eb58b03f952306a23e2458d856066e90532f0f4d
   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;
  41
  42        if (ce->ce_flags & CE_HASHED)
  43                return;
  44        ce->ce_flags |= CE_HASHED;
  45        ce->next = NULL;
  46        hash = hash_name(ce->name, ce_namelen(ce));
  47        pos = insert_hash(hash, ce, &istate->name_hash);
  48        if (pos) {
  49                ce->next = *pos;
  50                *pos = ce;
  51        }
  52}
  53
  54static void lazy_init_name_hash(struct index_state *istate)
  55{
  56        int nr;
  57
  58        if (istate->name_hash_initialized)
  59                return;
  60        for (nr = 0; nr < istate->cache_nr; nr++)
  61                hash_index_entry(istate, istate->cache[nr]);
  62        istate->name_hash_initialized = 1;
  63}
  64
  65static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  66{
  67        ce->ce_flags &= ~CE_UNHASHED;
  68        istate->cache[nr] = ce;
  69        if (istate->name_hash_initialized)
  70                hash_index_entry(istate, ce);
  71}
  72
  73/*
  74 * We don't actually *remove* it, we can just mark it invalid so that
  75 * we won't find it in lookups.
  76 *
  77 * Not only would we have to search the lists (simple enough), but
  78 * we'd also have to rehash other hash buckets in case this makes the
  79 * hash bucket empty (common). So it's much better to just mark
  80 * it.
  81 */
  82static void remove_hash_entry(struct index_state *istate, struct cache_entry *ce)
  83{
  84        ce->ce_flags |= CE_UNHASHED;
  85}
  86
  87static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  88{
  89        struct cache_entry *old = istate->cache[nr];
  90
  91        remove_hash_entry(istate, old);
  92        set_index_entry(istate, nr, ce);
  93        istate->cache_changed = 1;
  94}
  95
  96int index_name_exists(struct index_state *istate, const char *name, int namelen)
  97{
  98        unsigned int hash = hash_name(name, namelen);
  99        struct cache_entry *ce;
 100
 101        lazy_init_name_hash(istate);
 102        ce = lookup_hash(hash, &istate->name_hash);
 103
 104        while (ce) {
 105                if (!(ce->ce_flags & CE_UNHASHED)) {
 106                        if (!cache_name_compare(name, namelen, ce->name, ce->ce_flags))
 107                                return 1;
 108                }
 109                ce = ce->next;
 110        }
 111        return 0;
 112}
 113
 114/*
 115 * This only updates the "non-critical" parts of the directory
 116 * cache, ie the parts that aren't tracked by GIT, and only used
 117 * to validate the cache.
 118 */
 119void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 120{
 121        ce->ce_ctime = st->st_ctime;
 122        ce->ce_mtime = st->st_mtime;
 123        ce->ce_dev = st->st_dev;
 124        ce->ce_ino = st->st_ino;
 125        ce->ce_uid = st->st_uid;
 126        ce->ce_gid = st->st_gid;
 127        ce->ce_size = st->st_size;
 128
 129        if (assume_unchanged)
 130                ce->ce_flags |= CE_VALID;
 131
 132        if (S_ISREG(st->st_mode))
 133                ce_mark_uptodate(ce);
 134}
 135
 136static int ce_compare_data(struct cache_entry *ce, struct stat *st)
 137{
 138        int match = -1;
 139        int fd = open(ce->name, O_RDONLY);
 140
 141        if (fd >= 0) {
 142                unsigned char sha1[20];
 143                if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
 144                        match = hashcmp(sha1, ce->sha1);
 145                /* index_fd() closed the file descriptor already */
 146        }
 147        return match;
 148}
 149
 150static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
 151{
 152        int match = -1;
 153        char *target;
 154        void *buffer;
 155        unsigned long size;
 156        enum object_type type;
 157        int len;
 158
 159        target = xmalloc(expected_size);
 160        len = readlink(ce->name, target, expected_size);
 161        if (len != expected_size) {
 162                free(target);
 163                return -1;
 164        }
 165        buffer = read_sha1_file(ce->sha1, &type, &size);
 166        if (!buffer) {
 167                free(target);
 168                return -1;
 169        }
 170        if (size == expected_size)
 171                match = memcmp(buffer, target, size);
 172        free(buffer);
 173        free(target);
 174        return match;
 175}
 176
 177static int ce_compare_gitlink(struct cache_entry *ce)
 178{
 179        unsigned char sha1[20];
 180
 181        /*
 182         * We don't actually require that the .git directory
 183         * under GITLINK directory be a valid git directory. It
 184         * might even be missing (in case nobody populated that
 185         * sub-project).
 186         *
 187         * If so, we consider it always to match.
