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         unsigned 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, name_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                 name_len = ext2fs_dirent_name_len(dirent);
 146                 if (((dir_offset + rec_len) > fs->blocksize) ||
 147                     (rec_len < 8) ||
 148                     ((rec_len % 4) != 0) ||
 149                     (name_len + 8 > rec_len)) {
 150                         fd->err = EXT2_ET_DIR_CORRUPTED;
 151                         return BLOCK_ABORT;
 152                 }
 153                 dir_offset += rec_len;
 154                 if (dirent->inode == 0)
 155                         continue;
 156                 if (!fd->compress && (name_len == 1) &&
 157                     (dirent->name[0] == '.'))
 158                         continue;
 159                 if (!fd->compress && (name_len == 2) &&
 160                     (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
 161                         fd->parent = dirent->inode;
 162                         continue;
 163                 }
 164                 if (fd->num_array >= fd->max_array) {
 165                         new_array = realloc(fd->harray,
 166                             sizeof(struct hash_entry) * (fd->max_array+500));
 167                         if (!new_array) {
 168                                 fd->err = ENOMEM;
 169                                 return BLOCK_ABORT;
 170                         }
 171                         fd->harray = new_array;
 172                         fd->max_array += 500;
 173                 }
 174                 ent = fd->harray + fd->num_array++;
 175                 ent->dir = dirent;
 176                 fd->dir_size += EXT2_DIR_REC_LEN(name_len);
 177                 ent->ino = dirent->inode;
 178                 if (fd->compress)
 179                         ent->hash = ent->minor_hash = 0;
 180                 else {
 181                         fd->err = ext2fs_dirhash(hash_alg, dirent->name,
 182                                                  name_len,
 183                                                  fs->super->s_hash_seed,
 184                                                  &ent->hash, &ent->minor_hash);
 185                         if (fd->err)
 186                                 return BLOCK_ABORT;
 187                 }
 188         }
 189
 190         return 0;
 191 }
 192
 193 /* Used for sorting the hash entry */
 194 static EXT2_QSORT_TYPE ino_cmp(const void *a, const void *b)
 195 {
 196         const struct hash_entry *he_a = (const struct hash_entry *) a;
 197         const struct hash_entry *he_b = (const struct hash_entry *) b;
 198
 199         return (he_a->ino - he_b->ino);
 200 }
 201
 202 /* Used for sorting the hash entry */
 203 static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
 204 {
 205         const struct hash_entry *he_a = (const struct hash_entry *) a;
 206         const struct hash_entry *he_b = (const struct hash_entry *) b;
 207         unsigned int he_a_len, he_b_len;
 208         int     ret;
 209         int     min_len;
 210
 211         he_a_len = ext2fs_dirent_name_len(he_a->dir);
 212         he_b_len = ext2fs_dirent_name_len(he_b->dir);
 213         min_len = he_a_len;
 214         if (min_len > he_b_len)
 215                 min_len = he_b_len;
 216
 217         ret = strncmp(he_a->dir->name, he_b->dir->name, min_len);
 218         if (ret == 0) {
 219                 if (he_a_len > he_b_len)
 220                         ret = 1;
 221                 else if (he_a_len < he_b_len)
 222                         ret = -1;
 223                 else
 224                         ret = he_b->dir->inode - he_a->dir->inode;
 225         }
 226         return ret;
 227 }
 228
 229 /* Used for sorting the hash entry */
 230 static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
 231 {
 232         const struct hash_entry *he_a = (const struct hash_entry *) a;
 233         const struct hash_entry *he_b = (const struct hash_entry *) b;
 234         int     ret;
 235
 236         if (he_a->hash > he_b->hash)
