read-cache.con commit untracked-cache: temporarily disable with $GIT_DISABLE_UNTRACKED_CACHE (76e6b09)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 */
   6#define NO_THE_INDEX_COMPATIBILITY_MACROS
   7#include "cache.h"
   8#include "lockfile.h"
   9#include "cache-tree.h"
  10#include "refs.h"
  11#include "dir.h"
  12#include "tree.h"
  13#include "commit.h"
  14#include "blob.h"
  15#include "resolve-undo.h"
  16#include "strbuf.h"
  17#include "varint.h"
  18#include "split-index.h"
  19#include "sigchain.h"
  20#include "utf8.h"
  21
  22static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
  23                                               unsigned int options);
  24
  25/* Mask for the name length in ce_flags in the on-disk index */
  26
  27#define CE_NAMEMASK  (0x0fff)
  28
  29/* Index extensions.
  30 *
  31 * The first letter should be 'A'..'Z' for extensions that are not
  32 * necessary for a correct operation (i.e. optimization data).
  33 * When new extensions are added that _needs_ to be understood in
  34 * order to correctly interpret the index file, pick character that
  35 * is outside the range, to cause the reader to abort.
  36 */
  37
  38#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  39#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  40#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  41#define CACHE_EXT_LINK 0x6c696e6b         /* "link" */
  42#define CACHE_EXT_UNTRACKED 0x554E5452    /* "UNTR" */
  43
  44/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
  45#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
  46                 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
  47                 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
  48
  49struct index_state the_index;
  50static const char *alternate_index_output;
  51
  52static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  53{
  54        istate->cache[nr] = ce;
  55        add_name_hash(istate, ce);
  56}
  57
  58static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  59{
  60        struct cache_entry *old = istate->cache[nr];
  61
  62        replace_index_entry_in_base(istate, old, ce);
  63        remove_name_hash(istate, old);
  64        free(old);
  65        set_index_entry(istate, nr, ce);
  66        ce->ce_flags |= CE_UPDATE_IN_BASE;
  67        istate->cache_changed |= CE_ENTRY_CHANGED;
  68}
  69
  70void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  71{
  72        struct cache_entry *old = istate->cache[nr], *new;
  73        int namelen = strlen(new_name);
  74
  75        new = xmalloc(cache_entry_size(namelen));
  76        copy_cache_entry(new, old);
  77        new->ce_flags &= ~CE_HASHED;
  78        new->ce_namelen = namelen;
  79        new->index = 0;
  80        memcpy(new->name, new_name, namelen + 1);
  81
  82        cache_tree_invalidate_path(istate, old->name);
  83        untracked_cache_remove_from_index(istate, old->name);
  84        remove_index_entry_at(istate, nr);
  85        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  86}
  87
  88void fill_stat_data(struct stat_data *sd, struct stat *st)
  89{
  90        sd->sd_ctime.sec = (unsigned int)st->st_ctime;
  91        sd->sd_mtime.sec = (unsigned int)st->st_mtime;
  92        sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
  93        sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
  94        sd->sd_dev = st->st_dev;
  95        sd->sd_ino = st->st_ino;
  96        sd->sd_uid = st->st_uid;
  97        sd->sd_gid = st->st_gid;
  98        sd->sd_size = st->st_size;
  99}
 100
 101int match_stat_data(const struct stat_data *sd, struct stat *st)
 102{
 103        int changed = 0;
 104
 105        if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
 106                changed |= MTIME_CHANGED;
 107        if (trust_ctime && check_stat &&
 108            sd->sd_ctime.sec != (unsigned int)st->st_ctime)
 109                changed |= CTIME_CHANGED;
 110
 111#ifdef USE_NSEC
 112        if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
 113                changed |= MTIME_CHANGED;
 114        if (trust_ctime && check_stat &&
 115            sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
 116                changed |= CTIME_CHANGED;
 117#endif
 118
 119        if (check_stat) {
 120                if (sd->sd_uid != (unsigned int) st->st_uid ||
 121                        sd->sd_gid != (unsigned int) st->st_gid)
 122                        changed |= OWNER_CHANGED;
 123                if (sd->sd_ino != (unsigned int) st->st_ino)
 124                        changed |= INODE_CHANGED;
 125        }
 126
 127#ifdef USE_STDEV
 128        /*
 129         * st_dev breaks on network filesystems where different
 130         * clients will have different views of what "device"
 131         * the filesystem is on
 132         */
 133        if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
 134                        changed |= INODE_CHANGED;
 135#endif
 136
 137        if (sd->sd_size != (unsigned int) st->st_size)
 138                changed |= DATA_CHANGED;
 139
 140        return changed;
 141}
 142
 143/*
 144 * This only updates the "non-critical" parts of the directory
 145 * cache, ie the parts that aren't tracked by GIT, and only used
 146 * to validate the cache.
 147 */
 148void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 149{
 150        fill_stat_data(&ce->ce_stat_data, st);
 151
 152        if (assume_unchanged)
 153                ce->ce_flags |= CE_VALID;
 154
 155        if (S_ISREG(st->st_mode))
 156                ce_mark_uptodate(ce);
 157}
 158
 159static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
 160{
 161        int match = -1;
 162        int fd = open(ce->name, O_RDONLY);
 163
 164        if (fd >= 0) {
 165                unsigned char sha1[20];
 166                if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
 167                        match = hashcmp(sha1, ce->sha1);
 168                /* index_fd() closed the file descriptor already */
 169        }
 170        return match;
 171}
 172
 173static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
 174{
 175        int match = -1;
 176        void *buffer;
 177        unsigned long size;
 178        enum object_type type;
 179        struct strbuf sb = STRBUF_INIT;
 180
 181        if (strbuf_readlink(&sb, ce->name, expected_size))
 182                return -1;
 183
 184        buffer = read_sha1_file(ce->sha1, &type, &size);
 185        if (buffer) {
 186                if (size == sb.len)
 187                        match = memcmp(buffer, sb.buf, size);
 188                free(buffer);
 189        }
 190        strbuf_release(&sb);
 191        return match;
 192}
 193
 194static int ce_compare_gitlink(const struct cache_entry *ce)
 195{
 196        unsigned char sha1[20];
 197
 198        /*
 199         * We don't actually require that the .git directory
 200         * under GITLINK directory be a valid git directory. It
 201         * might even be missing (in case nobody populated that
 202         * sub-project).
 203         *
 204         * If so, we consider it always to match.
 205         */
 206        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 207                return 0;
 208        return hashcmp(sha1, ce->sha1);
 209}
 210
 211static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
 212{
 213        switch (st->st_mode & S_IFMT) {
 214        case S_IFREG:
 215                if (ce_compare_data(ce, st))
 216                        return DATA_CHANGED;
 217                break;
 218        case S_IFLNK:
 219                if (ce_compare_link(ce, xsize_t(st->st_size)))
 220                        return DATA_CHANGED;
 221                break;
 222        case S_IFDIR:
 223                if (S_ISGITLINK(ce->ce_mode))
 224                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 225        default:
 226                return TYPE_CHANGED;
 227        }
 228        return 0;
 229}
 230
 231static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
 232{
 233        unsigned int changed = 0;
 234
 235        if (ce->ce_flags & CE_REMOVE)
 236                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 237
 238        switch (ce->ce_mode & S_IFMT) {
 239        case S_IFREG:
 240                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 241                /* We consider only the owner x bit to be relevant for
 242                 * "mode changes"
 243                 */
 244                if (trust_executable_bit &&
 245                    (0100 & (ce->ce_mode ^ st->st_mode)))
 246                        changed |= MODE_CHANGED;
 247                break;
 248        case S_IFLNK:
 249                if (!S_ISLNK(st->st_mode) &&
 250                    (has_symlinks || !S_ISREG(st->st_mode)))
 251                        changed |= TYPE_CHANGED;
 252                break;
 253        case S_IFGITLINK:
 254                /* We ignore most of the st_xxx fields for gitlinks */
 255                if (!S_ISDIR(st->st_mode))
 256                        changed |= TYPE_CHANGED;
 257                else if (ce_compare_gitlink(ce))
 258                        changed |= DATA_CHANGED;
 259                return changed;
 260        default:
 261                die("internal error: ce_mode is %o", ce->ce_mode);
 262        }
 263
 264        changed |= match_stat_data(&ce->ce_stat_data, st);
 265
 266        /* Racily smudged entry? */
 267        if (!ce->ce_stat_data.sd_size) {
 268                if (!is_empty_blob_sha1(ce->sha1))
 269                        changed |= DATA_CHANGED;
 270        }
 271
 272        return changed;
 273}
 274
 275static int is_racy_stat(const struct index_state *istate,
 276                        const struct stat_data *sd)
 277{
 278        return (istate->timestamp.sec &&
 279#ifdef USE_NSEC
 280                 /* nanosecond timestamped files can also be racy! */
 281                (istate->timestamp.sec < sd->sd_mtime.sec ||
 282                 (istate->timestamp.sec == sd->sd_mtime.sec &&
 283                  istate->timestamp.nsec <= sd->sd_mtime.nsec))
 284#else
 285                istate->timestamp.sec <= sd->sd_mtime.sec
 286#endif
 287                );
 288}
 289
 290static int is_racy_timestamp(const struct index_state *istate,
 291                             const struct cache_entry *ce)
 292{
 293        return (!S_ISGITLINK(ce->ce_mode) &&
 294                is_racy_stat(istate, &ce->ce_stat_data));
 295}
 296
 297int match_stat_data_racy(const struct index_state *istate,
 298                         const struct stat_data *sd, struct stat *st)
 299{
 300        if (is_racy_stat(istate, sd))
 301                return MTIME_CHANGED;
 302        return match_stat_data(sd, st);
 303}
 304
 305int ie_match_stat(const struct index_state *istate,
 306                  const struct cache_entry *ce, struct stat *st,
 307                  unsigned int options)
 308{
 309        unsigned int changed;
 310        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 311        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 312        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 313
 314        /*
 315         * If it's marked as always valid in the index, it's
 316         * valid whatever the checked-out copy says.
