read-cache.c

   1 /*
   2  * GIT - The information manager from hell
   3  *
   4  * Copyright (C) Linus Torvalds, 2005
   5  */
   6 #include "cache.h"
   7 #include "cache-tree.h"
   8
   9 /* Index extensions.
  10  *
  11  * The first letter should be 'A'..'Z' for extensions that are not
  12  * necessary for a correct operation (i.e. optimization data).
  13  * When new extensions are added that _needs_ to be understood in
  14  * order to correctly interpret the index file, pick character that
  15  * is outside the range, to cause the reader to abort.
  16  */
  17
  18 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  19 #define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  20
  21 struct cache_entry **active_cache;
  22 static time_t index_file_timestamp;
  23 unsigned int active_nr, active_alloc, active_cache_changed;
  24
  25 struct cache_tree *active_cache_tree;
  26
  27 int cache_errno;
  28
  29 static void *cache_mmap;
  30 static size_t cache_mmap_size;
  31
  32 /*
  33  * This only updates the "non-critical" parts of the directory
  34  * cache, ie the parts that aren't tracked by GIT, and only used
  35  * to validate the cache.
  36  */
  37 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
  38 {
  39         ce->ce_ctime.sec = htonl(st->st_ctime);
  40         ce->ce_mtime.sec = htonl(st->st_mtime);
  41 #ifdef USE_NSEC
  42         ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
  43         ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
  44 #endif
  45         ce->ce_dev = htonl(st->st_dev);
  46         ce->ce_ino = htonl(st->st_ino);
  47         ce->ce_uid = htonl(st->st_uid);
  48         ce->ce_gid = htonl(st->st_gid);
  49         ce->ce_size = htonl(st->st_size);
  50
  51         if (assume_unchanged)
  52                 ce->ce_flags |= htons(CE_VALID);
  53 }
  54
  55 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
  56 {
  57         int match = -1;
  58         int fd = open(ce->name, O_RDONLY);
  59
  60         if (fd >= 0) {
  61                 unsigned char sha1[20];
  62                 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
  63                         match = hashcmp(sha1, ce->sha1);
  64                 /* index_fd() closed the file descriptor already */
  65         }
  66         return match;
  67 }
  68
  69 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
  70 {
  71         int match = -1;
  72         char *target;
  73         void *buffer;
  74         unsigned long size;
  75         enum object_type type;
  76         int len;
  77
  78         target = xmalloc(expected_size);
  79         len = readlink(ce->name, target, expected_size);
  80         if (len != expected_size) {
  81                 free(target);
  82                 return -1;
  83         }
  84         buffer = read_sha1_file(ce->sha1, &type, &size);
  85         if (!buffer) {
  86                 free(target);
  87                 return -1;
  88         }
  89         if (size == expected_size)
  90                 match = memcmp(buffer, target, size);
  91         free(buffer);
  92         free(target);
  93         return match;
  94 }
  95
  96 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
  97 {
  98         switch (st->st_mode & S_IFMT) {
  99         case S_IFREG:
 100                 if (ce_compare_data(ce, st))
 101                         return DATA_CHANGED;
 102                 break;
 103         case S_IFLNK:
 104                 if (ce_compare_link(ce, xsize_t(st->st_size)))
 105                         return DATA_CHANGED;
 106                 break;
 107         default:
 108                 return TYPE_CHANGED;
 109         }
 110         return 0;
 111 }
 112
 113 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 114 {
 115         unsigned int changed = 0;
 116
 117         switch (ntohl(ce->ce_mode) & S_IFMT) {
 118         case S_IFREG:
 119                 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 120                 /* We consider only the owner x bit to be relevant for
 121                  * "mode changes"
 122                  */
 123                 if (trust_executable_bit &&
 124                     (0100 & (ntohl(ce->ce_mode) ^ st->st_mode)))
 125                         changed |= MODE_CHANGED;
 126                 break;
 127         case S_IFLNK:
 128                 if (!S_ISLNK(st->st_mode) &&
 129                     (has_symlinks || !S_ISREG(st->st_mode)))
 130                         changed |= TYPE_CHANGED;
 131                 break;
 132         default:
 133                 die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
 134         }
 135         if (ce->ce_mtime.sec != htonl(st->st_mtime))
 136                 changed |= MTIME_CHANGED;
 137         if (ce->ce_ctime.sec != htonl(st->st_ctime))
 138                 changed |= CTIME_CHANGED;
 139
 140 #ifdef USE_NSEC
 141         /*
 142          * nsec seems unreliable - not all filesystems support it, so
 143          * as long as it is in the inode cache you get right nsec
 144          * but after it gets flushed, you get zero nsec.
