read-cache.con commit Add testcase for 'git cvsexportcommit -w $cvsdir ...' with relative $GIT_DIR (ab5a423)
   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#include "dir.h"
  11
  12/* Index extensions.
  13 *
  14 * The first letter should be 'A'..'Z' for extensions that are not
  15 * necessary for a correct operation (i.e. optimization data).
  16 * When new extensions are added that _needs_ to be understood in
  17 * order to correctly interpret the index file, pick character that
  18 * is outside the range, to cause the reader to abort.
  19 */
  20
  21#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  22#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  23
  24struct index_state the_index;
  25
  26static unsigned int hash_name(const char *name, int namelen)
  27{
  28        unsigned int hash = 0x123;
  29
  30        do {
  31                unsigned char c = *name++;
  32                hash = hash*101 + c;
  33        } while (--namelen);
  34        return hash;
  35}
  36
  37static void hash_index_entry(struct index_state *istate, struct cache_entry *ce)
  38{
  39        void **pos;
  40        unsigned int hash = hash_name(ce->name, ce_namelen(ce));
  41
  42        pos = insert_hash(hash, ce, &istate->name_hash);
  43        if (pos) {
  44                ce->next = *pos;
  45                *pos = ce;
  46        }
  47}
  48
  49static void lazy_init_name_hash(struct index_state *istate)
  50{
  51        int nr;
  52
  53        if (istate->name_hash_initialized)
  54                return;
  55        for (nr = 0; nr < istate->cache_nr; nr++)
  56                hash_index_entry(istate, istate->cache[nr]);
  57        istate->name_hash_initialized = 1;
  58}
  59
  60static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  61{
  62        istate->cache[nr] = ce;
  63        if (istate->name_hash_initialized)
  64                hash_index_entry(istate, ce);
  65}
  66
  67/*
  68 * We don't actually *remove* it, we can just mark it invalid so that
  69 * we won't find it in lookups.
  70 *
  71 * Not only would we have to search the lists (simple enough), but
  72 * we'd also have to rehash other hash buckets in case this makes the
  73 * hash bucket empty (common). So it's much better to just mark
  74 * it.
  75 */
  76static void remove_hash_entry(struct index_state *istate, struct cache_entry *ce)
  77{
  78        ce->ce_flags |= CE_UNHASHED;
  79}
  80
  81static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  82{
  83        struct cache_entry *old = istate->cache[nr];
  84
  85        if (ce != old) {
  86                remove_hash_entry(istate, old);
  87                set_index_entry(istate, nr, ce);
  88        }
  89        istate->cache_changed = 1;
  90}
  91
  92int index_name_exists(struct index_state *istate, const char *name, int namelen)
  93{
  94        unsigned int hash = hash_name(name, namelen);
  95        struct cache_entry *ce;
  96
  97        lazy_init_name_hash(istate);
  98        ce = lookup_hash(hash, &istate->name_hash);
  99
 100        while (ce) {
 101                if (!(ce->ce_flags & CE_UNHASHED)) {
 102                        if (!cache_name_compare(name, namelen, ce->name, ce->ce_flags))
 103                                return 1;
 104                }
 105                ce = ce->next;
 106        }
 107        return 0;
 108}
 109
 110/*
 111 * This only updates the "non-critical" parts of the directory
 112 * cache, ie the parts that aren't tracked by GIT, and only used
 113 * to validate the cache.
 114 */
 115void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 116{
 117        ce->ce_ctime = st->st_ctime;
 118        ce->ce_mtime = st->st_mtime;
 119        ce->ce_dev = st->st_dev;
 120        ce->ce_ino = st->st_ino;
 121        ce->ce_uid = st->st_uid;
 122        ce->ce_gid = st->st_gid;
 123        ce->ce_size = st->st_size;
 124
 125        if (assume_unchanged)
 126                ce->ce_flags |= CE_VALID;
 127
 128        if (S_ISREG(st->st_mode))
 129                ce_mark_uptodate(ce);
 130}
 131
 132static int ce_compare_data(struct cache_entry *ce, struct stat *st)
 133{
 134        int match = -1;
 135        int fd = open(ce->name, O_RDONLY);
 136
 137        if (fd >= 0) {
 138                unsigned char sha1[20];
 139                if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
 140                        match = hashcmp(sha1, ce->sha1);
 141                /* index_fd() closed the file descriptor already */
 142        }
 143        return match;
 144}
 145
 146static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
 147{
 148        int match = -1;
 149        char *target;
 150        void *buffer;
 151        unsigned long size;
 152        enum object_type type;
 153        int len;
 154
 155        target = xmalloc(expected_size);
 156        len = readlink(ce->name, target, expected_size);
 157        if (len != expected_size) {
 158                free(target);
 159                return -1;
 160        }
 161        buffer = read_sha1_file(ce->sha1, &type, &size);
 162        if (!buffer) {
 163                free(target);
 164                return -1;
 165        }
 166        if (size == expected_size)
 167                match = memcmp(buffer, target, size);
 168        free(buffer);
 169        free(target);
 170        return match;
 171}
 172
 173static int ce_compare_gitlink(struct cache_entry *ce)
 174{
 175        unsigned char sha1[20];
 176
 177        /*
 178         * We don't actually require that the .git directory
 179         * under GITLINK directory be a valid git directory. It
 180         * might even be missing (in case nobody populated that
 181         * sub-project).
 182         *
 183         * If so, we consider it always to match.
