read-cache.con commit Merge branch 'rs/clean-menu-item-defn' into maint (caea1a2)
   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#include "tree.h"
  12#include "commit.h"
  13#include "blob.h"
  14#include "resolve-undo.h"
  15#include "strbuf.h"
  16#include "varint.h"
  17#include "split-index.h"
  18#include "sigchain.h"
  19
  20static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
  21                                               unsigned int options);
  22
  23/* Mask for the name length in ce_flags in the on-disk index */
  24
  25#define CE_NAMEMASK  (0x0fff)
  26
  27/* Index extensions.
  28 *
  29 * The first letter should be 'A'..'Z' for extensions that are not
  30 * necessary for a correct operation (i.e. optimization data).
  31 * When new extensions are added that _needs_ to be understood in
  32 * order to correctly interpret the index file, pick character that
  33 * is outside the range, to cause the reader to abort.
  34 */
  35
  36#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  37#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  38#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  39#define CACHE_EXT_LINK 0x6c696e6b         /* "link" */
  40
  41/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
  42#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
  43                 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
  44                 SPLIT_INDEX_ORDERED)
  45
  46struct index_state the_index;
  47static const char *alternate_index_output;
  48
  49static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  50{
  51        istate->cache[nr] = ce;
  52        add_name_hash(istate, ce);
  53}
  54
  55static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  56{
  57        struct cache_entry *old = istate->cache[nr];
  58
  59        replace_index_entry_in_base(istate, old, ce);
  60        remove_name_hash(istate, old);
  61        free(old);
  62        set_index_entry(istate, nr, ce);
  63        ce->ce_flags |= CE_UPDATE_IN_BASE;
  64        istate->cache_changed |= CE_ENTRY_CHANGED;
  65}
  66
  67void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  68{
  69        struct cache_entry *old = istate->cache[nr], *new;
  70        int namelen = strlen(new_name);
  71
  72        new = xmalloc(cache_entry_size(namelen));
  73        copy_cache_entry(new, old);
  74        new->ce_flags &= ~CE_HASHED;
  75        new->ce_namelen = namelen;
  76        new->index = 0;
  77        memcpy(new->name, new_name, namelen + 1);
  78
  79        cache_tree_invalidate_path(istate, old->name);
  80        remove_index_entry_at(istate, nr);
  81        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  82}
  83
  84void fill_stat_data(struct stat_data *sd, struct stat *st)
  85{
  86        sd->sd_ctime.sec = (unsigned int)st->st_ctime;
  87        sd->sd_mtime.sec = (unsigned int)st->st_mtime;
  88        sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
  89        sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
  90        sd->sd_dev = st->st_dev;
  91        sd->sd_ino = st->st_ino;
  92        sd->sd_uid = st->st_uid;
  93        sd->sd_gid = st->st_gid;
  94        sd->sd_size = st->st_size;
  95}
  96
  97int match_stat_data(const struct stat_data *sd, struct stat *st)
  98{
  99        int changed = 0;
 100
 101        if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
 102                changed |= MTIME_CHANGED;
 103        if (trust_ctime && check_stat &&
 104            sd->sd_ctime.sec != (unsigned int)st->st_ctime)
 105                changed |= CTIME_CHANGED;
 106
 107#ifdef USE_NSEC
 108        if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
 109                changed |= MTIME_CHANGED;
 110        if (trust_ctime && check_stat &&
 111            sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
 112                changed |= CTIME_CHANGED;
 113#endif
 114
 115        if (check_stat) {
 116                if (sd->sd_uid != (unsigned int) st->st_uid ||
 117                        sd->sd_gid != (unsigned int) st->st_gid)
 118                        changed |= OWNER_CHANGED;
 119                if (sd->sd_ino != (unsigned int) st->st_ino)
 120                        changed |= INODE_CHANGED;
 121        }
 122
 123#ifdef USE_STDEV
 124        /*
 125         * st_dev breaks on network filesystems where different
 126         * clients will have different views of what "device"
 127         * the filesystem is on
 128         */
 129        if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
 130                        changed |= INODE_CHANGED;
 131#endif
 132
 133        if (sd->sd_size != (unsigned int) st->st_size)
 134                changed |= DATA_CHANGED;
 135
 136        return changed;
 137}
 138
 139/*
 140 * This only updates the "non-critical" parts of the directory
 141 * cache, ie the parts that aren't tracked by GIT, and only used
 142 * to validate the cache.
 143 */
 144void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 145{
 146        fill_stat_data(&ce->ce_stat_data, st);
 147
 148        if (assume_unchanged)
 149                ce->ce_flags |= CE_VALID;
 150
 151        if (S_ISREG(st->st_mode))
 152                ce_mark_uptodate(ce);
 153}
 154
 155static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
 156{
 157        int match = -1;
 158        int fd = open(ce->name, O_RDONLY);
 159
 160        if (fd >= 0) {
 161                unsigned char sha1[20];
 162                if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
 163                        match = hashcmp(sha1, ce->sha1);
 164                /* index_fd() closed the file descriptor already */
 165        }
 166        return match;
 167}
 168
 169static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
 170{
 171        int match = -1;
 172        void *buffer;
 173        unsigned long size;
 174        enum object_type type;
 175        struct strbuf sb = STRBUF_INIT;
 176
 177        if (strbuf_readlink(&sb, ce->name, expected_size))
 178                return -1;
 179
 180        buffer = read_sha1_file(ce->sha1, &type, &size);
 181        if (buffer) {
 182                if (size == sb.len)
 183                        match = memcmp(buffer, sb.buf, size);
 184                free(buffer);
 185        }
 186        strbuf_release(&sb);
 187        return match;
 188}
 189
 190static int ce_compare_gitlink(const struct cache_entry *ce)
 191{
 192        unsigned char sha1[20];
 193
 194        /*
 195         * We don't actually require that the .git directory
 196         * under GITLINK directory be a valid git directory. It
 197         * might even be missing (in case nobody populated that
 198         * sub-project).
 199         *
 200         * If so, we consider it always to match.
 201         */
 202        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 203                return 0;
 204        return hashcmp(sha1, ce->sha1);
 205}
 206
 207static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
 208{
 209        switch (st->st_mode & S_IFMT) {
 210        case S_IFREG:
 211                if (ce_compare_data(ce, st))
 212                        return DATA_CHANGED;
 213                break;
 214        case S_IFLNK:
 215                if (ce_compare_link(ce, xsize_t(st->st_size)))
 216                        return DATA_CHANGED;
 217                break;
 218        case S_IFDIR:
 219                if (S_ISGITLINK(ce->ce_mode))
 220                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 221        default:
 222                return TYPE_CHANGED;
 223        }
 224        return 0;
 225}
 226
 227static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
 228{
 229        unsigned int changed = 0;
 230
 231        if (ce->ce_flags & CE_REMOVE)
 232                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 233
 234        switch (ce->ce_mode & S_IFMT) {
 235        case S_IFREG:
 236                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 237                /* We consider only the owner x bit to be relevant for
 238                 * "mode changes"
 239                 */
 240                if (trust_executable_bit &&
 241                    (0100 & (ce->ce_mode ^ st->st_mode)))
 242                        changed |= MODE_CHANGED;
 243                break;
 244        case S_IFLNK:
 245                if (!S_ISLNK(st->st_mode) &&
 246                    (has_symlinks || !S_ISREG(st->st_mode)))
 247                        changed |= TYPE_CHANGED;
 248                break;
 249        case S_IFGITLINK:
 250                /* We ignore most of the st_xxx fields for gitlinks */
 251                if (!S_ISDIR(st->st_mode))
 252                        changed |= TYPE_CHANGED;
 253                else if (ce_compare_gitlink(ce))
 254                        changed |= DATA_CHANGED;
 255                return changed;
 256        default:
 257                die("internal error: ce_mode is %o", ce->ce_mode);
 258        }
 259
 260        changed |= match_stat_data(&ce->ce_stat_data, st);
 261
 262        /* Racily smudged entry? */
 263        if (!ce->ce_stat_data.sd_size) {
 264                if (!is_empty_blob_sha1(ce->sha1))
 265                        changed |= DATA_CHANGED;
 266        }
 267
 268        return changed;
 269}
 270
 271static int is_racy_timestamp(const struct index_state *istate,
 272                             const struct cache_entry *ce)
 273{
 274        return (!S_ISGITLINK(ce->ce_mode) &&
 275                istate->timestamp.sec &&
 276#ifdef USE_NSEC
 277                 /* nanosecond timestamped files can also be racy! */
 278                (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
 279                 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
 280                  istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
 281#else
 282                istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
 283#endif
 284                 );
 285}
 286
 287int ie_match_stat(const struct index_state *istate,
 288                  const struct cache_entry *ce, struct stat *st,
 289                  unsigned int options)
 290{
 291        unsigned int changed;
 292        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 293        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 294        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 295
 296        /*
 297         * If it's marked as always valid in the index, it's
 298         * valid whatever the checked-out copy says.
