read-cache.con commit Merge branch 'jc/push-default-explicit' into maint (af04b11)
   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 "tempfile.h"
   9#include "lockfile.h"
  10#include "cache-tree.h"
  11#include "refs.h"
  12#include "dir.h"
  13#include "tree.h"
  14#include "commit.h"
  15#include "blob.h"
  16#include "resolve-undo.h"
  17#include "strbuf.h"
  18#include "varint.h"
  19#include "split-index.h"
  20#include "utf8.h"
  21
  22/* Mask for the name length in ce_flags in the on-disk index */
  23
  24#define CE_NAMEMASK  (0x0fff)
  25
  26/* Index extensions.
  27 *
  28 * The first letter should be 'A'..'Z' for extensions that are not
  29 * necessary for a correct operation (i.e. optimization data).
  30 * When new extensions are added that _needs_ to be understood in
  31 * order to correctly interpret the index file, pick character that
  32 * is outside the range, to cause the reader to abort.
  33 */
  34
  35#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  36#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  37#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  38#define CACHE_EXT_LINK 0x6c696e6b         /* "link" */
  39#define CACHE_EXT_UNTRACKED 0x554E5452    /* "UNTR" */
  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 | UNTRACKED_CHANGED)
  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        untracked_cache_remove_from_index(istate, old->name);
  81        remove_index_entry_at(istate, nr);
  82        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  83}
  84
  85void fill_stat_data(struct stat_data *sd, struct stat *st)
  86{
  87        sd->sd_ctime.sec = (unsigned int)st->st_ctime;
  88        sd->sd_mtime.sec = (unsigned int)st->st_mtime;
  89        sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
  90        sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
  91        sd->sd_dev = st->st_dev;
  92        sd->sd_ino = st->st_ino;
  93        sd->sd_uid = st->st_uid;
  94        sd->sd_gid = st->st_gid;
  95        sd->sd_size = st->st_size;
  96}
  97
  98int match_stat_data(const struct stat_data *sd, struct stat *st)
  99{
 100        int changed = 0;
 101
 102        if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
 103                changed |= MTIME_CHANGED;
 104        if (trust_ctime && check_stat &&
 105            sd->sd_ctime.sec != (unsigned int)st->st_ctime)
 106                changed |= CTIME_CHANGED;
 107
 108#ifdef USE_NSEC
 109        if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
 110                changed |= MTIME_CHANGED;
 111        if (trust_ctime && check_stat &&
 112            sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
 113                changed |= CTIME_CHANGED;
 114#endif
 115
 116        if (check_stat) {
 117                if (sd->sd_uid != (unsigned int) st->st_uid ||
 118                        sd->sd_gid != (unsigned int) st->st_gid)
 119                        changed |= OWNER_CHANGED;
 120                if (sd->sd_ino != (unsigned int) st->st_ino)
 121                        changed |= INODE_CHANGED;
 122        }
 123
 124#ifdef USE_STDEV
 125        /*
 126         * st_dev breaks on network filesystems where different
 127         * clients will have different views of what "device"
 128         * the filesystem is on
 129         */
 130        if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
 131                        changed |= INODE_CHANGED;
 132#endif
 133
 134        if (sd->sd_size != (unsigned int) st->st_size)
 135                changed |= DATA_CHANGED;
 136
 137        return changed;
 138}
 139
 140/*
 141 * This only updates the "non-critical" parts of the directory
 142 * cache, ie the parts that aren't tracked by GIT, and only used
 143 * to validate the cache.
 144 */
 145void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 146{
 147        fill_stat_data(&ce->ce_stat_data, st);
 148
 149        if (assume_unchanged)
 150                ce->ce_flags |= CE_VALID;
 151
 152        if (S_ISREG(st->st_mode))
 153                ce_mark_uptodate(ce);
 154}
 155
 156static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
 157{
 158        int match = -1;
 159        static int cloexec = O_CLOEXEC;
 160        int fd = open(ce->name, O_RDONLY | cloexec);
 161
 162        if ((cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
 163                /* Try again w/o O_CLOEXEC: the kernel might not support it */
 164                cloexec &= ~O_CLOEXEC;
 165                fd = open(ce->name, O_RDONLY | cloexec);
 166        }
 167
 168        if (fd >= 0) {
 169                unsigned char sha1[20];
 170                if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
 171                        match = hashcmp(sha1, ce->oid.hash);
 172                /* index_fd() closed the file descriptor already */
 173        }
 174        return match;
 175}
 176
 177static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
 178{
 179        int match = -1;
 180        void *buffer;
 181        unsigned long size;
 182        enum object_type type;
 183        struct strbuf sb = STRBUF_INIT;
 184
 185        if (strbuf_readlink(&sb, ce->name, expected_size))
 186                return -1;
 187
 188        buffer = read_sha1_file(ce->oid.hash, &type, &size);
 189        if (buffer) {
 190                if (size == sb.len)
 191                        match = memcmp(buffer, sb.buf, size);
 192                free(buffer);
 193        }
 194        strbuf_release(&sb);
 195        return match;
 196}
 197
 198static int ce_compare_gitlink(const struct cache_entry *ce)
 199{
 200        unsigned char sha1[20];
 201
 202        /*
 203         * We don't actually require that the .git directory
 204         * under GITLINK directory be a valid git directory. It
 205         * might even be missing (in case nobody populated that
 206         * sub-project).
 207         *
 208         * If so, we consider it always to match.
 209         */
 210        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 211                return 0;
 212        return hashcmp(sha1, ce->oid.hash);
 213}
 214
 215static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
 216{
 217        switch (st->st_mode & S_IFMT) {
 218        case S_IFREG:
 219                if (ce_compare_data(ce, st))
 220                        return DATA_CHANGED;
 221                break;
 222        case S_IFLNK:
 223                if (ce_compare_link(ce, xsize_t(st->st_size)))
 224                        return DATA_CHANGED;
 225                break;
 226        case S_IFDIR:
 227                if (S_ISGITLINK(ce->ce_mode))
 228                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 229        default:
 230                return TYPE_CHANGED;
 231        }
 232        return 0;
 233}
 234
 235static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
 236{
 237        unsigned int changed = 0;
 238
 239        if (ce->ce_flags & CE_REMOVE)
 240                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 241
 242        switch (ce->ce_mode & S_IFMT) {
 243        case S_IFREG:
 244                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 245                /* We consider only the owner x bit to be relevant for
 246                 * "mode changes"
 247                 */
 248                if (trust_executable_bit &&
 249                    (0100 & (ce->ce_mode ^ st->st_mode)))
 250                        changed |= MODE_CHANGED;
 251                break;
 252        case S_IFLNK:
 253                if (!S_ISLNK(st->st_mode) &&
 254                    (has_symlinks || !S_ISREG(st->st_mode)))
 255                        changed |= TYPE_CHANGED;
 256                break;
 257        case S_IFGITLINK:
 258                /* We ignore most of the st_xxx fields for gitlinks */
 259                if (!S_ISDIR(st->st_mode))
 260                        changed |= TYPE_CHANGED;
 261                else if (ce_compare_gitlink(ce))
 262                        changed |= DATA_CHANGED;
 263                return changed;
 264        default:
 265                die("internal error: ce_mode is %o", ce->ce_mode);
 266        }
 267
 268        changed |= match_stat_data(&ce->ce_stat_data, st);
 269
 270        /* Racily smudged entry? */
 271        if (!ce->ce_stat_data.sd_size) {
 272                if (!is_empty_blob_sha1(ce->oid.hash))
 273                        changed |= DATA_CHANGED;
 274        }
 275
 276        return changed;
 277}
 278
 279static int is_racy_stat(const struct index_state *istate,
 280                        const struct stat_data *sd)
 281{
 282        return (istate->timestamp.sec &&
 283#ifdef USE_NSEC
 284                 /* nanosecond timestamped files can also be racy! */
 285                (istate->timestamp.sec < sd->sd_mtime.sec ||
 286                 (istate->timestamp.sec == sd->sd_mtime.sec &&
 287                  istate->timestamp.nsec <= sd->sd_mtime.nsec))
 288#else
 289                istate->timestamp.sec <= sd->sd_mtime.sec
 290#endif
 291                );
 292}
 293
 294static int is_racy_timestamp(const struct index_state *istate,
 295                             const struct cache_entry *ce)
 296{
 297        return (!S_ISGITLINK(ce->ce_mode) &&
 298                is_racy_stat(istate, &ce->ce_stat_data));
 299}
 300
 301int match_stat_data_racy(const struct index_state *istate,
 302                         const struct stat_data *sd, struct stat *st)
 303{
 304        if (is_racy_stat(istate, sd))
 305                return MTIME_CHANGED;
 306        return match_stat_data(sd, st);
 307}
 308
 309int ie_match_stat(const struct index_state *istate,
 310                  const struct cache_entry *ce, struct stat *st,
 311                  unsigned int options)
 312{
 313        unsigned int changed;
 314        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 315        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 316        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 317
 318        /*
 319         * If it's marked as always valid in the index, it's
 320         * valid whatever the checked-out copy says.
