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         int max_array, num_array;
  84         unsigned int dir_size;
  85         int compress;
  86         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         int             num;
 101         int             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;
 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_alg = fs->super->s_def_hash_version;
 145         if ((hash_alg <= EXT2_HASH_TEA) &&
 146             (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
 147                 hash_alg += 3;
 148         /* While the directory block is "hot", index it. */
 149         dir_offset = 0;
 150         while (dir_offset < fs->blocksize) {
 151                 dirent = (struct ext2_dir_entry *) (dir + dir_offset);
 152                 (void) ext2fs_get_rec_len(fs, dirent, &rec_len);
 153                 name_len = ext2fs_dirent_name_len(dirent);
 154                 if (((dir_offset + rec_len) > fs->blocksize) ||
 155                     (rec_len < 8) ||
 156                     ((rec_len % 4) != 0) ||
 157                     (name_len + 8 > rec_len)) {
 158                         fd->err = EXT2_ET_DIR_CORRUPTED;
 159                         return BLOCK_ABORT;
 160                 }
 161                 dir_offset += rec_len;
 162                 if (dirent->inode == 0)
 163                         continue;
 164                 if (!fd->compress && (name_len == 1) &&
 165                     (dirent->name[0] == '.')) {
 166                         fd->dot_de = dirent;
 167                         continue;
 168                 }
 169                 if (!fd->compress && (name_len == 2) &&
 170                     (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
 171                         fd->parent = dirent->inode;
 172                         fd->dotdot_de = dirent;
 173                         continue;
 174                 }
 175                 if (fd->num_array >= fd->max_array) {
 176                         new_array = realloc(fd->harray,
 177                             sizeof(struct hash_entry) * (fd->max_array+500));
 178                         if (!new_array) {
 179                                 fd->err = ENOMEM;
 180                                 return BLOCK_ABORT;
 181                         }
 182                         fd->harray = new_array;
 183                         fd->max_array += 500;
 184                 }
 185                 ent = fd->harray + fd->num_array++;
 186                 ent->dir = dirent;
 187                 fd->dir_size += EXT2_DIR_REC_LEN(dirent);
 188                 ent->ino = dirent->inode;
 189                 if (fd->compress)
 190                         ent->hash = ent->minor_hash = 0;
 191                 else {
 192                         fd->err = ext2fs_dirhash(hash_alg, dirent->name,
 193                                                  name_len,
 194                                                  fs->super->s_hash_seed,
 195                                                  &ent->hash, &ent->minor_hash);
 196                         if (fd->err)
 197                                 return BLOCK_ABORT;
 198                 }
 199         }
 200
 201         return 0;
 202 }
 203
 204 /* Used for sorting the hash entry */
 205 static EXT2_QSORT_TYPE ino_cmp(const void *a, const void *b)
 206 {
 207         const struct hash_entry *he_a = (const struct hash_entry *) a;
 208         const struct hash_entry *he_b = (const struct hash_entry *) b;
 209
 210         return (he_a->ino - he_b->ino);
 211 }
 212
 213 /* Used for sorting the hash entry */
 214 static EXT2_QSORT_TYPE name_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         unsigned int he_a_len, he_b_len, min_len;
 219         int     ret;
 220
 221         he_a_len = ext2fs_dirent_name_len(he_a->dir);
 222         he_b_len = ext2fs_dirent_name_len(he_b->dir);
 223         min_len = he_a_len;
 224         if (min_len > he_b_len)
 225                 min_len = he_b_len;
 226
 227         ret = memcmp(he_a->dir->name, he_b->dir->name, min_len);
 228         if (ret == 0) {
 229                 if (he_a_len > he_b_len)
 230                         ret = 1;
 231                 else if (he_a_len < he_b_len)
 232                         ret = -1;
 233                 else
 234                         ret = he_b->dir->inode - he_a->dir->inode;
 235         }
 236         return ret;
 237 }
 238
 239 /* Used for sorting the hash entry */
 240 static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
 241 {
 242         const struct hash_entry *he_a = (const struct hash_entry *) a;
 243         const struct hash_entry *he_b = (const struct hash_entry *) b;
 244         int     ret;
 245
 246         if (he_a->hash > he_b->hash)
 247                 ret = 1;
 248         else if (he_a->hash < he_b->hash)
 249                 ret = -1;