 188         */
 189        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 190                return 0;
 191        return hashcmp(sha1, ce->sha1);
 192}
 193
 194static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
 195{
 196        switch (st->st_mode & S_IFMT) {
 197        case S_IFREG:
 198                if (ce_compare_data(ce, st))
 199                        return DATA_CHANGED;
 200                break;
 201        case S_IFLNK:
 202                if (ce_compare_link(ce, xsize_t(st->st_size)))
 203                        return DATA_CHANGED;
 204                break;
 205        case S_IFDIR:
 206                if (S_ISGITLINK(ce->ce_mode))
 207                        return 0;
 208        default:
 209                return TYPE_CHANGED;
 210        }
 211        return 0;
 212}
 213
 214static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 215{
 216        unsigned int changed = 0;
 217
 218        if (ce->ce_flags & CE_REMOVE)
 219                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 220
 221        switch (ce->ce_mode & S_IFMT) {
 222        case S_IFREG:
 223                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 224                /* We consider only the owner x bit to be relevant for
 225                 * "mode changes"
 226                 */
 227                if (trust_executable_bit &&
 228                    (0100 & (ce->ce_mode ^ st->st_mode)))
 229                        changed |= MODE_CHANGED;
 230                break;
 231        case S_IFLNK:
 232                if (!S_ISLNK(st->st_mode) &&
 233                    (has_symlinks || !S_ISREG(st->st_mode)))
 234                        changed |= TYPE_CHANGED;
 235                break;
 236        case S_IFGITLINK:
 237                if (!S_ISDIR(st->st_mode))
 238                        changed |= TYPE_CHANGED;
 239                else if (ce_compare_gitlink(ce))
 240                        changed |= DATA_CHANGED;
 241                return changed;
 242        default:
 243                die("internal error: ce_mode is %o", ce->ce_mode);
 244        }
 245        if (ce->ce_mtime != (unsigned int) st->st_mtime)
 246                changed |= MTIME_CHANGED;
 247        if (ce->ce_ctime != (unsigned int) st->st_ctime)
 248                changed |= CTIME_CHANGED;
 249
 250        if (ce->ce_uid != (unsigned int) st->st_uid ||
 251            ce->ce_gid != (unsigned int) st->st_gid)
 252                changed |= OWNER_CHANGED;
 253        if (ce->ce_ino != (unsigned int) st->st_ino)
 254                changed |= INODE_CHANGED;
 255
 256#ifdef USE_STDEV
 257        /*
 258         * st_dev breaks on network filesystems where different
 259         * clients will have different views of what "device"
 260         * the filesystem is on
 261         */
 262        if (ce->ce_dev != (unsigned int) st->st_dev)
 263                changed |= INODE_CHANGED;
 264#endif
 265
 266        if (ce->ce_size != (unsigned int) st->st_size)
 267                changed |= DATA_CHANGED;
 268
 269        return changed;
 270}
 271
 272static int is_racy_timestamp(struct index_state *istate, struct cache_entry *ce)
 273{
 274        return (istate->timestamp &&
 275                ((unsigned int)istate->timestamp) <= ce->ce_mtime);
 276}
 277
 278int ie_match_stat(struct index_state *istate,
 279                  struct cache_entry *ce, struct stat *st,
 280                  unsigned int options)
 281{
 282        unsigned int changed;
 283        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 284        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 285
 286        /*
 287         * If it's marked as always valid in the index, it's
 288         * valid whatever the checked-out copy says.
 289         */
 290        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 291                return 0;
 292
 293        changed = ce_match_stat_basic(ce, st);
 294
 295        /*
 296         * Within 1 second of this sequence:
 297         *      echo xyzzy >file && git-update-index --add file
 298         * running this command:
 299         *      echo frotz >file
 300         * would give a falsely clean cache entry.  The mtime and
 301         * length match the cache, and other stat fields do not change.
 302         *
 303         * We could detect this at update-index time (the cache entry
 304         * being registered/updated records the same time as "now")
 305         * and delay the return from git-update-index, but that would
 306         * effectively mean we can make at most one commit per second,
 307         * which is not acceptable.  Instead, we check cache entries
 308         * whose mtime are the same as the index file timestamp more
 309         * carefully than others.
 310         */
 311        if (!changed && is_racy_timestamp(istate, ce)) {
 312                if (assume_racy_is_modified)
 313                        changed |= DATA_CHANGED;
 314                else
 315                        changed |= ce_modified_check_fs(ce, st);
 316        }
 317
 318        return changed;
 319}
 320
 321int ie_modified(struct index_state *istate,
 322                struct cache_entry *ce, struct stat *st, unsigned int options)
 323{
 324        int changed, changed_fs;
 325
 326        changed = ie_match_stat(istate, ce, st, options);
 327        if (!changed)
 328                return 0;
 329        /*
 330         * If the mode or type has changed, there's no point in trying
 331         * to refresh the entry - it's not going to match
 332         */
 333        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 334                return changed;
 335
 336        /* Immediately after read-tree or update-index --cacheinfo,
 337         * the length field is zero.  For other cases the ce_size
 338         * should match the SHA1 recorded in the index entry.