 237                 ret = 1;
 238         else if (he_a->hash < he_b->hash)
 239                 ret = -1;
 240         else {
 241                 if (he_a->minor_hash > he_b->minor_hash)
 242                         ret = 1;
 243                 else if (he_a->minor_hash < he_b->minor_hash)
 244                         ret = -1;
 245                 else
 246                         ret = name_cmp(a, b);
 247         }
 248         return ret;
 249 }
 250
 251 static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
 252                                 int blocks)
 253 {
 254         void                    *new_mem;
 255
 256         if (outdir->max) {
 257                 new_mem = realloc(outdir->buf, blocks * fs->blocksize);
 258                 if (!new_mem)
 259                         return ENOMEM;
 260                 outdir->buf = new_mem;
 261                 new_mem = realloc(outdir->hashes,
 262                                   blocks * sizeof(ext2_dirhash_t));
 263                 if (!new_mem)
 264                         return ENOMEM;
 265                 outdir->hashes = new_mem;
 266         } else {
 267                 outdir->buf = malloc(blocks * fs->blocksize);
 268                 outdir->hashes = malloc(blocks * sizeof(ext2_dirhash_t));
 269                 outdir->num = 0;
 270         }
 271         outdir->max = blocks;
 272         return 0;
 273 }
 274
 275 static void free_out_dir(struct out_dir *outdir)
 276 {
 277         free(outdir->buf);
 278         free(outdir->hashes);
 279         outdir->max = 0;
 280         outdir->num =0;
 281 }
 282
 283 static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
 284                          char ** ret)
 285 {
 286         errcode_t       retval;
 287
 288         if (outdir->num >= outdir->max) {
 289                 retval = alloc_size_dir(fs, outdir, outdir->max + 50);
 290                 if (retval)
 291                         return retval;
 292         }
 293         *ret = outdir->buf + (outdir->num++ * fs->blocksize);
 294         memset(*ret, 0, fs->blocksize);
 295         return 0;
 296 }
 297
 298 /*
 299  * This function is used to make a unique filename.  We do this by
 300  * appending ~0, and then incrementing the number.  However, we cannot
 301  * expand the length of the filename beyond the padding available in
 302  * the directory entry.
 303  */
 304 static void mutate_name(char *str, unsigned int *len)
 305 {
 306         int     i;
 307         unsigned int l = *len;
 308
 309         /*
 310          * First check to see if it looks the name has been mutated
 311          * already
 312          */
 313         for (i = l-1; i > 0; i--) {
 314                 if (!isdigit(str[i]))
 315                         break;
 316         }
 317         if ((i == l-1) || (str[i] != '~')) {
 318                 if (((l-1) & 3) < 2)
 319                         l += 2;
 320                 else
 321                         l = (l+3) & ~3;
 322                 str[l-2] = '~';
 323                 str[l-1] = '0';
 324                 *len = l;
 325                 return;
 326         }
 327         for (i = l-1; i >= 0; i--) {
 328                 if (isdigit(str[i])) {
 329                         if (str[i] == '9')
 330                                 str[i] = '0';
 331                         else {
 332                                 str[i]++;
 333                                 return;
 334                         }
 335                         continue;
 336                 }
 337                 if (i == 1) {
 338                         if (str[0] == 'z')
 339                                 str[0] = 'A';
 340                         else if (str[0] == 'Z') {
 341                                 str[0] = '~';
 342                                 str[1] = '0';
 343                         } else
 344                                 str[0]++;
 345                 } else if (i > 0) {
 346                         str[i] = '1';
 347                         str[i-1] = '~';
 348                 } else {
 349                         if (str[0] == '~')
 350                                 str[0] = 'a';
