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 struct fill_dir_struct {
  75         char *buf;
  76         struct ext2_inode *inode;
  77         int err;
  78         e2fsck_t ctx;
  79         struct hash_entry *harray;
  80         int max_array, num_array;
  81         int dir_size;
  82         int compress;
  83         ino_t parent;
  84         ext2_ino_t dir;
  85 };
  86
  87 struct hash_entry {
  88         ext2_dirhash_t  hash;
  89         ext2_dirhash_t  minor_hash;
  90         ino_t           ino;
  91         struct ext2_dir_entry   *dir;
  92 };
  93
  94 struct out_dir {
  95         int             num;
  96         int             max;
  97         char            *buf;
  98         ext2_dirhash_t  *hashes;
  99 };
 100
 101 static int fill_dir_block(ext2_filsys fs,
 102                           blk64_t *block_nr,
 103                           e2_blkcnt_t blockcnt,
 104                           blk64_t ref_block EXT2FS_ATTR((unused)),
 105                           int ref_offset EXT2FS_ATTR((unused)),
 106                           void *priv_data)
 107 {
 108         struct fill_dir_struct  *fd = (struct fill_dir_struct *) priv_data;
 109         struct hash_entry       *new_array, *ent;
 110         struct ext2_dir_entry   *dirent;
 111         char                    *dir;
 112         unsigned int            offset, dir_offset, rec_len;
 113         int                     hash_alg;
 114
 115         if (blockcnt < 0)
 116                 return 0;
 117
 118         offset = blockcnt * fs->blocksize;
 119         if (offset + fs->blocksize > fd->inode->i_size) {
 120                 fd->err = EXT2_ET_DIR_CORRUPTED;
 121                 return BLOCK_ABORT;
 122         }
 123         dir = (fd->buf+offset);
 124         if (HOLE_BLKADDR(*block_nr)) {
 125                 memset(dir, 0, fs->blocksize);
 126                 dirent = (struct ext2_dir_entry *) dir;
 127                 (void) ext2fs_set_rec_len(fs, fs->blocksize, dirent);
 128         } else {
 129                 fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
 130                 fd->err = ext2fs_read_dir_block4(fs, *block_nr, dir, 0,
 131                                                  fd->dir);
 132                 fs->flags &= ~EXT2_FLAG_IGNORE_CSUM_ERRORS;
 133                 if (fd->err)
 134                         return BLOCK_ABORT;
 135         }
 136         hash_alg = fs->super->s_def_hash_version;
 137         if ((hash_alg <= EXT2_HASH_TEA) &&
 138             (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
 139                 hash_alg += 3;
 140         /* While the directory block is "hot", index it. */
 141         dir_offset = 0;
 142         while (dir_offset < fs->blocksize) {
 143                 dirent = (struct ext2_dir_entry *) (dir + dir_offset);
 144                 (void) ext2fs_get_rec_len(fs, dirent, &rec_len);
 145                 if (((dir_offset + rec_len) > fs->blocksize) ||
 146                     (rec_len < 8) ||
 147                     ((rec_len % 4) != 0) ||
 148                     (((dirent->name_len & 0xFF)+8) > rec_len)) {
 149                         fd->err = EXT2_ET_DIR_CORRUPTED;
 150                         return BLOCK_ABORT;
 151                 }
 152                 dir_offset += rec_len;
 153                 if (dirent->inode == 0)
 154                         continue;
 155                 if (!fd->compress && ((dirent->name_len&0xFF) == 1) &&
 156                     (dirent->name[0] == '.'))