 317         *
 318         * skip-worktree has the same effect with higher precedence
 319         */
 320        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 321                return 0;
 322        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 323                return 0;
 324
 325        /*
 326         * Intent-to-add entries have not been added, so the index entry
 327         * by definition never matches what is in the work tree until it
 328         * actually gets added.
 329         */
 330        if (ce->ce_flags & CE_INTENT_TO_ADD)
 331                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 332
 333        changed = ce_match_stat_basic(ce, st);
 334
 335        /*
 336         * Within 1 second of this sequence:
 337         *      echo xyzzy >file && git-update-index --add file
 338         * running this command:
 339         *      echo frotz >file
 340         * would give a falsely clean cache entry.  The mtime and
 341         * length match the cache, and other stat fields do not change.
 342         *
 343         * We could detect this at update-index time (the cache entry
 344         * being registered/updated records the same time as "now")
 345         * and delay the return from git-update-index, but that would
 346         * effectively mean we can make at most one commit per second,
 347         * which is not acceptable.  Instead, we check cache entries
 348         * whose mtime are the same as the index file timestamp more
 349         * carefully than others.
 350         */
 351        if (!changed && is_racy_timestamp(istate, ce)) {
 352                if (assume_racy_is_modified)
 353                        changed |= DATA_CHANGED;
 354                else
 355                        changed |= ce_modified_check_fs(ce, st);
 356        }
 357
 358        return changed;
 359}
 360
 361int ie_modified(const struct index_state *istate,
 362                const struct cache_entry *ce,
 363                struct stat *st, unsigned int options)
 364{
 365        int changed, changed_fs;
 366
 367        changed = ie_match_stat(istate, ce, st, options);
 368        if (!changed)
 369                return 0;
 370        /*
 371         * If the mode or type has changed, there's no point in trying
 372         * to refresh the entry - it's not going to match
 373         */
 374        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 375                return changed;
 376
 377        /*
 378         * Immediately after read-tree or update-index --cacheinfo,
 379         * the length field is zero, as we have never even read the
 380         * lstat(2) information once, and we cannot trust DATA_CHANGED
 381         * returned by ie_match_stat() which in turn was returned by
 382         * ce_match_stat_basic() to signal that the filesize of the
 383         * blob changed.  We have to actually go to the filesystem to
 384         * see if the contents match, and if so, should answer "unchanged".
 385         *
 386         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 387         * already has checked the actual HEAD from the filesystem in the
 388         * subproject.  If ie_match_stat() already said it is different,
 389         * then we know it is.
 390         */
 391        if ((changed & DATA_CHANGED) &&
 392            (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
 393                return changed;
 394
 395        changed_fs = ce_modified_check_fs(ce, st);
 396        if (changed_fs)
 397                return changed | changed_fs;
 398        return 0;
 399}
 400
 401int base_name_compare(const char *name1, int len1, int mode1,
 402                      const char *name2, int len2, int mode2)
 403{
 404        unsigned char c1, c2;
 405        int len = len1 < len2 ? len1 : len2;
 406        int cmp;
 407
 408        cmp = memcmp(name1, name2, len);
 409        if (cmp)
 410                return cmp;
 411        c1 = name1[len];
 412        c2 = name2[len];
 413        if (!c1 && S_ISDIR(mode1))
 414                c1 = '/';
 415        if (!c2 && S_ISDIR(mode2))
 416                c2 = '/';
 417        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 418}
 419
 420/*
 421 * df_name_compare() is identical to base_name_compare(), except it
 422 * compares conflicting directory/file entries as equal. Note that
 423 * while a directory name compares as equal to a regular file, they
 424 * then individually compare _differently_ to a filename that has
 425 * a dot after the basename (because '\0' < '.' < '/').
 426 *
 427 * This is used by routines that want to traverse the git namespace
 428 * but then handle conflicting entries together when possible.
 429 */
 430int df_name_compare(const char *name1, int len1, int mode1,
 431                    const char *name2, int len2, int mode2)
 432{
 433        int len = len1 < len2 ? len1 : len2, cmp;
 434        unsigned char c1, c2;
 435
 436        cmp = memcmp(name1, name2, len);
 437        if (cmp)
 438                return cmp;
 439        /* Directories and files compare equal (same length, same name) */
 440        if (len1 == len2)
 441                return 0;
 442        c1 = name1[len];
 443        if (!c1 && S_ISDIR(mode1))
 444                c1 = '/';
 445        c2 = name2[len];
 446        if (!c2 && S_ISDIR(mode2))
 447                c2 = '/';
 448        if (c1 == '/' && !c2)
 449                return 0;
 450        if (c2 == '/' && !c1)
 451                return 0;
 452        return c1 - c2;
 453}
 454
 455int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
 456{
 457        size_t min_len = (len1 < len2) ? len1 : len2;
 458        int cmp = memcmp(name1, name2, min_len);
 459        if (cmp)
 460                return cmp;
 461        if (len1 < len2)
 462                return -1;
 463        if (len1 > len2)
 464                return 1;
 465        return 0;
 466}
 467
 468int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 469{
 470        int cmp;
 471
 472        cmp = name_compare(name1, len1, name2, len2);
 473        if (cmp)
 474                return cmp;
 475
 476        if (stage1 < stage2)
 477                return -1;
 478        if (stage1 > stage2)
 479                return 1;
 480        return 0;
 481}
 482
 483static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 484{
 485        int first, last;
 486
 487        first = 0;
 488        last = istate->cache_nr;
 489        while (last > first) {
 490                int next = (last + first) >> 1;
 491                struct cache_entry *ce = istate->cache[next];
 492                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 493                if (!cmp)
 494                        return next;
 495                if (cmp < 0) {
 496                        last = next;
 497                        continue;
 498                }
 499                first = next+1;
 500        }
 501        return -first-1;
 502}
 503
 504int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 505{
 506        return index_name_stage_pos(istate, name, namelen, 0);
 507}
 508
 509/* Remove entry, return true if there are more entries to go.. */
 510int remove_index_entry_at(struct index_state *istate, int pos)
 511{
 512        struct cache_entry *ce = istate->cache[pos];
 513
 514        record_resolve_undo(istate, ce);
 515        remove_name_hash(istate, ce);
 516        save_or_free_index_entry(istate, ce);
 517        istate->cache_changed |= CE_ENTRY_REMOVED;
 518        istate->cache_nr--;
 519        if (pos >= istate->cache_nr)
 520                return 0;
 521        memmove(istate->cache + pos,
 522                istate->cache + pos + 1,
 523                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 524        return 1;
 525}
 526
 527/*
 528 * Remove all cache entries marked for removal, that is where
 529 * CE_REMOVE is set in ce_flags.  This is much more effective than
 530 * calling remove_index_entry_at() for each entry to be removed.
 531 */
 532void remove_marked_cache_entries(struct index_state *istate)
 533{
 534        struct cache_entry **ce_array = istate->cache;
 535        unsigned int i, j;
 536
 537        for (i = j = 0; i < istate->cache_nr; i++) {
 538                if (ce_array[i]->ce_flags & CE_REMOVE) {
 539                        remove_name_hash(istate, ce_array[i]);
 540                        save_or_free_index_entry(istate, ce_array[i]);
 541                }
 542                else
 543                        ce_array[j++] = ce_array[i];
 544        }
 545        if (j == istate->cache_nr)
 546                return;
 547        istate->cache_changed |= CE_ENTRY_REMOVED;
 548        istate->cache_nr = j;
 549}
 550
 551int remove_file_from_index(struct index_state *istate, const char *path)
 552{
 553        int pos = index_name_pos(istate, path, strlen(path));
 554        if (pos < 0)
 555                pos = -pos-1;
 556        cache_tree_invalidate_path(istate, path);
 557        untracked_cache_remove_from_index(istate, path);
 558        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 559                remove_index_entry_at(istate, pos);
 560        return 0;
 561}
 562
 563static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 564{
 565        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 566}
 567
 568static int index_name_pos_also_unmerged(struct index_state *istate,
 569        const char *path, int namelen)
 570{
 571        int pos = index_name_pos(istate, path, namelen);
 572        struct cache_entry *ce;
 573
 574        if (pos >= 0)
 575                return pos;
 576
 577        /* maybe unmerged? */
 578        pos = -1 - pos;
 579        if (pos >= istate->cache_nr ||
 580                        compare_name((ce = istate->cache[pos]), path, namelen))
 581                return -1;
 582
 583        /* order of preference: stage 2, 1, 3 */
 584        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 585                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 586                        !compare_name(ce, path, namelen))
 587                pos++;
 588        return pos;
 589}
 590
 591static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 592{
 593        int len = ce_namelen(ce);
 594        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 595}
 596
 597/*
 598 * If we add a filename that aliases in the cache, we will use the
 599 * name that we already have - but we don't want to update the same
 600 * alias twice, because that implies that there were actually two
 601 * different files with aliasing names!
