read-cache.con commit Merge branch 'js/config-cb' (9bd81e4)
   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 void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  27{
  28        istate->cache[nr] = ce;
  29        add_name_hash(istate, ce);
  30}
  31
  32static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  33{
  34        struct cache_entry *old = istate->cache[nr];
  35
  36        remove_name_hash(old);
  37        set_index_entry(istate, nr, ce);
  38        istate->cache_changed = 1;
  39}
  40
  41/*
  42 * This only updates the "non-critical" parts of the directory
  43 * cache, ie the parts that aren't tracked by GIT, and only used
  44 * to validate the cache.
  45 */
  46void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
  47{
  48        ce->ce_ctime = st->st_ctime;
  49        ce->ce_mtime = st->st_mtime;
  50        ce->ce_dev = st->st_dev;
  51        ce->ce_ino = st->st_ino;
  52        ce->ce_uid = st->st_uid;
  53        ce->ce_gid = st->st_gid;
  54        ce->ce_size = st->st_size;
  55
  56        if (assume_unchanged)
  57                ce->ce_flags |= CE_VALID;
  58
  59        if (S_ISREG(st->st_mode))
  60                ce_mark_uptodate(ce);
  61}
  62
  63static int ce_compare_data(struct cache_entry *ce, struct stat *st)
  64{
  65        int match = -1;
  66        int fd = open(ce->name, O_RDONLY);
  67
  68        if (fd >= 0) {
  69                unsigned char sha1[20];
  70                if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
  71                        match = hashcmp(sha1, ce->sha1);
  72                /* index_fd() closed the file descriptor already */
  73        }
  74        return match;
  75}
  76
  77static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
  78{
  79        int match = -1;
  80        char *target;
  81        void *buffer;
  82        unsigned long size;
  83        enum object_type type;
  84        int len;
  85
  86        target = xmalloc(expected_size);
  87        len = readlink(ce->name, target, expected_size);
  88        if (len != expected_size) {
  89                free(target);
  90                return -1;
  91        }
  92        buffer = read_sha1_file(ce->sha1, &type, &size);
  93        if (!buffer) {
  94                free(target);
  95                return -1;
  96        }
  97        if (size == expected_size)
  98                match = memcmp(buffer, target, size);
  99        free(buffer);
 100        free(target);
 101        return match;
 102}
 103
 104static int ce_compare_gitlink(struct cache_entry *ce)
 105{
 106        unsigned char sha1[20];
 107
 108        /*
 109         * We don't actually require that the .git directory
 110         * under GITLINK directory be a valid git directory. It
 111         * might even be missing (in case nobody populated that
 112         * sub-project).
 113         *
 114         * If so, we consider it always to match.
 115         */
 116        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 117                return 0;
 118        return hashcmp(sha1, ce->sha1);
 119}
 120
 121static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
 122{
 123        switch (st->st_mode & S_IFMT) {
 124        case S_IFREG:
 125                if (ce_compare_data(ce, st))
 126                        return DATA_CHANGED;
 127                break;
 128        case S_IFLNK:
 129                if (ce_compare_link(ce, xsize_t(st->st_size)))
 130                        return DATA_CHANGED;
 131                break;
 132        case S_IFDIR:
 133                if (S_ISGITLINK(ce->ce_mode))
 134                        return 0;
 135        default:
 136                return TYPE_CHANGED;
 137        }
 138        return 0;
 139}
 140
 141static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 142{
 143        unsigned int changed = 0;
 144
 145        if (ce->ce_flags & CE_REMOVE)
 146                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 147
 148        switch (ce->ce_mode & S_IFMT) {
 149        case S_IFREG:
 150                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 151                /* We consider only the owner x bit to be relevant for
 152                 * "mode changes"
 153                 */
 154                if (trust_executable_bit &&
 155                    (0100 & (ce->ce_mode ^ st->st_mode)))
 156                        changed |= MODE_CHANGED;
 157                break;
 158        case S_IFLNK:
 159                if (!S_ISLNK(st->st_mode) &&
 160                    (has_symlinks || !S_ISREG(st->st_mode)))
 161                        changed |= TYPE_CHANGED;
 162                break;
 163        case S_IFGITLINK:
 164                if (!S_ISDIR(st->st_mode))
 165                        changed |= TYPE_CHANGED;
 166                else if (ce_compare_gitlink(ce))
 167                        changed |= DATA_CHANGED;
 168                return changed;
 169        default:
 170                die("internal error: ce_mode is %o", ce->ce_mode);
 171        }
 172        if (ce->ce_mtime != (unsigned int) st->st_mtime)
 173                changed |= MTIME_CHANGED;
 174        if (ce->ce_ctime != (unsigned int) st->st_ctime)
 175                changed |= CTIME_CHANGED;
 176
 177        if (ce->ce_uid != (unsigned int) st->st_uid ||
 178            ce->ce_gid != (unsigned int) st->st_gid)
 179                changed |= OWNER_CHANGED;
 180        if (ce->ce_ino != (unsigned int) st->st_ino)
 181                changed |= INODE_CHANGED;
 182
 183#ifdef USE_STDEV
 184        /*
 185         * st_dev breaks on network filesystems where different
 186         * clients will have different views of what "device"
 187         * the filesystem is on
 188         */
 189        if (ce->ce_dev != (unsigned int) st->st_dev)
 190                changed |= INODE_CHANGED;
 191#endif
 192
 193        if (ce->ce_size != (unsigned int) st->st_size)
 194                changed |= DATA_CHANGED;
 195
 196        return changed;
 197}
 198
 199static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
 200{
 201        return (!S_ISGITLINK(ce->ce_mode) &&
 202                istate->timestamp &&
 203                ((unsigned int)istate->timestamp) <= ce->ce_mtime);
 204}
 205
 206int ie_match_stat(const struct index_state *istate,
 207                  struct cache_entry *ce, struct stat *st,
 208                  unsigned int options)
 209{
 210        unsigned int changed;
 211        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 212        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 213
 214        /*
 215         * If it's marked as always valid in the index, it's
 216         * valid whatever the checked-out copy says.