 250         else {
 251                 if (he_a->minor_hash > he_b->minor_hash)
 252                         ret = 1;
 253                 else if (he_a->minor_hash < he_b->minor_hash)
 254                         ret = -1;
 255                 else
 256                         ret = name_cmp(a, b);
 257         }
 258         return ret;
 259 }
 260
 261 static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
 262                                 int blocks)
 263 {
 264         void                    *new_mem;
 265
 266         if (outdir->max) {
 267                 new_mem = realloc(outdir->buf, blocks * fs->blocksize);
 268                 if (!new_mem)
 269                         return ENOMEM;
 270                 outdir->buf = new_mem;
 271                 new_mem = realloc(outdir->hashes,
 272                                   blocks * sizeof(ext2_dirhash_t));
 273                 if (!new_mem)
 274                         return ENOMEM;
 275                 outdir->hashes = new_mem;
 276         } else {
 277                 outdir->buf = malloc(blocks * fs->blocksize);
 278                 outdir->hashes = malloc(blocks * sizeof(ext2_dirhash_t));
 279                 outdir->num = 0;
 280         }
 281         outdir->max = blocks;
 282         return 0;
 283 }
 284
 285 static void free_out_dir(struct out_dir *outdir)
 286 {
 287         free(outdir->buf);
 288         free(outdir->hashes);
 289         outdir->max = 0;
 290         outdir->num =0;
 291 }
 292
 293 static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
 294                          char ** ret)
 295 {
 296         errcode_t       retval;
 297
 298         if (outdir->num >= outdir->max) {
 299                 retval = alloc_size_dir(fs, outdir, outdir->max + 50);
 300                 if (retval)
 301                         return retval;
 302         }
 303         *ret = outdir->buf + (outdir->num++ * fs->blocksize);
 304         memset(*ret, 0, fs->blocksize);
 305         return 0;
 306 }
 307
 308 /*
 309  * This function is used to make a unique filename.  We do this by
 310  * appending ~0, and then incrementing the number.  However, we cannot
 311  * expand the length of the filename beyond the padding available in
 312  * the directory entry.
 313  */
 314 static void mutate_name(char *str, unsigned int *len)
 315 {
 316         int i;
 317         unsigned int l = *len;
 318
 319         /*
 320          * First check to see if it looks the name has been mutated
 321          * already
 322          */
 323         for (i = l-1; i > 0; i--) {
 324                 if (!isdigit(str[i]))
 325                         break;
 326         }
 327         if ((i == (int)l - 1) || (str[i] != '~')) {
 328                 if (((l-1) & 3) < 2)
 329                         l += 2;
 330                 else
 331                         l = (l+3) & ~3;
 332                 str[l-2] = '~';
 333                 str[l-1] = '0';
 334                 *len = l;
 335                 return;
 336         }
 337         for (i = l-1; i >= 0; i--) {
 338                 if (isdigit(str[i])) {
 339                         if (str[i] == '9')
 340                                 str[i] = '0';
 341                         else {
 342                                 str[i]++;
 343                                 return;
 344                         }
 345                         continue;
 346                 }
 347                 if (i == 1) {
 348                         if (str[0] == 'z')
 349                                 str[0] = 'A';
 350                         else if (str[0] == 'Z') {
 351                                 str[0] = '~';
 352                                 str[1] = '0';
 353                         } else
 354                                 str[0]++;
 355                 } else if (i > 0) {
 356                         str[i] = '1';
 357                         str[i-1] = '~';
 358                 } else {
 359                         if (str[0] == '~')
 360                                 str[0] = 'a';
 361                         else
 362                                 str[0]++;
 363                 }
 364                 break;
 365         }
 366 }
 367
 368 static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
 369                                     ext2_ino_t ino,
 370                                     struct fill_dir_struct *fd)
 371 {
 372         struct problem_context  pctx;
 373         struct hash_entry       *ent, *prev;
 374         int                     i, j;
 375         int                     fixed = 0;
 376         char                    new_name[256];
 377         unsigned int            new_len;
 378         int                     hash_alg;
 379
 380         clear_problem_context(&pctx);
 381         pctx.ino = ino;
 382
 383         hash_alg = fs->super->s_def_hash_version;
 384         if ((hash_alg <= EXT2_HASH_TEA) &&