 157                         continue;
 158                 if (!fd->compress && ((dirent->name_len&0xFF) == 2) &&
 159                     (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
 160                         fd->parent = dirent->inode;
 161                         continue;
 162                 }
 163                 if (fd->num_array >= fd->max_array) {
 164                         new_array = realloc(fd->harray,
 165                             sizeof(struct hash_entry) * (fd->max_array+500));
 166                         if (!new_array) {
 167                                 fd->err = ENOMEM;
 168                                 return BLOCK_ABORT;
 169                         }
 170                         fd->harray = new_array;
 171                         fd->max_array += 500;
 172                 }
 173                 ent = fd->harray + fd->num_array++;
 174                 ent->dir = dirent;
 175                 fd->dir_size += EXT2_DIR_REC_LEN(dirent->name_len & 0xFF);
 176                 ent->ino = dirent->inode;
 177                 if (fd->compress)
 178                         ent->hash = ent->minor_hash = 0;
 179                 else {
 180                         fd->err = ext2fs_dirhash(hash_alg, dirent->name,
 181                                                  dirent->name_len & 0xFF,
 182                                                  fs->super->s_hash_seed,
 183                                                  &ent->hash, &ent->minor_hash);
 184                         if (fd->err)
 185                                 return BLOCK_ABORT;
 186                 }
 187         }
 188
 189         return 0;
 190 }
 191
 192 /* Used for sorting the hash entry */
 193 static EXT2_QSORT_TYPE ino_cmp(const void *a, const void *b)
 194 {
 195         const struct hash_entry *he_a = (const struct hash_entry *) a;
 196         const struct hash_entry *he_b = (const struct hash_entry *) b;
 197
 198         return (he_a->ino - he_b->ino);
 199 }
 200
 201 /* Used for sorting the hash entry */
 202 static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
 203 {
 204         const struct hash_entry *he_a = (const struct hash_entry *) a;
 205         const struct hash_entry *he_b = (const struct hash_entry *) b;
 206         int     ret;
 207         int     min_len;
 208
 209         min_len = he_a->dir->name_len;
 210         if (min_len > he_b->dir->name_len)
 211                 min_len = he_b->dir->name_len;
 212
 213         ret = strncmp(he_a->dir->name, he_b->dir->name, min_len);
 214         if (ret == 0) {
 215                 if (he_a->dir->name_len > he_b->dir->name_len)
 216                         ret = 1;
 217                 else if (he_a->dir->name_len < he_b->dir->name_len)
 218                         ret = -1;
 219                 else
 220                         ret = he_b->dir->inode - he_a->dir->inode;
 221         }
 222         return ret;
 223 }
 224
 225 /* Used for sorting the hash entry */
 226 static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
 227 {
 228         const struct hash_entry *he_a = (const struct hash_entry *) a;
 229         const struct hash_entry *he_b = (const struct hash_entry *) b;
 230         int     ret;
 231
 232         if (he_a->hash > he_b->hash)
 233                 ret = 1;
 234         else if (he_a->hash < he_b->hash)
 235                 ret = -1;
 236         else {
 237                 if (he_a->minor_hash > he_b->minor_hash)
 238                         ret = 1;
 239                 else if (he_a->minor_hash < he_b->minor_hash)
 240                         ret = -1;
 241                 else
 242                         ret = name_cmp(a, b);
 243         }
 244         return ret;
 245 }
 246
 247 static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
 248                                 int blocks)
 249 {
 250         void                    *new_mem;
 251
 252         if (outdir->max) {
 253                 new_mem = realloc(outdir->buf, blocks * fs->blocksize);
 254                 if (!new_mem)
 255                         return ENOMEM;
 256                 outdir->buf = new_mem;
 257                 new_mem = realloc(outdir->hashes,
 258                                   blocks * sizeof(ext2_dirhash_t));
 259                 if (!new_mem)
 260                         return ENOMEM;
 261                 outdir->hashes = new_mem;
 262         } else {
 263                 outdir->buf = malloc(blocks * fs->blocksize);
 264                 outdir->hashes = malloc(blocks * sizeof(ext2_dirhash_t));
 265                 outdir->num = 0;
 266         }
 267         outdir->max = blocks;
 268         return 0;
 269 }
 270
 271 static void free_out_dir(struct out_dir *outdir)
 272 {
 273         free(outdir->buf);
 274         free(outdir->hashes);
 275         outdir->max = 0;
 276         outdir->num =0;
 277 }
 278
 279 static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
 280                          char ** ret)
 281 {
 282         errcode_t       retval;
 283
 284         if (outdir->num >= outdir->max) {
 285                 retval = alloc_size_dir(fs, outdir, outdir->max + 50);
 286                 if (retval)
 287                         return retval;
 288         }
 289         *ret = outdir->buf + (outdir->num++ * fs->blocksize);
 290         memset(*ret, 0, fs->blocksize);
 291         return 0;
 292 }
 293
 294 /*
 295  * This function is used to make a unique filename.  We do this by
 296  * appending ~0, and then incrementing the number.  However, we cannot
 297  * expand the length of the filename beyond the padding available in
 298  * the directory entry.
