read-cache.con commit Merge branch 'maint-1.5.1' into maint (cdd5b82)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 */
   6#define NO_THE_INDEX_COMPATIBILITY_MACROS
   7#include "cache.h"
   8#include "cache-tree.h"
   9#include "refs.h"
  10
  11/* Index extensions.
  12 *
  13 * The first letter should be 'A'..'Z' for extensions that are not
  14 * necessary for a correct operation (i.e. optimization data).
  15 * When new extensions are added that _needs_ to be understood in
  16 * order to correctly interpret the index file, pick character that
  17 * is outside the range, to cause the reader to abort.
  18 */
  19
  20#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  21#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  22
  23struct index_state the_index;
  24
  25/*
  26 * This only updates the "non-critical" parts of the directory
  27 * cache, ie the parts that aren't tracked by GIT, and only used
  28 * to validate the cache.
  29 */
  30void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
  31{
  32        ce->ce_ctime.sec = htonl(st->st_ctime);
  33        ce->ce_mtime.sec = htonl(st->st_mtime);
  34#ifdef USE_NSEC
  35        ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
  36        ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
  37#endif
  38        ce->ce_dev = htonl(st->st_dev);
  39        ce->ce_ino = htonl(st->st_ino);
  40        ce->ce_uid = htonl(st->st_uid);
  41        ce->ce_gid = htonl(st->st_gid);
  42        ce->ce_size = htonl(st->st_size);
  43
  44        if (assume_unchanged)
  45                ce->ce_flags |= htons(CE_VALID);
  46}
  47
  48static int ce_compare_data(struct cache_entry *ce, struct stat *st)
  49{
  50        int match = -1;
  51        int fd = open(ce->name, O_RDONLY);
  52
  53        if (fd >= 0) {
  54                unsigned char sha1[20];
  55                if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
  56                        match = hashcmp(sha1, ce->sha1);
  57                /* index_fd() closed the file descriptor already */
  58        }
  59        return match;
  60}
  61
  62static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
  63{
  64        int match = -1;
  65        char *target;
  66        void *buffer;
  67        unsigned long size;
  68        enum object_type type;
  69        int len;
  70
  71        target = xmalloc(expected_size);
  72        len = readlink(ce->name, target, expected_size);
  73        if (len != expected_size) {
  74                free(target);
  75                return -1;
  76        }
  77        buffer = read_sha1_file(ce->sha1, &type, &size);
  78        if (!buffer) {
  79                free(target);
  80                return -1;
  81        }
  82        if (size == expected_size)
  83                match = memcmp(buffer, target, size);
  84        free(buffer);
  85        free(target);
  86        return match;
  87}
  88
  89static int ce_compare_gitlink(struct cache_entry *ce)
  90{
  91        unsigned char sha1[20];
  92
  93        /*
  94         * We don't actually require that the .git directory
  95         * under DIRLNK directory be a valid git directory. It
  96         * might even be missing (in case nobody populated that
  97         * sub-project).
  98         *
  99         * If so, we consider it always to match.
 100         */
 101        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 102                return 0;
 103        return hashcmp(sha1, ce->sha1);
 104}
 105
 106static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
 107{
 108        switch (st->st_mode & S_IFMT) {
 109        case S_IFREG:
 110                if (ce_compare_data(ce, st))
 111                        return DATA_CHANGED;
 112                break;
 113        case S_IFLNK:
 114                if (ce_compare_link(ce, xsize_t(st->st_size)))
 115                        return DATA_CHANGED;
 116                break;
 117        case S_IFDIR:
 118                if (S_ISDIRLNK(ntohl(ce->ce_mode)))
 119                        return 0;
 120        default:
 121                return TYPE_CHANGED;
 122        }
 123        return 0;
 124}
 125
 126static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 127{
 128        unsigned int changed = 0;
 129
 130        switch (ntohl(ce->ce_mode) & S_IFMT) {
 131        case S_IFREG:
 132                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 133                /* We consider only the owner x bit to be relevant for
 134                 * "mode changes"
 135                 */
 136                if (trust_executable_bit &&
 137                    (0100 & (ntohl(ce->ce_mode) ^ st->st_mode)))
 138                        changed |= MODE_CHANGED;
 139                break;
 140        case S_IFLNK:
 141                if (!S_ISLNK(st->st_mode) &&
 142                    (has_symlinks || !S_ISREG(st->st_mode)))
 143                        changed |= TYPE_CHANGED;
 144                break;
 145        case S_IFDIRLNK:
 146                if (!S_ISDIR(st->st_mode))
 147                        changed |= TYPE_CHANGED;
 148                else if (ce_compare_gitlink(ce))
 149                        changed |= DATA_CHANGED;
 150                return changed;
 151        default:
 152                die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
 153        }
 154        if (ce->ce_mtime.sec != htonl(st->st_mtime))
 155                changed |= MTIME_CHANGED;
 156        if (ce->ce_ctime.sec != htonl(st->st_ctime))
 157                changed |= CTIME_CHANGED;
 158
 159#ifdef USE_NSEC
 160        /*
 161         * nsec seems unreliable - not all filesystems support it, so
 162         * as long as it is in the inode cache you get right nsec
 163         * but after it gets flushed, you get zero nsec.