 145          */
 146         if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
 147                 changed |= MTIME_CHANGED;
 148         if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
 149                 changed |= CTIME_CHANGED;
 150 #endif
 151
 152         if (ce->ce_uid != htonl(st->st_uid) ||
 153             ce->ce_gid != htonl(st->st_gid))
 154                 changed |= OWNER_CHANGED;
 155         if (ce->ce_ino != htonl(st->st_ino))
 156                 changed |= INODE_CHANGED;
 157
 158 #ifdef USE_STDEV
 159         /*
 160          * st_dev breaks on network filesystems where different
 161          * clients will have different views of what "device"
 162          * the filesystem is on
 163          */
 164         if (ce->ce_dev != htonl(st->st_dev))
 165                 changed |= INODE_CHANGED;
 166 #endif
 167
 168         if (ce->ce_size != htonl(st->st_size))
 169                 changed |= DATA_CHANGED;
 170
 171         return changed;
 172 }
 173
 174 int ce_match_stat(struct cache_entry *ce, struct stat *st, int options)
 175 {
 176         unsigned int changed;
 177         int ignore_valid = options & 01;
 178         int assume_racy_is_modified = options & 02;
 179
 180         /*
 181          * If it's marked as always valid in the index, it's
 182          * valid whatever the checked-out copy says.
 183          */
 184         if (!ignore_valid && (ce->ce_flags & htons(CE_VALID)))
 185                 return 0;
 186
 187         changed = ce_match_stat_basic(ce, st);
 188
 189         /*
 190          * Within 1 second of this sequence:
 191          *      echo xyzzy >file && git-update-index --add file
 192          * running this command:
 193          *      echo frotz >file
 194          * would give a falsely clean cache entry.  The mtime and
 195          * length match the cache, and other stat fields do not change.
 196          *
 197          * We could detect this at update-index time (the cache entry
 198          * being registered/updated records the same time as "now")
 199          * and delay the return from git-update-index, but that would
 200          * effectively mean we can make at most one commit per second,
 201          * which is not acceptable.  Instead, we check cache entries
 202          * whose mtime are the same as the index file timestamp more
 203          * carefully than others.
 204          */
 205         if (!changed &&
 206             index_file_timestamp &&
 207             index_file_timestamp <= ntohl(ce->ce_mtime.sec)) {
 208                 if (assume_racy_is_modified)
 209                         changed |= DATA_CHANGED;
 210                 else
 211                         changed |= ce_modified_check_fs(ce, st);
 212         }
 213
 214         return changed;
 215 }
 216
 217 int ce_modified(struct cache_entry *ce, struct stat *st, int really)
 218 {
 219         int changed, changed_fs;
 220         changed = ce_match_stat(ce, st, really);
 221         if (!changed)
 222                 return 0;
 223         /*
 224          * If the mode or type has changed, there's no point in trying
 225          * to refresh the entry - it's not going to match
 226          */
 227         if (changed & (MODE_CHANGED | TYPE_CHANGED))
 228                 return changed;
 229
 230         /* Immediately after read-tree or update-index --cacheinfo,
 231          * the length field is zero.  For other cases the ce_size
 232          * should match the SHA1 recorded in the index entry.