 339         */
 340        if ((changed & DATA_CHANGED) && ce->ce_size != 0)
 341                return changed;
 342
 343        changed_fs = ce_modified_check_fs(ce, st);
 344        if (changed_fs)
 345                return changed | changed_fs;
 346        return 0;
 347}
 348
 349int base_name_compare(const char *name1, int len1, int mode1,
 350                      const char *name2, int len2, int mode2)
 351{
 352        unsigned char c1, c2;
 353        int len = len1 < len2 ? len1 : len2;
 354        int cmp;
 355
 356        cmp = memcmp(name1, name2, len);
 357        if (cmp)
 358                return cmp;
 359        c1 = name1[len];
 360        c2 = name2[len];
 361        if (!c1 && S_ISDIR(mode1))
 362                c1 = '/';
 363        if (!c2 && S_ISDIR(mode2))
 364                c2 = '/';
 365        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 366}
 367
 368int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 369{
 370        int len1 = flags1 & CE_NAMEMASK;
 371        int len2 = flags2 & CE_NAMEMASK;
 372        int len = len1 < len2 ? len1 : len2;
 373        int cmp;
 374
 375        cmp = memcmp(name1, name2, len);
 376        if (cmp)
 377                return cmp;
 378        if (len1 < len2)
 379                return -1;
 380        if (len1 > len2)
 381                return 1;
 382
 383        /* Compare stages  */
 384        flags1 &= CE_STAGEMASK;
 385        flags2 &= CE_STAGEMASK;
 386
 387        if (flags1 < flags2)
 388                return -1;
 389        if (flags1 > flags2)
 390                return 1;
 391        return 0;
 392}
 393
 394int index_name_pos(struct index_state *istate, const char *name, int namelen)
 395{
 396        int first, last;
 397
 398        first = 0;
 399        last = istate->cache_nr;
 400        while (last > first) {
 401                int next = (last + first) >> 1;
 402                struct cache_entry *ce = istate->cache[next];
 403                int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
 404                if (!cmp)
 405                        return next;
 406                if (cmp < 0) {
 407                        last = next;
 408                        continue;
 409                }
 410                first = next+1;
 411        }
 412        return -first-1;
 413}
 414
 415/* Remove entry, return true if there are more entries to go.. */
 416int remove_index_entry_at(struct index_state *istate, int pos)
 417{
 418        struct cache_entry *ce = istate->cache[pos];
 419
 420        remove_hash_entry(istate, ce);
 421        istate->cache_changed = 1;
 422        istate->cache_nr--;
 423        if (pos >= istate->cache_nr)
 424                return 0;
 425        memmove(istate->cache + pos,
 426                istate->cache + pos + 1,
 427                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 428        return 1;
 429}
 430
 431int remove_file_from_index(struct index_state *istate, const char *path)
 432{
 433        int pos = index_name_pos(istate, path, strlen(path));
 434        if (pos < 0)
 435                pos = -pos-1;
 436        cache_tree_invalidate_path(istate->cache_tree, path);
 437        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 438                remove_index_entry_at(istate, pos);
 439        return 0;
 440}
 441
 442static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 443{
 444        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 445}
 446
 447static int index_name_pos_also_unmerged(struct index_state *istate,
 448        const char *path, int namelen)
 449{
 450        int pos = index_name_pos(istate, path, namelen);
 451        struct cache_entry *ce;
 452
 453        if (pos >= 0)
 454                return pos;
 455
 456        /* maybe unmerged? */
 457        pos = -1 - pos;
 458        if (pos >= istate->cache_nr ||
 459                        compare_name((ce = istate->cache[pos]), path, namelen))
 460                return -1;
 461
 462        /* order of preference: stage 2, 1, 3 */
 463        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 464                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 465                        !compare_name(ce, path, namelen))
 466                pos++;
 467        return pos;
 468}
 469
 470int add_file_to_index(struct index_state *istate, const char *path, int verbose)
 471{
 472        int size, namelen, pos;
 473        struct stat st;
 474        struct cache_entry *ce;
 475        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
 476
 477        if (lstat(path, &st))
 478                die("%s: unable to stat (%s)", path, strerror(errno));
 479
 480        if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
 481                die("%s: can only add regular files, symbolic links or git-directories", path);
 482
 483        namelen = strlen(path);
 484        if (S_ISDIR(st.st_mode)) {
 485                while (namelen && path[namelen-1] == '/')
 486                        namelen--;
 487        }
 488        size = cache_entry_size(namelen);
 489        ce = xcalloc(1, size);
 490        memcpy(ce->name, path, namelen);
 491        ce->ce_flags = namelen;
 492        fill_stat_cache_info(ce, &st);
 493
 494        if (trust_executable_bit && has_symlinks)
 495                ce->ce_mode = create_ce_mode(st.st_mode);
 496        else {
 497                /* If there is an existing entry, pick the mode bits and type
 498                 * from it, otherwise assume unexecutable regular file.
