((fs->blocksize - csum_size) * ctx->htree_slack_percentage)/100;
 502         if (slack < 12)
 503                 slack = 12;
 504         for (i = 0; i < fd->num_array; i++) {
 505                 ent = fd->harray + i;
 506                 if (ent->dir->inode == 0)
 507                         continue;
 508                 rec_len = EXT2_DIR_REC_LEN(ent->dir);
 509                 if (rec_len > left) {
 510                         if (left) {
 511                                 left += prev_rec_len;
 512                                 retval = ext2fs_set_rec_len(fs, left, dirent);
 513                                 if (retval)
 514                                         return retval;
 515                         }
 516                         if (csum_size) {
 517                                 t = EXT2_DIRENT_TAIL(block_start,
 518                                                      fs->blocksize);
 519                                 ext2fs_initialize_dirent_tail(fs, t);
 520                         }
 521                         if ((retval = get_next_block(fs, outdir,
 522                                                       &block_start)))
 523                                 return retval;
 524                         offset = 0;
 525                 }
 526                 left = (fs->blocksize - csum_size) - offset;
 527                 dirent = (struct ext2_dir_entry *) (block_start + offset);
 528                 if (offset == 0) {
 529                         if (ent->hash == prev_hash)
 530                                 outdir->hashes[outdir->num-1] = ent->hash | 1;
 531                         else
 532                                 outdir->hashes[outdir->num-1] = ent->hash;
 533                 }
 534                 dirent->inode = ent->dir->inode;
 535                 ext2fs_dirent_set_name_len(dirent,
 536                                            ext2fs_dirent_name_len(ent->dir));
 537                 ext2fs_dirent_set_file_type(dirent,
 538                                             ext2fs_dirent_file_type(ent->dir));
 539                 retval = ext2fs_set_rec_len(fs, rec_len, dirent);
 540                 if (retval)
 541                         return retval;
 542                 prev_rec_len = rec_len;
 543                 memcpy(dirent->name, ent->dir->name, rec_len);
 544                 offset += rec_len;
 545                 left -= rec_len;
 546                 if (left < slack) {
 547                         prev_rec_len += left;
 548                         retval = ext2fs_set_rec_len(fs, prev_rec_len, dirent);
 549                         if (retval)
 550                                 return retval;
 551                         offset += left;
 552                         left = 0;
 553                 }
 554                 prev_hash = ent->hash;
 555         }
 556         if (left)
 557                 retval = ext2fs_set_rec_len(fs, rec_len + left, dirent);
 558         if (csum_size) {
 559                 t = EXT2_DIRENT_TAIL(block_start, fs->blocksize);
 560                 ext2fs_initialize_dirent_tail(fs, t);
 561         }
 562
 563         return retval;
 564 }
 565
 566
 567 static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
 568                                         ext2_ino_t ino, ext2_ino_t parent,
 569                                         struct ext2_dir_entry *dot_de,
 570                                         struct ext2_dir_entry *dotdot_de)
 571 {
 572         struct ext2_dir_entry           *dirent;
 573         struct ext2_dx_root_info        *root;
 574         struct ext2_dx_countlimit       *limits;
 575         int                             csum_size = 0;
 576         int                             offset;
 577         int                             rec_len;
 578
 579         memset(buf, 0, fs->blocksize);
 580         dirent = (struct ext2_dir_entry *) buf;
 581         dirent->inode = ino;
 582
 583         dirent->name_len = dot_de->name_len;
 584         offset = rec_len = dirent->rec_len = dot_de->rec_len;
 585         memcpy(dirent->name, dot_de->name, rec_len);
 586
 587         dirent = EXT2_NEXT_DIRENT(dirent);
 588         /* set to jump over the index block */
 589
 590         dirent->inode = parent;
 591
 592         dirent->name_len = dotdot_de->name_len;
 593         dirent->rec_len = fs->blocksize - rec_len;
 594         rec_len = EXT2_DIR_REC_LEN(dotdot_de);
 595         memcpy(dirent->name, dotdot_de->name, rec_len);
 596         offset += rec_len;
 597
 598         root = (struct ext2_dx_root_info *)(buf + offset);
 599         root->reserved_zero = 0;
 600         root->hash_version = fs->super->s_def_hash_version;
 601         root->info_length = sizeof(*root);
 602         root->indirect_levels = 0;
 603         root->unused_flags = 0;
 604         offset += root->info_length;
 605
 606         if (ext2fs_has_feature_metadata_csum(fs->super))
 607                 csum_size = sizeof(struct ext2_dx_tail);
 608
 609         limits = (struct ext2_dx_countlimit *) (buf + offset);
 610         limits->limit = (fs->blocksize - (offset + csum_size)) /
 611                         sizeof(struct ext2_dx_entry);
 612         limits->count = 0;
 613
 614         return root;