 233          */
 234         if ((changed & DATA_CHANGED) && ce->ce_size != htonl(0))
 235                 return changed;
 236
 237         changed_fs = ce_modified_check_fs(ce, st);
 238         if (changed_fs)
 239                 return changed | changed_fs;
 240         return 0;
 241 }
 242
 243 int base_name_compare(const char *name1, int len1, int mode1,
 244                       const char *name2, int len2, int mode2)
 245 {
 246         unsigned char c1, c2;
 247         int len = len1 < len2 ? len1 : len2;
 248         int cmp;
 249
 250         cmp = memcmp(name1, name2, len);
 251         if (cmp)
 252                 return cmp;
 253         c1 = name1[len];
 254         c2 = name2[len];
 255         if (!c1 && S_ISDIR(mode1))
 256                 c1 = '/';
 257         if (!c2 && S_ISDIR(mode2))
 258                 c2 = '/';
 259         return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 260 }
 261
 262 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 263 {
 264         int len1 = flags1 & CE_NAMEMASK;
 265         int len2 = flags2 & CE_NAMEMASK;
 266         int len = len1 < len2 ? len1 : len2;
 267         int cmp;
 268
 269         cmp = memcmp(name1, name2, len);
 270         if (cmp)
 271                 return cmp;
 272         if (len1 < len2)
 273                 return -1;
 274         if (len1 > len2)
 275                 return 1;
 276
 277         /* Compare stages  */
 278         flags1 &= CE_STAGEMASK;
 279         flags2 &= CE_STAGEMASK;
 280
 281         if (flags1 < flags2)
 282                 return -1;
 283         if (flags1 > flags2)
 284                 return 1;
 285         return 0;
 286 }
 287
 288 int cache_name_pos(const char *name, int namelen)
 289 {
 290         int first, last;
 291
 292         first = 0;
 293         last = active_nr;
 294         while (last > first) {
 295                 int next = (last + first) >> 1;
 296                 struct cache_entry *ce = active_cache[next];
 297                 int cmp = cache_name_compare(name, namelen, ce->name, ntohs(ce->ce_flags));
 298                 if (!cmp)
 299                         return next;
 300                 if (cmp < 0) {
 301                         last = next;
 302                         continue;
 303                 }
 304                 first = next+1;
 305         }
 306         return -first-1;
 307 }
 308
 309 /* Remove entry, return true if there are more entries to go.. */
 310 int remove_cache_entry_at(int pos)
 311 {
 312         active_cache_changed = 1;
 313         active_nr--;
 314         if (pos >= active_nr)
 315                 return 0;
 316         memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos) * sizeof(struct cache_entry *));
 317         return 1;
 318 }
 319
 320 int remove_file_from_cache(const char *path)
 321 {
 322         int pos = cache_name_pos(path, strlen(path));
 323         if (pos < 0)
 324                 pos = -pos-1;
 325         while (pos < active_nr && !strcmp(active_cache[pos]->name, path))
 326                 remove_cache_entry_at(pos);
 327         return 0;
 328 }
 329
 330 int add_file_to_index(const char *path, int verbose)
 331 {
 332         int size, namelen;
 333         struct stat st;
 334         struct cache_entry *ce;
 335
 336         if (lstat(path, &st))
 337                 die("%s: unable to stat (%s)", path, strerror(errno));
 338
 339         if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode))
 340                 die("%s: can only add regular files or symbolic links", path);
 341
 342         namelen = strlen(path);
 343         size = cache_entry_size(namelen);
 344         ce = xcalloc(1, size);
 345         memcpy(ce->name, path, namelen);
 346         ce->ce_flags = htons(namelen);
 347         fill_stat_cache_info(ce, &st);
 348
 349         if (trust_executable_bit && has_symlinks)
 350                 ce->ce_mode = create_ce_mode(st.st_mode);
 351         else {
 352                 /* If there is an existing entry, pick the mode bits and type
 353                  * from it, otherwise assume unexecutable regular file.
 354                  */
 355                 struct cache_entry *ent;
 356                 int pos = cache_name_pos(path, namelen);
 357
 358                 ent = (0 <= pos) ? active_cache[pos] : NULL;
 359                 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
 360         }
 361
 362         if (index_path(ce->sha1, path, &st, 1))
 363                 die("unable to index file %s", path);
 364         if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
 365                 die("unable to add %s to index",path);
 366         if (verbose)
 367                 printf("add '%s'\n", path);
 368         cache_tree_invalidate_path(active_cache_tree, path);
 369         return 0;
 370 }
 371
 372 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 373 {
 374         int len = ce_namelen(a);
 375         return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 376 }
 377
 378 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
 379 {
 380         const char *match, *name;
 381         int len;
 382
 383         if (!pathspec)
 384                 return 1;
 385
 386         len = ce_namelen(ce);
 387         name = ce->name;
 388         while ((match = *pathspec++) != NULL) {
 389                 int matchlen = strlen(match);
 390                 if (matchlen > len)
 391                         continue;
 392                 if (memcmp(name, match, matchlen))
 393                         continue;
 394                 if (matchlen && name[matchlen-1] == '/')
 395                         return 1;
 396                 if (name[matchlen] == '/' || !name[matchlen])
 397                         return 1;
 398                 if (!matchlen)
 399                         return 1;
 400         }
 401         return 0;
 402 }
 403
 404 /*
 405  * We fundamentally don't like some paths: we don't want
 406  * dot or dot-dot anywhere, and for obvious reasons don't
 407  * want to recurse into ".git" either.