 184         */
 185        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 186                return 0;
 187        return hashcmp(sha1, ce->sha1);
 188}
 189
 190static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
 191{
 192        switch (st->st_mode & S_IFMT) {
 193        case S_IFREG:
 194                if (ce_compare_data(ce, st))
 195                        return DATA_CHANGED;
 196                break;
 197        case S_IFLNK:
 198                if (ce_compare_link(ce, xsize_t(st->st_size)))
 199                        return DATA_CHANGED;
 200                break;
 201        case S_IFDIR:
 202                if (S_ISGITLINK(ce->ce_mode))
 203                        return 0;
 204        default:
 205                return TYPE_CHANGED;
 206        }
 207        return 0;
 208}
 209
 210static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 211{
 212        unsigned int changed = 0;
 213
 214        if (ce->ce_flags & CE_REMOVE)
 215                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 216
 217        switch (ce->ce_mode & S_IFMT) {
 218        case S_IFREG:
 219                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 220                /* We consider only the owner x bit to be relevant for
 221                 * "mode changes"
 222                 */
 223                if (trust_executable_bit &&
 224                    (0100 & (ce->ce_mode ^ st->st_mode)))
 225                        changed |= MODE_CHANGED;
 226                break;
 227        case S_IFLNK:
 228                if (!S_ISLNK(st->st_mode) &&
 229                    (has_symlinks || !S_ISREG(st->st_mode)))
 230                        changed |= TYPE_CHANGED;
 231                break;
 232        case S_IFGITLINK:
 233                if (!S_ISDIR(st->st_mode))
 234                        changed |= TYPE_CHANGED;
 235                else if (ce_compare_gitlink(ce))
 236                        changed |= DATA_CHANGED;
 237                return changed;
 238        default:
 239                die("internal error: ce_mode is %o", ce->ce_mode);
 240        }
 241        if (ce->ce_mtime != (unsigned int) st->st_mtime)
 242                changed |= MTIME_CHANGED;
 243        if (ce->ce_ctime != (unsigned int) st->st_ctime)
 244                changed |= CTIME_CHANGED;
 245
 246        if (ce->ce_uid != (unsigned int) st->st_uid ||
 247            ce->ce_gid != (unsigned int) st->st_gid)
 248                changed |= OWNER_CHANGED;
 249        if (ce->ce_ino != (unsigned int) st->st_ino)
 250                changed |= INODE_CHANGED;
 251
 252#ifdef USE_STDEV
 253        /*
 254         * st_dev breaks on network filesystems where different
 255         * clients will have different views of what "device"
 256         * the filesystem is on
 257         */
 258        if (ce->ce_dev != (unsigned int) st->st_dev)
 259                changed |= INODE_CHANGED;
 260#endif
 261
 262        if (ce->ce_size != (unsigned int) st->st_size)
 263                changed |= DATA_CHANGED;
 264
 265        return changed;
 266}
 267
 268static int is_racy_timestamp(struct index_state *istate, struct cache_entry *ce)
 269{
 270        return (istate->timestamp &&
 271                ((unsigned int)istate->timestamp) <= ce->ce_mtime);
 272}
 273
 274int ie_match_stat(struct index_state *istate,
 275                  struct cache_entry *ce, struct stat *st,
 276                  unsigned int options)
 277{
 278        unsigned int changed;
 279        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 280        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 281
 282        /*
 283         * If it's marked as always valid in the index, it's
 284         * valid whatever the checked-out copy says.
 285         */
 286        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 287                return 0;
 288
 289        changed = ce_match_stat_basic(ce, st);
 290
 291        /*
 292         * Within 1 second of this sequence:
 293         *      echo xyzzy >file && git-update-index --add file
 294         * running this command:
 295         *      echo frotz >file
 296         * would give a falsely clean cache entry.  The mtime and
 297         * length match the cache, and other stat fields do not change.
 298         *
 299         * We could detect this at update-index time (the cache entry
 300         * being registered/updated records the same time as "now")
 301         * and delay the return from git-update-index, but that would
 302         * effectively mean we can make at most one commit per second,
 303         * which is not acceptable.  Instead, we check cache entries
 304         * whose mtime are the same as the index file timestamp more
 305         * carefully than others.
 306         */
 307        if (!changed && is_racy_timestamp(istate, ce)) {
 308                if (assume_racy_is_modified)
 309                        changed |= DATA_CHANGED;
 310                else
 311                        changed |= ce_modified_check_fs(ce, st);
 312        }
 313
 314        return changed;
 315}
 316
 317int ie_modified(struct index_state *istate,
 318                struct cache_entry *ce, struct stat *st, unsigned int options)
 319{
 320        int changed, changed_fs;
 321
 322        changed = ie_match_stat(istate, ce, st, options);
 323        if (!changed)
 324                return 0;
 325        /*
 326         * If the mode or type has changed, there's no point in trying
 327         * to refresh the entry - it's not going to match
 328         */
 329        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 330                return changed;
 331
 332        /* Immediately after read-tree or update-index --cacheinfo,
 333         * the length field is zero.  For other cases the ce_size
 334         * should match the SHA1 recorded in the index entry.
