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 };
  89
  90 struct hash_entry {
  91         ext2_dirhash_t  hash;
  92         ext2_dirhash_t  minor_hash;
  93         ino_t           ino;
  94         struct ext2_dir_entry   *dir;
  95 };
  96
  97 struct out_dir {
  98         int             num;
  99         int             max;
 100         char            *buf;
 101         ext2_dirhash_t  *hashes;
 102 };
 103
 104 static int fill_dir_block(ext2_filsys fs,
 105                           blk64_t *block_nr,
 106                           e2_blkcnt_t blockcnt,
 107                           blk64_t ref_block EXT2FS_ATTR((unused)),
 108                           int ref_offset EXT2FS_ATTR((unused)),
 109                           void *priv_data)
 110 {
 111         struct fill_dir_struct  *fd = (struct fill_dir_struct *) priv_data;
 112         struct hash_entry       *new_array, *ent;
 113         struct ext2_dir_entry   *dirent;
 114         char                    *dir;
 115         unsigned int            offset, dir_offset, rec_len, name_len;
 116         int                     hash_alg;
 117
 118         if (blockcnt < 0)
 119                 return 0;
 120
 121         offset = blockcnt * fs->blocksize;
 122         if (offset + fs->blocksize > fd->inode->i_size) {
 123                 fd->err = EXT2_ET_DIR_CORRUPTED;
 124                 return BLOCK_ABORT;
 125         }
 126
 127         dir = (fd->buf+offset);
 128         if (*block_nr == 0) {
 129                 memset(dir, 0, fs->blocksize);
 130                 dirent = (struct ext2_dir_entry *) dir;
 131                 (void) ext2fs_set_rec_len(fs, fs->blocksize, dirent);
 132         } else {
 133                 int flags = fs->flags;
 134                 fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
 135                 fd->err = ext2fs_read_dir_block4(fs, *block_nr, dir, 0,
 136                                                  fd->dir);
 137                 fs->flags = (flags & EXT2_FLAG_IGNORE_CSUM_ERRORS) |
 138                             (fs->flags & ~EXT2_FLAG_IGNORE_CSUM_ERRORS);
 139                 if (fd->err)
 140                         return BLOCK_ABORT;
 141         }
 142         hash_alg = fs->super->s_def_hash_version;
 143         if ((hash_alg <= EXT2_HASH_TEA) &&
 144             (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
 145                 hash_alg += 3;
 146         /* While the directory block is "hot", index it. */
 147         dir_offset = 0;
 148         while (dir_offset < fs->blocksize) {
 149                 dirent = (struct ext2_dir_entry *) (dir + dir_offset);
 150                 (void) ext2fs_get_rec_len(fs, dirent, &rec_len);
 151                 name_len = ext2fs_dirent_name_len(dirent);
 152                 if (((dir_offset + rec_len) > fs->blocksize) ||
 153                     (rec_len < 8) ||
 154                     ((rec_len % 4) != 0) ||
 155                     (name_len + 8 > rec_len)) {
 156                         fd->err = EXT2_ET_DIR_CORRUPTED;
 157                         return BLOCK_ABORT;
 158                 }
 159                 dir_offset += rec_len;
 160                 if (dirent->inode == 0)
 161                         continue;
 162                 if (!fd->compress && (name_len == 1) &&
 163                     (dirent->name[0] == '.'))