 299  */
 300 static void mutate_name(char *str, __u16 *len)
 301 {
 302         int     i;
 303         __u16   l = *len & 0xFF, h = *len & 0xff00;
 304
 305         /*
 306          * First check to see if it looks the name has been mutated
 307          * already
 308          */
 309         for (i = l-1; i > 0; i--) {
 310                 if (!isdigit(str[i]))
 311                         break;
 312         }
 313         if ((i == l-1) || (str[i] != '~')) {
 314                 if (((l-1) & 3) < 2)
 315                         l += 2;
 316                 else
 317                         l = (l+3) & ~3;
 318                 str[l-2] = '~';
 319                 str[l-1] = '0';
 320                 *len = l | h;
 321                 return;
 322         }
 323         for (i = l-1; i >= 0; i--) {
 324                 if (isdigit(str[i])) {
 325                         if (str[i] == '9')
 326                                 str[i] = '0';
 327                         else {
 328                                 str[i]++;
 329                                 return;
 330                         }
 331                         continue;
 332                 }
 333                 if (i == 1) {
 334                         if (str[0] == 'z')
 335                                 str[0] = 'A';
 336                         else if (str[0] == 'Z') {
 337                                 str[0] = '~';
 338                                 str[1] = '0';
 339                         } else
 340                                 str[0]++;
 341                 } else if (i > 0) {
 342                         str[i] = '1';
 343                         str[i-1] = '~';
 344                 } else {
 345                         if (str[0] == '~')
 346                                 str[0] = 'a';
 347                         else
 348                                 str[0]++;
 349                 }
 350                 break;
 351         }
 352 }
 353
 354 static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
 355                                     ext2_ino_t ino,
 356                                     struct fill_dir_struct *fd)
 357 {
 358         struct problem_context  pctx;
 359         struct hash_entry       *ent, *prev;
 360         int                     i, j;
 361         int                     fixed = 0;
 362         char                    new_name[256];
 363         __u16                   new_len;
 364         int                     hash_alg;
 365
 366         clear_problem_context(&pctx);
 367         pctx.ino = ino;
 368
 369         hash_alg = fs->super->s_def_hash_version;
 370         if ((hash_alg <= EXT2_HASH_TEA) &&
 371             (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
 372                 hash_alg += 3;
 373
 374         for (i=1; i < fd->num_array; i++) {
 375                 ent = fd->harray + i;
 376                 prev = ent - 1;
 377                 if (!ent->dir->inode ||
 378                     ((ent->dir->name_len & 0xFF) !=
 379                      (prev->dir->name_len & 0xFF)) ||
 380                     (strncmp(ent->dir->name, prev->dir->name,
 381                              ent->dir->name_len & 0xFF)))
 382                         continue;
 383                 pctx.dirent = ent->dir;
 384                 if ((ent->dir->inode == prev->dir->inode) &&
 385                     fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
 386                         e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
 387                         ent->dir->inode = 0;
 388                         fixed++;
 389                         continue;
 390                 }
 391                 memcpy(new_name, ent->dir->name, ent->dir->name_len & 0xFF);
 392                 new_len = ent->dir->name_len;
 393                 mutate_name(new_name, &new_len);
 394                 for (j=0; j < fd->num_array; j++) {
 395                         if ((i==j) ||
 396                             ((new_len & 0xFF) !=
 397                              (fd->harray[j].dir->name_len & 0xFF)) ||
 398                             (strncmp(new_name, fd->harray[j].dir->name,
 399                                      new_len & 0xFF)))
 400                                 continue;
 401                         mutate_name(new_name, &new_len);
 402
 403                         j = -1;
