0d96c753b74d450bf97e1471a58bdd9716a5206a
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 */
   6#define NO_THE_INDEX_COMPATIBILITY_MACROS
   7#include "cache.h"
   8#include "lockfile.h"
   9#include "cache-tree.h"
  10#include "refs.h"
  11#include "dir.h"
  12#include "tree.h"
  13#include "commit.h"
  14#include "blob.h"
  15#include "resolve-undo.h"
  16#include "strbuf.h"
  17#include "varint.h"
  18#include "split-index.h"
  19#include "sigchain.h"
  20#include "utf8.h"
  21
  22static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
  23                                               unsigned int options);
  24
  25/* Mask for the name length in ce_flags in the on-disk index */
  26
  27#define CE_NAMEMASK  (0x0fff)
  28
  29/* Index extensions.
  30 *
  31 * The first letter should be 'A'..'Z' for extensions that are not
  32 * necessary for a correct operation (i.e. optimization data).
  33 * When new extensions are added that _needs_ to be understood in
  34 * order to correctly interpret the index file, pick character that
  35 * is outside the range, to cause the reader to abort.
  36 */
  37
  38#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  39#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  40#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  41#define CACHE_EXT_LINK 0x6c696e6b         /* "link" */
  42#define CACHE_EXT_UNTRACKED 0x554E5452    /* "UNTR" */
  43
  44/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
  45#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
  46                 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
  47                 SPLIT_INDEX_ORDERED)
  48
  49struct index_state the_index;
  50static const char *alternate_index_output;
  51
  52static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  53{
  54        istate->cache[nr] = ce;
  55        add_name_hash(istate, ce);
  56}
  57
  58static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  59{
  60        struct cache_entry *old = istate->cache[nr];
  61
  62        replace_index_entry_in_base(istate, old, ce);
  63        remove_name_hash(istate, old);
  64        free(old);
  65        set_index_entry(istate, nr, ce);
  66        ce->ce_flags |= CE_UPDATE_IN_BASE;
  67        istate->cache_changed |= CE_ENTRY_CHANGED;
  68}
  69
  70void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  71{
  72        struct cache_entry *old = istate->cache[nr], *new;
  73        int namelen = strlen(new_name);
  74
  75        new = xmalloc(cache_entry_size(namelen));
  76        copy_cache_entry(new, old);
  77        new->ce_flags &= ~CE_HASHED;
  78        new->ce_namelen = namelen;
  79        new->index = 0;
  80        memcpy(new->name, new_name, namelen + 1);
  81
  82        cache_tree_invalidate_path(istate, old->name);
  83        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_timestamp(const struct index_state *istate,
 276                             const struct cache_entry *ce)
 277{
 278        return (!S_ISGITLINK(ce->ce_mode) &&
 279                istate->timestamp.sec &&
 280#ifdef USE_NSEC
 281                 /* nanosecond timestamped files can also be racy! */
 282                (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
 283                 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
 284                  istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
 285#else
 286                istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
 287#endif
 288                 );
 289}
 290
 291int ie_match_stat(const struct index_state *istate,
 292                  const struct cache_entry *ce, struct stat *st,
 293                  unsigned int options)
 294{
 295        unsigned int changed;
 296        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 297        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 298        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 299
 300        /*
 301         * If it's marked as always valid in the index, it's
 302         * valid whatever the checked-out copy says.
 303         *
 304         * skip-worktree has the same effect with higher precedence
 305         */
 306        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 307                return 0;
 308        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 309                return 0;
 310
 311        /*
 312         * Intent-to-add entries have not been added, so the index entry
 313         * by definition never matches what is in the work tree until it
 314         * actually gets added.
 315         */
 316        if (ce->ce_flags & CE_INTENT_TO_ADD)
 317                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 318
 319        changed = ce_match_stat_basic(ce, st);
 320
 321        /*
 322         * Within 1 second of this sequence:
 323         *      echo xyzzy >file && git-update-index --add file
 324         * running this command:
 325         *      echo frotz >file
 326         * would give a falsely clean cache entry.  The mtime and
 327         * length match the cache, and other stat fields do not change.
 328         *
 329         * We could detect this at update-index time (the cache entry
 330         * being registered/updated records the same time as "now")
 331         * and delay the return from git-update-index, but that would
 332         * effectively mean we can make at most one commit per second,
 333         * which is not acceptable.  Instead, we check cache entries
 334         * whose mtime are the same as the index file timestamp more
 335         * carefully than others.
 336         */
 337        if (!changed && is_racy_timestamp(istate, ce)) {
 338                if (assume_racy_is_modified)
 339                        changed |= DATA_CHANGED;
 340                else
 341                        changed |= ce_modified_check_fs(ce, st);
 342        }
 343
 344        return changed;
 345}
 346
 347int ie_modified(const struct index_state *istate,
 348                const struct cache_entry *ce,
 349                struct stat *st, unsigned int options)
 350{
 351        int changed, changed_fs;
 352
 353        changed = ie_match_stat(istate, ce, st, options);
 354        if (!changed)
 355                return 0;
 356        /*
 357         * If the mode or type has changed, there's no point in trying
 358         * to refresh the entry - it's not going to match
 359         */
 360        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 361                return changed;
 362
 363        /*
 364         * Immediately after read-tree or update-index --cacheinfo,
 365         * the length field is zero, as we have never even read the
 366         * lstat(2) information once, and we cannot trust DATA_CHANGED
 367         * returned by ie_match_stat() which in turn was returned by
 368         * ce_match_stat_basic() to signal that the filesize of the
 369         * blob changed.  We have to actually go to the filesystem to
 370         * see if the contents match, and if so, should answer "unchanged".
 371         *
 372         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 373         * already has checked the actual HEAD from the filesystem in the
 374         * subproject.  If ie_match_stat() already said it is different,
 375         * then we know it is.
 376         */
 377        if ((changed & DATA_CHANGED) &&
 378            (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
 379                return changed;
 380
 381        changed_fs = ce_modified_check_fs(ce, st);
 382        if (changed_fs)
 383                return changed | changed_fs;
 384        return 0;
 385}
 386
 387int base_name_compare(const char *name1, int len1, int mode1,
 388                      const char *name2, int len2, int mode2)
 389{
 390        unsigned char c1, c2;
 391        int len = len1 < len2 ? len1 : len2;
 392        int cmp;
 393
 394        cmp = memcmp(name1, name2, len);
 395        if (cmp)
 396                return cmp;
 397        c1 = name1[len];
 398        c2 = name2[len];
 399        if (!c1 && S_ISDIR(mode1))
 400                c1 = '/';
 401        if (!c2 && S_ISDIR(mode2))
 402                c2 = '/';
 403        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 404}
 405
 406/*
 407 * df_name_compare() is identical to base_name_compare(), except it
 408 * compares conflicting directory/file entries as equal. Note that
 409 * while a directory name compares as equal to a regular file, they
 410 * then individually compare _differently_ to a filename that has
 411 * a dot after the basename (because '\0' < '.' < '/').
 412 *
 413 * This is used by routines that want to traverse the git namespace
 414 * but then handle conflicting entries together when possible.