 164         */
 165        if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
 166                changed |= MTIME_CHANGED;
 167        if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
 168                changed |= CTIME_CHANGED;
 169#endif  
 170
 171        if (ce->ce_uid != htonl(st->st_uid) ||
 172            ce->ce_gid != htonl(st->st_gid))
 173                changed |= OWNER_CHANGED;
 174        if (ce->ce_ino != htonl(st->st_ino))
 175                changed |= INODE_CHANGED;
 176
 177#ifdef USE_STDEV
 178        /*
 179         * st_dev breaks on network filesystems where different
 180         * clients will have different views of what "device"
 181         * the filesystem is on
 182         */
 183        if (ce->ce_dev != htonl(st->st_dev))
 184                changed |= INODE_CHANGED;
 185#endif
 186
 187        if (ce->ce_size != htonl(st->st_size))
 188                changed |= DATA_CHANGED;
 189
 190        return changed;
 191}
 192
 193int ie_match_stat(struct index_state *istate,
 194                  struct cache_entry *ce, struct stat *st, int options)
 195{
 196        unsigned int changed;
 197        int ignore_valid = options & 01;
 198        int assume_racy_is_modified = options & 02;
 199
 200        /*
 201         * If it's marked as always valid in the index, it's
 202         * valid whatever the checked-out copy says.
 203         */
 204        if (!ignore_valid && (ce->ce_flags & htons(CE_VALID)))
 205                return 0;
 206
 207        changed = ce_match_stat_basic(ce, st);
 208
 209        /*
 210         * Within 1 second of this sequence:
 211         *      echo xyzzy >file && git-update-index --add file
 212         * running this command:
 213         *      echo frotz >file
 214         * would give a falsely clean cache entry.  The mtime and
 215         * length match the cache, and other stat fields do not change.
 216         *
 217         * We could detect this at update-index time (the cache entry
 218         * being registered/updated records the same time as "now")
 219         * and delay the return from git-update-index, but that would
 220         * effectively mean we can make at most one commit per second,
 221         * which is not acceptable.  Instead, we check cache entries
 222         * whose mtime are the same as the index file timestamp more
 223         * carefully than others.
 224         */
 225        if (!changed &&
 226            istate->timestamp &&
 227            istate->timestamp <= ntohl(ce->ce_mtime.sec)) {
 228                if (assume_racy_is_modified)
 229                        changed |= DATA_CHANGED;
 230                else
 231                        changed |= ce_modified_check_fs(ce, st);
 232        }
 233
 234        return changed;
 235}
 236
 237int ie_modified(struct index_state *istate,
 238                struct cache_entry *ce, struct stat *st, int really)
 239{
 240        int changed, changed_fs;
 241        changed = ie_match_stat(istate, ce, st, really);
 242        if (!changed)
 243                return 0;
 244        /*
 245         * If the mode or type has changed, there's no point in trying
 246         * to refresh the entry - it's not going to match
 247         */
 248        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 249                return changed;
 250
 251        /* Immediately after read-tree or update-index --cacheinfo,
 252         * the length field is zero.  For other cases the ce_size
 253         * should match the SHA1 recorded in the index entry.
