ee0ef049b84de5569af668c63f40c8701844981b
   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 "lockfile.h"
   9#include "cache-tree.h"
  10#include "refs.h"
  11#include "dir.h"
  12#include "tree.h"
  13#include "commit.h"
  14#include "blob.h"
  15#include "resolve-undo.h"
  16#include "strbuf.h"
  17#include "varint.h"
  18#include "split-index.h"
  19#include "sigchain.h"
  20#include "utf8.h"
  21
  22static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
  23                                               unsigned int options);
  24
  25/* Mask for the name length in ce_flags in the on-disk index */
  26
  27#define CE_NAMEMASK  (0x0fff)
  28
  29/* Index extensions.
  30 *
  31 * The first letter should be 'A'..'Z' for extensions that are not
  32 * necessary for a correct operation (i.e. optimization data).
  33 * When new extensions are added that _needs_ to be understood in
  34 * order to correctly interpret the index file, pick character that
  35 * is outside the range, to cause the reader to abort.
  36 */
  37
  38#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  39#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  40#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  41#define CACHE_EXT_LINK 0x6c696e6b         /* "link" */
  42#define CACHE_EXT_UNTRACKED 0x554E5452    /* "UNTR" */
  43
  44/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
  45#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
  46                 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
  47                 SPLIT_INDEX_ORDERED)
  48
  49struct index_state the_index;
  50static const char *alternate_index_output;
  51
  52static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  53{
  54        istate->cache[nr] = ce;
  55        add_name_hash(istate, ce);
  56}
  57
  58static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  59{
  60        struct cache_entry *old = istate->cache[nr];
  61
  62        replace_index_entry_in_base(istate, old, ce);
  63        remove_name_hash(istate, old);
  64        free(old);
  65        set_index_entry(istate, nr, ce);
  66        ce->ce_flags |= CE_UPDATE_IN_BASE;
  67        istate->cache_changed |= CE_ENTRY_CHANGED;
  68}
  69
  70void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  71{
  72        struct cache_entry *old = istate->cache[nr], *new;
  73        int namelen = strlen(new_name);
  74
  75        new = xmalloc(cache_entry_size(namelen));
  76        copy_cache_entry(new, old);
  77        new->ce_flags &= ~CE_HASHED;
  78        new->ce_namelen = namelen;
  79        new->index = 0;
  80        memcpy(new->name, new_name, namelen + 1);
  81
  82        cache_tree_invalidate_path(istate, old->name);
  83        remove_index_entry_at(istate, nr);
  84        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  85}
  86
  87void fill_stat_data(struct stat_data *sd, struct stat *st)
  88{
  89        sd->sd_ctime.sec = (unsigned int)st->st_ctime;
  90        sd->sd_mtime.sec = (unsigned int)st->st_mtime;
  91        sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
  92        sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
  93        sd->sd_dev = st->st_dev;
  94        sd->sd_ino = st->st_ino;
  95        sd->sd_uid = st->st_uid;
  96        sd->sd_gid = st->st_gid;
  97        sd->sd_size = st->st_size;
  98}
  99
 100int match_stat_data(const struct stat_data *sd, struct stat *st)
 101{
 102        int changed = 0;
 103
 104        if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
 105                changed |= MTIME_CHANGED;
 106        if (trust_ctime && check_stat &&
 107            sd->sd_ctime.sec != (unsigned int)st->st_ctime)
 108                changed |= CTIME_CHANGED;
 109
 110#ifdef USE_NSEC
 111        if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
 112                changed |= MTIME_CHANGED;
 113        if (trust_ctime && check_stat &&
 114            sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
 115                changed |= CTIME_CHANGED;
 116#endif
 117
 118        if (check_stat) {
 119                if (sd->sd_uid != (unsigned int) st->st_uid ||
 120                        sd->sd_gid != (unsigned int) st->st_gid)
 121                        changed |= OWNER_CHANGED;
 122                if (sd->sd_ino != (unsigned int) st->st_ino)
 123                        changed |= INODE_CHANGED;
 124        }
 125
 126#ifdef USE_STDEV
 127        /*
 128         * st_dev breaks on network filesystems where different
 129         * clients will have different views of what "device"
 130         * the filesystem is on
 131         */
 132        if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
 133                        changed |= INODE_CHANGED;
 134#endif
 135
 136        if (sd->sd_size != (unsigned int) st->st_size)
 137                changed |= DATA_CHANGED;
 138
 139        return changed;
 140}
 141
 142/*
 143 * This only updates the "non-critical" parts of the directory
 144 * cache, ie the parts that aren't tracked by GIT, and only used
 145 * to validate the cache.
 146 */
 147void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 148{
 149        fill_stat_data(&ce->ce_stat_data, st);
 150
 151        if (assume_unchanged)
 152                ce->ce_flags |= CE_VALID;
 153
 154        if (S_ISREG(st->st_mode))
 155                ce_mark_uptodate(ce);
 156}
 157
 158static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
 159{
 160        int match = -1;
 161        int fd = open(ce->name, O_RDONLY);
 162
 163        if (fd >= 0) {
 164                unsigned char sha1[20];
 165                if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
 166                        match = hashcmp(sha1, ce->sha1);
 167                /* index_fd() closed the file descriptor already */
 168        }
 169        return match;
 170}
 171
 172static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
 173{
 174        int match = -1;
 175        void *buffer;
 176        unsigned long size;
 177        enum object_type type;
 178        struct strbuf sb = STRBUF_INIT;
 179
 180        if (strbuf_readlink(&sb, ce->name, expected_size))
 181                return -1;
 182
 183        buffer = read_sha1_file(ce->sha1, &type, &size);
 184        if (buffer) {
 185                if (size == sb.len)
 186                        match = memcmp(buffer, sb.buf, size);
 187                free(buffer);
 188        }
 189        strbuf_release(&sb);
 190        return match;
 191}
 192
 193static int ce_compare_gitlink(const struct cache_entry *ce)
 194{
 195        unsigned char sha1[20];
 196
 197        /*
 198         * We don't actually require that the .git directory
 199         * under GITLINK directory be a valid git directory. It
 200         * might even be missing (in case nobody populated that
 201         * sub-project).
 202         *
 203         * If so, we consider it always to match.
 204         */
 205        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 206                return 0;
 207        return hashcmp(sha1, ce->sha1);
 208}
 209
 210static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
 211{
 212        switch (st->st_mode & S_IFMT) {
 213        case S_IFREG:
 214                if (ce_compare_data(ce, st))
 215                        return DATA_CHANGED;
 216                break;
 217        case S_IFLNK:
 218                if (ce_compare_link(ce, xsize_t(st->st_size)))
 219                        return DATA_CHANGED;
 220                break;
 221        case S_IFDIR:
 222                if (S_ISGITLINK(ce->ce_mode))
 223                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 224        default:
 225                return TYPE_CHANGED;
 226        }
 227        return 0;
 228}
 229
 230static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
 231{
 232        unsigned int changed = 0;
 233
 234        if (ce->ce_flags & CE_REMOVE)
 235                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 236
 237        switch (ce->ce_mode & S_IFMT) {
 238        case S_IFREG:
 239                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 240                /* We consider only the owner x bit to be relevant for
 241                 * "mode changes"
 242                 */
 243                if (trust_executable_bit &&
 244                    (0100 & (ce->ce_mode ^ st->st_mode)))
 245                        changed |= MODE_CHANGED;
 246                break;
 247        case S_IFLNK:
 248                if (!S_ISLNK(st->st_mode) &&
 249                    (has_symlinks || !S_ISREG(st->st_mode)))
 250                        changed |= TYPE_CHANGED;
 251                break;
 252        case S_IFGITLINK:
 253                /* We ignore most of the st_xxx fields for gitlinks */
 254                if (!S_ISDIR(st->st_mode))
 255                        changed |= TYPE_CHANGED;
 256                else if (ce_compare_gitlink(ce))
 257                        changed |= DATA_CHANGED;
 258                return changed;
 259        default:
 260                die("internal error: ce_mode is %o", ce->ce_mode);
 261        }
 262
 263        changed |= match_stat_data(&ce->ce_stat_data, st);
 264
 265        /* Racily smudged entry? */
 266        if (!ce->ce_stat_data.sd_size) {
 267                if (!is_empty_blob_sha1(ce->sha1))
 268                        changed |= DATA_CHANGED;
 269        }
 270
 271        return changed;
 272}
 273
 274static int is_racy_timestamp(const struct index_state *istate,
 275                             const struct cache_entry *ce)
 276{
 277        return (!S_ISGITLINK(ce->ce_mode) &&
 278                istate->timestamp.sec &&
 279#ifdef USE_NSEC
 280                 /* nanosecond timestamped files can also be racy! */
 281                (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
 282                 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
 283                  istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
 284#else
 285                istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
 286#endif
 287                 );
 288}
 289
 290int ie_match_stat(const struct index_state *istate,
 291                  const struct cache_entry *ce, struct stat *st,
 292                  unsigned int options)
 293{
 294        unsigned int changed;
 295        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 296        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 297        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 298
 299        /*
 300         * If it's marked as always valid in the index, it's
 301         * valid whatever the checked-out copy says.