 404                 }
 405                 new_name[new_len & 0xFF] = 0;
 406                 pctx.str = new_name;
 407                 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
 408                         memcpy(ent->dir->name, new_name, new_len & 0xFF);
 409                         ent->dir->name_len = new_len;
 410                         ext2fs_dirhash(hash_alg, ent->dir->name,
 411                                        ent->dir->name_len & 0xFF,
 412                                        fs->super->s_hash_seed,
 413                                        &ent->hash, &ent->minor_hash);
 414                         fixed++;
 415                 }
 416         }
 417         return fixed;
 418 }
 419
 420
 421 static errcode_t copy_dir_entries(e2fsck_t ctx,
 422                                   struct fill_dir_struct *fd,
 423                                   struct out_dir *outdir,
 424                                   ext2_ino_t ino)
 425 {
 426         ext2_filsys             fs = ctx->fs;
 427         errcode_t               retval;
 428         char                    *block_start;
 429         struct hash_entry       *ent;
 430         struct ext2_dir_entry   *dirent;
 431         unsigned int            rec_len, prev_rec_len;
 432         int                     i, left;
 433         ext2_dirhash_t          prev_hash;
 434         int                     offset, slack;
 435         int                     csum_size = 0;
 436         struct                  ext2_dir_entry_tail *t;
 437
 438         if (ctx->htree_slack_percentage == 255) {
 439                 profile_get_uint(ctx->profile, "options",
 440                                  "indexed_dir_slack_percentage",
 441                                  0, 20,
 442                                  &ctx->htree_slack_percentage);
 443                 if (ctx->htree_slack_percentage > 100)
 444                         ctx->htree_slack_percentage = 20;
 445         }
 446
 447         if (EXT2_HAS_RO_COMPAT_FEATURE(fs->super,
 448                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 449                 csum_size = sizeof(struct ext2_dir_entry_tail);
 450
 451         outdir->max = 0;
 452         retval = alloc_size_dir(fs, outdir,
 453                                 (fd->dir_size / fs->blocksize) + 2);
 454         if (retval)
 455                 return retval;
 456         outdir->num = fd->compress ? 0 : 1;
 457         offset = 0;
 458         outdir->hashes[0] = 0;
 459         prev_hash = 1;
 460         if ((retval = get_next_block(fs, outdir, &block_start)))
 461                 return retval;
 462         dirent = (struct ext2_dir_entry *) block_start;
 463         prev_rec_len = 0;
 464         rec_len = 0;
 465         left = fs->blocksize - csum_size;
 466         slack = fd->compress ? 12 :
 467                 ((fs->blocksize - csum_size) * ctx->htree_slack_percentage)/100;
 468         if (slack < 12)
 469                 slack = 12;
 470         for (i = 0; i < fd->num_array; i++) {
 471                 ent = fd->harray + i;
 472                 if (ent->dir->inode == 0)
 473                         continue;
 474                 rec_len = EXT2_DIR_REC_LEN(ent->dir->name_len & 0xFF);
 475                 if (rec_len > left) {
 476                         if (left) {
 477                                 left += prev_rec_len;
 478                                 retval = ext2fs_set_rec_len(fs, left, dirent);
 479                                 if (retval)
 480                                         return retval;
 481                         }
 482                         if (csum_size) {
 483                                 t = EXT2_DIRENT_TAIL(block_start,
 484                                                      fs->blocksize);
 485                                 ext2fs_initialize_dirent_tail(fs, t);
 486                         }
 487                         if ((retval = get_next_block(fs, outdir,
 488                                                       &block_start)))
 489                                 return retval;
 490                         offset = 0;
 491                 }
 492                 left = (fs->blocksize - csum_size) - offset;
 493                 dirent = (struct ext2_dir_entry *) (block_start + offset);
 494                 if (offset == 0) {