 499                 */
 500                struct cache_entry *ent;
 501                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 502
 503                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 504                ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
 505        }
 506
 507        pos = index_name_pos(istate, ce->name, namelen);
 508        if (0 <= pos &&
 509            !ce_stage(istate->cache[pos]) &&
 510            !ie_match_stat(istate, istate->cache[pos], &st, ce_option)) {
 511                /* Nothing changed, really */
 512                free(ce);
 513                ce_mark_uptodate(istate->cache[pos]);
 514                return 0;
 515        }
 516
 517        if (index_path(ce->sha1, path, &st, 1))
 518                die("unable to index file %s", path);
 519        if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
 520                die("unable to add %s to index",path);
 521        if (verbose)
 522                printf("add '%s'\n", path);
 523        return 0;
 524}
 525
 526struct cache_entry *make_cache_entry(unsigned int mode,
 527                const unsigned char *sha1, const char *path, int stage,
 528                int refresh)
 529{
 530        int size, len;
 531        struct cache_entry *ce;
 532
 533        if (!verify_path(path))
 534                return NULL;
 535
 536        len = strlen(path);
 537        size = cache_entry_size(len);
 538        ce = xcalloc(1, size);
 539
 540        hashcpy(ce->sha1, sha1);
 541        memcpy(ce->name, path, len);
 542        ce->ce_flags = create_ce_flags(len, stage);
 543        ce->ce_mode = create_ce_mode(mode);
 544
 545        if (refresh)
 546                return refresh_cache_entry(ce, 0);
 547
 548        return ce;
 549}
 550
 551int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 552{
 553        int len = ce_namelen(a);
 554        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 555}
 556
 557int ce_path_match(const struct cache_entry *ce, const char **pathspec)
 558{
 559        const char *match, *name;
 560        int len;
 561
 562        if (!pathspec)
 563                return 1;
 564
 565        len = ce_namelen(ce);
 566        name = ce->name;
 567        while ((match = *pathspec++) != NULL) {
 568                int matchlen = strlen(match);
 569                if (matchlen > len)
 570                        continue;
 571                if (memcmp(name, match, matchlen))
 572                        continue;
 573                if (matchlen && name[matchlen-1] == '/')
 574                        return 1;
 575                if (name[matchlen] == '/' || !name[matchlen])
 576                        return 1;
 577                if (!matchlen)
 578                        return 1;
 579        }
 580        return 0;
 581}
 582
 583/*
 584 * We fundamentally don't like some paths: we don't want
 585 * dot or dot-dot anywhere, and for obvious reasons don't
 586 * want to recurse into ".git" either.
 587 *
 588 * Also, we don't want double slashes or slashes at the
 589 * end that can make pathnames ambiguous.
 590 */
 591static int verify_dotfile(const char *rest)
 592{
 593        /*
 594         * The first character was '.', but that
 595         * has already been discarded, we now test
 596         * the rest.
 597         */
 598        switch (*rest) {
 599        /* "." is not allowed */
 600        case '\0': case '/':
 601                return 0;
 602
 603        /*
 604         * ".git" followed by  NUL or slash is bad. This
 605         * shares the path end test with the ".." case.
 606         */
 607        case 'g':
 608                if (rest[1] != 'i')
 609                        break;
 610                if (rest[2] != 't')
 611                        break;
 612                rest += 2;
 613        /* fallthrough */
 614        case '.':
 615                if (rest[1] == '\0' || rest[1] == '/')
 616                        return 0;
 617        }
 618        return 1;
 619}
 620
 621int verify_path(const char *path)
 622{
 623        char c;
 624
 625        goto inside;
 626        for (;;) {
 627                if (!c)
 628                        return 1;
 629                if (c == '/') {
 630inside:
 631                        c = *path++;
 632                        switch (c) {
 633                        default:
 634                                continue;
 635                        case '/': case '\0':
 636                                break;
 637                        case '.':
 638                                if (verify_dotfile(path))
 639                                        continue;
 640                        }
 641                        return 0;
 642                }
 643                c = *path++;
 644        }
 645}
 646
 647/*
 648 * Do we have another file that has the beginning components being a
 649 * proper superset of the name we're trying to add?
 650 */
 651static int has_file_name(struct index_state *istate,
 652                         const struct cache_entry *ce, int pos, int ok_to_replace)
 653{
 654        int retval = 0;
 655        int len = ce_namelen(ce);
 656        int stage = ce_stage(ce);
 657        const char *name = ce->name;
 658
 659        while (pos < istate->cache_nr) {
 660                struct cache_entry *p = istate->cache[pos++];
 661
 662                if (len >= ce_namelen(p))
 663                        break;
 664                if (memcmp(name, p->name, len))
 665                        break;
 666                if (ce_stage(p) != stage)
 667                        continue;
 668                if (p->name[len] != '/')
 669                        continue;
 670                if (p->ce_flags & CE_REMOVE)
 671                        continue;
 672                retval = -1;
 673                if (!ok_to_replace)
 674                        break;
 675                remove_index_entry_at(istate, --pos);
 676        }
 677        return retval;
 678}
 679
 680/*
 681 * Do we have another file with a pathname that is a proper
 682 * subset of the name we're trying to add?
 683 */
 684static int has_dir_name(struct index_state *istate,
 685                        const struct cache_entry *ce, int pos, int ok_to_replace)
 686{
 687        int retval = 0;
 688        int stage = ce_stage(ce);
 689        const char *name = ce->name;
 690        const char *slash = name + ce_namelen(ce);
 691
 692        for (;;) {
 693                int len;
 694
 695                for (;;) {
 696                        if (*--slash == '/')
 697                                break;
 698                        if (slash <= ce->name)
 699                                return retval;
 700                }
 701                len = slash - name;
 702
 703                pos = index_name_pos(istate, name, create_ce_flags(len, stage));
 704                if (pos >= 0) {
 705                        /*
 706                         * Found one, but not so fast.  This could
 707                         * be a marker that says "I was here, but
 708                         * I am being removed".  Such an entry is
 709                         * not a part of the resulting tree, and
 710                         * it is Ok to have a directory at the same
 711                         * path.
 712                         */
 713                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 714                                retval = -1;
 715                                if (!ok_to_replace)
 716                                        break;
 717                                remove_index_entry_at(istate, pos);
 718                                continue;
 719                        }
 720                }
 721                else
 722                        pos = -pos-1;
 723
 724                /*
 725                 * Trivial optimization: if we find an entry that
 726                 * already matches the sub-directory, then we know
 727                 * we're ok, and we can exit.