 415 */
 416int df_name_compare(const char *name1, int len1, int mode1,
 417                    const char *name2, int len2, int mode2)
 418{
 419        int len = len1 < len2 ? len1 : len2, cmp;
 420        unsigned char c1, c2;
 421
 422        cmp = memcmp(name1, name2, len);
 423        if (cmp)
 424                return cmp;
 425        /* Directories and files compare equal (same length, same name) */
 426        if (len1 == len2)
 427                return 0;
 428        c1 = name1[len];
 429        if (!c1 && S_ISDIR(mode1))
 430                c1 = '/';
 431        c2 = name2[len];
 432        if (!c2 && S_ISDIR(mode2))
 433                c2 = '/';
 434        if (c1 == '/' && !c2)
 435                return 0;
 436        if (c2 == '/' && !c1)
 437                return 0;
 438        return c1 - c2;
 439}
 440
 441int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
 442{
 443        size_t min_len = (len1 < len2) ? len1 : len2;
 444        int cmp = memcmp(name1, name2, min_len);
 445        if (cmp)
 446                return cmp;
 447        if (len1 < len2)
 448                return -1;
 449        if (len1 > len2)
 450                return 1;
 451        return 0;
 452}
 453
 454int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 455{
 456        int cmp;
 457
 458        cmp = name_compare(name1, len1, name2, len2);
 459        if (cmp)
 460                return cmp;
 461
 462        if (stage1 < stage2)
 463                return -1;
 464        if (stage1 > stage2)
 465                return 1;
 466        return 0;
 467}
 468
 469static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 470{
 471        int first, last;
 472
 473        first = 0;
 474        last = istate->cache_nr;
 475        while (last > first) {
 476                int next = (last + first) >> 1;
 477                struct cache_entry *ce = istate->cache[next];
 478                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 479                if (!cmp)
 480                        return next;
 481                if (cmp < 0) {
 482                        last = next;
 483                        continue;
 484                }
 485                first = next+1;
 486        }
 487        return -first-1;
 488}
 489
 490int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 491{
 492        return index_name_stage_pos(istate, name, namelen, 0);
 493}
 494
 495/* Remove entry, return true if there are more entries to go.. */
 496int remove_index_entry_at(struct index_state *istate, int pos)
 497{
 498        struct cache_entry *ce = istate->cache[pos];
 499
 500        record_resolve_undo(istate, ce);
 501        remove_name_hash(istate, ce);
 502        save_or_free_index_entry(istate, ce);
 503        istate->cache_changed |= CE_ENTRY_REMOVED;
 504        istate->cache_nr--;
 505        if (pos >= istate->cache_nr)
 506                return 0;
 507        memmove(istate->cache + pos,
 508                istate->cache + pos + 1,
 509                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 510        return 1;
 511}
 512
 513/*
 514 * Remove all cache entries marked for removal, that is where
 515 * CE_REMOVE is set in ce_flags.  This is much more effective than
 516 * calling remove_index_entry_at() for each entry to be removed.
 517 */
 518void remove_marked_cache_entries(struct index_state *istate)
 519{
 520        struct cache_entry **ce_array = istate->cache;
 521        unsigned int i, j;
 522
 523        for (i = j = 0; i < istate->cache_nr; i++) {
 524                if (ce_array[i]->ce_flags & CE_REMOVE) {
 525                        remove_name_hash(istate, ce_array[i]);
 526                        save_or_free_index_entry(istate, ce_array[i]);
 527                }
 528                else
 529                        ce_array[j++] = ce_array[i];
 530        }
 531        if (j == istate->cache_nr)
 532                return;
 533        istate->cache_changed |= CE_ENTRY_REMOVED;
 534        istate->cache_nr = j;
 535}
 536
 537int remove_file_from_index(struct index_state *istate, const char *path)
 538{
 539        int pos = index_name_pos(istate, path, strlen(path));
 540        if (pos < 0)
 541                pos = -pos-1;
 542        cache_tree_invalidate_path(istate, path);
 543        untracked_cache_remove_from_index(istate, path);
 544        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 545                remove_index_entry_at(istate, pos);
 546        return 0;
 547}
 548
 549static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 550{
 551        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 552}
 553
 554static int index_name_pos_also_unmerged(struct index_state *istate,
 555        const char *path, int namelen)
 556{
 557        int pos = index_name_pos(istate, path, namelen);
 558        struct cache_entry *ce;
 559
 560        if (pos >= 0)
 561                return pos;
 562
 563        /* maybe unmerged? */
 564        pos = -1 - pos;
 565        if (pos >= istate->cache_nr ||
 566                        compare_name((ce = istate->cache[pos]), path, namelen))
 567                return -1;
 568
 569        /* order of preference: stage 2, 1, 3 */
 570        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 571                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 572                        !compare_name(ce, path, namelen))
 573                pos++;
 574        return pos;
 575}
 576
 577static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 578{
 579        int len = ce_namelen(ce);
 580        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 581}
 582
 583/*
 584 * If we add a filename that aliases in the cache, we will use the
 585 * name that we already have - but we don't want to update the same
 586 * alias twice, because that implies that there were actually two
 587 * different files with aliasing names!
 588 *
 589 * So we use the CE_ADDED flag to verify that the alias was an old
 590 * one before we accept it as
 591 */
 592static struct cache_entry *create_alias_ce(struct index_state *istate,
 593                                           struct cache_entry *ce,
 594                                           struct cache_entry *alias)
 595{
 596        int len;
 597        struct cache_entry *new;
 598
 599        if (alias->ce_flags & CE_ADDED)
 600                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 601
 602        /* Ok, create the new entry using the name of the existing alias */
 603        len = ce_namelen(alias);
 604        new = xcalloc(1, cache_entry_size(len));
 605        memcpy(new->name, alias->name, len);
 606        copy_cache_entry(new, ce);
 607        save_or_free_index_entry(istate, ce);
 608        return new;
 609}
 610
 611void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
 612{
 613        unsigned char sha1[20];
 614        if (write_sha1_file("", 0, blob_type, sha1))
 615                die("cannot create an empty blob in the object database");
 616        hashcpy(ce->sha1, sha1);
 617}
 618
 619int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 620{
 621        int size, namelen, was_same;
 622        mode_t st_mode = st->st_mode;
 623        struct cache_entry *ce, *alias;
 624        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 625        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 626        int pretend = flags & ADD_CACHE_PRETEND;
 627        int intent_only = flags & ADD_CACHE_INTENT;
 628        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 629                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 630
 631        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 632                return error("%s: can only add regular files, symbolic links or git-directories", path);
 633
 634        namelen = strlen(path);
 635        if (S_ISDIR(st_mode)) {
 636                while (namelen && path[namelen-1] == '/')
 637                        namelen--;
 638        }
 639        size = cache_entry_size(namelen);
 640        ce = xcalloc(1, size);
 641        memcpy(ce->name, path, namelen);
 642        ce->ce_namelen = namelen;
 643        if (!intent_only)
 644                fill_stat_cache_info(ce, st);
 645        else
 646                ce->ce_flags |= CE_INTENT_TO_ADD;
 647
 648        if (trust_executable_bit && has_symlinks)
 649                ce->ce_mode = create_ce_mode(st_mode);
 650        else {
 651                /* If there is an existing entry, pick the mode bits and type
 652                 * from it, otherwise assume unexecutable regular file.
 653                 */
 654                struct cache_entry *ent;
 655                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 656
 657                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 658                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 659        }
 660
 661        /* When core.ignorecase=true, determine if a directory of the same name but differing
 662         * case already exists within the Git repository.  If it does, ensure the directory
 663         * case of the file being added to the repository matches (is folded into) the existing
 664         * entry's directory case.