 254         */
 255        if ((changed & DATA_CHANGED) && ce->ce_size != htonl(0))
 256                return changed;
 257
 258        changed_fs = ce_modified_check_fs(ce, st);
 259        if (changed_fs)
 260                return changed | changed_fs;
 261        return 0;
 262}
 263
 264int base_name_compare(const char *name1, int len1, int mode1,
 265                      const char *name2, int len2, int mode2)
 266{
 267        unsigned char c1, c2;
 268        int len = len1 < len2 ? len1 : len2;
 269        int cmp;
 270
 271        cmp = memcmp(name1, name2, len);
 272        if (cmp)
 273                return cmp;
 274        c1 = name1[len];
 275        c2 = name2[len];
 276        if (!c1 && S_ISDIR(mode1))
 277                c1 = '/';
 278        if (!c2 && S_ISDIR(mode2))
 279                c2 = '/';
 280        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 281}
 282
 283int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 284{
 285        int len1 = flags1 & CE_NAMEMASK;
 286        int len2 = flags2 & CE_NAMEMASK;
 287        int len = len1 < len2 ? len1 : len2;
 288        int cmp;
 289
 290        cmp = memcmp(name1, name2, len);
 291        if (cmp)
 292                return cmp;
 293        if (len1 < len2)
 294                return -1;
 295        if (len1 > len2)
 296                return 1;
 297
 298        /* Compare stages  */
 299        flags1 &= CE_STAGEMASK;
 300        flags2 &= CE_STAGEMASK;
 301
 302        if (flags1 < flags2)
 303                return -1;
 304        if (flags1 > flags2)
 305                return 1;
 306        return 0;
 307}
 308
 309int index_name_pos(struct index_state *istate, const char *name, int namelen)
 310{
 311        int first, last;
 312
 313        first = 0;
 314        last = istate->cache_nr;
 315        while (last > first) {
 316                int next = (last + first) >> 1;
 317                struct cache_entry *ce = istate->cache[next];
 318                int cmp = cache_name_compare(name, namelen, ce->name, ntohs(ce->ce_flags));
 319                if (!cmp)
 320                        return next;
 321                if (cmp < 0) {
 322                        last = next;
 323                        continue;
 324                }
 325                first = next+1;
 326        }
 327        return -first-1;
 328}
 329
 330/* Remove entry, return true if there are more entries to go.. */
 331int remove_index_entry_at(struct index_state *istate, int pos)
 332{
 333        istate->cache_changed = 1;
 334        istate->cache_nr--;
 335        if (pos >= istate->cache_nr)
 336                return 0;
 337        memmove(istate->cache + pos,
 338                istate->cache + pos + 1,
 339                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 340        return 1;
 341}
 342
 343int remove_file_from_index(struct index_state *istate, const char *path)
 344{
 345        int pos = index_name_pos(istate, path, strlen(path));
 346        if (pos < 0)
 347                pos = -pos-1;
 348        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 349                remove_index_entry_at(istate, pos);
 350        return 0;
 351}
 352
 353int add_file_to_index(struct index_state *istate, const char *path, int verbose)
 354{
 355        int size, namelen;
 356        struct stat st;
 357        struct cache_entry *ce;
 358
 359        if (lstat(path, &st))
 360                die("%s: unable to stat (%s)", path, strerror(errno));
 361
 362        if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
 363                die("%s: can only add regular files, symbolic links or git-directories", path);
 364
 365        namelen = strlen(path);
 366        if (S_ISDIR(st.st_mode)) {
 367                while (namelen && path[namelen-1] == '/')
 368                        namelen--;
 369        }
 370        size = cache_entry_size(namelen);
 371        ce = xcalloc(1, size);
 372        memcpy(ce->name, path, namelen);
 373        ce->ce_flags = htons(namelen);
 374        fill_stat_cache_info(ce, &st);
 375
 376        if (trust_executable_bit && has_symlinks)
 377                ce->ce_mode = create_ce_mode(st.st_mode);
 378        else {
 379                /* If there is an existing entry, pick the mode bits and type
 380                 * from it, otherwise assume unexecutable regular file.