 495                         if (ent->hash == prev_hash)
 496                                 outdir->hashes[outdir->num-1] = ent->hash | 1;
 497                         else
 498                                 outdir->hashes[outdir->num-1] = ent->hash;
 499                 }
 500                 dirent->inode = ent->dir->inode;
 501                 dirent->name_len = ent->dir->name_len;
 502                 retval = ext2fs_set_rec_len(fs, rec_len, dirent);
 503                 if (retval)
 504                         return retval;
 505                 prev_rec_len = rec_len;
 506                 memcpy(dirent->name, ent->dir->name, dirent->name_len & 0xFF);
 507                 offset += rec_len;
 508                 left -= rec_len;
 509                 if (left < slack) {
 510                         prev_rec_len += left;
 511                         retval = ext2fs_set_rec_len(fs, prev_rec_len, dirent);
 512                         if (retval)
 513                                 return retval;
 514                         offset += left;
 515                         left = 0;
 516                 }
 517                 prev_hash = ent->hash;
 518         }
 519         if (left)
 520                 retval = ext2fs_set_rec_len(fs, rec_len + left, dirent);
 521         if (csum_size) {
 522                 t = EXT2_DIRENT_TAIL(block_start, fs->blocksize);
 523                 ext2fs_initialize_dirent_tail(fs, t);
 524         }
 525
 526         return retval;
 527 }
 528
 529
 530 static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
 531                                     ext2_ino_t ino, ext2_ino_t parent)
 532 {
 533         struct ext2_dir_entry           *dir;
 534         struct ext2_dx_root_info        *root;
 535         struct ext2_dx_countlimit       *limits;
 536         int                             filetype = 0;
 537         int                             csum_size = 0;
 538
 539         if (fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_FILETYPE)
 540                 filetype = EXT2_FT_DIR << 8;
 541
 542         memset(buf, 0, fs->blocksize);
 543         dir = (struct ext2_dir_entry *) buf;
 544         dir->inode = ino;
 545         dir->name[0] = '.';
 546         dir->name_len = 1 | filetype;
 547         dir->rec_len = 12;
 548         dir = (struct ext2_dir_entry *) (buf + 12);
 549         dir->inode = parent;
 550         dir->name[0] = '.';
 551         dir->name[1] = '.';
 552         dir->name_len = 2 | filetype;
 553         dir->rec_len = fs->blocksize - 12;
 554
 555         root = (struct ext2_dx_root_info *) (buf+24);
 556         root->reserved_zero = 0;
 557         root->hash_version = fs->super->s_def_hash_version;
 558         root->info_length = 8;
 559         root->indirect_levels = 0;
 560         root->unused_flags = 0;
 561
 562         if (EXT2_HAS_RO_COMPAT_FEATURE(fs->super,
 563                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 564                 csum_size = sizeof(struct ext2_dx_tail);
 565
 566         limits = (struct ext2_dx_countlimit *) (buf+32);
 567         limits->limit = (fs->blocksize - (32 + csum_size)) /
 568                         sizeof(struct ext2_dx_entry);
 569         limits->count = 0;
 570
 571         return root;
 572 }
 573
 574
 575 static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
 576 {
 577         struct ext2_dir_entry           *dir;
 578         struct ext2_dx_countlimit       *limits;
 579         int                             csum_size = 0;
 580
 581         memset(buf, 0, fs->blocksize);
 582         dir = (struct ext2_dir_entry *) buf;
 583         dir->inode = 0;
 584         (void) ext2fs_set_rec_len(fs, fs->blocksize, dir);
 585
 586         if (EXT2_HAS_RO_COMPAT_FEATURE(fs->super,
 587                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 588                 csum_size = sizeof(struct ext2_dx_tail);
 589
 590         limits = (struct ext2_dx_countlimit *) (buf+8);
 591         limits->limit = (fs->blocksize - (8 + csum_size)) /
 592                         sizeof(struct ext2_dx_entry);
 593         limits->count = 0;
 594
 595         return (struct ext2_dx_entry *) limits;
 596 }
 597
 598 /*
 599  * This function takes the leaf nodes which have been written in
 600  * outdir, and populates the root node and any necessary interior nodes.