 665         */
 666        if (ignore_case) {
 667                const char *startPtr = ce->name;
 668                const char *ptr = startPtr;
 669                while (*ptr) {
 670                        while (*ptr && *ptr != '/')
 671                                ++ptr;
 672                        if (*ptr == '/') {
 673                                struct cache_entry *foundce;
 674                                ++ptr;
 675                                foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
 676                                if (foundce) {
 677                                        memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
 678                                        startPtr = ptr;
 679                                }
 680                        }
 681                }
 682        }
 683
 684        alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 685        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 686                /* Nothing changed, really */
 687                free(ce);
 688                if (!S_ISGITLINK(alias->ce_mode))
 689                        ce_mark_uptodate(alias);
 690                alias->ce_flags |= CE_ADDED;
 691                return 0;
 692        }
 693        if (!intent_only) {
 694                if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
 695                        return error("unable to index file %s", path);
 696        } else
 697                set_object_name_for_intent_to_add_entry(ce);
 698
 699        if (ignore_case && alias && different_name(ce, alias))
 700                ce = create_alias_ce(istate, ce, alias);
 701        ce->ce_flags |= CE_ADDED;
 702
 703        /* It was suspected to be racily clean, but it turns out to be Ok */
 704        was_same = (alias &&
 705                    !ce_stage(alias) &&
 706                    !hashcmp(alias->sha1, ce->sha1) &&
 707                    ce->ce_mode == alias->ce_mode);
 708
 709        if (pretend)
 710                ;
 711        else if (add_index_entry(istate, ce, add_option))
 712                return error("unable to add %s to index",path);
 713        if (verbose && !was_same)
 714                printf("add '%s'\n", path);
 715        return 0;
 716}
 717
 718int add_file_to_index(struct index_state *istate, const char *path, int flags)
 719{
 720        struct stat st;
 721        if (lstat(path, &st))
 722                die_errno("unable to stat '%s'", path);
 723        return add_to_index(istate, path, &st, flags);
 724}
 725
 726struct cache_entry *make_cache_entry(unsigned int mode,
 727                const unsigned char *sha1, const char *path, int stage,
 728                unsigned int refresh_options)
 729{
 730        int size, len;
 731        struct cache_entry *ce, *ret;
 732
 733        if (!verify_path(path)) {
 734                error("Invalid path '%s'", path);
 735                return NULL;
 736        }
 737
 738        len = strlen(path);
 739        size = cache_entry_size(len);
 740        ce = xcalloc(1, size);
 741
 742        hashcpy(ce->sha1, sha1);
 743        memcpy(ce->name, path, len);
 744        ce->ce_flags = create_ce_flags(stage);
 745        ce->ce_namelen = len;
 746        ce->ce_mode = create_ce_mode(mode);
 747
 748        ret = refresh_cache_entry(ce, refresh_options);
 749        if (!ret) {
 750                free(ce);
 751                return NULL;
 752        } else {
 753                return ret;
 754        }
 755}
 756
 757int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
 758{
 759        int len = ce_namelen(a);
 760        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 761}
 762
 763/*
 764 * We fundamentally don't like some paths: we don't want
 765 * dot or dot-dot anywhere, and for obvious reasons don't
 766 * want to recurse into ".git" either.
 767 *
 768 * Also, we don't want double slashes or slashes at the
 769 * end that can make pathnames ambiguous.
 770 */
 771static int verify_dotfile(const char *rest)
 772{
 773        /*
 774         * The first character was '.', but that
 775         * has already been discarded, we now test
 776         * the rest.
 777         */
 778
 779        /* "." is not allowed */
 780        if (*rest == '\0' || is_dir_sep(*rest))
 781                return 0;
 782
 783        switch (*rest) {
 784        /*
 785         * ".git" followed by  NUL or slash is bad. This
 786         * shares the path end test with the ".." case.
 787         */
 788        case 'g':
 789        case 'G':
 790                if (rest[1] != 'i' && rest[1] != 'I')
 791                        break;
 792                if (rest[2] != 't' && rest[2] != 'T')
 793                        break;
 794                rest += 2;
 795        /* fallthrough */
 796        case '.':
 797                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 798                        return 0;
 799        }
 800        return 1;
 801}
 802
 803int verify_path(const char *path)
 804{
 805        char c;
 806
 807        if (has_dos_drive_prefix(path))
 808                return 0;
 809
 810        goto inside;
 811        for (;;) {
 812                if (!c)
 813                        return 1;
 814                if (is_dir_sep(c)) {
 815inside:
 816                        if (protect_hfs && is_hfs_dotgit(path))
 817                                return 0;
 818                        if (protect_ntfs && is_ntfs_dotgit(path))
 819                                return 0;
 820                        c = *path++;
 821                        if ((c == '.' && !verify_dotfile(path)) ||
 822                            is_dir_sep(c) || c == '\0')
 823                                return 0;
 824                }
 825                c = *path++;
 826        }
 827}
 828
 829/*
 830 * Do we have another file that has the beginning components being a
 831 * proper superset of the name we're trying to add?
 832 */
 833static int has_file_name(struct index_state *istate,
 834                         const struct cache_entry *ce, int pos, int ok_to_replace)
 835{
 836        int retval = 0;
 837        int len = ce_namelen(ce);
 838        int stage = ce_stage(ce);
 839        const char *name = ce->name;
 840
 841        while (pos < istate->cache_nr) {
 842                struct cache_entry *p = istate->cache[pos++];
 843
 844                if (len >= ce_namelen(p))
 845                        break;
 846                if (memcmp(name, p->name, len))
 847                        break;
 848                if (ce_stage(p) != stage)
 849                        continue;
 850                if (p->name[len] != '/')
 851                        continue;
 852                if (p->ce_flags & CE_REMOVE)
 853                        continue;
 854                retval = -1;
 855                if (!ok_to_replace)
 856                        break;
 857                remove_index_entry_at(istate, --pos);
 858        }
 859        return retval;
 860}
 861
 862/*
 863 * Do we have another file with a pathname that is a proper
 864 * subset of the name we're trying to add?
 865 */
 866static int has_dir_name(struct index_state *istate,
 867                        const struct cache_entry *ce, int pos, int ok_to_replace)
 868{
 869        int retval = 0;
 870        int stage = ce_stage(ce);
 871        const char *name = ce->name;
 872        const char *slash = name + ce_namelen(ce);
 873
 874        for (;;) {
 875                int len;
 876
 877                for (;;) {
 878                        if (*--slash == '/')
 879                                break;
 880                        if (slash <= ce->name)
 881                                return retval;
 882                }
 883                len = slash - name;
 884
 885                pos = index_name_stage_pos(istate, name, len, stage);
 886                if (pos >= 0) {
 887                        /*
 888                         * Found one, but not so fast.  This could
 889                         * be a marker that says "I was here, but
 890                         * I am being removed".  Such an entry is
 891                         * not a part of the resulting tree, and
 892                         * it is Ok to have a directory at the same
 893                         * path.
 894                         */
 895                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 896                                retval = -1;
 897                                if (!ok_to_replace)
 898                                        break;
 899                                remove_index_entry_at(istate, pos);
 900                                continue;
 901                        }
 902                }
 903                else
 904                        pos = -pos-1;
 905
 906                /*
 907                 * Trivial optimization: if we find an entry that
 908                 * already matches the sub-directory, then we know
 909                 * we're ok, and we can exit.
 910                 */
 911                while (pos < istate->cache_nr) {
 912                        struct cache_entry *p = istate->cache[pos];
 913                        if ((ce_namelen(p) <= len) ||
 914                            (p->name[len] != '/') ||
 915                            memcmp(p->name, name, len))
 916                                break; /* not our subdirectory */
 917                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 918                                /*
 919                                 * p is at the same stage as our entry, and
 920                                 * is a subdirectory of what we are looking
 921                                 * at, so we cannot have conflicts at our
 922                                 * level or anything shorter.
 923                                 */
 924                                return retval;
 925                        pos++;
 926                }
 927        }
 928        return retval;
 929}
 930
 931/* We may be in a situation where we already have path/file and path
 932 * is being added, or we already have path and path/file is being
 933 * added.  Either one would result in a nonsense tree that has path
 934 * twice when git-write-tree tries to write it out.  Prevent it.
 935 *
 936 * If ok-to-replace is specified, we remove the conflicting entries
 937 * from the cache so the caller should recompute the insert position.
 938 * When this happens, we return non-zero.
 939 */
 940static int check_file_directory_conflict(struct index_state *istate,
 941                                         const struct cache_entry *ce,
 942                                         int pos, int ok_to_replace)
 943{
 944        int retval;
 945
 946        /*
 947         * When ce is an "I am going away" entry, we allow it to be added
 948         */
 949        if (ce->ce_flags & CE_REMOVE)
 950                return 0;
 951
 952        /*
 953         * We check if the path is a sub-path of a subsequent pathname
 954         * first, since removing those will not change the position
 955         * in the array.
 956         */
 957        retval = has_file_name(istate, ce, pos, ok_to_replace);
 958
 959        /*
 960         * Then check if the path might have a clashing sub-directory
 961         * before it.