 335         */
 336        if ((changed & DATA_CHANGED) && ce->ce_size != 0)
 337                return changed;
 338
 339        changed_fs = ce_modified_check_fs(ce, st);
 340        if (changed_fs)
 341                return changed | changed_fs;
 342        return 0;
 343}
 344
 345int base_name_compare(const char *name1, int len1, int mode1,
 346                      const char *name2, int len2, int mode2)
 347{
 348        unsigned char c1, c2;
 349        int len = len1 < len2 ? len1 : len2;
 350        int cmp;
 351
 352        cmp = memcmp(name1, name2, len);
 353        if (cmp)
 354                return cmp;
 355        c1 = name1[len];
 356        c2 = name2[len];
 357        if (!c1 && S_ISDIR(mode1))
 358                c1 = '/';
 359        if (!c2 && S_ISDIR(mode2))
 360                c2 = '/';
 361        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 362}
 363
 364int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 365{
 366        int len1 = flags1 & CE_NAMEMASK;
 367        int len2 = flags2 & CE_NAMEMASK;
 368        int len = len1 < len2 ? len1 : len2;
 369        int cmp;
 370
 371        cmp = memcmp(name1, name2, len);
 372        if (cmp)
 373                return cmp;
 374        if (len1 < len2)
 375                return -1;
 376        if (len1 > len2)
 377                return 1;
 378
 379        /* Compare stages  */
 380        flags1 &= CE_STAGEMASK;
 381        flags2 &= CE_STAGEMASK;
 382
 383        if (flags1 < flags2)
 384                return -1;
 385        if (flags1 > flags2)
 386                return 1;
 387        return 0;
 388}
 389
 390int index_name_pos(struct index_state *istate, const char *name, int namelen)
 391{
 392        int first, last;
 393
 394        first = 0;
 395        last = istate->cache_nr;
 396        while (last > first) {
 397                int next = (last + first) >> 1;
 398                struct cache_entry *ce = istate->cache[next];
 399                int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
 400                if (!cmp)
 401                        return next;
 402                if (cmp < 0) {
 403                        last = next;
 404                        continue;
 405                }
 406                first = next+1;
 407        }
 408        return -first-1;
 409}
 410
 411/* Remove entry, return true if there are more entries to go.. */
 412int remove_index_entry_at(struct index_state *istate, int pos)
 413{
 414        struct cache_entry *ce = istate->cache[pos];
 415
 416        remove_hash_entry(istate, ce);
 417        istate->cache_changed = 1;
 418        istate->cache_nr--;
 419        if (pos >= istate->cache_nr)
 420                return 0;
 421        memmove(istate->cache + pos,
 422                istate->cache + pos + 1,
 423                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 424        return 1;
 425}
 426
 427int remove_file_from_index(struct index_state *istate, const char *path)
 428{
 429        int pos = index_name_pos(istate, path, strlen(path));
 430        if (pos < 0)
 431                pos = -pos-1;
 432        cache_tree_invalidate_path(istate->cache_tree, path);
 433        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 434                remove_index_entry_at(istate, pos);
 435        return 0;
 436}
 437
 438static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 439{
 440        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 441}
 442
 443static int index_name_pos_also_unmerged(struct index_state *istate,
 444        const char *path, int namelen)
 445{
 446        int pos = index_name_pos(istate, path, namelen);
 447        struct cache_entry *ce;
 448
 449        if (pos >= 0)
 450                return pos;
 451
 452        /* maybe unmerged? */
 453        pos = -1 - pos;
 454        if (pos >= istate->cache_nr ||
 455                        compare_name((ce = istate->cache[pos]), path, namelen))
 456                return -1;
 457
 458        /* order of preference: stage 2, 1, 3 */
 459        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 460                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 461                        !compare_name(ce, path, namelen))
 462                pos++;
 463        return pos;
 464}
 465
 466int add_file_to_index(struct index_state *istate, const char *path, int verbose)
 467{
 468        int size, namelen, pos;
 469        struct stat st;
 470        struct cache_entry *ce;
 471        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
 472
 473        if (lstat(path, &st))
 474                die("%s: unable to stat (%s)", path, strerror(errno));
 475
 476        if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
 477                die("%s: can only add regular files, symbolic links or git-directories", path);
 478
 479        namelen = strlen(path);
 480        if (S_ISDIR(st.st_mode)) {
 481                while (namelen && path[namelen-1] == '/')
 482                        namelen--;
 483        }
 484        size = cache_entry_size(namelen);
 485        ce = xcalloc(1, size);
 486        memcpy(ce->name, path, namelen);
 487        ce->ce_flags = namelen;
 488        fill_stat_cache_info(ce, &st);
 489
 490        if (trust_executable_bit && has_symlinks)
 491                ce->ce_mode = create_ce_mode(st.st_mode);
 492        else {
 493                /* If there is an existing entry, pick the mode bits and type
 494                 * from it, otherwise assume unexecutable regular file.