 217         */
 218        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 219                return 0;
 220
 221        changed = ce_match_stat_basic(ce, st);
 222
 223        /*
 224         * Within 1 second of this sequence:
 225         *      echo xyzzy >file && git-update-index --add file
 226         * running this command:
 227         *      echo frotz >file
 228         * would give a falsely clean cache entry.  The mtime and
 229         * length match the cache, and other stat fields do not change.
 230         *
 231         * We could detect this at update-index time (the cache entry
 232         * being registered/updated records the same time as "now")
 233         * and delay the return from git-update-index, but that would
 234         * effectively mean we can make at most one commit per second,
 235         * which is not acceptable.  Instead, we check cache entries
 236         * whose mtime are the same as the index file timestamp more
 237         * carefully than others.
 238         */
 239        if (!changed && is_racy_timestamp(istate, ce)) {
 240                if (assume_racy_is_modified)
 241                        changed |= DATA_CHANGED;
 242                else
 243                        changed |= ce_modified_check_fs(ce, st);
 244        }
 245
 246        return changed;
 247}
 248
 249int ie_modified(const struct index_state *istate,
 250                struct cache_entry *ce, struct stat *st, unsigned int options)
 251{
 252        int changed, changed_fs;
 253
 254        changed = ie_match_stat(istate, ce, st, options);
 255        if (!changed)
 256                return 0;
 257        /*
 258         * If the mode or type has changed, there's no point in trying
 259         * to refresh the entry - it's not going to match
 260         */
 261        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 262                return changed;
 263
 264        /* Immediately after read-tree or update-index --cacheinfo,
 265         * the length field is zero.  For other cases the ce_size
 266         * should match the SHA1 recorded in the index entry.
 267         */
 268        if ((changed & DATA_CHANGED) && ce->ce_size != 0)
 269                return changed;
 270
 271        changed_fs = ce_modified_check_fs(ce, st);
 272        if (changed_fs)
 273                return changed | changed_fs;
 274        return 0;
 275}
 276
 277int base_name_compare(const char *name1, int len1, int mode1,
 278                      const char *name2, int len2, int mode2)
 279{
 280        unsigned char c1, c2;
 281        int len = len1 < len2 ? len1 : len2;
 282        int cmp;
 283
 284        cmp = memcmp(name1, name2, len);
 285        if (cmp)
 286                return cmp;
 287        c1 = name1[len];
 288        c2 = name2[len];
 289        if (!c1 && S_ISDIR(mode1))
 290                c1 = '/';
 291        if (!c2 && S_ISDIR(mode2))
 292                c2 = '/';
 293        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 294}
 295
 296/*
 297 * df_name_compare() is identical to base_name_compare(), except it
 298 * compares conflicting directory/file entries as equal. Note that
 299 * while a directory name compares as equal to a regular file, they
 300 * then individually compare _differently_ to a filename that has
 301 * a dot after the basename (because '\0' < '.' < '/').
 302 *
 303 * This is used by routines that want to traverse the git namespace
 304 * but then handle conflicting entries together when possible.
 305 */
 306int df_name_compare(const char *name1, int len1, int mode1,
 307                    const char *name2, int len2, int mode2)
 308{
 309        int len = len1 < len2 ? len1 : len2, cmp;
 310        unsigned char c1, c2;
 311
 312        cmp = memcmp(name1, name2, len);
 313        if (cmp)
 314                return cmp;
 315        /* Directories and files compare equal (same length, same name) */
 316        if (len1 == len2)
 317                return 0;
 318        c1 = name1[len];
 319        if (!c1 && S_ISDIR(mode1))
 320                c1 = '/';
 321        c2 = name2[len];
 322        if (!c2 && S_ISDIR(mode2))
 323                c2 = '/';
 324        if (c1 == '/' && !c2)
 325                return 0;
 326        if (c2 == '/' && !c1)
 327                return 0;
 328        return c1 - c2;
 329}
 330
 331int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 332{
 333        int len1 = flags1 & CE_NAMEMASK;
 334        int len2 = flags2 & CE_NAMEMASK;
 335        int len = len1 < len2 ? len1 : len2;
 336        int cmp;
 337
 338        cmp = memcmp(name1, name2, len);
 339        if (cmp)
 340                return cmp;
 341        if (len1 < len2)
 342                return -1;
 343        if (len1 > len2)
 344                return 1;
 345
 346        /* Compare stages  */
 347        flags1 &= CE_STAGEMASK;
 348        flags2 &= CE_STAGEMASK;
 349
 350        if (flags1 < flags2)
 351                return -1;
 352        if (flags1 > flags2)
 353                return 1;