 351                         else
 352                                 str[0]++;
 353                 }
 354                 break;
 355         }
 356 }
 357
 358 static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
 359                                     ext2_ino_t ino,
 360                                     struct fill_dir_struct *fd)
 361 {
 362         struct problem_context  pctx;
 363         struct hash_entry       *ent, *prev;
 364         int                     i, j;
 365         int                     fixed = 0;
 366         char                    new_name[256];
 367         unsigned int            new_len;
 368         int                     hash_alg;
 369
 370         clear_problem_context(&pctx);
 371         pctx.ino = ino;
 372
 373         hash_alg = fs->super->s_def_hash_version;
 374         if ((hash_alg <= EXT2_HASH_TEA) &&
 375             (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
 376                 hash_alg += 3;
 377
 378         for (i=1; i < fd->num_array; i++) {
 379                 ent = fd->harray + i;
 380                 prev = ent - 1;
 381                 if (!ent->dir->inode ||
 382                     (ext2fs_dirent_name_len(ent->dir) !=
 383                      ext2fs_dirent_name_len(prev->dir)) ||
 384                     strncmp(ent->dir->name, prev->dir->name,
 385                              ext2fs_dirent_name_len(ent->dir)))
 386                         continue;
 387                 pctx.dirent = ent->dir;
 388                 if ((ent->dir->inode == prev->dir->inode) &&
 389                     fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
 390                         e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
 391                         ent->dir->inode = 0;
 392                         fixed++;
 393                         continue;
 394                 }
 395                 new_len = ext2fs_dirent_name_len(ent->dir);
 396                 memcpy(new_name, ent->dir->name, new_len);
 397                 mutate_name(new_name, &new_len);
 398                 for (j=0; j < fd->num_array; j++) {
 399                         if ((i==j) ||
 400                             (new_len !=
 401                              ext2fs_dirent_name_len(fd->harray[j].dir)) ||
 402                             strncmp(new_name, fd->harray[j].dir->name, new_len))
 403                                 continue;
 404                         mutate_name(new_name, &new_len);
 405
 406                         j = -1;
 407                 }
 408                 new_name[new_len] = 0;
 409                 pctx.str = new_name;
 410                 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
 411                         memcpy(ent->dir->name, new_name, new_len);
 412                         ext2fs_dirent_set_name_len(ent->dir, new_len);
 413                         ext2fs_dirhash(hash_alg, ent->dir->name, new_len,
 414                                        fs->super->s_hash_seed,
 415                                        &ent->hash, &ent->minor_hash);
 416                         fixed++;
 417                 }
 418         }
 419         return fixed;
 420 }
 421
 422
 423 static errcode_t copy_dir_entries(e2fsck_t ctx,
 424                                   struct fill_dir_struct *fd,
 425                                   struct out_dir *outdir)
 426 {
 427         ext2_filsys             fs = ctx->fs;
 428         errcode_t               retval;
 429         char                    *block_start;
 430         struct hash_entry       *ent;
 431         struct ext2_dir_entry   *dirent;
 432         unsigned int            rec_len, prev_rec_len, left, slack, offset;
 433         int                     i;
 434         ext2_dirhash_t          prev_hash;
 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(ext2fs_dirent_name_len(ent->dir));
 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                 ext2fs_dirent_set_name_len(dirent,
 502                                            ext2fs_dirent_name_len(ent->dir));
 503                 ext2fs_dirent_set_file_type(dirent,
 504                                             ext2fs_dirent_file_type(ent->dir));