 164                         continue;
 165                 if (!fd->compress && (name_len == 2) &&
 166                     (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
 167                         fd->parent = dirent->inode;
 168                         continue;
 169                 }
 170                 if (fd->num_array >= fd->max_array) {
 171                         new_array = realloc(fd->harray,
 172                             sizeof(struct hash_entry) * (fd->max_array+500));
 173                         if (!new_array) {
 174                                 fd->err = ENOMEM;
 175                                 return BLOCK_ABORT;
 176                         }
 177                         fd->harray = new_array;
 178                         fd->max_array += 500;
 179                 }
 180                 ent = fd->harray + fd->num_array++;
 181                 ent->dir = dirent;
 182                 fd->dir_size += EXT2_DIR_REC_LEN(name_len);
 183                 ent->ino = dirent->inode;
 184                 if (fd->compress)
 185                         ent->hash = ent->minor_hash = 0;
 186                 else {
 187                         fd->err = ext2fs_dirhash(hash_alg, dirent->name,
 188                                                  name_len,
 189                                                  fs->super->s_hash_seed,
 190                                                  &ent->hash, &ent->minor_hash);
 191                         if (fd->err)
 192                                 return BLOCK_ABORT;
 193                 }
 194         }
 195
 196         return 0;
 197 }
 198
 199 /* Used for sorting the hash entry */
 200 static EXT2_QSORT_TYPE ino_cmp(const void *a, const void *b)
 201 {
 202         const struct hash_entry *he_a = (const struct hash_entry *) a;
 203         const struct hash_entry *he_b = (const struct hash_entry *) b;
 204
 205         return (he_a->ino - he_b->ino);
 206 }
 207
 208 /* Used for sorting the hash entry */
 209 static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
 210 {
 211         const struct hash_entry *he_a = (const struct hash_entry *) a;
 212         const struct hash_entry *he_b = (const struct hash_entry *) b;
 213         unsigned int he_a_len, he_b_len, min_len;
 214         int     ret;
 215
 216         he_a_len = ext2fs_dirent_name_len(he_a->dir);
 217         he_b_len = ext2fs_dirent_name_len(he_b->dir);
 218         min_len = he_a_len;
 219         if (min_len > he_b_len)
 220                 min_len = he_b_len;
 221
 222         ret = memcmp(he_a->dir->name, he_b->dir->name, min_len);
 223         if (ret == 0) {
 224                 if (he_a_len > he_b_len)
 225                         ret = 1;
 226                 else if (he_a_len < he_b_len)
 227                         ret = -1;
 228                 else
 229                         ret = he_b->dir->inode - he_a->dir->inode;
 230         }
 231         return ret;
 232 }
 233
 234 /* Used for sorting the hash entry */
 235 static EXT2_QSORT_TYPE hash_cmp(const void *a, const void *b)
 236 {
 237         const struct hash_entry *he_a = (const struct hash_entry *) a;
 238         const struct hash_entry *he_b = (const struct hash_entry *) b;
 239         int     ret;
 240
 241         if (he_a->hash > he_b->hash)
 242                 ret = 1;
 243         else if (he_a->hash < he_b->hash)
 244                 ret = -1;
 245         else {
 246                 if (he_a->minor_hash > he_b->minor_hash)
 247                         ret = 1;
 248                 else if (he_a->minor_hash < he_b->minor_hash)
 249                         ret = -1;
 250                 else
 251                         ret = name_cmp(a, b);
 252         }
 253         return ret;
 254 }
 255
 256 static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
 257                                 int blocks)
 258 {
 259         void                    *new_mem;
 260
 261         if (outdir->max) {
 262                 new_mem = realloc(outdir->buf, blocks * fs->blocksize);
 263                 if (!new_mem)
 264                         return ENOMEM;
 265                 outdir->buf = new_mem;
 266                 new_mem = realloc(outdir->hashes,
 267                                   blocks * sizeof(ext2_dirhash_t));
 268                 if (!new_mem)
 269                         return ENOMEM;
 270                 outdir->hashes = new_mem;
 271         } else {
 272                 outdir->buf = malloc(blocks * fs->blocksize);
 273                 outdir->hashes = malloc(blocks * sizeof(ext2_dirhash_t));
 274                 outdir->num = 0;
 275         }
 276         outdir->max = blocks;
 277         return 0;
 278 }
 279
 280 static void free_out_dir(struct out_dir *outdir)
 281 {
 282         free(outdir->buf);
 283         free(outdir->hashes);
 284         outdir->max = 0;
 285         outdir->num =0;
 286 }
 287
 288 static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
 289                          char ** ret)
 290 {
 291         errcode_t       retval;
 292
 293         if (outdir->num >= outdir->max) {
 294                 retval = alloc_size_dir(fs, outdir, outdir->max + 50);
 295                 if (retval)
 296                         return retval;
 297         }
 298         *ret = outdir->buf + (outdir->num++ * fs->blocksize);
 299         memset(*ret, 0, fs->blocksize);
 300         return 0;
 301 }
 302
 303 /*
 304  * This function is used to make a unique filename.  We do this by
 305  * appending ~0, and then incrementing the number.  However, we cannot
 306  * expand the length of the filename beyond the padding available in
 307  * the directory entry.