 962         */
 963        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 964}
 965
 966static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 967{
 968        int pos;
 969        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 970        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 971        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 972        int new_only = option & ADD_CACHE_NEW_ONLY;
 973
 974        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
 975                cache_tree_invalidate_path(istate, ce->name);
 976        pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 977
 978        /* existing match? Just replace it. */
 979        if (pos >= 0) {
 980                if (!new_only)
 981                        replace_index_entry(istate, pos, ce);
 982                return 0;
 983        }
 984        pos = -pos-1;
 985
 986        untracked_cache_add_to_index(istate, ce->name);
 987
 988        /*
 989         * Inserting a merged entry ("stage 0") into the index
 990         * will always replace all non-merged entries..
 991         */
 992        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 993                while (ce_same_name(istate->cache[pos], ce)) {
 994                        ok_to_add = 1;
 995                        if (!remove_index_entry_at(istate, pos))
 996                                break;
 997                }
 998        }
 999
1000        if (!ok_to_add)
1001                return -1;
1002        if (!verify_path(ce->name))
1003                return error("Invalid path '%s'", ce->name);
1004
1005        if (!skip_df_check &&
1006            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1007                if (!ok_to_replace)
1008                        return error("'%s' appears as both a file and as a directory",
1009                                     ce->name);
1010                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1011                pos = -pos-1;
1012        }
1013        return pos + 1;
1014}
1015
1016int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1017{
1018        int pos;
1019
1020        if (option & ADD_CACHE_JUST_APPEND)
1021                pos = istate->cache_nr;
1022        else {
1023                int ret;
1024                ret = add_index_entry_with_check(istate, ce, option);
1025                if (ret <= 0)
1026                        return ret;
1027                pos = ret - 1;
1028        }
1029
1030        /* Make sure the array is big enough .. */
1031        ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1032
1033        /* Add it in.. */
1034        istate->cache_nr++;
1035        if (istate->cache_nr > pos + 1)
1036                memmove(istate->cache + pos + 1,
1037                        istate->cache + pos,
1038                        (istate->cache_nr - pos - 1) * sizeof(ce));
1039        set_index_entry(istate, pos, ce);
1040        istate->cache_changed |= CE_ENTRY_ADDED;
1041        return 0;
1042}
1043
1044/*
1045 * "refresh" does not calculate a new sha1 file or bring the
1046 * cache up-to-date for mode/content changes. But what it
1047 * _does_ do is to "re-match" the stat information of a file
1048 * with the cache, so that you can refresh the cache for a
1049 * file that hasn't been changed but where the stat entry is
1050 * out of date.
1051 *
1052 * For example, you'd want to do this after doing a "git-read-tree",
1053 * to link up the stat cache details with the proper files.
1054 */
1055static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1056                                             struct cache_entry *ce,
1057                                             unsigned int options, int *err,
1058                                             int *changed_ret)
1059{
1060        struct stat st;
1061        struct cache_entry *updated;
1062        int changed, size;
1063        int refresh = options & CE_MATCH_REFRESH;
1064        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1065        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1066        int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1067
1068        if (!refresh || ce_uptodate(ce))
1069                return ce;
1070
1071        /*
1072         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1073         * that the change to the work tree does not matter and told
1074         * us not to worry.
1075         */
1076        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1077                ce_mark_uptodate(ce);
1078                return ce;
1079        }
1080        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1081                ce_mark_uptodate(ce);
1082                return ce;
1083        }
1084
1085        if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1086                if (ignore_missing)
1087                        return ce;
1088                if (err)
1089                        *err = ENOENT;
1090                return NULL;
1091        }
1092
1093        if (lstat(ce->name, &st) < 0) {
1094                if (ignore_missing && errno == ENOENT)
1095                        return ce;
1096                if (err)
1097                        *err = errno;
1098                return NULL;
1099        }
1100
1101        changed = ie_match_stat(istate, ce, &st, options);
1102        if (changed_ret)
1103                *changed_ret = changed;
1104        if (!changed) {
1105                /*
1106                 * The path is unchanged.  If we were told to ignore
1107                 * valid bit, then we did the actual stat check and
1108                 * found that the entry is unmodified.  If the entry
1109                 * is not marked VALID, this is the place to mark it
1110                 * valid again, under "assume unchanged" mode.
1111                 */
1112                if (ignore_valid && assume_unchanged &&
1113                    !(ce->ce_flags & CE_VALID))
1114                        ; /* mark this one VALID again */
1115                else {
1116                        /*
1117                         * We do not mark the index itself "modified"
1118                         * because CE_UPTODATE flag is in-core only;
1119                         * we are not going to write this change out.
1120                         */
1121                        if (!S_ISGITLINK(ce->ce_mode))
1122                                ce_mark_uptodate(ce);
1123                        return ce;
1124                }
1125        }
1126
1127        if (ie_modified(istate, ce, &st, options)) {
1128                if (err)
1129                        *err = EINVAL;
1130                return NULL;
1131        }
1132
1133        size = ce_size(ce);
1134        updated = xmalloc(size);
1135        memcpy(updated, ce, size);
1136        fill_stat_cache_info(updated, &st);
1137        /*
1138         * If ignore_valid is not set, we should leave CE_VALID bit
1139         * alone.  Otherwise, paths marked with --no-assume-unchanged
1140         * (i.e. things to be edited) will reacquire CE_VALID bit
1141         * automatically, which is not really what we want.
1142         */
1143        if (!ignore_valid && assume_unchanged &&
1144            !(ce->ce_flags & CE_VALID))
1145                updated->ce_flags &= ~CE_VALID;
1146
1147        /* istate->cache_changed is updated in the caller */
1148        return updated;
1149}
1150
1151static void show_file(const char * fmt, const char * name, int in_porcelain,
1152                      int * first, const char *header_msg)
1153{
1154        if (in_porcelain && *first && header_msg) {
1155                printf("%s\n", header_msg);
1156                *first = 0;
1157        }
1158        printf(fmt, name);
1159}
1160
1161int refresh_index(struct index_state *istate, unsigned int flags,
1162                  const struct pathspec *pathspec,
1163                  char *seen, const char *header_msg)
1164{
1165        int i;
1166        int has_errors = 0;
1167        int really = (flags & REFRESH_REALLY) != 0;
1168        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1169        int quiet = (flags & REFRESH_QUIET) != 0;
1170        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1171        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1172        int first = 1;
1173        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1174        unsigned int options = (CE_MATCH_REFRESH |
1175                                (really ? CE_MATCH_IGNORE_VALID : 0) |
1176                                (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1177        const char *modified_fmt;
1178        const char *deleted_fmt;
1179        const char *typechange_fmt;
1180        const char *added_fmt;
1181        const char *unmerged_fmt;
1182
1183        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1184        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1185        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1186        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1187        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1188        for (i = 0; i < istate->cache_nr; i++) {
1189                struct cache_entry *ce, *new;
1190                int cache_errno = 0;
1191                int changed = 0;
1192                int filtered = 0;
1193
1194                ce = istate->cache[i];
1195                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1196                        continue;
1197
1198                if (pathspec && !ce_path_match(ce, pathspec, seen))
1199                        filtered = 1;
1200
1201                if (ce_stage(ce)) {
1202                        while ((i < istate->cache_nr) &&
1203                               ! strcmp(istate->cache[i]->name, ce->name))
1204                                i++;
1205                        i--;
1206                        if (allow_unmerged)
1207                                continue;
1208                        if (!filtered)
1209                                show_file(unmerged_fmt, ce->name, in_porcelain,
1210                                          &first, header_msg);
1211                        has_errors = 1;
1212                        continue;
1213                }
1214
1215                if (filtered)
1216                        continue;
1217
1218                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1219                if (new == ce)
1220                        continue;
1221                if (!new) {
1222                        const char *fmt;
1223
1224                        if (really && cache_errno == EINVAL) {
1225                                /* If we are doing --really-refresh that
1226                                 * means the index is not valid anymore.
