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 struct fill_dir_struct {
  56         char *buf;
  57         struct ext2_inode *inode;
  58         errcode_t err;
  59         e2fsck_t ctx;
  60         struct hash_entry *harray;
  61         int max_array, num_array;
  62         unsigned int dir_size;
  63         int compress;
  64         ino_t parent;
  65 };
  66
  67 struct hash_entry {
  68         ext2_dirhash_t  hash;
  69         ext2_dirhash_t  minor_hash;
  70         ino_t           ino;
  71         struct ext2_dir_entry   *dir;
  72 };
  73
  74 struct out_dir {
  75         int             num;
  76         int             max;
  77         char            *buf;
  78         ext2_dirhash_t  *hashes;
  79 };
  80
  81 static int fill_dir_block(ext2_filsys fs,
  82                           blk64_t *block_nr,
  83                           e2_blkcnt_t blockcnt,
  84                           blk64_t ref_block EXT2FS_ATTR((unused)),
  85                           int ref_offset EXT2FS_ATTR((unused)),
  86                           void *priv_data)
  87 {
  88         struct fill_dir_struct  *fd = (struct fill_dir_struct *) priv_data;
  89         struct hash_entry       *new_array, *ent;
  90         struct ext2_dir_entry   *dirent;
  91         char                    *dir;
  92         unsigned int            offset, dir_offset, rec_len;
  93         int                     hash_alg;
  94
  95         if (blockcnt < 0)
  96                 return 0;
  97
  98         offset = blockcnt * fs->blocksize;
  99         if (offset + fs->blocksize > fd->inode->i_size) {
 100                 fd->err = EXT2_ET_DIR_CORRUPTED;
 101                 return BLOCK_ABORT;
 102         }
 103         dir = (fd->buf+offset);
 104         if (HOLE_BLKADDR(*block_nr)) {
 105                 memset(dir, 0, fs->blocksize);
 106                 dirent = (struct ext2_dir_entry *) dir;
 107                 (void) ext2fs_set_rec_len(fs, fs->blocksize, dirent);
 108         } else {
 109                 fd->err = ext2fs_read_dir_block3(fs, *block_nr, dir, 0);
 110                 if (fd->err)
 111                         return BLOCK_ABORT;
 112         }
 113         hash_alg = fs->super->s_def_hash_version;
 114         if ((hash_alg <= EXT2_HASH_TEA) &&
 115             (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
 116                 hash_alg += 3;
 117         /* While the directory block is "hot", index it. */
 118         dir_offset = 0;
 119         while (dir_offset < fs->blocksize) {
 120                 dirent = (struct ext2_dir_entry *) (dir + dir_offset);
 121                 (void) ext2fs_get_rec_len(fs, dirent, &rec_len);
 122                 if (((dir_offset + rec_len) > fs->blocksize) ||
 123                     (rec_len < 8) ||
 124                     ((rec_len % 4) != 0) ||
 125                     (((dirent->name_len & 0xFF)+8U) > rec_len)) {
 126                         fd->err = EXT2_ET_DIR_CORRUPTED;
 127                         return BLOCK_ABORT;
 128                 }
 129                 dir_offset += rec_len;
 130                 if (dirent->inode == 0)
 131                         continue;
 132                 if (!fd->compress && ((dirent->name_len&0xFF) == 1) &&
 133                     (dirent->name[0] == '.'))
