read-cache.con commit Teach git list-objects logic not to follow gitlinks (ea376fa)
   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#include "refs.h"
   9
  10/* Index extensions.
  11 *
  12 * The first letter should be 'A'..'Z' for extensions that are not
  13 * necessary for a correct operation (i.e. optimization data).
  14 * When new extensions are added that _needs_ to be understood in
  15 * order to correctly interpret the index file, pick character that
  16 * is outside the range, to cause the reader to abort.
  17 */
  18
  19#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  20#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  21
  22struct cache_entry **active_cache;
  23static time_t index_file_timestamp;
  24unsigned int active_nr, active_alloc, active_cache_changed;
  25
  26struct cache_tree *active_cache_tree;
  27
  28static void *cache_mmap;
  29static size_t cache_mmap_size;
  30
  31/*
  32 * This only updates the "non-critical" parts of the directory
  33 * cache, ie the parts that aren't tracked by GIT, and only used
  34 * to validate the cache.
  35 */
  36void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
  37{
  38        ce->ce_ctime.sec = htonl(st->st_ctime);
  39        ce->ce_mtime.sec = htonl(st->st_mtime);
  40#ifdef USE_NSEC
  41        ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
  42        ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
  43#endif
  44        ce->ce_dev = htonl(st->st_dev);
  45        ce->ce_ino = htonl(st->st_ino);
  46        ce->ce_uid = htonl(st->st_uid);
  47        ce->ce_gid = htonl(st->st_gid);
  48        ce->ce_size = htonl(st->st_size);
  49
  50        if (assume_unchanged)
  51                ce->ce_flags |= htons(CE_VALID);
  52}
  53
  54static int ce_compare_data(struct cache_entry *ce, struct stat *st)
  55{
  56        int match = -1;
  57        int fd = open(ce->name, O_RDONLY);
  58
  59        if (fd >= 0) {
  60                unsigned char sha1[20];
  61                if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
  62                        match = hashcmp(sha1, ce->sha1);
  63                /* index_fd() closed the file descriptor already */
  64        }
  65        return match;
  66}
  67
  68static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
  69{
  70        int match = -1;
  71        char *target;
  72        void *buffer;
  73        unsigned long size;
  74        enum object_type type;
  75        int len;
  76
  77        target = xmalloc(expected_size);
  78        len = readlink(ce->name, target, expected_size);
  79        if (len != expected_size) {
  80                free(target);
  81                return -1;
  82        }
  83        buffer = read_sha1_file(ce->sha1, &type, &size);
  84        if (!buffer) {
  85                free(target);
  86                return -1;
  87        }
  88        if (size == expected_size)
  89                match = memcmp(buffer, target, size);
  90        free(buffer);
  91        free(target);
  92        return match;
  93}
  94
  95static int ce_compare_gitlink(struct cache_entry *ce)
  96{
  97        unsigned char sha1[20];
  98
  99        /*
 100         * We don't actually require that the .git directory
 101         * under DIRLNK directory be a valid git directory. It
 102         * might even be missing (in case nobody populated that
 103         * sub-project).
 104         *
 105         * If so, we consider it always to match.
 106         */
 107        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 108                return 0;
 109        return hashcmp(sha1, ce->sha1);
 110}
 111
 112static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
 113{
 114        switch (st->st_mode & S_IFMT) {
 115        case S_IFREG:
 116                if (ce_compare_data(ce, st))
 117                        return DATA_CHANGED;
 118                break;
 119        case S_IFLNK:
 120                if (ce_compare_link(ce, xsize_t(st->st_size)))
 121                        return DATA_CHANGED;
 122                break;
 123        case S_IFDIRLNK:
 124                /* No need to do anything, we did the exact compare in "match_stat_basic" */
 125                break;
 126        default:
 127                return TYPE_CHANGED;
 128        }
 129        return 0;
 130}
 131
 132static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 133{
 134        unsigned int changed = 0;
 135
 136        switch (ntohl(ce->ce_mode) & S_IFMT) {
 137        case S_IFREG:
 138                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 139                /* We consider only the owner x bit to be relevant for
 140                 * "mode changes"
 141                 */
 142                if (trust_executable_bit &&
 143                    (0100 & (ntohl(ce->ce_mode) ^ st->st_mode)))
 144                        changed |= MODE_CHANGED;
 145                break;
 146        case S_IFLNK:
 147                if (!S_ISLNK(st->st_mode) &&
 148                    (has_symlinks || !S_ISREG(st->st_mode)))
 149                        changed |= TYPE_CHANGED;
 150                break;
 151        case S_IFDIRLNK:
 152                if (!S_ISDIR(st->st_mode))
 153                        changed |= TYPE_CHANGED;
 154                else if (ce_compare_gitlink(ce))
 155                        changed |= DATA_CHANGED;
 156                break;
 157        default:
 158                die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
 159        }
 160        if (ce->ce_mtime.sec != htonl(st->st_mtime))
 161                changed |= MTIME_CHANGED;
 162        if (ce->ce_ctime.sec != htonl(st->st_ctime))
 163                changed |= CTIME_CHANGED;
 164
 165#ifdef USE_NSEC
 166        /*
 167         * nsec seems unreliable - not all filesystems support it, so
 168         * as long as it is in the inode cache you get right nsec
 169         * but after it gets flushed, you get zero nsec.