 308  */
 309 static void mutate_name(char *str, unsigned int *len)
 310 {
 311         int i;
 312         unsigned int l = *len;
 313
 314         /*
 315          * First check to see if it looks the name has been mutated
 316          * already
 317          */
 318         for (i = l-1; i > 0; i--) {
 319                 if (!isdigit(str[i]))
 320                         break;
 321         }
 322         if ((i == (int)l - 1) || (str[i] != '~')) {
 323                 if (((l-1) & 3) < 2)
 324                         l += 2;
 325                 else
 326                         l = (l+3) & ~3;
 327                 str[l-2] = '~';
 328                 str[l-1] = '0';
 329                 *len = l;
 330                 return;
 331         }
 332         for (i = l-1; i >= 0; i--) {
 333                 if (isdigit(str[i])) {
 334                         if (str[i] == '9')
 335                                 str[i] = '0';
 336                         else {
 337                                 str[i]++;
 338                                 return;
 339                         }
 340                         continue;
 341                 }
 342                 if (i == 1) {
 343                         if (str[0] == 'z')
 344                                 str[0] = 'A';
 345                         else if (str[0] == 'Z') {
 346                                 str[0] = '~';
 347                                 str[1] = '0';
 348                         } else
 349                                 str[0]++;
 350                 } else if (i > 0) {
 351                         str[i] = '1';
 352                         str[i-1] = '~';
 353                 } else {
 354                         if (str[0] == '~')
 355                                 str[0] = 'a';
 356                         else
 357                                 str[0]++;
 358                 }
 359                 break;
 360         }
 361 }
 362
 363 static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
 364                                     ext2_ino_t ino,
 365                                     struct fill_dir_struct *fd)
 366 {
 367         struct problem_context  pctx;
 368         struct hash_entry       *ent, *prev;
 369         int                     i, j;
 370         int                     fixed = 0;
 371         char                    new_name[256];
 372         unsigned int            new_len;
 373         int                     hash_alg;
 374
 375         clear_problem_context(&pctx);
 376         pctx.ino = ino;
 377
 378         hash_alg = fs->super->s_def_hash_version;
 379         if ((hash_alg <= EXT2_HASH_TEA) &&
 380             (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
 381                 hash_alg += 3;
 382
 383         for (i=1; i < fd->num_array; i++) {
 384                 ent = fd->harray + i;
 385                 prev = ent - 1;
 386                 if (!ent->dir->inode ||
 387                     (ext2fs_dirent_name_len(ent->dir) !=
 388                      ext2fs_dirent_name_len(prev->dir)) ||
 389                     memcmp(ent->dir->name, prev->dir->name,
 390                              ext2fs_dirent_name_len(ent->dir)))
 391                         continue;
 392                 pctx.dirent = ent->dir;
 393                 if ((ent->dir->inode == prev->dir->inode) &&
 394                     fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
 395                         e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
 396                         ent->dir->inode = 0;
 397                         fixed++;
 398                         continue;
 399                 }
 400                 new_len = ext2fs_dirent_name_len(ent->dir);
 401                 memcpy(new_name, ent->dir->name, new_len);
 402                 mutate_name(new_name, &new_len);
 403                 for (j=0; j < fd->num_array; j++) {
 404                         if ((i==j) ||
 405                             (new_len !=
 406                              (unsigned) ext2fs_dirent_name_len(fd->harray[j].dir)) ||
 407                             memcmp(new_name, fd->harray[j].dir->name, new_len))
 408                                 continue;
 409                         mutate_name(new_name, &new_len);
 410
 411                         j = -1;
 412                 }
 413                 new_name[new_len] = 0;
 414                 pctx.str = new_name;
 415                 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
 416                         memcpy(ent->dir->name, new_name, new_len);
 417                         ext2fs_dirent_set_name_len(ent->dir, new_len);
 418                         ext2fs_dirhash(hash_alg, new_name, new_len,
 419                                        fs->super->s_hash_seed,
 420                                        &ent->hash, &ent->minor_hash);