 505                 retval = ext2fs_set_rec_len(fs, rec_len, dirent);
 506                 if (retval)
 507                         return retval;
 508                 prev_rec_len = rec_len;
 509                 memcpy(dirent->name, ent->dir->name,
 510                        ext2fs_dirent_name_len(dirent));
 511                 offset += rec_len;
 512                 left -= rec_len;
 513                 if (left < slack) {
 514                         prev_rec_len += left;
 515                         retval = ext2fs_set_rec_len(fs, prev_rec_len, dirent);
 516                         if (retval)
 517                                 return retval;
 518                         offset += left;
 519                         left = 0;
 520                 }
 521                 prev_hash = ent->hash;
 522         }
 523         if (left)
 524                 retval = ext2fs_set_rec_len(fs, rec_len + left, dirent);
 525         if (csum_size) {
 526                 t = EXT2_DIRENT_TAIL(block_start, fs->blocksize);
 527                 ext2fs_initialize_dirent_tail(fs, t);
 528         }
 529
 530         return retval;
 531 }
 532
 533
 534 static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
 535                                     ext2_ino_t ino, ext2_ino_t parent)
 536 {
 537         struct ext2_dir_entry           *dir;
 538         struct ext2_dx_root_info        *root;
 539         struct ext2_dx_countlimit       *limits;
 540         int                             filetype = 0;
 541         int                             csum_size = 0;
 542
 543         if (fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_FILETYPE)
 544                 filetype = EXT2_FT_DIR;
 545
 546         memset(buf, 0, fs->blocksize);
 547         dir = (struct ext2_dir_entry *) buf;
 548         dir->inode = ino;
 549         dir->name[0] = '.';
 550         ext2fs_dirent_set_name_len(dir, 1);
 551         ext2fs_dirent_set_file_type(dir, filetype);
 552         dir->rec_len = 12;
 553         dir = (struct ext2_dir_entry *) (buf + 12);
 554         dir->inode = parent;
 555         dir->name[0] = '.';
 556         dir->name[1] = '.';
 557         ext2fs_dirent_set_name_len(dir, 2);
 558         ext2fs_dirent_set_file_type(dir, filetype);
 559         dir->rec_len = fs->blocksize - 12;
 560
 561         root = (struct ext2_dx_root_info *) (buf+24);
 562         root->reserved_zero = 0;
 563         root->hash_version = fs->super->s_def_hash_version;
 564         root->info_length = 8;
 565         root->indirect_levels = 0;
 566         root->unused_flags = 0;
 567
 568         if (EXT2_HAS_RO_COMPAT_FEATURE(fs->super,
 569                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 570                 csum_size = sizeof(struct ext2_dx_tail);
 571
 572         limits = (struct ext2_dx_countlimit *) (buf+32);
 573         limits->limit = (fs->blocksize - (32 + csum_size)) /
 574                         sizeof(struct ext2_dx_entry);
 575         limits->count = 0;
 576
 577         return root;
 578 }
 579
 580
 581 static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
 582 {
 583         struct ext2_dir_entry           *dir;
 584         struct ext2_dx_countlimit       *limits;
 585         int                             csum_size = 0;
 586
 587         memset(buf, 0, fs->blocksize);
 588         dir = (struct ext2_dir_entry *) buf;
 589         dir->inode = 0;
 590         (void) ext2fs_set_rec_len(fs, fs->blocksize, dir);
 591
 592         if (EXT2_HAS_RO_COMPAT_FEATURE(fs->super,
 593                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 594                 csum_size = sizeof(struct ext2_dx_tail);
 595
 596         limits = (struct ext2_dx_countlimit *) (buf+8);
 597         limits->limit = (fs->blocksize - (8 + csum_size)) /
 598                         sizeof(struct ext2_dx_entry);
 599         limits->count = 0;
 600
 601         return (struct ext2_dx_entry *) limits;
 602 }
 603
 604 /*
 605  * This function takes the leaf nodes which have been written in
 606  * outdir, and populates the root node and any necessary interior nodes.