 615 }
 616
 617
 618 static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
 619 {
 620         struct ext2_dir_entry           *dir;
 621         struct ext2_dx_countlimit       *limits;
 622         int                             csum_size = 0;
 623
 624         memset(buf, 0, fs->blocksize);
 625         dir = (struct ext2_dir_entry *) buf;
 626         dir->inode = 0;
 627         (void) ext2fs_set_rec_len(fs, fs->blocksize, dir);
 628
 629         if (ext2fs_has_feature_metadata_csum(fs->super))
 630                 csum_size = sizeof(struct ext2_dx_tail);
 631
 632         limits = (struct ext2_dx_countlimit *) (buf+8);
 633         limits->limit = (fs->blocksize - (8 + csum_size)) /
 634                         sizeof(struct ext2_dx_entry);
 635         limits->count = 0;
 636
 637         return (struct ext2_dx_entry *) limits;
 638 }
 639
 640 static int alloc_blocks(ext2_filsys fs,
 641                         struct ext2_dx_countlimit **limit,
 642                         struct ext2_dx_entry **prev_ent,
 643                         struct ext2_dx_entry **next_ent,
 644                         int *prev_offset, int *next_offset,
 645                         struct out_dir *outdir, int i,
 646                         int *prev_count, int *next_count)
 647 {
 648         errcode_t       retval;
 649         char            *block_start;
 650
 651         if (*limit)
 652                 (*limit)->limit = (*limit)->count =
 653                         ext2fs_cpu_to_le16((*limit)->limit);
 654         *prev_ent = (struct ext2_dx_entry *) (outdir->buf + *prev_offset);
 655         (*prev_ent)->block = ext2fs_cpu_to_le32(outdir->num);
 656
 657         if (i != 1)
 658                 (*prev_ent)->hash =
 659                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 660
 661         retval = get_next_block(fs, outdir, &block_start);
 662         if (retval)
 663                 return retval;
 664
 665         /* outdir->buf might be reallocated */
 666         *prev_ent = (struct ext2_dx_entry *) (outdir->buf + *prev_offset);
 667
 668         *next_ent = set_int_node(fs, block_start);
 669         *limit = (struct ext2_dx_countlimit *)(*next_ent);
 670         if (next_offset)
 671                 *next_offset = ((char *) *next_ent - outdir->buf);
 672
 673         *next_count = (*limit)->limit;
 674         (*prev_offset) += sizeof(struct ext2_dx_entry);
 675         (*prev_count)--;
 676
 677         return 0;
 678 }
 679
 680 /*
 681  * This function takes the leaf nodes which have been written in
 682  * outdir, and populates the root node and any necessary interior nodes.
 683  */
 684 static errcode_t calculate_tree(ext2_filsys fs,
 685                                 struct out_dir *outdir,
 686                                 ext2_ino_t ino,
 687                                 ext2_ino_t parent,
 688                                 struct ext2_dir_entry *dot_de,
 689                                 struct ext2_dir_entry *dotdot_de)
 690 {
 691         struct ext2_dx_root_info        *root_info;
 692         struct ext2_dx_entry            *root, *int_ent, *dx_ent = 0;
 693         struct ext2_dx_countlimit       *root_limit, *int_limit, *limit;
 694         errcode_t                       retval;
 695         int                             i, c1, c2, c3, nblks;
 696         int                             limit_offset, int_offset, root_offset;
 697
 698         root_info = set_root_node(fs, outdir->buf, ino, parent, dot_de,
 699                                   dotdot_de);
 700
 701         root_offset = limit_offset = ((char *) root_info - outdir->buf) +
 702                 root_info->info_length;
 703         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 704         c1 = root_limit->limit;
 705         nblks = outdir->num;
 706
 707         /* Write out the pointer blocks */
 708         if (nblks - 1 <= c1) {
 709                 /* Just write out the root block, and we're done */
 710                 root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
 711                 for (i=1; i < nblks; i++) {
 712                         root->block = ext2fs_cpu_to_le32(i);
 713                         if (i != 1)
 714                                 root->hash =
 715                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 716                         root++;
 717                         c1--;
 718                 }
 719         } else if (nblks - 1 <= ext2fs_htree_intnode_maxrecs(fs, c1)) {
 720                 c2 = 0;
 721                 limit = NULL;
 722                 root_info->indirect_levels = 1;
 723                 for (i=1; i < nblks; i++) {
 724                         if (c2 == 0 && c1 == 0)
 725                                 return ENOSPC;
 726                         if (c2 == 0) {
 727                                 retval = alloc_blocks(fs, &limit, &root,
 728                                                       &dx_ent, &root_offset,
 729                                                       NULL, outdir, i, &c1,
 730                                                       &c2);
 731                                 if (retval)
 732                                         return retval;