 302         *
 303         * skip-worktree has the same effect with higher precedence
 304         */
 305        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 306                return 0;
 307        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 308                return 0;
 309
 310        /*
 311         * Intent-to-add entries have not been added, so the index entry
 312         * by definition never matches what is in the work tree until it
 313         * actually gets added.
 314         */
 315        if (ce->ce_flags & CE_INTENT_TO_ADD)
 316                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 317
 318        changed = ce_match_stat_basic(ce, st);
 319
 320        /*
 321         * Within 1 second of this sequence:
 322         *      echo xyzzy >file && git-update-index --add file
 323         * running this command:
 324         *      echo frotz >file
 325         * would give a falsely clean cache entry.  The mtime and
 326         * length match the cache, and other stat fields do not change.
 327         *
 328         * We could detect this at update-index time (the cache entry
 329         * being registered/updated records the same time as "now")
 330         * and delay the return from git-update-index, but that would
 331         * effectively mean we can make at most one commit per second,
 332         * which is not acceptable.  Instead, we check cache entries
 333         * whose mtime are the same as the index file timestamp more
 334         * carefully than others.
 335         */
 336        if (!changed && is_racy_timestamp(istate, ce)) {
 337                if (assume_racy_is_modified)
 338                        changed |= DATA_CHANGED;
 339                else
 340                        changed |= ce_modified_check_fs(ce, st);
 341        }
 342
 343        return changed;
 344}
 345
 346int ie_modified(const struct index_state *istate,
 347                const struct cache_entry *ce,
 348                struct stat *st, unsigned int options)
 349{
 350        int changed, changed_fs;
 351
 352        changed = ie_match_stat(istate, ce, st, options);
 353        if (!changed)
 354                return 0;
 355        /*
 356         * If the mode or type has changed, there's no point in trying
 357         * to refresh the entry - it's not going to match
 358         */
 359        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 360                return changed;
 361
 362        /*
 363         * Immediately after read-tree or update-index --cacheinfo,
 364         * the length field is zero, as we have never even read the
 365         * lstat(2) information once, and we cannot trust DATA_CHANGED
 366         * returned by ie_match_stat() which in turn was returned by
 367         * ce_match_stat_basic() to signal that the filesize of the
 368         * blob changed.  We have to actually go to the filesystem to
 369         * see if the contents match, and if so, should answer "unchanged".
 370         *
 371         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 372         * already has checked the actual HEAD from the filesystem in the
 373         * subproject.  If ie_match_stat() already said it is different,
 374         * then we know it is.
 375         */
 376        if ((changed & DATA_CHANGED) &&
 377            (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
 378                return changed;
 379
 380        changed_fs = ce_modified_check_fs(ce, st);
 381        if (changed_fs)
 382                return changed | changed_fs;
 383        return 0;
 384}
 385
 386int base_name_compare(const char *name1, int len1, int mode1,
 387                      const char *name2, int len2, int mode2)
 388{
 389        unsigned char c1, c2;
 390        int len = len1 < len2 ? len1 : len2;
 391        int cmp;
 392
 393        cmp = memcmp(name1, name2, len);
 394        if (cmp)
 395                return cmp;
 396        c1 = name1[len];
 397        c2 = name2[len];
 398        if (!c1 && S_ISDIR(mode1))
 399                c1 = '/';
 400        if (!c2 && S_ISDIR(mode2))
 401                c2 = '/';
 402        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 403}
 404
 405/*
 406 * df_name_compare() is identical to base_name_compare(), except it
 407 * compares conflicting directory/file entries as equal. Note that
 408 * while a directory name compares as equal to a regular file, they
 409 * then individually compare _differently_ to a filename that has
 410 * a dot after the basename (because '\0' < '.' < '/').
 411 *
 412 * This is used by routines that want to traverse the git namespace
 413 * but then handle conflicting entries together when possible.
 414 */
 415int df_name_compare(const char *name1, int len1, int mode1,
 416                    const char *name2, int len2, int mode2)
 417{
 418        int len = len1 < len2 ? len1 : len2, cmp;
 419        unsigned char c1, c2;
 420
 421        cmp = memcmp(name1, name2, len);
 422        if (cmp)
 423                return cmp;
 424        /* Directories and files compare equal (same length, same name) */
 425        if (len1 == len2)
 426                return 0;
 427        c1 = name1[len];
 428        if (!c1 && S_ISDIR(mode1))
 429                c1 = '/';
 430        c2 = name2[len];
 431        if (!c2 && S_ISDIR(mode2))
 432                c2 = '/';
 433        if (c1 == '/' && !c2)
 434                return 0;
 435        if (c2 == '/' && !c1)
 436                return 0;
 437        return c1 - c2;
 438}
 439
 440int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
 441{
 442        size_t min_len = (len1 < len2) ? len1 : len2;
 443        int cmp = memcmp(name1, name2, min_len);
 444        if (cmp)
 445                return cmp;
 446        if (len1 < len2)
 447                return -1;
 448        if (len1 > len2)
 449                return 1;
 450        return 0;
 451}
 452
 453int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 454{
 455        int cmp;
 456
 457        cmp = name_compare(name1, len1, name2, len2);
 458        if (cmp)
 459                return cmp;
 460
 461        if (stage1 < stage2)
 462                return -1;
 463        if (stage1 > stage2)
 464                return 1;
 465        return 0;
 466}
 467
 468static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 469{
 470        int first, last;
 471
 472        first = 0;
 473        last = istate->cache_nr;
 474        while (last > first) {
 475                int next = (last + first) >> 1;
 476                struct cache_entry *ce = istate->cache[next];
 477                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 478                if (!cmp)
 479                        return next;
 480                if (cmp < 0) {
 481                        last = next;
 482                        continue;
 483                }
 484                first = next+1;
 485        }
 486        return -first-1;
 487}
 488
 489int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 490{
 491        return index_name_stage_pos(istate, name, namelen, 0);
 492}
 493
 494/* Remove entry, return true if there are more entries to go.. */
 495int remove_index_entry_at(struct index_state *istate, int pos)
 496{
 497        struct cache_entry *ce = istate->cache[pos];
 498
 499        record_resolve_undo(istate, ce);
 500        remove_name_hash(istate, ce);
 501        save_or_free_index_entry(istate, ce);
 502        istate->cache_changed |= CE_ENTRY_REMOVED;
 503        istate->cache_nr--;
 504        if (pos >= istate->cache_nr)
 505                return 0;
 506        memmove(istate->cache + pos,
 507                istate->cache + pos + 1,
 508                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 509        return 1;
 510}
 511
 512/*
 513 * Remove all cache entries marked for removal, that is where
 514 * CE_REMOVE is set in ce_flags.  This is much more effective than
 515 * calling remove_index_entry_at() for each entry to be removed.
 516 */
 517void remove_marked_cache_entries(struct index_state *istate)
 518{
 519        struct cache_entry **ce_array = istate->cache;
 520        unsigned int i, j;
 521
 522        for (i = j = 0; i < istate->cache_nr; i++) {
 523                if (ce_array[i]->ce_flags & CE_REMOVE) {
 524                        remove_name_hash(istate, ce_array[i]);
 525                        save_or_free_index_entry(istate, ce_array[i]);
 526                }
 527                else
 528                        ce_array[j++] = ce_array[i];
 529        }
 530        if (j == istate->cache_nr)
 531                return;
 532        istate->cache_changed |= CE_ENTRY_REMOVED;
 533        istate->cache_nr = j;
 534}
 535
 536int remove_file_from_index(struct index_state *istate, const char *path)
 537{
 538        int pos = index_name_pos(istate, path, strlen(path));
 539        if (pos < 0)
 540                pos = -pos-1;
 541        cache_tree_invalidate_path(istate, path);
 542        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 543                remove_index_entry_at(istate, pos);
 544        return 0;
 545}
 546
 547static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 548{
 549        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 550}
 551
 552static int index_name_pos_also_unmerged(struct index_state *istate,
 553        const char *path, int namelen)
 554{
 555        int pos = index_name_pos(istate, path, namelen);
 556        struct cache_entry *ce;
 557
 558        if (pos >= 0)
 559                return pos;
 560
 561        /* maybe unmerged? */
 562        pos = -1 - pos;
 563        if (pos >= istate->cache_nr ||
 564                        compare_name((ce = istate->cache[pos]), path, namelen))
 565                return -1;
 566
 567        /* order of preference: stage 2, 1, 3 */
 568        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 569                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 570                        !compare_name(ce, path, namelen))
 571                pos++;
 572        return pos;
 573}
 574
 575static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 576{
 577        int len = ce_namelen(ce);
 578        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 579}
 580
 581/*
 582 * If we add a filename that aliases in the cache, we will use the
 583 * name that we already have - but we don't want to update the same
 584 * alias twice, because that implies that there were actually two
 585 * different files with aliasing names!