 728                 */
 729                while (pos < istate->cache_nr) {
 730                        struct cache_entry *p = istate->cache[pos];
 731                        if ((ce_namelen(p) <= len) ||
 732                            (p->name[len] != '/') ||
 733                            memcmp(p->name, name, len))
 734                                break; /* not our subdirectory */
 735                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 736                                /*
 737                                 * p is at the same stage as our entry, and
 738                                 * is a subdirectory of what we are looking
 739                                 * at, so we cannot have conflicts at our
 740                                 * level or anything shorter.
 741                                 */
 742                                return retval;
 743                        pos++;
 744                }
 745        }
 746        return retval;
 747}
 748
 749/* We may be in a situation where we already have path/file and path
 750 * is being added, or we already have path and path/file is being
 751 * added.  Either one would result in a nonsense tree that has path
 752 * twice when git-write-tree tries to write it out.  Prevent it.
 753 *
 754 * If ok-to-replace is specified, we remove the conflicting entries
 755 * from the cache so the caller should recompute the insert position.
 756 * When this happens, we return non-zero.
 757 */
 758static int check_file_directory_conflict(struct index_state *istate,
 759                                         const struct cache_entry *ce,
 760                                         int pos, int ok_to_replace)
 761{
 762        int retval;
 763
 764        /*
 765         * When ce is an "I am going away" entry, we allow it to be added
 766         */
 767        if (ce->ce_flags & CE_REMOVE)
 768                return 0;
 769
 770        /*
 771         * We check if the path is a sub-path of a subsequent pathname
 772         * first, since removing those will not change the position
 773         * in the array.
 774         */
 775        retval = has_file_name(istate, ce, pos, ok_to_replace);
 776
 777        /*
 778         * Then check if the path might have a clashing sub-directory
 779         * before it.
 780         */
 781        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 782}
 783
 784static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 785{
 786        int pos;
 787        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 788        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 789        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 790
 791        cache_tree_invalidate_path(istate->cache_tree, ce->name);
 792        pos = index_name_pos(istate, ce->name, ce->ce_flags);
 793
 794        /* existing match? Just replace it. */
 795        if (pos >= 0) {
 796                replace_index_entry(istate, pos, ce);
 797                return 0;
 798        }
 799        pos = -pos-1;
 800
 801        /*
 802         * Inserting a merged entry ("stage 0") into the index
 803         * will always replace all non-merged entries..
 804         */
 805        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 806                while (ce_same_name(istate->cache[pos], ce)) {
 807                        ok_to_add = 1;
 808                        if (!remove_index_entry_at(istate, pos))
 809                                break;
 810                }
 811        }
 812
 813        if (!ok_to_add)
 814                return -1;
 815        if (!verify_path(ce->name))
 816                return -1;
 817
 818        if (!skip_df_check &&
 819            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 820                if (!ok_to_replace)
 821                        return error("'%s' appears as both a file and as a directory",
 822                                     ce->name);
 823                pos = index_name_pos(istate, ce->name, ce->ce_flags);
 824                pos = -pos-1;
 825        }
 826        return pos + 1;
 827}
 828
 829int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 830{
 831        int pos;
 832
 833        if (option & ADD_CACHE_JUST_APPEND)
 834                pos = istate->cache_nr;
 835        else {
 836                int ret;
 837                ret = add_index_entry_with_check(istate, ce, option);
 838                if (ret <= 0)
 839                        return ret;
 840                pos = ret - 1;
 841        }
 842
 843        /* Make sure the array is big enough .. */
 844        if (istate->cache_nr == istate->cache_alloc) {
 845                istate->cache_alloc = alloc_nr(istate->cache_alloc);
 846                istate->cache = xrealloc(istate->cache,
 847                                        istate->cache_alloc * sizeof(struct cache_entry *));
 848        }
 849
 850        /* Add it in.. */
 851        istate->cache_nr++;
 852        if (istate->cache_nr > pos + 1)
 853                memmove(istate->cache + pos + 1,
 854                        istate->cache + pos,
 855                        (istate->cache_nr - pos - 1) * sizeof(ce));
 856        set_index_entry(istate, pos, ce);
 857        istate->cache_changed = 1;
 858        return 0;
 859}
 860
 861/*
 862 * "refresh" does not calculate a new sha1 file or bring the
 863 * cache up-to-date for mode/content changes. But what it
 864 * _does_ do is to "re-match" the stat information of a file
 865 * with the cache, so that you can refresh the cache for a
 866 * file that hasn't been changed but where the stat entry is
 867 * out of date.
 868 *
 869 * For example, you'd want to do this after doing a "git-read-tree",
 870 * to link up the stat cache details with the proper files.
 871 */
 872static struct cache_entry *refresh_cache_ent(struct index_state *istate,
 873                                             struct cache_entry *ce,
 874                                             unsigned int options, int *err)
 875{
 876        struct stat st;
 877        struct cache_entry *updated;
 878        int changed, size;
 879        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 880
 881        if (ce_uptodate(ce))
 882                return ce;
 883
 884        if (lstat(ce->name, &st) < 0) {
 885                if (err)
 886                        *err = errno;
 887                return NULL;
 888        }
 889
 890        changed = ie_match_stat(istate, ce, &st, options);
 891        if (!changed) {
 892                /*
 893                 * The path is unchanged.  If we were told to ignore
 894                 * valid bit, then we did the actual stat check and
 895                 * found that the entry is unmodified.  If the entry
 896                 * is not marked VALID, this is the place to mark it
 897                 * valid again, under "assume unchanged" mode.