 733                         }
 734                         dx_ent->block = ext2fs_cpu_to_le32(i);
 735                         if (c2 != limit->limit)
 736                                 dx_ent->hash =
 737                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 738                         dx_ent++;
 739                         c2--;
 740                 }
 741                 limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
 742                 limit->limit = ext2fs_cpu_to_le16(limit->limit);
 743         } else {
 744                 c2 = 0;
 745                 c3 = 0;
 746                 limit = NULL;
 747                 int_limit = 0;
 748                 root_info->indirect_levels = 2;
 749                 for (i = 1; i < nblks; i++) {
 750                         if (c3 == 0 && c2 == 0 && c1 == 0)
 751                                 return ENOSPC;
 752                         if (c3 == 0 && c2 == 0) {
 753                                 retval = alloc_blocks(fs, &int_limit, &root,
 754                                                       &int_ent, &root_offset,
 755                                                       &int_offset, outdir, i,
 756                                                       &c1, &c2);
 757                                 if (retval)
 758                                         return retval;
 759                         }
 760                         if (c3 == 0) {
 761                                 int delta1 = (char *)int_limit - outdir->buf;
 762                                 int delta2 = (char *)root - outdir->buf;
 763
 764                                 retval = alloc_blocks(fs, &limit, &int_ent,
 765                                                       &dx_ent, &int_offset,
 766                                                       NULL, outdir, i, &c2,
 767                                                       &c3);
 768                                 if (retval)
 769                                         return retval;
 770
 771                                 /* outdir->buf might be reallocated */
 772                                 int_limit = (struct ext2_dx_countlimit *)
 773                                         (outdir->buf + delta1);
 774                                 root = (struct ext2_dx_entry *)
 775                                         (outdir->buf + delta2);
 776                         }
 777                         dx_ent->block = ext2fs_cpu_to_le32(i);
 778                         if (c3 != limit->limit)
 779                                 dx_ent->hash =
 780                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 781                         dx_ent++;
 782                         c3--;
 783                 }
 784                 int_limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
 785                 int_limit->limit = ext2fs_cpu_to_le16(limit->limit);
 786
 787                 limit->count = ext2fs_cpu_to_le16(limit->limit - c3);
 788                 limit->limit = ext2fs_cpu_to_le16(limit->limit);
 789
 790         }
 791         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 792         root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
 793         root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
 794
 795         return 0;
 796 }
 797
 798 struct write_dir_struct {
 799         struct out_dir *outdir;
 800         errcode_t       err;
 801         ext2_ino_t      ino;
 802         e2fsck_t        ctx;
 803         ext2_ino_t      dir;
 804 };
 805
 806 /*
 807  * Helper function which writes out a directory block.
 808  */
 809 static int write_dir_block(ext2_filsys fs,
 810                            blk64_t *block_nr,
 811                            e2_blkcnt_t blockcnt,
 812                            blk64_t ref_block EXT2FS_ATTR((unused)),
 813                            int ref_offset EXT2FS_ATTR((unused)),
 814                            void *priv_data)
 815 {
 816         struct write_dir_struct *wd = (struct write_dir_struct *) priv_data;
 817         char    *dir, *buf = 0;
 818
 819 #ifdef REHASH_DEBUG
 820         printf("%u: write_dir_block %lld:%lld", wd->ino, blockcnt, *block_nr);
 821 #endif
 822         if ((*block_nr == 0) || (blockcnt < 0)) {
 823 #ifdef REHASH_DEBUG
 824                 printf(" - skip\n");
 825 #endif
 826                 return 0;
 827         }
 828         if (blockcnt < wd->outdir->num)
 829                 dir = wd->outdir->buf + (blockcnt * fs->blocksize);
 830         else if (wd->ctx->lost_and_found == wd->dir) {
 831                 /* Don't release any extra directory blocks for lost+found */
 832                 wd->err = ext2fs_new_dir_block(fs, 0, 0, &buf);
 833                 if (wd->err)
 834                         return BLOCK_ABORT;
 835                 dir = buf;
 836                 wd->outdir->num++;
 837         } else {
 838                 /* Don't free blocks at the end of the directory, they
 839                  * will be truncated by the caller. */
 840 #ifdef REHASH_DEBUG
 841                 printf(" - not freed\n");
 842 #endif
 843                 return 0;
 844         }
 845         wd->err = ext2fs_write_dir_block4(fs, *block_nr, dir, 0, wd->dir);