 495                 */
 496                struct cache_entry *ent;
 497                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 498
 499                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 500                ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
 501        }
 502
 503        pos = index_name_pos(istate, ce->name, namelen);
 504        if (0 <= pos &&
 505            !ce_stage(istate->cache[pos]) &&
 506            !ie_match_stat(istate, istate->cache[pos], &st, ce_option)) {
 507                /* Nothing changed, really */
 508                free(ce);
 509                ce_mark_uptodate(istate->cache[pos]);
 510                return 0;
 511        }
 512
 513        if (index_path(ce->sha1, path, &st, 1))
 514                die("unable to index file %s", path);
 515        if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
 516                die("unable to add %s to index",path);
 517        if (verbose)
 518                printf("add '%s'\n", path);
 519        return 0;
 520}
 521
 522struct cache_entry *make_cache_entry(unsigned int mode,
 523                const unsigned char *sha1, const char *path, int stage,
 524                int refresh)
 525{
 526        int size, len;
 527        struct cache_entry *ce;
 528
 529        if (!verify_path(path))
 530                return NULL;
 531
 532        len = strlen(path);
 533        size = cache_entry_size(len);
 534        ce = xcalloc(1, size);
 535
 536        hashcpy(ce->sha1, sha1);
 537        memcpy(ce->name, path, len);
 538        ce->ce_flags = create_ce_flags(len, stage);
 539        ce->ce_mode = create_ce_mode(mode);
 540
 541        if (refresh)
 542                return refresh_cache_entry(ce, 0);
 543
 544        return ce;
 545}
 546
 547int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 548{
 549        int len = ce_namelen(a);
 550        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 551}
 552
 553int ce_path_match(const struct cache_entry *ce, const char **pathspec)
 554{
 555        const char *match, *name;
 556        int len;
 557
 558        if (!pathspec)
 559                return 1;
 560
 561        len = ce_namelen(ce);
 562        name = ce->name;
 563        while ((match = *pathspec++) != NULL) {
 564                int matchlen = strlen(match);
 565                if (matchlen > len)
 566                        continue;
 567                if (memcmp(name, match, matchlen))
 568                        continue;
 569                if (matchlen && name[matchlen-1] == '/')
 570                        return 1;
 571                if (name[matchlen] == '/' || !name[matchlen])
 572                        return 1;
 573                if (!matchlen)
 574                        return 1;
 575        }
 576        return 0;
 577}
 578
 579/*
 580 * We fundamentally don't like some paths: we don't want
 581 * dot or dot-dot anywhere, and for obvious reasons don't
 582 * want to recurse into ".git" either.
 583 *
 584 * Also, we don't want double slashes or slashes at the
 585 * end that can make pathnames ambiguous.
 586 */
 587static int verify_dotfile(const char *rest)
 588{
 589        /*
 590         * The first character was '.', but that
 591         * has already been discarded, we now test
 592         * the rest.
 593         */
 594        switch (*rest) {
 595        /* "." is not allowed */
 596        case '\0': case '/':
 597                return 0;
 598
 599        /*
 600         * ".git" followed by  NUL or slash is bad. This
 601         * shares the path end test with the ".." case.
 602         */
 603        case 'g':
 604                if (rest[1] != 'i')
 605                        break;
 606                if (rest[2] != 't')
 607                        break;
 608                rest += 2;
 609        /* fallthrough */
 610        case '.':
 611                if (rest[1] == '\0' || rest[1] == '/')
 612                        return 0;
 613        }
 614        return 1;
 615}
 616
 617int verify_path(const char *path)
 618{
 619        char c;
 620
 621        goto inside;
 622        for (;;) {
 623                if (!c)
 624                        return 1;
 625                if (c == '/') {
 626inside:
 627                        c = *path++;
 628                        switch (c) {
 629                        default:
 630                                continue;
 631                        case '/': case '\0':
 632                                break;
 633                        case '.':
 634                                if (verify_dotfile(path))
 635                                        continue;
 636                        }
 637                        return 0;
 638                }
 639                c = *path++;
 640        }
 641}
 642
 643/*
 644 * Do we have another file that has the beginning components being a
 645 * proper superset of the name we're trying to add?
 646 */
 647static int has_file_name(struct index_state *istate,
 648                         const struct cache_entry *ce, int pos, int ok_to_replace)
 649{
 650        int retval = 0;
 651        int len = ce_namelen(ce);
 652        int stage = ce_stage(ce);
 653        const char *name = ce->name;
 654
 655        while (pos < istate->cache_nr) {
 656                struct cache_entry *p = istate->cache[pos++];
 657
 658                if (len >= ce_namelen(p))
 659                        break;
 660                if (memcmp(name, p->name, len))
 661                        break;
 662                if (ce_stage(p) != stage)
 663                        continue;
 664                if (p->name[len] != '/')
 665                        continue;
 666                if (p->ce_flags & CE_REMOVE)
 667                        continue;
 668                retval = -1;
 669                if (!ok_to_replace)
 670                        break;
 671                remove_index_entry_at(istate, --pos);
 672        }
 673        return retval;
 674}
 675
 676/*
 677 * Do we have another file with a pathname that is a proper
 678 * subset of the name we're trying to add?
 679 */
 680static int has_dir_name(struct index_state *istate,
 681                        const struct cache_entry *ce, int pos, int ok_to_replace)
 682{
 683        int retval = 0;
 684        int stage = ce_stage(ce);
 685        const char *name = ce->name;
 686        const char *slash = name + ce_namelen(ce);
 687
 688        for (;;) {
 689                int len;
 690
 691                for (;;) {
 692                        if (*--slash == '/')
 693                                break;
 694                        if (slash <= ce->name)
 695                                return retval;
 696                }
 697                len = slash - name;
 698
 699                pos = index_name_pos(istate, name, create_ce_flags(len, stage));
 700                if (pos >= 0) {
 701                        /*
 702                         * Found one, but not so fast.  This could
 703                         * be a marker that says "I was here, but
 704                         * I am being removed".  Such an entry is
 705                         * not a part of the resulting tree, and
 706                         * it is Ok to have a directory at the same
 707                         * path.