 408  *
 409  * Also, we don't want double slashes or slashes at the
 410  * end that can make pathnames ambiguous.
 411  */
 412 static int verify_dotfile(const char *rest)
 413 {
 414         /*
 415          * The first character was '.', but that
 416          * has already been discarded, we now test
 417          * the rest.
 418          */
 419         switch (*rest) {
 420         /* "." is not allowed */
 421         case '\0': case '/':
 422                 return 0;
 423
 424         /*
 425          * ".git" followed by  NUL or slash is bad. This
 426          * shares the path end test with the ".." case.
 427          */
 428         case 'g':
 429                 if (rest[1] != 'i')
 430                         break;
 431                 if (rest[2] != 't')
 432                         break;
 433                 rest += 2;
 434         /* fallthrough */
 435         case '.':
 436                 if (rest[1] == '\0' || rest[1] == '/')
 437                         return 0;
 438         }
 439         return 1;
 440 }
 441
 442 int verify_path(const char *path)
 443 {
 444         char c;
 445
 446         goto inside;
 447         for (;;) {
 448                 if (!c)
 449                         return 1;
 450                 if (c == '/') {
 451 inside:
 452                         c = *path++;
 453                         switch (c) {
 454                         default:
 455                                 continue;
 456                         case '/': case '\0':
 457                                 break;
 458                         case '.':
 459                                 if (verify_dotfile(path))
 460                                         continue;
 461                         }
 462                         return 0;
 463                 }
 464                 c = *path++;
 465         }
 466 }
 467
 468 /*
 469  * Do we have another file that has the beginning components being a
 470  * proper superset of the name we're trying to add?
 471  */
 472 static int has_file_name(const struct cache_entry *ce, int pos, int ok_to_replace)
 473 {
 474         int retval = 0;
 475         int len = ce_namelen(ce);
 476         int stage = ce_stage(ce);
 477         const char *name = ce->name;
 478
 479         while (pos < active_nr) {
 480                 struct cache_entry *p = active_cache[pos++];
 481
 482                 if (len >= ce_namelen(p))
 483                         break;
 484                 if (memcmp(name, p->name, len))
 485                         break;
 486                 if (ce_stage(p) != stage)
 487                         continue;
 488                 if (p->name[len] != '/')
 489                         continue;
 490                 if (!ce_stage(p) && !p->ce_mode)
 491                         continue;
 492                 retval = -1;
 493                 if (!ok_to_replace)
 494                         break;
 495                 remove_cache_entry_at(--pos);
 496         }
 497         return retval;
 498 }
 499
 500 /*
 501  * Do we have another file with a pathname that is a proper
 502  * subset of the name we're trying to add?
 503  */
 504 static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace)
 505 {
 506         int retval = 0;
 507         int stage = ce_stage(ce);
 508         const char *name = ce->name;
 509         const char *slash = name + ce_namelen(ce);
 510
 511         for (;;) {
 512                 int len;
 513
 514                 for (;;) {
 515                         if (*--slash == '/')
 516                                 break;
 517                         if (slash <= ce->name)
 518                                 return retval;
 519                 }
 520                 len = slash - name;
 521
 522                 pos = cache_name_pos(name, ntohs(create_ce_flags(len, stage)));
 523                 if (pos >= 0) {
 524                         /*
 525                          * Found one, but not so fast.  This could
 526                          * be a marker that says "I was here, but
 527                          * I am being removed".  Such an entry is
 528                          * not a part of the resulting tree, and
 529                          * it is Ok to have a directory at the same
 530                          * path.
 531                          */
 532                         if (stage || active_cache[pos]->ce_mode) {
 533                                 retval = -1;
 534                                 if (!ok_to_replace)
 535                                         break;
 536                                 remove_cache_entry_at(pos);
 537                                 continue;
 538                         }
 539                 }
 540                 else
 541                         pos = -pos-1;
 542
 543                 /*
 544                  * Trivial optimization: if we find an entry that
 545                  * already matches the sub-directory, then we know
 546                  * we're ok, and we can exit.