 299         *
 300         * skip-worktree has the same effect with higher precedence
 301         */
 302        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 303                return 0;
 304        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 305                return 0;
 306
 307        /*
 308         * Intent-to-add entries have not been added, so the index entry
 309         * by definition never matches what is in the work tree until it
 310         * actually gets added.
 311         */
 312        if (ce->ce_flags & CE_INTENT_TO_ADD)
 313                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 314
 315        changed = ce_match_stat_basic(ce, st);
 316
 317        /*
 318         * Within 1 second of this sequence:
 319         *      echo xyzzy >file && git-update-index --add file
 320         * running this command:
 321         *      echo frotz >file
 322         * would give a falsely clean cache entry.  The mtime and
 323         * length match the cache, and other stat fields do not change.
 324         *
 325         * We could detect this at update-index time (the cache entry
 326         * being registered/updated records the same time as "now")
 327         * and delay the return from git-update-index, but that would
 328         * effectively mean we can make at most one commit per second,
 329         * which is not acceptable.  Instead, we check cache entries
 330         * whose mtime are the same as the index file timestamp more
 331         * carefully than others.
 332         */
 333        if (!changed && is_racy_timestamp(istate, ce)) {
 334                if (assume_racy_is_modified)
 335                        changed |= DATA_CHANGED;
 336                else
 337                        changed |= ce_modified_check_fs(ce, st);
 338        }
 339
 340        return changed;
 341}
 342
 343int ie_modified(const struct index_state *istate,
 344                const struct cache_entry *ce,
 345                struct stat *st, unsigned int options)
 346{
 347        int changed, changed_fs;
 348
 349        changed = ie_match_stat(istate, ce, st, options);
 350        if (!changed)
 351                return 0;
 352        /*
 353         * If the mode or type has changed, there's no point in trying
 354         * to refresh the entry - it's not going to match
 355         */
 356        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 357                return changed;
 358
 359        /*
 360         * Immediately after read-tree or update-index --cacheinfo,
 361         * the length field is zero, as we have never even read the
 362         * lstat(2) information once, and we cannot trust DATA_CHANGED
 363         * returned by ie_match_stat() which in turn was returned by
 364         * ce_match_stat_basic() to signal that the filesize of the
 365         * blob changed.  We have to actually go to the filesystem to
 366         * see if the contents match, and if so, should answer "unchanged".
 367         *
 368         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 369         * already has checked the actual HEAD from the filesystem in the
 370         * subproject.  If ie_match_stat() already said it is different,
 371         * then we know it is.
 372         */
 373        if ((changed & DATA_CHANGED) &&
 374            (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
 375                return changed;
 376
 377        changed_fs = ce_modified_check_fs(ce, st);
 378        if (changed_fs)
 379                return changed | changed_fs;
 380        return 0;
 381}
 382
 383int base_name_compare(const char *name1, int len1, int mode1,
 384                      const char *name2, int len2, int mode2)
 385{
 386        unsigned char c1, c2;
 387        int len = len1 < len2 ? len1 : len2;
 388        int cmp;
 389
 390        cmp = memcmp(name1, name2, len);
 391        if (cmp)
 392                return cmp;
 393        c1 = name1[len];
 394        c2 = name2[len];
 395        if (!c1 && S_ISDIR(mode1))
 396                c1 = '/';
 397        if (!c2 && S_ISDIR(mode2))
 398                c2 = '/';
 399        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 400}
 401
 402/*
 403 * df_name_compare() is identical to base_name_compare(), except it
 404 * compares conflicting directory/file entries as equal. Note that
 405 * while a directory name compares as equal to a regular file, they
 406 * then individually compare _differently_ to a filename that has
 407 * a dot after the basename (because '\0' < '.' < '/').
 408 *
 409 * This is used by routines that want to traverse the git namespace
 410 * but then handle conflicting entries together when possible.
 411 */
 412int df_name_compare(const char *name1, int len1, int mode1,
 413                    const char *name2, int len2, int mode2)
 414{
 415        int len = len1 < len2 ? len1 : len2, cmp;
 416        unsigned char c1, c2;
 417
 418        cmp = memcmp(name1, name2, len);
 419        if (cmp)
 420                return cmp;
 421        /* Directories and files compare equal (same length, same name) */
 422        if (len1 == len2)
 423                return 0;
 424        c1 = name1[len];
 425        if (!c1 && S_ISDIR(mode1))
 426                c1 = '/';
 427        c2 = name2[len];
 428        if (!c2 && S_ISDIR(mode2))
 429                c2 = '/';
 430        if (c1 == '/' && !c2)
 431                return 0;
 432        if (c2 == '/' && !c1)
 433                return 0;
 434        return c1 - c2;
 435}
 436
 437int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
 438{
 439        size_t min_len = (len1 < len2) ? len1 : len2;
 440        int cmp = memcmp(name1, name2, min_len);
 441        if (cmp)
 442                return cmp;
 443        if (len1 < len2)
 444                return -1;
 445        if (len1 > len2)
 446                return 1;
 447        return 0;
 448}
 449
 450int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 451{
 452        int cmp;
 453
 454        cmp = name_compare(name1, len1, name2, len2);
 455        if (cmp)
 456                return cmp;
 457
 458        if (stage1 < stage2)
 459                return -1;
 460        if (stage1 > stage2)
 461                return 1;
 462        return 0;
 463}
 464
 465static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 466{
 467        int first, last;
 468
 469        first = 0;
 470        last = istate->cache_nr;
 471        while (last > first) {
 472                int next = (last + first) >> 1;
 473                struct cache_entry *ce = istate->cache[next];
 474                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 475                if (!cmp)
 476                        return next;
 477                if (cmp < 0) {
 478                        last = next;
 479                        continue;
 480                }
 481                first = next+1;
 482        }
 483        return -first-1;
 484}
 485
 486int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 487{
 488        return index_name_stage_pos(istate, name, namelen, 0);
 489}
 490
 491/* Remove entry, return true if there are more entries to go.. */
 492int remove_index_entry_at(struct index_state *istate, int pos)
 493{
 494        struct cache_entry *ce = istate->cache[pos];
 495
 496        record_resolve_undo(istate, ce);
 497        remove_name_hash(istate, ce);
 498        save_or_free_index_entry(istate, ce);
 499        istate->cache_changed |= CE_ENTRY_REMOVED;
 500        istate->cache_nr--;
 501        if (pos >= istate->cache_nr)
 502                return 0;
 503        memmove(istate->cache + pos,
 504                istate->cache + pos + 1,
 505                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 506        return 1;
 507}
 508
 509/*
 510 * Remove all cache entries marked for removal, that is where
 511 * CE_REMOVE is set in ce_flags.  This is much more effective than
 512 * calling remove_index_entry_at() for each entry to be removed.
 513 */
 514void remove_marked_cache_entries(struct index_state *istate)
 515{
 516        struct cache_entry **ce_array = istate->cache;
 517        unsigned int i, j;
 518
 519        for (i = j = 0; i < istate->cache_nr; i++) {
 520                if (ce_array[i]->ce_flags & CE_REMOVE) {
 521                        remove_name_hash(istate, ce_array[i]);
 522                        save_or_free_index_entry(istate, ce_array[i]);
 523                }
 524                else
 525                        ce_array[j++] = ce_array[i];
 526        }
 527        if (j == istate->cache_nr)
 528                return;
 529        istate->cache_changed |= CE_ENTRY_REMOVED;
 530        istate->cache_nr = j;
 531}
 532
 533int remove_file_from_index(struct index_state *istate, const char *path)
 534{
 535        int pos = index_name_pos(istate, path, strlen(path));
 536        if (pos < 0)
 537                pos = -pos-1;
 538        cache_tree_invalidate_path(istate, path);
 539        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 540                remove_index_entry_at(istate, pos);
 541        return 0;
 542}
 543
 544static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 545{
 546        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 547}
 548
 549static int index_name_pos_also_unmerged(struct index_state *istate,
 550        const char *path, int namelen)
 551{
 552        int pos = index_name_pos(istate, path, namelen);
 553        struct cache_entry *ce;
 554
 555        if (pos >= 0)
 556                return pos;
 557
 558        /* maybe unmerged? */
 559        pos = -1 - pos;
 560        if (pos >= istate->cache_nr ||
 561                        compare_name((ce = istate->cache[pos]), path, namelen))
 562                return -1;
 563
 564        /* order of preference: stage 2, 1, 3 */
 565        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 566                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 567                        !compare_name(ce, path, namelen))
 568                pos++;
 569        return pos;
 570}
 571
 572static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 573{
 574        int len = ce_namelen(ce);
 575        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 576}
 577
 578/*
 579 * If we add a filename that aliases in the cache, we will use the
 580 * name that we already have - but we don't want to update the same
 581 * alias twice, because that implies that there were actually two
 582 * different files with aliasing names!