 607  */
 608 static errcode_t calculate_tree(ext2_filsys fs,
 609                                 struct out_dir *outdir,
 610                                 ext2_ino_t ino,
 611                                 ext2_ino_t parent)
 612 {
 613         struct ext2_dx_root_info        *root_info;
 614         struct ext2_dx_entry            *root, *dx_ent = 0;
 615         struct ext2_dx_countlimit       *root_limit, *limit;
 616         errcode_t                       retval;
 617         char                            * block_start;
 618         int                             i, c1, c2, nblks;
 619         int                             limit_offset, root_offset;
 620
 621         root_info = set_root_node(fs, outdir->buf, ino, parent);
 622         root_offset = limit_offset = ((char *) root_info - outdir->buf) +
 623                 root_info->info_length;
 624         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 625         c1 = root_limit->limit;
 626         nblks = outdir->num;
 627
 628         /* Write out the pointer blocks */
 629         if (nblks-1 <= c1) {
 630                 /* Just write out the root block, and we're done */
 631                 root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
 632                 for (i=1; i < nblks; i++) {
 633                         root->block = ext2fs_cpu_to_le32(i);
 634                         if (i != 1)
 635                                 root->hash =
 636                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 637                         root++;
 638                         c1--;
 639                 }
 640         } else {
 641                 c2 = 0;
 642                 limit = 0;
 643                 root_info->indirect_levels = 1;
 644                 for (i=1; i < nblks; i++) {
 645                         if (c1 == 0)
 646                                 return ENOSPC;
 647                         if (c2 == 0) {
 648                                 if (limit)
 649                                         limit->limit = limit->count =
 650                 ext2fs_cpu_to_le16(limit->limit);
 651                                 root = (struct ext2_dx_entry *)
 652                                         (outdir->buf + root_offset);
 653                                 root->block = ext2fs_cpu_to_le32(outdir->num);
 654                                 if (i != 1)
 655                                         root->hash =
 656                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 657                                 if ((retval =  get_next_block(fs, outdir,
 658                                                               &block_start)))
 659                                         return retval;
 660                                 dx_ent = set_int_node(fs, block_start);
 661                                 limit = (struct ext2_dx_countlimit *) dx_ent;
 662                                 c2 = limit->limit;
 663                                 root_offset += sizeof(struct ext2_dx_entry);
 664                                 c1--;
 665                         }
 666                         dx_ent->block = ext2fs_cpu_to_le32(i);
 667                         if (c2 != limit->limit)
 668                                 dx_ent->hash =
 669                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 670                         dx_ent++;
 671                         c2--;
 672                 }
 673                 limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
 674                 limit->limit = ext2fs_cpu_to_le16(limit->limit);
 675         }
 676         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 677         root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
 678         root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
 679
 680         return 0;
 681 }
 682
 683 struct write_dir_struct {
 684         struct out_dir *outdir;
 685         errcode_t       err;
 686         e2fsck_t        ctx;
 687         int             cleared;
 688         ext2_ino_t      dir;
 689 };
 690
 691 /*
 692  * Helper function which writes out a directory block.
 693  */
 694 static int write_dir_block(ext2_filsys fs,
 695                            blk64_t *block_nr,
 696                            e2_blkcnt_t blockcnt,
 697                            blk64_t ref_block EXT2FS_ATTR((unused)),
 698                            int ref_offset EXT2FS_ATTR((unused)),
 699                            void *priv_data)
 700 {
 701         struct write_dir_struct *wd = (struct write_dir_struct *) priv_data;
 702         blk64_t blk;
 703         char    *dir, *buf = 0;
 704
 705         if (*block_nr == 0)
 706                 return 0;
 707         if (blockcnt < 0)
 708                 return 0;
 709         if (blockcnt < wd->outdir->num)
 710                 dir = wd->outdir->buf + (blockcnt * fs->blocksize);
 711         else if (wd->ctx->lost_and_found == wd->dir) {
 712                 /* Don't release any extra directory blocks for lost+found */
 713                 wd->err = ext2fs_new_dir_block(fs, 0, 0, &buf);
 714                 if (wd->err)
 715                         return BLOCK_ABORT;
 716                 dir = buf;