 381                 */
 382                struct cache_entry *ent;
 383                int pos = index_name_pos(istate, path, namelen);
 384
 385                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 386                ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
 387        }
 388
 389        if (index_path(ce->sha1, path, &st, 1))
 390                die("unable to index file %s", path);
 391        if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
 392                die("unable to add %s to index",path);
 393        if (verbose)
 394                printf("add '%s'\n", path);
 395        cache_tree_invalidate_path(istate->cache_tree, path);
 396        return 0;
 397}
 398
 399int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 400{
 401        int len = ce_namelen(a);
 402        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 403}
 404
 405int ce_path_match(const struct cache_entry *ce, const char **pathspec)
 406{
 407        const char *match, *name;
 408        int len;
 409
 410        if (!pathspec)
 411                return 1;
 412
 413        len = ce_namelen(ce);
 414        name = ce->name;
 415        while ((match = *pathspec++) != NULL) {
 416                int matchlen = strlen(match);
 417                if (matchlen > len)
 418                        continue;
 419                if (memcmp(name, match, matchlen))
 420                        continue;
 421                if (matchlen && name[matchlen-1] == '/')
 422                        return 1;
 423                if (name[matchlen] == '/' || !name[matchlen])
 424                        return 1;
 425                if (!matchlen)
 426                        return 1;
 427        }
 428        return 0;
 429}
 430
 431/*
 432 * We fundamentally don't like some paths: we don't want
 433 * dot or dot-dot anywhere, and for obvious reasons don't
 434 * want to recurse into ".git" either.
 435 *
 436 * Also, we don't want double slashes or slashes at the
 437 * end that can make pathnames ambiguous.
 438 */
 439static int verify_dotfile(const char *rest)
 440{
 441        /*
 442         * The first character was '.', but that
 443         * has already been discarded, we now test
 444         * the rest.
 445         */
 446        switch (*rest) {
 447        /* "." is not allowed */
 448        case '\0': case '/':
 449                return 0;
 450
 451        /*
 452         * ".git" followed by  NUL or slash is bad. This
 453         * shares the path end test with the ".." case.
 454         */
 455        case 'g':
 456                if (rest[1] != 'i')
 457                        break;
 458                if (rest[2] != 't')
 459                        break;
 460                rest += 2;
 461        /* fallthrough */
 462        case '.':
 463                if (rest[1] == '\0' || rest[1] == '/')
 464                        return 0;
 465        }
 466        return 1;
 467}
 468
 469int verify_path(const char *path)
 470{
 471        char c;
 472
 473        goto inside;
 474        for (;;) {
 475                if (!c)
 476                        return 1;
 477                if (c == '/') {
 478inside:
 479                        c = *path++;
 480                        switch (c) {
 481                        default:
 482                                continue;
 483                        case '/': case '\0':
 484                                break;
 485                        case '.':
 486                                if (verify_dotfile(path))
 487                                        continue;
 488                        }
 489                        return 0;
 490                }
 491                c = *path++;
 492        }
 493}
 494
 495/*
 496 * Do we have another file that has the beginning components being a
 497 * proper superset of the name we're trying to add?
 498 */
 499static int has_file_name(struct index_state *istate,
 500                         const struct cache_entry *ce, int pos, int ok_to_replace)
 501{
 502        int retval = 0;
 503        int len = ce_namelen(ce);
 504        int stage = ce_stage(ce);
 505        const char *name = ce->name;
 506
 507        while (pos < istate->cache_nr) {
 508                struct cache_entry *p = istate->cache[pos++];
 509
 510                if (len >= ce_namelen(p))
 511                        break;
 512                if (memcmp(name, p->name, len))
 513                        break;
 514                if (ce_stage(p) != stage)
 515                        continue;
 516                if (p->name[len] != '/')
 517                        continue;
 518                if (!ce_stage(p) && !p->ce_mode)
 519                        continue;
 520                retval = -1;
 521                if (!ok_to_replace)
 522                        break;
 523                remove_index_entry_at(istate, --pos);
 524        }
 525        return retval;
 526}
 527
 528/*
 529 * Do we have another file with a pathname that is a proper
 530 * subset of the name we're trying to add?