 602 *
 603 * So we use the CE_ADDED flag to verify that the alias was an old
 604 * one before we accept it as
 605 */
 606static struct cache_entry *create_alias_ce(struct index_state *istate,
 607                                           struct cache_entry *ce,
 608                                           struct cache_entry *alias)
 609{
 610        int len;
 611        struct cache_entry *new;
 612
 613        if (alias->ce_flags & CE_ADDED)
 614                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 615
 616        /* Ok, create the new entry using the name of the existing alias */
 617        len = ce_namelen(alias);
 618        new = xcalloc(1, cache_entry_size(len));
 619        memcpy(new->name, alias->name, len);
 620        copy_cache_entry(new, ce);
 621        save_or_free_index_entry(istate, ce);
 622        return new;
 623}
 624
 625void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
 626{
 627        unsigned char sha1[20];
 628        if (write_sha1_file("", 0, blob_type, sha1))
 629                die("cannot create an empty blob in the object database");
 630        hashcpy(ce->sha1, sha1);
 631}
 632
 633int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 634{
 635        int size, namelen, was_same;
 636        mode_t st_mode = st->st_mode;
 637        struct cache_entry *ce, *alias;
 638        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 639        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 640        int pretend = flags & ADD_CACHE_PRETEND;
 641        int intent_only = flags & ADD_CACHE_INTENT;
 642        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 643                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 644
 645        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 646                return error("%s: can only add regular files, symbolic links or git-directories", path);
 647
 648        namelen = strlen(path);
 649        if (S_ISDIR(st_mode)) {
 650                while (namelen && path[namelen-1] == '/')
 651                        namelen--;
 652        }
 653        size = cache_entry_size(namelen);
 654        ce = xcalloc(1, size);
 655        memcpy(ce->name, path, namelen);
 656        ce->ce_namelen = namelen;
 657        if (!intent_only)
 658                fill_stat_cache_info(ce, st);
 659        else
 660                ce->ce_flags |= CE_INTENT_TO_ADD;
 661
 662        if (trust_executable_bit && has_symlinks)
 663                ce->ce_mode = create_ce_mode(st_mode);
 664        else {
 665                /* If there is an existing entry, pick the mode bits and type
 666                 * from it, otherwise assume unexecutable regular file.
 667                 */
 668                struct cache_entry *ent;
 669                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 670
 671                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 672                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 673        }
 674
 675        /* When core.ignorecase=true, determine if a directory of the same name but differing
 676         * case already exists within the Git repository.  If it does, ensure the directory
 677         * case of the file being added to the repository matches (is folded into) the existing
 678         * entry's directory case.
 679         */
 680        if (ignore_case) {
 681                const char *startPtr = ce->name;
 682                const char *ptr = startPtr;
 683                while (*ptr) {
 684                        while (*ptr && *ptr != '/')
 685                                ++ptr;
 686                        if (*ptr == '/') {
 687                                struct cache_entry *foundce;
 688                                ++ptr;
 689                                foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
 690                                if (foundce) {
 691                                        memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
 692                                        startPtr = ptr;
 693                                }
 694                        }
 695                }
 696        }
 697
 698        alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 699        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 700                /* Nothing changed, really */
 701                free(ce);
 702                if (!S_ISGITLINK(alias->ce_mode))
 703                        ce_mark_uptodate(alias);
 704                alias->ce_flags |= CE_ADDED;
 705                return 0;
 706        }
 707        if (!intent_only) {
 708                if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
 709                        return error("unable to index file %s", path);
 710        } else
 711                set_object_name_for_intent_to_add_entry(ce);
 712
 713        if (ignore_case && alias && different_name(ce, alias))
 714                ce = create_alias_ce(istate, ce, alias);
 715        ce->ce_flags |= CE_ADDED;
 716
 717        /* It was suspected to be racily clean, but it turns out to be Ok */
 718        was_same = (alias &&
 719                    !ce_stage(alias) &&
 720                    !hashcmp(alias->sha1, ce->sha1) &&
 721                    ce->ce_mode == alias->ce_mode);
 722
 723        if (pretend)
 724                ;
 725        else if (add_index_entry(istate, ce, add_option))
 726                return error("unable to add %s to index",path);
 727        if (verbose && !was_same)
 728                printf("add '%s'\n", path);
 729        return 0;
 730}
 731
 732int add_file_to_index(struct index_state *istate, const char *path, int flags)
 733{
 734        struct stat st;
 735        if (lstat(path, &st))
 736                die_errno("unable to stat '%s'", path);
 737        return add_to_index(istate, path, &st, flags);
 738}
 739
 740struct cache_entry *make_cache_entry(unsigned int mode,
 741                const unsigned char *sha1, const char *path, int stage,
 742                unsigned int refresh_options)
 743{
 744        int size, len;
 745        struct cache_entry *ce, *ret;
 746
 747        if (!verify_path(path)) {
 748                error("Invalid path '%s'", path);
 749                return NULL;
 750        }
 751
 752        len = strlen(path);
 753        size = cache_entry_size(len);
 754        ce = xcalloc(1, size);
 755
 756        hashcpy(ce->sha1, sha1);
 757        memcpy(ce->name, path, len);
 758        ce->ce_flags = create_ce_flags(stage);
 759        ce->ce_namelen = len;
 760        ce->ce_mode = create_ce_mode(mode);
 761
 762        ret = refresh_cache_entry(ce, refresh_options);
 763        if (!ret) {
 764                free(ce);
 765                return NULL;
 766        } else {
 767                return ret;
 768        }
 769}
 770
 771int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
 772{
 773        int len = ce_namelen(a);
 774        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 775}
 776
 777/*
 778 * We fundamentally don't like some paths: we don't want
 779 * dot or dot-dot anywhere, and for obvious reasons don't
 780 * want to recurse into ".git" either.
 781 *
 782 * Also, we don't want double slashes or slashes at the
 783 * end that can make pathnames ambiguous.
 784 */
 785static int verify_dotfile(const char *rest)
 786{
 787        /*
 788         * The first character was '.', but that
 789         * has already been discarded, we now test
 790         * the rest.
 791         */
 792
 793        /* "." is not allowed */
 794        if (*rest == '\0' || is_dir_sep(*rest))
 795                return 0;
 796
 797        switch (*rest) {
 798        /*
 799         * ".git" followed by  NUL or slash is bad. This
 800         * shares the path end test with the ".." case.
 801         */
 802        case 'g':
 803        case 'G':
 804                if (rest[1] != 'i' && rest[1] != 'I')
 805                        break;
 806                if (rest[2] != 't' && rest[2] != 'T')
 807                        break;
 808                rest += 2;
 809        /* fallthrough */
 810        case '.':
 811                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 812                        return 0;
 813        }
 814        return 1;
 815}
 816
 817int verify_path(const char *path)
 818{
 819        char c;
 820
 821        if (has_dos_drive_prefix(path))
 822                return 0;
 823
 824        goto inside;
 825        for (;;) {
 826                if (!c)
 827                        return 1;
 828                if (is_dir_sep(c)) {
 829inside:
 830                        if (protect_hfs && is_hfs_dotgit(path))
 831                                return 0;
 832                        if (protect_ntfs && is_ntfs_dotgit(path))
 833                                return 0;
 834                        c = *path++;
 835                        if ((c == '.' && !verify_dotfile(path)) ||
 836                            is_dir_sep(c) || c == '\0')
 837                                return 0;
 838                }
 839                c = *path++;
 840        }
 841}
 842
 843/*
 844 * Do we have another file that has the beginning components being a
 845 * proper superset of the name we're trying to add?
 846 */
 847static int has_file_name(struct index_state *istate,
 848                         const struct cache_entry *ce, int pos, int ok_to_replace)
 849{
 850        int retval = 0;
 851        int len = ce_namelen(ce);
 852        int stage = ce_stage(ce);
 853        const char *name = ce->name;
 854
 855        while (pos < istate->cache_nr) {
 856                struct cache_entry *p = istate->cache[pos++];
 857
 858                if (len >= ce_namelen(p))
 859                        break;
 860                if (memcmp(name, p->name, len))
 861                        break;
 862                if (ce_stage(p) != stage)
 863                        continue;
 864                if (p->name[len] != '/')
 865                        continue;
 866                if (p->ce_flags & CE_REMOVE)
 867                        continue;
 868                retval = -1;
 869                if (!ok_to_replace)
 870                        break;
 871                remove_index_entry_at(istate, --pos);
 872        }
 873        return retval;
 874}
 875
 876/*
 877 * Do we have another file with a pathname that is a proper
 878 * subset of the name we're trying to add?