 354        return 0;
 355}
 356
 357int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 358{
 359        int first, last;
 360
 361        first = 0;
 362        last = istate->cache_nr;
 363        while (last > first) {
 364                int next = (last + first) >> 1;
 365                struct cache_entry *ce = istate->cache[next];
 366                int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
 367                if (!cmp)
 368                        return next;
 369                if (cmp < 0) {
 370                        last = next;
 371                        continue;
 372                }
 373                first = next+1;
 374        }
 375        return -first-1;
 376}
 377
 378/* Remove entry, return true if there are more entries to go.. */
 379int remove_index_entry_at(struct index_state *istate, int pos)
 380{
 381        struct cache_entry *ce = istate->cache[pos];
 382
 383        remove_name_hash(ce);
 384        istate->cache_changed = 1;
 385        istate->cache_nr--;
 386        if (pos >= istate->cache_nr)
 387                return 0;
 388        memmove(istate->cache + pos,
 389                istate->cache + pos + 1,
 390                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 391        return 1;
 392}
 393
 394int remove_file_from_index(struct index_state *istate, const char *path)
 395{
 396        int pos = index_name_pos(istate, path, strlen(path));
 397        if (pos < 0)
 398                pos = -pos-1;
 399        cache_tree_invalidate_path(istate->cache_tree, path);
 400        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 401                remove_index_entry_at(istate, pos);
 402        return 0;
 403}
 404
 405static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 406{
 407        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 408}
 409
 410static int index_name_pos_also_unmerged(struct index_state *istate,
 411        const char *path, int namelen)
 412{
 413        int pos = index_name_pos(istate, path, namelen);
 414        struct cache_entry *ce;
 415
 416        if (pos >= 0)
 417                return pos;
 418
 419        /* maybe unmerged? */
 420        pos = -1 - pos;
 421        if (pos >= istate->cache_nr ||
 422                        compare_name((ce = istate->cache[pos]), path, namelen))
 423                return -1;
 424
 425        /* order of preference: stage 2, 1, 3 */
 426        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 427                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 428                        !compare_name(ce, path, namelen))
 429                pos++;
 430        return pos;
 431}
 432
 433static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 434{
 435        int len = ce_namelen(ce);
 436        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 437}
 438
 439/*
 440 * If we add a filename that aliases in the cache, we will use the
 441 * name that we already have - but we don't want to update the same
 442 * alias twice, because that implies that there were actually two
 443 * different files with aliasing names!
 444 *
 445 * So we use the CE_ADDED flag to verify that the alias was an old
 446 * one before we accept it as
 447 */
 448static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
 449{
 450        int len;
 451        struct cache_entry *new;
 452
 453        if (alias->ce_flags & CE_ADDED)
 454                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 455
 456        /* Ok, create the new entry using the name of the existing alias */
 457        len = ce_namelen(alias);
 458        new = xcalloc(1, cache_entry_size(len));
 459        memcpy(new->name, alias->name, len);
 460        copy_cache_entry(new, ce);
 461        free(ce);
 462        return new;
 463}
 464
 465int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 466{
 467        int size, namelen, was_same;
 468        mode_t st_mode = st->st_mode;
 469        struct cache_entry *ce, *alias;
 470        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
 471        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 472        int pretend = flags & ADD_CACHE_PRETEND;
 473
 474        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 475                return error("%s: can only add regular files, symbolic links or git-directories", path);
 476
 477        namelen = strlen(path);
 478        if (S_ISDIR(st_mode)) {
 479                while (namelen && path[namelen-1] == '/')
 480                        namelen--;
 481        }
 482        size = cache_entry_size(namelen);
 483        ce = xcalloc(1, size);
 484        memcpy(ce->name, path, namelen);
 485        ce->ce_flags = namelen;
 486        fill_stat_cache_info(ce, st);
 487
 488        if (trust_executable_bit && has_symlinks)
 489                ce->ce_mode = create_ce_mode(st_mode);
 490        else {
 491                /* If there is an existing entry, pick the mode bits and type
 492                 * from it, otherwise assume unexecutable regular file.