1227                                 */
1228                                ce->ce_flags &= ~CE_VALID;
1229                                ce->ce_flags |= CE_UPDATE_IN_BASE;
1230                                istate->cache_changed |= CE_ENTRY_CHANGED;
1231                        }
1232                        if (quiet)
1233                                continue;
1234
1235                        if (cache_errno == ENOENT)
1236                                fmt = deleted_fmt;
1237                        else if (ce->ce_flags & CE_INTENT_TO_ADD)
1238                                fmt = added_fmt; /* must be before other checks */
1239                        else if (changed & TYPE_CHANGED)
1240                                fmt = typechange_fmt;
1241                        else
1242                                fmt = modified_fmt;
1243                        show_file(fmt,
1244                                  ce->name, in_porcelain, &first, header_msg);
1245                        has_errors = 1;
1246                        continue;
1247                }
1248
1249                replace_index_entry(istate, i, new);
1250        }
1251        return has_errors;
1252}
1253
1254static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1255                                               unsigned int options)
1256{
1257        return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1258}
1259
1260
1261/*****************************************************************
1262 * Index File I/O
1263 *****************************************************************/
1264
1265#define INDEX_FORMAT_DEFAULT 3
1266
1267static unsigned int get_index_format_default(void)
1268{
1269        char *envversion = getenv("GIT_INDEX_VERSION");
1270        char *endp;
1271        int value;
1272        unsigned int version = INDEX_FORMAT_DEFAULT;
1273
1274        if (!envversion) {
1275                if (!git_config_get_int("index.version", &value))
1276                        version = value;
1277                if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1278                        warning(_("index.version set, but the value is invalid.\n"
1279                                  "Using version %i"), INDEX_FORMAT_DEFAULT);
1280                        return INDEX_FORMAT_DEFAULT;
1281                }
1282                return version;
1283        }
1284
1285        version = strtoul(envversion, &endp, 10);
1286        if (*endp ||
1287            version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1288                warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1289                          "Using version %i"), INDEX_FORMAT_DEFAULT);
1290                version = INDEX_FORMAT_DEFAULT;
1291        }
1292        return version;
1293}
1294
1295/*
1296 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1297 * Again - this is just a (very strong in practice) heuristic that
1298 * the inode hasn't changed.
1299 *
1300 * We save the fields in big-endian order to allow using the
1301 * index file over NFS transparently.
1302 */
1303struct ondisk_cache_entry {
1304        struct cache_time ctime;
1305        struct cache_time mtime;
1306        uint32_t dev;
1307        uint32_t ino;
1308        uint32_t mode;
1309        uint32_t uid;
1310        uint32_t gid;
1311        uint32_t size;
1312        unsigned char sha1[20];
1313        uint16_t flags;
1314        char name[FLEX_ARRAY]; /* more */
1315};
1316
1317/*
1318 * This struct is used when CE_EXTENDED bit is 1
1319 * The struct must match ondisk_cache_entry exactly from
1320 * ctime till flags
1321 */
1322struct ondisk_cache_entry_extended {
1323        struct cache_time ctime;
1324        struct cache_time mtime;
1325        uint32_t dev;
1326        uint32_t ino;
1327        uint32_t mode;
1328        uint32_t uid;
1329        uint32_t gid;
1330        uint32_t size;
1331        unsigned char sha1[20];
1332        uint16_t flags;
1333        uint16_t flags2;
1334        char name[FLEX_ARRAY]; /* more */
1335};
1336
1337/* These are only used for v3 or lower */
1338#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1339#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1340#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1341#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1342                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1343                            ondisk_cache_entry_size(ce_namelen(ce)))
1344
1345static int verify_hdr(struct cache_header *hdr, unsigned long size)
1346{
1347        git_SHA_CTX c;
1348        unsigned char sha1[20];
1349        int hdr_version;
1350
1351        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1352                return error("bad signature");
1353        hdr_version = ntohl(hdr->hdr_version);
1354        if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1355                return error("bad index version %d", hdr_version);
1356        git_SHA1_Init(&c);
1357        git_SHA1_Update(&c, hdr, size - 20);
1358        git_SHA1_Final(sha1, &c);
1359        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1360                return error("bad index file sha1 signature");
1361        return 0;
1362}
1363
1364static int read_index_extension(struct index_state *istate,
1365                                const char *ext, void *data, unsigned long sz)
1366{
1367        switch (CACHE_EXT(ext)) {
1368        case CACHE_EXT_TREE:
1369                istate->cache_tree = cache_tree_read(data, sz);
1370                break;
1371        case CACHE_EXT_RESOLVE_UNDO:
1372                istate->resolve_undo = resolve_undo_read(data, sz);
1373                break;
1374        case CACHE_EXT_LINK:
1375                if (read_link_extension(istate, data, sz))
1376                        return -1;
1377                break;
1378        case CACHE_EXT_UNTRACKED:
1379                istate->untracked = read_untracked_extension(data, sz);
1380                break;
1381        default:
1382                if (*ext < 'A' || 'Z' < *ext)
1383                        return error("index uses %.4s extension, which we do not understand",
1384                                     ext);
1385                fprintf(stderr, "ignoring %.4s extension\n", ext);
1386                break;
1387        }
1388        return 0;
1389}
1390
1391int hold_locked_index(struct lock_file *lk, int die_on_error)
1392{
1393        return hold_lock_file_for_update(lk, get_index_file(),
1394                                         die_on_error
1395                                         ? LOCK_DIE_ON_ERROR
1396                                         : 0);
1397}
1398
1399int read_index(struct index_state *istate)
1400{
1401        return read_index_from(istate, get_index_file());
1402}
1403
1404static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1405                                                   unsigned int flags,
1406                                                   const char *name,
1407                                                   size_t len)
1408{
1409        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1410
1411        ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1412        ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1413        ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1414        ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1415        ce->ce_stat_data.sd_dev   = get_be32(&ondisk->dev);
1416        ce->ce_stat_data.sd_ino   = get_be32(&ondisk->ino);
1417        ce->ce_mode  = get_be32(&ondisk->mode);
1418        ce->ce_stat_data.sd_uid   = get_be32(&ondisk->uid);
1419        ce->ce_stat_data.sd_gid   = get_be32(&ondisk->gid);
1420        ce->ce_stat_data.sd_size  = get_be32(&ondisk->size);
1421        ce->ce_flags = flags & ~CE_NAMEMASK;
1422        ce->ce_namelen = len;
1423        ce->index = 0;
1424        hashcpy(ce->sha1, ondisk->sha1);
1425        memcpy(ce->name, name, len);
1426        ce->name[len] = '\0';
1427        return ce;
1428}
1429
1430/*
1431 * Adjacent cache entries tend to share the leading paths, so it makes
1432 * sense to only store the differences in later entries.  In the v4
1433 * on-disk format of the index, each on-disk cache entry stores the
1434 * number of bytes to be stripped from the end of the previous name,
1435 * and the bytes to append to the result, to come up with its name.