 879 */
 880static int has_dir_name(struct index_state *istate,
 881                        const struct cache_entry *ce, int pos, int ok_to_replace)
 882{
 883        int retval = 0;
 884        int stage = ce_stage(ce);
 885        const char *name = ce->name;
 886        const char *slash = name + ce_namelen(ce);
 887
 888        for (;;) {
 889                int len;
 890
 891                for (;;) {
 892                        if (*--slash == '/')
 893                                break;
 894                        if (slash <= ce->name)
 895                                return retval;
 896                }
 897                len = slash - name;
 898
 899                pos = index_name_stage_pos(istate, name, len, stage);
 900                if (pos >= 0) {
 901                        /*
 902                         * Found one, but not so fast.  This could
 903                         * be a marker that says "I was here, but
 904                         * I am being removed".  Such an entry is
 905                         * not a part of the resulting tree, and
 906                         * it is Ok to have a directory at the same
 907                         * path.
 908                         */
 909                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 910                                retval = -1;
 911                                if (!ok_to_replace)
 912                                        break;
 913                                remove_index_entry_at(istate, pos);
 914                                continue;
 915                        }
 916                }
 917                else
 918                        pos = -pos-1;
 919
 920                /*
 921                 * Trivial optimization: if we find an entry that
 922                 * already matches the sub-directory, then we know
 923                 * we're ok, and we can exit.
 924                 */
 925                while (pos < istate->cache_nr) {
 926                        struct cache_entry *p = istate->cache[pos];
 927                        if ((ce_namelen(p) <= len) ||
 928                            (p->name[len] != '/') ||
 929                            memcmp(p->name, name, len))
 930                                break; /* not our subdirectory */
 931                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 932                                /*
 933                                 * p is at the same stage as our entry, and
 934                                 * is a subdirectory of what we are looking
 935                                 * at, so we cannot have conflicts at our
 936                                 * level or anything shorter.
 937                                 */
 938                                return retval;
 939                        pos++;
 940                }
 941        }
 942        return retval;
 943}
 944
 945/* We may be in a situation where we already have path/file and path
 946 * is being added, or we already have path and path/file is being
 947 * added.  Either one would result in a nonsense tree that has path
 948 * twice when git-write-tree tries to write it out.  Prevent it.
 949 *
 950 * If ok-to-replace is specified, we remove the conflicting entries
 951 * from the cache so the caller should recompute the insert position.
 952 * When this happens, we return non-zero.
 953 */
 954static int check_file_directory_conflict(struct index_state *istate,
 955                                         const struct cache_entry *ce,
 956                                         int pos, int ok_to_replace)
 957{
 958        int retval;
 959
 960        /*
 961         * When ce is an "I am going away" entry, we allow it to be added
 962         */
 963        if (ce->ce_flags & CE_REMOVE)
 964                return 0;
 965
 966        /*
 967         * We check if the path is a sub-path of a subsequent pathname
 968         * first, since removing those will not change the position
 969         * in the array.
 970         */
 971        retval = has_file_name(istate, ce, pos, ok_to_replace);
 972
 973        /*
 974         * Then check if the path might have a clashing sub-directory
 975         * before it.
 976         */
 977        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 978}
 979
 980static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 981{
 982        int pos;
 983        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 984        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 985        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 986        int new_only = option & ADD_CACHE_NEW_ONLY;
 987
 988        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
 989                cache_tree_invalidate_path(istate, ce->name);
 990        pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 991
 992        /* existing match? Just replace it. */
 993        if (pos >= 0) {
 994                if (!new_only)
 995                        replace_index_entry(istate, pos, ce);
 996                return 0;
 997        }
 998        pos = -pos-1;
 999
1000        untracked_cache_add_to_index(istate, ce->name);
1001
1002        /*
1003         * Inserting a merged entry ("stage 0") into the index
1004         * will always replace all non-merged entries..
1005         */
1006        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1007                while (ce_same_name(istate->cache[pos], ce)) {
1008                        ok_to_add = 1;
1009                        if (!remove_index_entry_at(istate, pos))
1010                                break;
1011                }
1012        }
1013
1014        if (!ok_to_add)
1015                return -1;
1016        if (!verify_path(ce->name))
1017                return error("Invalid path '%s'", ce->name);
1018
1019        if (!skip_df_check &&
1020            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1021                if (!ok_to_replace)
1022                        return error("'%s' appears as both a file and as a directory",
1023                                     ce->name);
1024                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1025                pos = -pos-1;
1026        }
1027        return pos + 1;
1028}
1029
1030int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1031{
1032        int pos;
1033
1034        if (option & ADD_CACHE_JUST_APPEND)
1035                pos = istate->cache_nr;
1036        else {
1037                int ret;
1038                ret = add_index_entry_with_check(istate, ce, option);
1039                if (ret <= 0)
1040                        return ret;
1041                pos = ret - 1;
1042        }
1043
1044        /* Make sure the array is big enough .. */
1045        ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1046
1047        /* Add it in.. */
1048        istate->cache_nr++;
1049        if (istate->cache_nr > pos + 1)
1050                memmove(istate->cache + pos + 1,
1051                        istate->cache + pos,
1052                        (istate->cache_nr - pos - 1) * sizeof(ce));
1053        set_index_entry(istate, pos, ce);
1054        istate->cache_changed |= CE_ENTRY_ADDED;
1055        return 0;
1056}
1057
1058/*
1059 * "refresh" does not calculate a new sha1 file or bring the
1060 * cache up-to-date for mode/content changes. But what it
1061 * _does_ do is to "re-match" the stat information of a file
1062 * with the cache, so that you can refresh the cache for a
1063 * file that hasn't been changed but where the stat entry is
1064 * out of date.
1065 *
1066 * For example, you'd want to do this after doing a "git-read-tree",
1067 * to link up the stat cache details with the proper files.
1068 */
1069static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1070                                             struct cache_entry *ce,
1071                                             unsigned int options, int *err,
1072                                             int *changed_ret)
1073{
1074        struct stat st;
1075        struct cache_entry *updated;
1076        int changed, size;
1077        int refresh = options & CE_MATCH_REFRESH;
1078        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1079        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1080        int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1081
1082        if (!refresh || ce_uptodate(ce))
1083                return ce;
1084
1085        /*
1086         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1087         * that the change to the work tree does not matter and told
1088         * us not to worry.
1089         */
1090        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1091                ce_mark_uptodate(ce);
1092                return ce;
1093        }
1094        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1095                ce_mark_uptodate(ce);
1096                return ce;
1097        }
1098
1099        if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1100                if (ignore_missing)
1101                        return ce;
1102                if (err)
1103                        *err = ENOENT;
1104                return NULL;
1105        }
1106
1107        if (lstat(ce->name, &st) < 0) {
1108                if (ignore_missing && errno == ENOENT)
1109                        return ce;
1110                if (err)
1111                        *err = errno;
1112                return NULL;
1113        }
1114
1115        changed = ie_match_stat(istate, ce, &st, options);
1116        if (changed_ret)
1117                *changed_ret = changed;
1118        if (!changed) {
1119                /*
1120                 * The path is unchanged.  If we were told to ignore
1121                 * valid bit, then we did the actual stat check and
1122                 * found that the entry is unmodified.  If the entry
1123                 * is not marked VALID, this is the place to mark it
1124                 * valid again, under "assume unchanged" mode.
1125                 */
1126                if (ignore_valid && assume_unchanged &&
1127                    !(ce->ce_flags & CE_VALID))
1128                        ; /* mark this one VALID again */
1129                else {
1130                        /*
1131                         * We do not mark the index itself "modified"
1132                         * because CE_UPTODATE flag is in-core only;
1133                         * we are not going to write this change out.
1134                         */
1135                        if (!S_ISGITLINK(ce->ce_mode))
1136                                ce_mark_uptodate(ce);
1137                        return ce;
1138                }
1139        }
1140
1141        if (ie_modified(istate, ce, &st, options)) {
1142                if (err)
1143                        *err = EINVAL;
1144                return NULL;
1145        }
1146
1147        size = ce_size(ce);
1148        updated = xmalloc(size);
1149        memcpy(updated, ce, size);
1150        fill_stat_cache_info(updated, &st);
1151        /*
1152         * If ignore_valid is not set, we should leave CE_VALID bit
1153         * alone.  Otherwise, paths marked with --no-assume-unchanged
1154         * (i.e. things to be edited) will reacquire CE_VALID bit
1155         * automatically, which is not really what we want.