 134                         continue;
 135                 if (!fd->compress && ((dirent->name_len&0xFF) == 2) &&
 136                     (dirent->name[0] == '.') && (dirent->name[1] == '.')) {
 137                         fd->parent = dirent->inode;
 138                         continue;
 139                 }
 140                 if (fd->num_array >= fd->max_array) {
 141                         new_array = realloc(fd->harray,
 142                             sizeof(struct hash_entry) * (fd->max_array+500));
 143                         if (!new_array) {
 144                                 fd->err = ENOMEM;
 145                                 return BLOCK_ABORT;
 146                         }
 147                         fd->harray = new_array;
 148                         fd->max_array += 500;
 149                 }
 150                 ent = fd->harray + fd->num_array++;
 151                 ent->dir = dirent;
 152                 fd->dir_size += EXT2_DIR_REC_LEN(dirent->name_len & 0xFF);
 153                 ent->ino = dirent->inode;
 154                 if (fd->compress)
 155                         ent->hash = ent->minor_hash = 0;
 156                 else {
 157                         fd->err = ext2fs_dirhash(hash_alg, dirent->name,
 158                                                  dirent->name_len & 0xFF,
 159                                                  fs->super->s_hash_seed,
 160                                                  &ent->hash, &ent->minor_hash);
 161                         if (fd->err)
 162                                 return BLOCK_ABORT;
 163                 }
 164         }
 165
 166         return 0;
 167 }
 168
 169 /* Used for sorting the hash entry */
 170 static EXT2_QSORT_TYPE ino_cmp(const void *a, const void *b)
 171 {
 172         const struct hash_entry *he_a = (const struct hash_entry *) a;
 173         const struct hash_entry *he_b = (const struct hash_entry *) b;
 174
 175         return (he_a->ino - he_b->ino);
 176 }
 177
 178 /* Used for sorting the hash entry */
 179 static EXT2_QSORT_TYPE name_cmp(const void *a, const void *b)
 180 {
 181         const struct hash_entry *he_a = (const struct hash_entry *) a;
 182         const struct hash_entry *he_b = (const struct hash_entry *) b;
 183         int     ret;
 184         int     min_len;
 185
 186         min_len = he_a->dir->name_len;
 187         if (min_len > he_b->dir->name_len)
 188                 min_len = he_b->dir->name_len;
 189
 190         ret = strncmp(he_a->dir->name, he_b->dir->name, min_len);
 191         if (ret == 0) {
 192                 if (he_a->dir->name_len > he_b->dir->name_len)
 193                         ret = 1;
 194                 else if (he_a->dir->name_len < he_b->dir->name_len)
 195                         ret = -1;
 196                 else
 197                         ret = he_b->dir->inode - he_a->dir->inode;
 198         }
 199         return ret;
 200 }
 201
 202 /* Used for sorting the hash entry */
 203 static EXT2_QSORT_TYPE hash_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         int     ret;
 208
 209         if (he_a->hash > he_b->hash)
 210                 ret = 1;
 211         else if (he_a->hash < he_b->hash)
 212                 ret = -1;
 213         else {
 214                 if (he_a->minor_hash > he_b->minor_hash)
 215                         ret = 1;
 216                 else if (he_a->minor_hash < he_b->minor_hash)
 217                         ret = -1;
 218                 else
 219                         ret = name_cmp(a, b);
 220         }
 221         return ret;
 222 }
 223
 224 static errcode_t alloc_size_dir(ext2_filsys fs, struct out_dir *outdir,
 225                                 int blocks)
 226 {
 227         void                    *new_mem;
 228
 229         if (outdir->max) {
 230                 new_mem = realloc(outdir->buf, blocks * fs->blocksize);
 231                 if (!new_mem)
 232                         return ENOMEM;
 233                 outdir->buf = new_mem;
 234                 new_mem = realloc(outdir->hashes,
 235                                   blocks * sizeof(ext2_dirhash_t));
 236                 if (!new_mem)
 237                         return ENOMEM;
 238                 outdir->hashes = new_mem;
 239         } else {
 240                 outdir->buf = malloc(blocks * fs->blocksize);
 241                 outdir->hashes = malloc(blocks * sizeof(ext2_dirhash_t));
 242                 outdir->num = 0;
 243         }
 244         outdir->max = blocks;
 245         return 0;
 246 }
 247
 248 static void free_out_dir(struct out_dir *outdir)
 249 {
 250         free(outdir->buf);
 251         free(outdir->hashes);
 252         outdir->max = 0;
 253         outdir->num =0;
 254 }
 255
 256 static errcode_t get_next_block(ext2_filsys fs, struct out_dir *outdir,
 257                          char ** ret)
 258 {
 259         errcode_t       retval;
 260
 261         if (outdir->num >= outdir->max) {
 262                 retval = alloc_size_dir(fs, outdir, outdir->max + 50);
 263                 if (retval)
 264                         return retval;
 265         }
 266         *ret = outdir->buf + (outdir->num++ * fs->blocksize);
 267         memset(*ret, 0, fs->blocksize);
 268         return 0;
 269 }
 270
 271 /*
 272  * This function is used to make a unique filename.  We do this by
 273  * appending ~0, and then incrementing the number.  However, we cannot
 274  * expand the length of the filename beyond the padding available in
 275  * the directory entry.