 321         *
 322         * skip-worktree has the same effect with higher precedence
 323         */
 324        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 325                return 0;
 326        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 327                return 0;
 328
 329        /*
 330         * Intent-to-add entries have not been added, so the index entry
 331         * by definition never matches what is in the work tree until it
 332         * actually gets added.
 333         */
 334        if (ce_intent_to_add(ce))
 335                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 336
 337        changed = ce_match_stat_basic(ce, st);
 338
 339        /*
 340         * Within 1 second of this sequence:
 341         *      echo xyzzy >file && git-update-index --add file
 342         * running this command:
 343         *      echo frotz >file
 344         * would give a falsely clean cache entry.  The mtime and
 345         * length match the cache, and other stat fields do not change.
 346         *
 347         * We could detect this at update-index time (the cache entry
 348         * being registered/updated records the same time as "now")
 349         * and delay the return from git-update-index, but that would
 350         * effectively mean we can make at most one commit per second,
 351         * which is not acceptable.  Instead, we check cache entries
 352         * whose mtime are the same as the index file timestamp more
 353         * carefully than others.
 354         */
 355        if (!changed && is_racy_timestamp(istate, ce)) {
 356                if (assume_racy_is_modified)
 357                        changed |= DATA_CHANGED;
 358                else
 359                        changed |= ce_modified_check_fs(ce, st);
 360        }
 361
 362        return changed;
 363}
 364
 365int ie_modified(const struct index_state *istate,
 366                const struct cache_entry *ce,
 367                struct stat *st, unsigned int options)
 368{
 369        int changed, changed_fs;
 370
 371        changed = ie_match_stat(istate, ce, st, options);
 372        if (!changed)
 373                return 0;
 374        /*
 375         * If the mode or type has changed, there's no point in trying
 376         * to refresh the entry - it's not going to match
 377         */
 378        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 379                return changed;
 380
 381        /*
 382         * Immediately after read-tree or update-index --cacheinfo,
 383         * the length field is zero, as we have never even read the
 384         * lstat(2) information once, and we cannot trust DATA_CHANGED
 385         * returned by ie_match_stat() which in turn was returned by
 386         * ce_match_stat_basic() to signal that the filesize of the
 387         * blob changed.  We have to actually go to the filesystem to
 388         * see if the contents match, and if so, should answer "unchanged".
 389         *
 390         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 391         * already has checked the actual HEAD from the filesystem in the
 392         * subproject.  If ie_match_stat() already said it is different,
 393         * then we know it is.
 394         */
 395        if ((changed & DATA_CHANGED) &&
 396            (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
 397                return changed;
 398
 399        changed_fs = ce_modified_check_fs(ce, st);
 400        if (changed_fs)
 401                return changed | changed_fs;
 402        return 0;
 403}
 404
 405int base_name_compare(const char *name1, int len1, int mode1,
 406                      const char *name2, int len2, int mode2)
 407{
 408        unsigned char c1, c2;
 409        int len = len1 < len2 ? len1 : len2;
 410        int cmp;
 411
 412        cmp = memcmp(name1, name2, len);
 413        if (cmp)
 414                return cmp;
 415        c1 = name1[len];
 416        c2 = name2[len];
 417        if (!c1 && S_ISDIR(mode1))
 418                c1 = '/';
 419        if (!c2 && S_ISDIR(mode2))
 420                c2 = '/';
 421        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 422}
 423
 424/*
 425 * df_name_compare() is identical to base_name_compare(), except it
 426 * compares conflicting directory/file entries as equal. Note that
 427 * while a directory name compares as equal to a regular file, they
 428 * then individually compare _differently_ to a filename that has
 429 * a dot after the basename (because '\0' < '.' < '/').
 430 *
 431 * This is used by routines that want to traverse the git namespace
 432 * but then handle conflicting entries together when possible.
 433 */
 434int df_name_compare(const char *name1, int len1, int mode1,
 435                    const char *name2, int len2, int mode2)
 436{
 437        int len = len1 < len2 ? len1 : len2, cmp;
 438        unsigned char c1, c2;
 439
 440        cmp = memcmp(name1, name2, len);
 441        if (cmp)
 442                return cmp;
 443        /* Directories and files compare equal (same length, same name) */
 444        if (len1 == len2)
 445                return 0;
 446        c1 = name1[len];
 447        if (!c1 && S_ISDIR(mode1))
 448                c1 = '/';
 449        c2 = name2[len];
 450        if (!c2 && S_ISDIR(mode2))
 451                c2 = '/';
 452        if (c1 == '/' && !c2)
 453                return 0;
 454        if (c2 == '/' && !c1)
 455                return 0;
 456        return c1 - c2;
 457}
 458
 459int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
 460{
 461        size_t min_len = (len1 < len2) ? len1 : len2;
 462        int cmp = memcmp(name1, name2, min_len);
 463        if (cmp)
 464                return cmp;
 465        if (len1 < len2)
 466                return -1;
 467        if (len1 > len2)
 468                return 1;
 469        return 0;
 470}
 471
 472int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 473{
 474        int cmp;
 475
 476        cmp = name_compare(name1, len1, name2, len2);
 477        if (cmp)
 478                return cmp;
 479
 480        if (stage1 < stage2)
 481                return -1;
 482        if (stage1 > stage2)
 483                return 1;
 484        return 0;
 485}
 486
 487static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 488{
 489        int first, last;
 490
 491        first = 0;
 492        last = istate->cache_nr;
 493        while (last > first) {
 494                int next = (last + first) >> 1;
 495                struct cache_entry *ce = istate->cache[next];
 496                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 497                if (!cmp)
 498                        return next;
 499                if (cmp < 0) {
 500                        last = next;
 501                        continue;
 502                }
 503                first = next+1;
 504        }
 505        return -first-1;
 506}
 507
 508int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 509{
 510        return index_name_stage_pos(istate, name, namelen, 0);
 511}
 512
 513/* Remove entry, return true if there are more entries to go.. */
 514int remove_index_entry_at(struct index_state *istate, int pos)
 515{
 516        struct cache_entry *ce = istate->cache[pos];
 517
 518        record_resolve_undo(istate, ce);
 519        remove_name_hash(istate, ce);
 520        save_or_free_index_entry(istate, ce);
 521        istate->cache_changed |= CE_ENTRY_REMOVED;
 522        istate->cache_nr--;
 523        if (pos >= istate->cache_nr)
 524                return 0;
 525        memmove(istate->cache + pos,
 526                istate->cache + pos + 1,
 527                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 528        return 1;
 529}
 530
 531/*
 532 * Remove all cache entries marked for removal, that is where
 533 * CE_REMOVE is set in ce_flags.  This is much more effective than
 534 * calling remove_index_entry_at() for each entry to be removed.
 535 */
 536void remove_marked_cache_entries(struct index_state *istate)
 537{
 538        struct cache_entry **ce_array = istate->cache;
 539        unsigned int i, j;
 540
 541        for (i = j = 0; i < istate->cache_nr; i++) {
 542                if (ce_array[i]->ce_flags & CE_REMOVE) {
 543                        remove_name_hash(istate, ce_array[i]);
 544                        save_or_free_index_entry(istate, ce_array[i]);
 545                }
 546                else
 547                        ce_array[j++] = ce_array[i];
 548        }
 549        if (j == istate->cache_nr)
 550                return;
 551        istate->cache_changed |= CE_ENTRY_REMOVED;
 552        istate->cache_nr = j;
 553}
 554
 555int remove_file_from_index(struct index_state *istate, const char *path)
 556{
 557        int pos = index_name_pos(istate, path, strlen(path));
 558        if (pos < 0)
 559                pos = -pos-1;
 560        cache_tree_invalidate_path(istate, path);
 561        untracked_cache_remove_from_index(istate, path);
 562        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 563                remove_index_entry_at(istate, pos);
 564        return 0;
 565}
 566
 567static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 568{
 569        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 570}
 571
 572static int index_name_pos_also_unmerged(struct index_state *istate,
 573        const char *path, int namelen)
 574{
 575        int pos = index_name_pos(istate, path, namelen);
 576        struct cache_entry *ce;
 577
 578        if (pos >= 0)
 579                return pos;
 580
 581        /* maybe unmerged? */
 582        pos = -1 - pos;
 583        if (pos >= istate->cache_nr ||
 584                        compare_name((ce = istate->cache[pos]), path, namelen))
 585                return -1;
 586
 587        /* order of preference: stage 2, 1, 3 */
 588        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 589                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 590                        !compare_name(ce, path, namelen))
 591                pos++;
 592        return pos;
 593}
 594
 595static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 596{
 597        int len = ce_namelen(ce);
 598        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 599}
 600
 601/*
 602 * If we add a filename that aliases in the cache, we will use the
 603 * name that we already have - but we don't want to update the same
 604 * alias twice, because that implies that there were actually two
 605 * different files with aliasing names!