 717                 wd->outdir->num++;
 718         } else {
 719                 /* We don't need this block, so release it */
 720                 e2fsck_read_bitmaps(wd->ctx);
 721                 blk = *block_nr;
 722                 ext2fs_unmark_block_bitmap2(wd->ctx->block_found_map, blk);
 723                 ext2fs_block_alloc_stats2(fs, blk, -1);
 724                 *block_nr = 0;
 725                 wd->cleared++;
 726                 return BLOCK_CHANGED;
 727         }
 728
 729         wd->err = ext2fs_write_dir_block4(fs, *block_nr, dir, 0, wd->dir);
 730         if (buf)
 731                 ext2fs_free_mem(&buf);
 732
 733         if (wd->err)
 734                 return BLOCK_ABORT;
 735         return 0;
 736 }
 737
 738 static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
 739                                  struct out_dir *outdir,
 740                                  ext2_ino_t ino, int compress)
 741 {
 742         struct write_dir_struct wd;
 743         errcode_t       retval;
 744         struct ext2_inode       inode;
 745
 746         retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
 747         if (retval)
 748                 return retval;
 749
 750         wd.outdir = outdir;
 751         wd.err = 0;
 752         wd.ctx = ctx;
 753         wd.cleared = 0;
 754         wd.dir = ino;
 755
 756         retval = ext2fs_block_iterate3(fs, ino, 0, 0,
 757                                        write_dir_block, &wd);
 758         if (retval)
 759                 return retval;
 760         if (wd.err)
 761                 return wd.err;
 762
 763         e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
 764         if (compress)
 765                 inode.i_flags &= ~EXT2_INDEX_FL;
 766         else
 767                 inode.i_flags |= EXT2_INDEX_FL;
 768         inode.i_size = outdir->num * fs->blocksize;
 769         ext2fs_iblk_sub_blocks(fs, &inode, wd.cleared);
 770         e2fsck_write_inode(ctx, ino, &inode, "rehash_dir");
 771
 772         return 0;
 773 }
 774
 775 errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino)
 776 {
 777         ext2_filsys             fs = ctx->fs;
 778         errcode_t               retval;
 779         struct ext2_inode       inode;
 780         char                    *dir_buf = 0;
 781         struct fill_dir_struct  fd;
 782         struct out_dir          outdir;
 783
 784         outdir.max = outdir.num = 0;
 785         outdir.buf = 0;
 786         outdir.hashes = 0;
 787         e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
 788
 789         retval = ENOMEM;
 790         fd.harray = 0;
 791         dir_buf = malloc(inode.i_size);
 792         if (!dir_buf)
 793                 goto errout;
 794
 795         fd.max_array = inode.i_size / 32;
 796         fd.num_array = 0;
 797         fd.harray = malloc(fd.max_array * sizeof(struct hash_entry));
 798         if (!fd.harray)
 799                 goto errout;
 800
 801         fd.ctx = ctx;
 802         fd.buf = dir_buf;
 803         fd.inode = &inode;
 804         fd.err = 0;
 805         fd.dir_size = 0;
 806         fd.compress = 0;
 807         fd.dir = ino;
 808         if (!(fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_DIR_INDEX) ||
 809             (inode.i_size / fs->blocksize) < 2)
 810                 fd.compress = 1;
 811         fd.parent = 0;
 812
 813 retry_nohash:
 814         /* Read in the entire directory into memory */
 815         retval = ext2fs_block_iterate3(fs, ino, 0, 0,
 816                                        fill_dir_block, &fd);
 817         if (fd.err) {
 818                 retval = fd.err;
 819                 goto errout;
 820         }
 821
 822         /*
 823          * If the entries read are less than a block, then don't index
 824          * the directory
 825          */
 826         if (!fd.compress && (fd.dir_size < (fs->blocksize - 24))) {
 827                 fd.compress = 1;
 828                 fd.dir_size = 0;
 829                 fd.num_array = 0;
 830                 goto retry_nohash;
 831         }
 832
 833 #if 0
 834         printf("%d entries (%d bytes) found in inode %d\n",
 835                fd.num_array, fd.dir_size, ino);
 836 #endif
 837
 838         /* Sort the list */
 839 resort:
 840         if (fd.compress)
 841                 qsort(fd.harray+2, fd.num_array-2, sizeof(struct hash_entry),
 842                       hash_cmp);
 843         else
 844                 qsort(fd.harray, fd.num_array, sizeof(struct hash_entry),
 845                       hash_cmp);
 846
 847         /*
 848          * Look for duplicates
 849          */
 850         if (duplicate_search_and_fix(ctx, fs, ino, &fd))
 851                 goto resort;
 852
 853         if (ctx->options & E2F_OPT_NO) {
 854                 retval = 0;
 855                 goto errout;
 856         }
 857
 858         /* Sort non-hashed directories by inode number */
 859         if (fd.compress)
 860                 qsort(fd.harray+2, fd.num_array-2,
 861                       sizeof(struct hash_entry), ino_cmp);
 862
 863         /*
 864          * Copy the directory entries.  In a htree directory these
 865          * will become the leaf nodes.