 493                 */
 494                struct cache_entry *ent;
 495                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 496
 497                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 498                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 499        }
 500
 501        alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 502        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 503                /* Nothing changed, really */
 504                free(ce);
 505                ce_mark_uptodate(alias);
 506                alias->ce_flags |= CE_ADDED;
 507                return 0;
 508        }
 509        if (index_path(ce->sha1, path, st, 1))
 510                return error("unable to index file %s", path);
 511        if (ignore_case && alias && different_name(ce, alias))
 512                ce = create_alias_ce(ce, alias);
 513        ce->ce_flags |= CE_ADDED;
 514
 515        /* It was suspected to be recily clean, but it turns out to be Ok */
 516        was_same = (alias &&
 517                    !ce_stage(alias) &&
 518                    !hashcmp(alias->sha1, ce->sha1) &&
 519                    ce->ce_mode == alias->ce_mode);
 520
 521        if (pretend)
 522                ;
 523        else if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
 524                return error("unable to add %s to index",path);
 525        if (verbose && !was_same)
 526                printf("add '%s'\n", path);
 527        return 0;
 528}
 529
 530int add_file_to_index(struct index_state *istate, const char *path, int flags)
 531{
 532        struct stat st;
 533        if (lstat(path, &st))
 534                die("%s: unable to stat (%s)", path, strerror(errno));
 535        return add_to_index(istate, path, &st, flags);
 536}
 537
 538struct cache_entry *make_cache_entry(unsigned int mode,
 539                const unsigned char *sha1, const char *path, int stage,
 540                int refresh)
 541{
 542        int size, len;
 543        struct cache_entry *ce;
 544
 545        if (!verify_path(path))
 546                return NULL;
 547
 548        len = strlen(path);
 549        size = cache_entry_size(len);
 550        ce = xcalloc(1, size);
 551
 552        hashcpy(ce->sha1, sha1);
 553        memcpy(ce->name, path, len);
 554        ce->ce_flags = create_ce_flags(len, stage);
 555        ce->ce_mode = create_ce_mode(mode);
 556
 557        if (refresh)
 558                return refresh_cache_entry(ce, 0);
 559
 560        return ce;
 561}
 562
 563int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 564{
 565        int len = ce_namelen(a);
 566        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 567}
 568
 569int ce_path_match(const struct cache_entry *ce, const char **pathspec)
 570{
 571        const char *match, *name;
 572        int len;
 573
 574        if (!pathspec)
 575                return 1;
 576
 577        len = ce_namelen(ce);
 578        name = ce->name;
 579        while ((match = *pathspec++) != NULL) {
 580                int matchlen = strlen(match);
 581                if (matchlen > len)
 582                        continue;
 583                if (memcmp(name, match, matchlen))
 584                        continue;
 585                if (matchlen && name[matchlen-1] == '/')
 586                        return 1;
 587                if (name[matchlen] == '/' || !name[matchlen])
 588                        return 1;
 589                if (!matchlen)
 590                        return 1;
 591        }
 592        return 0;
 593}
 594
 595/*
 596 * We fundamentally don't like some paths: we don't want
 597 * dot or dot-dot anywhere, and for obvious reasons don't
 598 * want to recurse into ".git" either.
 599 *
 600 * Also, we don't want double slashes or slashes at the
 601 * end that can make pathnames ambiguous.
 602 */
 603static int verify_dotfile(const char *rest)
 604{
 605        /*
 606         * The first character was '.', but that
 607         * has already been discarded, we now test
 608         * the rest.
 609         */
 610        switch (*rest) {
 611        /* "." is not allowed */
 612        case '\0': case '/':
 613                return 0;
 614
 615        /*
 616         * ".git" followed by  NUL or slash is bad. This
 617         * shares the path end test with the ".." case.
 618         */
 619        case 'g':
 620                if (rest[1] != 'i')
 621                        break;
 622                if (rest[2] != 't')
 623                        break;
 624                rest += 2;
 625        /* fallthrough */
 626        case '.':
 627                if (rest[1] == '\0' || rest[1] == '/')
 628                        return 0;
 629        }
 630        return 1;
 631}
 632
 633int verify_path(const char *path)
 634{
 635        char c;
 636
 637        goto inside;
 638        for (;;) {
 639                if (!c)
 640                        return 1;
 641                if (c == '/') {
 642inside:
 643                        c = *path++;
 644                        switch (c) {
 645                        default:
 646                                continue;
 647                        case '/': case '\0':
 648                                break;
 649                        case '.':
 650                                if (verify_dotfile(path))
 651                                        continue;
 652                        }
 653                        return 0;
 654                }
 655                c = *path++;
 656        }
 657}
 658
 659/*
 660 * Do we have another file that has the beginning components being a
 661 * proper superset of the name we're trying to add?
 662 */
 663static int has_file_name(struct index_state *istate,
 664                         const struct cache_entry *ce, int pos, int ok_to_replace)
 665{
 666        int retval = 0;
 667        int len = ce_namelen(ce);
 668        int stage = ce_stage(ce);
 669        const char *name = ce->name;
 670
 671        while (pos < istate->cache_nr) {
 672                struct cache_entry *p = istate->cache[pos++];
 673
 674                if (len >= ce_namelen(p))
 675                        break;
 676                if (memcmp(name, p->name, len))
 677                        break;
 678                if (ce_stage(p) != stage)
 679                        continue;
 680                if (p->name[len] != '/')
 681                        continue;
 682                if (p->ce_flags & CE_REMOVE)
 683                        continue;
 684                retval = -1;
 685                if (!ok_to_replace)
 686                        break;
 687                remove_index_entry_at(istate, --pos);
 688        }
 689        return retval;
 690}
 691
 692/*
 693 * Do we have another file with a pathname that is a proper
 694 * subset of the name we're trying to add?
 695 */
 696static int has_dir_name(struct index_state *istate,
 697                        const struct cache_entry *ce, int pos, int ok_to_replace)
 698{
 699        int retval = 0;
 700        int stage = ce_stage(ce);
 701        const char *name = ce->name;
 702        const char *slash = name + ce_namelen(ce);
 703
 704        for (;;) {
 705                int len;
 706
 707                for (;;) {
 708                        if (*--slash == '/')
 709                                break;
 710                        if (slash <= ce->name)
 711                                return retval;
 712                }
 713                len = slash - name;
 714
 715                pos = index_name_pos(istate, name, create_ce_flags(len, stage));
 716                if (pos >= 0) {
 717                        /*
 718                         * Found one, but not so fast.  This could
 719                         * be a marker that says "I was here, but
 720                         * I am being removed".  Such an entry is
 721                         * not a part of the resulting tree, and
 722                         * it is Ok to have a directory at the same
 723                         * path.