 846         if (buf)
 847                 ext2fs_free_mem(&buf);
 848
 849 #ifdef REHASH_DEBUG
 850         printf(" - write (%d)\n", wd->err);
 851 #endif
 852         if (wd->err)
 853                 return BLOCK_ABORT;
 854         return 0;
 855 }
 856
 857 static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
 858                                  struct out_dir *outdir,
 859                                  ext2_ino_t ino, struct ext2_inode *inode,
 860                                  int compress)
 861 {
 862         struct write_dir_struct wd;
 863         errcode_t       retval;
 864
 865         retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
 866         if (retval)
 867                 return retval;
 868
 869         wd.outdir = outdir;
 870         wd.err = 0;
 871         wd.ino = ino;
 872         wd.ctx = ctx;
 873         wd.dir = ino;
 874
 875         retval = ext2fs_block_iterate3(fs, ino, 0, NULL,
 876                                        write_dir_block, &wd);
 877         if (retval)
 878                 return retval;
 879         if (wd.err)
 880                 return wd.err;
 881
 882         e2fsck_read_inode(ctx, ino, inode, "rehash_dir");
 883         if (compress)
 884                 inode->i_flags &= ~EXT2_INDEX_FL;
 885         else
 886                 inode->i_flags |= EXT2_INDEX_FL;
 887 #ifdef REHASH_DEBUG
 888         printf("%u: set inode size to %u blocks = %u bytes\n",
 889                ino, outdir->num, outdir->num * fs->blocksize);
 890 #endif
 891         retval = ext2fs_inode_size_set(fs, inode, (ext2_off64_t)outdir->num *
 892                                                    fs->blocksize);
 893         if (retval)
 894                 return retval;
 895
 896         /* ext2fs_punch() calls ext2fs_write_inode() which writes the size */
 897         return ext2fs_punch(fs, ino, inode, NULL, outdir->num, ~0ULL);
 898 }
 899
 900 errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino,
 901                             struct problem_context *pctx)
 902 {
 903         ext2_filsys             fs = ctx->fs;
 904         errcode_t               retval;
 905         struct ext2_inode       inode;
 906         char                    *dir_buf = 0;
 907         struct fill_dir_struct  fd = { NULL, NULL, 0, 0, 0, NULL,
 908                                        0, 0, 0, 0, 0, 0 };
 909         struct out_dir          outdir = { 0, 0, 0, 0 };
 910
 911         e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
 912
 913         if (ext2fs_has_feature_inline_data(fs->super) &&
 914            (inode.i_flags & EXT4_INLINE_DATA_FL))
 915                 return 0;
 916
 917         retval = ext2fs_get_mem(inode.i_size, &dir_buf);
 918         if (retval)
 919                 goto errout;
 920
 921         fd.max_array = inode.i_size / 32;
 922         retval = ext2fs_get_array(sizeof(struct hash_entry),
 923                                   fd.max_array, &fd.harray);
 924         if (retval)
 925                 goto errout;
 926
 927         fd.ino = ino;
 928         fd.ctx = ctx;
 929         fd.buf = dir_buf;
 930         fd.inode = &inode;
 931         fd.dir = ino;
 932         if (!ext2fs_has_feature_dir_index(fs->super) ||
 933             (inode.i_size / fs->blocksize) < 2)
 934                 fd.compress = 1;
 935         fd.parent = 0;
 936
 937 retry_nohash:
 938         /* Read in the entire directory into memory */
 939         retval = ext2fs_block_iterate3(fs, ino, 0, 0,
 940                                        fill_dir_block, &fd);
 941         if (fd.err) {
 942                 retval = fd.err;
 943                 goto errout;
 944         }
 945
 946         /*
 947          * If the entries read are less than a block, then don't index
 948          * the directory
 949          */
 950         if (!fd.compress && (fd.dir_size < (fs->blocksize - 24))) {
 951                 fd.compress = 1;
 952                 fd.dir_size = 0;
 953                 fd.num_array = 0;
 954                 goto retry_nohash;
 955         }
 956
 957 #if 0
 958         printf("%d entries (%d bytes) found in inode %d\n",
 959                fd.num_array, fd.dir_size, ino);
 960 #endif
 961
 962         /* Sort the list */
 963 resort:
 964         if (fd.compress && fd.num_array > 1)
 965                 qsort(fd.harray+2, fd.num_array-2, sizeof(struct hash_entry),
 966                       hash_cmp);
 967         else
 968                 qsort(fd.harray, fd.num_array, sizeof(struct hash_entry),
 969                       hash_cmp);
 970
 971         /*
 972          * Look for duplicates
 973          */
 974         if (duplicate_search_and_fix(ctx, fs, ino, &fd))
 975                 goto resort;
 976
 977         if (ctx->options & E2F_OPT_NO) {
 978                 retval = 0;
 979                 goto errout;
 980         }
 981
 982         /* Sort non-hashed directories by inode number */
 983         if (fd.compress && fd.num_array > 1)
 984                 qsort(fd.harray+2, fd.num_array-2,
 985                       sizeof(struct hash_entry), ino_cmp);
 986
 987         /*
 988          * Copy the directory entries.  In a htree directory these
 989          * will become the leaf nodes.