 385             (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
 386                 hash_alg += 3;
 387
 388         for (i=1; i < fd->num_array; i++) {
 389                 ent = fd->harray + i;
 390                 prev = ent - 1;
 391                 if (!ent->dir->inode ||
 392                     (ext2fs_dirent_name_len(ent->dir) !=
 393                      ext2fs_dirent_name_len(prev->dir)) ||
 394                     memcmp(ent->dir->name, prev->dir->name,
 395                              ext2fs_dirent_name_len(ent->dir)))
 396                         continue;
 397                 pctx.dirent = ent->dir;
 398                 if ((ent->dir->inode == prev->dir->inode) &&
 399                     fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
 400                         e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
 401                         ent->dir->inode = 0;
 402                         fixed++;
 403                         continue;
 404                 }
 405                 new_len = ext2fs_dirent_name_len(ent->dir);
 406                 memcpy(new_name, ent->dir->name, new_len);
 407                 mutate_name(new_name, &new_len);
 408                 for (j=0; j < fd->num_array; j++) {
 409                         if ((i==j) ||
 410                             (new_len !=
 411                              (unsigned) ext2fs_dirent_name_len(fd->harray[j].dir)) ||
 412                             memcmp(new_name, fd->harray[j].dir->name, new_len))
 413                                 continue;
 414                         mutate_name(new_name, &new_len);
 415
 416                         j = -1;
 417                 }
 418                 new_name[new_len] = 0;
 419                 pctx.str = new_name;
 420                 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
 421                         memcpy(ent->dir->name, new_name, new_len);
 422                         ext2fs_dirent_set_name_len(ent->dir, new_len);
 423                         ext2fs_dirhash(hash_alg, new_name, new_len,
 424                                        fs->super->s_hash_seed,
 425                                        &ent->hash, &ent->minor_hash);
 426                         fixed++;
 427                 }
 428         }
 429         return fixed;
 430 }
 431
 432
 433 static errcode_t copy_dir_entries(e2fsck_t ctx,
 434                                   struct fill_dir_struct *fd,
 435                                   struct out_dir *outdir)
 436 {
 437         ext2_filsys             fs = ctx->fs;
 438         errcode_t               retval;
 439         char                    *block_start;
 440         struct hash_entry       *ent;
 441         struct ext2_dir_entry   *dirent;
 442         unsigned int            rec_len, prev_rec_len, left, slack, offset;
 443         int                     i;
 444         ext2_dirhash_t          prev_hash;
 445         int                     csum_size = 0;
 446         struct                  ext2_dir_entry_tail *t;
 447
 448         if (ctx->htree_slack_percentage == 255) {
 449                 profile_get_uint(ctx->profile, "options",
 450                                  "indexed_dir_slack_percentage",
 451                                  0, 20,
 452                                  &ctx->htree_slack_percentage);
 453                 if (ctx->htree_slack_percentage > 100)
 454                         ctx->htree_slack_percentage = 20;
 455         }
 456
 457         if (ext2fs_has_feature_metadata_csum(fs->super))
 458                 csum_size = sizeof(struct ext2_dir_entry_tail);
 459
 460         outdir->max = 0;
 461         retval = alloc_size_dir(fs, outdir,
 462                                 (fd->dir_size / fs->blocksize) + 2);
 463         if (retval)
 464                 return retval;
 465         outdir->num = fd->compress ? 0 : 1;
 466         offset = 0;
 467         outdir->hashes[0] = 0;
 468         prev_hash = 1;
 469         if ((retval = get_next_block(fs, outdir, &block_start)))
 470                 return retval;
 471         dirent = (struct ext2_dir_entry *) block_start;
 472         prev_rec_len = 0;
 473         rec_len = 0;
 474         left = fs->blocksize - csum_size;
 475         slack = fd->compress ? 12 :
 476                 ((fs->blocksize - csum_size) * ctx->htree_slack_percentage)/100;
 477         if (slack < 12)
 478                 slack = 12;
 479         for (i = 0; i < fd->num_array; i++) {
 480                 ent = fd->harray + i;
 481                 if (ent->dir->inode == 0)
 482                         continue;
 483                 rec_len = EXT2_DIR_REC_LEN(ent->dir);
 484                 if (rec_len > left) {
 485                         if (left) {
 486                                 left += prev_rec_len;
 487                                 retval = ext2fs_set_rec_len(fs, left, dirent);
 488                                 if (retval)