 583 *
 584 * So we use the CE_ADDED flag to verify that the alias was an old
 585 * one before we accept it as
 586 */
 587static struct cache_entry *create_alias_ce(struct index_state *istate,
 588                                           struct cache_entry *ce,
 589                                           struct cache_entry *alias)
 590{
 591        int len;
 592        struct cache_entry *new;
 593
 594        if (alias->ce_flags & CE_ADDED)
 595                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 596
 597        /* Ok, create the new entry using the name of the existing alias */
 598        len = ce_namelen(alias);
 599        new = xcalloc(1, cache_entry_size(len));
 600        memcpy(new->name, alias->name, len);
 601        copy_cache_entry(new, ce);
 602        save_or_free_index_entry(istate, ce);
 603        return new;
 604}
 605
 606void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
 607{
 608        unsigned char sha1[20];
 609        if (write_sha1_file("", 0, blob_type, sha1))
 610                die("cannot create an empty blob in the object database");
 611        hashcpy(ce->sha1, sha1);
 612}
 613
 614int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 615{
 616        int size, namelen, was_same;
 617        mode_t st_mode = st->st_mode;
 618        struct cache_entry *ce, *alias;
 619        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 620        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 621        int pretend = flags & ADD_CACHE_PRETEND;
 622        int intent_only = flags & ADD_CACHE_INTENT;
 623        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 624                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 625
 626        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 627                return error("%s: can only add regular files, symbolic links or git-directories", path);
 628
 629        namelen = strlen(path);
 630        if (S_ISDIR(st_mode)) {
 631                while (namelen && path[namelen-1] == '/')
 632                        namelen--;
 633        }
 634        size = cache_entry_size(namelen);
 635        ce = xcalloc(1, size);
 636        memcpy(ce->name, path, namelen);
 637        ce->ce_namelen = namelen;
 638        if (!intent_only)
 639                fill_stat_cache_info(ce, st);
 640        else
 641                ce->ce_flags |= CE_INTENT_TO_ADD;
 642
 643        if (trust_executable_bit && has_symlinks)
 644                ce->ce_mode = create_ce_mode(st_mode);
 645        else {
 646                /* If there is an existing entry, pick the mode bits and type
 647                 * from it, otherwise assume unexecutable regular file.
 648                 */
 649                struct cache_entry *ent;
 650                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 651
 652                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 653                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 654        }
 655
 656        /* When core.ignorecase=true, determine if a directory of the same name but differing
 657         * case already exists within the Git repository.  If it does, ensure the directory
 658         * case of the file being added to the repository matches (is folded into) the existing
 659         * entry's directory case.
 660         */
 661        if (ignore_case) {
 662                const char *startPtr = ce->name;
 663                const char *ptr = startPtr;
 664                while (*ptr) {
 665                        while (*ptr && *ptr != '/')
 666                                ++ptr;
 667                        if (*ptr == '/') {
 668                                struct cache_entry *foundce;
 669                                ++ptr;
 670                                foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
 671                                if (foundce) {
 672                                        memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
 673                                        startPtr = ptr;
 674                                }
 675                        }
 676                }
 677        }
 678
 679        alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 680        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 681                /* Nothing changed, really */
 682                free(ce);
 683                if (!S_ISGITLINK(alias->ce_mode))
 684                        ce_mark_uptodate(alias);
 685                alias->ce_flags |= CE_ADDED;
 686                return 0;
 687        }
 688        if (!intent_only) {
 689                if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
 690                        return error("unable to index file %s", path);
 691        } else
 692                set_object_name_for_intent_to_add_entry(ce);
 693
 694        if (ignore_case && alias && different_name(ce, alias))
 695                ce = create_alias_ce(istate, ce, alias);
 696        ce->ce_flags |= CE_ADDED;
 697
 698        /* It was suspected to be racily clean, but it turns out to be Ok */
 699        was_same = (alias &&
 700                    !ce_stage(alias) &&
 701                    !hashcmp(alias->sha1, ce->sha1) &&
 702                    ce->ce_mode == alias->ce_mode);
 703
 704        if (pretend)
 705                ;
 706        else if (add_index_entry(istate, ce, add_option))
 707                return error("unable to add %s to index",path);
 708        if (verbose && !was_same)
 709                printf("add '%s'\n", path);
 710        return 0;
 711}
 712
 713int add_file_to_index(struct index_state *istate, const char *path, int flags)
 714{
 715        struct stat st;
 716        if (lstat(path, &st))
 717                die_errno("unable to stat '%s'", path);
 718        return add_to_index(istate, path, &st, flags);
 719}
 720
 721struct cache_entry *make_cache_entry(unsigned int mode,
 722                const unsigned char *sha1, const char *path, int stage,
 723                unsigned int refresh_options)
 724{
 725        int size, len;
 726        struct cache_entry *ce;
 727
 728        if (!verify_path(path)) {
 729                error("Invalid path '%s'", path);
 730                return NULL;
 731        }
 732
 733        len = strlen(path);
 734        size = cache_entry_size(len);
 735        ce = xcalloc(1, size);
 736
 737        hashcpy(ce->sha1, sha1);
 738        memcpy(ce->name, path, len);
 739        ce->ce_flags = create_ce_flags(stage);
 740        ce->ce_namelen = len;
 741        ce->ce_mode = create_ce_mode(mode);
 742
 743        return refresh_cache_entry(ce, refresh_options);
 744}
 745
 746int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
 747{
 748        int len = ce_namelen(a);
 749        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 750}
 751
 752/*
 753 * We fundamentally don't like some paths: we don't want
 754 * dot or dot-dot anywhere, and for obvious reasons don't
 755 * want to recurse into ".git" either.
 756 *
 757 * Also, we don't want double slashes or slashes at the
 758 * end that can make pathnames ambiguous.
 759 */
 760static int verify_dotfile(const char *rest)
 761{
 762        /*
 763         * The first character was '.', but that
 764         * has already been discarded, we now test
 765         * the rest.
 766         */
 767
 768        /* "." is not allowed */
 769        if (*rest == '\0' || is_dir_sep(*rest))
 770                return 0;
 771
 772        switch (*rest) {
 773        /*
 774         * ".git" followed by  NUL or slash is bad. This
 775         * shares the path end test with the ".." case.
 776         */
 777        case 'g':
 778                if (rest[1] != 'i')
 779                        break;
 780                if (rest[2] != 't')
 781                        break;
 782                rest += 2;
 783        /* fallthrough */
 784        case '.':
 785                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 786                        return 0;
 787        }
 788        return 1;
 789}
 790
 791int verify_path(const char *path)
 792{
 793        char c;
 794
 795        if (has_dos_drive_prefix(path))
 796                return 0;
 797
 798        goto inside;
 799        for (;;) {
 800                if (!c)
 801                        return 1;
 802                if (is_dir_sep(c)) {
 803inside:
 804                        c = *path++;
 805                        if ((c == '.' && !verify_dotfile(path)) ||
 806                            is_dir_sep(c) || c == '\0')
 807                                return 0;
 808                }
 809                c = *path++;
 810        }
 811}
 812
 813/*
 814 * Do we have another file that has the beginning components being a
 815 * proper superset of the name we're trying to add?
 816 */
 817static int has_file_name(struct index_state *istate,
 818                         const struct cache_entry *ce, int pos, int ok_to_replace)
 819{
 820        int retval = 0;
 821        int len = ce_namelen(ce);
 822        int stage = ce_stage(ce);
 823        const char *name = ce->name;
 824
 825        while (pos < istate->cache_nr) {
 826                struct cache_entry *p = istate->cache[pos++];
 827
 828                if (len >= ce_namelen(p))
 829                        break;
 830                if (memcmp(name, p->name, len))
 831                        break;
 832                if (ce_stage(p) != stage)
 833                        continue;
 834                if (p->name[len] != '/')
 835                        continue;
 836                if (p->ce_flags & CE_REMOVE)
 837                        continue;
 838                retval = -1;
 839                if (!ok_to_replace)
 840                        break;
 841                remove_index_entry_at(istate, --pos);
 842        }
 843        return retval;
 844}
 845
 846/*
 847 * Do we have another file with a pathname that is a proper
 848 * subset of the name we're trying to add?
 849 */
 850static int has_dir_name(struct index_state *istate,
 851                        const struct cache_entry *ce, int pos, int ok_to_replace)
 852{
 853        int retval = 0;
 854        int stage = ce_stage(ce);
 855        const char *name = ce->name;
 856        const char *slash = name + ce_namelen(ce);
 857
 858        for (;;) {
 859                int len;
 860
 861                for (;;) {
 862                        if (*--slash == '/')
 863                                break;
 864                        if (slash <= ce->name)
 865                                return retval;
 866                }
 867                len = slash - name;
 868
 869                pos = index_name_stage_pos(istate, name, len, stage);
 870                if (pos >= 0) {
 871                        /*
 872                         * Found one, but not so fast.  This could
 873                         * be a marker that says "I was here, but
 874                         * I am being removed".  Such an entry is
 875                         * not a part of the resulting tree, and
 876                         * it is Ok to have a directory at the same
 877                         * path.