1436 */
1437static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1438{
1439        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1440        size_t len = decode_varint(&cp);
1441
1442        if (name->len < len)
1443                die("malformed name field in the index");
1444        strbuf_remove(name, name->len - len, len);
1445        for (ep = cp; *ep; ep++)
1446                ; /* find the end */
1447        strbuf_add(name, cp, ep - cp);
1448        return (const char *)ep + 1 - cp_;
1449}
1450
1451static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1452                                            unsigned long *ent_size,
1453                                            struct strbuf *previous_name)
1454{
1455        struct cache_entry *ce;
1456        size_t len;
1457        const char *name;
1458        unsigned int flags;
1459
1460        /* On-disk flags are just 16 bits */
1461        flags = get_be16(&ondisk->flags);
1462        len = flags & CE_NAMEMASK;
1463
1464        if (flags & CE_EXTENDED) {
1465                struct ondisk_cache_entry_extended *ondisk2;
1466                int extended_flags;
1467                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1468                extended_flags = get_be16(&ondisk2->flags2) << 16;
1469                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1470                if (extended_flags & ~CE_EXTENDED_FLAGS)
1471                        die("Unknown index entry format %08x", extended_flags);
1472                flags |= extended_flags;
1473                name = ondisk2->name;
1474        }
1475        else
1476                name = ondisk->name;
1477
1478        if (!previous_name) {
1479                /* v3 and earlier */
1480                if (len == CE_NAMEMASK)
1481                        len = strlen(name);
1482                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1483
1484                *ent_size = ondisk_ce_size(ce);
1485        } else {
1486                unsigned long consumed;
1487                consumed = expand_name_field(previous_name, name);
1488                ce = cache_entry_from_ondisk(ondisk, flags,
1489                                             previous_name->buf,
1490                                             previous_name->len);
1491
1492                *ent_size = (name - ((char *)ondisk)) + consumed;
1493        }
1494        return ce;
1495}
1496
1497static void check_ce_order(struct cache_entry *ce, struct cache_entry *next_ce)
1498{
1499        int name_compare = strcmp(ce->name, next_ce->name);
1500        if (0 < name_compare)
1501                die("unordered stage entries in index");
1502        if (!name_compare) {
1503                if (!ce_stage(ce))
1504                        die("multiple stage entries for merged file '%s'",
1505                                ce->name);
1506                if (ce_stage(ce) > ce_stage(next_ce))
1507                        die("unordered stage entries for '%s'",
1508                                ce->name);
1509        }
1510}
1511
1512/* remember to discard_cache() before reading a different cache! */
1513int do_read_index(struct index_state *istate, const char *path, int must_exist)
1514{
1515        int fd, i;
1516        struct stat st;
1517        unsigned long src_offset;
1518        struct cache_header *hdr;
1519        void *mmap;
1520        size_t mmap_size;
1521        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1522
1523        if (istate->initialized)
1524                return istate->cache_nr;
1525
1526        istate->timestamp.sec = 0;
1527        istate->timestamp.nsec = 0;
1528        fd = open(path, O_RDONLY);
1529        if (fd < 0) {
1530                if (!must_exist && errno == ENOENT)
1531                        return 0;
1532                die_errno("%s: index file open failed", path);
1533        }
1534
1535        if (fstat(fd, &st))
1536                die_errno("cannot stat the open index");
1537
1538        mmap_size = xsize_t(st.st_size);
1539        if (mmap_size < sizeof(struct cache_header) + 20)
1540                die("index file smaller than expected");
1541
1542        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1543        if (mmap == MAP_FAILED)
1544                die_errno("unable to map index file");
1545        close(fd);
1546
1547        hdr = mmap;
1548        if (verify_hdr(hdr, mmap_size) < 0)
1549                goto unmap;
1550
1551        hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1552        istate->version = ntohl(hdr->hdr_version);
1553        istate->cache_nr = ntohl(hdr->hdr_entries);
1554        istate->cache_alloc = alloc_nr(istate->cache_nr);
1555        istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1556        istate->initialized = 1;
1557
1558        if (istate->version == 4)
1559                previous_name = &previous_name_buf;
1560        else
1561                previous_name = NULL;
1562
1563        src_offset = sizeof(*hdr);
1564        for (i = 0; i < istate->cache_nr; i++) {
1565                struct ondisk_cache_entry *disk_ce;
1566                struct cache_entry *ce;
1567                unsigned long consumed;
1568
1569                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1570                ce = create_from_disk(disk_ce, &consumed, previous_name);
1571                set_index_entry(istate, i, ce);
1572
1573                if (i > 0)
1574                        check_ce_order(istate->cache[i - 1], ce);
1575
1576                src_offset += consumed;
1577        }
1578        strbuf_release(&previous_name_buf);
1579        istate->timestamp.sec = st.st_mtime;
1580        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1581
1582        while (src_offset <= mmap_size - 20 - 8) {
1583                /* After an array of active_nr index entries,
1584                 * there can be arbitrary number of extended
1585                 * sections, each of which is prefixed with
1586                 * extension name (4-byte) and section length
1587                 * in 4-byte network byte order.
1588                 */
1589                uint32_t extsize;
1590                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1591                extsize = ntohl(extsize);
1592                if (read_index_extension(istate,
1593                                         (const char *) mmap + src_offset,
1594                                         (char *) mmap + src_offset + 8,
1595                                         extsize) < 0)
1596                        goto unmap;
1597                src_offset += 8;
1598                src_offset += extsize;
1599        }
1600        munmap(mmap, mmap_size);
1601        return istate->cache_nr;
1602
1603unmap:
1604        munmap(mmap, mmap_size);
1605        die("index file corrupt");
1606}
1607
1608int read_index_from(struct index_state *istate, const char *path)
1609{
1610        struct split_index *split_index;
1611        int ret;
1612
1613        /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1614        if (istate->initialized)
1615                return istate->cache_nr;
1616
1617        ret = do_read_index(istate, path, 0);
1618        split_index = istate->split_index;
1619        if (!split_index)
1620                return ret;
1621
1622        if (is_null_sha1(split_index->base_sha1))
1623                return ret;
1624
1625        if (split_index->base)
1626                discard_index(split_index->base);
1627        else
1628                split_index->base = xcalloc(1, sizeof(*split_index->base));
1629        ret = do_read_index(split_index->base,
1630                            git_path("sharedindex.%s",
1631                                     sha1_to_hex(split_index->base_sha1)), 1);
1632        if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1633                die("broken index, expect %s in %s, got %s",
1634                    sha1_to_hex(split_index->base_sha1),
1635                    git_path("sharedindex.%s",
1636                                     sha1_to_hex(split_index->base_sha1)),
1637                    sha1_to_hex(split_index->base->sha1));
1638        merge_base_index(istate);
1639        return ret;
1640}
1641
1642int is_index_unborn(struct index_state *istate)
1643{
1644        return (!istate->cache_nr && !istate->timestamp.sec);
1645}
1646
1647int discard_index(struct index_state *istate)
1648{
1649        int i;
1650
1651        for (i = 0; i < istate->cache_nr; i++) {
1652                if (istate->cache[i]->index &&
1653                    istate->split_index &&
1654                    istate->split_index->base &&
1655                    istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1656                    istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1657                        continue;
1658                free(istate->cache[i]);
1659        }
1660        resolve_undo_clear_index(istate);
1661        istate->cache_nr = 0;
1662        istate->cache_changed = 0;
1663        istate->timestamp.sec = 0;
1664        istate->timestamp.nsec = 0;
1665        free_name_hash(istate);
1666        cache_tree_free(&(istate->cache_tree));
1667        istate->initialized = 0;
1668        free(istate->cache);
1669        istate->cache = NULL;
1670        istate->cache_alloc = 0;
1671        discard_split_index(istate);
1672        free_untracked_cache(istate->untracked);
1673        istate->untracked = NULL;
1674        return 0;
1675}
1676
1677int unmerged_index(const struct index_state *istate)
1678{
1679        int i;
1680        for (i = 0; i < istate->cache_nr; i++) {
1681                if (ce_stage(istate->cache[i]))
1682                        return 1;
1683        }
1684        return 0;
1685}
1686
1687#define WRITE_BUFFER_SIZE 8192
1688static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1689static unsigned long write_buffer_len;
1690
1691static int ce_write_flush(git_SHA_CTX *context, int fd)
1692{
1693        unsigned int buffered = write_buffer_len;
1694        if (buffered) {
1695                git_SHA1_Update(context, write_buffer, buffered);
1696                if (write_in_full(fd, write_buffer, buffered) != buffered)
1697                        return -1;
1698                write_buffer_len = 0;
1699        }
1700        return 0;
1701}
1702
1703static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1704{
1705        while (len) {
1706                unsigned int buffered = write_buffer_len;
1707                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1708                if (partial > len)
1709                        partial = len;
1710                memcpy(write_buffer + buffered, data, partial);
1711                buffered += partial;
1712                if (buffered == WRITE_BUFFER_SIZE) {
1713                        write_buffer_len = buffered;
1714                        if (ce_write_flush(context, fd))
1715                                return -1;
1716                        buffered = 0;
1717                }
1718                write_buffer_len = buffered;
1719                len -= partial;
1720                data = (char *) data + partial;
1721        }
1722        return 0;
1723}
1724
1725static int write_index_ext_header(git_SHA_CTX *context, int fd,
1726                                  unsigned int ext, unsigned int sz)
1727{
1728        ext = htonl(ext);
1729        sz = htonl(sz);
1730        return ((ce_write(context, fd, &ext, 4) < 0) ||
1731                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1732}
1733
1734static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1735{
1736        unsigned int left = write_buffer_len;
1737
1738        if (left) {
1739                write_buffer_len = 0;
1740                git_SHA1_Update(context, write_buffer, left);
1741        }
1742
1743        /* Flush first if not enough space for SHA1 signature */
1744        if (left + 20 > WRITE_BUFFER_SIZE) {
1745                if (write_in_full(fd, write_buffer, left) != left)
1746                        return -1;
1747                left = 0;
1748        }
1749
1750        /* Append the SHA1 signature at the end */
1751        git_SHA1_Final(write_buffer + left, context);
1752        hashcpy(sha1, write_buffer + left);
1753        left += 20;
1754        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1755}
1756
1757static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1758{
1759        /*
1760         * The only thing we care about in this function is to smudge the
1761         * falsely clean entry due to touch-update-touch race, so we leave
1762         * everything else as they are.  We are called for entries whose
1763         * ce_stat_data.sd_mtime match the index file mtime.