 586 *
 587 * So we use the CE_ADDED flag to verify that the alias was an old
 588 * one before we accept it as
 589 */
 590static struct cache_entry *create_alias_ce(struct index_state *istate,
 591                                           struct cache_entry *ce,
 592                                           struct cache_entry *alias)
 593{
 594        int len;
 595        struct cache_entry *new;
 596
 597        if (alias->ce_flags & CE_ADDED)
 598                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 599
 600        /* Ok, create the new entry using the name of the existing alias */
 601        len = ce_namelen(alias);
 602        new = xcalloc(1, cache_entry_size(len));
 603        memcpy(new->name, alias->name, len);
 604        copy_cache_entry(new, ce);
 605        save_or_free_index_entry(istate, ce);
 606        return new;
 607}
 608
 609void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
 610{
 611        unsigned char sha1[20];
 612        if (write_sha1_file("", 0, blob_type, sha1))
 613                die("cannot create an empty blob in the object database");
 614        hashcpy(ce->sha1, sha1);
 615}
 616
 617int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 618{
 619        int size, namelen, was_same;
 620        mode_t st_mode = st->st_mode;
 621        struct cache_entry *ce, *alias;
 622        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 623        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 624        int pretend = flags & ADD_CACHE_PRETEND;
 625        int intent_only = flags & ADD_CACHE_INTENT;
 626        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 627                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 628
 629        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 630                return error("%s: can only add regular files, symbolic links or git-directories", path);
 631
 632        namelen = strlen(path);
 633        if (S_ISDIR(st_mode)) {
 634                while (namelen && path[namelen-1] == '/')
 635                        namelen--;
 636        }
 637        size = cache_entry_size(namelen);
 638        ce = xcalloc(1, size);
 639        memcpy(ce->name, path, namelen);
 640        ce->ce_namelen = namelen;
 641        if (!intent_only)
 642                fill_stat_cache_info(ce, st);
 643        else
 644                ce->ce_flags |= CE_INTENT_TO_ADD;
 645
 646        if (trust_executable_bit && has_symlinks)
 647                ce->ce_mode = create_ce_mode(st_mode);
 648        else {
 649                /* If there is an existing entry, pick the mode bits and type
 650                 * from it, otherwise assume unexecutable regular file.
 651                 */
 652                struct cache_entry *ent;
 653                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 654
 655                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 656                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 657        }
 658
 659        /* When core.ignorecase=true, determine if a directory of the same name but differing
 660         * case already exists within the Git repository.  If it does, ensure the directory
 661         * case of the file being added to the repository matches (is folded into) the existing
 662         * entry's directory case.
 663         */
 664        if (ignore_case) {
 665                const char *startPtr = ce->name;
 666                const char *ptr = startPtr;
 667                while (*ptr) {
 668                        while (*ptr && *ptr != '/')
 669                                ++ptr;
 670                        if (*ptr == '/') {
 671                                struct cache_entry *foundce;
 672                                ++ptr;
 673                                foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
 674                                if (foundce) {
 675                                        memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
 676                                        startPtr = ptr;
 677                                }
 678                        }
 679                }
 680        }
 681
 682        alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 683        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 684                /* Nothing changed, really */
 685                free(ce);
 686                if (!S_ISGITLINK(alias->ce_mode))
 687                        ce_mark_uptodate(alias);
 688                alias->ce_flags |= CE_ADDED;
 689                return 0;
 690        }
 691        if (!intent_only) {
 692                if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
 693                        return error("unable to index file %s", path);
 694        } else
 695                set_object_name_for_intent_to_add_entry(ce);
 696
 697        if (ignore_case && alias && different_name(ce, alias))
 698                ce = create_alias_ce(istate, ce, alias);
 699        ce->ce_flags |= CE_ADDED;
 700
 701        /* It was suspected to be racily clean, but it turns out to be Ok */
 702        was_same = (alias &&
 703                    !ce_stage(alias) &&
 704                    !hashcmp(alias->sha1, ce->sha1) &&
 705                    ce->ce_mode == alias->ce_mode);
 706
 707        if (pretend)
 708                ;
 709        else if (add_index_entry(istate, ce, add_option))
 710                return error("unable to add %s to index",path);
 711        if (verbose && !was_same)
 712                printf("add '%s'\n", path);
 713        return 0;
 714}
 715
 716int add_file_to_index(struct index_state *istate, const char *path, int flags)
 717{
 718        struct stat st;
 719        if (lstat(path, &st))
 720                die_errno("unable to stat '%s'", path);
 721        return add_to_index(istate, path, &st, flags);
 722}
 723
 724struct cache_entry *make_cache_entry(unsigned int mode,
 725                const unsigned char *sha1, const char *path, int stage,
 726                unsigned int refresh_options)
 727{
 728        int size, len;
 729        struct cache_entry *ce, *ret;
 730
 731        if (!verify_path(path)) {
 732                error("Invalid path '%s'", path);
 733                return NULL;
 734        }
 735
 736        len = strlen(path);
 737        size = cache_entry_size(len);
 738        ce = xcalloc(1, size);
 739
 740        hashcpy(ce->sha1, sha1);
 741        memcpy(ce->name, path, len);
 742        ce->ce_flags = create_ce_flags(stage);
 743        ce->ce_namelen = len;
 744        ce->ce_mode = create_ce_mode(mode);
 745
 746        ret = refresh_cache_entry(ce, refresh_options);
 747        if (!ret) {
 748                free(ce);
 749                return NULL;
 750        } else {
 751                return ret;
 752        }
 753}
 754
 755int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
 756{
 757        int len = ce_namelen(a);
 758        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 759}
 760
 761/*
 762 * We fundamentally don't like some paths: we don't want
 763 * dot or dot-dot anywhere, and for obvious reasons don't
 764 * want to recurse into ".git" either.
 765 *
 766 * Also, we don't want double slashes or slashes at the
 767 * end that can make pathnames ambiguous.
 768 */
 769static int verify_dotfile(const char *rest)
 770{
 771        /*
 772         * The first character was '.', but that
 773         * has already been discarded, we now test
 774         * the rest.
 775         */
 776
 777        /* "." is not allowed */
 778        if (*rest == '\0' || is_dir_sep(*rest))
 779                return 0;
 780
 781        switch (*rest) {
 782        /*
 783         * ".git" followed by  NUL or slash is bad. This
 784         * shares the path end test with the ".." case.
 785         */
 786        case 'g':
 787        case 'G':
 788                if (rest[1] != 'i' && rest[1] != 'I')
 789                        break;
 790                if (rest[2] != 't' && rest[2] != 'T')
 791                        break;
 792                rest += 2;
 793        /* fallthrough */
 794        case '.':
 795                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 796                        return 0;
 797        }
 798        return 1;
 799}
 800
 801int verify_path(const char *path)
 802{
 803        char c;
 804
 805        if (has_dos_drive_prefix(path))
 806                return 0;
 807
 808        goto inside;
 809        for (;;) {
 810                if (!c)
 811                        return 1;
 812                if (is_dir_sep(c)) {
 813inside:
 814                        if (protect_hfs && is_hfs_dotgit(path))
 815                                return 0;
 816                        if (protect_ntfs && is_ntfs_dotgit(path))
 817                                return 0;
 818                        c = *path++;
 819                        if ((c == '.' && !verify_dotfile(path)) ||
 820                            is_dir_sep(c) || c == '\0')
 821                                return 0;
 822                }
 823                c = *path++;
 824        }
 825}
 826
 827/*
 828 * Do we have another file that has the beginning components being a
 829 * proper superset of the name we're trying to add?
 830 */
 831static int has_file_name(struct index_state *istate,
 832                         const struct cache_entry *ce, int pos, int ok_to_replace)
 833{
 834        int retval = 0;
 835        int len = ce_namelen(ce);
 836        int stage = ce_stage(ce);
 837        const char *name = ce->name;
 838
 839        while (pos < istate->cache_nr) {
 840                struct cache_entry *p = istate->cache[pos++];
 841
 842                if (len >= ce_namelen(p))
 843                        break;
 844                if (memcmp(name, p->name, len))
 845                        break;
 846                if (ce_stage(p) != stage)
 847                        continue;
 848                if (p->name[len] != '/')
 849                        continue;
 850                if (p->ce_flags & CE_REMOVE)
 851                        continue;
 852                retval = -1;
 853                if (!ok_to_replace)
 854                        break;
 855                remove_index_entry_at(istate, --pos);
 856        }
 857        return retval;
 858}
 859
 860/*
 861 * Do we have another file with a pathname that is a proper
 862 * subset of the name we're trying to add?
 863 */
 864static int has_dir_name(struct index_state *istate,
 865                        const struct cache_entry *ce, int pos, int ok_to_replace)
 866{
 867        int retval = 0;
 868        int stage = ce_stage(ce);
 869        const char *name = ce->name;
 870        const char *slash = name + ce_namelen(ce);
 871
 872        for (;;) {
 873                int len;
 874
 875                for (;;) {
 876                        if (*--slash == '/')
 877                                break;
 878                        if (slash <= ce->name)
 879                                return retval;
 880                }
 881                len = slash - name;
 882
 883                pos = index_name_stage_pos(istate, name, len, stage);
 884                if (pos >= 0) {
 885                        /*
 886                         * Found one, but not so fast.  This could
 887                         * be a marker that says "I was here, but
 888                         * I am being removed".  Such an entry is
 889                         * not a part of the resulting tree, and
 890                         * it is Ok to have a directory at the same
 891                         * path.