 724                         */
 725                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 726                                retval = -1;
 727                                if (!ok_to_replace)
 728                                        break;
 729                                remove_index_entry_at(istate, pos);
 730                                continue;
 731                        }
 732                }
 733                else
 734                        pos = -pos-1;
 735
 736                /*
 737                 * Trivial optimization: if we find an entry that
 738                 * already matches the sub-directory, then we know
 739                 * we're ok, and we can exit.
 740                 */
 741                while (pos < istate->cache_nr) {
 742                        struct cache_entry *p = istate->cache[pos];
 743                        if ((ce_namelen(p) <= len) ||
 744                            (p->name[len] != '/') ||
 745                            memcmp(p->name, name, len))
 746                                break; /* not our subdirectory */
 747                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 748                                /*
 749                                 * p is at the same stage as our entry, and
 750                                 * is a subdirectory of what we are looking
 751                                 * at, so we cannot have conflicts at our
 752                                 * level or anything shorter.
 753                                 */
 754                                return retval;
 755                        pos++;
 756                }
 757        }
 758        return retval;
 759}
 760
 761/* We may be in a situation where we already have path/file and path
 762 * is being added, or we already have path and path/file is being
 763 * added.  Either one would result in a nonsense tree that has path
 764 * twice when git-write-tree tries to write it out.  Prevent it.
 765 *
 766 * If ok-to-replace is specified, we remove the conflicting entries
 767 * from the cache so the caller should recompute the insert position.
 768 * When this happens, we return non-zero.
 769 */
 770static int check_file_directory_conflict(struct index_state *istate,
 771                                         const struct cache_entry *ce,
 772                                         int pos, int ok_to_replace)
 773{
 774        int retval;
 775
 776        /*
 777         * When ce is an "I am going away" entry, we allow it to be added
 778         */
 779        if (ce->ce_flags & CE_REMOVE)
 780                return 0;
 781
 782        /*
 783         * We check if the path is a sub-path of a subsequent pathname
 784         * first, since removing those will not change the position
 785         * in the array.
 786         */
 787        retval = has_file_name(istate, ce, pos, ok_to_replace);
 788
 789        /*
 790         * Then check if the path might have a clashing sub-directory
 791         * before it.
 792         */
 793        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 794}
 795
 796static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 797{
 798        int pos;
 799        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 800        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 801        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 802
 803        cache_tree_invalidate_path(istate->cache_tree, ce->name);
 804        pos = index_name_pos(istate, ce->name, ce->ce_flags);
 805
 806        /* existing match? Just replace it. */
 807        if (pos >= 0) {
 808                replace_index_entry(istate, pos, ce);
 809                return 0;
 810        }
 811        pos = -pos-1;
 812
 813        /*
 814         * Inserting a merged entry ("stage 0") into the index
 815         * will always replace all non-merged entries..
 816         */
 817        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 818                while (ce_same_name(istate->cache[pos], ce)) {
 819                        ok_to_add = 1;
 820                        if (!remove_index_entry_at(istate, pos))
 821                                break;
 822                }
 823        }
 824
 825        if (!ok_to_add)
 826                return -1;
 827        if (!verify_path(ce->name))
 828                return -1;
 829
 830        if (!skip_df_check &&
 831            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 832                if (!ok_to_replace)
 833                        return error("'%s' appears as both a file and as a directory",
 834                                     ce->name);
 835                pos = index_name_pos(istate, ce->name, ce->ce_flags);
 836                pos = -pos-1;
 837        }
 838        return pos + 1;
 839}
 840
 841int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 842{
 843        int pos;
 844
 845        if (option & ADD_CACHE_JUST_APPEND)
 846                pos = istate->cache_nr;
 847        else {
 848                int ret;
 849                ret = add_index_entry_with_check(istate, ce, option);
 850                if (ret <= 0)
 851                        return ret;
 852                pos = ret - 1;
 853        }
 854
 855        /* Make sure the array is big enough .. */
 856        if (istate->cache_nr == istate->cache_alloc) {
 857                istate->cache_alloc = alloc_nr(istate->cache_alloc);
 858                istate->cache = xrealloc(istate->cache,
 859                                        istate->cache_alloc * sizeof(struct cache_entry *));
 860        }
 861
 862        /* Add it in.. */
 863        istate->cache_nr++;
 864        if (istate->cache_nr > pos + 1)
 865                memmove(istate->cache + pos + 1,
 866                        istate->cache + pos,
 867                        (istate->cache_nr - pos - 1) * sizeof(ce));
 868        set_index_entry(istate, pos, ce);
 869        istate->cache_changed = 1;
 870        return 0;
 871}
 872
 873/*
 874 * "refresh" does not calculate a new sha1 file or bring the
 875 * cache up-to-date for mode/content changes. But what it
 876 * _does_ do is to "re-match" the stat information of a file
 877 * with the cache, so that you can refresh the cache for a
 878 * file that hasn't been changed but where the stat entry is
 879 * out of date.
 880 *
 881 * For example, you'd want to do this after doing a "git-read-tree",
 882 * to link up the stat cache details with the proper files.
 883 */
 884static struct cache_entry *refresh_cache_ent(struct index_state *istate,
 885                                             struct cache_entry *ce,
 886                                             unsigned int options, int *err)
 887{
 888        struct stat st;
 889        struct cache_entry *updated;
 890        int changed, size;
 891        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 892
 893        if (ce_uptodate(ce))
 894                return ce;
 895
 896        if (lstat(ce->name, &st) < 0) {
 897                if (err)
 898                        *err = errno;
 899                return NULL;
 900        }
 901
 902        changed = ie_match_stat(istate, ce, &st, options);
 903        if (!changed) {
 904                /*
 905                 * The path is unchanged.  If we were told to ignore
 906                 * valid bit, then we did the actual stat check and
 907                 * found that the entry is unmodified.  If the entry
 908                 * is not marked VALID, this is the place to mark it
 909                 * valid again, under "assume unchanged" mode.