 531 */
 532static int has_dir_name(struct index_state *istate,
 533                        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 = index_name_pos(istate, 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 || istate->cache[pos]->ce_mode) {
 562                                retval = -1;
 563                                if (!ok_to_replace)
 564                                        break;
 565                                remove_index_entry_at(istate, 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 < istate->cache_nr) {
 578                        struct cache_entry *p = istate->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(struct index_state *istate,
 606                                         const struct cache_entry *ce,
 607                                         int pos, int ok_to_replace)
 608{
 609        int retval;
 610
 611        /*
 612         * When ce is an "I am going away" entry, we allow it to be added
 613         */
 614        if (!ce_stage(ce) && !ce->ce_mode)
 615                return 0;
 616
 617        /*
 618         * We check if the path is a sub-path of a subsequent pathname
 619         * first, since removing those will not change the position
 620         * in the array.
 621         */
 622        retval = has_file_name(istate, ce, pos, ok_to_replace);
 623
 624        /*
 625         * Then check if the path might have a clashing sub-directory
 626         * before it.
 627         */
 628        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 629}
 630
 631int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 632{
 633        int pos;
 634        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 635        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 636        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 637
 638        pos = index_name_pos(istate, ce->name, ntohs(ce->ce_flags));
 639
 640        /* existing match? Just replace it. */
 641        if (pos >= 0) {
 642                istate->cache_changed = 1;
 643                istate->cache[pos] = ce;
 644                return 0;
 645        }
 646        pos = -pos-1;
 647
 648        /*
 649         * Inserting a merged entry ("stage 0") into the index
 650         * will always replace all non-merged entries..
 651         */
 652        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 653                while (ce_same_name(istate->cache[pos], ce)) {
 654                        ok_to_add = 1;
 655                        if (!remove_index_entry_at(istate, pos))
 656                                break;
 657                }
 658        }
 659
 660        if (!ok_to_add)
 661                return -1;
 662        if (!verify_path(ce->name))
 663                return -1;
 664
 665        if (!skip_df_check &&
 666            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 667                if (!ok_to_replace)
 668                        return error("'%s' appears as both a file and as a directory",
 669                                     ce->name);
 670                pos = index_name_pos(istate, ce->name, ntohs(ce->ce_flags));
 671                pos = -pos-1;
 672        }
 673
 674        /* Make sure the array is big enough .. */
 675        if (istate->cache_nr == istate->cache_alloc) {
 676                istate->cache_alloc = alloc_nr(istate->cache_alloc);
 677                istate->cache = xrealloc(istate->cache,
 678                                        istate->cache_alloc * sizeof(struct cache_entry *));
 679        }
 680
 681        /* Add it in.. */
 682        istate->cache_nr++;
 683        if (istate->cache_nr > pos)
 684                memmove(istate->cache + pos + 1,
 685                        istate->cache + pos,
 686                        (istate->cache_nr - pos - 1) * sizeof(ce));
 687        istate->cache[pos] = ce;
 688        istate->cache_changed = 1;
 689        return 0;
 690}
 691
 692/*
 693 * "refresh" does not calculate a new sha1 file or bring the
 694 * cache up-to-date for mode/content changes. But what it
 695 * _does_ do is to "re-match" the stat information of a file
 696 * with the cache, so that you can refresh the cache for a
 697 * file that hasn't been changed but where the stat entry is
 698 * out of date.
 699 *
 700 * For example, you'd want to do this after doing a "git-read-tree",
 701 * to link up the stat cache details with the proper files.
 702 */
 703static struct cache_entry *refresh_cache_ent(struct index_state *istate,
 704                                             struct cache_entry *ce, int really, int *err)
 705{
 706        struct stat st;
 707        struct cache_entry *updated;
 708        int changed, size;
 709
 710        if (lstat(ce->name, &st) < 0) {
 711                if (err)
 712                        *err = errno;
 713                return NULL;
 714        }
 715
 716        changed = ie_match_stat(istate, ce, &st, really);
 717        if (!changed) {
 718                if (really && assume_unchanged &&
 719                    !(ce->ce_flags & htons(CE_VALID)))
 720                        ; /* mark this one VALID again */
 721                else
 722                        return ce;
 723        }
 724
 725        if (ie_modified(istate, ce, &st, really)) {
 726                if (err)
 727                        *err = EINVAL;
 728                return NULL;
 729        }
 730
 731        size = ce_size(ce);
 732        updated = xmalloc(size);
 733        memcpy(updated, ce, size);
 734        fill_stat_cache_info(updated, &st);
 735
 736        /* In this case, if really is not set, we should leave
 737         * CE_VALID bit alone.  Otherwise, paths marked with
 738         * --no-assume-unchanged (i.e. things to be edited) will
 739         * reacquire CE_VALID bit automatically, which is not
 740         * really what we want.