 892                         */
 893                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 894                                retval = -1;
 895                                if (!ok_to_replace)
 896                                        break;
 897                                remove_index_entry_at(istate, pos);
 898                                continue;
 899                        }
 900                }
 901                else
 902                        pos = -pos-1;
 903
 904                /*
 905                 * Trivial optimization: if we find an entry that
 906                 * already matches the sub-directory, then we know
 907                 * we're ok, and we can exit.
 908                 */
 909                while (pos < istate->cache_nr) {
 910                        struct cache_entry *p = istate->cache[pos];
 911                        if ((ce_namelen(p) <= len) ||
 912                            (p->name[len] != '/') ||
 913                            memcmp(p->name, name, len))
 914                                break; /* not our subdirectory */
 915                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 916                                /*
 917                                 * p is at the same stage as our entry, and
 918                                 * is a subdirectory of what we are looking
 919                                 * at, so we cannot have conflicts at our
 920                                 * level or anything shorter.
 921                                 */
 922                                return retval;
 923                        pos++;
 924                }
 925        }
 926        return retval;
 927}
 928
 929/* We may be in a situation where we already have path/file and path
 930 * is being added, or we already have path and path/file is being
 931 * added.  Either one would result in a nonsense tree that has path
 932 * twice when git-write-tree tries to write it out.  Prevent it.
 933 *
 934 * If ok-to-replace is specified, we remove the conflicting entries
 935 * from the cache so the caller should recompute the insert position.
 936 * When this happens, we return non-zero.
 937 */
 938static int check_file_directory_conflict(struct index_state *istate,
 939                                         const struct cache_entry *ce,
 940                                         int pos, int ok_to_replace)
 941{
 942        int retval;
 943
 944        /*
 945         * When ce is an "I am going away" entry, we allow it to be added
 946         */
 947        if (ce->ce_flags & CE_REMOVE)
 948                return 0;
 949
 950        /*
 951         * We check if the path is a sub-path of a subsequent pathname
 952         * first, since removing those will not change the position
 953         * in the array.
 954         */
 955        retval = has_file_name(istate, ce, pos, ok_to_replace);
 956
 957        /*
 958         * Then check if the path might have a clashing sub-directory
 959         * before it.
 960         */
 961        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 962}
 963
 964static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 965{
 966        int pos;
 967        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 968        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 969        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 970        int new_only = option & ADD_CACHE_NEW_ONLY;
 971
 972        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
 973                cache_tree_invalidate_path(istate, ce->name);
 974        pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 975
 976        /* existing match? Just replace it. */
 977        if (pos >= 0) {
 978                if (!new_only)
 979                        replace_index_entry(istate, pos, ce);
 980                return 0;
 981        }
 982        pos = -pos-1;
 983
 984        /*
 985         * Inserting a merged entry ("stage 0") into the index
 986         * will always replace all non-merged entries..
 987         */
 988        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 989                while (ce_same_name(istate->cache[pos], ce)) {
 990                        ok_to_add = 1;
 991                        if (!remove_index_entry_at(istate, pos))
 992                                break;
 993                }
 994        }
 995
 996        if (!ok_to_add)
 997                return -1;
 998        if (!verify_path(ce->name))
 999                return error("Invalid path '%s'", ce->name);
1000
1001        if (!skip_df_check &&
1002            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1003                if (!ok_to_replace)
1004                        return error("'%s' appears as both a file and as a directory",
1005                                     ce->name);
1006                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1007                pos = -pos-1;
1008        }
1009        return pos + 1;
1010}
1011
1012int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1013{
1014        int pos;
1015
1016        if (option & ADD_CACHE_JUST_APPEND)
1017                pos = istate->cache_nr;
1018        else {
1019                int ret;
1020                ret = add_index_entry_with_check(istate, ce, option);
1021                if (ret <= 0)
1022                        return ret;
1023                pos = ret - 1;
1024        }
1025
1026        /* Make sure the array is big enough .. */
1027        ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1028
1029        /* Add it in.. */
1030        istate->cache_nr++;
1031        if (istate->cache_nr > pos + 1)
1032                memmove(istate->cache + pos + 1,
1033                        istate->cache + pos,
1034                        (istate->cache_nr - pos - 1) * sizeof(ce));
1035        set_index_entry(istate, pos, ce);
1036        istate->cache_changed |= CE_ENTRY_ADDED;
1037        return 0;
1038}
1039
1040/*
1041 * "refresh" does not calculate a new sha1 file or bring the
1042 * cache up-to-date for mode/content changes. But what it
1043 * _does_ do is to "re-match" the stat information of a file
1044 * with the cache, so that you can refresh the cache for a
1045 * file that hasn't been changed but where the stat entry is
1046 * out of date.
1047 *
1048 * For example, you'd want to do this after doing a "git-read-tree",
1049 * to link up the stat cache details with the proper files.
1050 */
1051static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1052                                             struct cache_entry *ce,
1053                                             unsigned int options, int *err,
1054                                             int *changed_ret)
1055{
1056        struct stat st;
1057        struct cache_entry *updated;
1058        int changed, size;
1059        int refresh = options & CE_MATCH_REFRESH;
1060        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1061        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1062        int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1063
1064        if (!refresh || ce_uptodate(ce))
1065                return ce;
1066
1067        /*
1068         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1069         * that the change to the work tree does not matter and told
1070         * us not to worry.
1071         */
1072        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1073                ce_mark_uptodate(ce);
1074                return ce;
1075        }
1076        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1077                ce_mark_uptodate(ce);
1078                return ce;
1079        }
1080
1081        if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1082                if (ignore_missing)
1083                        return ce;
1084                if (err)
1085                        *err = ENOENT;
1086                return NULL;
1087        }
1088
1089        if (lstat(ce->name, &st) < 0) {
1090                if (ignore_missing && errno == ENOENT)
1091                        return ce;
1092                if (err)
1093                        *err = errno;
1094                return NULL;
1095        }
1096
1097        changed = ie_match_stat(istate, ce, &st, options);
1098        if (changed_ret)
1099                *changed_ret = changed;
1100        if (!changed) {
1101                /*
1102                 * The path is unchanged.  If we were told to ignore
1103                 * valid bit, then we did the actual stat check and
1104                 * found that the entry is unmodified.  If the entry
1105                 * is not marked VALID, this is the place to mark it
1106                 * valid again, under "assume unchanged" mode.
1107                 */
1108                if (ignore_valid && assume_unchanged &&
1109                    !(ce->ce_flags & CE_VALID))
1110                        ; /* mark this one VALID again */
1111                else {
1112                        /*
1113                         * We do not mark the index itself "modified"
1114                         * because CE_UPTODATE flag is in-core only;
1115                         * we are not going to write this change out.
1116                         */
1117                        if (!S_ISGITLINK(ce->ce_mode))
1118                                ce_mark_uptodate(ce);
1119                        return ce;
1120                }
1121        }
1122
1123        if (ie_modified(istate, ce, &st, options)) {
1124                if (err)
1125                        *err = EINVAL;
1126                return NULL;
1127        }
1128
1129        size = ce_size(ce);
1130        updated = xmalloc(size);
1131        memcpy(updated, ce, size);
1132        fill_stat_cache_info(updated, &st);
1133        /*
1134         * If ignore_valid is not set, we should leave CE_VALID bit
1135         * alone.  Otherwise, paths marked with --no-assume-unchanged
1136         * (i.e. things to be edited) will reacquire CE_VALID bit
1137         * automatically, which is not really what we want.
1138         */
1139        if (!ignore_valid && assume_unchanged &&
1140            !(ce->ce_flags & CE_VALID))
1141                updated->ce_flags &= ~CE_VALID;
1142
1143        /* istate->cache_changed is updated in the caller */
1144        return updated;
1145}
1146
1147static void show_file(const char * fmt, const char * name, int in_porcelain,
1148                      int * first, const char *header_msg)
1149{
1150        if (in_porcelain && *first && header_msg) {
1151                printf("%s\n", header_msg);
1152                *first = 0;
1153        }
1154        printf(fmt, name);
1155}
1156
1157int refresh_index(struct index_state *istate, unsigned int flags,
1158                  const struct pathspec *pathspec,
1159                  char *seen, const char *header_msg)
1160{
1161        int i;
1162        int has_errors = 0;
1163        int really = (flags & REFRESH_REALLY) != 0;
1164        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1165        int quiet = (flags & REFRESH_QUIET) != 0;
1166        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1167        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1168        int first = 1;
1169        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1170        unsigned int options = (CE_MATCH_REFRESH |
1171                                (really ? CE_MATCH_IGNORE_VALID : 0) |
1172                                (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1173        const char *modified_fmt;
1174        const char *deleted_fmt;
1175        const char *typechange_fmt;
1176        const char *added_fmt;
1177        const char *unmerged_fmt;
1178
1179        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1180        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1181        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1182        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1183        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1184        for (i = 0; i < istate->cache_nr; i++) {
1185                struct cache_entry *ce, *new;
1186                int cache_errno = 0;
1187                int changed = 0;
1188                int filtered = 0;
1189
1190                ce = istate->cache[i];
1191                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1192                        continue;
1193
1194                if (pathspec && !ce_path_match(ce, pathspec, seen))
1195                        filtered = 1;
1196
1197                if (ce_stage(ce)) {
1198                        while ((i < istate->cache_nr) &&
1199                               ! strcmp(istate->cache[i]->name, ce->name))
1200                                i++;
1201                        i--;
1202                        if (allow_unmerged)
1203                                continue;
1204                        if (!filtered)
1205                                show_file(unmerged_fmt, ce->name, in_porcelain,
1206                                          &first, header_msg);
1207                        has_errors = 1;
1208                        continue;
1209                }
1210
1211                if (filtered)
1212                        continue;
1213
1214                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1215                if (new == ce)
1216                        continue;
1217                if (!new) {
1218                        const char *fmt;
1219
1220                        if (really && cache_errno == EINVAL) {
1221                                /* If we are doing --really-refresh that
1222                                 * means the index is not valid anymore.