 910                 */
 911                if (ignore_valid && assume_unchanged &&
 912                    !(ce->ce_flags & CE_VALID))
 913                        ; /* mark this one VALID again */
 914                else {
 915                        /*
 916                         * We do not mark the index itself "modified"
 917                         * because CE_UPTODATE flag is in-core only;
 918                         * we are not going to write this change out.
 919                         */
 920                        ce_mark_uptodate(ce);
 921                        return ce;
 922                }
 923        }
 924
 925        if (ie_modified(istate, ce, &st, options)) {
 926                if (err)
 927                        *err = EINVAL;
 928                return NULL;
 929        }
 930
 931        size = ce_size(ce);
 932        updated = xmalloc(size);
 933        memcpy(updated, ce, size);
 934        fill_stat_cache_info(updated, &st);
 935        /*
 936         * If ignore_valid is not set, we should leave CE_VALID bit
 937         * alone.  Otherwise, paths marked with --no-assume-unchanged
 938         * (i.e. things to be edited) will reacquire CE_VALID bit
 939         * automatically, which is not really what we want.
 940         */
 941        if (!ignore_valid && assume_unchanged &&
 942            !(ce->ce_flags & CE_VALID))
 943                updated->ce_flags &= ~CE_VALID;
 944
 945        return updated;
 946}
 947
 948int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
 949{
 950        int i;
 951        int has_errors = 0;
 952        int really = (flags & REFRESH_REALLY) != 0;
 953        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
 954        int quiet = (flags & REFRESH_QUIET) != 0;
 955        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
 956        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
 957        unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
 958
 959        for (i = 0; i < istate->cache_nr; i++) {
 960                struct cache_entry *ce, *new;
 961                int cache_errno = 0;
 962
 963                ce = istate->cache[i];
 964                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
 965                        continue;
 966
 967                if (ce_stage(ce)) {
 968                        while ((i < istate->cache_nr) &&
 969                               ! strcmp(istate->cache[i]->name, ce->name))
 970                                i++;
 971                        i--;
 972                        if (allow_unmerged)
 973                                continue;
 974                        printf("%s: needs merge\n", ce->name);
 975                        has_errors = 1;
 976                        continue;
 977                }
 978
 979                if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
 980                        continue;
 981
 982                new = refresh_cache_ent(istate, ce, options, &cache_errno);
 983                if (new == ce)
 984                        continue;
 985                if (!new) {
 986                        if (not_new && cache_errno == ENOENT)
 987                                continue;
 988                        if (really && cache_errno == EINVAL) {
 989                                /* If we are doing --really-refresh that
 990                                 * means the index is not valid anymore.
 991                                 */
 992                                ce->ce_flags &= ~CE_VALID;
 993                                istate->cache_changed = 1;
 994                        }
 995                        if (quiet)
 996                                continue;
 997                        printf("%s: needs update\n", ce->name);
 998                        has_errors = 1;
 999                        continue;
1000                }
1001
1002                replace_index_entry(istate, i, new);
1003        }
1004        return has_errors;
1005}
1006
1007struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1008{
1009        return refresh_cache_ent(&the_index, ce, really, NULL);
1010}
1011
1012static int verify_hdr(struct cache_header *hdr, unsigned long size)
1013{
1014        SHA_CTX c;
1015        unsigned char sha1[20];
1016
1017        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1018                return error("bad signature");
1019        if (hdr->hdr_version != htonl(2))
1020                return error("bad index version");
1021        SHA1_Init(&c);
1022        SHA1_Update(&c, hdr, size - 20);
1023        SHA1_Final(sha1, &c);
1024        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1025                return error("bad index file sha1 signature");
1026        return 0;
1027}
1028
1029static int read_index_extension(struct index_state *istate,
1030                                const char *ext, void *data, unsigned long sz)
1031{
1032        switch (CACHE_EXT(ext)) {
1033        case CACHE_EXT_TREE:
1034                istate->cache_tree = cache_tree_read(data, sz);
1035                break;
1036        default:
1037                if (*ext < 'A' || 'Z' < *ext)
1038                        return error("index uses %.4s extension, which we do not understand",
1039                                     ext);
1040                fprintf(stderr, "ignoring %.4s extension\n", ext);
1041                break;
1042        }
1043        return 0;
1044}
1045
1046int read_index(struct index_state *istate)
1047{
1048        return read_index_from(istate, get_index_file());
1049}
1050
1051static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1052{
1053        size_t len;
1054
1055        ce->ce_ctime = ntohl(ondisk->ctime.sec);
1056        ce->ce_mtime = ntohl(ondisk->mtime.sec);
1057        ce->ce_dev   = ntohl(ondisk->dev);
1058        ce->ce_ino   = ntohl(ondisk->ino);
1059        ce->ce_mode  = ntohl(ondisk->mode);
1060        ce->ce_uid   = ntohl(ondisk->uid);
1061        ce->ce_gid   = ntohl(ondisk->gid);
1062        ce->ce_size  = ntohl(ondisk->size);
1063        /* On-disk flags are just 16 bits */
1064        ce->ce_flags = ntohs(ondisk->flags);
1065        hashcpy(ce->sha1, ondisk->sha1);
1066
1067        len = ce->ce_flags & CE_NAMEMASK;
1068        if (len == CE_NAMEMASK)
1069                len = strlen(ondisk->name);
1070        /*
1071         * NEEDSWORK: If the original index is crafted, this copy could
1072         * go unchecked.