 741         */
 742        if (!really && assume_unchanged && !(ce->ce_flags & htons(CE_VALID)))
 743                updated->ce_flags &= ~htons(CE_VALID);
 744
 745        return updated;
 746}
 747
 748int refresh_index(struct index_state *istate, unsigned int flags)
 749{
 750        int i;
 751        int has_errors = 0;
 752        int really = (flags & REFRESH_REALLY) != 0;
 753        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
 754        int quiet = (flags & REFRESH_QUIET) != 0;
 755        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
 756
 757        for (i = 0; i < istate->cache_nr; i++) {
 758                struct cache_entry *ce, *new;
 759                int cache_errno = 0;
 760
 761                ce = istate->cache[i];
 762                if (ce_stage(ce)) {
 763                        while ((i < istate->cache_nr) &&
 764                               ! strcmp(istate->cache[i]->name, ce->name))
 765                                i++;
 766                        i--;
 767                        if (allow_unmerged)
 768                                continue;
 769                        printf("%s: needs merge\n", ce->name);
 770                        has_errors = 1;
 771                        continue;
 772                }
 773
 774                new = refresh_cache_ent(istate, ce, really, &cache_errno);
 775                if (new == ce)
 776                        continue;
 777                if (!new) {
 778                        if (not_new && cache_errno == ENOENT)
 779                                continue;
 780                        if (really && cache_errno == EINVAL) {
 781                                /* If we are doing --really-refresh that
 782                                 * means the index is not valid anymore.
 783                                 */
 784                                ce->ce_flags &= ~htons(CE_VALID);
 785                                istate->cache_changed = 1;
 786                        }
 787                        if (quiet)
 788                                continue;
 789                        printf("%s: needs update\n", ce->name);
 790                        has_errors = 1;
 791                        continue;
 792                }
 793                istate->cache_changed = 1;
 794                /* You can NOT just free istate->cache[i] here, since it
 795                 * might not be necessarily malloc()ed but can also come
 796                 * from mmap(). */
 797                istate->cache[i] = new;
 798        }
 799        return has_errors;
 800}
 801
 802struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
 803{
 804        return refresh_cache_ent(&the_index, ce, really, NULL);
 805}
 806
 807static int verify_hdr(struct cache_header *hdr, unsigned long size)
 808{
 809        SHA_CTX c;
 810        unsigned char sha1[20];
 811
 812        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
 813                return error("bad signature");
 814        if (hdr->hdr_version != htonl(2))
 815                return error("bad index version");
 816        SHA1_Init(&c);
 817        SHA1_Update(&c, hdr, size - 20);
 818        SHA1_Final(sha1, &c);
 819        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
 820                return error("bad index file sha1 signature");
 821        return 0;
 822}
 823
 824static int read_index_extension(struct index_state *istate,
 825                                const char *ext, void *data, unsigned long sz)
 826{
 827        switch (CACHE_EXT(ext)) {
 828        case CACHE_EXT_TREE:
 829                istate->cache_tree = cache_tree_read(data, sz);
 830                break;
 831        default:
 832                if (*ext < 'A' || 'Z' < *ext)
 833                        return error("index uses %.4s extension, which we do not understand",
 834                                     ext);
 835                fprintf(stderr, "ignoring %.4s extension\n", ext);
 836                break;
 837        }
 838        return 0;
 839}
 840
 841int read_index(struct index_state *istate)
 842{
 843        return read_index_from(istate, get_index_file());
 844}
 845
 846/* remember to discard_cache() before reading a different cache! */
 847int read_index_from(struct index_state *istate, const char *path)
 848{
 849        int fd, i;
 850        struct stat st;
 851        unsigned long offset;
 852        struct cache_header *hdr;
 853
 854        errno = EBUSY;
 855        if (istate->mmap)
 856                return istate->cache_nr;
 857
 858        errno = ENOENT;
 859        istate->timestamp = 0;
 860        fd = open(path, O_RDONLY);
 861        if (fd < 0) {
 862                if (errno == ENOENT)