 421                         fixed++;
 422                 }
 423         }
 424         return fixed;
 425 }
 426
 427
 428 static errcode_t copy_dir_entries(e2fsck_t ctx,
 429                                   struct fill_dir_struct *fd,
 430                                   struct out_dir *outdir)
 431 {
 432         ext2_filsys             fs = ctx->fs;
 433         errcode_t               retval;
 434         char                    *block_start;
 435         struct hash_entry       *ent;
 436         struct ext2_dir_entry   *dirent;
 437         unsigned int            rec_len, prev_rec_len, left, slack, offset;
 438         int                     i;
 439         ext2_dirhash_t          prev_hash;
 440         int                     csum_size = 0;
 441         struct                  ext2_dir_entry_tail *t;
 442
 443         if (ctx->htree_slack_percentage == 255) {
 444                 profile_get_uint(ctx->profile, "options",
 445                                  "indexed_dir_slack_percentage",
 446                                  0, 20,
 447                                  &ctx->htree_slack_percentage);
 448                 if (ctx->htree_slack_percentage > 100)
 449                         ctx->htree_slack_percentage = 20;
 450         }
 451
 452         if (ext2fs_has_feature_metadata_csum(fs->super))
 453                 csum_size = sizeof(struct ext2_dir_entry_tail);
 454
 455         outdir->max = 0;
 456         retval = alloc_size_dir(fs, outdir,
 457                                 (fd->dir_size / fs->blocksize) + 2);
 458         if (retval)
 459                 return retval;
 460         outdir->num = fd->compress ? 0 : 1;
 461         offset = 0;
 462         outdir->hashes[0] = 0;
 463         prev_hash = 1;
 464         if ((retval = get_next_block(fs, outdir, &block_start)))
 465                 return retval;
 466         dirent = (struct ext2_dir_entry *) block_start;
 467         prev_rec_len = 0;
 468         rec_len = 0;
 469         left = fs->blocksize - csum_size;
 470         slack = fd->compress ? 12 :
 471                 ((fs->blocksize - csum_size) * ctx->htree_slack_percentage)/100;
 472         if (slack < 12)
 473                 slack = 12;
 474         for (i = 0; i < fd->num_array; i++) {
 475                 ent = fd->harray + i;
 476                 if (ent->dir->inode == 0)
 477                         continue;
 478                 rec_len = EXT2_DIR_REC_LEN(ext2fs_dirent_name_len(ent->dir));
 479                 if (rec_len > left) {
 480                         if (left) {
 481                                 left += prev_rec_len;
 482                                 retval = ext2fs_set_rec_len(fs, left, dirent);
 483                                 if (retval)
 484                                         return retval;
 485                         }
 486                         if (csum_size) {
 487                                 t = EXT2_DIRENT_TAIL(block_start,
 488                                                      fs->blocksize);
 489                                 ext2fs_initialize_dirent_tail(fs, t);
 490                         }
 491                         if ((retval = get_next_block(fs, outdir,
 492                                                       &block_start)))
 493                                 return retval;
 494                         offset = 0;
 495                 }
 496                 left = (fs->blocksize - csum_size) - offset;
 497                 dirent = (struct ext2_dir_entry *) (block_start + offset);
 498                 if (offset == 0) {
 499                         if (ent->hash == prev_hash)
 500                                 outdir->hashes[outdir->num-1] = ent->hash | 1;
 501                         else
 502                                 outdir->hashes[outdir->num-1] = ent->hash;
 503                 }
 504                 dirent->inode = ent->dir->inode;
 505                 ext2fs_dirent_set_name_len(dirent,
 506                                            ext2fs_dirent_name_len(ent->dir));
 507                 ext2fs_dirent_set_file_type(dirent,
 508                                             ext2fs_dirent_file_type(ent->dir));
 509                 retval = ext2fs_set_rec_len(fs, rec_len, dirent);
 510                 if (retval)
 511                         return retval;
 512                 prev_rec_len = rec_len;
 513                 memcpy(dirent->name, ent->dir->name,
 514                        ext2fs_dirent_name_len(dirent));
 515                 offset += rec_len;
 516                 left -= rec_len;
 517                 if (left < slack) {
 518                         prev_rec_len += left;
 519                         retval = ext2fs_set_rec_len(fs, prev_rec_len, dirent);
 520                         if (retval)