 489                                         return retval;
 490                         }
 491                         if (csum_size) {
 492                                 t = EXT2_DIRENT_TAIL(block_start,
 493                                                      fs->blocksize);
 494                                 ext2fs_initialize_dirent_tail(fs, t);
 495                         }
 496                         if ((retval = get_next_block(fs, outdir,
 497                                                       &block_start)))
 498                                 return retval;
 499                         offset = 0;
 500                 }
 501                 left = (fs->blocksize - csum_size) - offset;
 502                 dirent = (struct ext2_dir_entry *) (block_start + offset);
 503                 if (offset == 0) {
 504                         if (ent->hash == prev_hash)
 505                                 outdir->hashes[outdir->num-1] = ent->hash | 1;
 506                         else
 507                                 outdir->hashes[outdir->num-1] = ent->hash;
 508                 }
 509                 dirent->inode = ent->dir->inode;
 510                 ext2fs_dirent_set_name_len(dirent,
 511                                            ext2fs_dirent_name_len(ent->dir));
 512                 ext2fs_dirent_set_file_type(dirent,
 513                                             ext2fs_dirent_file_type(ent->dir));
 514                 retval = ext2fs_set_rec_len(fs, rec_len, dirent);
 515                 if (retval)
 516                         return retval;
 517                 prev_rec_len = rec_len;
 518                 memcpy(dirent->name, ent->dir->name, rec_len);
 519                 offset += rec_len;
 520                 left -= rec_len;
 521                 if (left < slack) {
 522                         prev_rec_len += left;
 523                         retval = ext2fs_set_rec_len(fs, prev_rec_len, dirent);
 524                         if (retval)
 525                                 return retval;
 526                         offset += left;
 527                         left = 0;
 528                 }
 529                 prev_hash = ent->hash;
 530         }
 531         if (left)
 532                 retval = ext2fs_set_rec_len(fs, rec_len + left, dirent);
 533         if (csum_size) {
 534                 t = EXT2_DIRENT_TAIL(block_start, fs->blocksize);
 535                 ext2fs_initialize_dirent_tail(fs, t);
 536         }
 537
 538         return retval;
 539 }
 540
 541
 542 static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
 543                                         ext2_ino_t ino, ext2_ino_t parent,
 544                                         struct ext2_dir_entry *dot_de,
 545                                         struct ext2_dir_entry *dotdot_de)
 546 {
 547         struct ext2_dir_entry           *dirent;
 548         struct ext2_dx_root_info        *root;
 549         struct ext2_dx_countlimit       *limits;
 550         int                             csum_size = 0;
 551         int                             offset;
 552         int                             rec_len;
 553
 554         memset(buf, 0, fs->blocksize);
 555         dirent = (struct ext2_dir_entry *) buf;
 556         dirent->inode = ino;
 557
 558         dirent->name_len = dot_de->name_len;
 559         offset = rec_len = dirent->rec_len = dot_de->rec_len;
 560         memcpy(dirent->name, dot_de->name, rec_len);
 561
 562         dirent = EXT2_NEXT_DIRENT(dirent);
 563         /* set to jump over the index block */
 564
 565         dirent->inode = parent;
 566
 567         dirent->name_len = dotdot_de->name_len;
 568         dirent->rec_len = fs->blocksize - rec_len;
 569         rec_len = EXT2_DIR_REC_LEN(dotdot_de);
 570         memcpy(dirent->name, dotdot_de->name, rec_len);
 571         offset += rec_len;
 572
 573         root = (struct ext2_dx_root_info *)(buf + offset);
 574         root->reserved_zero = 0;
 575         root->hash_version = fs->super->s_def_hash_version;
 576         root->info_length = sizeof(*root);
 577         root->indirect_levels = 0;
 578         root->unused_flags = 0;
 579         offset += root->info_length;
 580
 581         if (ext2fs_has_feature_metadata_csum(fs->super))
 582                 csum_size = sizeof(struct ext2_dx_tail);
 583
 584         limits = (struct ext2_dx_countlimit *) (buf + offset);
 585         limits->limit = (fs->blocksize - (offset + csum_size)) /
 586                         sizeof(struct ext2_dx_entry);
 587         limits->count = 0;
 588
 589         return root;
 590 }
 591
 592
 593 static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
 594 {
 595         struct ext2_dir_entry           *dir;
 596         struct ext2_dx_countlimit       *limits;
 597         int                             csum_size = 0;
 598
 599         memset(buf, 0, fs->blocksize);