1223                                 */
1224                                ce->ce_flags &= ~CE_VALID;
1225                                ce->ce_flags |= CE_UPDATE_IN_BASE;
1226                                istate->cache_changed |= CE_ENTRY_CHANGED;
1227                        }
1228                        if (quiet)
1229                                continue;
1230
1231                        if (cache_errno == ENOENT)
1232                                fmt = deleted_fmt;
1233                        else if (ce->ce_flags & CE_INTENT_TO_ADD)
1234                                fmt = added_fmt; /* must be before other checks */
1235                        else if (changed & TYPE_CHANGED)
1236                                fmt = typechange_fmt;
1237                        else
1238                                fmt = modified_fmt;
1239                        show_file(fmt,
1240                                  ce->name, in_porcelain, &first, header_msg);
1241                        has_errors = 1;
1242                        continue;
1243                }
1244
1245                replace_index_entry(istate, i, new);
1246        }
1247        return has_errors;
1248}
1249
1250static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1251                                               unsigned int options)
1252{
1253        return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1254}
1255
1256
1257/*****************************************************************
1258 * Index File I/O
1259 *****************************************************************/
1260
1261#define INDEX_FORMAT_DEFAULT 3
1262
1263static unsigned int get_index_format_default(void)
1264{
1265        char *envversion = getenv("GIT_INDEX_VERSION");
1266        char *endp;
1267        int value;
1268        unsigned int version = INDEX_FORMAT_DEFAULT;
1269
1270        if (!envversion) {
1271                if (!git_config_get_int("index.version", &value))
1272                        version = value;
1273                if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1274                        warning(_("index.version set, but the value is invalid.\n"
1275                                  "Using version %i"), INDEX_FORMAT_DEFAULT);
1276                        return INDEX_FORMAT_DEFAULT;
1277                }
1278                return version;
1279        }
1280
1281        version = strtoul(envversion, &endp, 10);
1282        if (*endp ||
1283            version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1284                warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1285                          "Using version %i"), INDEX_FORMAT_DEFAULT);
1286                version = INDEX_FORMAT_DEFAULT;
1287        }
1288        return version;
1289}
1290
1291/*
1292 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1293 * Again - this is just a (very strong in practice) heuristic that
1294 * the inode hasn't changed.
1295 *
1296 * We save the fields in big-endian order to allow using the
1297 * index file over NFS transparently.
1298 */
1299struct ondisk_cache_entry {
1300        struct cache_time ctime;
1301        struct cache_time mtime;
1302        uint32_t dev;
1303        uint32_t ino;
1304        uint32_t mode;
1305        uint32_t uid;
1306        uint32_t gid;
1307        uint32_t size;
1308        unsigned char sha1[20];
1309        uint16_t flags;
1310        char name[FLEX_ARRAY]; /* more */
1311};
1312
1313/*
1314 * This struct is used when CE_EXTENDED bit is 1
1315 * The struct must match ondisk_cache_entry exactly from
1316 * ctime till flags
1317 */
1318struct ondisk_cache_entry_extended {
1319        struct cache_time ctime;
1320        struct cache_time mtime;
1321        uint32_t dev;
1322        uint32_t ino;
1323        uint32_t mode;
1324        uint32_t uid;
1325        uint32_t gid;
1326        uint32_t size;
1327        unsigned char sha1[20];
1328        uint16_t flags;
1329        uint16_t flags2;
1330        char name[FLEX_ARRAY]; /* more */
1331};
1332
1333/* These are only used for v3 or lower */
1334#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1335#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1336#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1337#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1338                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1339                            ondisk_cache_entry_size(ce_namelen(ce)))
1340
1341static int verify_hdr(struct cache_header *hdr, unsigned long size)
1342{
1343        git_SHA_CTX c;
1344        unsigned char sha1[20];
1345        int hdr_version;
1346
1347        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1348                return error("bad signature");
1349        hdr_version = ntohl(hdr->hdr_version);
1350        if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1351                return error("bad index version %d", hdr_version);
1352        git_SHA1_Init(&c);
1353        git_SHA1_Update(&c, hdr, size - 20);
1354        git_SHA1_Final(sha1, &c);
1355        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1356                return error("bad index file sha1 signature");
1357        return 0;
1358}
1359
1360static int read_index_extension(struct index_state *istate,
1361                                const char *ext, void *data, unsigned long sz)
1362{
1363        switch (CACHE_EXT(ext)) {
1364        case CACHE_EXT_TREE:
1365                istate->cache_tree = cache_tree_read(data, sz);
1366                break;
1367        case CACHE_EXT_RESOLVE_UNDO:
1368                istate->resolve_undo = resolve_undo_read(data, sz);
1369                break;
1370        case CACHE_EXT_LINK:
1371                if (read_link_extension(istate, data, sz))
1372                        return -1;
1373                break;
1374        case CACHE_EXT_UNTRACKED:
1375                istate->untracked = read_untracked_extension(data, sz);
1376                break;
1377        default:
1378                if (*ext < 'A' || 'Z' < *ext)
1379                        return error("index uses %.4s extension, which we do not understand",
1380                                     ext);
1381                fprintf(stderr, "ignoring %.4s extension\n", ext);
1382                break;
1383        }
1384        return 0;
1385}
1386
1387int hold_locked_index(struct lock_file *lk, int die_on_error)
1388{
1389        return hold_lock_file_for_update(lk, get_index_file(),
1390                                         die_on_error
1391                                         ? LOCK_DIE_ON_ERROR
1392                                         : 0);
1393}
1394
1395int read_index(struct index_state *istate)
1396{
1397        return read_index_from(istate, get_index_file());
1398}
1399
1400static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1401                                                   unsigned int flags,
1402                                                   const char *name,
1403                                                   size_t len)
1404{
1405        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1406
1407        ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1408        ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1409        ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1410        ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1411        ce->ce_stat_data.sd_dev   = get_be32(&ondisk->dev);
1412        ce->ce_stat_data.sd_ino   = get_be32(&ondisk->ino);
1413        ce->ce_mode  = get_be32(&ondisk->mode);
1414        ce->ce_stat_data.sd_uid   = get_be32(&ondisk->uid);
1415        ce->ce_stat_data.sd_gid   = get_be32(&ondisk->gid);
1416        ce->ce_stat_data.sd_size  = get_be32(&ondisk->size);
1417        ce->ce_flags = flags & ~CE_NAMEMASK;
1418        ce->ce_namelen = len;
1419        ce->index = 0;
1420        hashcpy(ce->sha1, ondisk->sha1);
1421        memcpy(ce->name, name, len);
1422        ce->name[len] = '\0';
1423        return ce;
1424}
1425
1426/*
1427 * Adjacent cache entries tend to share the leading paths, so it makes
1428 * sense to only store the differences in later entries.  In the v4
1429 * on-disk format of the index, each on-disk cache entry stores the
1430 * number of bytes to be stripped from the end of the previous name,
1431 * and the bytes to append to the result, to come up with its name.