 521                                 return retval;
 522                         offset += left;
 523                         left = 0;
 524                 }
 525                 prev_hash = ent->hash;
 526         }
 527         if (left)
 528                 retval = ext2fs_set_rec_len(fs, rec_len + left, dirent);
 529         if (csum_size) {
 530                 t = EXT2_DIRENT_TAIL(block_start, fs->blocksize);
 531                 ext2fs_initialize_dirent_tail(fs, t);
 532         }
 533
 534         return retval;
 535 }
 536
 537
 538 static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
 539                                     ext2_ino_t ino, ext2_ino_t parent)
 540 {
 541         struct ext2_dir_entry           *dir;
 542         struct ext2_dx_root_info        *root;
 543         struct ext2_dx_countlimit       *limits;
 544         int                             filetype = 0;
 545         int                             csum_size = 0;
 546
 547         if (ext2fs_has_feature_filetype(fs->super))
 548                 filetype = EXT2_FT_DIR;
 549
 550         memset(buf, 0, fs->blocksize);
 551         dir = (struct ext2_dir_entry *) buf;
 552         dir->inode = ino;
 553         dir->name[0] = '.';
 554         ext2fs_dirent_set_name_len(dir, 1);
 555         ext2fs_dirent_set_file_type(dir, filetype);
 556         dir->rec_len = 12;
 557         dir = (struct ext2_dir_entry *) (buf + 12);
 558         dir->inode = parent;
 559         dir->name[0] = '.';
 560         dir->name[1] = '.';
 561         ext2fs_dirent_set_name_len(dir, 2);
 562         ext2fs_dirent_set_file_type(dir, filetype);
 563         dir->rec_len = fs->blocksize - 12;
 564
 565         root = (struct ext2_dx_root_info *) (buf+24);
 566         root->reserved_zero = 0;
 567         root->hash_version = fs->super->s_def_hash_version;
 568         root->info_length = 8;
 569         root->indirect_levels = 0;
 570         root->unused_flags = 0;
 571
 572         if (ext2fs_has_feature_metadata_csum(fs->super))
 573                 csum_size = sizeof(struct ext2_dx_tail);
 574
 575         limits = (struct ext2_dx_countlimit *) (buf+32);
 576         limits->limit = (fs->blocksize - (32 + csum_size)) /
 577                         sizeof(struct ext2_dx_entry);
 578         limits->count = 0;
 579
 580         return root;
 581 }
 582
 583
 584 static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
 585 {
 586         struct ext2_dir_entry           *dir;
 587         struct ext2_dx_countlimit       *limits;
 588         int                             csum_size = 0;
 589
 590         memset(buf, 0, fs->blocksize);
 591         dir = (struct ext2_dir_entry *) buf;
 592         dir->inode = 0;
 593         (void) ext2fs_set_rec_len(fs, fs->blocksize, dir);
 594
 595         if (ext2fs_has_feature_metadata_csum(fs->super))
 596                 csum_size = sizeof(struct ext2_dx_tail);
 597
 598         limits = (struct ext2_dx_countlimit *) (buf+8);
 599         limits->limit = (fs->blocksize - (8 + csum_size)) /
 600                         sizeof(struct ext2_dx_entry);
 601         limits->count = 0;
 602
 603         return (struct ext2_dx_entry *) limits;
 604 }
 605
 606 static int alloc_blocks(ext2_filsys fs,
 607                         struct ext2_dx_countlimit **limit,
 608                         struct ext2_dx_entry **prev_ent,
 609                         struct ext2_dx_entry **next_ent,
 610                         int *prev_offset, int *next_offset,
 611                         struct out_dir *outdir, int i,
 612                         int *prev_count, int *next_count)
 613 {
 614         errcode_t       retval;
 615         char            *block_start;
 616
 617         if (*limit)
 618                 (*limit)->limit = (*limit)->count =
 619                         ext2fs_cpu_to_le16((*limit)->limit);
 620         *prev_ent = (struct ext2_dx_entry *) (outdir->buf + *prev_offset);
 621         (*prev_ent)->block = ext2fs_cpu_to_le32(outdir->num);
 622
 623         if (i != 1)
 624                 (*prev_ent)->hash =
 625                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 626
 627         retval = get_next_block(fs, outdir, &block_start);
 628         if (retval)
 629                 return retval;
 630
 631         *next_ent = set_int_node(fs, block_start);
 632         *limit = (struct ext2_dx_countlimit *)(*next_ent);
 633         if (next_offset)
 634                 *next_offset = ((char *) *next_ent - outdir->buf);
 635
 636         *next_count = (*limit)->limit;
 637         (*prev_offset) += sizeof(struct ext2_dx_entry);
 638         (*prev_count)--;
 639
 640         return 0;
 641 }
 642
 643 /*
 644  * This function takes the leaf nodes which have been written in
 645  * outdir, and populates the root node and any necessary interior nodes.