 547                  */
 548                 while (pos < active_nr) {
 549                         struct cache_entry *p = active_cache[pos];
 550                         if ((ce_namelen(p) <= len) ||
 551                             (p->name[len] != '/') ||
 552                             memcmp(p->name, name, len))
 553                                 break; /* not our subdirectory */
 554                         if (ce_stage(p) == stage && (stage || p->ce_mode))
 555                                 /* p is at the same stage as our entry, and
 556                                  * is a subdirectory of what we are looking
 557                                  * at, so we cannot have conflicts at our
 558                                  * level or anything shorter.
 559                                  */
 560                                 return retval;
 561                         pos++;
 562                 }
 563         }
 564         return retval;
 565 }
 566
 567 /* We may be in a situation where we already have path/file and path
 568  * is being added, or we already have path and path/file is being
 569  * added.  Either one would result in a nonsense tree that has path
 570  * twice when git-write-tree tries to write it out.  Prevent it.
 571  *
 572  * If ok-to-replace is specified, we remove the conflicting entries
 573  * from the cache so the caller should recompute the insert position.
 574  * When this happens, we return non-zero.
 575  */
 576 static int check_file_directory_conflict(const struct cache_entry *ce, int pos, int ok_to_replace)
 577 {
 578         int retval;
 579
 580         /*
 581          * When ce is an "I am going away" entry, we allow it to be added
 582          */
 583         if (!ce_stage(ce) && !ce->ce_mode)
 584                 return 0;
 585
 586         /*
 587          * We check if the path is a sub-path of a subsequent pathname
 588          * first, since removing those will not change the position
 589          * in the array.
 590          */
 591         retval = has_file_name(ce, pos, ok_to_replace);
 592
 593         /*
 594          * Then check if the path might have a clashing sub-directory
 595          * before it.
 596          */
 597         return retval + has_dir_name(ce, pos, ok_to_replace);
 598 }
 599
 600 int add_cache_entry(struct cache_entry *ce, int option)
 601 {
 602         int pos;
 603         int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 604         int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 605         int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 606
 607         pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
 608
 609         /* existing match? Just replace it. */
 610         if (pos >= 0) {
 611                 active_cache_changed = 1;
 612                 active_cache[pos] = ce;
 613                 return 0;
 614         }
 615         pos = -pos-1;
 616
 617         /*
 618          * Inserting a merged entry ("stage 0") into the index
 619          * will always replace all non-merged entries..
 620          */
 621         if (pos < active_nr && ce_stage(ce) == 0) {
 622                 while (ce_same_name(active_cache[pos], ce)) {
 623                         ok_to_add = 1;
 624                         if (!remove_cache_entry_at(pos))
 625                                 break;
 626                 }
 627         }
 628
 629         if (!ok_to_add)
 630                 return -1;
 631         if (!verify_path(ce->name))
 632                 return -1;
 633
 634         if (!skip_df_check &&
 635             check_file_directory_conflict(ce, pos, ok_to_replace)) {
 636                 if (!ok_to_replace)
 637                         return error("'%s' appears as both a file and as a directory", ce->name);
 638                 pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
 639                 pos = -pos-1;
 640         }
 641
 642         /* Make sure the array is big enough .. */
 643         if (active_nr == active_alloc) {
 644                 active_alloc = alloc_nr(active_alloc);
 645                 active_cache = xrealloc(active_cache, active_alloc * sizeof(struct cache_entry *));
 646         }
 647
 648         /* Add it in.. */
 649         active_nr++;
 650         if (active_nr > pos)
 651                 memmove(active_cache + pos + 1, active_cache + pos, (active_nr - pos - 1) * sizeof(ce));
 652         active_cache[pos] = ce;
 653         active_cache_changed = 1;
 654         return 0;
 655 }
 656
 657 /*
 658  * "refresh" does not calculate a new sha1 file or bring the
 659  * cache up-to-date for mode/content changes. But what it
 660  * _does_ do is to "re-match" the stat information of a file
 661  * with the cache, so that you can refresh the cache for a
 662  * file that hasn't been changed but where the stat entry is
 663  * out of date.
 664  *
 665  * For example, you'd want to do this after doing a "git-read-tree",
 666  * to link up the stat cache details with the proper files.