 863                        return 0;
 864                die("index file open failed (%s)", strerror(errno));
 865        }
 866
 867        if (fstat(fd, &st))
 868                die("cannot stat the open index (%s)", strerror(errno));
 869
 870        errno = EINVAL;
 871        istate->mmap_size = xsize_t(st.st_size);
 872        if (istate->mmap_size < sizeof(struct cache_header) + 20)
 873                die("index file smaller than expected");
 874
 875        istate->mmap = xmmap(NULL, istate->mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
 876        close(fd);
 877
 878        hdr = istate->mmap;
 879        if (verify_hdr(hdr, istate->mmap_size) < 0)
 880                goto unmap;
 881
 882        istate->cache_nr = ntohl(hdr->hdr_entries);
 883        istate->cache_alloc = alloc_nr(istate->cache_nr);
 884        istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
 885
 886        offset = sizeof(*hdr);
 887        for (i = 0; i < istate->cache_nr; i++) {
 888                struct cache_entry *ce;
 889
 890                ce = (struct cache_entry *)((char *)(istate->mmap) + offset);
 891                offset = offset + ce_size(ce);
 892                istate->cache[i] = ce;
 893        }
 894        istate->timestamp = st.st_mtime;
 895        while (offset <= istate->mmap_size - 20 - 8) {
 896                /* After an array of active_nr index entries,
 897                 * there can be arbitrary number of extended
 898                 * sections, each of which is prefixed with
 899                 * extension name (4-byte) and section length
 900                 * in 4-byte network byte order.
 901                 */
 902                unsigned long extsize;
 903                memcpy(&extsize, (char *)(istate->mmap) + offset + 4, 4);
 904                extsize = ntohl(extsize);
 905                if (read_index_extension(istate,
 906                                         ((const char *) (istate->mmap)) + offset,
 907                                         (char *) (istate->mmap) + offset + 8,
 908                                         extsize) < 0)
 909                        goto unmap;
 910                offset += 8;
 911                offset += extsize;
 912        }
 913        return istate->cache_nr;
 914
 915unmap:
 916        munmap(istate->mmap, istate->mmap_size);
 917        errno = EINVAL;
 918        die("index file corrupt");
 919}
 920
 921int discard_index(struct index_state *istate)
 922{
 923        int ret;
 924
 925        istate->cache_nr = 0;
 926        istate->cache_changed = 0;
 927        istate->timestamp = 0;
 928        cache_tree_free(&(istate->cache_tree));
 929        if (istate->mmap == NULL)
 930                return 0;
 931        ret = munmap(istate->mmap, istate->mmap_size);
 932        istate->mmap = NULL;
 933        istate->mmap_size = 0;
 934
 935        /* no need to throw away allocated active_cache */
 936        return ret;
 937}
 938
 939#define WRITE_BUFFER_SIZE 8192
 940static unsigned char write_buffer[WRITE_BUFFER_SIZE];
 941static unsigned long write_buffer_len;
 942
 943static int ce_write_flush(SHA_CTX *context, int fd)
 944{
 945        unsigned int buffered = write_buffer_len;
 946        if (buffered) {
 947                SHA1_Update(context, write_buffer, buffered);
 948                if (write_in_full(fd, write_buffer, buffered) != buffered)
 949                        return -1;
 950                write_buffer_len = 0;
 951        }
 952        return 0;
 953}
 954
 955static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
 956{
 957        while (len) {
 958                unsigned int buffered = write_buffer_len;
 959                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
 960                if (partial > len)
 961                        partial = len;
 962                memcpy(write_buffer + buffered, data, partial);
 963                buffered += partial;
 964                if (buffered == WRITE_BUFFER_SIZE) {
 965                        write_buffer_len = buffered;
 966                        if (ce_write_flush(context, fd))
 967                                return -1;
 968                        buffered = 0;
 969                }
 970                write_buffer_len = buffered;
 971                len -= partial;
 972                data = (char *) data + partial;
 973        }
 974        return 0;
 975}
 976
 977static int write_index_ext_header(SHA_CTX *context, int fd,
 978                                  unsigned int ext, unsigned int sz)
 979{
 980        ext = htonl(ext);
 981        sz = htonl(sz);
 982        return ((ce_write(context, fd, &ext, 4) < 0) ||
 983                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