1432 */
1433static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1434{
1435        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1436        size_t len = decode_varint(&cp);
1437
1438        if (name->len < len)
1439                die("malformed name field in the index");
1440        strbuf_remove(name, name->len - len, len);
1441        for (ep = cp; *ep; ep++)
1442                ; /* find the end */
1443        strbuf_add(name, cp, ep - cp);
1444        return (const char *)ep + 1 - cp_;
1445}
1446
1447static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1448                                            unsigned long *ent_size,
1449                                            struct strbuf *previous_name)
1450{
1451        struct cache_entry *ce;
1452        size_t len;
1453        const char *name;
1454        unsigned int flags;
1455
1456        /* On-disk flags are just 16 bits */
1457        flags = get_be16(&ondisk->flags);
1458        len = flags & CE_NAMEMASK;
1459
1460        if (flags & CE_EXTENDED) {
1461                struct ondisk_cache_entry_extended *ondisk2;
1462                int extended_flags;
1463                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1464                extended_flags = get_be16(&ondisk2->flags2) << 16;
1465                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1466                if (extended_flags & ~CE_EXTENDED_FLAGS)
1467                        die("Unknown index entry format %08x", extended_flags);
1468                flags |= extended_flags;
1469                name = ondisk2->name;
1470        }
1471        else
1472                name = ondisk->name;
1473
1474        if (!previous_name) {
1475                /* v3 and earlier */
1476                if (len == CE_NAMEMASK)
1477                        len = strlen(name);
1478                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1479
1480                *ent_size = ondisk_ce_size(ce);
1481        } else {
1482                unsigned long consumed;
1483                consumed = expand_name_field(previous_name, name);
1484                ce = cache_entry_from_ondisk(ondisk, flags,
1485                                             previous_name->buf,
1486                                             previous_name->len);
1487
1488                *ent_size = (name - ((char *)ondisk)) + consumed;
1489        }
1490        return ce;
1491}
1492
1493static void check_ce_order(struct cache_entry *ce, struct cache_entry *next_ce)
1494{
1495        int name_compare = strcmp(ce->name, next_ce->name);
1496        if (0 < name_compare)
1497                die("unordered stage entries in index");
1498        if (!name_compare) {
1499                if (!ce_stage(ce))
1500                        die("multiple stage entries for merged file '%s'",
1501                                ce->name);
1502                if (ce_stage(ce) > ce_stage(next_ce))
1503                        die("unordered stage entries for '%s'",
1504                                ce->name);
1505        }
1506}
1507
1508/* remember to discard_cache() before reading a different cache! */
1509int do_read_index(struct index_state *istate, const char *path, int must_exist)
1510{
1511        int fd, i;
1512        struct stat st;
1513        unsigned long src_offset;
1514        struct cache_header *hdr;
1515        void *mmap;
1516        size_t mmap_size;
1517        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1518
1519        if (istate->initialized)
1520                return istate->cache_nr;
1521
1522        istate->timestamp.sec = 0;
1523        istate->timestamp.nsec = 0;
1524        fd = open(path, O_RDONLY);
1525        if (fd < 0) {
1526                if (!must_exist && errno == ENOENT)
1527                        return 0;
1528                die_errno("%s: index file open failed", path);
1529        }
1530
1531        if (fstat(fd, &st))
1532                die_errno("cannot stat the open index");
1533
1534        mmap_size = xsize_t(st.st_size);
1535        if (mmap_size < sizeof(struct cache_header) + 20)
1536                die("index file smaller than expected");
1537
1538        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1539        if (mmap == MAP_FAILED)
1540                die_errno("unable to map index file");
1541        close(fd);
1542
1543        hdr = mmap;
1544        if (verify_hdr(hdr, mmap_size) < 0)
1545                goto unmap;
1546
1547        hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1548        istate->version = ntohl(hdr->hdr_version);
1549        istate->cache_nr = ntohl(hdr->hdr_entries);
1550        istate->cache_alloc = alloc_nr(istate->cache_nr);
1551        istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1552        istate->initialized = 1;
1553
1554        if (istate->version == 4)
1555                previous_name = &previous_name_buf;
1556        else
1557                previous_name = NULL;
1558
1559        src_offset = sizeof(*hdr);
1560        for (i = 0; i < istate->cache_nr; i++) {
1561                struct ondisk_cache_entry *disk_ce;
1562                struct cache_entry *ce;
1563                unsigned long consumed;
1564
1565                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1566                ce = create_from_disk(disk_ce, &consumed, previous_name);
1567                set_index_entry(istate, i, ce);
1568
1569                if (i > 0)
1570                        check_ce_order(istate->cache[i - 1], ce);
1571
1572                src_offset += consumed;
1573        }
1574        strbuf_release(&previous_name_buf);
1575        istate->timestamp.sec = st.st_mtime;
1576        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1577
1578        while (src_offset <= mmap_size - 20 - 8) {
1579                /* After an array of active_nr index entries,
1580                 * there can be arbitrary number of extended
1581                 * sections, each of which is prefixed with
1582                 * extension name (4-byte) and section length
1583                 * in 4-byte network byte order.
1584                 */
1585                uint32_t extsize;
1586                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1587                extsize = ntohl(extsize);
1588                if (read_index_extension(istate,
1589                                         (const char *) mmap + src_offset,
1590                                         (char *) mmap + src_offset + 8,
1591                                         extsize) < 0)
1592                        goto unmap;
1593                src_offset += 8;
1594                src_offset += extsize;
1595        }
1596        munmap(mmap, mmap_size);
1597        return istate->cache_nr;
1598
1599unmap:
1600        munmap(mmap, mmap_size);
1601        die("index file corrupt");
1602}
1603
1604int read_index_from(struct index_state *istate, const char *path)
1605{
1606        struct split_index *split_index;
1607        int ret;
1608
1609        /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1610        if (istate->initialized)
1611                return istate->cache_nr;
1612
1613        ret = do_read_index(istate, path, 0);
1614        split_index = istate->split_index;
1615        if (!split_index)
1616                return ret;
1617
1618        if (is_null_sha1(split_index->base_sha1))
1619                return ret;
1620
1621        if (split_index->base)
1622                discard_index(split_index->base);
1623        else
1624                split_index->base = xcalloc(1, sizeof(*split_index->base));
1625        ret = do_read_index(split_index->base,
1626                            git_path("sharedindex.%s",
1627                                     sha1_to_hex(split_index->base_sha1)), 1);
1628        if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1629                die("broken index, expect %s in %s, got %s",
1630                    sha1_to_hex(split_index->base_sha1),
1631                    git_path("sharedindex.%s",
1632                                     sha1_to_hex(split_index->base_sha1)),
1633                    sha1_to_hex(split_index->base->sha1));
1634        merge_base_index(istate);
1635        return ret;
1636}
1637
1638int is_index_unborn(struct index_state *istate)
1639{
1640        return (!istate->cache_nr && !istate->timestamp.sec);
1641}
1642
1643int discard_index(struct index_state *istate)
1644{
1645        int i;
1646
1647        for (i = 0; i < istate->cache_nr; i++) {
1648                if (istate->cache[i]->index &&
1649                    istate->split_index &&
1650                    istate->split_index->base &&
1651                    istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1652                    istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1653                        continue;
1654                free(istate->cache[i]);
1655        }
1656        resolve_undo_clear_index(istate);
1657        istate->cache_nr = 0;
1658        istate->cache_changed = 0;
1659        istate->timestamp.sec = 0;
1660        istate->timestamp.nsec = 0;
1661        free_name_hash(istate);
1662        cache_tree_free(&(istate->cache_tree));
1663        istate->initialized = 0;
1664        free(istate->cache);
1665        istate->cache = NULL;
1666        istate->cache_alloc = 0;
1667        discard_split_index(istate);
1668        free_untracked_cache(istate->untracked);
1669        istate->untracked = NULL;
1670        return 0;
1671}
1672
1673int unmerged_index(const struct index_state *istate)
1674{
1675        int i;
1676        for (i = 0; i < istate->cache_nr; i++) {
1677                if (ce_stage(istate->cache[i]))
1678                        return 1;
1679        }
1680        return 0;
1681}
1682
1683#define WRITE_BUFFER_SIZE 8192
1684static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1685static unsigned long write_buffer_len;
1686
1687static int ce_write_flush(git_SHA_CTX *context, int fd)
1688{
1689        unsigned int buffered = write_buffer_len;
1690        if (buffered) {
1691                git_SHA1_Update(context, write_buffer, buffered);
1692                if (write_in_full(fd, write_buffer, buffered) != buffered)
1693                        return -1;
1694                write_buffer_len = 0;
1695        }
1696        return 0;
1697}
1698
1699static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1700{
1701        while (len) {
1702                unsigned int buffered = write_buffer_len;
1703                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1704                if (partial > len)
1705                        partial = len;
1706                memcpy(write_buffer + buffered, data, partial);
1707                buffered += partial;
1708                if (buffered == WRITE_BUFFER_SIZE) {
1709                        write_buffer_len = buffered;
1710                        if (ce_write_flush(context, fd))
1711                                return -1;
1712                        buffered = 0;
1713                }
1714                write_buffer_len = buffered;
1715                len -= partial;
1716                data = (char *) data + partial;
1717        }
1718        return 0;
1719}
1720
1721static int write_index_ext_header(git_SHA_CTX *context, int fd,
1722                                  unsigned int ext, unsigned int sz)
1723{
1724        ext = htonl(ext);
1725        sz = htonl(sz);
1726        return ((ce_write(context, fd, &ext, 4) < 0) ||
1727                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1728}
1729
1730static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1731{
1732        unsigned int left = write_buffer_len;
1733
1734        if (left) {
1735                write_buffer_len = 0;
1736                git_SHA1_Update(context, write_buffer, left);
1737        }
1738
1739        /* Flush first if not enough space for SHA1 signature */
1740        if (left + 20 > WRITE_BUFFER_SIZE) {
1741                if (write_in_full(fd, write_buffer, left) != left)
1742                        return -1;
1743                left = 0;
1744        }
1745
1746        /* Append the SHA1 signature at the end */
1747        git_SHA1_Final(write_buffer + left, context);
1748        hashcpy(sha1, write_buffer + left);
1749        left += 20;
1750        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1751}
1752
1753static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1754{
1755        /*
1756         * The only thing we care about in this function is to smudge the
1757         * falsely clean entry due to touch-update-touch race, so we leave
1758         * everything else as they are.  We are called for entries whose
1759         * ce_stat_data.sd_mtime match the index file mtime.