 667  */
 668 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
 669 {
 670         struct stat st;
 671         struct cache_entry *updated;
 672         int changed, size;
 673
 674         if (lstat(ce->name, &st) < 0) {
 675                 cache_errno = errno;
 676                 return NULL;
 677         }
 678
 679         changed = ce_match_stat(ce, &st, really);
 680         if (!changed) {
 681                 if (really && assume_unchanged &&
 682                     !(ce->ce_flags & htons(CE_VALID)))
 683                         ; /* mark this one VALID again */
 684                 else
 685                         return ce;
 686         }
 687
 688         if (ce_modified(ce, &st, really)) {
 689                 cache_errno = EINVAL;
 690                 return NULL;
 691         }
 692
 693         size = ce_size(ce);
 694         updated = xmalloc(size);
 695         memcpy(updated, ce, size);
 696         fill_stat_cache_info(updated, &st);
 697
 698         /* In this case, if really is not set, we should leave
 699          * CE_VALID bit alone.  Otherwise, paths marked with
 700          * --no-assume-unchanged (i.e. things to be edited) will
 701          * reacquire CE_VALID bit automatically, which is not
 702          * really what we want.
 703          */
 704         if (!really && assume_unchanged && !(ce->ce_flags & htons(CE_VALID)))
 705                 updated->ce_flags &= ~htons(CE_VALID);
 706
 707         return updated;
 708 }
 709
 710 int refresh_cache(unsigned int flags)
 711 {
 712         int i;
 713         int has_errors = 0;
 714         int really = (flags & REFRESH_REALLY) != 0;
 715         int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
 716         int quiet = (flags & REFRESH_QUIET) != 0;
 717         int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
 718
 719         for (i = 0; i < active_nr; i++) {
 720                 struct cache_entry *ce, *new;
 721                 ce = active_cache[i];
 722                 if (ce_stage(ce)) {
 723                         while ((i < active_nr) &&
 724                                ! strcmp(active_cache[i]->name, ce->name))
 725                                 i++;
 726                         i--;
 727                         if (allow_unmerged)
 728                                 continue;
 729                         printf("%s: needs merge\n", ce->name);
 730                         has_errors = 1;
 731                         continue;
 732                 }
 733
 734                 new = refresh_cache_entry(ce, really);
 735                 if (new == ce)
 736                         continue;
 737                 if (!new) {
 738                         if (not_new && cache_errno == ENOENT)
 739                                 continue;
 740                         if (really && cache_errno == EINVAL) {
 741                                 /* If we are doing --really-refresh that
 742                                  * means the index is not valid anymore.
 743                                  */
 744                                 ce->ce_flags &= ~htons(CE_VALID);
 745                                 active_cache_changed = 1;
 746                         }
 747                         if (quiet)
 748                                 continue;
 749                         printf("%s: needs update\n", ce->name);
 750                         has_errors = 1;
 751                         continue;
 752                 }
 753                 active_cache_changed = 1;
 754                 /* You can NOT just free active_cache[i] here, since it
 755                  * might not be necessarily malloc()ed but can also come
 756                  * from mmap(). */
 757                 active_cache[i] = new;
 758         }
 759         return has_errors;
 760 }
 761
 762 static int verify_hdr(struct cache_header *hdr, unsigned long size)
 763 {
 764         SHA_CTX c;
 765         unsigned char sha1[20];
 766
 767         if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
 768                 return error("bad signature");
 769         if (hdr->hdr_version != htonl(2))
 770                 return error("bad index version");
 771         SHA1_Init(&c);
 772         SHA1_Update(&c, hdr, size - 20);
 773         SHA1_Final(sha1, &c);
 774         if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
 775                 return error("bad index file sha1 signature");
 776         return 0;
 777 }
 778
 779 static int read_index_extension(const char *ext, void *data, unsigned long sz)
 780 {
 781         switch (CACHE_EXT(ext)) {
 782         case CACHE_EXT_TREE:
 783                 active_cache_tree = cache_tree_read(data, sz);
 784                 break;
 785         default:
 786                 if (*ext < 'A' || 'Z' < *ext)
 787                         return error("index uses %.4s extension, which we do not understand",
 788                                      ext);
 789                 fprintf(stderr, "ignoring %.4s extension\n", ext);
 790                 break;