 878                         */
 879                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 880                                retval = -1;
 881                                if (!ok_to_replace)
 882                                        break;
 883                                remove_index_entry_at(istate, pos);
 884                                continue;
 885                        }
 886                }
 887                else
 888                        pos = -pos-1;
 889
 890                /*
 891                 * Trivial optimization: if we find an entry that
 892                 * already matches the sub-directory, then we know
 893                 * we're ok, and we can exit.
 894                 */
 895                while (pos < istate->cache_nr) {
 896                        struct cache_entry *p = istate->cache[pos];
 897                        if ((ce_namelen(p) <= len) ||
 898                            (p->name[len] != '/') ||
 899                            memcmp(p->name, name, len))
 900                                break; /* not our subdirectory */
 901                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 902                                /*
 903                                 * p is at the same stage as our entry, and
 904                                 * is a subdirectory of what we are looking
 905                                 * at, so we cannot have conflicts at our
 906                                 * level or anything shorter.
 907                                 */
 908                                return retval;
 909                        pos++;
 910                }
 911        }
 912        return retval;
 913}
 914
 915/* We may be in a situation where we already have path/file and path
 916 * is being added, or we already have path and path/file is being
 917 * added.  Either one would result in a nonsense tree that has path
 918 * twice when git-write-tree tries to write it out.  Prevent it.
 919 *
 920 * If ok-to-replace is specified, we remove the conflicting entries
 921 * from the cache so the caller should recompute the insert position.
 922 * When this happens, we return non-zero.
 923 */
 924static int check_file_directory_conflict(struct index_state *istate,
 925                                         const struct cache_entry *ce,
 926                                         int pos, int ok_to_replace)
 927{
 928        int retval;
 929
 930        /*
 931         * When ce is an "I am going away" entry, we allow it to be added
 932         */
 933        if (ce->ce_flags & CE_REMOVE)
 934                return 0;
 935
 936        /*
 937         * We check if the path is a sub-path of a subsequent pathname
 938         * first, since removing those will not change the position
 939         * in the array.
 940         */
 941        retval = has_file_name(istate, ce, pos, ok_to_replace);
 942
 943        /*
 944         * Then check if the path might have a clashing sub-directory
 945         * before it.
 946         */
 947        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 948}
 949
 950static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 951{
 952        int pos;
 953        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 954        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 955        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 956        int new_only = option & ADD_CACHE_NEW_ONLY;
 957
 958        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
 959                cache_tree_invalidate_path(istate, ce->name);
 960        pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 961
 962        /* existing match? Just replace it. */
 963        if (pos >= 0) {
 964                if (!new_only)
 965                        replace_index_entry(istate, pos, ce);
 966                return 0;
 967        }
 968        pos = -pos-1;
 969
 970        /*
 971         * Inserting a merged entry ("stage 0") into the index
 972         * will always replace all non-merged entries..
 973         */
 974        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 975                while (ce_same_name(istate->cache[pos], ce)) {
 976                        ok_to_add = 1;
 977                        if (!remove_index_entry_at(istate, pos))
 978                                break;
 979                }
 980        }
 981
 982        if (!ok_to_add)
 983                return -1;
 984        if (!verify_path(ce->name))
 985                return error("Invalid path '%s'", ce->name);
 986
 987        if (!skip_df_check &&
 988            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 989                if (!ok_to_replace)
 990                        return error("'%s' appears as both a file and as a directory",
 991                                     ce->name);
 992                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 993                pos = -pos-1;
 994        }
 995        return pos + 1;
 996}
 997
 998int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 999{
1000        int pos;
1001
1002        if (option & ADD_CACHE_JUST_APPEND)
1003                pos = istate->cache_nr;
1004        else {
1005                int ret;
1006                ret = add_index_entry_with_check(istate, ce, option);
1007                if (ret <= 0)
1008                        return ret;
1009                pos = ret - 1;
1010        }
1011
1012        /* Make sure the array is big enough .. */
1013        ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1014
1015        /* Add it in.. */
1016        istate->cache_nr++;
1017        if (istate->cache_nr > pos + 1)
1018                memmove(istate->cache + pos + 1,
1019                        istate->cache + pos,
1020                        (istate->cache_nr - pos - 1) * sizeof(ce));
1021        set_index_entry(istate, pos, ce);
1022        istate->cache_changed |= CE_ENTRY_ADDED;
1023        return 0;
1024}
1025
1026/*
1027 * "refresh" does not calculate a new sha1 file or bring the
1028 * cache up-to-date for mode/content changes. But what it
1029 * _does_ do is to "re-match" the stat information of a file
1030 * with the cache, so that you can refresh the cache for a
1031 * file that hasn't been changed but where the stat entry is
1032 * out of date.
1033 *
1034 * For example, you'd want to do this after doing a "git-read-tree",
1035 * to link up the stat cache details with the proper files.
1036 */
1037static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1038                                             struct cache_entry *ce,
1039                                             unsigned int options, int *err,
1040                                             int *changed_ret)
1041{
1042        struct stat st;
1043        struct cache_entry *updated;
1044        int changed, size;
1045        int refresh = options & CE_MATCH_REFRESH;
1046        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1047        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1048        int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1049
1050        if (!refresh || ce_uptodate(ce))
1051                return ce;
1052
1053        /*
1054         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1055         * that the change to the work tree does not matter and told
1056         * us not to worry.
1057         */
1058        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1059                ce_mark_uptodate(ce);
1060                return ce;
1061        }
1062        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1063                ce_mark_uptodate(ce);
1064                return ce;
1065        }
1066
1067        if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1068                if (ignore_missing)
1069                        return ce;
1070                if (err)
1071                        *err = ENOENT;
1072                return NULL;
1073        }
1074
1075        if (lstat(ce->name, &st) < 0) {
1076                if (ignore_missing && errno == ENOENT)
1077                        return ce;
1078                if (err)
1079                        *err = errno;
1080                return NULL;
1081        }
1082
1083        changed = ie_match_stat(istate, ce, &st, options);
1084        if (changed_ret)
1085                *changed_ret = changed;
1086        if (!changed) {
1087                /*
1088                 * The path is unchanged.  If we were told to ignore
1089                 * valid bit, then we did the actual stat check and
1090                 * found that the entry is unmodified.  If the entry
1091                 * is not marked VALID, this is the place to mark it
1092                 * valid again, under "assume unchanged" mode.
1093                 */
1094                if (ignore_valid && assume_unchanged &&
1095                    !(ce->ce_flags & CE_VALID))
1096                        ; /* mark this one VALID again */
1097                else {
1098                        /*
1099                         * We do not mark the index itself "modified"
1100                         * because CE_UPTODATE flag is in-core only;
1101                         * we are not going to write this change out.
1102                         */
1103                        if (!S_ISGITLINK(ce->ce_mode))
1104                                ce_mark_uptodate(ce);
1105                        return ce;
1106                }
1107        }
1108
1109        if (ie_modified(istate, ce, &st, options)) {
1110                if (err)
1111                        *err = EINVAL;
1112                return NULL;
1113        }
1114
1115        size = ce_size(ce);
1116        updated = xmalloc(size);
1117        memcpy(updated, ce, size);
1118        fill_stat_cache_info(updated, &st);
1119        /*
1120         * If ignore_valid is not set, we should leave CE_VALID bit
1121         * alone.  Otherwise, paths marked with --no-assume-unchanged
1122         * (i.e. things to be edited) will reacquire CE_VALID bit
1123         * automatically, which is not really what we want.
1124         */
1125        if (!ignore_valid && assume_unchanged &&
1126            !(ce->ce_flags & CE_VALID))
1127                updated->ce_flags &= ~CE_VALID;
1128
1129        /* istate->cache_changed is updated in the caller */
1130        return updated;
1131}
1132
1133static void show_file(const char * fmt, const char * name, int in_porcelain,
1134                      int * first, const char *header_msg)
1135{
1136        if (in_porcelain && *first && header_msg) {
1137                printf("%s\n", header_msg);
1138                *first = 0;
1139        }
1140        printf(fmt, name);
1141}
1142
1143int refresh_index(struct index_state *istate, unsigned int flags,
1144                  const struct pathspec *pathspec,
1145                  char *seen, const char *header_msg)
1146{
1147        int i;
1148        int has_errors = 0;
1149        int really = (flags & REFRESH_REALLY) != 0;
1150        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1151        int quiet = (flags & REFRESH_QUIET) != 0;
1152        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1153        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1154        int first = 1;
1155        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1156        unsigned int options = (CE_MATCH_REFRESH |
1157                                (really ? CE_MATCH_IGNORE_VALID : 0) |
1158                                (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1159        const char *modified_fmt;
1160        const char *deleted_fmt;
1161        const char *typechange_fmt;
1162        const char *added_fmt;
1163        const char *unmerged_fmt;
1164
1165        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1166        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1167        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1168        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1169        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1170        for (i = 0; i < istate->cache_nr; i++) {
1171                struct cache_entry *ce, *new;
1172                int cache_errno = 0;
1173                int changed = 0;
1174                int filtered = 0;
1175
1176                ce = istate->cache[i];
1177                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1178                        continue;
1179
1180                if (pathspec && !ce_path_match(ce, pathspec, seen))
1181                        filtered = 1;
1182
1183                if (ce_stage(ce)) {
1184                        while ((i < istate->cache_nr) &&
1185                               ! strcmp(istate->cache[i]->name, ce->name))
1186                                i++;
1187                        i--;
1188                        if (allow_unmerged)
1189                                continue;
1190                        if (!filtered)
1191                                show_file(unmerged_fmt, ce->name, in_porcelain,
1192                                          &first, header_msg);
1193                        has_errors = 1;
1194                        continue;
1195                }
1196
1197                if (filtered)
1198                        continue;
1199
1200                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1201                if (new == ce)
1202                        continue;
1203                if (!new) {
1204                        const char *fmt;
1205
1206                        if (really && cache_errno == EINVAL) {
1207                                /* If we are doing --really-refresh that
1208                                 * means the index is not valid anymore.