1764         *
1765         * Note that this actually does not do much for gitlinks, for
1766         * which ce_match_stat_basic() always goes to the actual
1767         * contents.  The caller checks with is_racy_timestamp() which
1768         * always says "no" for gitlinks, so we are not called for them ;-)
1769         */
1770        struct stat st;
1771
1772        if (lstat(ce->name, &st) < 0)
1773                return;
1774        if (ce_match_stat_basic(ce, &st))
1775                return;
1776        if (ce_modified_check_fs(ce, &st)) {
1777                /* This is "racily clean"; smudge it.  Note that this
1778                 * is a tricky code.  At first glance, it may appear
1779                 * that it can break with this sequence:
1780                 *
1781                 * $ echo xyzzy >frotz
1782                 * $ git-update-index --add frotz
1783                 * $ : >frotz
1784                 * $ sleep 3
1785                 * $ echo filfre >nitfol
1786                 * $ git-update-index --add nitfol
1787                 *
1788                 * but it does not.  When the second update-index runs,
1789                 * it notices that the entry "frotz" has the same timestamp
1790                 * as index, and if we were to smudge it by resetting its
1791                 * size to zero here, then the object name recorded
1792                 * in index is the 6-byte file but the cached stat information
1793                 * becomes zero --- which would then match what we would
1794                 * obtain from the filesystem next time we stat("frotz").
1795                 *
1796                 * However, the second update-index, before calling
1797                 * this function, notices that the cached size is 6
1798                 * bytes and what is on the filesystem is an empty
1799                 * file, and never calls us, so the cached size information
1800                 * for "frotz" stays 6 which does not match the filesystem.
1801                 */
1802                ce->ce_stat_data.sd_size = 0;
1803        }
1804}
1805
1806/* Copy miscellaneous fields but not the name */
1807static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1808                                       struct cache_entry *ce)
1809{
1810        short flags;
1811
1812        ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1813        ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1814        ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1815        ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1816        ondisk->dev  = htonl(ce->ce_stat_data.sd_dev);
1817        ondisk->ino  = htonl(ce->ce_stat_data.sd_ino);
1818        ondisk->mode = htonl(ce->ce_mode);
1819        ondisk->uid  = htonl(ce->ce_stat_data.sd_uid);
1820        ondisk->gid  = htonl(ce->ce_stat_data.sd_gid);
1821        ondisk->size = htonl(ce->ce_stat_data.sd_size);
1822        hashcpy(ondisk->sha1, ce->sha1);
1823
1824        flags = ce->ce_flags & ~CE_NAMEMASK;
1825        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1826        ondisk->flags = htons(flags);
1827        if (ce->ce_flags & CE_EXTENDED) {
1828                struct ondisk_cache_entry_extended *ondisk2;
1829                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1830                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1831                return ondisk2->name;
1832        }
1833        else {
1834                return ondisk->name;
1835        }
1836}
1837
1838static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1839                          struct strbuf *previous_name)
1840{
1841        int size;
1842        struct ondisk_cache_entry *ondisk;
1843        int saved_namelen = saved_namelen; /* compiler workaround */
1844        char *name;
1845        int result;
1846
1847        if (ce->ce_flags & CE_STRIP_NAME) {
1848                saved_namelen = ce_namelen(ce);
1849                ce->ce_namelen = 0;
1850        }
1851
1852        if (!previous_name) {
1853                size = ondisk_ce_size(ce);
1854                ondisk = xcalloc(1, size);
1855                name = copy_cache_entry_to_ondisk(ondisk, ce);
1856                memcpy(name, ce->name, ce_namelen(ce));
1857        } else {
1858                int common, to_remove, prefix_size;
1859                unsigned char to_remove_vi[16];
1860                for (common = 0;
1861                     (ce->name[common] &&
1862                      common < previous_name->len &&
1863                      ce->name[common] == previous_name->buf[common]);
1864                     common++)
1865                        ; /* still matching */
1866                to_remove = previous_name->len - common;
1867                prefix_size = encode_varint(to_remove, to_remove_vi);
1868
1869                if (ce->ce_flags & CE_EXTENDED)
1870                        size = offsetof(struct ondisk_cache_entry_extended, name);
1871                else
1872                        size = offsetof(struct ondisk_cache_entry, name);
1873                size += prefix_size + (ce_namelen(ce) - common + 1);
1874
1875                ondisk = xcalloc(1, size);
1876                name = copy_cache_entry_to_ondisk(ondisk, ce);
1877                memcpy(name, to_remove_vi, prefix_size);
1878                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1879
1880                strbuf_splice(previous_name, common, to_remove,
1881                              ce->name + common, ce_namelen(ce) - common);
1882        }
1883        if (ce->ce_flags & CE_STRIP_NAME) {
1884                ce->ce_namelen = saved_namelen;
1885                ce->ce_flags &= ~CE_STRIP_NAME;
1886        }
1887
1888        result = ce_write(c, fd, ondisk, size);
1889        free(ondisk);
1890        return result;
1891}
1892
1893/*
1894 * This function verifies if index_state has the correct sha1 of the
1895 * index file.  Don't die if we have any other failure, just return 0.