1760         *
1761         * Note that this actually does not do much for gitlinks, for
1762         * which ce_match_stat_basic() always goes to the actual
1763         * contents.  The caller checks with is_racy_timestamp() which
1764         * always says "no" for gitlinks, so we are not called for them ;-)
1765         */
1766        struct stat st;
1767
1768        if (lstat(ce->name, &st) < 0)
1769                return;
1770        if (ce_match_stat_basic(ce, &st))
1771                return;
1772        if (ce_modified_check_fs(ce, &st)) {
1773                /* This is "racily clean"; smudge it.  Note that this
1774                 * is a tricky code.  At first glance, it may appear
1775                 * that it can break with this sequence:
1776                 *
1777                 * $ echo xyzzy >frotz
1778                 * $ git-update-index --add frotz
1779                 * $ : >frotz
1780                 * $ sleep 3
1781                 * $ echo filfre >nitfol
1782                 * $ git-update-index --add nitfol
1783                 *
1784                 * but it does not.  When the second update-index runs,
1785                 * it notices that the entry "frotz" has the same timestamp
1786                 * as index, and if we were to smudge it by resetting its
1787                 * size to zero here, then the object name recorded
1788                 * in index is the 6-byte file but the cached stat information
1789                 * becomes zero --- which would then match what we would
1790                 * obtain from the filesystem next time we stat("frotz").
1791                 *
1792                 * However, the second update-index, before calling
1793                 * this function, notices that the cached size is 6
1794                 * bytes and what is on the filesystem is an empty
1795                 * file, and never calls us, so the cached size information
1796                 * for "frotz" stays 6 which does not match the filesystem.
1797                 */
1798                ce->ce_stat_data.sd_size = 0;
1799        }
1800}
1801
1802/* Copy miscellaneous fields but not the name */
1803static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1804                                       struct cache_entry *ce)
1805{
1806        short flags;
1807
1808        ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1809        ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1810        ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1811        ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1812        ondisk->dev  = htonl(ce->ce_stat_data.sd_dev);
1813        ondisk->ino  = htonl(ce->ce_stat_data.sd_ino);
1814        ondisk->mode = htonl(ce->ce_mode);
1815        ondisk->uid  = htonl(ce->ce_stat_data.sd_uid);
1816        ondisk->gid  = htonl(ce->ce_stat_data.sd_gid);
1817        ondisk->size = htonl(ce->ce_stat_data.sd_size);
1818        hashcpy(ondisk->sha1, ce->sha1);
1819
1820        flags = ce->ce_flags & ~CE_NAMEMASK;
1821        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1822        ondisk->flags = htons(flags);
1823        if (ce->ce_flags & CE_EXTENDED) {
1824                struct ondisk_cache_entry_extended *ondisk2;
1825                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1826                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1827                return ondisk2->name;
1828        }
1829        else {
1830                return ondisk->name;
1831        }
1832}
1833
1834static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1835                          struct strbuf *previous_name)
1836{
1837        int size;
1838        struct ondisk_cache_entry *ondisk;
1839        int saved_namelen = saved_namelen; /* compiler workaround */
1840        char *name;
1841        int result;
1842
1843        if (ce->ce_flags & CE_STRIP_NAME) {
1844                saved_namelen = ce_namelen(ce);
1845                ce->ce_namelen = 0;
1846        }
1847
1848        if (!previous_name) {
1849                size = ondisk_ce_size(ce);
1850                ondisk = xcalloc(1, size);
1851                name = copy_cache_entry_to_ondisk(ondisk, ce);
1852                memcpy(name, ce->name, ce_namelen(ce));
1853        } else {
1854                int common, to_remove, prefix_size;
1855                unsigned char to_remove_vi[16];
1856                for (common = 0;
1857                     (ce->name[common] &&
1858                      common < previous_name->len &&
1859                      ce->name[common] == previous_name->buf[common]);
1860                     common++)
1861                        ; /* still matching */
1862                to_remove = previous_name->len - common;
1863                prefix_size = encode_varint(to_remove, to_remove_vi);
1864
1865                if (ce->ce_flags & CE_EXTENDED)
1866                        size = offsetof(struct ondisk_cache_entry_extended, name);
1867                else
1868                        size = offsetof(struct ondisk_cache_entry, name);
1869                size += prefix_size + (ce_namelen(ce) - common + 1);
1870
1871                ondisk = xcalloc(1, size);
1872                name = copy_cache_entry_to_ondisk(ondisk, ce);
1873                memcpy(name, to_remove_vi, prefix_size);
1874                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1875
1876                strbuf_splice(previous_name, common, to_remove,
1877                              ce->name + common, ce_namelen(ce) - common);
1878        }
1879        if (ce->ce_flags & CE_STRIP_NAME) {
1880                ce->ce_namelen = saved_namelen;
1881                ce->ce_flags &= ~CE_STRIP_NAME;
1882        }
1883
1884        result = ce_write(c, fd, ondisk, size);
1885        free(ondisk);
1886        return result;
1887}
1888
1889/*
1890 * This function verifies if index_state has the correct sha1 of the
1891 * index file.  Don't die if we have any other failure, just return 0.
1892 */
1893static int verify_index_from(const struct index_state *istate, const char *path)
1894{
1895        int fd;
1896        ssize_t n;
1897        struct stat st;
1898        unsigned char sha1[20];
1899
1900        if (!istate->initialized)
1901                return 0;
1902
1903        fd = open(path, O_RDONLY);
1904        if (fd < 0)
1905                return 0;
1906
1907        if (fstat(fd, &st))
1908                goto out;
1909
1910        if (st.st_size < sizeof(struct cache_header) + 20)
1911                goto out;
1912
1913        n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1914        if (n != 20)
1915                goto out;
1916
1917        if (hashcmp(istate->sha1, sha1))
1918                goto out;
1919
1920        close(fd);
1921        return 1;
1922
1923out:
1924        close(fd);
1925        return 0;
1926}
1927
1928static int verify_index(const struct index_state *istate)
1929{
1930        return verify_index_from(istate, get_index_file());
1931}
1932
1933static int has_racy_timestamp(struct index_state *istate)
1934{
1935        int entries = istate->cache_nr;
1936        int i;
1937
1938        for (i = 0; i < entries; i++) {
1939                struct cache_entry *ce = istate->cache[i];
1940                if (is_racy_timestamp(istate, ce))
1941                        return 1;
1942        }
1943        return 0;
1944}
1945
1946/*
1947 * Opportunistically update the index but do not complain if we can't
1948 */
1949void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1950{
1951        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1952            verify_index(istate) &&
1953            write_locked_index(istate, lockfile, COMMIT_LOCK))
1954                rollback_lock_file(lockfile);
1955}
1956
1957static int do_write_index(struct index_state *istate, int newfd,
1958                          int strip_extensions)
1959{
1960        git_SHA_CTX c;
1961        struct cache_header hdr;
1962        int i, err, removed, extended, hdr_version;
1963        struct cache_entry **cache = istate->cache;
1964        int entries = istate->cache_nr;
1965        struct stat st;
1966        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1967
1968        for (i = removed = extended = 0; i < entries; i++) {
1969                if (cache[i]->ce_flags & CE_REMOVE)
1970                        removed++;
1971
1972                /* reduce extended entries if possible */
1973                cache[i]->ce_flags &= ~CE_EXTENDED;
1974                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1975                        extended++;
1976                        cache[i]->ce_flags |= CE_EXTENDED;
1977                }
1978        }
1979
1980        if (!istate->version) {
1981                istate->version = get_index_format_default();
1982                if (getenv("GIT_TEST_SPLIT_INDEX"))
1983                        init_split_index(istate);
1984        }
1985
1986        /* demote version 3 to version 2 when the latter suffices */
1987        if (istate->version == 3 || istate->version == 2)
1988                istate->version = extended ? 3 : 2;
1989
1990        hdr_version = istate->version;
1991
1992        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1993        hdr.hdr_version = htonl(hdr_version);
1994        hdr.hdr_entries = htonl(entries - removed);
1995
1996        git_SHA1_Init(&c);
1997        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1998                return -1;
1999
2000        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2001        for (i = 0; i < entries; i++) {
2002                struct cache_entry *ce = cache[i];
2003                if (ce->ce_flags & CE_REMOVE)
2004                        continue;
2005                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2006                        ce_smudge_racily_clean_entry(ce);
2007                if (is_null_sha1(ce->sha1)) {
2008                        static const char msg[] = "cache entry has null sha1: %s";
2009                        static int allow = -1;
2010
2011                        if (allow < 0)
2012                                allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2013                        if (allow)
2014                                warning(msg, ce->name);
2015                        else
2016                                return error(msg, ce->name);
2017                }
2018                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2019                        return -1;
2020        }
2021        strbuf_release(&previous_name_buf);
2022
2023        /* Write extension data here */
2024        if (!strip_extensions && istate->split_index) {
2025                struct strbuf sb = STRBUF_INIT;
2026
2027                err = write_link_extension(&sb, istate) < 0 ||