1209                                 */
1210                                ce->ce_flags &= ~CE_VALID;
1211                                ce->ce_flags |= CE_UPDATE_IN_BASE;
1212                                istate->cache_changed |= CE_ENTRY_CHANGED;
1213                        }
1214                        if (quiet)
1215                                continue;
1216
1217                        if (cache_errno == ENOENT)
1218                                fmt = deleted_fmt;
1219                        else if (ce->ce_flags & CE_INTENT_TO_ADD)
1220                                fmt = added_fmt; /* must be before other checks */
1221                        else if (changed & TYPE_CHANGED)
1222                                fmt = typechange_fmt;
1223                        else
1224                                fmt = modified_fmt;
1225                        show_file(fmt,
1226                                  ce->name, in_porcelain, &first, header_msg);
1227                        has_errors = 1;
1228                        continue;
1229                }
1230
1231                replace_index_entry(istate, i, new);
1232        }
1233        return has_errors;
1234}
1235
1236static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1237                                               unsigned int options)
1238{
1239        return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1240}
1241
1242
1243/*****************************************************************
1244 * Index File I/O
1245 *****************************************************************/
1246
1247#define INDEX_FORMAT_DEFAULT 3
1248
1249static int index_format_config(const char *var, const char *value, void *cb)
1250{
1251        unsigned int *version = cb;
1252        if (!strcmp(var, "index.version")) {
1253                *version = git_config_int(var, value);
1254                return 0;
1255        }
1256        return 1;
1257}
1258
1259static unsigned int get_index_format_default(void)
1260{
1261        char *envversion = getenv("GIT_INDEX_VERSION");
1262        char *endp;
1263        unsigned int version = INDEX_FORMAT_DEFAULT;
1264
1265        if (!envversion) {
1266                git_config(index_format_config, &version);
1267                if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1268                        warning(_("index.version set, but the value is invalid.\n"
1269                                  "Using version %i"), INDEX_FORMAT_DEFAULT);
1270                        return INDEX_FORMAT_DEFAULT;
1271                }
1272                return version;
1273        }
1274
1275        version = strtoul(envversion, &endp, 10);
1276        if (*endp ||
1277            version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1278                warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1279                          "Using version %i"), INDEX_FORMAT_DEFAULT);
1280                version = INDEX_FORMAT_DEFAULT;
1281        }
1282        return version;
1283}
1284
1285/*
1286 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1287 * Again - this is just a (very strong in practice) heuristic that
1288 * the inode hasn't changed.
1289 *
1290 * We save the fields in big-endian order to allow using the
1291 * index file over NFS transparently.
1292 */
1293struct ondisk_cache_entry {
1294        struct cache_time ctime;
1295        struct cache_time mtime;
1296        uint32_t dev;
1297        uint32_t ino;
1298        uint32_t mode;
1299        uint32_t uid;
1300        uint32_t gid;
1301        uint32_t size;
1302        unsigned char sha1[20];
1303        uint16_t flags;
1304        char name[FLEX_ARRAY]; /* more */
1305};
1306
1307/*
1308 * This struct is used when CE_EXTENDED bit is 1
1309 * The struct must match ondisk_cache_entry exactly from
1310 * ctime till flags
1311 */
1312struct ondisk_cache_entry_extended {
1313        struct cache_time ctime;
1314        struct cache_time mtime;
1315        uint32_t dev;
1316        uint32_t ino;
1317        uint32_t mode;
1318        uint32_t uid;
1319        uint32_t gid;
1320        uint32_t size;
1321        unsigned char sha1[20];
1322        uint16_t flags;
1323        uint16_t flags2;
1324        char name[FLEX_ARRAY]; /* more */
1325};
1326
1327/* These are only used for v3 or lower */
1328#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1329#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1330#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1331#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1332                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1333                            ondisk_cache_entry_size(ce_namelen(ce)))
1334
1335static int verify_hdr(struct cache_header *hdr, unsigned long size)
1336{
1337        git_SHA_CTX c;
1338        unsigned char sha1[20];
1339        int hdr_version;
1340
1341        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1342                return error("bad signature");
1343        hdr_version = ntohl(hdr->hdr_version);
1344        if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1345                return error("bad index version %d", hdr_version);
1346        git_SHA1_Init(&c);
1347        git_SHA1_Update(&c, hdr, size - 20);
1348        git_SHA1_Final(sha1, &c);
1349        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1350                return error("bad index file sha1 signature");
1351        return 0;
1352}
1353
1354static int read_index_extension(struct index_state *istate,
1355                                const char *ext, void *data, unsigned long sz)
1356{
1357        switch (CACHE_EXT(ext)) {
1358        case CACHE_EXT_TREE:
1359                istate->cache_tree = cache_tree_read(data, sz);
1360                break;
1361        case CACHE_EXT_RESOLVE_UNDO:
1362                istate->resolve_undo = resolve_undo_read(data, sz);
1363                break;
1364        case CACHE_EXT_LINK:
1365                if (read_link_extension(istate, data, sz))
1366                        return -1;
1367                break;
1368        default:
1369                if (*ext < 'A' || 'Z' < *ext)
1370                        return error("index uses %.4s extension, which we do not understand",
1371                                     ext);
1372                fprintf(stderr, "ignoring %.4s extension\n", ext);
1373                break;
1374        }
1375        return 0;
1376}
1377
1378int read_index(struct index_state *istate)
1379{
1380        return read_index_from(istate, get_index_file());
1381}
1382
1383static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1384                                                   unsigned int flags,
1385                                                   const char *name,
1386                                                   size_t len)
1387{
1388        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1389
1390        ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1391        ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1392        ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1393        ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1394        ce->ce_stat_data.sd_dev   = get_be32(&ondisk->dev);
1395        ce->ce_stat_data.sd_ino   = get_be32(&ondisk->ino);
1396        ce->ce_mode  = get_be32(&ondisk->mode);
1397        ce->ce_stat_data.sd_uid   = get_be32(&ondisk->uid);
1398        ce->ce_stat_data.sd_gid   = get_be32(&ondisk->gid);
1399        ce->ce_stat_data.sd_size  = get_be32(&ondisk->size);
1400        ce->ce_flags = flags & ~CE_NAMEMASK;
1401        ce->ce_namelen = len;
1402        ce->index = 0;
1403        hashcpy(ce->sha1, ondisk->sha1);
1404        memcpy(ce->name, name, len);
1405        ce->name[len] = '\0';
1406        return ce;
1407}
1408
1409/*
1410 * Adjacent cache entries tend to share the leading paths, so it makes
1411 * sense to only store the differences in later entries.  In the v4
1412 * on-disk format of the index, each on-disk cache entry stores the
1413 * number of bytes to be stripped from the end of the previous name,
1414 * and the bytes to append to the result, to come up with its name.