 276  */
 277 static void mutate_name(char *str, __u16 *len)
 278 {
 279         int     i;
 280         __u16   l = *len & 0xFF, h = *len & 0xff00;
 281
 282         /*
 283          * First check to see if it looks the name has been mutated
 284          * already
 285          */
 286         for (i = l-1; i > 0; i--) {
 287                 if (!isdigit(str[i]))
 288                         break;
 289         }
 290         if ((i == l-1) || (str[i] != '~')) {
 291                 if (((l-1) & 3) < 2)
 292                         l += 2;
 293                 else
 294                         l = (l+3) & ~3;
 295                 str[l-2] = '~';
 296                 str[l-1] = '0';
 297                 *len = l | h;
 298                 return;
 299         }
 300         for (i = l-1; i >= 0; i--) {
 301                 if (isdigit(str[i])) {
 302                         if (str[i] == '9')
 303                                 str[i] = '0';
 304                         else {
 305                                 str[i]++;
 306                                 return;
 307                         }
 308                         continue;
 309                 }
 310                 if (i == 1) {
 311                         if (str[0] == 'z')
 312                                 str[0] = 'A';
 313                         else if (str[0] == 'Z') {
 314                                 str[0] = '~';
 315                                 str[1] = '0';
 316                         } else
 317                                 str[0]++;
 318                 } else if (i > 0) {
 319                         str[i] = '1';
 320                         str[i-1] = '~';
 321                 } else {
 322                         if (str[0] == '~')
 323                                 str[0] = 'a';
 324                         else
 325                                 str[0]++;
 326                 }
 327                 break;
 328         }
 329 }
 330
 331 static int duplicate_search_and_fix(e2fsck_t ctx, ext2_filsys fs,
 332                                     ext2_ino_t ino,
 333                                     struct fill_dir_struct *fd)
 334 {
 335         struct problem_context  pctx;
 336         struct hash_entry       *ent, *prev;
 337         int                     i, j;
 338         int                     fixed = 0;
 339         char                    new_name[256];
 340         __u16                   new_len;
 341         int                     hash_alg;
 342
 343         clear_problem_context(&pctx);
 344         pctx.ino = ino;
 345
 346         hash_alg = fs->super->s_def_hash_version;
 347         if ((hash_alg <= EXT2_HASH_TEA) &&
 348             (fs->super->s_flags & EXT2_FLAGS_UNSIGNED_HASH))
 349                 hash_alg += 3;
 350
 351         for (i=1; i < fd->num_array; i++) {
 352                 ent = fd->harray + i;
 353                 prev = ent - 1;
 354                 if (!ent->dir->inode ||
 355                     ((ent->dir->name_len & 0xFF) !=
 356                      (prev->dir->name_len & 0xFF)) ||
 357                     (strncmp(ent->dir->name, prev->dir->name,
 358                              ent->dir->name_len & 0xFF)))
 359                         continue;
 360                 pctx.dirent = ent->dir;
 361                 if ((ent->dir->inode == prev->dir->inode) &&
 362                     fix_problem(ctx, PR_2_DUPLICATE_DIRENT, &pctx)) {
 363                         e2fsck_adjust_inode_count(ctx, ent->dir->inode, -1);
 364                         ent->dir->inode = 0;
 365                         fixed++;
 366                         continue;
 367                 }
 368                 memcpy(new_name, ent->dir->name, ent->dir->name_len & 0xFF);
 369                 new_len = ent->dir->name_len;
 370                 mutate_name(new_name, &new_len);