1156         */
1157        if (!ignore_valid && assume_unchanged &&
1158            !(ce->ce_flags & CE_VALID))
1159                updated->ce_flags &= ~CE_VALID;
1160
1161        /* istate->cache_changed is updated in the caller */
1162        return updated;
1163}
1164
1165static void show_file(const char * fmt, const char * name, int in_porcelain,
1166                      int * first, const char *header_msg)
1167{
1168        if (in_porcelain && *first && header_msg) {
1169                printf("%s\n", header_msg);
1170                *first = 0;
1171        }
1172        printf(fmt, name);
1173}
1174
1175int refresh_index(struct index_state *istate, unsigned int flags,
1176                  const struct pathspec *pathspec,
1177                  char *seen, const char *header_msg)
1178{
1179        int i;
1180        int has_errors = 0;
1181        int really = (flags & REFRESH_REALLY) != 0;
1182        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1183        int quiet = (flags & REFRESH_QUIET) != 0;
1184        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1185        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1186        int first = 1;
1187        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1188        unsigned int options = (CE_MATCH_REFRESH |
1189                                (really ? CE_MATCH_IGNORE_VALID : 0) |
1190                                (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1191        const char *modified_fmt;
1192        const char *deleted_fmt;
1193        const char *typechange_fmt;
1194        const char *added_fmt;
1195        const char *unmerged_fmt;
1196
1197        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1198        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1199        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1200        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1201        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1202        for (i = 0; i < istate->cache_nr; i++) {
1203                struct cache_entry *ce, *new;
1204                int cache_errno = 0;
1205                int changed = 0;
1206                int filtered = 0;
1207
1208                ce = istate->cache[i];
1209                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1210                        continue;
1211
1212                if (pathspec && !ce_path_match(ce, pathspec, seen))
1213                        filtered = 1;
1214
1215                if (ce_stage(ce)) {
1216                        while ((i < istate->cache_nr) &&
1217                               ! strcmp(istate->cache[i]->name, ce->name))
1218                                i++;
1219                        i--;
1220                        if (allow_unmerged)
1221                                continue;
1222                        if (!filtered)
1223                                show_file(unmerged_fmt, ce->name, in_porcelain,
1224                                          &first, header_msg);
1225                        has_errors = 1;
1226                        continue;
1227                }
1228
1229                if (filtered)
1230                        continue;
1231
1232                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1233                if (new == ce)
1234                        continue;
1235                if (!new) {
1236                        const char *fmt;
1237
1238                        if (really && cache_errno == EINVAL) {
1239                                /* If we are doing --really-refresh that
1240                                 * means the index is not valid anymore.
1241                                 */
1242                                ce->ce_flags &= ~CE_VALID;
1243                                ce->ce_flags |= CE_UPDATE_IN_BASE;
1244                                istate->cache_changed |= CE_ENTRY_CHANGED;
1245                        }
1246                        if (quiet)
1247                                continue;
1248
1249                        if (cache_errno == ENOENT)
1250                                fmt = deleted_fmt;
1251                        else if (ce->ce_flags & CE_INTENT_TO_ADD)
1252                                fmt = added_fmt; /* must be before other checks */
1253                        else if (changed & TYPE_CHANGED)
1254                                fmt = typechange_fmt;
1255                        else
1256                                fmt = modified_fmt;
1257                        show_file(fmt,
1258                                  ce->name, in_porcelain, &first, header_msg);
1259                        has_errors = 1;
1260                        continue;
1261                }
1262
1263                replace_index_entry(istate, i, new);
1264        }
1265        return has_errors;
1266}
1267
1268static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1269                                               unsigned int options)
1270{
1271        return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1272}
1273
1274
1275/*****************************************************************
1276 * Index File I/O
1277 *****************************************************************/
1278
1279#define INDEX_FORMAT_DEFAULT 3
1280
1281static unsigned int get_index_format_default(void)
1282{
1283        char *envversion = getenv("GIT_INDEX_VERSION");
1284        char *endp;
1285        int value;
1286        unsigned int version = INDEX_FORMAT_DEFAULT;
1287
1288        if (!envversion) {
1289                if (!git_config_get_int("index.version", &value))
1290                        version = value;
1291                if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1292                        warning(_("index.version set, but the value is invalid.\n"
1293                                  "Using version %i"), INDEX_FORMAT_DEFAULT);
1294                        return INDEX_FORMAT_DEFAULT;
1295                }
1296                return version;
1297        }
1298
1299        version = strtoul(envversion, &endp, 10);
1300        if (*endp ||
1301            version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1302                warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1303                          "Using version %i"), INDEX_FORMAT_DEFAULT);
1304                version = INDEX_FORMAT_DEFAULT;
1305        }
1306        return version;
1307}
1308
1309/*
1310 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1311 * Again - this is just a (very strong in practice) heuristic that
1312 * the inode hasn't changed.
1313 *
1314 * We save the fields in big-endian order to allow using the
1315 * index file over NFS transparently.
1316 */
1317struct ondisk_cache_entry {
1318        struct cache_time ctime;
1319        struct cache_time mtime;
1320        uint32_t dev;
1321        uint32_t ino;
1322        uint32_t mode;
1323        uint32_t uid;
1324        uint32_t gid;
1325        uint32_t size;
1326        unsigned char sha1[20];
1327        uint16_t flags;
1328        char name[FLEX_ARRAY]; /* more */
1329};
1330
1331/*
1332 * This struct is used when CE_EXTENDED bit is 1
1333 * The struct must match ondisk_cache_entry exactly from
1334 * ctime till flags
1335 */
1336struct ondisk_cache_entry_extended {
1337        struct cache_time ctime;
1338        struct cache_time mtime;
1339        uint32_t dev;
1340        uint32_t ino;
1341        uint32_t mode;
1342        uint32_t uid;
1343        uint32_t gid;
1344        uint32_t size;
1345        unsigned char sha1[20];
1346        uint16_t flags;
1347        uint16_t flags2;
1348        char name[FLEX_ARRAY]; /* more */
1349};
1350
1351/* These are only used for v3 or lower */
1352#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1353#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1354#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1355#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1356                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1357                            ondisk_cache_entry_size(ce_namelen(ce)))
1358
1359static int verify_hdr(struct cache_header *hdr, unsigned long size)
1360{
1361        git_SHA_CTX c;
1362        unsigned char sha1[20];
1363        int hdr_version;
1364
1365        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1366                return error("bad signature");
1367        hdr_version = ntohl(hdr->hdr_version);
1368        if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1369                return error("bad index version %d", hdr_version);
1370        git_SHA1_Init(&c);
1371        git_SHA1_Update(&c, hdr, size - 20);
1372        git_SHA1_Final(sha1, &c);
1373        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1374                return error("bad index file sha1 signature");
1375        return 0;
1376}
1377
1378static int read_index_extension(struct index_state *istate,
1379                                const char *ext, void *data, unsigned long sz)
1380{
1381        switch (CACHE_EXT(ext)) {
1382        case CACHE_EXT_TREE:
1383                istate->cache_tree = cache_tree_read(data, sz);
1384                break;
1385        case CACHE_EXT_RESOLVE_UNDO:
1386                istate->resolve_undo = resolve_undo_read(data, sz);
1387                break;
1388        case CACHE_EXT_LINK:
1389                if (read_link_extension(istate, data, sz))
1390                        return -1;
1391                break;
1392        case CACHE_EXT_UNTRACKED:
1393                istate->untracked = read_untracked_extension(data, sz);
1394                break;
1395        default:
1396                if (*ext < 'A' || 'Z' < *ext)
1397                        return error("index uses %.4s extension, which we do not understand",
1398                                     ext);
1399                fprintf(stderr, "ignoring %.4s extension\n", ext);
1400                break;
1401        }
1402        return 0;
1403}
1404
1405int hold_locked_index(struct lock_file *lk, int die_on_error)
1406{
1407        return hold_lock_file_for_update(lk, get_index_file(),
1408                                         die_on_error
1409                                         ? LOCK_DIE_ON_ERROR
1410                                         : 0);
1411}
1412
1413int read_index(struct index_state *istate)
1414{
1415        return read_index_from(istate, get_index_file());
1416}
1417
1418static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1419                                                   unsigned int flags,
1420                                                   const char *name,
1421                                                   size_t len)
1422{
1423        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1424
1425        ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1426        ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1427        ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1428        ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1429        ce->ce_stat_data.sd_dev   = get_be32(&ondisk->dev);
1430        ce->ce_stat_data.sd_ino   = get_be32(&ondisk->ino);
1431        ce->ce_mode  = get_be32(&ondisk->mode);
1432        ce->ce_stat_data.sd_uid   = get_be32(&ondisk->uid);
1433        ce->ce_stat_data.sd_gid   = get_be32(&ondisk->gid);
1434        ce->ce_stat_data.sd_size  = get_be32(&ondisk->size);
1435        ce->ce_flags = flags & ~CE_NAMEMASK;
1436        ce->ce_namelen = len;
1437        ce->index = 0;
1438        hashcpy(ce->sha1, ondisk->sha1);
1439        memcpy(ce->name, name, len);
1440        ce->name[len] = '\0';
1441        return ce;
1442}
1443
1444/*
1445 * Adjacent cache entries tend to share the leading paths, so it makes
1446 * sense to only store the differences in later entries.  In the v4
1447 * on-disk format of the index, each on-disk cache entry stores the
1448 * number of bytes to be stripped from the end of the previous name,
1449 * and the bytes to append to the result, to come up with its name.