 606 *
 607 * So we use the CE_ADDED flag to verify that the alias was an old
 608 * one before we accept it as
 609 */
 610static struct cache_entry *create_alias_ce(struct index_state *istate,
 611                                           struct cache_entry *ce,
 612                                           struct cache_entry *alias)
 613{
 614        int len;
 615        struct cache_entry *new;
 616
 617        if (alias->ce_flags & CE_ADDED)
 618                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 619
 620        /* Ok, create the new entry using the name of the existing alias */
 621        len = ce_namelen(alias);
 622        new = xcalloc(1, cache_entry_size(len));
 623        memcpy(new->name, alias->name, len);
 624        copy_cache_entry(new, ce);
 625        save_or_free_index_entry(istate, ce);
 626        return new;
 627}
 628
 629void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
 630{
 631        unsigned char sha1[20];
 632        if (write_sha1_file("", 0, blob_type, sha1))
 633                die("cannot create an empty blob in the object database");
 634        hashcpy(ce->oid.hash, sha1);
 635}
 636
 637int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 638{
 639        int size, namelen, was_same;
 640        mode_t st_mode = st->st_mode;
 641        struct cache_entry *ce, *alias;
 642        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 643        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 644        int pretend = flags & ADD_CACHE_PRETEND;
 645        int intent_only = flags & ADD_CACHE_INTENT;
 646        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 647                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 648
 649        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 650                return error("%s: can only add regular files, symbolic links or git-directories", path);
 651
 652        namelen = strlen(path);
 653        if (S_ISDIR(st_mode)) {
 654                while (namelen && path[namelen-1] == '/')
 655                        namelen--;
 656        }
 657        size = cache_entry_size(namelen);
 658        ce = xcalloc(1, size);
 659        memcpy(ce->name, path, namelen);
 660        ce->ce_namelen = namelen;
 661        if (!intent_only)
 662                fill_stat_cache_info(ce, st);
 663        else
 664                ce->ce_flags |= CE_INTENT_TO_ADD;
 665
 666
 667        if (trust_executable_bit && has_symlinks) {
 668                ce->ce_mode = create_ce_mode(st_mode);
 669        } else {
 670                /* If there is an existing entry, pick the mode bits and type
 671                 * from it, otherwise assume unexecutable regular file.
 672                 */
 673                struct cache_entry *ent;
 674                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 675
 676                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 677                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 678        }
 679
 680        /* When core.ignorecase=true, determine if a directory of the same name but differing
 681         * case already exists within the Git repository.  If it does, ensure the directory
 682         * case of the file being added to the repository matches (is folded into) the existing
 683         * entry's directory case.
 684         */
 685        if (ignore_case) {
 686                adjust_dirname_case(istate, ce->name);
 687        }
 688
 689        alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 690        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 691                /* Nothing changed, really */
 692                if (!S_ISGITLINK(alias->ce_mode))
 693                        ce_mark_uptodate(alias);
 694                alias->ce_flags |= CE_ADDED;
 695
 696                free(ce);
 697                return 0;
 698        }
 699        if (!intent_only) {
 700                if (index_path(ce->oid.hash, path, st, HASH_WRITE_OBJECT)) {
 701                        free(ce);
 702                        return error("unable to index file %s", path);
 703                }
 704        } else
 705                set_object_name_for_intent_to_add_entry(ce);
 706
 707        if (ignore_case && alias && different_name(ce, alias))
 708                ce = create_alias_ce(istate, ce, alias);
 709        ce->ce_flags |= CE_ADDED;
 710
 711        /* It was suspected to be racily clean, but it turns out to be Ok */
 712        was_same = (alias &&
 713                    !ce_stage(alias) &&
 714                    !oidcmp(&alias->oid, &ce->oid) &&
 715                    ce->ce_mode == alias->ce_mode);
 716
 717        if (pretend)
 718                free(ce);
 719        else if (add_index_entry(istate, ce, add_option)) {
 720                free(ce);
 721                return error("unable to add %s to index", path);
 722        }
 723        if (verbose && !was_same)
 724                printf("add '%s'\n", path);
 725        return 0;
 726}
 727
 728int add_file_to_index(struct index_state *istate, const char *path, int flags)
 729{
 730        struct stat st;
 731        if (lstat(path, &st))
 732                die_errno("unable to stat '%s'", path);
 733        return add_to_index(istate, path, &st, flags);
 734}
 735
 736struct cache_entry *make_cache_entry(unsigned int mode,
 737                const unsigned char *sha1, const char *path, int stage,
 738                unsigned int refresh_options)
 739{
 740        int size, len;
 741        struct cache_entry *ce, *ret;
 742
 743        if (!verify_path(path)) {
 744                error("Invalid path '%s'", path);
 745                return NULL;
 746        }
 747
 748        len = strlen(path);
 749        size = cache_entry_size(len);
 750        ce = xcalloc(1, size);
 751
 752        hashcpy(ce->oid.hash, sha1);
 753        memcpy(ce->name, path, len);
 754        ce->ce_flags = create_ce_flags(stage);
 755        ce->ce_namelen = len;
 756        ce->ce_mode = create_ce_mode(mode);
 757
 758        ret = refresh_cache_entry(ce, refresh_options);
 759        if (ret != ce)
 760                free(ce);
 761        return ret;
 762}
 763
 764/*
 765 * Chmod an index entry with either +x or -x.
 766 *
 767 * Returns -1 if the chmod for the particular cache entry failed (if it's
 768 * not a regular file), -2 if an invalid flip argument is passed in, 0
 769 * otherwise.
 770 */
 771int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
 772                      char flip)
 773{
 774        if (!S_ISREG(ce->ce_mode))
 775                return -1;
 776        switch (flip) {
 777        case '+':
 778                ce->ce_mode |= 0111;
 779                break;
 780        case '-':
 781                ce->ce_mode &= ~0111;
 782                break;
 783        default:
 784                return -2;
 785        }
 786        cache_tree_invalidate_path(istate, ce->name);
 787        ce->ce_flags |= CE_UPDATE_IN_BASE;
 788        istate->cache_changed |= CE_ENTRY_CHANGED;
 789
 790        return 0;
 791}
 792
 793int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
 794{
 795        int len = ce_namelen(a);
 796        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 797}
 798
 799/*
 800 * We fundamentally don't like some paths: we don't want
 801 * dot or dot-dot anywhere, and for obvious reasons don't
 802 * want to recurse into ".git" either.
 803 *
 804 * Also, we don't want double slashes or slashes at the
 805 * end that can make pathnames ambiguous.
 806 */
 807static int verify_dotfile(const char *rest)
 808{
 809        /*
 810         * The first character was '.', but that
 811         * has already been discarded, we now test
 812         * the rest.
 813         */
 814
 815        /* "." is not allowed */
 816        if (*rest == '\0' || is_dir_sep(*rest))
 817                return 0;
 818
 819        switch (*rest) {
 820        /*
 821         * ".git" followed by  NUL or slash is bad. This
 822         * shares the path end test with the ".." case.
 823         */
 824        case 'g':
 825        case 'G':
 826                if (rest[1] != 'i' && rest[1] != 'I')
 827                        break;
 828                if (rest[2] != 't' && rest[2] != 'T')
 829                        break;
 830                rest += 2;
 831        /* fallthrough */
 832        case '.':
 833                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 834                        return 0;
 835        }
 836        return 1;
 837}
 838
 839int verify_path(const char *path)
 840{
 841        char c;
 842
 843        if (has_dos_drive_prefix(path))
 844                return 0;
 845
 846        goto inside;
 847        for (;;) {
 848                if (!c)
 849                        return 1;
 850                if (is_dir_sep(c)) {
 851inside:
 852                        if (protect_hfs && is_hfs_dotgit(path))
 853                                return 0;
 854                        if (protect_ntfs && is_ntfs_dotgit(path))
 855                                return 0;
 856                        c = *path++;
 857                        if ((c == '.' && !verify_dotfile(path)) ||
 858                            is_dir_sep(c) || c == '\0')
 859                                return 0;
 860                }
 861                c = *path++;
 862        }
 863}
 864
 865/*
 866 * Do we have another file that has the beginning components being a
 867 * proper superset of the name we're trying to add?
 868 */
 869static int has_file_name(struct index_state *istate,
 870                         const struct cache_entry *ce, int pos, int ok_to_replace)
 871{
 872        int retval = 0;
 873        int len = ce_namelen(ce);
 874        int stage = ce_stage(ce);
 875        const char *name = ce->name;
 876
 877        while (pos < istate->cache_nr) {
 878                struct cache_entry *p = istate->cache[pos++];
 879
 880                if (len >= ce_namelen(p))
 881                        break;
 882                if (memcmp(name, p->name, len))
 883                        break;
 884                if (ce_stage(p) != stage)
 885                        continue;
 886                if (p->name[len] != '/')
 887                        continue;
 888                if (p->ce_flags & CE_REMOVE)
 889                        continue;
 890                retval = -1;
 891                if (!ok_to_replace)
 892                        break;
 893                remove_index_entry_at(istate, --pos);
 894        }
 895        return retval;
 896}
 897
 898/*
 899 * Do we have another file with a pathname that is a proper
 900 * subset of the name we're trying to add?