1073         */
1074        memcpy(ce->name, ondisk->name, len + 1);
1075}
1076
1077static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1078{
1079        long per_entry;
1080
1081        per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1082
1083        /*
1084         * Alignment can cause differences. This should be "alignof", but
1085         * since that's a gcc'ism, just use the size of a pointer.
1086         */
1087        per_entry += sizeof(void *);
1088        return ondisk_size + entries*per_entry;
1089}
1090
1091/* remember to discard_cache() before reading a different cache! */
1092int read_index_from(struct index_state *istate, const char *path)
1093{
1094        int fd, i;
1095        struct stat st;
1096        unsigned long src_offset, dst_offset;
1097        struct cache_header *hdr;
1098        void *mmap;
1099        size_t mmap_size;
1100
1101        errno = EBUSY;
1102        if (istate->alloc)
1103                return istate->cache_nr;
1104
1105        errno = ENOENT;
1106        istate->timestamp = 0;
1107        fd = open(path, O_RDONLY);
1108        if (fd < 0) {
1109                if (errno == ENOENT)
1110                        return 0;
1111                die("index file open failed (%s)", strerror(errno));
1112        }
1113
1114        if (fstat(fd, &st))
1115                die("cannot stat the open index (%s)", strerror(errno));
1116
1117        errno = EINVAL;
1118        mmap_size = xsize_t(st.st_size);
1119        if (mmap_size < sizeof(struct cache_header) + 20)
1120                die("index file smaller than expected");
1121
1122        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1123        close(fd);
1124        if (mmap == MAP_FAILED)
1125                die("unable to map index file");
1126
1127        hdr = mmap;
1128        if (verify_hdr(hdr, mmap_size) < 0)
1129                goto unmap;
1130
1131        istate->cache_nr = ntohl(hdr->hdr_entries);
1132        istate->cache_alloc = alloc_nr(istate->cache_nr);
1133        istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1134
1135        /*
1136         * The disk format is actually larger than the in-memory format,
1137         * due to space for nsec etc, so even though the in-memory one
1138         * has room for a few  more flags, we can allocate using the same
1139         * index size
1140         */
1141        istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1142
1143        src_offset = sizeof(*hdr);
1144        dst_offset = 0;
1145        for (i = 0; i < istate->cache_nr; i++) {
1146                struct ondisk_cache_entry *disk_ce;
1147                struct cache_entry *ce;
1148
1149                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1150                ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1151                convert_from_disk(disk_ce, ce);
1152                set_index_entry(istate, i, ce);
1153
1154                src_offset += ondisk_ce_size(ce);
1155                dst_offset += ce_size(ce);
1156        }
1157        istate->timestamp = st.st_mtime;
1158        while (src_offset <= mmap_size - 20 - 8) {
1159                /* After an array of active_nr index entries,
1160                 * there can be arbitrary number of extended
1161                 * sections, each of which is prefixed with
1162                 * extension name (4-byte) and section length
1163                 * in 4-byte network byte order.
1164                 */
1165                unsigned long extsize;
1166                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1167                extsize = ntohl(extsize);
1168                if (read_index_extension(istate,
1169                                         (const char *) mmap + src_offset,
1170                                         (char *) mmap + src_offset + 8,
1171                                         extsize) < 0)
1172                        goto unmap;
1173                src_offset += 8;
1174                src_offset += extsize;
1175        }
1176        munmap(mmap, mmap_size);
1177        return istate->cache_nr;
1178
1179unmap:
1180        munmap(mmap, mmap_size);
1181        errno = EINVAL;
1182        die("index file corrupt");
1183}
1184
1185int discard_index(struct index_state *istate)
1186{
1187        istate->cache_nr = 0;
1188        istate->cache_changed = 0;
1189        istate->timestamp = 0;
1190        free_hash(&istate->name_hash);
1191        cache_tree_free(&(istate->cache_tree));
1192        free(istate->alloc);
1193        istate->alloc = NULL;
1194
1195        /* no need to throw away allocated active_cache */
1196        return 0;
1197}
1198
1199int unmerged_index(const struct index_state *istate)
1200{
1201        int i;
1202        for (i = 0; i < istate->cache_nr; i++) {
1203                if (ce_stage(istate->cache[i]))
1204                        return 1;
1205        }
1206        return 0;
1207}
1208
1209#define WRITE_BUFFER_SIZE 8192
1210static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1211static unsigned long write_buffer_len;
1212
1213static int ce_write_flush(SHA_CTX *context, int fd)
1214{
1215        unsigned int buffered = write_buffer_len;
1216        if (buffered) {
1217                SHA1_Update(context, write_buffer, buffered);
1218                if (write_in_full(fd, write_buffer, buffered) != buffered)
1219                        return -1;
1220                write_buffer_len = 0;
1221        }
1222        return 0;
1223}
1224
1225static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1226{
1227        while (len) {
1228                unsigned int buffered = write_buffer_len;
1229                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1230                if (partial > len)
1231                        partial = len;
1232                memcpy(write_buffer + buffered, data, partial);
1233                buffered += partial;
1234                if (buffered == WRITE_BUFFER_SIZE) {
1235                        write_buffer_len = buffered;
1236                        if (ce_write_flush(context, fd))
1237                                return -1;
1238                        buffered = 0;
1239                }
1240                write_buffer_len = buffered;
1241                len -= partial;
1242                data = (char *) data + partial;
1243        }
1244        return 0;
1245}
1246
1247static int write_index_ext_header(SHA_CTX *context, int fd,
1248                                  unsigned int ext, unsigned int sz)
1249{
1250        ext = htonl(ext);
1251        sz = htonl(sz);
1252        return ((ce_write(context, fd, &ext, 4) < 0) ||
1253                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1254}
1255
1256static int ce_flush(SHA_CTX *context, int fd)
1257{
1258        unsigned int left = write_buffer_len;
1259
1260        if (left) {
1261                write_buffer_len = 0;
1262                SHA1_Update(context, write_buffer, left);
1263        }
1264
1265        /* Flush first if not enough space for SHA1 signature */
1266        if (left + 20 > WRITE_BUFFER_SIZE) {
1267                if (write_in_full(fd, write_buffer, left) != left)
1268                        return -1;
1269                left = 0;
1270        }
1271
1272        /* Append the SHA1 signature at the end */
1273        SHA1_Final(write_buffer + left, context);
1274        left += 20;
1275        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1276}
1277
1278static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1279{
1280        /*
1281         * The only thing we care about in this function is to smudge the
1282         * falsely clean entry due to touch-update-touch race, so we leave
1283         * everything else as they are.  We are called for entries whose
1284         * ce_mtime match the index file mtime.