 898                 */
 899                if (ignore_valid && assume_unchanged &&
 900                    !(ce->ce_flags & CE_VALID))
 901                        ; /* mark this one VALID again */
 902                else {
 903                        /*
 904                         * We do not mark the index itself "modified"
 905                         * because CE_UPTODATE flag is in-core only;
 906                         * we are not going to write this change out.
 907                         */
 908                        ce_mark_uptodate(ce);
 909                        return ce;
 910                }
 911        }
 912
 913        if (ie_modified(istate, ce, &st, options)) {
 914                if (err)
 915                        *err = EINVAL;
 916                return NULL;
 917        }
 918
 919        size = ce_size(ce);
 920        updated = xmalloc(size);
 921        memcpy(updated, ce, size);
 922        fill_stat_cache_info(updated, &st);
 923        /*
 924         * If ignore_valid is not set, we should leave CE_VALID bit
 925         * alone.  Otherwise, paths marked with --no-assume-unchanged
 926         * (i.e. things to be edited) will reacquire CE_VALID bit
 927         * automatically, which is not really what we want.
 928         */
 929        if (!ignore_valid && assume_unchanged &&
 930            !(ce->ce_flags & CE_VALID))
 931                updated->ce_flags &= ~CE_VALID;
 932
 933        return updated;
 934}
 935
 936int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
 937{
 938        int i;
 939        int has_errors = 0;
 940        int really = (flags & REFRESH_REALLY) != 0;
 941        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
 942        int quiet = (flags & REFRESH_QUIET) != 0;
 943        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
 944        unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
 945
 946        for (i = 0; i < istate->cache_nr; i++) {
 947                struct cache_entry *ce, *new;
 948                int cache_errno = 0;
 949
 950                ce = istate->cache[i];
 951                if (ce_stage(ce)) {
 952                        while ((i < istate->cache_nr) &&
 953                               ! strcmp(istate->cache[i]->name, ce->name))
 954                                i++;
 955                        i--;
 956                        if (allow_unmerged)
 957                                continue;
 958                        printf("%s: needs merge\n", ce->name);
 959                        has_errors = 1;
 960                        continue;
 961                }
 962
 963                if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
 964                        continue;
 965
 966                new = refresh_cache_ent(istate, ce, options, &cache_errno);
 967                if (new == ce)
 968                        continue;
 969                if (!new) {
 970                        if (not_new && cache_errno == ENOENT)
 971                                continue;
 972                        if (really && cache_errno == EINVAL) {
 973                                /* If we are doing --really-refresh that
 974                                 * means the index is not valid anymore.
 975                                 */
 976                                ce->ce_flags &= ~CE_VALID;
 977                                istate->cache_changed = 1;
 978                        }
 979                        if (quiet)
 980                                continue;
 981                        printf("%s: needs update\n", ce->name);
 982                        has_errors = 1;
 983                        continue;
 984                }
 985
 986                replace_index_entry(istate, i, new);
 987        }
 988        return has_errors;
 989}
 990
 991struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
 992{
 993        return refresh_cache_ent(&the_index, ce, really, NULL);
 994}
 995
 996static int verify_hdr(struct cache_header *hdr, unsigned long size)
 997{
 998        SHA_CTX c;
 999        unsigned char sha1[20];
1000
1001        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1002                return error("bad signature");
1003        if (hdr->hdr_version != htonl(2))
1004                return error("bad index version");
1005        SHA1_Init(&c);
1006        SHA1_Update(&c, hdr, size - 20);
1007        SHA1_Final(sha1, &c);
1008        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1009                return error("bad index file sha1 signature");
1010        return 0;
1011}
1012
1013static int read_index_extension(struct index_state *istate,
1014                                const char *ext, void *data, unsigned long sz)
1015{
1016        switch (CACHE_EXT(ext)) {
1017        case CACHE_EXT_TREE:
1018                istate->cache_tree = cache_tree_read(data, sz);
1019                break;
1020        default:
1021                if (*ext < 'A' || 'Z' < *ext)
1022                        return error("index uses %.4s extension, which we do not understand",
1023                                     ext);
1024                fprintf(stderr, "ignoring %.4s extension\n", ext);
1025                break;
1026        }
1027        return 0;
1028}
1029
1030int read_index(struct index_state *istate)
1031{
1032        return read_index_from(istate, get_index_file());
1033}
1034
1035static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1036{
1037        size_t len;
1038
1039        ce->ce_ctime = ntohl(ondisk->ctime.sec);
1040        ce->ce_mtime = ntohl(ondisk->mtime.sec);
1041        ce->ce_dev   = ntohl(ondisk->dev);
1042        ce->ce_ino   = ntohl(ondisk->ino);
1043        ce->ce_mode  = ntohl(ondisk->mode);
1044        ce->ce_uid   = ntohl(ondisk->uid);
1045        ce->ce_gid   = ntohl(ondisk->gid);
1046        ce->ce_size  = ntohl(ondisk->size);
1047        /* On-disk flags are just 16 bits */
1048        ce->ce_flags = ntohs(ondisk->flags);
1049        hashcpy(ce->sha1, ondisk->sha1);
1050
1051        len = ce->ce_flags & CE_NAMEMASK;
1052        if (len == CE_NAMEMASK)
1053                len = strlen(ondisk->name);
1054        /*
1055         * NEEDSWORK: If the original index is crafted, this copy could
1056         * go unchecked.