 170         */
 171        if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
 172                changed |= MTIME_CHANGED;
 173        if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
 174                changed |= CTIME_CHANGED;
 175#endif  
 176
 177        if (ce->ce_uid != htonl(st->st_uid) ||
 178            ce->ce_gid != htonl(st->st_gid))
 179                changed |= OWNER_CHANGED;
 180        if (ce->ce_ino != htonl(st->st_ino))
 181                changed |= INODE_CHANGED;
 182
 183#ifdef USE_STDEV
 184        /*
 185         * st_dev breaks on network filesystems where different
 186         * clients will have different views of what "device"
 187         * the filesystem is on
 188         */
 189        if (ce->ce_dev != htonl(st->st_dev))
 190                changed |= INODE_CHANGED;
 191#endif
 192
 193        if (ce->ce_size != htonl(st->st_size))
 194                changed |= DATA_CHANGED;
 195
 196        return changed;
 197}
 198
 199int ce_match_stat(struct cache_entry *ce, struct stat *st, int options)
 200{
 201        unsigned int changed;
 202        int ignore_valid = options & 01;
 203        int assume_racy_is_modified = options & 02;
 204
 205        /*
 206         * If it's marked as always valid in the index, it's
 207         * valid whatever the checked-out copy says.
 208         */
 209        if (!ignore_valid && (ce->ce_flags & htons(CE_VALID)))
 210                return 0;
 211
 212        changed = ce_match_stat_basic(ce, st);
 213
 214        /*
 215         * Within 1 second of this sequence:
 216         *      echo xyzzy >file && git-update-index --add file
 217         * running this command:
 218         *      echo frotz >file
 219         * would give a falsely clean cache entry.  The mtime and
 220         * length match the cache, and other stat fields do not change.
 221         *
 222         * We could detect this at update-index time (the cache entry
 223         * being registered/updated records the same time as "now")
 224         * and delay the return from git-update-index, but that would
 225         * effectively mean we can make at most one commit per second,
 226         * which is not acceptable.  Instead, we check cache entries
 227         * whose mtime are the same as the index file timestamp more
 228         * carefully than others.
 229         */
 230        if (!changed &&
 231            index_file_timestamp &&
 232            index_file_timestamp <= ntohl(ce->ce_mtime.sec)) {
 233                if (assume_racy_is_modified)
 234                        changed |= DATA_CHANGED;
 235                else
 236                        changed |= ce_modified_check_fs(ce, st);
 237        }
 238
 239        return changed;
 240}
 241
 242int ce_modified(struct cache_entry *ce, struct stat *st, int really)
 243{
 244        int changed, changed_fs;
 245        changed = ce_match_stat(ce, st, really);
 246        if (!changed)
 247                return 0;
 248        /*
 249         * If the mode or type has changed, there's no point in trying
 250         * to refresh the entry - it's not going to match
 251         */
 252        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 253                return changed;
 254
 255        /* Immediately after read-tree or update-index --cacheinfo,
 256         * the length field is zero.  For other cases the ce_size
 257         * should match the SHA1 recorded in the index entry.