 901 */
 902static int has_dir_name(struct index_state *istate,
 903                        const struct cache_entry *ce, int pos, int ok_to_replace)
 904{
 905        int retval = 0;
 906        int stage = ce_stage(ce);
 907        const char *name = ce->name;
 908        const char *slash = name + ce_namelen(ce);
 909
 910        for (;;) {
 911                int len;
 912
 913                for (;;) {
 914                        if (*--slash == '/')
 915                                break;
 916                        if (slash <= ce->name)
 917                                return retval;
 918                }
 919                len = slash - name;
 920
 921                pos = index_name_stage_pos(istate, name, len, stage);
 922                if (pos >= 0) {
 923                        /*
 924                         * Found one, but not so fast.  This could
 925                         * be a marker that says "I was here, but
 926                         * I am being removed".  Such an entry is
 927                         * not a part of the resulting tree, and
 928                         * it is Ok to have a directory at the same
 929                         * path.
 930                         */
 931                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 932                                retval = -1;
 933                                if (!ok_to_replace)
 934                                        break;
 935                                remove_index_entry_at(istate, pos);
 936                                continue;
 937                        }
 938                }
 939                else
 940                        pos = -pos-1;
 941
 942                /*
 943                 * Trivial optimization: if we find an entry that
 944                 * already matches the sub-directory, then we know
 945                 * we're ok, and we can exit.
 946                 */
 947                while (pos < istate->cache_nr) {
 948                        struct cache_entry *p = istate->cache[pos];
 949                        if ((ce_namelen(p) <= len) ||
 950                            (p->name[len] != '/') ||
 951                            memcmp(p->name, name, len))
 952                                break; /* not our subdirectory */
 953                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 954                                /*
 955                                 * p is at the same stage as our entry, and
 956                                 * is a subdirectory of what we are looking
 957                                 * at, so we cannot have conflicts at our
 958                                 * level or anything shorter.
 959                                 */
 960                                return retval;
 961                        pos++;
 962                }
 963        }
 964        return retval;
 965}
 966
 967/* We may be in a situation where we already have path/file and path
 968 * is being added, or we already have path and path/file is being
 969 * added.  Either one would result in a nonsense tree that has path
 970 * twice when git-write-tree tries to write it out.  Prevent it.
 971 *
 972 * If ok-to-replace is specified, we remove the conflicting entries
 973 * from the cache so the caller should recompute the insert position.
 974 * When this happens, we return non-zero.
 975 */
 976static int check_file_directory_conflict(struct index_state *istate,
 977                                         const struct cache_entry *ce,
 978                                         int pos, int ok_to_replace)
 979{
 980        int retval;
 981
 982        /*
 983         * When ce is an "I am going away" entry, we allow it to be added
 984         */
 985        if (ce->ce_flags & CE_REMOVE)
 986                return 0;
 987
 988        /*
 989         * We check if the path is a sub-path of a subsequent pathname
 990         * first, since removing those will not change the position
 991         * in the array.
 992         */
 993        retval = has_file_name(istate, ce, pos, ok_to_replace);
 994
 995        /*
 996         * Then check if the path might have a clashing sub-directory
 997         * before it.
 998         */
 999        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1000}
1001
1002static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1003{
1004        int pos;
1005        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1006        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1007        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1008        int new_only = option & ADD_CACHE_NEW_ONLY;
1009
1010        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1011                cache_tree_invalidate_path(istate, ce->name);
1012        pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1013
1014        /* existing match? Just replace it. */
1015        if (pos >= 0) {
1016                if (!new_only)
1017                        replace_index_entry(istate, pos, ce);
1018                return 0;
1019        }
1020        pos = -pos-1;
1021
1022        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1023                untracked_cache_add_to_index(istate, ce->name);
1024
1025        /*
1026         * Inserting a merged entry ("stage 0") into the index
1027         * will always replace all non-merged entries..
1028         */
1029        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1030                while (ce_same_name(istate->cache[pos], ce)) {
1031                        ok_to_add = 1;
1032                        if (!remove_index_entry_at(istate, pos))
1033                                break;
1034                }
1035        }
1036
1037        if (!ok_to_add)
1038                return -1;
1039        if (!verify_path(ce->name))
1040                return error("Invalid path '%s'", ce->name);
1041
1042        if (!skip_df_check &&
1043            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1044                if (!ok_to_replace)
1045                        return error("'%s' appears as both a file and as a directory",
1046                                     ce->name);
1047                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1048                pos = -pos-1;
1049        }
1050        return pos + 1;
1051}
1052
1053int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1054{
1055        int pos;
1056
1057        if (option & ADD_CACHE_JUST_APPEND)
1058                pos = istate->cache_nr;
1059        else {
1060                int ret;
1061                ret = add_index_entry_with_check(istate, ce, option);
1062                if (ret <= 0)
1063                        return ret;
1064                pos = ret - 1;
1065        }
1066
1067        /* Make sure the array is big enough .. */
1068        ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1069
1070        /* Add it in.. */
1071        istate->cache_nr++;
1072        if (istate->cache_nr > pos + 1)
1073                memmove(istate->cache + pos + 1,
1074                        istate->cache + pos,
1075                        (istate->cache_nr - pos - 1) * sizeof(ce));
1076        set_index_entry(istate, pos, ce);
1077        istate->cache_changed |= CE_ENTRY_ADDED;
1078        return 0;
1079}
1080
1081/*
1082 * "refresh" does not calculate a new sha1 file or bring the
1083 * cache up-to-date for mode/content changes. But what it
1084 * _does_ do is to "re-match" the stat information of a file
1085 * with the cache, so that you can refresh the cache for a
1086 * file that hasn't been changed but where the stat entry is
1087 * out of date.
1088 *
1089 * For example, you'd want to do this after doing a "git-read-tree",
1090 * to link up the stat cache details with the proper files.
1091 */
1092static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1093                                             struct cache_entry *ce,
1094                                             unsigned int options, int *err,
1095                                             int *changed_ret)
1096{
1097        struct stat st;
1098        struct cache_entry *updated;
1099        int changed, size;
1100        int refresh = options & CE_MATCH_REFRESH;
1101        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1102        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1103        int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1104
1105        if (!refresh || ce_uptodate(ce))
1106                return ce;
1107
1108        /*
1109         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1110         * that the change to the work tree does not matter and told
1111         * us not to worry.
1112         */
1113        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1114                ce_mark_uptodate(ce);
1115                return ce;
1116        }
1117        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1118                ce_mark_uptodate(ce);
1119                return ce;
1120        }
1121
1122        if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1123                if (ignore_missing)
1124                        return ce;
1125                if (err)
1126                        *err = ENOENT;
1127                return NULL;
1128        }
1129
1130        if (lstat(ce->name, &st) < 0) {
1131                if (ignore_missing && errno == ENOENT)
1132                        return ce;
1133                if (err)
1134                        *err = errno;
1135                return NULL;
1136        }
1137
1138        changed = ie_match_stat(istate, ce, &st, options);
1139        if (changed_ret)
1140                *changed_ret = changed;
1141        if (!changed) {
1142                /*
1143                 * The path is unchanged.  If we were told to ignore
1144                 * valid bit, then we did the actual stat check and
1145                 * found that the entry is unmodified.  If the entry
1146                 * is not marked VALID, this is the place to mark it
1147                 * valid again, under "assume unchanged" mode.
1148                 */
1149                if (ignore_valid && assume_unchanged &&
1150                    !(ce->ce_flags & CE_VALID))
1151                        ; /* mark this one VALID again */
1152                else {
1153                        /*
1154                         * We do not mark the index itself "modified"
1155                         * because CE_UPTODATE flag is in-core only;
1156                         * we are not going to write this change out.
1157                         */
1158                        if (!S_ISGITLINK(ce->ce_mode))
1159                                ce_mark_uptodate(ce);
1160                        return ce;
1161                }
1162        }
1163
1164        if (ie_modified(istate, ce, &st, options)) {
1165                if (err)
1166                        *err = EINVAL;
1167                return NULL;
1168        }
1169
1170        size = ce_size(ce);
1171        updated = xmalloc(size);
1172        memcpy(updated, ce, size);
1173        fill_stat_cache_info(updated, &st);
1174        /*
1175         * If ignore_valid is not set, we should leave CE_VALID bit
1176         * alone.  Otherwise, paths marked with --no-assume-unchanged
1177         * (i.e. things to be edited) will reacquire CE_VALID bit
1178         * automatically, which is not really what we want.