 708                         */
 709                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 710                                retval = -1;
 711                                if (!ok_to_replace)
 712                                        break;
 713                                remove_index_entry_at(istate, pos);
 714                                continue;
 715                        }
 716                }
 717                else
 718                        pos = -pos-1;
 719
 720                /*
 721                 * Trivial optimization: if we find an entry that
 722                 * already matches the sub-directory, then we know
 723                 * we're ok, and we can exit.
 724                 */
 725                while (pos < istate->cache_nr) {
 726                        struct cache_entry *p = istate->cache[pos];
 727                        if ((ce_namelen(p) <= len) ||
 728                            (p->name[len] != '/') ||
 729                            memcmp(p->name, name, len))
 730                                break; /* not our subdirectory */
 731                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 732                                /*
 733                                 * p is at the same stage as our entry, and
 734                                 * is a subdirectory of what we are looking
 735                                 * at, so we cannot have conflicts at our
 736                                 * level or anything shorter.
 737                                 */
 738                                return retval;
 739                        pos++;
 740                }
 741        }
 742        return retval;
 743}
 744
 745/* We may be in a situation where we already have path/file and path
 746 * is being added, or we already have path and path/file is being
 747 * added.  Either one would result in a nonsense tree that has path
 748 * twice when git-write-tree tries to write it out.  Prevent it.
 749 *
 750 * If ok-to-replace is specified, we remove the conflicting entries
 751 * from the cache so the caller should recompute the insert position.
 752 * When this happens, we return non-zero.
 753 */
 754static int check_file_directory_conflict(struct index_state *istate,
 755                                         const struct cache_entry *ce,
 756                                         int pos, int ok_to_replace)
 757{
 758        int retval;
 759
 760        /*
 761         * When ce is an "I am going away" entry, we allow it to be added
 762         */
 763        if (ce->ce_flags & CE_REMOVE)
 764                return 0;
 765
 766        /*
 767         * We check if the path is a sub-path of a subsequent pathname
 768         * first, since removing those will not change the position
 769         * in the array.
 770         */
 771        retval = has_file_name(istate, ce, pos, ok_to_replace);
 772
 773        /*
 774         * Then check if the path might have a clashing sub-directory
 775         * before it.
 776         */
 777        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 778}
 779
 780static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 781{
 782        int pos;
 783        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 784        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 785        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 786
 787        cache_tree_invalidate_path(istate->cache_tree, ce->name);
 788        pos = index_name_pos(istate, ce->name, ce->ce_flags);
 789
 790        /* existing match? Just replace it. */
 791        if (pos >= 0) {
 792                replace_index_entry(istate, pos, ce);
 793                return 0;
 794        }
 795        pos = -pos-1;
 796
 797        /*
 798         * Inserting a merged entry ("stage 0") into the index
 799         * will always replace all non-merged entries..
 800         */
 801        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 802                while (ce_same_name(istate->cache[pos], ce)) {
 803                        ok_to_add = 1;
 804                        if (!remove_index_entry_at(istate, pos))
 805                                break;
 806                }
 807        }
 808
 809        if (!ok_to_add)
 810                return -1;
 811        if (!verify_path(ce->name))
 812                return -1;
 813
 814        if (!skip_df_check &&
 815            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 816                if (!ok_to_replace)
 817                        return error("'%s' appears as both a file and as a directory",
 818                                     ce->name);
 819                pos = index_name_pos(istate, ce->name, ce->ce_flags);
 820                pos = -pos-1;
 821        }
 822        return pos + 1;
 823}
 824
 825int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 826{
 827        int pos;
 828
 829        if (option & ADD_CACHE_JUST_APPEND)
 830                pos = istate->cache_nr;
 831        else {
 832                int ret;
 833                ret = add_index_entry_with_check(istate, ce, option);
 834                if (ret <= 0)
 835                        return ret;
 836                pos = ret - 1;
 837        }
 838
 839        /* Make sure the array is big enough .. */
 840        if (istate->cache_nr == istate->cache_alloc) {
 841                istate->cache_alloc = alloc_nr(istate->cache_alloc);
 842                istate->cache = xrealloc(istate->cache,
 843                                        istate->cache_alloc * sizeof(struct cache_entry *));
 844        }
 845
 846        /* Add it in.. */
 847        istate->cache_nr++;
 848        if (istate->cache_nr > pos + 1)
 849                memmove(istate->cache + pos + 1,
 850                        istate->cache + pos,
 851                        (istate->cache_nr - pos - 1) * sizeof(ce));
 852        set_index_entry(istate, pos, ce);
 853        istate->cache_changed = 1;
 854        return 0;
 855}
 856
 857/*
 858 * "refresh" does not calculate a new sha1 file or bring the
 859 * cache up-to-date for mode/content changes. But what it
 860 * _does_ do is to "re-match" the stat information of a file
 861 * with the cache, so that you can refresh the cache for a
 862 * file that hasn't been changed but where the stat entry is
 863 * out of date.
 864 *
 865 * For example, you'd want to do this after doing a "git-read-tree",
 866 * to link up the stat cache details with the proper files.