 646  */
 647 static errcode_t calculate_tree(ext2_filsys fs,
 648                                 struct out_dir *outdir,
 649                                 ext2_ino_t ino,
 650                                 ext2_ino_t parent)
 651 {
 652         struct ext2_dx_root_info        *root_info;
 653         struct ext2_dx_entry            *root, *int_ent, *dx_ent = 0;
 654         struct ext2_dx_countlimit       *root_limit, *int_limit, *limit;
 655         errcode_t                       retval;
 656         int                             i, c1, c2, c3, nblks;
 657         int                             limit_offset, int_offset, root_offset;
 658
 659         root_info = set_root_node(fs, outdir->buf, ino, parent);
 660         root_offset = limit_offset = ((char *) root_info - outdir->buf) +
 661                 root_info->info_length;
 662         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 663         c1 = root_limit->limit;
 664         nblks = outdir->num;
 665
 666         /* Write out the pointer blocks */
 667         if (nblks - 1 <= c1) {
 668                 /* Just write out the root block, and we're done */
 669                 root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
 670                 for (i=1; i < nblks; i++) {
 671                         root->block = ext2fs_cpu_to_le32(i);
 672                         if (i != 1)
 673                                 root->hash =
 674                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 675                         root++;
 676                         c1--;
 677                 }
 678         } else if (nblks - 1 <= ext2fs_htree_intnode_maxrecs(fs, c1)) {
 679                 c2 = 0;
 680                 limit = NULL;
 681                 root_info->indirect_levels = 1;
 682                 for (i=1; i < nblks; i++) {
 683                         if (c2 == 0 && c1 == 0)
 684                                 return ENOSPC;
 685                         if (c2 == 0) {
 686                                 retval = alloc_blocks(fs, &limit, &root,
 687                                                       &dx_ent, &root_offset,
 688                                                       NULL, outdir, i, &c1,
 689                                                       &c2);
 690                                 if (retval)
 691                                         return retval;
 692                         }
 693                         dx_ent->block = ext2fs_cpu_to_le32(i);
 694                         if (c2 != limit->limit)
 695                                 dx_ent->hash =
 696                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 697                         dx_ent++;
 698                         c2--;
 699                 }
 700                 limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
 701                 limit->limit = ext2fs_cpu_to_le16(limit->limit);
 702         } else {
 703                 c2 = 0;
 704                 c3 = 0;
 705                 limit = NULL;
 706                 int_limit = 0;
 707                 root_info->indirect_levels = 2;
 708                 for (i = 1; i < nblks; i++) {
 709                         if (c3 == 0 && c2 == 0 && c1 == 0)
 710                                 return ENOSPC;
 711                         if (c3 == 0 && c2 == 0) {
 712                                 retval = alloc_blocks(fs, &int_limit, &root,
 713                                                       &int_ent, &root_offset,
 714                                                       &int_offset, outdir, i,
 715                                                       &c1, &c2);
 716                                 if (retval)
 717                                         return retval;
 718                         }
 719                         if (c3 == 0) {
 720                                 retval = alloc_blocks(fs, &limit, &int_ent,
 721                                                       &dx_ent, &int_offset,
 722                                                       NULL, outdir, i, &c2,
 723                                                       &c3);
 724                                 if (retval)
 725                                         return retval;
 726
 727                         }
 728                         dx_ent->block = ext2fs_cpu_to_le32(i);
 729                         if (c3 != limit->limit)
 730                                 dx_ent->hash =
 731                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 732                         dx_ent++;
 733                         c3--;
 734                 }
 735                 int_limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
 736                 int_limit->limit = ext2fs_cpu_to_le16(limit->limit);
 737
 738                 limit->count = ext2fs_cpu_to_le16(limit->limit - c3);
 739                 limit->limit = ext2fs_cpu_to_le16(limit->limit);
 740
 741         }
 742         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 743         root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
 744         root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
 745
 746         return 0;
 747 }
 748
 749 struct write_dir_struct {
 750         struct out_dir *outdir;
 751         errcode_t       err;
 752         ext2_ino_t      ino;
 753         e2fsck_t        ctx;
 754         ext2_ino_t      dir;
 755 };
 756
 757 /*
 758  * Helper function which writes out a directory block.
 759  */
 760 static int write_dir_block(ext2_filsys fs,
 761                            blk64_t *block_nr,
 762                            e2_blkcnt_t blockcnt,
 763                            blk64_t ref_block EXT2FS_ATTR((unused)),
 764                            int ref_offset EXT2FS_ATTR((unused)),
 765                            void *priv_data)
 766 {
 767         struct write_dir_struct *wd = (struct write_dir_struct *) priv_data;
 768         char    *dir, *buf = 0;
 769
 770 #ifdef REHASH_DEBUG
 771         printf("%u: write_dir_block %lld:%lld", wd->ino, blockcnt, *block_nr);
 772 #endif
 773         if ((*block_nr == 0) || (blockcnt < 0)) {
 774 #ifdef REHASH_DEBUG