 371                 for (j=0; j < fd->num_array; j++) {
 372                         if ((i==j) ||
 373                             ((new_len & 0xFF) !=
 374                              (fd->harray[j].dir->name_len & 0xFF)) ||
 375                             (strncmp(new_name, fd->harray[j].dir->name,
 376                                      new_len & 0xFF)))
 377                                 continue;
 378                         mutate_name(new_name, &new_len);
 379
 380                         j = -1;
 381                 }
 382                 new_name[new_len & 0xFF] = 0;
 383                 pctx.str = new_name;
 384                 if (fix_problem(ctx, PR_2_NON_UNIQUE_FILE, &pctx)) {
 385                         memcpy(ent->dir->name, new_name, new_len & 0xFF);
 386                         ent->dir->name_len = new_len;
 387                         ext2fs_dirhash(hash_alg, ent->dir->name,
 388                                        ent->dir->name_len & 0xFF,
 389                                        fs->super->s_hash_seed,
 390                                        &ent->hash, &ent->minor_hash);
 391                         fixed++;
 392                 }
 393         }
 394         return fixed;
 395 }
 396
 397
 398 static errcode_t copy_dir_entries(e2fsck_t ctx,
 399                                   struct fill_dir_struct *fd,
 400                                   struct out_dir *outdir)
 401 {
 402         ext2_filsys             fs = ctx->fs;
 403         errcode_t               retval;
 404         char                    *block_start;
 405         struct hash_entry       *ent;
 406         struct ext2_dir_entry   *dirent;
 407         unsigned int            rec_len, prev_rec_len, left, slack, offset;
 408         int                     i;
 409         ext2_dirhash_t          prev_hash;
 410
 411         if (ctx->htree_slack_percentage == 255) {
 412                 profile_get_uint(ctx->profile, "options",
 413                                  "indexed_dir_slack_percentage",
 414                                  0, 20,
 415                                  &ctx->htree_slack_percentage);
 416                 if (ctx->htree_slack_percentage > 100)
 417                         ctx->htree_slack_percentage = 20;
 418         }
 419
 420         outdir->max = 0;
 421         retval = alloc_size_dir(fs, outdir,
 422                                 (fd->dir_size / fs->blocksize) + 2);
 423         if (retval)
 424                 return retval;
 425         outdir->num = fd->compress ? 0 : 1;
 426         offset = 0;
 427         outdir->hashes[0] = 0;
 428         prev_hash = 1;
 429         if ((retval = get_next_block(fs, outdir, &block_start)))
 430                 return retval;
 431         dirent = (struct ext2_dir_entry *) block_start;
 432         prev_rec_len = 0;
 433         rec_len = 0;
 434         left = fs->blocksize;
 435         slack = fd->compress ? 12 :
 436                 (fs->blocksize * ctx->htree_slack_percentage)/100;
 437         if (slack < 12)
 438                 slack = 12;
 439         for (i = 0; i < fd->num_array; i++) {
 440                 ent = fd->harray + i;
 441                 if (ent->dir->inode == 0)
 442                         continue;
 443                 rec_len = EXT2_DIR_REC_LEN(ent->dir->name_len & 0xFF);
 444                 if (rec_len > left) {
 445                         if (left) {
 446                                 left += prev_rec_len;
 447                                 retval = ext2fs_set_rec_len(fs, left, dirent);
 448                                 if (retval)
 449                                         return retval;
 450                         }
 451                         if ((retval = get_next_block(fs, outdir,
 452                                                       &block_start)))
 453                                 return retval;