 867 */
 868static struct cache_entry *refresh_cache_ent(struct index_state *istate,
 869                                             struct cache_entry *ce,
 870                                             unsigned int options, int *err)
 871{
 872        struct stat st;
 873        struct cache_entry *updated;
 874        int changed, size;
 875        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 876
 877        if (ce_uptodate(ce))
 878                return ce;
 879
 880        if (lstat(ce->name, &st) < 0) {
 881                if (err)
 882                        *err = errno;
 883                return NULL;
 884        }
 885
 886        changed = ie_match_stat(istate, ce, &st, options);
 887        if (!changed) {
 888                /*
 889                 * The path is unchanged.  If we were told to ignore
 890                 * valid bit, then we did the actual stat check and
 891                 * found that the entry is unmodified.  If the entry
 892                 * is not marked VALID, this is the place to mark it
 893                 * valid again, under "assume unchanged" mode.
 894                 */
 895                if (ignore_valid && assume_unchanged &&
 896                    !(ce->ce_flags & CE_VALID))
 897                        ; /* mark this one VALID again */
 898                else {
 899                        /*
 900                         * We do not mark the index itself "modified"
 901                         * because CE_UPTODATE flag is in-core only;
 902                         * we are not going to write this change out.
 903                         */
 904                        ce_mark_uptodate(ce);
 905                        return ce;
 906                }
 907        }
 908
 909        if (ie_modified(istate, ce, &st, options)) {
 910                if (err)
 911                        *err = EINVAL;
 912                return NULL;
 913        }
 914
 915        size = ce_size(ce);
 916        updated = xmalloc(size);
 917        memcpy(updated, ce, size);
 918        fill_stat_cache_info(updated, &st);
 919        /*
 920         * If ignore_valid is not set, we should leave CE_VALID bit
 921         * alone.  Otherwise, paths marked with --no-assume-unchanged
 922         * (i.e. things to be edited) will reacquire CE_VALID bit
 923         * automatically, which is not really what we want.
 924         */
 925        if (!ignore_valid && assume_unchanged &&
 926            !(ce->ce_flags & CE_VALID))
 927                updated->ce_flags &= ~CE_VALID;
 928
 929        return updated;
 930}
 931
 932int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
 933{
 934        int i;
 935        int has_errors = 0;
 936        int really = (flags & REFRESH_REALLY) != 0;
 937        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
 938        int quiet = (flags & REFRESH_QUIET) != 0;
 939        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
 940        unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
 941
 942        for (i = 0; i < istate->cache_nr; i++) {
 943                struct cache_entry *ce, *new;
 944                int cache_errno = 0;
 945
 946                ce = istate->cache[i];
 947                if (ce_stage(ce)) {
 948                        while ((i < istate->cache_nr) &&
 949                               ! strcmp(istate->cache[i]->name, ce->name))
 950                                i++;
 951                        i--;
 952                        if (allow_unmerged)
 953                                continue;
 954                        printf("%s: needs merge\n", ce->name);
 955                        has_errors = 1;
 956                        continue;
 957                }
 958
 959                if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
 960                        continue;
 961
 962                new = refresh_cache_ent(istate, ce, options, &cache_errno);
 963                if (new == ce)
 964                        continue;
 965                if (!new) {
 966                        if (not_new && cache_errno == ENOENT)
 967                                continue;
 968                        if (really && cache_errno == EINVAL) {
 969                                /* If we are doing --really-refresh that
 970                                 * means the index is not valid anymore.
 971                                 */
 972                                ce->ce_flags &= ~CE_VALID;
 973                                istate->cache_changed = 1;
 974                        }
 975                        if (quiet)
 976                                continue;
 977                        printf("%s: needs update\n", ce->name);
 978                        has_errors = 1;
 979                        continue;
 980                }
 981
 982                replace_index_entry(istate, i, new);
 983        }
 984        return has_errors;
 985}
 986
 987struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
 988{
 989        return refresh_cache_ent(&the_index, ce, really, NULL);
 990}
 991
 992static int verify_hdr(struct cache_header *hdr, unsigned long size)
 993{
 994        SHA_CTX c;
 995        unsigned char sha1[20];
 996
 997        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
 998                return error("bad signature");
 999        if (hdr->hdr_version != htonl(2))
1000                return error("bad index version");
1001        SHA1_Init(&c);
1002        SHA1_Update(&c, hdr, size - 20);
1003        SHA1_Final(sha1, &c);
1004        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1005                return error("bad index file sha1 signature");
1006        return 0;
1007}
1008
1009static int read_index_extension(struct index_state *istate,
1010                                const char *ext, void *data, unsigned long sz)
1011{
1012        switch (CACHE_EXT(ext)) {
1013        case CACHE_EXT_TREE:
1014                istate->cache_tree = cache_tree_read(data, sz);
1015                break;
1016        default:
1017                if (*ext < 'A' || 'Z' < *ext)
1018                        return error("index uses %.4s extension, which we do not understand",
1019                                     ext);
1020                fprintf(stderr, "ignoring %.4s extension\n", ext);
1021                break;
1022        }
1023        return 0;
1024}
1025
1026int read_index(struct index_state *istate)
1027{
1028        return read_index_from(istate, get_index_file());
1029}
1030
1031static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1032{
1033        size_t len;
1034
1035        ce->ce_ctime = ntohl(ondisk->ctime.sec);
1036        ce->ce_mtime = ntohl(ondisk->mtime.sec);
1037        ce->ce_dev   = ntohl(ondisk->dev);
1038        ce->ce_ino   = ntohl(ondisk->ino);
1039        ce->ce_mode  = ntohl(ondisk->mode);
1040        ce->ce_uid   = ntohl(ondisk->uid);
1041        ce->ce_gid   = ntohl(ondisk->gid);
1042        ce->ce_size  = ntohl(ondisk->size);
1043        /* On-disk flags are just 16 bits */
1044        ce->ce_flags = ntohs(ondisk->flags);
1045        hashcpy(ce->sha1, ondisk->sha1);
1046
1047        len = ce->ce_flags & CE_NAMEMASK;
1048        if (len == CE_NAMEMASK)
1049                len = strlen(ondisk->name);
1050        /*
1051         * NEEDSWORK: If the original index is crafted, this copy could
1052         * go unchecked.