 258         */
 259        if ((changed & DATA_CHANGED) && ce->ce_size != htonl(0))
 260                return changed;
 261
 262        changed_fs = ce_modified_check_fs(ce, st);
 263        if (changed_fs)
 264                return changed | changed_fs;
 265        return 0;
 266}
 267
 268int base_name_compare(const char *name1, int len1, int mode1,
 269                      const char *name2, int len2, int mode2)
 270{
 271        unsigned char c1, c2;
 272        int len = len1 < len2 ? len1 : len2;
 273        int cmp;
 274
 275        cmp = memcmp(name1, name2, len);
 276        if (cmp)
 277                return cmp;
 278        c1 = name1[len];
 279        c2 = name2[len];
 280        if (!c1 && S_ISDIR(mode1))
 281                c1 = '/';
 282        if (!c2 && S_ISDIR(mode2))
 283                c2 = '/';
 284        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 285}
 286
 287int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 288{
 289        int len1 = flags1 & CE_NAMEMASK;
 290        int len2 = flags2 & CE_NAMEMASK;
 291        int len = len1 < len2 ? len1 : len2;
 292        int cmp;
 293
 294        cmp = memcmp(name1, name2, len);
 295        if (cmp)
 296                return cmp;
 297        if (len1 < len2)
 298                return -1;
 299        if (len1 > len2)
 300                return 1;
 301
 302        /* Compare stages  */
 303        flags1 &= CE_STAGEMASK;
 304        flags2 &= CE_STAGEMASK;
 305
 306        if (flags1 < flags2)
 307                return -1;
 308        if (flags1 > flags2)
 309                return 1;
 310        return 0;
 311}
 312
 313int cache_name_pos(const char *name, int namelen)
 314{
 315        int first, last;
 316
 317        first = 0;
 318        last = active_nr;
 319        while (last > first) {
 320                int next = (last + first) >> 1;
 321                struct cache_entry *ce = active_cache[next];
 322                int cmp = cache_name_compare(name, namelen, ce->name, ntohs(ce->ce_flags));
 323                if (!cmp)
 324                        return next;
 325                if (cmp < 0) {
 326                        last = next;
 327                        continue;
 328                }
 329                first = next+1;
 330        }
 331        return -first-1;
 332}
 333
 334/* Remove entry, return true if there are more entries to go.. */
 335int remove_cache_entry_at(int pos)
 336{
 337        active_cache_changed = 1;
 338        active_nr--;
 339        if (pos >= active_nr)
 340                return 0;
 341        memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos) * sizeof(struct cache_entry *));
 342        return 1;
 343}
 344
 345int remove_file_from_cache(const char *path)
 346{
 347        int pos = cache_name_pos(path, strlen(path));
 348        if (pos < 0)
 349                pos = -pos-1;
 350        while (pos < active_nr && !strcmp(active_cache[pos]->name, path))
 351                remove_cache_entry_at(pos);
 352        return 0;
 353}
 354
 355int add_file_to_cache(const char *path, int verbose)
 356{
 357        int size, namelen;
 358        struct stat st;
 359        struct cache_entry *ce;
 360
 361        if (lstat(path, &st))
 362                die("%s: unable to stat (%s)", path, strerror(errno));
 363
 364        if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
 365                die("%s: can only add regular files, symbolic links or git-directories", path);
 366
 367        namelen = strlen(path);
 368        if (S_ISDIR(st.st_mode)) {
 369                while (namelen && path[namelen-1] == '/')
 370                        namelen--;
 371        }
 372        size = cache_entry_size(namelen);
 373        ce = xcalloc(1, size);
 374        memcpy(ce->name, path, namelen);
 375        ce->ce_flags = htons(namelen);
 376        fill_stat_cache_info(ce, &st);
 377
 378        if (trust_executable_bit && has_symlinks)
 379                ce->ce_mode = create_ce_mode(st.st_mode);
 380        else {
 381                /* If there is an existing entry, pick the mode bits and type
 382                 * from it, otherwise assume unexecutable regular file.