1179         */
1180        if (!ignore_valid && assume_unchanged &&
1181            !(ce->ce_flags & CE_VALID))
1182                updated->ce_flags &= ~CE_VALID;
1183
1184        /* istate->cache_changed is updated in the caller */
1185        return updated;
1186}
1187
1188static void show_file(const char * fmt, const char * name, int in_porcelain,
1189                      int * first, const char *header_msg)
1190{
1191        if (in_porcelain && *first && header_msg) {
1192                printf("%s\n", header_msg);
1193                *first = 0;
1194        }
1195        printf(fmt, name);
1196}
1197
1198int refresh_index(struct index_state *istate, unsigned int flags,
1199                  const struct pathspec *pathspec,
1200                  char *seen, const char *header_msg)
1201{
1202        int i;
1203        int has_errors = 0;
1204        int really = (flags & REFRESH_REALLY) != 0;
1205        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1206        int quiet = (flags & REFRESH_QUIET) != 0;
1207        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1208        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1209        int first = 1;
1210        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1211        unsigned int options = (CE_MATCH_REFRESH |
1212                                (really ? CE_MATCH_IGNORE_VALID : 0) |
1213                                (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1214        const char *modified_fmt;
1215        const char *deleted_fmt;
1216        const char *typechange_fmt;
1217        const char *added_fmt;
1218        const char *unmerged_fmt;
1219
1220        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1221        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1222        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1223        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1224        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1225        for (i = 0; i < istate->cache_nr; i++) {
1226                struct cache_entry *ce, *new;
1227                int cache_errno = 0;
1228                int changed = 0;
1229                int filtered = 0;
1230
1231                ce = istate->cache[i];
1232                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1233                        continue;
1234
1235                if (pathspec && !ce_path_match(ce, pathspec, seen))
1236                        filtered = 1;
1237
1238                if (ce_stage(ce)) {
1239                        while ((i < istate->cache_nr) &&
1240                               ! strcmp(istate->cache[i]->name, ce->name))
1241                                i++;
1242                        i--;
1243                        if (allow_unmerged)
1244                                continue;
1245                        if (!filtered)
1246                                show_file(unmerged_fmt, ce->name, in_porcelain,
1247                                          &first, header_msg);
1248                        has_errors = 1;
1249                        continue;
1250                }
1251
1252                if (filtered)
1253                        continue;
1254
1255                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1256                if (new == ce)
1257                        continue;
1258                if (!new) {
1259                        const char *fmt;
1260
1261                        if (really && cache_errno == EINVAL) {
1262                                /* If we are doing --really-refresh that
1263                                 * means the index is not valid anymore.
1264                                 */
1265                                ce->ce_flags &= ~CE_VALID;
1266                                ce->ce_flags |= CE_UPDATE_IN_BASE;
1267                                istate->cache_changed |= CE_ENTRY_CHANGED;
1268                        }
1269                        if (quiet)
1270                                continue;
1271
1272                        if (cache_errno == ENOENT)
1273                                fmt = deleted_fmt;
1274                        else if (ce_intent_to_add(ce))
1275                                fmt = added_fmt; /* must be before other checks */
1276                        else if (changed & TYPE_CHANGED)
1277                                fmt = typechange_fmt;
1278                        else
1279                                fmt = modified_fmt;
1280                        show_file(fmt,
1281                                  ce->name, in_porcelain, &first, header_msg);
1282                        has_errors = 1;
1283                        continue;
1284                }
1285
1286                replace_index_entry(istate, i, new);
1287        }
1288        return has_errors;
1289}
1290
1291struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1292                                               unsigned int options)
1293{
1294        return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1295}
1296
1297
1298/*****************************************************************
1299 * Index File I/O
1300 *****************************************************************/
1301
1302#define INDEX_FORMAT_DEFAULT 3
1303
1304static unsigned int get_index_format_default(void)
1305{
1306        char *envversion = getenv("GIT_INDEX_VERSION");
1307        char *endp;
1308        int value;
1309        unsigned int version = INDEX_FORMAT_DEFAULT;
1310
1311        if (!envversion) {
1312                if (!git_config_get_int("index.version", &value))
1313                        version = value;
1314                if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1315                        warning(_("index.version set, but the value is invalid.\n"
1316                                  "Using version %i"), INDEX_FORMAT_DEFAULT);
1317                        return INDEX_FORMAT_DEFAULT;
1318                }
1319                return version;
1320        }
1321
1322        version = strtoul(envversion, &endp, 10);
1323        if (*endp ||
1324            version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1325                warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1326                          "Using version %i"), INDEX_FORMAT_DEFAULT);
1327                version = INDEX_FORMAT_DEFAULT;
1328        }
1329        return version;
1330}
1331
1332/*
1333 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1334 * Again - this is just a (very strong in practice) heuristic that
1335 * the inode hasn't changed.
1336 *
1337 * We save the fields in big-endian order to allow using the
1338 * index file over NFS transparently.
1339 */
1340struct ondisk_cache_entry {
1341        struct cache_time ctime;
1342        struct cache_time mtime;
1343        uint32_t dev;
1344        uint32_t ino;
1345        uint32_t mode;
1346        uint32_t uid;
1347        uint32_t gid;
1348        uint32_t size;
1349        unsigned char sha1[20];
1350        uint16_t flags;
1351        char name[FLEX_ARRAY]; /* more */
1352};
1353
1354/*
1355 * This struct is used when CE_EXTENDED bit is 1
1356 * The struct must match ondisk_cache_entry exactly from
1357 * ctime till flags
1358 */
1359struct ondisk_cache_entry_extended {
1360        struct cache_time ctime;
1361        struct cache_time mtime;
1362        uint32_t dev;
1363        uint32_t ino;
1364        uint32_t mode;
1365        uint32_t uid;
1366        uint32_t gid;
1367        uint32_t size;
1368        unsigned char sha1[20];
1369        uint16_t flags;
1370        uint16_t flags2;
1371        char name[FLEX_ARRAY]; /* more */
1372};
1373
1374/* These are only used for v3 or lower */
1375#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1376#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1377#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1378#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1379                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1380                            ondisk_cache_entry_size(ce_namelen(ce)))
1381
1382static int verify_hdr(struct cache_header *hdr, unsigned long size)
1383{
1384        git_SHA_CTX c;
1385        unsigned char sha1[20];
1386        int hdr_version;
1387
1388        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1389                return error("bad signature");
1390        hdr_version = ntohl(hdr->hdr_version);
1391        if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1392                return error("bad index version %d", hdr_version);
1393        git_SHA1_Init(&c);
1394        git_SHA1_Update(&c, hdr, size - 20);
1395        git_SHA1_Final(sha1, &c);
1396        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1397                return error("bad index file sha1 signature");
1398        return 0;
1399}
1400
1401static int read_index_extension(struct index_state *istate,
1402                                const char *ext, void *data, unsigned long sz)
1403{
1404        switch (CACHE_EXT(ext)) {
1405        case CACHE_EXT_TREE:
1406                istate->cache_tree = cache_tree_read(data, sz);
1407                break;
1408        case CACHE_EXT_RESOLVE_UNDO:
1409                istate->resolve_undo = resolve_undo_read(data, sz);
1410                break;
1411        case CACHE_EXT_LINK:
1412                if (read_link_extension(istate, data, sz))
1413                        return -1;
1414                break;
1415        case CACHE_EXT_UNTRACKED:
1416                istate->untracked = read_untracked_extension(data, sz);
1417                break;
1418        default:
1419                if (*ext < 'A' || 'Z' < *ext)
1420                        return error("index uses %.4s extension, which we do not understand",
1421                                     ext);
1422                fprintf(stderr, "ignoring %.4s extension\n", ext);
1423                break;
1424        }
1425        return 0;
1426}
1427
1428int hold_locked_index(struct lock_file *lk, int die_on_error)
1429{
1430        return hold_lock_file_for_update(lk, get_index_file(),
1431                                         die_on_error
1432                                         ? LOCK_DIE_ON_ERROR
1433                                         : 0);
1434}
1435
1436int read_index(struct index_state *istate)
1437{
1438        return read_index_from(istate, get_index_file());
1439}
1440
1441static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1442                                                   unsigned int flags,
1443                                                   const char *name,
1444                                                   size_t len)
1445{
1446        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1447
1448        ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1449        ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1450        ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1451        ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1452        ce->ce_stat_data.sd_dev   = get_be32(&ondisk->dev);
1453        ce->ce_stat_data.sd_ino   = get_be32(&ondisk->ino);
1454        ce->ce_mode  = get_be32(&ondisk->mode);
1455        ce->ce_stat_data.sd_uid   = get_be32(&ondisk->uid);
1456        ce->ce_stat_data.sd_gid   = get_be32(&ondisk->gid);
1457        ce->ce_stat_data.sd_size  = get_be32(&ondisk->size);
1458        ce->ce_flags = flags & ~CE_NAMEMASK;
1459        ce->ce_namelen = len;
1460        ce->index = 0;
1461        hashcpy(ce->oid.hash, ondisk->sha1);
1462        memcpy(ce->name, name, len);
1463        ce->name[len] = '\0';
1464        return ce;
1465}
1466
1467/*
1468 * Adjacent cache entries tend to share the leading paths, so it makes
1469 * sense to only store the differences in later entries.  In the v4
1470 * on-disk format of the index, each on-disk cache entry stores the
1471 * number of bytes to be stripped from the end of the previous name,
1472 * and the bytes to append to the result, to come up with its name.