 775                 printf(" - skip\n");
 776 #endif
 777                 return 0;
 778         }
 779         if (blockcnt < wd->outdir->num)
 780                 dir = wd->outdir->buf + (blockcnt * fs->blocksize);
 781         else if (wd->ctx->lost_and_found == wd->dir) {
 782                 /* Don't release any extra directory blocks for lost+found */
 783                 wd->err = ext2fs_new_dir_block(fs, 0, 0, &buf);
 784                 if (wd->err)
 785                         return BLOCK_ABORT;
 786                 dir = buf;
 787                 wd->outdir->num++;
 788         } else {
 789                 /* Don't free blocks at the end of the directory, they
 790                  * will be truncated by the caller. */
 791 #ifdef REHASH_DEBUG
 792                 printf(" - not freed\n");
 793 #endif
 794                 return 0;
 795         }
 796         wd->err = ext2fs_write_dir_block4(fs, *block_nr, dir, 0, wd->dir);
 797         if (buf)
 798                 ext2fs_free_mem(&buf);
 799
 800 #ifdef REHASH_DEBUG
 801         printf(" - write (%d)\n", wd->err);
 802 #endif
 803         if (wd->err)
 804                 return BLOCK_ABORT;
 805         return 0;
 806 }
 807
 808 static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
 809                                  struct out_dir *outdir,
 810                                  ext2_ino_t ino, struct ext2_inode *inode,
 811                                  int compress)
 812 {
 813         struct write_dir_struct wd;
 814         errcode_t       retval;
 815
 816         retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
 817         if (retval)
 818                 return retval;
 819
 820         wd.outdir = outdir;
 821         wd.err = 0;
 822         wd.ino = ino;
 823         wd.ctx = ctx;
 824         wd.dir = ino;
 825
 826         retval = ext2fs_block_iterate3(fs, ino, 0, NULL,
 827                                        write_dir_block, &wd);
 828         if (retval)
 829                 return retval;
 830         if (wd.err)
 831                 return wd.err;
 832
 833         e2fsck_read_inode(ctx, ino, inode, "rehash_dir");
 834         if (compress)
 835                 inode->i_flags &= ~EXT2_INDEX_FL;
 836         else
 837                 inode->i_flags |= EXT2_INDEX_FL;
 838 #ifdef REHASH_DEBUG
 839         printf("%u: set inode size to %u blocks = %u bytes\n",
 840                ino, outdir->num, outdir->num * fs->blocksize);
 841 #endif
 842         retval = ext2fs_inode_size_set(fs, inode, (ext2_off64_t)outdir->num *
 843                                                    fs->blocksize);
 844         if (retval)
 845                 return retval;
 846
 847         /* ext2fs_punch() calls ext2fs_write_inode() which writes the size */
 848         return ext2fs_punch(fs, ino, inode, NULL, outdir->num, ~0ULL);
 849 }
 850
 851 errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino,
 852                             struct problem_context *pctx)
 853 {
 854         ext2_filsys             fs = ctx->fs;
 855         errcode_t               retval;
 856         struct ext2_inode       inode;
 857         char                    *dir_buf = 0;
 858         struct fill_dir_struct  fd = { NULL, NULL, 0, 0, 0, NULL,
 859                                        0, 0, 0, 0, 0, 0 };
 860         struct out_dir          outdir = { 0, 0, 0, 0 };
 861
 862         e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
 863
 864         if (ext2fs_has_feature_inline_data(fs->super) &&
 865            (inode.i_flags & EXT4_INLINE_DATA_FL))
 866                 return 0;
 867
 868         retval = ENOMEM;
 869         dir_buf = malloc(inode.i_size);
 870         if (!dir_buf)
 871                 goto errout;
 872
 873         fd.max_array = inode.i_size / 32;
 874         fd.harray = malloc(fd.max_array * sizeof(struct hash_entry));
 875         if (!fd.harray)
 876                 goto errout;
 877
 878         fd.ino = ino;
 879         fd.ctx = ctx;
 880         fd.buf = dir_buf;
 881         fd.inode = &inode;
 882         fd.dir = ino;
 883         if (!ext2fs_has_feature_dir_index(fs->super) ||
 884             (inode.i_size / fs->blocksize) < 2)
 885                 fd.compress = 1;
 886         fd.parent = 0;
 887
 888 retry_nohash:
 889         /* Read in the entire directory into memory */
 890         retval = ext2fs_block_iterate3(fs, ino, 0, 0,
 891                                        fill_dir_block, &fd);
 892         if (fd.err) {
 893                 retval = fd.err;
 894                 goto errout;
 895         }
 896
 897         /*
 898          * If the entries read are less than a block, then don't index
 899          * the directory
 900          */
 901         if (!fd.compress && (fd.dir_size < (fs->blocksize - 24))) {
 902                 fd.compress = 1;
 903                 fd.dir_size = 0;
 904                 fd.num_array = 0;
 905                 goto retry_nohash;
 906         }
 907
 908 #if 0
 909         printf("%d entries (%d bytes) found in inode %d\n",
 910                fd.num_array, fd.dir_size, ino);
 911 #endif
 912
 913         /* Sort the list */
 914 resort:
 915         if (fd.compress && fd.num_array > 1)
 916                 qsort(fd.harray+2, fd.num_array-2, sizeof(struct hash_entry),
 917                       hash_cmp);
 918         else
 919                 qsort(fd.harray, fd.num_array, sizeof(struct hash_entry),
 920                       hash_cmp);
 921
 922         /*
 923          * Look for duplicates
 924          */
 925         if (duplicate_search_and_fix(ctx, fs, ino, &fd))
 926                 goto resort;
 927
 928         if (ctx->options & E2F_OPT_NO) {
 929                 retval = 0;
 930                 goto errout;
 931         }
 932
 933         /* Sort non-hashed directories by inode number */
 934         if (fd.compress && fd.num_array > 1)
 935                 qsort(fd.harray+2, fd.num_array-2,
 936                       sizeof(struct hash_entry), ino_cmp);
 937
 938         /*
 939          * Copy the directory entries.  In a htree directory these
 940          * will become the leaf nodes.