 601  */
 602 static errcode_t calculate_tree(ext2_filsys fs,
 603                                 struct out_dir *outdir,
 604                                 ext2_ino_t ino,
 605                                 ext2_ino_t parent)
 606 {
 607         struct ext2_dx_root_info        *root_info;
 608         struct ext2_dx_entry            *root, *dx_ent = 0;
 609         struct ext2_dx_countlimit       *root_limit, *limit;
 610         errcode_t                       retval;
 611         char                            * block_start;
 612         int                             i, c1, c2, nblks;
 613         int                             limit_offset, root_offset;
 614
 615         root_info = set_root_node(fs, outdir->buf, ino, parent);
 616         root_offset = limit_offset = ((char *) root_info - outdir->buf) +
 617                 root_info->info_length;
 618         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 619         c1 = root_limit->limit;
 620         nblks = outdir->num;
 621
 622         /* Write out the pointer blocks */
 623         if (nblks-1 <= c1) {
 624                 /* Just write out the root block, and we're done */
 625                 root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
 626                 for (i=1; i < nblks; i++) {
 627                         root->block = ext2fs_cpu_to_le32(i);
 628                         if (i != 1)
 629                                 root->hash =
 630                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 631                         root++;
 632                         c1--;
 633                 }
 634         } else {
 635                 c2 = 0;
 636                 limit = 0;
 637                 root_info->indirect_levels = 1;
 638                 for (i=1; i < nblks; i++) {
 639                         if (c1 == 0)
 640                                 return ENOSPC;
 641                         if (c2 == 0) {
 642                                 if (limit)
 643                                         limit->limit = limit->count =
 644                 ext2fs_cpu_to_le16(limit->limit);
 645                                 root = (struct ext2_dx_entry *)
 646                                         (outdir->buf + root_offset);
 647                                 root->block = ext2fs_cpu_to_le32(outdir->num);
 648                                 if (i != 1)
 649                                         root->hash =
 650                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 651                                 if ((retval =  get_next_block(fs, outdir,
 652                                                               &block_start)))
 653                                         return retval;
 654                                 dx_ent = set_int_node(fs, block_start);
 655                                 limit = (struct ext2_dx_countlimit *) dx_ent;
 656                                 c2 = limit->limit;
 657                                 root_offset += sizeof(struct ext2_dx_entry);
 658                                 c1--;
 659                         }
 660                         dx_ent->block = ext2fs_cpu_to_le32(i);
 661                         if (c2 != limit->limit)
 662                                 dx_ent->hash =
 663                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 664                         dx_ent++;
 665                         c2--;
 666                 }
 667                 limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
 668                 limit->limit = ext2fs_cpu_to_le16(limit->limit);
 669         }
 670         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 671         root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
 672         root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
 673
 674         return 0;
 675 }
 676
 677 struct write_dir_struct {
 678         struct out_dir *outdir;
 679         errcode_t       err;
 680         e2fsck_t        ctx;
 681         int             cleared;
 682         ext2_ino_t      dir;
 683 };
 684
 685 /*
 686  * Helper function which writes out a directory block.
 687  */
 688 static int write_dir_block(ext2_filsys fs,
 689                            blk64_t *block_nr,
 690                            e2_blkcnt_t blockcnt,
 691                            blk64_t ref_block EXT2FS_ATTR((unused)),
 692                            int ref_offset EXT2FS_ATTR((unused)),
 693                            void *priv_data)
 694 {
 695         struct write_dir_struct *wd = (struct write_dir_struct *) priv_data;
 696         blk64_t blk;
 697         char    *dir, *buf = 0;
 698
 699         if (*block_nr == 0)
 700                 return 0;
 701         if (blockcnt < 0)
 702                 return 0;
 703         if (blockcnt < wd->outdir->num)
 704                 dir = wd->outdir->buf + (blockcnt * fs->blocksize);
 705         else if (wd->ctx->lost_and_found == wd->dir) {
 706                 /* Don't release any extra directory blocks for lost+found */
 707                 wd->err = ext2fs_new_dir_block(fs, 0, 0, &buf);
 708                 if (wd->err)
 709                         return BLOCK_ABORT;
 710                 dir = buf;