 454                         offset = 0;
 455                 }
 456                 left = fs->blocksize - offset;
 457                 dirent = (struct ext2_dir_entry *) (block_start + offset);
 458                 if (offset == 0) {
 459                         if (ent->hash == prev_hash)
 460                                 outdir->hashes[outdir->num-1] = ent->hash | 1;
 461                         else
 462                                 outdir->hashes[outdir->num-1] = ent->hash;
 463                 }
 464                 dirent->inode = ent->dir->inode;
 465                 dirent->name_len = ent->dir->name_len;
 466                 retval = ext2fs_set_rec_len(fs, rec_len, dirent);
 467                 if (retval)
 468                         return retval;
 469                 prev_rec_len = rec_len;
 470                 memcpy(dirent->name, ent->dir->name, dirent->name_len & 0xFF);
 471                 offset += rec_len;
 472                 left -= rec_len;
 473                 if (left < slack) {
 474                         prev_rec_len += left;
 475                         retval = ext2fs_set_rec_len(fs, prev_rec_len, dirent);
 476                         if (retval)
 477                                 return retval;
 478                         offset += left;
 479                         left = 0;
 480                 }
 481                 prev_hash = ent->hash;
 482         }
 483         if (left)
 484                 retval = ext2fs_set_rec_len(fs, rec_len + left, dirent);
 485
 486         return retval;
 487 }
 488
 489
 490 static struct ext2_dx_root_info *set_root_node(ext2_filsys fs, char *buf,
 491                                     ext2_ino_t ino, ext2_ino_t parent)
 492 {
 493         struct ext2_dir_entry           *dir;
 494         struct ext2_dx_root_info        *root;
 495         struct ext2_dx_countlimit       *limits;
 496         int                             filetype = 0;
 497
 498         if (fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_FILETYPE)
 499                 filetype = EXT2_FT_DIR << 8;
 500
 501         memset(buf, 0, fs->blocksize);
 502         dir = (struct ext2_dir_entry *) buf;
 503         dir->inode = ino;
 504         dir->name[0] = '.';
 505         dir->name_len = 1 | filetype;
 506         dir->rec_len = 12;
 507         dir = (struct ext2_dir_entry *) (buf + 12);
 508         dir->inode = parent;
 509         dir->name[0] = '.';
 510         dir->name[1] = '.';
 511         dir->name_len = 2 | filetype;
 512         dir->rec_len = fs->blocksize - 12;
 513
 514         root = (struct ext2_dx_root_info *) (buf+24);
 515         root->reserved_zero = 0;
 516         root->hash_version = fs->super->s_def_hash_version;
 517         root->info_length = 8;
 518         root->indirect_levels = 0;
 519         root->unused_flags = 0;
 520
 521         limits = (struct ext2_dx_countlimit *) (buf+32);
 522         limits->limit = (fs->blocksize - 32) / sizeof(struct ext2_dx_entry);
 523         limits->count = 0;
 524
 525         return root;
 526 }
 527
 528
 529 static struct ext2_dx_entry *set_int_node(ext2_filsys fs, char *buf)
 530 {
 531         struct ext2_dir_entry           *dir;
 532         struct ext2_dx_countlimit       *limits;
 533
 534         memset(buf, 0, fs->blocksize);
 535         dir = (struct ext2_dir_entry *) buf;
 536         dir->inode = 0;
 537         (void) ext2fs_set_rec_len(fs, fs->blocksize, dir);
 538
 539         limits = (struct ext2_dx_countlimit *) (buf+8);
 540         limits->limit = (fs->blocksize - 8) / sizeof(struct ext2_dx_entry);
 541         limits->count = 0;
 542
 543         return (struct ext2_dx_entry *) limits;
 544 }
 545
 546 /*
 547  * This function takes the leaf nodes which have been written in
 548  * outdir, and populates the root node and any necessary interior nodes.