 941          */
 942         retval = copy_dir_entries(ctx, &fd, &outdir);
 943         if (retval)
 944                 goto errout;
 945
 946         free(dir_buf); dir_buf = 0;
 947
 948         if (!fd.compress) {
 949                 /* Calculate the interior nodes */
 950                 retval = calculate_tree(fs, &outdir, ino, fd.parent);
 951                 if (retval)
 952                         goto errout;
 953         }
 954
 955         retval = write_directory(ctx, fs, &outdir, ino, &inode, fd.compress);
 956         if (retval)
 957                 goto errout;
 958
 959         if (ctx->options & E2F_OPT_CONVERT_BMAP)
 960                 retval = e2fsck_rebuild_extents_later(ctx, ino);
 961         else
 962                 retval = e2fsck_check_rebuild_extents(ctx, ino, &inode, pctx);
 963 errout:
 964         free(dir_buf);
 965         free(fd.harray);
 966
 967         free_out_dir(&outdir);
 968         return retval;
 969 }
 970
 971 void e2fsck_rehash_directories(e2fsck_t ctx)
 972 {
 973         struct problem_context  pctx;
 974 #ifdef RESOURCE_TRACK
 975         struct resource_track   rtrack;
 976 #endif
 977         struct dir_info         *dir;
 978         ext2_u32_iterate        iter;
 979         struct dir_info_iter *  dirinfo_iter = 0;
 980         ext2_ino_t              ino;
 981         errcode_t               retval;
 982         int                     cur, max, all_dirs, first = 1;
 983
 984         init_resource_track(&rtrack, ctx->fs->io);
 985         all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
 986
 987         if (!ctx->dirs_to_hash && !all_dirs)
 988                 return;
 989
 990         (void) e2fsck_get_lost_and_found(ctx, 0);
 991
 992         clear_problem_context(&pctx);
 993
 994         cur = 0;
 995         if (all_dirs) {
 996                 dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
 997                 max = e2fsck_get_num_dirinfo(ctx);
 998         } else {
 999                 retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
1000                                                        &iter);
1001                 if (retval) {
1002                         pctx.errcode = retval;
1003                         fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
1004                         return;
1005                 }
1006                 max = ext2fs_u32_list_count(ctx->dirs_to_hash);
1007         }
1008         while (1) {
1009                 if (all_dirs) {
1010                         if ((dir = e2fsck_dir_info_iter(ctx,
1011                                                         dirinfo_iter)) == 0)
1012                                 break;
1013                         ino = dir->ino;
1014                 } else {
1015                         if (!ext2fs_u32_list_iterate(iter, &ino))
1016                                 break;
1017                 }
1018
1019                 pctx.dir = ino;
1020                 if (first) {
1021                         fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
1022                         first = 0;
1023                 }
1024 #if 0
1025                 fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
1026 #endif
1027                 pctx.errcode = e2fsck_rehash_dir(ctx, ino, &pctx);
1028                 if (pctx.errcode) {
1029                         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
1030                         fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
1031                 }
1032                 if (ctx->progress && !ctx->progress_fd)
1033                         e2fsck_simple_progress(ctx, "Rebuilding directory",
1034                                100.0 * (float) (++cur) / (float) max, ino);
1035         }
1036         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
1037         if (all_dirs)
1038                 e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
1039         else
1040                 ext2fs_u32_list_iterate_end(iter);
1041
1042         if (ctx->dirs_to_hash)
1043                 ext2fs_u32_list_free(ctx->dirs_to_hash);
1044         ctx->dirs_to_hash = 0;
1045
1046         print_resource_track(ctx, "Pass 3A", &rtrack, ctx->fs->io);
1047 }