 711                 wd->outdir->num++;
 712         } else {
 713                 /* We don't need this block, so release it */
 714                 e2fsck_read_bitmaps(wd->ctx);
 715                 blk = *block_nr;
 716                 ext2fs_unmark_block_bitmap2(wd->ctx->block_found_map, blk);
 717                 ext2fs_block_alloc_stats2(fs, blk, -1);
 718                 *block_nr = 0;
 719                 wd->cleared++;
 720                 return BLOCK_CHANGED;
 721         }
 722
 723         wd->err = ext2fs_write_dir_block4(fs, *block_nr, dir, 0, wd->dir);
 724         if (buf)
 725                 ext2fs_free_mem(&buf);
 726
 727         if (wd->err)
 728                 return BLOCK_ABORT;
 729         return 0;
 730 }
 731
 732 static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
 733                                  struct out_dir *outdir,
 734                                  ext2_ino_t ino, int compress)
 735 {
 736         struct write_dir_struct wd;
 737         errcode_t       retval;
 738         struct ext2_inode       inode;
 739
 740         retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
 741         if (retval)
 742                 return retval;
 743
 744         wd.outdir = outdir;
 745         wd.err = 0;
 746         wd.ctx = ctx;
 747         wd.cleared = 0;
 748         wd.dir = ino;
 749
 750         retval = ext2fs_block_iterate3(fs, ino, 0, 0,
 751                                        write_dir_block, &wd);
 752         if (retval)
 753                 return retval;
 754         if (wd.err)
 755                 return wd.err;
 756
 757         e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
 758         if (compress)
 759                 inode.i_flags &= ~EXT2_INDEX_FL;
 760         else
 761                 inode.i_flags |= EXT2_INDEX_FL;
 762         inode.i_size = outdir->num * fs->blocksize;
 763         ext2fs_iblk_sub_blocks(fs, &inode, wd.cleared);
 764         e2fsck_write_inode(ctx, ino, &inode, "rehash_dir");
 765
 766         return 0;
 767 }
 768
 769 errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino)
 770 {
 771         ext2_filsys             fs = ctx->fs;
 772         errcode_t               retval;
 773         struct ext2_inode       inode;
 774         char                    *dir_buf = 0;
 775         struct fill_dir_struct  fd;
 776         struct out_dir          outdir;
 777
 778         outdir.max = outdir.num = 0;
 779         outdir.buf = 0;
 780         outdir.hashes = 0;
 781         e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
 782
 783         retval = ENOMEM;
 784         fd.harray = 0;
 785         dir_buf = malloc(inode.i_size);
 786         if (!dir_buf)
 787                 goto errout;
 788
 789         fd.max_array = inode.i_size / 32;
 790         fd.num_array = 0;
 791         fd.harray = malloc(fd.max_array * sizeof(struct hash_entry));
 792         if (!fd.harray)
 793                 goto errout;
 794
 795         fd.ctx = ctx;
 796         fd.buf = dir_buf;
 797         fd.inode = &inode;
 798         fd.err = 0;
 799         fd.dir_size = 0;
 800         fd.compress = 0;
 801         fd.dir = ino;
 802         if (!(fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_DIR_INDEX) ||
 803             (inode.i_size / fs->blocksize) < 2)
 804                 fd.compress = 1;
 805         fd.parent = 0;
 806
 807 retry_nohash:
 808         /* Read in the entire directory into memory */
 809         retval = ext2fs_block_iterate3(fs, ino, 0, 0,
 810                                        fill_dir_block, &fd);
 811         if (fd.err) {
 812                 retval = fd.err;
 813                 goto errout;
 814         }
 815
 816         /*
 817          * If the entries read are less than a block, then don't index
 818          * the directory
 819          */
 820         if (!fd.compress && (fd.dir_size < (fs->blocksize - 24))) {
 821                 fd.compress = 1;
 822                 fd.dir_size = 0;
 823                 fd.num_array = 0;
 824                 goto retry_nohash;
 825         }
 826
 827 #if 0
 828         printf("%d entries (%d bytes) found in inode %d\n",
 829                fd.num_array, fd.dir_size, ino);
 830 #endif
 831
 832         /* Sort the list */
 833 resort:
 834         if (fd.compress)
 835                 qsort(fd.harray+2, fd.num_array-2, sizeof(struct hash_entry),
 836                       hash_cmp);
 837         else
 838                 qsort(fd.harray, fd.num_array, sizeof(struct hash_entry),
 839                       hash_cmp);
 840
 841         /*
 842          * Look for duplicates
 843          */
 844         if (duplicate_search_and_fix(ctx, fs, ino, &fd))
 845                 goto resort;
 846
 847         if (ctx->options & E2F_OPT_NO) {
 848                 retval = 0;
 849                 goto errout;
 850         }
 851
 852         /* Sort non-hashed directories by inode number */
 853         if (fd.compress)
 854                 qsort(fd.harray+2, fd.num_array-2,
 855                       sizeof(struct hash_entry), ino_cmp);
 856
 857         /*
 858          * Copy the directory entries.  In a htree directory these
 859          * will become the leaf nodes.