1473 */
1474static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1475{
1476        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1477        size_t len = decode_varint(&cp);
1478
1479        if (name->len < len)
1480                die("malformed name field in the index");
1481        strbuf_remove(name, name->len - len, len);
1482        for (ep = cp; *ep; ep++)
1483                ; /* find the end */
1484        strbuf_add(name, cp, ep - cp);
1485        return (const char *)ep + 1 - cp_;
1486}
1487
1488static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1489                                            unsigned long *ent_size,
1490                                            struct strbuf *previous_name)
1491{
1492        struct cache_entry *ce;
1493        size_t len;
1494        const char *name;
1495        unsigned int flags;
1496
1497        /* On-disk flags are just 16 bits */
1498        flags = get_be16(&ondisk->flags);
1499        len = flags & CE_NAMEMASK;
1500
1501        if (flags & CE_EXTENDED) {
1502                struct ondisk_cache_entry_extended *ondisk2;
1503                int extended_flags;
1504                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1505                extended_flags = get_be16(&ondisk2->flags2) << 16;
1506                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1507                if (extended_flags & ~CE_EXTENDED_FLAGS)
1508                        die("Unknown index entry format %08x", extended_flags);
1509                flags |= extended_flags;
1510                name = ondisk2->name;
1511        }
1512        else
1513                name = ondisk->name;
1514
1515        if (!previous_name) {
1516                /* v3 and earlier */
1517                if (len == CE_NAMEMASK)
1518                        len = strlen(name);
1519                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1520
1521                *ent_size = ondisk_ce_size(ce);
1522        } else {
1523                unsigned long consumed;
1524                consumed = expand_name_field(previous_name, name);
1525                ce = cache_entry_from_ondisk(ondisk, flags,
1526                                             previous_name->buf,
1527                                             previous_name->len);
1528
1529                *ent_size = (name - ((char *)ondisk)) + consumed;
1530        }
1531        return ce;
1532}
1533
1534static void check_ce_order(struct index_state *istate)
1535{
1536        unsigned int i;
1537
1538        for (i = 1; i < istate->cache_nr; i++) {
1539                struct cache_entry *ce = istate->cache[i - 1];
1540                struct cache_entry *next_ce = istate->cache[i];
1541                int name_compare = strcmp(ce->name, next_ce->name);
1542
1543                if (0 < name_compare)
1544                        die("unordered stage entries in index");
1545                if (!name_compare) {
1546                        if (!ce_stage(ce))
1547                                die("multiple stage entries for merged file '%s'",
1548                                    ce->name);
1549                        if (ce_stage(ce) > ce_stage(next_ce))
1550                                die("unordered stage entries for '%s'",
1551                                    ce->name);
1552                }
1553        }
1554}
1555
1556static void tweak_untracked_cache(struct index_state *istate)
1557{
1558        switch (git_config_get_untracked_cache()) {
1559        case -1: /* keep: do nothing */
1560                break;
1561        case 0: /* false */
1562                remove_untracked_cache(istate);
1563                break;
1564        case 1: /* true */
1565                add_untracked_cache(istate);
1566                break;
1567        default: /* unknown value: do nothing */
1568                break;
1569        }
1570}
1571
1572static void post_read_index_from(struct index_state *istate)
1573{
1574        check_ce_order(istate);
1575        tweak_untracked_cache(istate);
1576}
1577
1578/* remember to discard_cache() before reading a different cache! */
1579int do_read_index(struct index_state *istate, const char *path, int must_exist)
1580{
1581        int fd, i;
1582        struct stat st;
1583        unsigned long src_offset;
1584        struct cache_header *hdr;
1585        void *mmap;
1586        size_t mmap_size;
1587        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1588
1589        if (istate->initialized)
1590                return istate->cache_nr;
1591
1592        istate->timestamp.sec = 0;
1593        istate->timestamp.nsec = 0;
1594        fd = open(path, O_RDONLY);
1595        if (fd < 0) {
1596                if (!must_exist && errno == ENOENT)
1597                        return 0;
1598                die_errno("%s: index file open failed", path);
1599        }
1600
1601        if (fstat(fd, &st))
1602                die_errno("cannot stat the open index");
1603
1604        mmap_size = xsize_t(st.st_size);
1605        if (mmap_size < sizeof(struct cache_header) + 20)
1606                die("index file smaller than expected");
1607
1608        mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1609        if (mmap == MAP_FAILED)
1610                die_errno("unable to map index file");
1611        close(fd);
1612
1613        hdr = mmap;
1614        if (verify_hdr(hdr, mmap_size) < 0)
1615                goto unmap;
1616
1617        hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1618        istate->version = ntohl(hdr->hdr_version);
1619        istate->cache_nr = ntohl(hdr->hdr_entries);
1620        istate->cache_alloc = alloc_nr(istate->cache_nr);
1621        istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1622        istate->initialized = 1;
1623
1624        if (istate->version == 4)
1625                previous_name = &previous_name_buf;
1626        else
1627                previous_name = NULL;
1628
1629        src_offset = sizeof(*hdr);
1630        for (i = 0; i < istate->cache_nr; i++) {
1631                struct ondisk_cache_entry *disk_ce;
1632                struct cache_entry *ce;
1633                unsigned long consumed;
1634
1635                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1636                ce = create_from_disk(disk_ce, &consumed, previous_name);
1637                set_index_entry(istate, i, ce);
1638
1639                src_offset += consumed;
1640        }
1641        strbuf_release(&previous_name_buf);
1642        istate->timestamp.sec = st.st_mtime;
1643        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1644
1645        while (src_offset <= mmap_size - 20 - 8) {
1646                /* After an array of active_nr index entries,
1647                 * there can be arbitrary number of extended
1648                 * sections, each of which is prefixed with
1649                 * extension name (4-byte) and section length
1650                 * in 4-byte network byte order.
1651                 */
1652                uint32_t extsize;
1653                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1654                extsize = ntohl(extsize);
1655                if (read_index_extension(istate,
1656                                         (const char *) mmap + src_offset,
1657                                         (char *) mmap + src_offset + 8,
1658                                         extsize) < 0)
1659                        goto unmap;
1660                src_offset += 8;
1661                src_offset += extsize;
1662        }
1663        munmap(mmap, mmap_size);
1664        return istate->cache_nr;
1665
1666unmap:
1667        munmap(mmap, mmap_size);
1668        die("index file corrupt");
1669}
1670
1671int read_index_from(struct index_state *istate, const char *path)
1672{
1673        struct split_index *split_index;
1674        int ret;
1675
1676        /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1677        if (istate->initialized)
1678                return istate->cache_nr;
1679
1680        ret = do_read_index(istate, path, 0);
1681
1682        split_index = istate->split_index;
1683        if (!split_index || is_null_sha1(split_index->base_sha1)) {
1684                post_read_index_from(istate);
1685                return ret;
1686        }
1687
1688        if (split_index->base)
1689                discard_index(split_index->base);
1690        else
1691                split_index->base = xcalloc(1, sizeof(*split_index->base));
1692        ret = do_read_index(split_index->base,
1693                            git_path("sharedindex.%s",
1694                                     sha1_to_hex(split_index->base_sha1)), 1);
1695        if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1696                die("broken index, expect %s in %s, got %s",
1697                    sha1_to_hex(split_index->base_sha1),
1698                    git_path("sharedindex.%s",
1699                             sha1_to_hex(split_index->base_sha1)),
1700                    sha1_to_hex(split_index->base->sha1));
1701        merge_base_index(istate);
1702        post_read_index_from(istate);
1703        return ret;
1704}
1705
1706int is_index_unborn(struct index_state *istate)
1707{
1708        return (!istate->cache_nr && !istate->timestamp.sec);
1709}
1710
1711int discard_index(struct index_state *istate)
1712{
1713        int i;
1714
1715        for (i = 0; i < istate->cache_nr; i++) {
1716                if (istate->cache[i]->index &&
1717                    istate->split_index &&
1718                    istate->split_index->base &&
1719                    istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1720                    istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1721                        continue;
1722                free(istate->cache[i]);
1723        }
1724        resolve_undo_clear_index(istate);
1725        istate->cache_nr = 0;
1726        istate->cache_changed = 0;
1727        istate->timestamp.sec = 0;
1728        istate->timestamp.nsec = 0;
1729        free_name_hash(istate);
1730        cache_tree_free(&(istate->cache_tree));
1731        istate->initialized = 0;
1732        free(istate->cache);
1733        istate->cache = NULL;
1734        istate->cache_alloc = 0;
1735        discard_split_index(istate);
1736        free_untracked_cache(istate->untracked);
1737        istate->untracked = NULL;
1738        return 0;
1739}
1740
1741int unmerged_index(const struct index_state *istate)
1742{
1743        int i;
1744        for (i = 0; i < istate->cache_nr; i++) {
1745                if (ce_stage(istate->cache[i]))
1746                        return 1;
1747        }
1748        return 0;
1749}
1750
1751#define WRITE_BUFFER_SIZE 8192
1752static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1753static unsigned long write_buffer_len;
1754
1755static int ce_write_flush(git_SHA_CTX *context, int fd)
1756{
1757        unsigned int buffered = write_buffer_len;
1758        if (buffered) {
1759                git_SHA1_Update(context, write_buffer, buffered);
1760                if (write_in_full(fd, write_buffer, buffered) != buffered)
1761                        return -1;
1762                write_buffer_len = 0;
1763        }
1764        return 0;
1765}
1766
1767static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1768{
1769        while (len) {
1770                unsigned int buffered = write_buffer_len;
1771                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1772                if (partial > len)
1773                        partial = len;
1774                memcpy(write_buffer + buffered, data, partial);
1775                buffered += partial;
1776                if (buffered == WRITE_BUFFER_SIZE) {
1777                        write_buffer_len = buffered;
1778                        if (ce_write_flush(context, fd))
1779                                return -1;
1780                        buffered = 0;
1781                }
1782                write_buffer_len = buffered;
1783                len -= partial;
1784                data = (char *) data + partial;
1785        }
1786        return 0;
1787}
1788
1789static int write_index_ext_header(git_SHA_CTX *context, int fd,
1790                                  unsigned int ext, unsigned int sz)
1791{
1792        ext = htonl(ext);
1793        sz = htonl(sz);
1794        return ((ce_write(context, fd, &ext, 4) < 0) ||
1795                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1796}
1797
1798static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1799{
1800        unsigned int left = write_buffer_len;
1801
1802        if (left) {
1803                write_buffer_len = 0;
1804                git_SHA1_Update(context, write_buffer, left);
1805        }
1806
1807        /* Flush first if not enough space for SHA1 signature */
1808        if (left + 20 > WRITE_BUFFER_SIZE) {
1809                if (write_in_full(fd, write_buffer, left) != left)
1810                        return -1;
1811                left = 0;
1812        }
1813
1814        /* Append the SHA1 signature at the end */
1815        git_SHA1_Final(write_buffer + left, context);
1816        hashcpy(sha1, write_buffer + left);
1817        left += 20;
1818        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1819}
1820
1821static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1822{
1823        /*
1824         * The only thing we care about in this function is to smudge the
1825         * falsely clean entry due to touch-update-touch race, so we leave
1826         * everything else as they are.  We are called for entries whose
1827         * ce_stat_data.sd_mtime match the index file mtime.