 600         dir = (struct ext2_dir_entry *) buf;
 601         dir->inode = 0;
 602         (void) ext2fs_set_rec_len(fs, fs->blocksize, dir);
 603
 604         if (ext2fs_has_feature_metadata_csum(fs->super))
 605                 csum_size = sizeof(struct ext2_dx_tail);
 606
 607         limits = (struct ext2_dx_countlimit *) (buf+8);
 608         limits->limit = (fs->blocksize - (8 + csum_size)) /
 609                         sizeof(struct ext2_dx_entry);
 610         limits->count = 0;
 611
 612         return (struct ext2_dx_entry *) limits;
 613 }
 614
 615 static int alloc_blocks(ext2_filsys fs,
 616                         struct ext2_dx_countlimit **limit,
 617                         struct ext2_dx_entry **prev_ent,
 618                         struct ext2_dx_entry **next_ent,
 619                         int *prev_offset, int *next_offset,
 620                         struct out_dir *outdir, int i,
 621                         int *prev_count, int *next_count)
 622 {
 623         errcode_t       retval;
 624         char            *block_start;
 625
 626         if (*limit)
 627                 (*limit)->limit = (*limit)->count =
 628                         ext2fs_cpu_to_le16((*limit)->limit);
 629         *prev_ent = (struct ext2_dx_entry *) (outdir->buf + *prev_offset);
 630         (*prev_ent)->block = ext2fs_cpu_to_le32(outdir->num);
 631
 632         if (i != 1)
 633                 (*prev_ent)->hash =
 634                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 635
 636         retval = get_next_block(fs, outdir, &block_start);
 637         if (retval)
 638                 return retval;
 639
 640         *next_ent = set_int_node(fs, block_start);
 641         *limit = (struct ext2_dx_countlimit *)(*next_ent);
 642         if (next_offset)
 643                 *next_offset = ((char *) *next_ent - outdir->buf);
 644
 645         *next_count = (*limit)->limit;
 646         (*prev_offset) += sizeof(struct ext2_dx_entry);
 647         (*prev_count)--;
 648
 649         return 0;
 650 }
 651
 652 /*
 653  * This function takes the leaf nodes which have been written in
 654  * outdir, and populates the root node and any necessary interior nodes.
 655  */
 656 static errcode_t calculate_tree(ext2_filsys fs,
 657                                 struct out_dir *outdir,
 658                                 ext2_ino_t ino,
 659                                 ext2_ino_t parent,
 660                                 struct ext2_dir_entry *dot_de,
 661                                 struct ext2_dir_entry *dotdot_de)
 662 {
 663         struct ext2_dx_root_info        *root_info;
 664         struct ext2_dx_entry            *root, *int_ent, *dx_ent = 0;
 665         struct ext2_dx_countlimit       *root_limit, *int_limit, *limit;
 666         errcode_t                       retval;
 667         int                             i, c1, c2, c3, nblks;
 668         int                             limit_offset, int_offset, root_offset;
 669
 670         root_info = set_root_node(fs, outdir->buf, ino, parent, dot_de,
 671                                   dotdot_de);
 672
 673         root_offset = limit_offset = ((char *) root_info - outdir->buf) +
 674                 root_info->info_length;
 675         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 676         c1 = root_limit->limit;
 677         nblks = outdir->num;
 678
 679         /* Write out the pointer blocks */
 680         if (nblks - 1 <= c1) {
 681                 /* Just write out the root block, and we're done */
 682                 root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
 683                 for (i=1; i < nblks; i++) {
 684                         root->block = ext2fs_cpu_to_le32(i);
 685                         if (i != 1)
 686                                 root->hash =
 687                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 688                         root++;
 689                         c1--;
 690                 }
 691         } else if (nblks - 1 <= ext2fs_htree_intnode_maxrecs(fs, c1)) {
 692                 c2 = 0;
 693                 limit = NULL;
 694                 root_info->indirect_levels = 1;
 695                 for (i=1; i < nblks; i++) {
 696                         if (c2 == 0 && c1 == 0)
 697                                 return ENOSPC;
 698                         if (c2 == 0) {
 699                                 retval = alloc_blocks(fs, &limit, &root,
 700                                                       &dx_ent, &root_offset,
 701                                                       NULL, outdir, i, &c1,
 702                                                       &c2);
 703                                 if (retval)
 704                                         return retval;
 705                         }
 706                         dx_ent->block = ext2fs_cpu_to_le32(i);