 383                 */
 384                struct cache_entry *ent;
 385                int pos = cache_name_pos(path, namelen);
 386
 387                ent = (0 <= pos) ? active_cache[pos] : NULL;
 388                ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
 389        }
 390
 391        if (index_path(ce->sha1, path, &st, 1))
 392                die("unable to index file %s", path);
 393        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
 394                die("unable to add %s to index",path);
 395        if (verbose)
 396                printf("add '%s'\n", path);
 397        cache_tree_invalidate_path(active_cache_tree, path);
 398        return 0;
 399}
 400
 401int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 402{
 403        int len = ce_namelen(a);
 404        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 405}
 406
 407int ce_path_match(const struct cache_entry *ce, const char **pathspec)
 408{
 409        const char *match, *name;
 410        int len;
 411
 412        if (!pathspec)
 413                return 1;
 414
 415        len = ce_namelen(ce);
 416        name = ce->name;
 417        while ((match = *pathspec++) != NULL) {
 418                int matchlen = strlen(match);
 419                if (matchlen > len)
 420                        continue;
 421                if (memcmp(name, match, matchlen))
 422                        continue;
 423                if (matchlen && name[matchlen-1] == '/')
 424                        return 1;
 425                if (name[matchlen] == '/' || !name[matchlen])
 426                        return 1;
 427                if (!matchlen)
 428                        return 1;
 429        }
 430        return 0;
 431}
 432
 433/*
 434 * We fundamentally don't like some paths: we don't want
 435 * dot or dot-dot anywhere, and for obvious reasons don't
 436 * want to recurse into ".git" either.
 437 *
 438 * Also, we don't want double slashes or slashes at the
 439 * end that can make pathnames ambiguous.
 440 */
 441static int verify_dotfile(const char *rest)
 442{
 443        /*
 444         * The first character was '.', but that
 445         * has already been discarded, we now test
 446         * the rest.
 447         */
 448        switch (*rest) {
 449        /* "." is not allowed */
 450        case '\0': case '/':
 451                return 0;
 452
 453        /*
 454         * ".git" followed by  NUL or slash is bad. This
 455         * shares the path end test with the ".." case.
 456         */
 457        case 'g':
 458                if (rest[1] != 'i')
 459                        break;
 460                if (rest[2] != 't')
 461                        break;
 462                rest += 2;
 463        /* fallthrough */
 464        case '.':
 465                if (rest[1] == '\0' || rest[1] == '/')
 466                        return 0;
 467        }
 468        return 1;
 469}
 470
 471int verify_path(const char *path)
 472{
 473        char c;
 474
 475        goto inside;
 476        for (;;) {
 477                if (!c)
 478                        return 1;
 479                if (c == '/') {
 480inside:
 481                        c = *path++;
 482                        switch (c) {
 483                        default:
 484                                continue;
 485                        case '/': case '\0':
 486                                break;
 487                        case '.':
 488                                if (verify_dotfile(path))
 489                                        continue;
 490                        }
 491                        return 0;
 492                }
 493                c = *path++;
 494        }
 495}
 496
 497/*
 498 * Do we have another file that has the beginning components being a
 499 * proper superset of the name we're trying to add?
 500 */
 501static int has_file_name(const struct cache_entry *ce, int pos, int ok_to_replace)
 502{
 503        int retval = 0;
 504        int len = ce_namelen(ce);
 505        int stage = ce_stage(ce);
 506        const char *name = ce->name;
 507
 508        while (pos < active_nr) {
 509                struct cache_entry *p = active_cache[pos++];
 510
 511                if (len >= ce_namelen(p))
 512                        break;
 513                if (memcmp(name, p->name, len))
 514                        break;
 515                if (ce_stage(p) != stage)
 516                        continue;
 517                if (p->name[len] != '/')
 518                        continue;
 519                if (!ce_stage(p) && !p->ce_mode)
 520                        continue;
 521                retval = -1;
 522                if (!ok_to_replace)
 523                        break;
 524                remove_cache_entry_at(--pos);
 525        }
 526        return retval;
 527}
 528
 529/*
 530 * Do we have another file with a pathname that is a proper
 531 * subset of the name we're trying to add?
 532 */
 533static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace)
 534{
 535        int retval = 0;
 536        int stage = ce_stage(ce);
 537        const char *name = ce->name;
 538        const char *slash = name + ce_namelen(ce);
 539
 540        for (;;) {
 541                int len;
 542
 543                for (;;) {
 544                        if (*--slash == '/')
 545                                break;
 546                        if (slash <= ce->name)
 547                                return retval;
 548                }
 549                len = slash - name;
 550
 551                pos = cache_name_pos(name, ntohs(create_ce_flags(len, stage)));
 552                if (pos >= 0) {
 553                        /*
 554                         * Found one, but not so fast.  This could
 555                         * be a marker that says "I was here, but
 556                         * I am being removed".  Such an entry is
 557                         * not a part of the resulting tree, and
 558                         * it is Ok to have a directory at the same
 559                         * path.