1450 */
1451static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1452{
1453        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1454        size_t len = decode_varint(&cp);
1455
1456        if (name->len < len)
1457                die("malformed name field in the index");
1458        strbuf_remove(name, name->len - len, len);
1459        for (ep = cp; *ep; ep++)
1460                ; /* find the end */
1461        strbuf_add(name, cp, ep - cp);
1462        return (const char *)ep + 1 - cp_;
1463}
1464
1465static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1466                                            unsigned long *ent_size,
1467                                            struct strbuf *previous_name)
1468{
1469        struct cache_entry *ce;
1470        size_t len;
1471        const char *name;
1472        unsigned int flags;
1473
1474        /* On-disk flags are just 16 bits */
1475        flags = get_be16(&ondisk->flags);
1476        len = flags & CE_NAMEMASK;
1477
1478        if (flags & CE_EXTENDED) {
1479                struct ondisk_cache_entry_extended *ondisk2;
1480                int extended_flags;
1481                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1482                extended_flags = get_be16(&ondisk2->flags2) << 16;
1483                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1484                if (extended_flags & ~CE_EXTENDED_FLAGS)
1485                        die("Unknown index entry format %08x", extended_flags);
1486                flags |= extended_flags;
1487                name = ondisk2->name;
1488        }
1489        else
1490                name = ondisk->name;
1491
1492        if (!previous_name) {
1493                /* v3 and earlier */
1494                if (len == CE_NAMEMASK)
1495                        len = strlen(name);
1496                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1497
1498                *ent_size = ondisk_ce_size(ce);
1499        } else {
1500                unsigned long consumed;
1501                consumed = expand_name_field(previous_name, name);
1502                ce = cache_entry_from_ondisk(ondisk, flags,
1503                                             previous_name->buf,
1504                                             previous_name->len);
1505
1506                *ent_size = (name - ((char *)ondisk)) + consumed;
1507        }
1508        return ce;
1509}
1510
1511static void check_ce_order(struct cache_entry *ce, struct cache_entry *next_ce)
1512{
1513        int name_compare = strcmp(ce->name, next_ce->name);
1514        if (0 < name_compare)
1515                die("unordered stage entries in index");
1516        if (!name_compare) {
1517                if (!ce_stage(ce))
1518                        die("multiple stage entries for merged file '%s'",
1519                                ce->name);
1520                if (ce_stage(ce) > ce_stage(next_ce))
1521                        die("unordered stage entries for '%s'",
1522                                ce->name);
1523        }
1524}
1525
1526/* remember to discard_cache() before reading a different cache! */
1527int do_read_index(struct index_state *istate, const char *path, int must_exist)
1528{
1529        int fd, i;
1530        struct stat st;
1531        unsigned long src_offset;
1532        struct cache_header *hdr;
1533        void *mmap;
1534        size_t mmap_size;
1535        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1536
1537        if (istate->initialized)
1538                return istate->cache_nr;
1539
1540        istate->timestamp.sec = 0;
1541        istate->timestamp.nsec = 0;
1542        fd = open(path, O_RDONLY);
1543        if (fd < 0) {
1544                if (!must_exist && errno == ENOENT)
1545                        return 0;
1546                die_errno("%s: index file open failed", path);
1547        }
1548
1549        if (fstat(fd, &st))
1550                die_errno("cannot stat the open index");
1551
1552        mmap_size = xsize_t(st.st_size);
1553        if (mmap_size < sizeof(struct cache_header) + 20)
1554                die("index file smaller than expected");
1555
1556        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1557        if (mmap == MAP_FAILED)
1558                die_errno("unable to map index file");
1559        close(fd);
1560
1561        hdr = mmap;
1562        if (verify_hdr(hdr, mmap_size) < 0)
1563                goto unmap;
1564
1565        hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1566        istate->version = ntohl(hdr->hdr_version);
1567        istate->cache_nr = ntohl(hdr->hdr_entries);
1568        istate->cache_alloc = alloc_nr(istate->cache_nr);
1569        istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1570        istate->initialized = 1;
1571
1572        if (istate->version == 4)
1573                previous_name = &previous_name_buf;
1574        else
1575                previous_name = NULL;
1576
1577        src_offset = sizeof(*hdr);
1578        for (i = 0; i < istate->cache_nr; i++) {
1579                struct ondisk_cache_entry *disk_ce;
1580                struct cache_entry *ce;
1581                unsigned long consumed;
1582
1583                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1584                ce = create_from_disk(disk_ce, &consumed, previous_name);
1585                set_index_entry(istate, i, ce);
1586
1587                if (i > 0)
1588                        check_ce_order(istate->cache[i - 1], ce);
1589
1590                src_offset += consumed;
1591        }
1592        strbuf_release(&previous_name_buf);
1593        istate->timestamp.sec = st.st_mtime;
1594        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1595
1596        while (src_offset <= mmap_size - 20 - 8) {
1597                /* After an array of active_nr index entries,
1598                 * there can be arbitrary number of extended
1599                 * sections, each of which is prefixed with
1600                 * extension name (4-byte) and section length
1601                 * in 4-byte network byte order.
1602                 */
1603                uint32_t extsize;
1604                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1605                extsize = ntohl(extsize);
1606                if (read_index_extension(istate,
1607                                         (const char *) mmap + src_offset,
1608                                         (char *) mmap + src_offset + 8,
1609                                         extsize) < 0)
1610                        goto unmap;
1611                src_offset += 8;
1612                src_offset += extsize;
1613        }
1614        munmap(mmap, mmap_size);
1615        return istate->cache_nr;
1616
1617unmap:
1618        munmap(mmap, mmap_size);
1619        die("index file corrupt");
1620}
1621
1622int read_index_from(struct index_state *istate, const char *path)
1623{
1624        struct split_index *split_index;
1625        int ret;
1626
1627        /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1628        if (istate->initialized)
1629                return istate->cache_nr;
1630
1631        ret = do_read_index(istate, path, 0);
1632        split_index = istate->split_index;
1633        if (!split_index)
1634                return ret;
1635
1636        if (is_null_sha1(split_index->base_sha1))
1637                return ret;
1638
1639        if (split_index->base)
1640                discard_index(split_index->base);
1641        else
1642                split_index->base = xcalloc(1, sizeof(*split_index->base));
1643        ret = do_read_index(split_index->base,
1644                            git_path("sharedindex.%s",
1645                                     sha1_to_hex(split_index->base_sha1)), 1);
1646        if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1647                die("broken index, expect %s in %s, got %s",
1648                    sha1_to_hex(split_index->base_sha1),
1649                    git_path("sharedindex.%s",
1650                                     sha1_to_hex(split_index->base_sha1)),
1651                    sha1_to_hex(split_index->base->sha1));
1652        merge_base_index(istate);
1653        return ret;
1654}
1655
1656int is_index_unborn(struct index_state *istate)
1657{
1658        return (!istate->cache_nr && !istate->timestamp.sec);
1659}
1660
1661int discard_index(struct index_state *istate)
1662{
1663        int i;
1664
1665        for (i = 0; i < istate->cache_nr; i++) {
1666                if (istate->cache[i]->index &&
1667                    istate->split_index &&
1668                    istate->split_index->base &&
1669                    istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1670                    istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1671                        continue;
1672                free(istate->cache[i]);
1673        }
1674        resolve_undo_clear_index(istate);
1675        istate->cache_nr = 0;
1676        istate->cache_changed = 0;
1677        istate->timestamp.sec = 0;
1678        istate->timestamp.nsec = 0;
1679        free_name_hash(istate);
1680        cache_tree_free(&(istate->cache_tree));
1681        istate->initialized = 0;
1682        free(istate->cache);
1683        istate->cache = NULL;
1684        istate->cache_alloc = 0;
1685        discard_split_index(istate);
1686        free_untracked_cache(istate->untracked);
1687        istate->untracked = NULL;
1688        return 0;
1689}
1690
1691int unmerged_index(const struct index_state *istate)
1692{
1693        int i;
1694        for (i = 0; i < istate->cache_nr; i++) {
1695                if (ce_stage(istate->cache[i]))
1696                        return 1;
1697        }
1698        return 0;
1699}
1700
1701#define WRITE_BUFFER_SIZE 8192
1702static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1703static unsigned long write_buffer_len;
1704
1705static int ce_write_flush(git_SHA_CTX *context, int fd)
1706{
1707        unsigned int buffered = write_buffer_len;
1708        if (buffered) {
1709                git_SHA1_Update(context, write_buffer, buffered);
1710                if (write_in_full(fd, write_buffer, buffered) != buffered)
1711                        return -1;
1712                write_buffer_len = 0;
1713        }
1714        return 0;
1715}
1716
1717static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1718{
1719        while (len) {
1720                unsigned int buffered = write_buffer_len;
1721                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1722                if (partial > len)
1723                        partial = len;
1724                memcpy(write_buffer + buffered, data, partial);
1725                buffered += partial;
1726                if (buffered == WRITE_BUFFER_SIZE) {
1727                        write_buffer_len = buffered;
1728                        if (ce_write_flush(context, fd))
1729                                return -1;
1730                        buffered = 0;
1731                }
1732                write_buffer_len = buffered;
1733                len -= partial;
1734                data = (char *) data + partial;
1735        }
1736        return 0;
1737}
1738
1739static int write_index_ext_header(git_SHA_CTX *context, int fd,
1740                                  unsigned int ext, unsigned int sz)
1741{
1742        ext = htonl(ext);
1743        sz = htonl(sz);
1744        return ((ce_write(context, fd, &ext, 4) < 0) ||
1745                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1746}
1747
1748static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1749{
1750        unsigned int left = write_buffer_len;
1751
1752        if (left) {
1753                write_buffer_len = 0;
1754                git_SHA1_Update(context, write_buffer, left);
1755        }
1756
1757        /* Flush first if not enough space for SHA1 signature */
1758        if (left + 20 > WRITE_BUFFER_SIZE) {
1759                if (write_in_full(fd, write_buffer, left) != left)
1760                        return -1;
1761                left = 0;
1762        }
1763
1764        /* Append the SHA1 signature at the end */
1765        git_SHA1_Final(write_buffer + left, context);
1766        hashcpy(sha1, write_buffer + left);
1767        left += 20;
1768        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1769}
1770
1771static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1772{
1773        /*
1774         * The only thing we care about in this function is to smudge the
1775         * falsely clean entry due to touch-update-touch race, so we leave
1776         * everything else as they are.  We are called for entries whose
1777         * ce_stat_data.sd_mtime match the index file mtime.