1828         *
1829         * Note that this actually does not do much for gitlinks, for
1830         * which ce_match_stat_basic() always goes to the actual
1831         * contents.  The caller checks with is_racy_timestamp() which
1832         * always says "no" for gitlinks, so we are not called for them ;-)
1833         */
1834        struct stat st;
1835
1836        if (lstat(ce->name, &st) < 0)
1837                return;
1838        if (ce_match_stat_basic(ce, &st))
1839                return;
1840        if (ce_modified_check_fs(ce, &st)) {
1841                /* This is "racily clean"; smudge it.  Note that this
1842                 * is a tricky code.  At first glance, it may appear
1843                 * that it can break with this sequence:
1844                 *
1845                 * $ echo xyzzy >frotz
1846                 * $ git-update-index --add frotz
1847                 * $ : >frotz
1848                 * $ sleep 3
1849                 * $ echo filfre >nitfol
1850                 * $ git-update-index --add nitfol
1851                 *
1852                 * but it does not.  When the second update-index runs,
1853                 * it notices that the entry "frotz" has the same timestamp
1854                 * as index, and if we were to smudge it by resetting its
1855                 * size to zero here, then the object name recorded
1856                 * in index is the 6-byte file but the cached stat information
1857                 * becomes zero --- which would then match what we would
1858                 * obtain from the filesystem next time we stat("frotz").
1859                 *
1860                 * However, the second update-index, before calling
1861                 * this function, notices that the cached size is 6
1862                 * bytes and what is on the filesystem is an empty
1863                 * file, and never calls us, so the cached size information
1864                 * for "frotz" stays 6 which does not match the filesystem.
1865                 */
1866                ce->ce_stat_data.sd_size = 0;
1867        }
1868}
1869
1870/* Copy miscellaneous fields but not the name */
1871static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1872                                       struct cache_entry *ce)
1873{
1874        short flags;
1875
1876        ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1877        ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1878        ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1879        ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1880        ondisk->dev  = htonl(ce->ce_stat_data.sd_dev);
1881        ondisk->ino  = htonl(ce->ce_stat_data.sd_ino);
1882        ondisk->mode = htonl(ce->ce_mode);
1883        ondisk->uid  = htonl(ce->ce_stat_data.sd_uid);
1884        ondisk->gid  = htonl(ce->ce_stat_data.sd_gid);
1885        ondisk->size = htonl(ce->ce_stat_data.sd_size);
1886        hashcpy(ondisk->sha1, ce->oid.hash);
1887
1888        flags = ce->ce_flags & ~CE_NAMEMASK;
1889        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1890        ondisk->flags = htons(flags);
1891        if (ce->ce_flags & CE_EXTENDED) {
1892                struct ondisk_cache_entry_extended *ondisk2;
1893                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1894                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1895                return ondisk2->name;
1896        }
1897        else {
1898                return ondisk->name;
1899        }
1900}
1901
1902static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1903                          struct strbuf *previous_name)
1904{
1905        int size;
1906        struct ondisk_cache_entry *ondisk;
1907        int saved_namelen = saved_namelen; /* compiler workaround */
1908        char *name;
1909        int result;
1910
1911        if (ce->ce_flags & CE_STRIP_NAME) {
1912                saved_namelen = ce_namelen(ce);
1913                ce->ce_namelen = 0;
1914        }
1915
1916        if (!previous_name) {
1917                size = ondisk_ce_size(ce);
1918                ondisk = xcalloc(1, size);
1919                name = copy_cache_entry_to_ondisk(ondisk, ce);
1920                memcpy(name, ce->name, ce_namelen(ce));
1921        } else {
1922                int common, to_remove, prefix_size;
1923                unsigned char to_remove_vi[16];
1924                for (common = 0;
1925                     (ce->name[common] &&
1926                      common < previous_name->len &&
1927                      ce->name[common] == previous_name->buf[common]);
1928                     common++)
1929                        ; /* still matching */
1930                to_remove = previous_name->len - common;
1931                prefix_size = encode_varint(to_remove, to_remove_vi);
1932
1933                if (ce->ce_flags & CE_EXTENDED)
1934                        size = offsetof(struct ondisk_cache_entry_extended, name);
1935                else
1936                        size = offsetof(struct ondisk_cache_entry, name);
1937                size += prefix_size + (ce_namelen(ce) - common + 1);
1938
1939                ondisk = xcalloc(1, size);
1940                name = copy_cache_entry_to_ondisk(ondisk, ce);
1941                memcpy(name, to_remove_vi, prefix_size);
1942                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1943
1944                strbuf_splice(previous_name, common, to_remove,
1945                              ce->name + common, ce_namelen(ce) - common);
1946        }
1947        if (ce->ce_flags & CE_STRIP_NAME) {
1948                ce->ce_namelen = saved_namelen;
1949                ce->ce_flags &= ~CE_STRIP_NAME;
1950        }
1951
1952        result = ce_write(c, fd, ondisk, size);
1953        free(ondisk);
1954        return result;
1955}
1956
1957/*
1958 * This function verifies if index_state has the correct sha1 of the
1959 * index file.  Don't die if we have any other failure, just return 0.
1960 */
1961static int verify_index_from(const struct index_state *istate, const char *path)
1962{
1963        int fd;
1964        ssize_t n;
1965        struct stat st;
1966        unsigned char sha1[20];
1967
1968        if (!istate->initialized)
1969                return 0;
1970
1971        fd = open(path, O_RDONLY);
1972        if (fd < 0)
1973                return 0;
1974
1975        if (fstat(fd, &st))
1976                goto out;
1977
1978        if (st.st_size < sizeof(struct cache_header) + 20)
1979                goto out;
1980
1981        n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1982        if (n != 20)
1983                goto out;
1984
1985        if (hashcmp(istate->sha1, sha1))
1986                goto out;
1987
1988        close(fd);
1989        return 1;
1990
1991out:
1992        close(fd);
1993        return 0;
1994}
1995
1996static int verify_index(const struct index_state *istate)
1997{
1998        return verify_index_from(istate, get_index_file());
1999}
2000
2001static int has_racy_timestamp(struct index_state *istate)
2002{
2003        int entries = istate->cache_nr;
2004        int i;
2005
2006        for (i = 0; i < entries; i++) {
2007                struct cache_entry *ce = istate->cache[i];
2008                if (is_racy_timestamp(istate, ce))
2009                        return 1;
2010        }
2011        return 0;
2012}
2013
2014/*
2015 * Opportunistically update the index but do not complain if we can't
2016 */
2017void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2018{
2019        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2020            verify_index(istate) &&
2021            write_locked_index(istate, lockfile, COMMIT_LOCK))
2022                rollback_lock_file(lockfile);
2023}
2024
2025static int do_write_index(struct index_state *istate, int newfd,
2026                          int strip_extensions)
2027{
2028        git_SHA_CTX c;
2029        struct cache_header hdr;
2030        int i, err, removed, extended, hdr_version;
2031        struct cache_entry **cache = istate->cache;
2032        int entries = istate->cache_nr;
2033        struct stat st;
2034        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2035
2036        for (i = removed = extended = 0; i < entries; i++) {
2037                if (cache[i]->ce_flags & CE_REMOVE)
2038                        removed++;
2039
2040                /* reduce extended entries if possible */
2041                cache[i]->ce_flags &= ~CE_EXTENDED;
2042                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2043                        extended++;
2044                        cache[i]->ce_flags |= CE_EXTENDED;
2045                }
2046        }
2047
2048        if (!istate->version) {
2049                istate->version = get_index_format_default();
2050                if (getenv("GIT_TEST_SPLIT_INDEX"))
2051                        init_split_index(istate);
2052        }
2053
2054        /* demote version 3 to version 2 when the latter suffices */
2055        if (istate->version == 3 || istate->version == 2)
2056                istate->version = extended ? 3 : 2;
2057
2058        hdr_version = istate->version;
2059
2060        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2061        hdr.hdr_version = htonl(hdr_version);
2062        hdr.hdr_entries = htonl(entries - removed);
2063
2064        git_SHA1_Init(&c);
2065        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2066                return -1;
2067
2068        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2069        for (i = 0; i < entries; i++) {
2070                struct cache_entry *ce = cache[i];
2071                if (ce->ce_flags & CE_REMOVE)
2072                        continue;
2073                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2074                        ce_smudge_racily_clean_entry(ce);
2075                if (is_null_oid(&ce->oid)) {
2076                        static const char msg[] = "cache entry has null sha1: %s";
2077                        static int allow = -1;
2078
2079                        if (allow < 0)
2080                                allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2081                        if (allow)
2082                                warning(msg, ce->name);
2083                        else
2084                                return error(msg, ce->name);