 707                         if (c2 != limit->limit)
 708                                 dx_ent->hash =
 709                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 710                         dx_ent++;
 711                         c2--;
 712                 }
 713                 limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
 714                 limit->limit = ext2fs_cpu_to_le16(limit->limit);
 715         } else {
 716                 c2 = 0;
 717                 c3 = 0;
 718                 limit = NULL;
 719                 int_limit = 0;
 720                 root_info->indirect_levels = 2;
 721                 for (i = 1; i < nblks; i++) {
 722                         if (c3 == 0 && c2 == 0 && c1 == 0)
 723                                 return ENOSPC;
 724                         if (c3 == 0 && c2 == 0) {
 725                                 retval = alloc_blocks(fs, &int_limit, &root,
 726                                                       &int_ent, &root_offset,
 727                                                       &int_offset, outdir, i,
 728                                                       &c1, &c2);
 729                                 if (retval)
 730                                         return retval;
 731                         }
 732                         if (c3 == 0) {
 733                                 retval = alloc_blocks(fs, &limit, &int_ent,
 734                                                       &dx_ent, &int_offset,
 735                                                       NULL, outdir, i, &c2,
 736                                                       &c3);
 737                                 if (retval)
 738                                         return retval;
 739
 740                         }
 741                         dx_ent->block = ext2fs_cpu_to_le32(i);
 742                         if (c3 != limit->limit)
 743                                 dx_ent->hash =
 744                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 745                         dx_ent++;
 746                         c3--;
 747                 }
 748                 int_limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
 749                 int_limit->limit = ext2fs_cpu_to_le16(limit->limit);
 750
 751                 limit->count = ext2fs_cpu_to_le16(limit->limit - c3);
 752                 limit->limit = ext2fs_cpu_to_le16(limit->limit);
 753
 754         }
 755         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 756         root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
 757         root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
 758
 759         return 0;
 760 }
 761
 762 struct write_dir_struct {
 763         struct out_dir *outdir;
 764         errcode_t       err;
 765         ext2_ino_t      ino;
 766         e2fsck_t        ctx;
 767         ext2_ino_t      dir;
 768 };
 769
 770 /*
 771  * Helper function which writes out a directory block.
 772  */
 773 static int write_dir_block(ext2_filsys fs,
 774                            blk64_t *block_nr,
 775                            e2_blkcnt_t blockcnt,
 776                            blk64_t ref_block EXT2FS_ATTR((unused)),
 777                            int ref_offset EXT2FS_ATTR((unused)),
 778                            void *priv_data)
 779 {
 780         struct write_dir_struct *wd = (struct write_dir_struct *) priv_data;
 781         char    *dir, *buf = 0;
 782
 783 #ifdef REHASH_DEBUG
 784         printf("%u: write_dir_block %lld:%lld", wd->ino, blockcnt, *block_nr);
 785 #endif
 786         if ((*block_nr == 0) || (blockcnt < 0)) {
 787 #ifdef REHASH_DEBUG
 788                 printf(" - skip\n");
 789 #endif
 790                 return 0;
 791         }
 792         if (blockcnt < wd->outdir->num)
 793                 dir = wd->outdir->buf + (blockcnt * fs->blocksize);
 794         else if (wd->ctx->lost_and_found == wd->dir) {
 795                 /* Don't release any extra directory blocks for lost+found */
 796                 wd->err = ext2fs_new_dir_block(fs, 0, 0, &buf);
 797                 if (wd->err)
 798                         return BLOCK_ABORT;
 799                 dir = buf;
 800                 wd->outdir->num++;
 801         } else {
 802                 /* Don't free blocks at the end of the directory, they
 803                  * will be truncated by the caller. */
 804 #ifdef REHASH_DEBUG
 805                 printf(" - not freed\n");
 806 #endif
 807                 return 0;
 808         }
 809         wd->err = ext2fs_write_dir_block4(fs, *block_nr, dir, 0, wd->dir);
 810         if (buf)
 811                 ext2fs_free_mem(&buf);
 812
 813 #ifdef REHASH_DEBUG
 814         printf(" - write (%d)\n", wd->err);
 815 #endif
 816         if (wd->err)
 817                 return BLOCK_ABORT;
 818         return 0;
 819 }
 820