1285         */
1286        struct stat st;
1287
1288        if (lstat(ce->name, &st) < 0)
1289                return;
1290        if (ce_match_stat_basic(ce, &st))
1291                return;
1292        if (ce_modified_check_fs(ce, &st)) {
1293                /* This is "racily clean"; smudge it.  Note that this
1294                 * is a tricky code.  At first glance, it may appear
1295                 * that it can break with this sequence:
1296                 *
1297                 * $ echo xyzzy >frotz
1298                 * $ git-update-index --add frotz
1299                 * $ : >frotz
1300                 * $ sleep 3
1301                 * $ echo filfre >nitfol
1302                 * $ git-update-index --add nitfol
1303                 *
1304                 * but it does not.  When the second update-index runs,
1305                 * it notices that the entry "frotz" has the same timestamp
1306                 * as index, and if we were to smudge it by resetting its
1307                 * size to zero here, then the object name recorded
1308                 * in index is the 6-byte file but the cached stat information
1309                 * becomes zero --- which would then match what we would
1310                 * obtain from the filesystem next time we stat("frotz").
1311                 *
1312                 * However, the second update-index, before calling
1313                 * this function, notices that the cached size is 6
1314                 * bytes and what is on the filesystem is an empty
1315                 * file, and never calls us, so the cached size information
1316                 * for "frotz" stays 6 which does not match the filesystem.
1317                 */
1318                ce->ce_size = 0;
1319        }
1320}
1321
1322static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1323{
1324        int size = ondisk_ce_size(ce);
1325        struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1326
1327        ondisk->ctime.sec = htonl(ce->ce_ctime);
1328        ondisk->ctime.nsec = 0;
1329        ondisk->mtime.sec = htonl(ce->ce_mtime);
1330        ondisk->mtime.nsec = 0;
1331        ondisk->dev  = htonl(ce->ce_dev);
1332        ondisk->ino  = htonl(ce->ce_ino);
1333        ondisk->mode = htonl(ce->ce_mode);
1334        ondisk->uid  = htonl(ce->ce_uid);
1335        ondisk->gid  = htonl(ce->ce_gid);
1336        ondisk->size = htonl(ce->ce_size);
1337        hashcpy(ondisk->sha1, ce->sha1);
1338        ondisk->flags = htons(ce->ce_flags);
1339        memcpy(ondisk->name, ce->name, ce_namelen(ce));
1340
1341        return ce_write(c, fd, ondisk, size);
1342}
1343
1344int write_index(const struct index_state *istate, int newfd)
1345{
1346        SHA_CTX c;
1347        struct cache_header hdr;
1348        int i, err, removed;
1349        struct cache_entry **cache = istate->cache;
1350        int entries = istate->cache_nr;
1351
1352        for (i = removed = 0; i < entries; i++)
1353                if (cache[i]->ce_flags & CE_REMOVE)
1354                        removed++;
1355
1356        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1357        hdr.hdr_version = htonl(2);
1358        hdr.hdr_entries = htonl(entries - removed);
1359
1360        SHA1_Init(&c);
1361        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1362                return -1;
1363
1364        for (i = 0; i < entries; i++) {
1365                struct cache_entry *ce = cache[i];
1366                if (ce->ce_flags & CE_REMOVE)
1367                        continue;
1368                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1369                        ce_smudge_racily_clean_entry(ce);
1370                if (ce_write_entry(&c, newfd, ce) < 0)
1371                        return -1;
1372        }
1373
1374        /* Write extension data here */
1375        if (istate->cache_tree) {
1376                struct strbuf sb;
1377
1378                strbuf_init(&sb, 0);
1379                cache_tree_write(&sb, istate->cache_tree);
1380                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1381                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1382                strbuf_release(&sb);
1383                if (err)
1384                        return -1;
1385        }
1386        return ce_flush(&c, newfd);
1387}