1415 */
1416static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1417{
1418        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1419        size_t len = decode_varint(&cp);
1420
1421        if (name->len < len)
1422                die("malformed name field in the index");
1423        strbuf_remove(name, name->len - len, len);
1424        for (ep = cp; *ep; ep++)
1425                ; /* find the end */
1426        strbuf_add(name, cp, ep - cp);
1427        return (const char *)ep + 1 - cp_;
1428}
1429
1430static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1431                                            unsigned long *ent_size,
1432                                            struct strbuf *previous_name)
1433{
1434        struct cache_entry *ce;
1435        size_t len;
1436        const char *name;
1437        unsigned int flags;
1438
1439        /* On-disk flags are just 16 bits */
1440        flags = get_be16(&ondisk->flags);
1441        len = flags & CE_NAMEMASK;
1442
1443        if (flags & CE_EXTENDED) {
1444                struct ondisk_cache_entry_extended *ondisk2;
1445                int extended_flags;
1446                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1447                extended_flags = get_be16(&ondisk2->flags2) << 16;
1448                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1449                if (extended_flags & ~CE_EXTENDED_FLAGS)
1450                        die("Unknown index entry format %08x", extended_flags);
1451                flags |= extended_flags;
1452                name = ondisk2->name;
1453        }
1454        else
1455                name = ondisk->name;
1456
1457        if (!previous_name) {
1458                /* v3 and earlier */
1459                if (len == CE_NAMEMASK)
1460                        len = strlen(name);
1461                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1462
1463                *ent_size = ondisk_ce_size(ce);
1464        } else {
1465                unsigned long consumed;
1466                consumed = expand_name_field(previous_name, name);
1467                ce = cache_entry_from_ondisk(ondisk, flags,
1468                                             previous_name->buf,
1469                                             previous_name->len);
1470
1471                *ent_size = (name - ((char *)ondisk)) + consumed;
1472        }
1473        return ce;
1474}
1475
1476/* remember to discard_cache() before reading a different cache! */
1477int do_read_index(struct index_state *istate, const char *path, int must_exist)
1478{
1479        int fd, i;
1480        struct stat st;
1481        unsigned long src_offset;
1482        struct cache_header *hdr;
1483        void *mmap;
1484        size_t mmap_size;
1485        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1486
1487        if (istate->initialized)
1488                return istate->cache_nr;
1489
1490        istate->timestamp.sec = 0;
1491        istate->timestamp.nsec = 0;
1492        fd = open(path, O_RDONLY);
1493        if (fd < 0) {
1494                if (!must_exist && errno == ENOENT)
1495                        return 0;
1496                die_errno("%s: index file open failed", path);
1497        }
1498
1499        if (fstat(fd, &st))
1500                die_errno("cannot stat the open index");
1501
1502        mmap_size = xsize_t(st.st_size);
1503        if (mmap_size < sizeof(struct cache_header) + 20)
1504                die("index file smaller than expected");
1505
1506        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1507        if (mmap == MAP_FAILED)
1508                die_errno("unable to map index file");
1509        close(fd);
1510
1511        hdr = mmap;
1512        if (verify_hdr(hdr, mmap_size) < 0)
1513                goto unmap;
1514
1515        hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1516        istate->version = ntohl(hdr->hdr_version);
1517        istate->cache_nr = ntohl(hdr->hdr_entries);
1518        istate->cache_alloc = alloc_nr(istate->cache_nr);
1519        istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1520        istate->initialized = 1;
1521
1522        if (istate->version == 4)
1523                previous_name = &previous_name_buf;
1524        else
1525                previous_name = NULL;
1526
1527        src_offset = sizeof(*hdr);
1528        for (i = 0; i < istate->cache_nr; i++) {
1529                struct ondisk_cache_entry *disk_ce;
1530                struct cache_entry *ce;
1531                unsigned long consumed;
1532
1533                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1534                ce = create_from_disk(disk_ce, &consumed, previous_name);
1535                set_index_entry(istate, i, ce);
1536
1537                src_offset += consumed;
1538        }
1539        strbuf_release(&previous_name_buf);
1540        istate->timestamp.sec = st.st_mtime;
1541        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1542
1543        while (src_offset <= mmap_size - 20 - 8) {
1544                /* After an array of active_nr index entries,
1545                 * there can be arbitrary number of extended
1546                 * sections, each of which is prefixed with
1547                 * extension name (4-byte) and section length
1548                 * in 4-byte network byte order.
1549                 */
1550                uint32_t extsize;
1551                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1552                extsize = ntohl(extsize);
1553                if (read_index_extension(istate,
1554                                         (const char *) mmap + src_offset,
1555                                         (char *) mmap + src_offset + 8,
1556                                         extsize) < 0)
1557                        goto unmap;
1558                src_offset += 8;
1559                src_offset += extsize;
1560        }
1561        munmap(mmap, mmap_size);
1562        return istate->cache_nr;
1563
1564unmap:
1565        munmap(mmap, mmap_size);
1566        die("index file corrupt");
1567}
1568
1569int read_index_from(struct index_state *istate, const char *path)
1570{
1571        struct split_index *split_index;
1572        int ret;
1573
1574        /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1575        if (istate->initialized)
1576                return istate->cache_nr;
1577
1578        ret = do_read_index(istate, path, 0);
1579        split_index = istate->split_index;
1580        if (!split_index)
1581                return ret;
1582
1583        if (is_null_sha1(split_index->base_sha1))
1584                return ret;
1585
1586        if (split_index->base)
1587                discard_index(split_index->base);
1588        else
1589                split_index->base = xcalloc(1, sizeof(*split_index->base));
1590        ret = do_read_index(split_index->base,
1591                            git_path("sharedindex.%s",
1592                                     sha1_to_hex(split_index->base_sha1)), 1);
1593        if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1594                die("broken index, expect %s in %s, got %s",
1595                    sha1_to_hex(split_index->base_sha1),
1596                    git_path("sharedindex.%s",
1597                                     sha1_to_hex(split_index->base_sha1)),
1598                    sha1_to_hex(split_index->base->sha1));
1599        merge_base_index(istate);
1600        return ret;
1601}
1602
1603int is_index_unborn(struct index_state *istate)
1604{
1605        return (!istate->cache_nr && !istate->timestamp.sec);
1606}
1607
1608int discard_index(struct index_state *istate)
1609{
1610        int i;
1611
1612        for (i = 0; i < istate->cache_nr; i++) {
1613                if (istate->cache[i]->index &&
1614                    istate->split_index &&
1615                    istate->split_index->base &&
1616                    istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1617                    istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1618                        continue;
1619                free(istate->cache[i]);
1620        }
1621        resolve_undo_clear_index(istate);
1622        istate->cache_nr = 0;
1623        istate->cache_changed = 0;
1624        istate->timestamp.sec = 0;
1625        istate->timestamp.nsec = 0;
1626        free_name_hash(istate);
1627        cache_tree_free(&(istate->cache_tree));
1628        istate->initialized = 0;
1629        free(istate->cache);
1630        istate->cache = NULL;
1631        istate->cache_alloc = 0;
1632        discard_split_index(istate);
1633        return 0;
1634}
1635
1636int unmerged_index(const struct index_state *istate)
1637{
1638        int i;
1639        for (i = 0; i < istate->cache_nr; i++) {
1640                if (ce_stage(istate->cache[i]))
1641                        return 1;
1642        }
1643        return 0;
1644}
1645
1646#define WRITE_BUFFER_SIZE 8192
1647static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1648static unsigned long write_buffer_len;
1649
1650static int ce_write_flush(git_SHA_CTX *context, int fd)
1651{
1652        unsigned int buffered = write_buffer_len;
1653        if (buffered) {
1654                git_SHA1_Update(context, write_buffer, buffered);
1655                if (write_in_full(fd, write_buffer, buffered) != buffered)
1656                        return -1;
1657                write_buffer_len = 0;
1658        }
1659        return 0;
1660}
1661
1662static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1663{
1664        while (len) {
1665                unsigned int buffered = write_buffer_len;
1666                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1667                if (partial > len)
1668                        partial = len;
1669                memcpy(write_buffer + buffered, data, partial);
1670                buffered += partial;
1671                if (buffered == WRITE_BUFFER_SIZE) {
1672                        write_buffer_len = buffered;
1673                        if (ce_write_flush(context, fd))
1674                                return -1;
1675                        buffered = 0;
1676                }
1677                write_buffer_len = buffered;
1678                len -= partial;
1679                data = (char *) data + partial;
1680        }
1681        return 0;
1682}
1683
1684static int write_index_ext_header(git_SHA_CTX *context, int fd,
1685                                  unsigned int ext, unsigned int sz)
1686{
1687        ext = htonl(ext);
1688        sz = htonl(sz);
1689        return ((ce_write(context, fd, &ext, 4) < 0) ||
1690                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1691}
1692
1693static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1694{
1695        unsigned int left = write_buffer_len;
1696
1697        if (left) {
1698                write_buffer_len = 0;
1699                git_SHA1_Update(context, write_buffer, left);
1700        }
1701
1702        /* Flush first if not enough space for SHA1 signature */
1703        if (left + 20 > WRITE_BUFFER_SIZE) {
1704                if (write_in_full(fd, write_buffer, left) != left)
1705                        return -1;
1706                left = 0;
1707        }
1708
1709        /* Append the SHA1 signature at the end */
1710        git_SHA1_Final(write_buffer + left, context);
1711        hashcpy(sha1, write_buffer + left);
1712        left += 20;
1713        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1714}
1715
1716static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1717{
1718        /*
1719         * The only thing we care about in this function is to smudge the
1720         * falsely clean entry due to touch-update-touch race, so we leave
1721         * everything else as they are.  We are called for entries whose
1722         * ce_stat_data.sd_mtime match the index file mtime.