 791         }
 792         return 0;
 793 }
 794
 795 int read_cache(void)
 796 {
 797         return read_cache_from(get_index_file());
 798 }
 799
 800 /* remember to discard_cache() before reading a different cache! */
 801 int read_cache_from(const char *path)
 802 {
 803         int fd, i;
 804         struct stat st;
 805         unsigned long offset;
 806         struct cache_header *hdr;
 807
 808         errno = EBUSY;
 809         if (cache_mmap)
 810                 return active_nr;
 811
 812         errno = ENOENT;
 813         index_file_timestamp = 0;
 814         fd = open(path, O_RDONLY);
 815         if (fd < 0) {
 816                 if (errno == ENOENT)
 817                         return 0;
 818                 die("index file open failed (%s)", strerror(errno));
 819         }
 820
 821         if (!fstat(fd, &st)) {
 822                 cache_mmap_size = xsize_t(st.st_size);
 823                 errno = EINVAL;
 824                 if (cache_mmap_size >= sizeof(struct cache_header) + 20)
 825                         cache_mmap = xmmap(NULL, cache_mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
 826                 else
 827                         die("index file smaller than expected");
 828         } else
 829                 die("cannot stat the open index (%s)", strerror(errno));
 830         close(fd);
 831
 832         hdr = cache_mmap;
 833         if (verify_hdr(hdr, cache_mmap_size) < 0)
 834                 goto unmap;
 835
 836         active_nr = ntohl(hdr->hdr_entries);
 837         active_alloc = alloc_nr(active_nr);
 838         active_cache = xcalloc(active_alloc, sizeof(struct cache_entry *));
 839
 840         offset = sizeof(*hdr);
 841         for (i = 0; i < active_nr; i++) {
 842                 struct cache_entry *ce = (struct cache_entry *) ((char *) cache_mmap + offset);
 843                 offset = offset + ce_size(ce);
 844                 active_cache[i] = ce;
 845         }
 846         index_file_timestamp = st.st_mtime;
 847         while (offset <= cache_mmap_size - 20 - 8) {
 848                 /* After an array of active_nr index entries,
 849                  * there can be arbitrary number of extended
 850                  * sections, each of which is prefixed with
 851                  * extension name (4-byte) and section length
 852                  * in 4-byte network byte order.
 853                  */
 854                 unsigned long extsize;
 855                 memcpy(&extsize, (char *) cache_mmap + offset + 4, 4);
 856                 extsize = ntohl(extsize);
 857                 if (read_index_extension(((const char *) cache_mmap) + offset,
 858                                          (char *) cache_mmap + offset + 8,
 859                                          extsize) < 0)
 860                         goto unmap;
 861                 offset += 8;
 862                 offset += extsize;
 863         }
 864         return active_nr;
 865
 866 unmap:
 867         munmap(cache_mmap, cache_mmap_size);
 868         errno = EINVAL;
 869         die("index file corrupt");
 870 }
 871
 872 int discard_cache(void)
 873 {
 874         int ret;
 875
 876         active_nr = active_cache_changed = 0;
 877         index_file_timestamp = 0;
 878         cache_tree_free(&active_cache_tree);
 879         if (cache_mmap == NULL)
 880                 return 0;
 881         ret = munmap(cache_mmap, cache_mmap_size);
 882         cache_mmap = NULL;
 883         cache_mmap_size = 0;
 884
 885         /* no need to throw away allocated active_cache */
 886         return ret;
 887 }
 888
 889 #define WRITE_BUFFER_SIZE 8192
 890 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
 891 static unsigned long write_buffer_len;
 892
 893 static int ce_write_flush(SHA_CTX *context, int fd)
 894 {
 895         unsigned int buffered = write_buffer_len;
 896         if (buffered) {
 897                 SHA1_Update(context, write_buffer, buffered);
 898                 if (write_in_full(fd, write_buffer, buffered) != buffered)
 899                         return -1;
 900                 write_buffer_len = 0;
 901         }
 902         return 0;
 903 }
 904
 905 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
 906 {
 907         while (len) {
 908                 unsigned int buffered = write_buffer_len;
 909                 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
 910                 if (partial > len)
 911                         partial = len;
 912                 memcpy(write_buffer + buffered, data, partial);
 913                 buffered += partial;
 914                 if (buffered == WRITE_BUFFER_SIZE) {
 915                         write_buffer_len = buffered;
 916                         if (ce_write_flush(context, fd))
 917                                 return -1;
 918                         buffered = 0;
 919                 }
 920                 write_buffer_len = buffered;
 921                 len -= partial;
 922                 data = (char *) data + partial;
 923         }
 924         return 0;