 984}
 985
 986static int ce_flush(SHA_CTX *context, int fd)
 987{
 988        unsigned int left = write_buffer_len;
 989
 990        if (left) {
 991                write_buffer_len = 0;
 992                SHA1_Update(context, write_buffer, left);
 993        }
 994
 995        /* Flush first if not enough space for SHA1 signature */
 996        if (left + 20 > WRITE_BUFFER_SIZE) {
 997                if (write_in_full(fd, write_buffer, left) != left)
 998                        return -1;
 999                left = 0;
1000        }
1001
1002        /* Append the SHA1 signature at the end */
1003        SHA1_Final(write_buffer + left, context);
1004        left += 20;
1005        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1006}
1007
1008static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1009{
1010        /*
1011         * The only thing we care about in this function is to smudge the
1012         * falsely clean entry due to touch-update-touch race, so we leave
1013         * everything else as they are.  We are called for entries whose
1014         * ce_mtime match the index file mtime.
1015         */
1016        struct stat st;
1017
1018        if (lstat(ce->name, &st) < 0)
1019                return;
1020        if (ce_match_stat_basic(ce, &st))
1021                return;
1022        if (ce_modified_check_fs(ce, &st)) {
1023                /* This is "racily clean"; smudge it.  Note that this
1024                 * is a tricky code.  At first glance, it may appear
1025                 * that it can break with this sequence:
1026                 *
1027                 * $ echo xyzzy >frotz
1028                 * $ git-update-index --add frotz
1029                 * $ : >frotz
1030                 * $ sleep 3
1031                 * $ echo filfre >nitfol
1032                 * $ git-update-index --add nitfol
1033                 *
1034                 * but it does not.  When the second update-index runs,
1035                 * it notices that the entry "frotz" has the same timestamp
1036                 * as index, and if we were to smudge it by resetting its
1037                 * size to zero here, then the object name recorded
1038                 * in index is the 6-byte file but the cached stat information
1039                 * becomes zero --- which would then match what we would
1040                 * obtain from the filesystem next time we stat("frotz"). 
1041                 *
1042                 * However, the second update-index, before calling
1043                 * this function, notices that the cached size is 6
1044                 * bytes and what is on the filesystem is an empty
1045                 * file, and never calls us, so the cached size information
1046                 * for "frotz" stays 6 which does not match the filesystem.
1047                 */
1048                ce->ce_size = htonl(0);
1049        }
1050}
1051
1052int write_index(struct index_state *istate, int newfd)
1053{
1054        SHA_CTX c;
1055        struct cache_header hdr;
1056        int i, removed;
1057        struct cache_entry **cache = istate->cache;
1058        int entries = istate->cache_nr;
1059
1060        for (i = removed = 0; i < entries; i++)
1061                if (!cache[i]->ce_mode)
1062                        removed++;
1063
1064        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1065        hdr.hdr_version = htonl(2);
1066        hdr.hdr_entries = htonl(entries - removed);
1067
1068        SHA1_Init(&c);
1069        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1070                return -1;
1071
1072        for (i = 0; i < entries; i++) {
1073                struct cache_entry *ce = cache[i];
1074                if (!ce->ce_mode)
1075                        continue;
1076                if (istate->timestamp &&
1077                    istate->timestamp <= ntohl(ce->ce_mtime.sec))
1078                        ce_smudge_racily_clean_entry(ce);
1079                if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
1080                        return -1;
1081        }
1082
1083        /* Write extension data here */
1084        if (istate->cache_tree) {
1085                unsigned long sz;
1086                void *data = cache_tree_write(istate->cache_tree, &sz);
1087                if (data &&
1088                    !write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sz) &&
1089                    !ce_write(&c, newfd, data, sz))
1090                        free(data);
1091                else {
1092                        free(data);
1093                        return -1;
1094                }
1095        }
1096        return ce_flush(&c, newfd);
1097}