 821 static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
 822                                  struct out_dir *outdir,
 823                                  ext2_ino_t ino, struct ext2_inode *inode,
 824                                  int compress)
 825 {
 826         struct write_dir_struct wd;
 827         errcode_t       retval;
 828
 829         retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
 830         if (retval)
 831                 return retval;
 832
 833         wd.outdir = outdir;
 834         wd.err = 0;
 835         wd.ino = ino;
 836         wd.ctx = ctx;
 837         wd.dir = ino;
 838
 839         retval = ext2fs_block_iterate3(fs, ino, 0, NULL,
 840                                        write_dir_block, &wd);
 841         if (retval)
 842                 return retval;
 843         if (wd.err)
 844                 return wd.err;
 845
 846         e2fsck_read_inode(ctx, ino, inode, "rehash_dir");
 847         if (compress)
 848                 inode->i_flags &= ~EXT2_INDEX_FL;
 849         else
 850                 inode->i_flags |= EXT2_INDEX_FL;
 851 #ifdef REHASH_DEBUG
 852         printf("%u: set inode size to %u blocks = %u bytes\n",
 853                ino, outdir->num, outdir->num * fs->blocksize);
 854 #endif
 855         retval = ext2fs_inode_size_set(fs, inode, (ext2_off64_t)outdir->num *
 856                                                    fs->blocksize);
 857         if (retval)
 858                 return retval;
 859
 860         /* ext2fs_punch() calls ext2fs_write_inode() which writes the size */
 861         return ext2fs_punch(fs, ino, inode, NULL, outdir->num, ~0ULL);
 862 }
 863
 864 errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino,
 865                             struct problem_context *pctx)
 866 {
 867         ext2_filsys             fs = ctx->fs;
 868         errcode_t               retval;
 869         struct ext2_inode       inode;
 870         char                    *dir_buf = 0;
 871         struct fill_dir_struct  fd = { NULL, NULL, 0, 0, 0, NULL,
 872                                        0, 0, 0, 0, 0, 0 };
 873         struct out_dir          outdir = { 0, 0, 0, 0 };
 874
 875         e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
 876
 877         if (ext2fs_has_feature_inline_data(fs->super) &&
 878            (inode.i_flags & EXT4_INLINE_DATA_FL))
 879                 return 0;
 880
 881         retval = ENOMEM;
 882         dir_buf = malloc(inode.i_size);
 883         if (!dir_buf)
 884                 goto errout;
 885
 886         fd.max_array = inode.i_size / 32;
 887         fd.harray = malloc(fd.max_array * sizeof(struct hash_entry));
 888         if (!fd.harray)
 889                 goto errout;
 890
 891         fd.ino = ino;
 892         fd.ctx = ctx;
 893         fd.buf = dir_buf;
 894         fd.inode = &inode;
 895         fd.dir = ino;
 896         if (!ext2fs_has_feature_dir_index(fs->super) ||
 897             (inode.i_size / fs->blocksize) < 2)
 898                 fd.compress = 1;
 899         fd.parent = 0;
 900
 901 retry_nohash:
 902         /* Read in the entire directory into memory */
 903         retval = ext2fs_block_iterate3(fs, ino, 0, 0,
 904                                        fill_dir_block, &fd);
 905         if (fd.err) {
 906                 retval = fd.err;
 907                 goto errout;
 908         }
 909
 910         /*
 911          * If the entries read are less than a block, then don't index
 912          * the directory
 913          */
 914         if (!fd.compress && (fd.dir_size < (fs->blocksize - 24))) {
 915                 fd.compress = 1;
 916                 fd.dir_size = 0;
 917                 fd.num_array = 0;
 918                 goto retry_nohash;
 919         }
 920
 921 #if 0
 922         printf("%d entries (%d bytes) found in inode %d\n",
 923                fd.num_array, fd.dir_size, ino);
 924 #endif
 925
 926         /* Sort the list */
 927 resort:
 928         if (fd.compress && fd.num_array > 1)
 929                 qsort(fd.harray+2, fd.num_array-2, sizeof(struct hash_entry),
 930                       hash_cmp);
 931         else
 932                 qsort(fd.harray, fd.num_array, sizeof(struct hash_entry),
 933                       hash_cmp);
 934
 935         /*
 936          * Look for duplicates
 937          */
 938         if (duplicate_search_and_fix(ctx, fs, ino, &fd))
 939                 goto resort;
 940
 941         if (ctx->options & E2F_OPT_NO) {
 942                 retval = 0;
 943                 goto errout;
 944         }
 945
 946         /* Sort non-hashed directories by inode number */
 947         if (fd.compress && fd.num_array > 1)
 948                 qsort(fd.harray+2, fd.num_array-2,
 949                       sizeof(struct hash_entry), ino_cmp);
 950
 951         /*
 952          * Copy the directory entries.  In a htree directory these
 953          * will become the leaf nodes.