 549  */
 550 static errcode_t calculate_tree(ext2_filsys fs,
 551                                 struct out_dir *outdir,
 552                                 ext2_ino_t ino,
 553                                 ext2_ino_t parent)
 554 {
 555         struct ext2_dx_root_info        *root_info;
 556         struct ext2_dx_entry            *root, *dx_ent = 0;
 557         struct ext2_dx_countlimit       *root_limit, *limit;
 558         errcode_t                       retval;
 559         char                            * block_start;
 560         int                             i, c1, c2, nblks;
 561         int                             limit_offset, root_offset;
 562
 563         root_info = set_root_node(fs, outdir->buf, ino, parent);
 564         root_offset = limit_offset = ((char *) root_info - outdir->buf) +
 565                 root_info->info_length;
 566         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 567         c1 = root_limit->limit;
 568         nblks = outdir->num;
 569
 570         /* Write out the pointer blocks */
 571         if (nblks-1 <= c1) {
 572                 /* Just write out the root block, and we're done */
 573                 root = (struct ext2_dx_entry *) (outdir->buf + root_offset);
 574                 for (i=1; i < nblks; i++) {
 575                         root->block = ext2fs_cpu_to_le32(i);
 576                         if (i != 1)
 577                                 root->hash =
 578                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 579                         root++;
 580                         c1--;
 581                 }
 582         } else {
 583                 c2 = 0;
 584                 limit = 0;
 585                 root_info->indirect_levels = 1;
 586                 for (i=1; i < nblks; i++) {
 587                         if (c1 == 0)
 588                                 return ENOSPC;
 589                         if (c2 == 0) {
 590                                 if (limit)
 591                                         limit->limit = limit->count =
 592                 ext2fs_cpu_to_le16(limit->limit);
 593                                 root = (struct ext2_dx_entry *)
 594                                         (outdir->buf + root_offset);
 595                                 root->block = ext2fs_cpu_to_le32(outdir->num);
 596                                 if (i != 1)
 597                                         root->hash =
 598                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 599                                 if ((retval =  get_next_block(fs, outdir,
 600                                                               &block_start)))
 601                                         return retval;
 602                                 dx_ent = set_int_node(fs, block_start);
 603                                 limit = (struct ext2_dx_countlimit *) dx_ent;
 604                                 c2 = limit->limit;
 605                                 root_offset += sizeof(struct ext2_dx_entry);
 606                                 c1--;
 607                         }
 608                         dx_ent->block = ext2fs_cpu_to_le32(i);
 609                         if (c2 != limit->limit)
 610                                 dx_ent->hash =
 611                                         ext2fs_cpu_to_le32(outdir->hashes[i]);
 612                         dx_ent++;
 613                         c2--;
 614                 }
 615                 limit->count = ext2fs_cpu_to_le16(limit->limit - c2);
 616                 limit->limit = ext2fs_cpu_to_le16(limit->limit);
 617         }
 618         root_limit = (struct ext2_dx_countlimit *) (outdir->buf + limit_offset);
 619         root_limit->count = ext2fs_cpu_to_le16(root_limit->limit - c1);
 620         root_limit->limit = ext2fs_cpu_to_le16(root_limit->limit);
 621
 622         return 0;
 623 }
 624
 625 struct write_dir_struct {
 626         struct out_dir *outdir;
 627         errcode_t       err;
 628         e2fsck_t        ctx;
 629         blk64_t         cleared;
 630 };
 631
 632 /*
 633  * Helper function which writes out a directory block.
 634  */
 635 static int write_dir_block(ext2_filsys fs,
 636                            blk64_t *block_nr,
 637                            e2_blkcnt_t blockcnt,
 638                            blk64_t ref_block EXT2FS_ATTR((unused)),
 639                            int ref_offset EXT2FS_ATTR((unused)),
 640                            void *priv_data)
 641 {
 642         struct write_dir_struct *wd = (struct write_dir_struct *) priv_data;
 643         blk64_t blk;
 644         char    *dir;
 645
 646         if (*block_nr == 0)
 647                 return 0;
 648         if (blockcnt >= wd->outdir->num) {
 649                 e2fsck_read_bitmaps(wd->ctx);
 650                 blk = *block_nr;
 651                 /*
 652                  * In theory, we only release blocks from the end of the
 653                  * directory file, so it's fine to clobber a whole cluster at
 654                  * once.