1057         */
1058        memcpy(ce->name, ondisk->name, len + 1);
1059}
1060
1061static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1062{
1063        long per_entry;
1064
1065        per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1066
1067        /*
1068         * Alignment can cause differences. This should be "alignof", but
1069         * since that's a gcc'ism, just use the size of a pointer.
1070         */
1071        per_entry += sizeof(void *);
1072        return ondisk_size + entries*per_entry;
1073}
1074
1075/* remember to discard_cache() before reading a different cache! */
1076int read_index_from(struct index_state *istate, const char *path)
1077{
1078        int fd, i;
1079        struct stat st;
1080        unsigned long src_offset, dst_offset;
1081        struct cache_header *hdr;
1082        void *mmap;
1083        size_t mmap_size;
1084
1085        errno = EBUSY;
1086        if (istate->alloc)
1087                return istate->cache_nr;
1088
1089        errno = ENOENT;
1090        istate->timestamp = 0;
1091        fd = open(path, O_RDONLY);
1092        if (fd < 0) {
1093                if (errno == ENOENT)
1094                        return 0;
1095                die("index file open failed (%s)", strerror(errno));
1096        }
1097
1098        if (fstat(fd, &st))
1099                die("cannot stat the open index (%s)", strerror(errno));
1100
1101        errno = EINVAL;
1102        mmap_size = xsize_t(st.st_size);
1103        if (mmap_size < sizeof(struct cache_header) + 20)
1104                die("index file smaller than expected");
1105
1106        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1107        close(fd);
1108        if (mmap == MAP_FAILED)
1109                die("unable to map index file");
1110
1111        hdr = mmap;
1112        if (verify_hdr(hdr, mmap_size) < 0)
1113                goto unmap;
1114
1115        istate->cache_nr = ntohl(hdr->hdr_entries);
1116        istate->cache_alloc = alloc_nr(istate->cache_nr);
1117        istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1118
1119        /*
1120         * The disk format is actually larger than the in-memory format,
1121         * due to space for nsec etc, so even though the in-memory one
1122         * has room for a few  more flags, we can allocate using the same
1123         * index size
1124         */
1125        istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1126
1127        src_offset = sizeof(*hdr);
1128        dst_offset = 0;
1129        for (i = 0; i < istate->cache_nr; i++) {
1130                struct ondisk_cache_entry *disk_ce;
1131                struct cache_entry *ce;
1132
1133                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1134                ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1135                convert_from_disk(disk_ce, ce);
1136                set_index_entry(istate, i, ce);
1137
1138                src_offset += ondisk_ce_size(ce);
1139                dst_offset += ce_size(ce);
1140        }
1141        istate->timestamp = st.st_mtime;
1142        while (src_offset <= mmap_size - 20 - 8) {
1143                /* After an array of active_nr index entries,
1144                 * there can be arbitrary number of extended
1145                 * sections, each of which is prefixed with
1146                 * extension name (4-byte) and section length
1147                 * in 4-byte network byte order.
1148                 */
1149                unsigned long extsize;
1150                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1151                extsize = ntohl(extsize);
1152                if (read_index_extension(istate,
1153                                         (const char *) mmap + src_offset,
1154                                         (char *) mmap + src_offset + 8,
1155                                         extsize) < 0)
1156                        goto unmap;
1157                src_offset += 8;
1158                src_offset += extsize;
1159        }
1160        munmap(mmap, mmap_size);
1161        return istate->cache_nr;
1162
1163unmap:
1164        munmap(mmap, mmap_size);
1165        errno = EINVAL;
1166        die("index file corrupt");
1167}
1168
1169int discard_index(struct index_state *istate)
1170{
1171        istate->cache_nr = 0;
1172        istate->cache_changed = 0;
1173        istate->timestamp = 0;
1174        free_hash(&istate->name_hash);
1175        cache_tree_free(&(istate->cache_tree));
1176        free(istate->alloc);
1177        istate->alloc = NULL;
1178
1179        /* no need to throw away allocated active_cache */
1180        return 0;
1181}
1182
1183#define WRITE_BUFFER_SIZE 8192
1184static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1185static unsigned long write_buffer_len;
1186
1187static int ce_write_flush(SHA_CTX *context, int fd)
1188{
1189        unsigned int buffered = write_buffer_len;
1190        if (buffered) {
1191                SHA1_Update(context, write_buffer, buffered);
1192                if (write_in_full(fd, write_buffer, buffered) != buffered)
1193                        return -1;
1194                write_buffer_len = 0;
1195        }
1196        return 0;
1197}
1198
1199static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1200{
1201        while (len) {
1202                unsigned int buffered = write_buffer_len;
1203                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1204                if (partial > len)
1205                        partial = len;
1206                memcpy(write_buffer + buffered, data, partial);
1207                buffered += partial;
1208                if (buffered == WRITE_BUFFER_SIZE) {
1209                        write_buffer_len = buffered;
1210                        if (ce_write_flush(context, fd))
1211                                return -1;
1212                        buffered = 0;
1213                }
1214                write_buffer_len = buffered;
1215                len -= partial;
1216                data = (char *) data + partial;
1217        }
1218        return 0;
1219}
1220
1221static int write_index_ext_header(SHA_CTX *context, int fd,
1222                                  unsigned int ext, unsigned int sz)
1223{
1224        ext = htonl(ext);
1225        sz = htonl(sz);
1226        return ((ce_write(context, fd, &ext, 4) < 0) ||
1227                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1228}
1229
1230static int ce_flush(SHA_CTX *context, int fd)
1231{
1232        unsigned int left = write_buffer_len;
1233
1234        if (left) {
1235                write_buffer_len = 0;
1236                SHA1_Update(context, write_buffer, left);
1237        }
1238
1239        /* Flush first if not enough space for SHA1 signature */
1240        if (left + 20 > WRITE_BUFFER_SIZE) {
1241                if (write_in_full(fd, write_buffer, left) != left)
1242                        return -1;
1243                left = 0;
1244        }
1245
1246        /* Append the SHA1 signature at the end */
1247        SHA1_Final(write_buffer + left, context);
1248        left += 20;
1249        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1250}
1251
1252static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1253{
1254        /*
1255         * The only thing we care about in this function is to smudge the
1256         * falsely clean entry due to touch-update-touch race, so we leave
1257         * everything else as they are.  We are called for entries whose
1258         * ce_mtime match the index file mtime.