 954          */
 955         retval = copy_dir_entries(ctx, &fd, &outdir);
 956         if (retval)
 957                 goto errout;
 958
 959         if (!fd.compress) {
 960                 /* Calculate the interior nodes */
 961                 retval = calculate_tree(fs, &outdir, ino, fd.parent,
 962                                         fd.dot_de, fd.dotdot_de);
 963                 if (retval)
 964                         goto errout;
 965         }
 966
 967         retval = write_directory(ctx, fs, &outdir, ino, &inode, fd.compress);
 968         if (retval)
 969                 goto errout;
 970
 971         if (ctx->options & E2F_OPT_CONVERT_BMAP)
 972                 retval = e2fsck_rebuild_extents_later(ctx, ino);
 973         else
 974                 retval = e2fsck_check_rebuild_extents(ctx, ino, &inode, pctx);
 975 errout:
 976         free(dir_buf);
 977         free(fd.harray);
 978
 979         free_out_dir(&outdir);
 980         return retval;
 981 }
 982
 983 void e2fsck_rehash_directories(e2fsck_t ctx)
 984 {
 985         struct problem_context  pctx;
 986 #ifdef RESOURCE_TRACK
 987         struct resource_track   rtrack;
 988 #endif
 989         struct dir_info         *dir;
 990         ext2_u32_iterate        iter;
 991         struct dir_info_iter *  dirinfo_iter = 0;
 992         ext2_ino_t              ino;
 993         errcode_t               retval;
 994         int                     cur, max, all_dirs, first = 1;
 995
 996         init_resource_track(&rtrack, ctx->fs->io);
 997         all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
 998
 999         if (!ctx->dirs_to_hash && !all_dirs)
1000                 return;
1001
1002         (void) e2fsck_get_lost_and_found(ctx, 0);
1003
1004         clear_problem_context(&pctx);
1005
1006         cur = 0;
1007         if (all_dirs) {
1008                 dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
1009                 max = e2fsck_get_num_dirinfo(ctx);
1010         } else {
1011                 retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
1012                                                        &iter);
1013                 if (retval) {
1014                         pctx.errcode = retval;
1015                         fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
1016                         return;
1017                 }
1018                 max = ext2fs_u32_list_count(ctx->dirs_to_hash);
1019         }
1020         while (1) {
1021                 if (all_dirs) {
1022                         if ((dir = e2fsck_dir_info_iter(ctx,
1023                                                         dirinfo_iter)) == 0)
1024                                 break;
1025                         ino = dir->ino;
1026                 } else {
1027                         if (!ext2fs_u32_list_iterate(iter, &ino))
1028                                 break;
1029                 }
1030
1031                 pctx.dir = ino;
1032                 if (first) {
1033                         fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
1034                         first = 0;
1035                 }
1036 #if 0
1037                 fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
1038 #endif
1039                 pctx.errcode = e2fsck_rehash_dir(ctx, ino, &pctx);
1040                 if (pctx.errcode) {
1041                         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
1042                         fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
1043                 }
1044                 if (ctx->progress && !ctx->progress_fd)
1045                         e2fsck_simple_progress(ctx, "Rebuilding directory",
1046                                100.0 * (float) (++cur) / (float) max, ino);
1047         }
1048         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
1049         if (all_dirs)
1050                 e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
1051         else
1052                 ext2fs_u32_list_iterate_end(iter);
1053
1054         if (ctx->dirs_to_hash)
1055                 ext2fs_u32_list_free(ctx->dirs_to_hash);
1056         ctx->dirs_to_hash = 0;
1057
1058         print_resource_track(ctx, "Pass 3A", &rtrack, ctx->fs->io);
1059 }