 866          */
 867         retval = copy_dir_entries(ctx, &fd, &outdir);
 868         if (retval)
 869                 goto errout;
 870
 871         free(dir_buf); dir_buf = 0;
 872
 873         if (!fd.compress) {
 874                 /* Calculate the interior nodes */
 875                 retval = calculate_tree(fs, &outdir, ino, fd.parent);
 876                 if (retval)
 877                         goto errout;
 878         }
 879
 880         retval = write_directory(ctx, fs, &outdir, ino, fd.compress);
 881         if (retval)
 882                 goto errout;
 883
 884 errout:
 885         free(dir_buf);
 886         free(fd.harray);
 887
 888         free_out_dir(&outdir);
 889         return retval;
 890 }
 891
 892 void e2fsck_rehash_directories(e2fsck_t ctx)
 893 {
 894         struct problem_context  pctx;
 895 #ifdef RESOURCE_TRACK
 896         struct resource_track   rtrack;
 897 #endif
 898         struct dir_info         *dir;
 899         ext2_u32_iterate        iter;
 900         struct dir_info_iter *  dirinfo_iter = 0;
 901         ext2_ino_t              ino;
 902         errcode_t               retval;
 903         int                     cur, max, all_dirs, first = 1;
 904
 905         init_resource_track(&rtrack, ctx->fs->io);
 906         all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
 907
 908         if (!ctx->dirs_to_hash && !all_dirs)
 909                 return;
 910
 911         (void) e2fsck_get_lost_and_found(ctx, 0);
 912
 913         clear_problem_context(&pctx);
 914
 915         cur = 0;
 916         if (all_dirs) {
 917                 dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
 918                 max = e2fsck_get_num_dirinfo(ctx);
 919         } else {
 920                 retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
 921                                                        &iter);
 922                 if (retval) {
 923                         pctx.errcode = retval;
 924                         fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
 925                         return;
 926                 }
 927                 max = ext2fs_u32_list_count(ctx->dirs_to_hash);
 928         }
 929         while (1) {
 930                 if (all_dirs) {
 931                         if ((dir = e2fsck_dir_info_iter(ctx,
 932                                                         dirinfo_iter)) == 0)
 933                                 break;
 934                         ino = dir->ino;
 935                 } else {
 936                         if (!ext2fs_u32_list_iterate(iter, &ino))
 937                                 break;
 938                 }
 939
 940                 pctx.dir = ino;
 941                 if (first) {
 942                         fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
 943                         first = 0;
 944                 }
 945 #if 0
 946                 fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
 947 #endif
 948                 pctx.errcode = e2fsck_rehash_dir(ctx, ino);
 949                 if (pctx.errcode) {
 950                         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
 951                         fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
 952                 }
 953                 if (ctx->progress && !ctx->progress_fd)
 954                         e2fsck_simple_progress(ctx, "Rebuilding directory",
 955                                100.0 * (float) (++cur) / (float) max, ino);
 956         }
 957         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
 958         if (all_dirs)
 959                 e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
 960         else
 961                 ext2fs_u32_list_iterate_end(iter);
 962
 963         if (ctx->dirs_to_hash)
 964                 ext2fs_u32_list_free(ctx->dirs_to_hash);
 965         ctx->dirs_to_hash = 0;
 966
 967         print_resource_track(ctx, "Pass 3A", &rtrack, ctx->fs->io);
 968 }