1053         */
1054        memcpy(ce->name, ondisk->name, len + 1);
1055}
1056
1057static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1058{
1059        long per_entry;
1060
1061        per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1062
1063        /*
1064         * Alignment can cause differences. This should be "alignof", but
1065         * since that's a gcc'ism, just use the size of a pointer.
1066         */
1067        per_entry += sizeof(void *);
1068        return ondisk_size + entries*per_entry;
1069}
1070
1071/* remember to discard_cache() before reading a different cache! */
1072int read_index_from(struct index_state *istate, const char *path)
1073{
1074        int fd, i;
1075        struct stat st;
1076        unsigned long src_offset, dst_offset;
1077        struct cache_header *hdr;
1078        void *mmap;
1079        size_t mmap_size;
1080
1081        errno = EBUSY;
1082        if (istate->alloc)
1083                return istate->cache_nr;
1084
1085        errno = ENOENT;
1086        istate->timestamp = 0;
1087        fd = open(path, O_RDONLY);
1088        if (fd < 0) {
1089                if (errno == ENOENT)
1090                        return 0;
1091                die("index file open failed (%s)", strerror(errno));
1092        }
1093
1094        if (fstat(fd, &st))
1095                die("cannot stat the open index (%s)", strerror(errno));
1096
1097        errno = EINVAL;
1098        mmap_size = xsize_t(st.st_size);
1099        if (mmap_size < sizeof(struct cache_header) + 20)
1100                die("index file smaller than expected");
1101
1102        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1103        close(fd);
1104        if (mmap == MAP_FAILED)
1105                die("unable to map index file");
1106
1107        hdr = mmap;
1108        if (verify_hdr(hdr, mmap_size) < 0)
1109                goto unmap;
1110
1111        istate->cache_nr = ntohl(hdr->hdr_entries);
1112        istate->cache_alloc = alloc_nr(istate->cache_nr);
1113        istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1114
1115        /*
1116         * The disk format is actually larger than the in-memory format,
1117         * due to space for nsec etc, so even though the in-memory one
1118         * has room for a few  more flags, we can allocate using the same
1119         * index size
1120         */
1121        istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1122
1123        src_offset = sizeof(*hdr);
1124        dst_offset = 0;
1125        for (i = 0; i < istate->cache_nr; i++) {
1126                struct ondisk_cache_entry *disk_ce;
1127                struct cache_entry *ce;
1128
1129                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1130                ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1131                convert_from_disk(disk_ce, ce);
1132                set_index_entry(istate, i, ce);
1133
1134                src_offset += ondisk_ce_size(ce);
1135                dst_offset += ce_size(ce);
1136        }
1137        istate->timestamp = st.st_mtime;
1138        while (src_offset <= mmap_size - 20 - 8) {
1139                /* After an array of active_nr index entries,
1140                 * there can be arbitrary number of extended
1141                 * sections, each of which is prefixed with
1142                 * extension name (4-byte) and section length
1143                 * in 4-byte network byte order.
1144                 */
1145                unsigned long extsize;
1146                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1147                extsize = ntohl(extsize);
1148                if (read_index_extension(istate,
1149                                         (const char *) mmap + src_offset,
1150                                         (char *) mmap + src_offset + 8,
1151                                         extsize) < 0)
1152                        goto unmap;
1153                src_offset += 8;
1154                src_offset += extsize;
1155        }
1156        munmap(mmap, mmap_size);
1157        return istate->cache_nr;
1158
1159unmap:
1160        munmap(mmap, mmap_size);
1161        errno = EINVAL;
1162        die("index file corrupt");
1163}
1164
1165int discard_index(struct index_state *istate)
1166{
1167        istate->cache_nr = 0;
1168        istate->cache_changed = 0;
1169        istate->timestamp = 0;
1170        free_hash(&istate->name_hash);
1171        cache_tree_free(&(istate->cache_tree));
1172        free(istate->alloc);
1173        istate->alloc = NULL;
1174
1175        /* no need to throw away allocated active_cache */
1176        return 0;
1177}
1178
1179#define WRITE_BUFFER_SIZE 8192
1180static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1181static unsigned long write_buffer_len;
1182
1183static int ce_write_flush(SHA_CTX *context, int fd)
1184{
1185        unsigned int buffered = write_buffer_len;
1186        if (buffered) {
1187                SHA1_Update(context, write_buffer, buffered);
1188                if (write_in_full(fd, write_buffer, buffered) != buffered)
1189                        return -1;
1190                write_buffer_len = 0;
1191        }
1192        return 0;
1193}
1194
1195static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1196{
1197        while (len) {
1198                unsigned int buffered = write_buffer_len;
1199                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1200                if (partial > len)
1201                        partial = len;
1202                memcpy(write_buffer + buffered, data, partial);
1203                buffered += partial;
1204                if (buffered == WRITE_BUFFER_SIZE) {
1205                        write_buffer_len = buffered;
1206                        if (ce_write_flush(context, fd))
1207                                return -1;
1208                        buffered = 0;
1209                }
1210                write_buffer_len = buffered;
1211                len -= partial;
1212                data = (char *) data + partial;
1213        }
1214        return 0;
1215}
1216
1217static int write_index_ext_header(SHA_CTX *context, int fd,
1218                                  unsigned int ext, unsigned int sz)
1219{
1220        ext = htonl(ext);
1221        sz = htonl(sz);
1222        return ((ce_write(context, fd, &ext, 4) < 0) ||
1223                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1224}
1225
1226static int ce_flush(SHA_CTX *context, int fd)
1227{
1228        unsigned int left = write_buffer_len;
1229
1230        if (left) {
1231                write_buffer_len = 0;
1232                SHA1_Update(context, write_buffer, left);
1233        }
1234
1235        /* Flush first if not enough space for SHA1 signature */
1236        if (left + 20 > WRITE_BUFFER_SIZE) {
1237                if (write_in_full(fd, write_buffer, left) != left)
1238                        return -1;
1239                left = 0;
1240        }
1241
1242        /* Append the SHA1 signature at the end */
1243        SHA1_Final(write_buffer + left, context);
1244        left += 20;
1245        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1246}
1247
1248static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1249{
1250        /*
1251         * The only thing we care about in this function is to smudge the
1252         * falsely clean entry due to touch-update-touch race, so we leave
1253         * everything else as they are.  We are called for entries whose
1254         * ce_mtime match the index file mtime.