 655                  */
 656                 if (blk % EXT2FS_CLUSTER_RATIO(fs) == 0) {
 657                         ext2fs_block_alloc_stats2(fs, blk, -1);
 658                         wd->cleared++;
 659                 }
 660                 *block_nr = 0;
 661                 return BLOCK_CHANGED;
 662         }
 663         if (blockcnt < 0)
 664                 return 0;
 665
 666         dir = wd->outdir->buf + (blockcnt * fs->blocksize);
 667         wd->err = ext2fs_write_dir_block3(fs, *block_nr, dir, 0);
 668         if (wd->err)
 669                 return BLOCK_ABORT;
 670         return 0;
 671 }
 672
 673 static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
 674                                  struct out_dir *outdir,
 675                                  ext2_ino_t ino, int compress)
 676 {
 677         struct write_dir_struct wd;
 678         errcode_t       retval;
 679         struct ext2_inode       inode;
 680
 681         retval = e2fsck_expand_directory(ctx, ino, -1, outdir->num);
 682         if (retval)
 683                 return retval;
 684
 685         wd.outdir = outdir;
 686         wd.err = 0;
 687         wd.ctx = ctx;
 688         wd.cleared = 0;
 689
 690         retval = ext2fs_block_iterate3(fs, ino, 0, 0,
 691                                        write_dir_block, &wd);
 692         if (retval)
 693                 return retval;
 694         if (wd.err)
 695                 return wd.err;
 696
 697         e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
 698         if (compress)
 699                 inode.i_flags &= ~EXT2_INDEX_FL;
 700         else
 701                 inode.i_flags |= EXT2_INDEX_FL;
 702         inode.i_size = outdir->num * fs->blocksize;
 703         ext2fs_iblk_sub_blocks(fs, &inode, wd.cleared);
 704         e2fsck_write_inode(ctx, ino, &inode, "rehash_dir");
 705
 706         return 0;
 707 }
 708
 709 errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino)
 710 {
 711         ext2_filsys             fs = ctx->fs;
 712         errcode_t               retval;
 713         struct ext2_inode       inode;
 714         char                    *dir_buf = 0;
 715         struct fill_dir_struct  fd;
 716         struct out_dir          outdir;
 717
 718         outdir.max = outdir.num = 0;
 719         outdir.buf = 0;
 720         outdir.hashes = 0;
 721         e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
 722
 723         retval = ENOMEM;
 724         fd.harray = 0;
 725         dir_buf = malloc(inode.i_size);
 726         if (!dir_buf)
 727                 goto errout;
 728
 729         fd.max_array = inode.i_size / 32;
 730         fd.num_array = 0;
 731         fd.harray = malloc(fd.max_array * sizeof(struct hash_entry));
 732         if (!fd.harray)
 733                 goto errout;
 734
 735         fd.ctx = ctx;
 736         fd.buf = dir_buf;
 737         fd.inode = &inode;
 738         fd.err = 0;
 739         fd.dir_size = 0;
 740         fd.compress = 0;
 741         if (!(fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_DIR_INDEX) ||
 742             (inode.i_size / fs->blocksize) < 2)
 743                 fd.compress = 1;
 744         fd.parent = 0;
 745
 746 retry_nohash:
 747         /* Read in the entire directory into memory */
 748         retval = ext2fs_block_iterate3(fs, ino, 0, 0,
 749                                        fill_dir_block, &fd);
 750         if (fd.err) {
 751                 retval = fd.err;
 752                 goto errout;
 753         }
 754
 755         /*
 756          * If the entries read are less than a block, then don't index
 757          * the directory
 758          */
 759         if (!fd.compress && (fd.dir_size < (fs->blocksize - 24))) {
 760                 fd.compress = 1;
 761                 fd.dir_size = 0;
 762                 fd.num_array = 0;
 763                 goto retry_nohash;
 764         }
 765
 766 #if 0
 767         printf("%d entries (%d bytes) found in inode %d\n",
 768                fd.num_array, fd.dir_size, ino);
 769 #endif
 770
 771         /* Sort the list */
 772 resort:
 773         if (fd.compress)
 774                 qsort(fd.harray+2, fd.num_array-2, sizeof(struct hash_entry),
 775                       hash_cmp);
 776         else
 777                 qsort(fd.harray, fd.num_array, sizeof(struct hash_entry),
 778                       hash_cmp);
 779
 780         /*
 781          * Look for duplicates
 782          */
 783         if (duplicate_search_and_fix(ctx, fs, ino, &fd))
 784                 goto resort;
 785
 786         if (ctx->options & E2F_OPT_NO) {
 787                 retval = 0;
 788                 goto errout;
 789         }
 790
 791         /* Sort non-hashed directories by inode number */
 792         if (fd.compress)
 793                 qsort(fd.harray+2, fd.num_array-2,
 794                       sizeof(struct hash_entry), ino_cmp);
 795
 796         /*
 797          * Copy the directory entries.  In a htree directory these
 798          * will become the leaf nodes.