 560                         */
 561                        if (stage || active_cache[pos]->ce_mode) {
 562                                retval = -1;
 563                                if (!ok_to_replace)
 564                                        break;
 565                                remove_cache_entry_at(pos);
 566                                continue;
 567                        }
 568                }
 569                else
 570                        pos = -pos-1;
 571
 572                /*
 573                 * Trivial optimization: if we find an entry that
 574                 * already matches the sub-directory, then we know
 575                 * we're ok, and we can exit.
 576                 */
 577                while (pos < active_nr) {
 578                        struct cache_entry *p = active_cache[pos];
 579                        if ((ce_namelen(p) <= len) ||
 580                            (p->name[len] != '/') ||
 581                            memcmp(p->name, name, len))
 582                                break; /* not our subdirectory */
 583                        if (ce_stage(p) == stage && (stage || p->ce_mode))
 584                                /* p is at the same stage as our entry, and
 585                                 * is a subdirectory of what we are looking
 586                                 * at, so we cannot have conflicts at our
 587                                 * level or anything shorter.
 588                                 */
 589                                return retval;
 590                        pos++;
 591                }
 592        }
 593        return retval;
 594}
 595
 596/* We may be in a situation where we already have path/file and path
 597 * is being added, or we already have path and path/file is being
 598 * added.  Either one would result in a nonsense tree that has path
 599 * twice when git-write-tree tries to write it out.  Prevent it.
 600 * 
 601 * If ok-to-replace is specified, we remove the conflicting entries
 602 * from the cache so the caller should recompute the insert position.
 603 * When this happens, we return non-zero.
 604 */
 605static int check_file_directory_conflict(const struct cache_entry *ce, int pos, int ok_to_replace)
 606{
 607        int retval;
 608
 609        /*
 610         * When ce is an "I am going away" entry, we allow it to be added
 611         */
 612        if (!ce_stage(ce) && !ce->ce_mode)
 613                return 0;
 614
 615        /*
 616         * We check if the path is a sub-path of a subsequent pathname
 617         * first, since removing those will not change the position
 618         * in the array.
 619         */
 620        retval = has_file_name(ce, pos, ok_to_replace);
 621
 622        /*
 623         * Then check if the path might have a clashing sub-directory
 624         * before it.
 625         */
 626        return retval + has_dir_name(ce, pos, ok_to_replace);
 627}
 628
 629int add_cache_entry(struct cache_entry *ce, int option)
 630{
 631        int pos;
 632        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 633        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 634        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 635
 636        pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
 637
 638        /* existing match? Just replace it. */
 639        if (pos >= 0) {
 640                active_cache_changed = 1;
 641                active_cache[pos] = ce;
 642                return 0;
 643        }
 644        pos = -pos-1;
 645
 646        /*
 647         * Inserting a merged entry ("stage 0") into the index
 648         * will always replace all non-merged entries..
 649         */
 650        if (pos < active_nr && ce_stage(ce) == 0) {
 651                while (ce_same_name(active_cache[pos], ce)) {
 652                        ok_to_add = 1;
 653                        if (!remove_cache_entry_at(pos))
 654                                break;
 655                }
 656        }
 657
 658        if (!ok_to_add)
 659                return -1;
 660        if (!verify_path(ce->name))
 661                return -1;
 662
 663        if (!skip_df_check &&
 664            check_file_directory_conflict(ce, pos, ok_to_replace)) {
 665                if (!ok_to_replace)
 666                        return error("'%s' appears as both a file and as a directory", ce->name);
 667                pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
 668                pos = -pos-1;
 669        }
 670
 671        /* Make sure the array is big enough .. */
 672        if (active_nr == active_alloc) {
 673                active_alloc = alloc_nr(active_alloc);
 674                active_cache = xrealloc(active_cache, active_alloc * sizeof(struct cache_entry *));
 675        }
 676
 677        /* Add it in.. */
 678        active_nr++;
 679        if (active_nr > pos)
 680                memmove(active_cache + pos + 1, active_cache + pos, (active_nr - pos - 1) * sizeof(ce));
 681        active_cache[pos] = ce;
 682        active_cache_changed = 1;
 683        return 0;
 684}
 685
 686/*
 687 * "refresh" does not calculate a new sha1 file or bring the
 688 * cache up-to-date for mode/content changes. But what it
 689 * _does_ do is to "re-match" the stat information of a file
 690 * with the cache, so that you can refresh the cache for a
 691 * file that hasn't been changed but where the stat entry is
 692 * out of date.