1723         *
1724         * Note that this actually does not do much for gitlinks, for
1725         * which ce_match_stat_basic() always goes to the actual
1726         * contents.  The caller checks with is_racy_timestamp() which
1727         * always says "no" for gitlinks, so we are not called for them ;-)
1728         */
1729        struct stat st;
1730
1731        if (lstat(ce->name, &st) < 0)
1732                return;
1733        if (ce_match_stat_basic(ce, &st))
1734                return;
1735        if (ce_modified_check_fs(ce, &st)) {
1736                /* This is "racily clean"; smudge it.  Note that this
1737                 * is a tricky code.  At first glance, it may appear
1738                 * that it can break with this sequence:
1739                 *
1740                 * $ echo xyzzy >frotz
1741                 * $ git-update-index --add frotz
1742                 * $ : >frotz
1743                 * $ sleep 3
1744                 * $ echo filfre >nitfol
1745                 * $ git-update-index --add nitfol
1746                 *
1747                 * but it does not.  When the second update-index runs,
1748                 * it notices that the entry "frotz" has the same timestamp
1749                 * as index, and if we were to smudge it by resetting its
1750                 * size to zero here, then the object name recorded
1751                 * in index is the 6-byte file but the cached stat information
1752                 * becomes zero --- which would then match what we would
1753                 * obtain from the filesystem next time we stat("frotz").
1754                 *
1755                 * However, the second update-index, before calling
1756                 * this function, notices that the cached size is 6
1757                 * bytes and what is on the filesystem is an empty
1758                 * file, and never calls us, so the cached size information
1759                 * for "frotz" stays 6 which does not match the filesystem.
1760                 */
1761                ce->ce_stat_data.sd_size = 0;
1762        }
1763}
1764
1765/* Copy miscellaneous fields but not the name */
1766static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1767                                       struct cache_entry *ce)
1768{
1769        short flags;
1770
1771        ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1772        ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1773        ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1774        ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1775        ondisk->dev  = htonl(ce->ce_stat_data.sd_dev);
1776        ondisk->ino  = htonl(ce->ce_stat_data.sd_ino);
1777        ondisk->mode = htonl(ce->ce_mode);
1778        ondisk->uid  = htonl(ce->ce_stat_data.sd_uid);
1779        ondisk->gid  = htonl(ce->ce_stat_data.sd_gid);
1780        ondisk->size = htonl(ce->ce_stat_data.sd_size);
1781        hashcpy(ondisk->sha1, ce->sha1);
1782
1783        flags = ce->ce_flags & ~CE_NAMEMASK;
1784        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1785        ondisk->flags = htons(flags);
1786        if (ce->ce_flags & CE_EXTENDED) {
1787                struct ondisk_cache_entry_extended *ondisk2;
1788                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1789                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1790                return ondisk2->name;
1791        }
1792        else {
1793                return ondisk->name;
1794        }
1795}
1796
1797static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1798                          struct strbuf *previous_name)
1799{
1800        int size;
1801        struct ondisk_cache_entry *ondisk;
1802        int saved_namelen = saved_namelen; /* compiler workaround */
1803        char *name;
1804        int result;
1805
1806        if (ce->ce_flags & CE_STRIP_NAME) {
1807                saved_namelen = ce_namelen(ce);
1808                ce->ce_namelen = 0;
1809        }
1810
1811        if (!previous_name) {
1812                size = ondisk_ce_size(ce);
1813                ondisk = xcalloc(1, size);
1814                name = copy_cache_entry_to_ondisk(ondisk, ce);
1815                memcpy(name, ce->name, ce_namelen(ce));
1816        } else {
1817                int common, to_remove, prefix_size;
1818                unsigned char to_remove_vi[16];
1819                for (common = 0;
1820                     (ce->name[common] &&
1821                      common < previous_name->len &&
1822                      ce->name[common] == previous_name->buf[common]);
1823                     common++)
1824                        ; /* still matching */
1825                to_remove = previous_name->len - common;
1826                prefix_size = encode_varint(to_remove, to_remove_vi);
1827
1828                if (ce->ce_flags & CE_EXTENDED)
1829                        size = offsetof(struct ondisk_cache_entry_extended, name);
1830                else
1831                        size = offsetof(struct ondisk_cache_entry, name);
1832                size += prefix_size + (ce_namelen(ce) - common + 1);
1833
1834                ondisk = xcalloc(1, size);
1835                name = copy_cache_entry_to_ondisk(ondisk, ce);
1836                memcpy(name, to_remove_vi, prefix_size);
1837                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1838
1839                strbuf_splice(previous_name, common, to_remove,
1840                              ce->name + common, ce_namelen(ce) - common);
1841        }
1842        if (ce->ce_flags & CE_STRIP_NAME) {
1843                ce->ce_namelen = saved_namelen;
1844                ce->ce_flags &= ~CE_STRIP_NAME;
1845        }
1846
1847        result = ce_write(c, fd, ondisk, size);
1848        free(ondisk);
1849        return result;
1850}
1851
1852/*
1853 * This function verifies if index_state has the correct sha1 of the
1854 * index file.  Don't die if we have any other failure, just return 0.
1855 */
1856static int verify_index_from(const struct index_state *istate, const char *path)
1857{
1858        int fd;
1859        ssize_t n;
1860        struct stat st;
1861        unsigned char sha1[20];
1862
1863        if (!istate->initialized)
1864                return 0;
1865
1866        fd = open(path, O_RDONLY);
1867        if (fd < 0)
1868                return 0;
1869
1870        if (fstat(fd, &st))
1871                goto out;
1872
1873        if (st.st_size < sizeof(struct cache_header) + 20)
1874                goto out;
1875
1876        n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1877        if (n != 20)
1878                goto out;
1879
1880        if (hashcmp(istate->sha1, sha1))
1881                goto out;
1882
1883        close(fd);
1884        return 1;
1885
1886out:
1887        close(fd);
1888        return 0;
1889}
1890
1891static int verify_index(const struct index_state *istate)
1892{
1893        return verify_index_from(istate, get_index_file());
1894}
1895
1896static int has_racy_timestamp(struct index_state *istate)
1897{
1898        int entries = istate->cache_nr;
1899        int i;
1900
1901        for (i = 0; i < entries; i++) {
1902                struct cache_entry *ce = istate->cache[i];
1903                if (is_racy_timestamp(istate, ce))
1904                        return 1;
1905        }
1906        return 0;
1907}
1908
1909/*
1910 * Opportunistically update the index but do not complain if we can't
1911 */
1912void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1913{
1914        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1915            verify_index(istate) &&
1916            write_locked_index(istate, lockfile, COMMIT_LOCK))
1917                rollback_lock_file(lockfile);
1918}
1919
1920static int do_write_index(struct index_state *istate, int newfd,
1921                          int strip_extensions)
1922{
1923        git_SHA_CTX c;
1924        struct cache_header hdr;
1925        int i, err, removed, extended, hdr_version;
1926        struct cache_entry **cache = istate->cache;
1927        int entries = istate->cache_nr;
1928        struct stat st;
1929        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1930
1931        for (i = removed = extended = 0; i < entries; i++) {
1932                if (cache[i]->ce_flags & CE_REMOVE)
1933                        removed++;
1934
1935                /* reduce extended entries if possible */
1936                cache[i]->ce_flags &= ~CE_EXTENDED;
1937                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1938                        extended++;
1939                        cache[i]->ce_flags |= CE_EXTENDED;
1940                }
1941        }
1942
1943        if (!istate->version) {
1944                istate->version = get_index_format_default();
1945                if (getenv("GIT_TEST_SPLIT_INDEX"))
1946                        init_split_index(istate);
1947        }
1948
1949        /* demote version 3 to version 2 when the latter suffices */
1950        if (istate->version == 3 || istate->version == 2)
1951                istate->version = extended ? 3 : 2;
1952
1953        hdr_version = istate->version;
1954
1955        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1956        hdr.hdr_version = htonl(hdr_version);
1957        hdr.hdr_entries = htonl(entries - removed);
1958
1959        git_SHA1_Init(&c);
1960        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1961                return -1;
1962
1963        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1964        for (i = 0; i < entries; i++) {
1965                struct cache_entry *ce = cache[i];
1966                if (ce->ce_flags & CE_REMOVE)
1967                        continue;
1968                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1969                        ce_smudge_racily_clean_entry(ce);
1970                if (is_null_sha1(ce->sha1)) {
1971                        static const char msg[] = "cache entry has null sha1: %s";
1972                        static int allow = -1;
1973
1974                        if (allow < 0)
1975                                allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1976                        if (allow)
1977                                warning(msg, ce->name);
1978                        else
1979                                return error(msg, ce->name);
1980                }
1981                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1982                        return -1;
1983        }
1984        strbuf_release(&previous_name_buf);
1985
1986        /* Write extension data here */
1987        if (!strip_extensions && istate->split_index) {