2085                }
2086                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2087                        return -1;
2088        }
2089        strbuf_release(&previous_name_buf);
2090
2091        /* Write extension data here */
2092        if (!strip_extensions && istate->split_index) {
2093                struct strbuf sb = STRBUF_INIT;
2094
2095                err = write_link_extension(&sb, istate) < 0 ||
2096                        write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2097                                               sb.len) < 0 ||
2098                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2099                strbuf_release(&sb);
2100                if (err)
2101                        return -1;
2102        }
2103        if (!strip_extensions && istate->cache_tree) {
2104                struct strbuf sb = STRBUF_INIT;
2105
2106                cache_tree_write(&sb, istate->cache_tree);
2107                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2108                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2109                strbuf_release(&sb);
2110                if (err)
2111                        return -1;
2112        }
2113        if (!strip_extensions && istate->resolve_undo) {
2114                struct strbuf sb = STRBUF_INIT;
2115
2116                resolve_undo_write(&sb, istate->resolve_undo);
2117                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2118                                             sb.len) < 0
2119                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2120                strbuf_release(&sb);
2121                if (err)
2122                        return -1;
2123        }
2124        if (!strip_extensions && istate->untracked) {
2125                struct strbuf sb = STRBUF_INIT;
2126
2127                write_untracked_extension(&sb, istate->untracked);
2128                err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2129                                             sb.len) < 0 ||
2130                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2131                strbuf_release(&sb);
2132                if (err)
2133                        return -1;
2134        }
2135
2136        if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2137                return -1;
2138        istate->timestamp.sec = (unsigned int)st.st_mtime;
2139        istate->timestamp.nsec = ST_MTIME_NSEC(st);
2140        return 0;
2141}
2142
2143void set_alternate_index_output(const char *name)
2144{
2145        alternate_index_output = name;
2146}
2147
2148static int commit_locked_index(struct lock_file *lk)
2149{
2150        if (alternate_index_output)
2151                return commit_lock_file_to(lk, alternate_index_output);
2152        else
2153                return commit_lock_file(lk);
2154}
2155
2156static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2157                                 unsigned flags)
2158{
2159        int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
2160        if (ret)
2161                return ret;
2162        assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2163               (COMMIT_LOCK | CLOSE_LOCK));
2164        if (flags & COMMIT_LOCK)
2165                return commit_locked_index(lock);
2166        else if (flags & CLOSE_LOCK)
2167                return close_lock_file(lock);
2168        else
2169                return ret;
2170}
2171
2172static int write_split_index(struct index_state *istate,
2173                             struct lock_file *lock,
2174                             unsigned flags)
2175{
2176        int ret;
2177        prepare_to_write_split_index(istate);
2178        ret = do_write_locked_index(istate, lock, flags);
2179        finish_writing_split_index(istate);
2180        return ret;
2181}
2182
2183static struct tempfile temporary_sharedindex;
2184
2185static int write_shared_index(struct index_state *istate,
2186                              struct lock_file *lock, unsigned flags)
2187{
2188        struct split_index *si = istate->split_index;
2189        int fd, ret;
2190
2191        fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2192        if (fd < 0) {
2193                hashclr(si->base_sha1);
2194                return do_write_locked_index(istate, lock, flags);
2195        }
2196        move_cache_to_base_index(istate);
2197        ret = do_write_index(si->base, fd, 1);
2198        if (ret) {
2199                delete_tempfile(&temporary_sharedindex);
2200                return ret;
2201        }
2202        ret = rename_tempfile(&temporary_sharedindex,
2203                              git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2204        if (!ret)
2205                hashcpy(si->base_sha1, si->base->sha1);
2206        return ret;
2207}
2208
2209int write_locked_index(struct index_state *istate, struct lock_file *lock,
2210                       unsigned flags)
2211{
2212        struct split_index *si = istate->split_index;
2213
2214        if (!si || alternate_index_output ||
2215            (istate->cache_changed & ~EXTMASK)) {
2216                if (si)
2217                        hashclr(si->base_sha1);
2218                return do_write_locked_index(istate, lock, flags);
2219        }
2220
2221        if (getenv("GIT_TEST_SPLIT_INDEX")) {
2222                int v = si->base_sha1[0];
2223                if ((v & 15) < 6)
2224                        istate->cache_changed |= SPLIT_INDEX_ORDERED;
2225        }
2226        if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2227                int ret = write_shared_index(istate, lock, flags);
2228                if (ret)
2229                        return ret;
2230        }
2231
2232        return write_split_index(istate, lock, flags);
2233}
2234
2235/*
2236 * Read the index file that is potentially unmerged into given
2237 * index_state, dropping any unmerged entries.  Returns true if
2238 * the index is unmerged.  Callers who want to refuse to work
2239 * from an unmerged state can call this and check its return value,
2240 * instead of calling read_cache().
2241 */
2242int read_index_unmerged(struct index_state *istate)
2243{
2244        int i;
2245        int unmerged = 0;
2246
2247        read_index(istate);
2248        for (i = 0; i < istate->cache_nr; i++) {
2249                struct cache_entry *ce = istate->cache[i];
2250                struct cache_entry *new_ce;
2251                int size, len;
2252
2253                if (!ce_stage(ce))
2254                        continue;
2255                unmerged = 1;
2256                len = ce_namelen(ce);
2257                size = cache_entry_size(len);
2258                new_ce = xcalloc(1, size);
2259                memcpy(new_ce->name, ce->name, len);
2260                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2261                new_ce->ce_namelen = len;
2262                new_ce->ce_mode = ce->ce_mode;
2263                if (add_index_entry(istate, new_ce, 0))
2264                        return error("%s: cannot drop to stage #0",
2265                                     new_ce->name);
2266        }
2267        return unmerged;
2268}
2269
2270/*
2271 * Returns 1 if the path is an "other" path with respect to
2272 * the index; that is, the path is not mentioned in the index at all,
2273 * either as a file, a directory with some files in the index,
2274 * or as an unmerged entry.
2275 *
2276 * We helpfully remove a trailing "/" from directories so that
2277 * the output of read_directory can be used as-is.
2278 */
2279int index_name_is_other(const struct index_state *istate, const char *name,
2280                int namelen)
2281{
2282        int pos;
2283        if (namelen && name[namelen - 1] == '/')
2284                namelen--;
2285        pos = index_name_pos(istate, name, namelen);
2286        if (0 <= pos)
2287                return 0;       /* exact match */
2288        pos = -pos - 1;
2289        if (pos < istate->cache_nr) {
2290                struct cache_entry *ce = istate->cache[pos];
2291                if (ce_namelen(ce) == namelen &&
2292                    !memcmp(ce->name, name, namelen))
2293                        return 0; /* Yup, this one exists unmerged */
2294        }
2295        return 1;
2296}
2297
2298void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2299{
2300        int pos, len;
2301        unsigned long sz;
2302        enum object_type type;
2303        void *data;
2304
2305        len = strlen(path);
2306        pos = index_name_pos(istate, path, len);
2307        if (pos < 0) {
2308                /*
2309                 * We might be in the middle of a merge, in which
2310                 * case we would read stage #2 (ours).
2311                 */
2312                int i;
2313                for (i = -pos - 1;
2314                     (pos < 0 && i < istate->cache_nr &&
2315                      !strcmp(istate->cache[i]->name, path));
2316                     i++)
2317                        if (ce_stage(istate->cache[i]) == 2)
2318                                pos = i;
2319        }
2320        if (pos < 0)
2321                return NULL;
2322        data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2323        if (!data || type != OBJ_BLOB) {
2324                free(data);
2325                return NULL;
2326        }
2327        if (size)
2328                *size = sz;
2329        return data;
2330}
2331
2332void stat_validity_clear(struct stat_validity *sv)
2333{
2334        free(sv->sd);
2335        sv->sd = NULL;
2336}
2337
2338int stat_validity_check(struct stat_validity *sv, const char *path)
2339{
2340        struct stat st;
2341
2342        if (stat(path, &st) < 0)
2343                return sv->sd == NULL;
2344        if (!sv->sd)
2345                return 0;
2346        return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2347}
2348
2349void stat_validity_update(struct stat_validity *sv, int fd)
2350{
2351        struct stat st;
2352
2353        if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2354                stat_validity_clear(sv);
2355        else {
2356                if (!sv->sd)
2357                        sv->sd = xcalloc(1, sizeof(struct stat_data));
2358                fill_stat_data(sv->sd, &st);
2359        }
2360}