 860          */
 861         retval = copy_dir_entries(ctx, &fd, &outdir, ino);
 862         if (retval)
 863                 goto errout;
 864
 865         free(dir_buf); dir_buf = 0;
 866
 867         if (!fd.compress) {
 868                 /* Calculate the interior nodes */
 869                 retval = calculate_tree(fs, &outdir, ino, fd.parent);
 870                 if (retval)
 871                         goto errout;
 872         }
 873
 874         retval = write_directory(ctx, fs, &outdir, ino, fd.compress);
 875         if (retval)
 876                 goto errout;
 877
 878 errout:
 879         free(dir_buf);
 880         free(fd.harray);
 881
 882         free_out_dir(&outdir);
 883         return retval;
 884 }
 885
 886 void e2fsck_rehash_directories(e2fsck_t ctx)
 887 {
 888         struct problem_context  pctx;
 889 #ifdef RESOURCE_TRACK
 890         struct resource_track   rtrack;
 891 #endif
 892         struct dir_info         *dir;
 893         ext2_u32_iterate        iter;
 894         struct dir_info_iter *  dirinfo_iter = 0;
 895         ext2_ino_t              ino;
 896         errcode_t               retval;
 897         int                     cur, max, all_dirs, first = 1;
 898
 899         init_resource_track(&rtrack, ctx->fs->io);
 900         all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
 901
 902         if (!ctx->dirs_to_hash && !all_dirs)
 903                 return;
 904
 905         (void) e2fsck_get_lost_and_found(ctx, 0);
 906
 907         clear_problem_context(&pctx);
 908
 909         cur = 0;
 910         if (all_dirs) {
 911                 dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
 912                 max = e2fsck_get_num_dirinfo(ctx);
 913         } else {
 914                 retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
 915                                                        &iter);
 916                 if (retval) {
 917                         pctx.errcode = retval;
 918                         fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
 919                         return;
 920                 }
 921                 max = ext2fs_u32_list_count(ctx->dirs_to_hash);
 922         }
 923         while (1) {
 924                 if (all_dirs) {
 925                         if ((dir = e2fsck_dir_info_iter(ctx,
 926                                                         dirinfo_iter)) == 0)
 927                                 break;
 928                         ino = dir->ino;
 929                 } else {
 930                         if (!ext2fs_u32_list_iterate(iter, &ino))
 931                                 break;
 932                 }
 933
 934                 pctx.dir = ino;
 935                 if (first) {
 936                         fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
 937                         first = 0;
 938                 }
 939 #if 0
 940                 fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
 941 #endif
 942                 pctx.errcode = e2fsck_rehash_dir(ctx, ino);
 943                 if (pctx.errcode) {
 944                         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
 945                         fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
 946                 }
 947                 if (ctx->progress && !ctx->progress_fd)
 948                         e2fsck_simple_progress(ctx, "Rebuilding directory",
 949                                100.0 * (float) (++cur) / (float) max, ino);
 950         }
 951         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
 952         if (all_dirs)
 953                 e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
 954         else
 955                 ext2fs_u32_list_iterate_end(iter);
 956
 957         if (ctx->dirs_to_hash)
 958                 ext2fs_u32_list_free(ctx->dirs_to_hash);
 959         ctx->dirs_to_hash = 0;
 960
 961         print_resource_track(ctx, "Pass 3A", &rtrack, ctx->fs->io);
 962 }