 693 *
 694 * For example, you'd want to do this after doing a "git-read-tree",
 695 * to link up the stat cache details with the proper files.
 696 */
 697static struct cache_entry *refresh_cache_ent(struct cache_entry *ce, int really, int *err)
 698{
 699        struct stat st;
 700        struct cache_entry *updated;
 701        int changed, size;
 702
 703        if (lstat(ce->name, &st) < 0) {
 704                if (err)
 705                        *err = errno;
 706                return NULL;
 707        }
 708
 709        changed = ce_match_stat(ce, &st, really);
 710        if (!changed) {
 711                if (really && assume_unchanged &&
 712                    !(ce->ce_flags & htons(CE_VALID)))
 713                        ; /* mark this one VALID again */
 714                else
 715                        return ce;
 716        }
 717
 718        if (ce_modified(ce, &st, really)) {
 719                if (err)
 720                        *err = EINVAL;
 721                return NULL;
 722        }
 723
 724        size = ce_size(ce);
 725        updated = xmalloc(size);
 726        memcpy(updated, ce, size);
 727        fill_stat_cache_info(updated, &st);
 728
 729        /* In this case, if really is not set, we should leave
 730         * CE_VALID bit alone.  Otherwise, paths marked with
 731         * --no-assume-unchanged (i.e. things to be edited) will
 732         * reacquire CE_VALID bit automatically, which is not
 733         * really what we want.
 734         */
 735        if (!really && assume_unchanged && !(ce->ce_flags & htons(CE_VALID)))
 736                updated->ce_flags &= ~htons(CE_VALID);
 737
 738        return updated;
 739}
 740
 741int refresh_cache(unsigned int flags)
 742{
 743        int i;
 744        int has_errors = 0;
 745        int really = (flags & REFRESH_REALLY) != 0;
 746        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
 747        int quiet = (flags & REFRESH_QUIET) != 0;
 748        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
 749
 750        for (i = 0; i < active_nr; i++) {
 751                struct cache_entry *ce, *new;
 752                int cache_errno = 0;
 753
 754                ce = active_cache[i];
 755                if (ce_stage(ce)) {
 756                        while ((i < active_nr) &&
 757                               ! strcmp(active_cache[i]->name, ce->name))
 758                                i++;
 759                        i--;
 760                        if (allow_unmerged)
 761                                continue;
 762                        printf("%s: needs merge\n", ce->name);
 763                        has_errors = 1;
 764                        continue;
 765                }
 766
 767                new = refresh_cache_ent(ce, really, &cache_errno);
 768                if (new == ce)
 769                        continue;
 770                if (!new) {
 771                        if (not_new && cache_errno == ENOENT)
 772                                continue;
 773                        if (really && cache_errno == EINVAL) {
 774                                /* If we are doing --really-refresh that
 775                                 * means the index is not valid anymore.
 776                                 */
 777                                ce->ce_flags &= ~htons(CE_VALID);
 778                                active_cache_changed = 1;
 779                        }
 780                        if (quiet)
 781                                continue;
 782                        printf("%s: needs update\n", ce->name);
 783                        has_errors = 1;
 784                        continue;
 785                }
 786                active_cache_changed = 1;
 787                /* You can NOT just free active_cache[i] here, since it
 788                 * might not be necessarily malloc()ed but can also come
 789                 * from mmap(). */
 790                active_cache[i] = new;
 791        }
 792        return has_errors;
 793}
 794
 795struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
 796{
 797        return refresh_cache_ent(ce, really, NULL);
 798}
 799
 800static int verify_hdr(struct cache_header *hdr, unsigned long size)
 801{
 802        SHA_CTX c;
 803        unsigned char sha1[20];
 804
 805        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
 806                return error("bad signature");
 807        if (hdr->hdr_version != htonl(2))
 808                return error("bad index version");
 809        SHA1_Init(&c);
 810        SHA1_Update(&c, hdr, size - 20);
 811        SHA1_Final(sha1, &c);
 812        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
 813                return error("bad index file sha1 signature");
 814        return 0;
 815}
 816
 817static int read_index_extension(const char *ext, void *data, unsigned long sz)
 818{
 819        switch (CACHE_EXT(ext)) {
 820        case CACHE_EXT_TREE:
 821                active_cache_tree = cache_tree_read(data, sz);
 822                break;
 823        default:
 824                if (*ext < 'A' || 'Z' < *ext)
 825                        return error("index uses %.4s extension, which we do not understand",
 826                                     ext);
 827                fprintf(stderr, "ignoring %.4s extension\n", ext);
 828                break;
 829        }
 830        return 0;
 831}
 832
 833int read_cache(void)
 834{
 835        return read_cache_from(get_index_file());
 836}
 837
 838/* remember to discard_cache() before reading a different cache! */
 839int read_cache_from(const char *path)
 840{
 841        int fd, i;
 842        struct stat st;
 843        unsigned long offset;
 844        struct cache_header *hdr;
 845
 846        errno = EBUSY;
 847        if (cache_mmap)
 848                return active_nr;
 849
 850        errno = ENOENT;
 851        index_file_timestamp = 0;
 852        fd = open(path, O_RDONLY);
 853        if (fd < 0) {
 854                if (errno == ENOENT)
 855                        return 0;
 856                die("index file open failed (%s)", strerror(errno));
 857        }
 858
 859        if (!fstat(fd, &st)) {
 860                cache_mmap_size = xsize_t(st.st_size);
 861                errno = EINVAL;
 862                if (cache_mmap_size >= sizeof(struct cache_header) + 20)
 863                        cache_mmap = xmmap(NULL, cache_mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
 864                else
 865                        die("index file smaller than expected");
 866        } else
 867                die("cannot stat the open index (%s)", strerror(errno));
 868        close(fd);
 869
 870        hdr = cache_mmap;
 871        if (verify_hdr(hdr, cache_mmap_size) < 0)
 872                goto unmap;
 873
 874        active_nr = ntohl(hdr->hdr_entries);
 875        active_alloc = alloc_nr(active_nr);
 876        active_cache = xcalloc(active_alloc, sizeof(struct cache_entry *));
 877
 878        offset = sizeof(*hdr);
 879        for (i = 0; i < active_nr; i++) {
 880                struct cache_entry *ce = (struct cache_entry *) ((char *) cache_mmap + offset);
 881                offset = offset + ce_size(ce);
 882                active_cache[i] = ce;
 883        }
 884        index_file_timestamp = st.st_mtime;
 885        while (offset <= cache_mmap_size - 20 - 8) {
 886                /* After an array of active_nr index entries,
 887                 * there can be arbitrary number of extended
 888                 * sections, each of which is prefixed with
 889                 * extension name (4-byte) and section length
 890                 * in 4-byte network byte order.
 891                 */
 892                unsigned long extsize;
 893                memcpy(&extsize, (char *) cache_mmap + offset + 4, 4);
 894                extsize = ntohl(extsize);
 895                if (read_index_extension(((const char *) cache_mmap) + offset,
 896                                         (char *) cache_mmap + offset + 8,
 897                                         extsize) < 0)
 898                        goto unmap;
 899                offset += 8;
 900                offset += extsize;
 901        }
 902        return active_nr;
 903
 904unmap:
 905        munmap(cache_mmap, cache_mmap_size);
 906        errno = EINVAL;
 907        die("index file corrupt");
 908}
 909
 910int discard_cache(void)
 911{
 912        int ret;
 913
 914        active_nr = active_cache_changed = 0;
 915        index_file_timestamp = 0;
 916        cache_tree_free(&active_cache_tree);
 917        if (cache_mmap == NULL)
 918                return 0;
 919        ret = munmap(cache_mmap, cache_mmap_size);
 920        cache_mmap = NULL;
 921        cache_mmap_size = 0;
 922
 923        /* no need to throw away allocated active_cache */
 924        return ret;
 925}
 926
 927#define WRITE_BUFFER_SIZE 8192
 928static unsigned char write_buffer[WRITE_BUFFER_SIZE];
 929static unsigned long write_buffer_len;
 930
 931static int ce_write_flush(SHA_CTX *context, int fd)
 932{
 933        unsigned int buffered = write_buffer_len;
 934        if (buffered) {
 935                SHA1_Update(context, write_buffer, buffered);
 936                if (write_in_full(fd, write_buffer, buffered) != buffered)
 937                        return -1;
 938                write_buffer_len = 0;
 939        }
 940        return 0;
 941}
 942
 943static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
 944{
 945        while (len) {
 946                unsigned int buffered = write_buffer_len;
 947                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
 948                if (partial > len)
 949                        partial = len;
 950                memcpy(write_buffer + buffered, data, partial);
 951                buffered += partial;
 952                if (buffered == WRITE_BUFFER_SIZE) {
 953                        write_buffer_len = buffered;
 954                        if (ce_write_flush(context, fd))
 955                                return -1;
 956                        buffered = 0;
 957                }
 958                write_buffer_len = buffered;