1778         *
1779         * Note that this actually does not do much for gitlinks, for
1780         * which ce_match_stat_basic() always goes to the actual
1781         * contents.  The caller checks with is_racy_timestamp() which
1782         * always says "no" for gitlinks, so we are not called for them ;-)
1783         */
1784        struct stat st;
1785
1786        if (lstat(ce->name, &st) < 0)
1787                return;
1788        if (ce_match_stat_basic(ce, &st))
1789                return;
1790        if (ce_modified_check_fs(ce, &st)) {
1791                /* This is "racily clean"; smudge it.  Note that this
1792                 * is a tricky code.  At first glance, it may appear
1793                 * that it can break with this sequence:
1794                 *
1795                 * $ echo xyzzy >frotz
1796                 * $ git-update-index --add frotz
1797                 * $ : >frotz
1798                 * $ sleep 3
1799                 * $ echo filfre >nitfol
1800                 * $ git-update-index --add nitfol
1801                 *
1802                 * but it does not.  When the second update-index runs,
1803                 * it notices that the entry "frotz" has the same timestamp
1804                 * as index, and if we were to smudge it by resetting its
1805                 * size to zero here, then the object name recorded
1806                 * in index is the 6-byte file but the cached stat information
1807                 * becomes zero --- which would then match what we would
1808                 * obtain from the filesystem next time we stat("frotz").
1809                 *
1810                 * However, the second update-index, before calling
1811                 * this function, notices that the cached size is 6
1812                 * bytes and what is on the filesystem is an empty
1813                 * file, and never calls us, so the cached size information
1814                 * for "frotz" stays 6 which does not match the filesystem.
1815                 */
1816                ce->ce_stat_data.sd_size = 0;
1817        }
1818}
1819
1820/* Copy miscellaneous fields but not the name */
1821static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1822                                       struct cache_entry *ce)
1823{
1824        short flags;
1825
1826        ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1827        ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1828        ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1829        ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1830        ondisk->dev  = htonl(ce->ce_stat_data.sd_dev);
1831        ondisk->ino  = htonl(ce->ce_stat_data.sd_ino);
1832        ondisk->mode = htonl(ce->ce_mode);
1833        ondisk->uid  = htonl(ce->ce_stat_data.sd_uid);
1834        ondisk->gid  = htonl(ce->ce_stat_data.sd_gid);
1835        ondisk->size = htonl(ce->ce_stat_data.sd_size);
1836        hashcpy(ondisk->sha1, ce->sha1);
1837
1838        flags = ce->ce_flags & ~CE_NAMEMASK;
1839        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1840        ondisk->flags = htons(flags);
1841        if (ce->ce_flags & CE_EXTENDED) {
1842                struct ondisk_cache_entry_extended *ondisk2;
1843                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1844                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1845                return ondisk2->name;
1846        }
1847        else {
1848                return ondisk->name;
1849        }
1850}
1851
1852static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1853                          struct strbuf *previous_name)
1854{
1855        int size;
1856        struct ondisk_cache_entry *ondisk;
1857        int saved_namelen = saved_namelen; /* compiler workaround */
1858        char *name;
1859        int result;
1860
1861        if (ce->ce_flags & CE_STRIP_NAME) {
1862                saved_namelen = ce_namelen(ce);
1863                ce->ce_namelen = 0;
1864        }
1865
1866        if (!previous_name) {
1867                size = ondisk_ce_size(ce);
1868                ondisk = xcalloc(1, size);
1869                name = copy_cache_entry_to_ondisk(ondisk, ce);
1870                memcpy(name, ce->name, ce_namelen(ce));
1871        } else {
1872                int common, to_remove, prefix_size;
1873                unsigned char to_remove_vi[16];
1874                for (common = 0;
1875                     (ce->name[common] &&
1876                      common < previous_name->len &&
1877                      ce->name[common] == previous_name->buf[common]);
1878                     common++)
1879                        ; /* still matching */
1880                to_remove = previous_name->len - common;
1881                prefix_size = encode_varint(to_remove, to_remove_vi);
1882
1883                if (ce->ce_flags & CE_EXTENDED)
1884                        size = offsetof(struct ondisk_cache_entry_extended, name);
1885                else
1886                        size = offsetof(struct ondisk_cache_entry, name);
1887                size += prefix_size + (ce_namelen(ce) - common + 1);
1888
1889                ondisk = xcalloc(1, size);
1890                name = copy_cache_entry_to_ondisk(ondisk, ce);
1891                memcpy(name, to_remove_vi, prefix_size);
1892                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1893
1894                strbuf_splice(previous_name, common, to_remove,
1895                              ce->name + common, ce_namelen(ce) - common);
1896        }
1897        if (ce->ce_flags & CE_STRIP_NAME) {
1898                ce->ce_namelen = saved_namelen;
1899                ce->ce_flags &= ~CE_STRIP_NAME;
1900        }
1901
1902        result = ce_write(c, fd, ondisk, size);
1903        free(ondisk);
1904        return result;
1905}
1906
1907/*
1908 * This function verifies if index_state has the correct sha1 of the
1909 * index file.  Don't die if we have any other failure, just return 0.
1910 */
1911static int verify_index_from(const struct index_state *istate, const char *path)
1912{
1913        int fd;
1914        ssize_t n;
1915        struct stat st;
1916        unsigned char sha1[20];
1917
1918        if (!istate->initialized)
1919                return 0;
1920
1921        fd = open(path, O_RDONLY);
1922        if (fd < 0)
1923                return 0;
1924
1925        if (fstat(fd, &st))
1926                goto out;
1927
1928        if (st.st_size < sizeof(struct cache_header) + 20)
1929                goto out;
1930
1931        n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1932        if (n != 20)
1933                goto out;
1934
1935        if (hashcmp(istate->sha1, sha1))
1936                goto out;
1937
1938        close(fd);
1939        return 1;
1940
1941out:
1942        close(fd);
1943        return 0;
1944}
1945
1946static int verify_index(const struct index_state *istate)
1947{
1948        return verify_index_from(istate, get_index_file());
1949}
1950
1951static int has_racy_timestamp(struct index_state *istate)
1952{
1953        int entries = istate->cache_nr;
1954        int i;
1955
1956        for (i = 0; i < entries; i++) {
1957                struct cache_entry *ce = istate->cache[i];
1958                if (is_racy_timestamp(istate, ce))
1959                        return 1;
1960        }
1961        return 0;
1962}
1963
1964/*
1965 * Opportunistically update the index but do not complain if we can't
1966 */
1967void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1968{
1969        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1970            verify_index(istate) &&
1971            write_locked_index(istate, lockfile, COMMIT_LOCK))
1972                rollback_lock_file(lockfile);
1973}
1974
1975static int do_write_index(struct index_state *istate, int newfd,
1976                          int strip_extensions)
1977{
1978        git_SHA_CTX c;
1979        struct cache_header hdr;
1980        int i, err, removed, extended, hdr_version;
1981        struct cache_entry **cache = istate->cache;
1982        int entries = istate->cache_nr;
1983        struct stat st;
1984        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1985
1986        for (i = removed = extended = 0; i < entries; i++) {
1987                if (cache[i]->ce_flags & CE_REMOVE)
1988                        removed++;
1989
1990                /* reduce extended entries if possible */
1991                cache[i]->ce_flags &= ~CE_EXTENDED;
1992                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1993                        extended++;
1994                        cache[i]->ce_flags |= CE_EXTENDED;
1995                }
1996        }
1997
1998        if (!istate->version) {
1999                istate->version = get_index_format_default();
2000                if (getenv("GIT_TEST_SPLIT_INDEX"))
2001                        init_split_index(istate);
2002        }
2003
2004        /* demote version 3 to version 2 when the latter suffices */
2005        if (istate->version == 3 || istate->version == 2)
2006                istate->version = extended ? 3 : 2;
2007
2008        hdr_version = istate->version;
2009
2010        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2011        hdr.hdr_version = htonl(hdr_version);
2012        hdr.hdr_entries = htonl(entries - removed);
2013
2014        git_SHA1_Init(&c);
2015        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2016                return -1;
2017
2018        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2019        for (i = 0; i < entries; i++) {
2020                struct cache_entry *ce = cache[i];
2021                if (ce->ce_flags & CE_REMOVE)
2022                        continue;
2023                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2024                        ce_smudge_racily_clean_entry(ce);
2025                if (is_null_sha1(ce->sha1)) {
2026                        static const char msg[] = "cache entry has null sha1: %s";
2027                        static int allow = -1;
2028
2029                        if (allow < 0)
2030                                allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2031                        if (allow)
2032                                warning(msg, ce->name);
2033                        else
2034                                return error(msg, ce->name);
2035                }
2036                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2037                        return -1;
2038        }
2039        strbuf_release(&previous_name_buf);
2040
2041        /* Write extension data here */
2042        if (!strip_extensions && istate->split_index) {
2043                struct strbuf sb = STRBUF_INIT;
2044
2045                err = write_link_extension(&sb, istate) < 0 ||