1988                struct strbuf sb = STRBUF_INIT;
1989
1990                err = write_link_extension(&sb, istate) < 0 ||
1991                        write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
1992                                               sb.len) < 0 ||
1993                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
1994                strbuf_release(&sb);
1995                if (err)
1996                        return -1;
1997        }
1998        if (!strip_extensions && istate->cache_tree) {
1999                struct strbuf sb = STRBUF_INIT;
2000
2001                cache_tree_write(&sb, istate->cache_tree);
2002                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2003                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2004                strbuf_release(&sb);
2005                if (err)
2006                        return -1;
2007        }
2008        if (!strip_extensions && istate->resolve_undo) {
2009                struct strbuf sb = STRBUF_INIT;
2010
2011                resolve_undo_write(&sb, istate->resolve_undo);
2012                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2013                                             sb.len) < 0
2014                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2015                strbuf_release(&sb);
2016                if (err)
2017                        return -1;
2018        }
2019
2020        if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2021                return -1;
2022        istate->timestamp.sec = (unsigned int)st.st_mtime;
2023        istate->timestamp.nsec = ST_MTIME_NSEC(st);
2024        return 0;
2025}
2026
2027void set_alternate_index_output(const char *name)
2028{
2029        alternate_index_output = name;
2030}
2031
2032static int commit_locked_index(struct lock_file *lk)
2033{
2034        if (alternate_index_output) {
2035                if (lk->fd >= 0 && close_lock_file(lk))
2036                        return -1;
2037                if (rename(lk->filename, alternate_index_output))
2038                        return -1;
2039                lk->filename[0] = 0;
2040                return 0;
2041        } else {
2042                return commit_lock_file(lk);
2043        }
2044}
2045
2046static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2047                                 unsigned flags)
2048{
2049        int ret = do_write_index(istate, lock->fd, 0);
2050        if (ret)
2051                return ret;
2052        assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2053               (COMMIT_LOCK | CLOSE_LOCK));
2054        if (flags & COMMIT_LOCK)
2055                return commit_locked_index(lock);
2056        else if (flags & CLOSE_LOCK)
2057                return close_lock_file(lock);
2058        else
2059                return ret;
2060}
2061
2062static int write_split_index(struct index_state *istate,
2063                             struct lock_file *lock,
2064                             unsigned flags)
2065{
2066        int ret;
2067        prepare_to_write_split_index(istate);
2068        ret = do_write_locked_index(istate, lock, flags);
2069        finish_writing_split_index(istate);
2070        return ret;
2071}
2072
2073static char *temporary_sharedindex;
2074
2075static void remove_temporary_sharedindex(void)
2076{
2077        if (temporary_sharedindex) {
2078                unlink_or_warn(temporary_sharedindex);
2079                free(temporary_sharedindex);
2080                temporary_sharedindex = NULL;
2081        }
2082}
2083
2084static void remove_temporary_sharedindex_on_signal(int signo)
2085{
2086        remove_temporary_sharedindex();
2087        sigchain_pop(signo);
2088        raise(signo);
2089}
2090
2091static int write_shared_index(struct index_state *istate,
2092                              struct lock_file *lock, unsigned flags)
2093{
2094        struct split_index *si = istate->split_index;
2095        static int installed_handler;
2096        int fd, ret;
2097
2098        temporary_sharedindex = git_pathdup("sharedindex_XXXXXX");
2099        fd = mkstemp(temporary_sharedindex);
2100        if (fd < 0) {
2101                free(temporary_sharedindex);
2102                temporary_sharedindex = NULL;
2103                hashclr(si->base_sha1);
2104                return do_write_locked_index(istate, lock, flags);
2105        }
2106        if (!installed_handler) {
2107                atexit(remove_temporary_sharedindex);
2108                sigchain_push_common(remove_temporary_sharedindex_on_signal);
2109        }
2110        move_cache_to_base_index(istate);
2111        ret = do_write_index(si->base, fd, 1);
2112        close(fd);
2113        if (ret) {
2114                remove_temporary_sharedindex();
2115                return ret;
2116        }
2117        ret = rename(temporary_sharedindex,
2118                     git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2119        free(temporary_sharedindex);
2120        temporary_sharedindex = NULL;
2121        if (!ret)
2122                hashcpy(si->base_sha1, si->base->sha1);
2123        return ret;
2124}
2125
2126int write_locked_index(struct index_state *istate, struct lock_file *lock,
2127                       unsigned flags)
2128{
2129        struct split_index *si = istate->split_index;
2130
2131        if (!si || alternate_index_output ||
2132            (istate->cache_changed & ~EXTMASK)) {
2133                if (si)
2134                        hashclr(si->base_sha1);
2135                return do_write_locked_index(istate, lock, flags);
2136        }
2137
2138        if (getenv("GIT_TEST_SPLIT_INDEX")) {
2139                int v = si->base_sha1[0];
2140                if ((v & 15) < 6)
2141                        istate->cache_changed |= SPLIT_INDEX_ORDERED;
2142        }
2143        if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2144                int ret = write_shared_index(istate, lock, flags);
2145                if (ret)
2146                        return ret;
2147        }
2148
2149        return write_split_index(istate, lock, flags);
2150}
2151
2152/*
2153 * Read the index file that is potentially unmerged into given
2154 * index_state, dropping any unmerged entries.  Returns true if
2155 * the index is unmerged.  Callers who want to refuse to work
2156 * from an unmerged state can call this and check its return value,
2157 * instead of calling read_cache().
2158 */
2159int read_index_unmerged(struct index_state *istate)
2160{
2161        int i;
2162        int unmerged = 0;
2163
2164        read_index(istate);
2165        for (i = 0; i < istate->cache_nr; i++) {
2166                struct cache_entry *ce = istate->cache[i];
2167                struct cache_entry *new_ce;
2168                int size, len;
2169
2170                if (!ce_stage(ce))
2171                        continue;
2172                unmerged = 1;
2173                len = ce_namelen(ce);
2174                size = cache_entry_size(len);
2175                new_ce = xcalloc(1, size);
2176                memcpy(new_ce->name, ce->name, len);
2177                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2178                new_ce->ce_namelen = len;
2179                new_ce->ce_mode = ce->ce_mode;
2180                if (add_index_entry(istate, new_ce, 0))
2181                        return error("%s: cannot drop to stage #0",
2182                                     new_ce->name);
2183                i = index_name_pos(istate, new_ce->name, len);
2184        }
2185        return unmerged;
2186}
2187
2188/*
2189 * Returns 1 if the path is an "other" path with respect to
2190 * the index; that is, the path is not mentioned in the index at all,
2191 * either as a file, a directory with some files in the index,
2192 * or as an unmerged entry.
2193 *
2194 * We helpfully remove a trailing "/" from directories so that
2195 * the output of read_directory can be used as-is.
2196 */
2197int index_name_is_other(const struct index_state *istate, const char *name,
2198                int namelen)
2199{
2200        int pos;
2201        if (namelen && name[namelen - 1] == '/')
2202                namelen--;
2203        pos = index_name_pos(istate, name, namelen);
2204        if (0 <= pos)
2205                return 0;       /* exact match */
2206        pos = -pos - 1;
2207        if (pos < istate->cache_nr) {
2208                struct cache_entry *ce = istate->cache[pos];
2209                if (ce_namelen(ce) == namelen &&
2210                    !memcmp(ce->name, name, namelen))
2211                        return 0; /* Yup, this one exists unmerged */
2212        }
2213        return 1;
2214}
2215
2216void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2217{
2218        int pos, len;
2219        unsigned long sz;
2220        enum object_type type;
2221        void *data;
2222
2223        len = strlen(path);
2224        pos = index_name_pos(istate, path, len);
2225        if (pos < 0) {
2226                /*
2227                 * We might be in the middle of a merge, in which
2228                 * case we would read stage #2 (ours).
2229                 */
2230                int i;
2231                for (i = -pos - 1;
2232                     (pos < 0 && i < istate->cache_nr &&
2233                      !strcmp(istate->cache[i]->name, path));
2234                     i++)
2235                        if (ce_stage(istate->cache[i]) == 2)
2236                                pos = i;
2237        }
2238        if (pos < 0)
2239                return NULL;
2240        data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2241        if (!data || type != OBJ_BLOB) {
2242                free(data);
2243                return NULL;
2244        }
2245        if (size)
2246                *size = sz;
2247        return data;
2248}
2249
2250void stat_validity_clear(struct stat_validity *sv)
2251{
2252        free(sv->sd);
2253        sv->sd = NULL;
2254}
2255
2256int stat_validity_check(struct stat_validity *sv, const char *path)
2257{
2258        struct stat st;
2259
2260        if (stat(path, &st) < 0)
2261                return sv->sd == NULL;
2262        if (!sv->sd)
2263                return 0;
2264        return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2265}
2266
2267void stat_validity_update(struct stat_validity *sv, int fd)
2268{
2269        struct stat st;
2270
2271        if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2272                stat_validity_clear(sv);
2273        else {
2274                if (!sv->sd)
2275                        sv->sd = xcalloc(1, sizeof(struct stat_data));
2276                fill_stat_data(sv->sd, &st);
2277        }
2278}