 799          */
 800         retval = copy_dir_entries(ctx, &fd, &outdir);
 801         if (retval)
 802                 goto errout;
 803
 804         free(dir_buf); dir_buf = 0;
 805
 806         if (!fd.compress) {
 807                 /* Calculate the interior nodes */
 808                 retval = calculate_tree(fs, &outdir, ino, fd.parent);
 809                 if (retval)
 810                         goto errout;
 811         }
 812
 813         retval = write_directory(ctx, fs, &outdir, ino, fd.compress);
 814         if (retval)
 815                 goto errout;
 816
 817 errout:
 818         free(dir_buf);
 819         free(fd.harray);
 820
 821         free_out_dir(&outdir);
 822         return retval;
 823 }
 824
 825 void e2fsck_rehash_directories(e2fsck_t ctx)
 826 {
 827         struct problem_context  pctx;
 828 #ifdef RESOURCE_TRACK
 829         struct resource_track   rtrack;
 830 #endif
 831         struct dir_info         *dir;
 832         ext2_u32_iterate        iter;
 833         struct dir_info_iter *  dirinfo_iter = 0;
 834         ext2_ino_t              ino;
 835         errcode_t               retval;
 836         int                     cur, max, all_dirs, first = 1;
 837
 838         init_resource_track(&rtrack, ctx->fs->io);
 839         all_dirs = ctx->options & E2F_OPT_COMPRESS_DIRS;
 840
 841         if (!ctx->dirs_to_hash && !all_dirs)
 842                 return;
 843
 844         e2fsck_get_lost_and_found(ctx, 0);
 845
 846         clear_problem_context(&pctx);
 847
 848         cur = 0;
 849         if (all_dirs) {
 850                 dirinfo_iter = e2fsck_dir_info_iter_begin(ctx);
 851                 max = e2fsck_get_num_dirinfo(ctx);
 852         } else {
 853                 retval = ext2fs_u32_list_iterate_begin(ctx->dirs_to_hash,
 854                                                        &iter);
 855                 if (retval) {
 856                         pctx.errcode = retval;
 857                         fix_problem(ctx, PR_3A_OPTIMIZE_ITER, &pctx);
 858                         return;
 859                 }
 860                 max = ext2fs_u32_list_count(ctx->dirs_to_hash);
 861         }
 862         while (1) {
 863                 if (all_dirs) {
 864                         if ((dir = e2fsck_dir_info_iter(ctx,
 865                                                         dirinfo_iter)) == 0)
 866                                 break;
 867                         ino = dir->ino;
 868                 } else {
 869                         if (!ext2fs_u32_list_iterate(iter, &ino))
 870                                 break;
 871                 }
 872                 if (ino == ctx->lost_and_found)
 873                         continue;
 874                 pctx.dir = ino;
 875                 if (first) {
 876                         fix_problem(ctx, PR_3A_PASS_HEADER, &pctx);
 877                         first = 0;
 878                 }
 879 #if 0
 880                 fix_problem(ctx, PR_3A_OPTIMIZE_DIR, &pctx);
 881 #endif
 882                 pctx.errcode = e2fsck_rehash_dir(ctx, ino);
 883                 if (pctx.errcode) {
 884                         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
 885                         fix_problem(ctx, PR_3A_OPTIMIZE_DIR_ERR, &pctx);
 886                 }
 887                 if (ctx->progress && !ctx->progress_fd)
 888                         e2fsck_simple_progress(ctx, "Rebuilding directory",
 889                                100.0 * (float) (++cur) / (float) max, ino);
 890         }
 891         end_problem_latch(ctx, PR_LATCH_OPTIMIZE_DIR);
 892         if (all_dirs)
 893                 e2fsck_dir_info_iter_end(ctx, dirinfo_iter);
 894         else
 895                 ext2fs_u32_list_iterate_end(iter);
 896
 897         if (ctx->dirs_to_hash)
 898                 ext2fs_u32_list_free(ctx->dirs_to_hash);
 899         ctx->dirs_to_hash = 0;
 900
 901         print_resource_track(ctx, "Pass 3A", &rtrack, ctx->fs->io);
 902 }