2046                        write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2047                                               sb.len) < 0 ||
2048                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2049                strbuf_release(&sb);
2050                if (err)
2051                        return -1;
2052        }
2053        if (!strip_extensions && istate->cache_tree) {
2054                struct strbuf sb = STRBUF_INIT;
2055
2056                cache_tree_write(&sb, istate->cache_tree);
2057                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2058                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2059                strbuf_release(&sb);
2060                if (err)
2061                        return -1;
2062        }
2063        if (!strip_extensions && istate->resolve_undo) {
2064                struct strbuf sb = STRBUF_INIT;
2065
2066                resolve_undo_write(&sb, istate->resolve_undo);
2067                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2068                                             sb.len) < 0
2069                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2070                strbuf_release(&sb);
2071                if (err)
2072                        return -1;
2073        }
2074        if (!strip_extensions && istate->untracked) {
2075                struct strbuf sb = STRBUF_INIT;
2076
2077                write_untracked_extension(&sb, istate->untracked);
2078                err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2079                                             sb.len) < 0 ||
2080                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2081                strbuf_release(&sb);
2082                if (err)
2083                        return -1;
2084        }
2085
2086        if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2087                return -1;
2088        istate->timestamp.sec = (unsigned int)st.st_mtime;
2089        istate->timestamp.nsec = ST_MTIME_NSEC(st);
2090        return 0;
2091}
2092
2093void set_alternate_index_output(const char *name)
2094{
2095        alternate_index_output = name;
2096}
2097
2098static int commit_locked_index(struct lock_file *lk)
2099{
2100        if (alternate_index_output)
2101                return commit_lock_file_to(lk, alternate_index_output);
2102        else
2103                return commit_lock_file(lk);
2104}
2105
2106static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2107                                 unsigned flags)
2108{
2109        int ret = do_write_index(istate, lock->fd, 0);
2110        if (ret)
2111                return ret;
2112        assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2113               (COMMIT_LOCK | CLOSE_LOCK));
2114        if (flags & COMMIT_LOCK)
2115                return commit_locked_index(lock);
2116        else if (flags & CLOSE_LOCK)
2117                return close_lock_file(lock);
2118        else
2119                return ret;
2120}
2121
2122static int write_split_index(struct index_state *istate,
2123                             struct lock_file *lock,
2124                             unsigned flags)
2125{
2126        int ret;
2127        prepare_to_write_split_index(istate);
2128        ret = do_write_locked_index(istate, lock, flags);
2129        finish_writing_split_index(istate);
2130        return ret;
2131}
2132
2133static char *temporary_sharedindex;
2134
2135static void remove_temporary_sharedindex(void)
2136{
2137        if (temporary_sharedindex) {
2138                unlink_or_warn(temporary_sharedindex);
2139                free(temporary_sharedindex);
2140                temporary_sharedindex = NULL;
2141        }
2142}
2143
2144static void remove_temporary_sharedindex_on_signal(int signo)
2145{
2146        remove_temporary_sharedindex();
2147        sigchain_pop(signo);
2148        raise(signo);
2149}
2150
2151static int write_shared_index(struct index_state *istate,
2152                              struct lock_file *lock, unsigned flags)
2153{
2154        struct split_index *si = istate->split_index;
2155        static int installed_handler;
2156        int fd, ret;
2157
2158        temporary_sharedindex = git_pathdup("sharedindex_XXXXXX");
2159        fd = mkstemp(temporary_sharedindex);
2160        if (fd < 0) {
2161                free(temporary_sharedindex);
2162                temporary_sharedindex = NULL;
2163                hashclr(si->base_sha1);
2164                return do_write_locked_index(istate, lock, flags);
2165        }
2166        if (!installed_handler) {
2167                atexit(remove_temporary_sharedindex);
2168                sigchain_push_common(remove_temporary_sharedindex_on_signal);
2169        }
2170        move_cache_to_base_index(istate);
2171        ret = do_write_index(si->base, fd, 1);
2172        close(fd);
2173        if (ret) {
2174                remove_temporary_sharedindex();
2175                return ret;
2176        }
2177        ret = rename(temporary_sharedindex,
2178                     git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2179        free(temporary_sharedindex);
2180        temporary_sharedindex = NULL;
2181        if (!ret)
2182                hashcpy(si->base_sha1, si->base->sha1);
2183        return ret;
2184}
2185
2186int write_locked_index(struct index_state *istate, struct lock_file *lock,
2187                       unsigned flags)
2188{
2189        struct split_index *si = istate->split_index;
2190
2191        if (!si || alternate_index_output ||
2192            (istate->cache_changed & ~EXTMASK)) {
2193                if (si)
2194                        hashclr(si->base_sha1);
2195                return do_write_locked_index(istate, lock, flags);
2196        }
2197
2198        if (getenv("GIT_TEST_SPLIT_INDEX")) {
2199                int v = si->base_sha1[0];
2200                if ((v & 15) < 6)
2201                        istate->cache_changed |= SPLIT_INDEX_ORDERED;
2202        }
2203        if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2204                int ret = write_shared_index(istate, lock, flags);
2205                if (ret)
2206                        return ret;
2207        }
2208
2209        return write_split_index(istate, lock, flags);
2210}
2211
2212/*
2213 * Read the index file that is potentially unmerged into given
2214 * index_state, dropping any unmerged entries.  Returns true if
2215 * the index is unmerged.  Callers who want to refuse to work
2216 * from an unmerged state can call this and check its return value,
2217 * instead of calling read_cache().
2218 */
2219int read_index_unmerged(struct index_state *istate)
2220{
2221        int i;
2222        int unmerged = 0;
2223
2224        read_index(istate);
2225        for (i = 0; i < istate->cache_nr; i++) {
2226                struct cache_entry *ce = istate->cache[i];
2227                struct cache_entry *new_ce;
2228                int size, len;
2229
2230                if (!ce_stage(ce))
2231                        continue;
2232                unmerged = 1;
2233                len = ce_namelen(ce);
2234                size = cache_entry_size(len);
2235                new_ce = xcalloc(1, size);
2236                memcpy(new_ce->name, ce->name, len);
2237                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2238                new_ce->ce_namelen = len;
2239                new_ce->ce_mode = ce->ce_mode;
2240                if (add_index_entry(istate, new_ce, 0))
2241                        return error("%s: cannot drop to stage #0",
2242                                     new_ce->name);
2243        }
2244        return unmerged;
2245}
2246
2247/*
2248 * Returns 1 if the path is an "other" path with respect to
2249 * the index; that is, the path is not mentioned in the index at all,
2250 * either as a file, a directory with some files in the index,
2251 * or as an unmerged entry.
2252 *
2253 * We helpfully remove a trailing "/" from directories so that
2254 * the output of read_directory can be used as-is.
2255 */
2256int index_name_is_other(const struct index_state *istate, const char *name,
2257                int namelen)
2258{
2259        int pos;
2260        if (namelen && name[namelen - 1] == '/')
2261                namelen--;
2262        pos = index_name_pos(istate, name, namelen);
2263        if (0 <= pos)
2264                return 0;       /* exact match */
2265        pos = -pos - 1;
2266        if (pos < istate->cache_nr) {
2267                struct cache_entry *ce = istate->cache[pos];
2268                if (ce_namelen(ce) == namelen &&
2269                    !memcmp(ce->name, name, namelen))
2270                        return 0; /* Yup, this one exists unmerged */
2271        }
2272        return 1;
2273}
2274
2275void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2276{
2277        int pos, len;
2278        unsigned long sz;
2279        enum object_type type;
2280        void *data;
2281
2282        len = strlen(path);
2283        pos = index_name_pos(istate, path, len);
2284        if (pos < 0) {
2285                /*
2286                 * We might be in the middle of a merge, in which
2287                 * case we would read stage #2 (ours).
2288                 */
2289                int i;
2290                for (i = -pos - 1;
2291                     (pos < 0 && i < istate->cache_nr &&
2292                      !strcmp(istate->cache[i]->name, path));
2293                     i++)
2294                        if (ce_stage(istate->cache[i]) == 2)
2295                                pos = i;
2296        }
2297        if (pos < 0)
2298                return NULL;
2299        data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2300        if (!data || type != OBJ_BLOB) {
2301                free(data);
2302                return NULL;
2303        }
2304        if (size)
2305                *size = sz;
2306        return data;
2307}
2308
2309void stat_validity_clear(struct stat_validity *sv)
2310{
2311        free(sv->sd);
2312        sv->sd = NULL;
2313}
2314
2315int stat_validity_check(struct stat_validity *sv, const char *path)
2316{
2317        struct stat st;
2318
2319        if (stat(path, &st) < 0)
2320                return sv->sd == NULL;
2321        if (!sv->sd)
2322                return 0;
2323        return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2324}
2325
2326void stat_validity_update(struct stat_validity *sv, int fd)
2327{
2328        struct stat st;
2329
2330        if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2331                stat_validity_clear(sv);
2332        else {
2333                if (!sv->sd)
2334                        sv->sd = xcalloc(1, sizeof(struct stat_data));
2335                fill_stat_data(sv->sd, &st);
2336        }
2337}