2028                        write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2029                                               sb.len) < 0 ||
2030                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2031                strbuf_release(&sb);
2032                if (err)
2033                        return -1;
2034        }
2035        if (!strip_extensions && istate->cache_tree) {
2036                struct strbuf sb = STRBUF_INIT;
2037
2038                cache_tree_write(&sb, istate->cache_tree);
2039                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2040                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2041                strbuf_release(&sb);
2042                if (err)
2043                        return -1;
2044        }
2045        if (!strip_extensions && istate->resolve_undo) {
2046                struct strbuf sb = STRBUF_INIT;
2047
2048                resolve_undo_write(&sb, istate->resolve_undo);
2049                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2050                                             sb.len) < 0
2051                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2052                strbuf_release(&sb);
2053                if (err)
2054                        return -1;
2055        }
2056        if (!strip_extensions && istate->untracked) {
2057                struct strbuf sb = STRBUF_INIT;
2058
2059                write_untracked_extension(&sb, istate->untracked);
2060                err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2061                                             sb.len) < 0 ||
2062                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2063                strbuf_release(&sb);
2064                if (err)
2065                        return -1;
2066        }
2067
2068        if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2069                return -1;
2070        istate->timestamp.sec = (unsigned int)st.st_mtime;
2071        istate->timestamp.nsec = ST_MTIME_NSEC(st);
2072        return 0;
2073}
2074
2075void set_alternate_index_output(const char *name)
2076{
2077        alternate_index_output = name;
2078}
2079
2080static int commit_locked_index(struct lock_file *lk)
2081{
2082        if (alternate_index_output)
2083                return commit_lock_file_to(lk, alternate_index_output);
2084        else
2085                return commit_lock_file(lk);
2086}
2087
2088static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2089                                 unsigned flags)
2090{
2091        int ret = do_write_index(istate, lock->fd, 0);
2092        if (ret)
2093                return ret;
2094        assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2095               (COMMIT_LOCK | CLOSE_LOCK));
2096        if (flags & COMMIT_LOCK)
2097                return commit_locked_index(lock);
2098        else if (flags & CLOSE_LOCK)
2099                return close_lock_file(lock);
2100        else
2101                return ret;
2102}
2103
2104static int write_split_index(struct index_state *istate,
2105                             struct lock_file *lock,
2106                             unsigned flags)
2107{
2108        int ret;
2109        prepare_to_write_split_index(istate);
2110        ret = do_write_locked_index(istate, lock, flags);
2111        finish_writing_split_index(istate);
2112        return ret;
2113}
2114
2115static char *temporary_sharedindex;
2116
2117static void remove_temporary_sharedindex(void)
2118{
2119        if (temporary_sharedindex) {
2120                unlink_or_warn(temporary_sharedindex);
2121                free(temporary_sharedindex);
2122                temporary_sharedindex = NULL;
2123        }
2124}
2125
2126static void remove_temporary_sharedindex_on_signal(int signo)
2127{
2128        remove_temporary_sharedindex();
2129        sigchain_pop(signo);
2130        raise(signo);
2131}
2132
2133static int write_shared_index(struct index_state *istate,
2134                              struct lock_file *lock, unsigned flags)
2135{
2136        struct split_index *si = istate->split_index;
2137        static int installed_handler;
2138        int fd, ret;
2139
2140        temporary_sharedindex = git_pathdup("sharedindex_XXXXXX");
2141        fd = mkstemp(temporary_sharedindex);
2142        if (fd < 0) {
2143                free(temporary_sharedindex);
2144                temporary_sharedindex = NULL;
2145                hashclr(si->base_sha1);
2146                return do_write_locked_index(istate, lock, flags);
2147        }
2148        if (!installed_handler) {
2149                atexit(remove_temporary_sharedindex);
2150                sigchain_push_common(remove_temporary_sharedindex_on_signal);
2151        }
2152        move_cache_to_base_index(istate);
2153        ret = do_write_index(si->base, fd, 1);
2154        close(fd);
2155        if (ret) {
2156                remove_temporary_sharedindex();
2157                return ret;
2158        }
2159        ret = rename(temporary_sharedindex,
2160                     git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2161        free(temporary_sharedindex);
2162        temporary_sharedindex = NULL;
2163        if (!ret)
2164                hashcpy(si->base_sha1, si->base->sha1);
2165        return ret;
2166}
2167
2168int write_locked_index(struct index_state *istate, struct lock_file *lock,
2169                       unsigned flags)
2170{
2171        struct split_index *si = istate->split_index;
2172
2173        if (!si || alternate_index_output ||
2174            (istate->cache_changed & ~EXTMASK)) {
2175                if (si)
2176                        hashclr(si->base_sha1);
2177                return do_write_locked_index(istate, lock, flags);
2178        }
2179
2180        if (getenv("GIT_TEST_SPLIT_INDEX")) {
2181                int v = si->base_sha1[0];
2182                if ((v & 15) < 6)
2183                        istate->cache_changed |= SPLIT_INDEX_ORDERED;
2184        }
2185        if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2186                int ret = write_shared_index(istate, lock, flags);
2187                if (ret)
2188                        return ret;
2189        }
2190
2191        return write_split_index(istate, lock, flags);
2192}
2193
2194/*
2195 * Read the index file that is potentially unmerged into given
2196 * index_state, dropping any unmerged entries.  Returns true if
2197 * the index is unmerged.  Callers who want to refuse to work
2198 * from an unmerged state can call this and check its return value,
2199 * instead of calling read_cache().
2200 */
2201int read_index_unmerged(struct index_state *istate)
2202{
2203        int i;
2204        int unmerged = 0;
2205
2206        read_index(istate);
2207        for (i = 0; i < istate->cache_nr; i++) {
2208                struct cache_entry *ce = istate->cache[i];
2209                struct cache_entry *new_ce;
2210                int size, len;
2211
2212                if (!ce_stage(ce))
2213                        continue;
2214                unmerged = 1;
2215                len = ce_namelen(ce);
2216                size = cache_entry_size(len);
2217                new_ce = xcalloc(1, size);
2218                memcpy(new_ce->name, ce->name, len);
2219                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2220                new_ce->ce_namelen = len;
2221                new_ce->ce_mode = ce->ce_mode;
2222                if (add_index_entry(istate, new_ce, 0))
2223                        return error("%s: cannot drop to stage #0",
2224                                     new_ce->name);
2225        }
2226        return unmerged;
2227}
2228
2229/*
2230 * Returns 1 if the path is an "other" path with respect to
2231 * the index; that is, the path is not mentioned in the index at all,
2232 * either as a file, a directory with some files in the index,
2233 * or as an unmerged entry.
2234 *
2235 * We helpfully remove a trailing "/" from directories so that
2236 * the output of read_directory can be used as-is.
2237 */
2238int index_name_is_other(const struct index_state *istate, const char *name,
2239                int namelen)
2240{
2241        int pos;
2242        if (namelen && name[namelen - 1] == '/')
2243                namelen--;
2244        pos = index_name_pos(istate, name, namelen);
2245        if (0 <= pos)
2246                return 0;       /* exact match */
2247        pos = -pos - 1;
2248        if (pos < istate->cache_nr) {
2249                struct cache_entry *ce = istate->cache[pos];
2250                if (ce_namelen(ce) == namelen &&
2251                    !memcmp(ce->name, name, namelen))
2252                        return 0; /* Yup, this one exists unmerged */
2253        }
2254        return 1;
2255}
2256
2257void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2258{
2259        int pos, len;
2260        unsigned long sz;
2261        enum object_type type;
2262        void *data;
2263
2264        len = strlen(path);
2265        pos = index_name_pos(istate, path, len);
2266        if (pos < 0) {
2267                /*
2268                 * We might be in the middle of a merge, in which
2269                 * case we would read stage #2 (ours).
2270                 */
2271                int i;
2272                for (i = -pos - 1;
2273                     (pos < 0 && i < istate->cache_nr &&
2274                      !strcmp(istate->cache[i]->name, path));
2275                     i++)
2276                        if (ce_stage(istate->cache[i]) == 2)
2277                                pos = i;
2278        }
2279        if (pos < 0)
2280                return NULL;
2281        data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2282        if (!data || type != OBJ_BLOB) {
2283                free(data);
2284                return NULL;
2285        }
2286        if (size)
2287                *size = sz;
2288        return data;
2289}
2290
2291void stat_validity_clear(struct stat_validity *sv)
2292{
2293        free(sv->sd);
2294        sv->sd = NULL;
2295}
2296
2297int stat_validity_check(struct stat_validity *sv, const char *path)
2298{
2299        struct stat st;
2300
2301        if (stat(path, &st) < 0)
2302                return sv->sd == NULL;
2303        if (!sv->sd)
2304                return 0;
2305        return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2306}
2307
2308void stat_validity_update(struct stat_validity *sv, int fd)
2309{
2310        struct stat st;
2311
2312        if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2313                stat_validity_clear(sv);
2314        else {
2315                if (!sv->sd)
2316                        sv->sd = xcalloc(1, sizeof(struct stat_data));
2317                fill_stat_data(sv->sd, &st);
2318        }
2319}