 959                len -= partial;
 960                data = (char *) data + partial;
 961        }
 962        return 0;
 963}
 964
 965static int write_index_ext_header(SHA_CTX *context, int fd,
 966                                  unsigned int ext, unsigned int sz)
 967{
 968        ext = htonl(ext);
 969        sz = htonl(sz);
 970        return ((ce_write(context, fd, &ext, 4) < 0) ||
 971                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
 972}
 973
 974static int ce_flush(SHA_CTX *context, int fd)
 975{
 976        unsigned int left = write_buffer_len;
 977
 978        if (left) {
 979                write_buffer_len = 0;
 980                SHA1_Update(context, write_buffer, left);
 981        }
 982
 983        /* Flush first if not enough space for SHA1 signature */
 984        if (left + 20 > WRITE_BUFFER_SIZE) {
 985                if (write_in_full(fd, write_buffer, left) != left)
 986                        return -1;
 987                left = 0;
 988        }
 989
 990        /* Append the SHA1 signature at the end */
 991        SHA1_Final(write_buffer + left, context);
 992        left += 20;
 993        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
 994}
 995
 996static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
 997{
 998        /*
 999         * The only thing we care about in this function is to smudge the
1000         * falsely clean entry due to touch-update-touch race, so we leave
1001         * everything else as they are.  We are called for entries whose
1002         * ce_mtime match the index file mtime.
1003         */
1004        struct stat st;
1005
1006        if (lstat(ce->name, &st) < 0)
1007                return;
1008        if (ce_match_stat_basic(ce, &st))
1009                return;
1010        if (ce_modified_check_fs(ce, &st)) {
1011                /* This is "racily clean"; smudge it.  Note that this
1012                 * is a tricky code.  At first glance, it may appear
1013                 * that it can break with this sequence:
1014                 *
1015                 * $ echo xyzzy >frotz
1016                 * $ git-update-index --add frotz
1017                 * $ : >frotz
1018                 * $ sleep 3
1019                 * $ echo filfre >nitfol
1020                 * $ git-update-index --add nitfol
1021                 *
1022                 * but it does not.  When the second update-index runs,
1023                 * it notices that the entry "frotz" has the same timestamp
1024                 * as index, and if we were to smudge it by resetting its
1025                 * size to zero here, then the object name recorded
1026                 * in index is the 6-byte file but the cached stat information
1027                 * becomes zero --- which would then match what we would
1028                 * obtain from the filesystem next time we stat("frotz"). 
1029                 *
1030                 * However, the second update-index, before calling
1031                 * this function, notices that the cached size is 6
1032                 * bytes and what is on the filesystem is an empty
1033                 * file, and never calls us, so the cached size information
1034                 * for "frotz" stays 6 which does not match the filesystem.
1035                 */
1036                ce->ce_size = htonl(0);
1037        }
1038}
1039
1040int write_cache(int newfd, struct cache_entry **cache, int entries)
1041{
1042        SHA_CTX c;
1043        struct cache_header hdr;
1044        int i, removed;
1045
1046        for (i = removed = 0; i < entries; i++)
1047                if (!cache[i]->ce_mode)
1048                        removed++;
1049
1050        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1051        hdr.hdr_version = htonl(2);
1052        hdr.hdr_entries = htonl(entries - removed);
1053
1054        SHA1_Init(&c);
1055        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1056                return -1;
1057
1058        for (i = 0; i < entries; i++) {
1059                struct cache_entry *ce = cache[i];
1060                if (!ce->ce_mode)
1061                        continue;
1062                if (index_file_timestamp &&
1063                    index_file_timestamp <= ntohl(ce->ce_mtime.sec))
1064                        ce_smudge_racily_clean_entry(ce);
1065                if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
1066                        return -1;
1067        }
1068
1069        /* Write extension data here */
1070        if (active_cache_tree) {
1071                unsigned long sz;
1072                void *data = cache_tree_write(active_cache_tree, &sz);
1073                if (data &&
1074                    !write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sz) &&
1075                    !ce_write(&c, newfd, data, sz))
1076                        free(data);
1077                else {
1078                        free(data);
1079                        return -1;
1080                }
1081        }
1082        return ce_flush(&c, newfd);
1083}