1259         */
1260        struct stat st;
1261
1262        if (lstat(ce->name, &st) < 0)
1263                return;
1264        if (ce_match_stat_basic(ce, &st))
1265                return;
1266        if (ce_modified_check_fs(ce, &st)) {
1267                /* This is "racily clean"; smudge it.  Note that this
1268                 * is a tricky code.  At first glance, it may appear
1269                 * that it can break with this sequence:
1270                 *
1271                 * $ echo xyzzy >frotz
1272                 * $ git-update-index --add frotz
1273                 * $ : >frotz
1274                 * $ sleep 3
1275                 * $ echo filfre >nitfol
1276                 * $ git-update-index --add nitfol
1277                 *
1278                 * but it does not.  When the second update-index runs,
1279                 * it notices that the entry "frotz" has the same timestamp
1280                 * as index, and if we were to smudge it by resetting its
1281                 * size to zero here, then the object name recorded
1282                 * in index is the 6-byte file but the cached stat information
1283                 * becomes zero --- which would then match what we would
1284                 * obtain from the filesystem next time we stat("frotz").
1285                 *
1286                 * However, the second update-index, before calling
1287                 * this function, notices that the cached size is 6
1288                 * bytes and what is on the filesystem is an empty
1289                 * file, and never calls us, so the cached size information
1290                 * for "frotz" stays 6 which does not match the filesystem.
1291                 */
1292                ce->ce_size = 0;
1293        }
1294}
1295
1296static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1297{
1298        int size = ondisk_ce_size(ce);
1299        struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1300
1301        ondisk->ctime.sec = htonl(ce->ce_ctime);
1302        ondisk->ctime.nsec = 0;
1303        ondisk->mtime.sec = htonl(ce->ce_mtime);
1304        ondisk->mtime.nsec = 0;
1305        ondisk->dev  = htonl(ce->ce_dev);
1306        ondisk->ino  = htonl(ce->ce_ino);
1307        ondisk->mode = htonl(ce->ce_mode);
1308        ondisk->uid  = htonl(ce->ce_uid);
1309        ondisk->gid  = htonl(ce->ce_gid);
1310        ondisk->size = htonl(ce->ce_size);
1311        hashcpy(ondisk->sha1, ce->sha1);
1312        ondisk->flags = htons(ce->ce_flags);
1313        memcpy(ondisk->name, ce->name, ce_namelen(ce));
1314
1315        return ce_write(c, fd, ondisk, size);
1316}
1317
1318int write_index(struct index_state *istate, int newfd)
1319{
1320        SHA_CTX c;
1321        struct cache_header hdr;
1322        int i, err, removed;
1323        struct cache_entry **cache = istate->cache;
1324        int entries = istate->cache_nr;
1325
1326        for (i = removed = 0; i < entries; i++)
1327                if (cache[i]->ce_flags & CE_REMOVE)
1328                        removed++;
1329
1330        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1331        hdr.hdr_version = htonl(2);
1332        hdr.hdr_entries = htonl(entries - removed);
1333
1334        SHA1_Init(&c);
1335        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1336                return -1;
1337
1338        for (i = 0; i < entries; i++) {
1339                struct cache_entry *ce = cache[i];
1340                if (ce->ce_flags & CE_REMOVE)
1341                        continue;
1342                if (is_racy_timestamp(istate, ce))
1343                        ce_smudge_racily_clean_entry(ce);
1344                if (ce_write_entry(&c, newfd, ce) < 0)
1345                        return -1;
1346        }
1347
1348        /* Write extension data here */
1349        if (istate->cache_tree) {
1350                struct strbuf sb;
1351
1352                strbuf_init(&sb, 0);
1353                cache_tree_write(&sb, istate->cache_tree);
1354                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1355                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1356                strbuf_release(&sb);
1357                if (err)
1358                        return -1;
1359        }
1360        return ce_flush(&c, newfd);
1361}