1255         */
1256        struct stat st;
1257
1258        if (lstat(ce->name, &st) < 0)
1259                return;
1260        if (ce_match_stat_basic(ce, &st))
1261                return;
1262        if (ce_modified_check_fs(ce, &st)) {
1263                /* This is "racily clean"; smudge it.  Note that this
1264                 * is a tricky code.  At first glance, it may appear
1265                 * that it can break with this sequence:
1266                 *
1267                 * $ echo xyzzy >frotz
1268                 * $ git-update-index --add frotz
1269                 * $ : >frotz
1270                 * $ sleep 3
1271                 * $ echo filfre >nitfol
1272                 * $ git-update-index --add nitfol
1273                 *
1274                 * but it does not.  When the second update-index runs,
1275                 * it notices that the entry "frotz" has the same timestamp
1276                 * as index, and if we were to smudge it by resetting its
1277                 * size to zero here, then the object name recorded
1278                 * in index is the 6-byte file but the cached stat information
1279                 * becomes zero --- which would then match what we would
1280                 * obtain from the filesystem next time we stat("frotz").
1281                 *
1282                 * However, the second update-index, before calling
1283                 * this function, notices that the cached size is 6
1284                 * bytes and what is on the filesystem is an empty
1285                 * file, and never calls us, so the cached size information
1286                 * for "frotz" stays 6 which does not match the filesystem.
1287                 */
1288                ce->ce_size = 0;
1289        }
1290}
1291
1292static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1293{
1294        int size = ondisk_ce_size(ce);
1295        struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1296
1297        ondisk->ctime.sec = htonl(ce->ce_ctime);
1298        ondisk->ctime.nsec = 0;
1299        ondisk->mtime.sec = htonl(ce->ce_mtime);
1300        ondisk->mtime.nsec = 0;
1301        ondisk->dev  = htonl(ce->ce_dev);
1302        ondisk->ino  = htonl(ce->ce_ino);
1303        ondisk->mode = htonl(ce->ce_mode);
1304        ondisk->uid  = htonl(ce->ce_uid);
1305        ondisk->gid  = htonl(ce->ce_gid);
1306        ondisk->size = htonl(ce->ce_size);
1307        hashcpy(ondisk->sha1, ce->sha1);
1308        ondisk->flags = htons(ce->ce_flags);
1309        memcpy(ondisk->name, ce->name, ce_namelen(ce));
1310
1311        return ce_write(c, fd, ondisk, size);
1312}
1313
1314int write_index(struct index_state *istate, int newfd)
1315{
1316        SHA_CTX c;
1317        struct cache_header hdr;
1318        int i, err, removed;
1319        struct cache_entry **cache = istate->cache;
1320        int entries = istate->cache_nr;
1321
1322        for (i = removed = 0; i < entries; i++)
1323                if (cache[i]->ce_flags & CE_REMOVE)
1324                        removed++;
1325
1326        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1327        hdr.hdr_version = htonl(2);
1328        hdr.hdr_entries = htonl(entries - removed);
1329
1330        SHA1_Init(&c);
1331        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1332                return -1;
1333
1334        for (i = 0; i < entries; i++) {
1335                struct cache_entry *ce = cache[i];
1336                if (ce->ce_flags & CE_REMOVE)
1337                        continue;
1338                if (is_racy_timestamp(istate, ce))
1339                        ce_smudge_racily_clean_entry(ce);
1340                if (ce_write_entry(&c, newfd, ce) < 0)
1341                        return -1;
1342        }
1343
1344        /* Write extension data here */
1345        if (istate->cache_tree) {
1346                struct strbuf sb;
1347
1348                strbuf_init(&sb, 0);
1349                cache_tree_write(&sb, istate->cache_tree);
1350                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1351                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1352                strbuf_release(&sb);
1353                if (err)
1354                        return -1;
1355        }
1356        return ce_flush(&c, newfd);
1357}