 990          */
 991         retval = copy_dir_entries(ctx, &fd, &outdir);
 992         if (retval)
 993                 goto errout;
 994
 995         if (!fd.compress) {
 996                 /* Calculate the interior nodes */
 997                 retval = calculate_tree(fs, &outdir, ino, fd.parent,
 998                                         fd.dot_de, fd.dotdot_de);
 999                 if (retval)
1000                         goto errout;
1001         }
1002
1003         retval = write_directory(ctx, fs, &outdir, ino, &inode, fd.compress);
1004         if (retval)
1005                 goto errout;
1006
1007         if (ctx->options & E2F_OPT_CONVERT_BMAP)
1008                 retval = e2fsck_rebuild_extents_later(ctx, ino);
1009         else
1010                 retval = e2fsck_check_rebuild_extents(ctx, ino, &inode, pctx);
1011 errout:
1012         ext2fs_free_mem(&dir_buf);
1013         ext2fs_free_mem(&fd.harray);
1014
1015         free_out_dir(&outdir);
1016         return retval;
1017 }
1018
1019 void e2fsck_rehash_directories(e2fsck_t ctx)
1020 {
1021         struct problem_context  pctx;
1022 #ifdef RESOURCE_TRACK
1023         struct resource_track   rtrack;
1024 #endif
1025         struct dir_info         *dir;
1026         ext2_u32_iterate        iter;
1027         struct dir_info_iter *  dirinfo_iter = 0;
1028         ext2_ino_t              ino;
1029         errcode_t               retval;
1030         int                     cur, max, all_dirs, first = 1;
1031
1032         init_resource_track(&rtrack, ctx->fs->io);
1033         all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
1034
1035         if (!ctx->dirs_to_hash && !all_dirs)
1036                 return;
1037
1038         (void) e2fsck_get_lost_and_found(ctx, 0);
1039
1040         clear_problem_context(&pctx);
1041
1042         cur = 0;
1043         if (all_dirs) {
1044                 dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
1045                 max = e2fsck_get_num_dirinfo(ctx);
1046         } else {
1047                 retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
1048                                                        &iter);
1049                 if (retval) {
1050                         pctx.errcode = retval;
1051                         fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
1052                         return;
1053                 }
1054                 max = ext2fs_u32_list_count(ctx->dirs_to_hash);
1055         }
1056         while (1) {
1057                 if (all_dirs) {
1058                         if ((dir = e2fsck_dir_info_iter(ctx,
1059                                                         dirinfo_iter)) == 0)
1060                                 break;
1061                         ino = dir->ino;
1062                 } else {
1063                         if (!ext2fs_u32_list_iterate(iter, &ino))
1064                                 break;
1065                 }
1066                 if (!ext2fs_test_inode_bitmap2(ctx->inode_dir_map, ino))
1067                         continue;
1068
1069                 pctx.dir = ino;
1070                 if (first) {
1071                         fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
1072                         first = 0;
1073                 }
1074 #if 0
1075                 fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
1076 #endif
1077                 pctx.errcode = e2fsck_rehash_dir(ctx, ino, &pctx);
1078                 if (pctx.errcode) {
1079                         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
1080                         fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
1081                 }
1082                 if (ctx->progress && !ctx->progress_fd)
1083                         e2fsck_simple_progress(ctx, "Rebuilding directory",
1084                                100.0 * (float) (++cur) / (float) max, ino);
1085         }
1086         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
1087         if (all_dirs)
1088                 e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
1089         else
1090                 ext2fs_u32_list_iterate_end(iter);
1091
1092         if (ctx->dirs_to_hash)
1093                 ext2fs_u32_list_free(ctx->dirs_to_hash);
1094         ctx->dirs_to_hash = 0;
1095
1096         print_resource_track(ctx, "Pass 3A", &rtrack, ctx->fs->io);
1097 }