 925 }
 926
 927 static int write_index_ext_header(SHA_CTX *context, int fd,
 928                                   unsigned int ext, unsigned int sz)
 929 {
 930         ext = htonl(ext);
 931         sz = htonl(sz);
 932         return ((ce_write(context, fd, &ext, 4) < 0) ||
 933                 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
 934 }
 935
 936 static int ce_flush(SHA_CTX *context, int fd)
 937 {
 938         unsigned int left = write_buffer_len;
 939
 940         if (left) {
 941                 write_buffer_len = 0;
 942                 SHA1_Update(context, write_buffer, left);
 943         }
 944
 945         /* Flush first if not enough space for SHA1 signature */
 946         if (left + 20 > WRITE_BUFFER_SIZE) {
 947                 if (write_in_full(fd, write_buffer, left) != left)
 948                         return -1;
 949                 left = 0;
 950         }
 951
 952         /* Append the SHA1 signature at the end */
 953         SHA1_Final(write_buffer + left, context);
 954         left += 20;
 955         return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
 956 }
 957
 958 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
 959 {
 960         /*
 961          * The only thing we care about in this function is to smudge the
 962          * falsely clean entry due to touch-update-touch race, so we leave
 963          * everything else as they are.  We are called for entries whose
 964          * ce_mtime match the index file mtime.
 965          */
 966         struct stat st;
 967
 968         if (lstat(ce->name, &st) < 0)
 969                 return;
 970         if (ce_match_stat_basic(ce, &st))
 971                 return;
 972         if (ce_modified_check_fs(ce, &st)) {
 973                 /* This is "racily clean"; smudge it.  Note that this
 974                  * is a tricky code.  At first glance, it may appear
 975                  * that it can break with this sequence:
 976                  *
 977                  * $ echo xyzzy >frotz
 978                  * $ git-update-index --add frotz
 979                  * $ : >frotz
 980                  * $ sleep 3
 981                  * $ echo filfre >nitfol
 982                  * $ git-update-index --add nitfol
 983                  *
 984                  * but it does not.  When the second update-index runs,
 985                  * it notices that the entry "frotz" has the same timestamp
 986                  * as index, and if we were to smudge it by resetting its
 987                  * size to zero here, then the object name recorded
 988                  * in index is the 6-byte file but the cached stat information
 989                  * becomes zero --- which would then match what we would
 990                  * obtain from the filesystem next time we stat("frotz").
 991                  *
 992                  * However, the second update-index, before calling
 993                  * this function, notices that the cached size is 6
 994                  * bytes and what is on the filesystem is an empty
 995                  * file, and never calls us, so the cached size information
 996                  * for "frotz" stays 6 which does not match the filesystem.
 997                  */
 998                 ce->ce_size = htonl(0);
 999         }
1000 }
1001
1002 int write_cache(int newfd, struct cache_entry **cache, int entries)
1003 {
1004         SHA_CTX c;
1005         struct cache_header hdr;
1006         int i, removed;
1007
1008         for (i = removed = 0; i < entries; i++)
1009                 if (!cache[i]->ce_mode)
1010                         removed++;
1011
1012         hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1013         hdr.hdr_version = htonl(2);
1014         hdr.hdr_entries = htonl(entries - removed);
1015
1016         SHA1_Init(&c);
1017         if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1018                 return -1;
1019
1020         for (i = 0; i < entries; i++) {
1021                 struct cache_entry *ce = cache[i];
1022                 if (!ce->ce_mode)
1023                         continue;
1024                 if (index_file_timestamp &&
1025                     index_file_timestamp <= ntohl(ce->ce_mtime.sec))
1026                         ce_smudge_racily_clean_entry(ce);
1027                 if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
1028                         return -1;
1029         }
1030
1031         /* Write extension data here */
1032         if (active_cache_tree) {
1033                 unsigned long sz;
1034                 void *data = cache_tree_write(active_cache_tree, &sz);
1035                 if (data &&
1036                     !write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sz) &&
1037                     !ce_write(&c, newfd, data, sz))
1038                         free(data);
1039                 else {
1040                         free(data);
1041                         return -1;
1042                 }
1043         }
1044         return ce_flush(&c, newfd);
1045 }