1896 */
1897static int verify_index_from(const struct index_state *istate, const char *path)
1898{
1899        int fd;
1900        ssize_t n;
1901        struct stat st;
1902        unsigned char sha1[20];
1903
1904        if (!istate->initialized)
1905                return 0;
1906
1907        fd = open(path, O_RDONLY);
1908        if (fd < 0)
1909                return 0;
1910
1911        if (fstat(fd, &st))
1912                goto out;
1913
1914        if (st.st_size < sizeof(struct cache_header) + 20)
1915                goto out;
1916
1917        n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1918        if (n != 20)
1919                goto out;
1920
1921        if (hashcmp(istate->sha1, sha1))
1922                goto out;
1923
1924        close(fd);
1925        return 1;
1926
1927out:
1928        close(fd);
1929        return 0;
1930}
1931
1932static int verify_index(const struct index_state *istate)
1933{
1934        return verify_index_from(istate, get_index_file());
1935}
1936
1937static int has_racy_timestamp(struct index_state *istate)
1938{
1939        int entries = istate->cache_nr;
1940        int i;
1941
1942        for (i = 0; i < entries; i++) {
1943                struct cache_entry *ce = istate->cache[i];
1944                if (is_racy_timestamp(istate, ce))
1945                        return 1;
1946        }
1947        return 0;
1948}
1949
1950/*
1951 * Opportunistically update the index but do not complain if we can't
1952 */
1953void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1954{
1955        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1956            verify_index(istate) &&
1957            write_locked_index(istate, lockfile, COMMIT_LOCK))
1958                rollback_lock_file(lockfile);
1959}
1960
1961static int do_write_index(struct index_state *istate, int newfd,
1962                          int strip_extensions)
1963{
1964        git_SHA_CTX c;
1965        struct cache_header hdr;
1966        int i, err, removed, extended, hdr_version;
1967        struct cache_entry **cache = istate->cache;
1968        int entries = istate->cache_nr;
1969        struct stat st;
1970        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1971
1972        for (i = removed = extended = 0; i < entries; i++) {
1973                if (cache[i]->ce_flags & CE_REMOVE)
1974                        removed++;
1975
1976                /* reduce extended entries if possible */
1977                cache[i]->ce_flags &= ~CE_EXTENDED;
1978                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1979                        extended++;
1980                        cache[i]->ce_flags |= CE_EXTENDED;
1981                }
1982        }
1983
1984        if (!istate->version) {
1985                istate->version = get_index_format_default();
1986                if (getenv("GIT_TEST_SPLIT_INDEX"))
1987                        init_split_index(istate);
1988        }
1989
1990        /* demote version 3 to version 2 when the latter suffices */
1991        if (istate->version == 3 || istate->version == 2)
1992                istate->version = extended ? 3 : 2;
1993
1994        hdr_version = istate->version;
1995
1996        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1997        hdr.hdr_version = htonl(hdr_version);
1998        hdr.hdr_entries = htonl(entries - removed);
1999
2000        git_SHA1_Init(&c);
2001        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2002                return -1;
2003
2004        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2005        for (i = 0; i < entries; i++) {
2006                struct cache_entry *ce = cache[i];
2007                if (ce->ce_flags & CE_REMOVE)
2008                        continue;
2009                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2010                        ce_smudge_racily_clean_entry(ce);
2011                if (is_null_sha1(ce->sha1)) {
2012                        static const char msg[] = "cache entry has null sha1: %s";
2013                        static int allow = -1;
2014
2015                        if (allow < 0)
2016                                allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2017                        if (allow)
2018                                warning(msg, ce->name);
2019                        else
2020                                return error(msg, ce->name);
2021                }
2022                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2023                        return -1;
2024        }
2025        strbuf_release(&previous_name_buf);
2026
2027        /* Write extension data here */
2028        if (!strip_extensions && istate->split_index) {
2029                struct strbuf sb = STRBUF_INIT;
2030
2031                err = write_link_extension(&sb, istate) < 0 ||
2032                        write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2033                                               sb.len) < 0 ||
2034                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2035                strbuf_release(&sb);
2036                if (err)
2037                        return -1;
2038        }
2039        if (!strip_extensions && istate->cache_tree) {
2040                struct strbuf sb = STRBUF_INIT;
2041
2042                cache_tree_write(&sb, istate->cache_tree);
2043                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2044                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2045                strbuf_release(&sb);
2046                if (err)
2047                        return -1;
2048        }
2049        if (!strip_extensions && istate->resolve_undo) {
2050                struct strbuf sb = STRBUF_INIT;
2051
2052                resolve_undo_write(&sb, istate->resolve_undo);
2053                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2054                                             sb.len) < 0
2055                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2056                strbuf_release(&sb);
2057                if (err)
2058                        return -1;
2059        }
2060        if (!strip_extensions && istate->untracked) {
2061                struct strbuf sb = STRBUF_INIT;
2062
2063                write_untracked_extension(&sb, istate->untracked);
2064                err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2065                                             sb.len) < 0 ||
2066                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2067                strbuf_release(&sb);
2068                if (err)
2069                        return -1;
2070        }
2071
2072        if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2073                return -1;
2074        istate->timestamp.sec = (unsigned int)st.st_mtime;
2075        istate->timestamp.nsec = ST_MTIME_NSEC(st);
2076        return 0;
2077}
2078
2079void set_alternate_index_output(const char *name)
2080{
2081        alternate_index_output = name;
2082}
2083
2084static int commit_locked_index(struct lock_file *lk)
2085{
2086        if (alternate_index_output)
2087                return commit_lock_file_to(lk, alternate_index_output);
2088        else
2089                return commit_lock_file(lk);
2090}
2091
2092static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2093                                 unsigned flags)
2094{
2095        int ret = do_write_index(istate, lock->fd, 0);
2096        if (ret)
2097                return ret;
2098        assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2099               (COMMIT_LOCK | CLOSE_LOCK));
2100        if (flags & COMMIT_LOCK)
2101                return commit_locked_index(lock);
2102        else if (flags & CLOSE_LOCK)
2103                return close_lock_file(lock);
2104        else
2105                return ret;
2106}
2107
2108static int write_split_index(struct index_state *istate,
2109                             struct lock_file *lock,
2110                             unsigned flags)
2111{
2112        int ret;
2113        prepare_to_write_split_index(istate);
2114        ret = do_write_locked_index(istate, lock, flags);
2115        finish_writing_split_index(istate);
2116        return ret;
2117}
2118
2119static char *temporary_sharedindex;
2120
2121static void remove_temporary_sharedindex(void)
2122{
2123        if (temporary_sharedindex) {
2124                unlink_or_warn(temporary_sharedindex);
2125                free(temporary_sharedindex);
2126                temporary_sharedindex = NULL;
2127        }
2128}
2129
2130static void remove_temporary_sharedindex_on_signal(int signo)
2131{
2132        remove_temporary_sharedindex();
2133        sigchain_pop(signo);
2134        raise(signo);
2135}
2136
2137static int write_shared_index(struct index_state *istate,
2138                              struct lock_file *lock, unsigned flags)
2139{
2140        struct split_index *si = istate->split_index;
2141        static int installed_handler;
2142        int fd, ret;
2143
2144        temporary_sharedindex = git_pathdup("sharedindex_XXXXXX");
2145        fd = mkstemp(temporary_sharedindex);
2146        if (fd < 0) {
2147                free(temporary_sharedindex);
2148                temporary_sharedindex = NULL;
2149                hashclr(si->base_sha1);
2150                return do_write_locked_index(istate, lock, flags);
2151        }
2152        if (!installed_handler) {
2153                atexit(remove_temporary_sharedindex);
2154                sigchain_push_common(remove_temporary_sharedindex_on_signal);
2155        }
2156        move_cache_to_base_index(istate);
2157        ret = do_write_index(si->base, fd, 1);
2158        close(fd);
2159        if (ret) {
2160                remove_temporary_sharedindex();
2161                return ret;
2162        }
2163        ret = rename(temporary_sharedindex,
2164                     git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2165        free(temporary_sharedindex);
2166        temporary_sharedindex = NULL;
2167        if (!ret)
2168                hashcpy(si->base_sha1, si->base->sha1);
2169        return ret;
2170}
2171
2172int write_locked_index(struct index_state *istate, struct lock_file *lock,
2173                       unsigned flags)
2174{
2175        struct split_index *si = istate->split_index;
2176
2177        if (!si || alternate_index_output ||
2178            (istate->cache_changed & ~EXTMASK)) {
2179                if (si)
2180                        hashclr(si->base_sha1);
2181                return do_write_locked_index(istate, lock, flags);
2182        }
2183
2184        if (getenv("GIT_TEST_SPLIT_INDEX")) {
2185                int v = si->base_sha1[0];
2186                if ((v & 15) < 6)
2187                        istate->cache_changed |= SPLIT_INDEX_ORDERED;
2188        }
2189        if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2190                int ret = write_shared_index(istate, lock, flags);
2191                if (ret)
2192                        return ret;
2193        }
2194
2195        return write_split_index(istate, lock, flags);
2196}
2197
2198/*
2199 * Read the index file that is potentially unmerged into given
2200 * index_state, dropping any unmerged entries.  Returns true if
2201 * the index is unmerged.  Callers who want to refuse to work
2202 * from an unmerged state can call this and check its return value,
2203 * instead of calling read_cache().
2204 */
2205int read_index_unmerged(struct index_state *istate)
2206{
2207        int i;
2208        int unmerged = 0;
2209
2210        read_index(istate);
2211        for (i = 0; i < istate->cache_nr; i++) {
2212                struct cache_entry *ce = istate->cache[i];
2213                struct cache_entry *new_ce;
2214                int size, len;
2215
2216                if (!ce_stage(ce))
2217                        continue;
2218                unmerged = 1;
2219                len = ce_namelen(ce);
2220                size = cache_entry_size(len);
2221                new_ce = xcalloc(1, size);
2222                memcpy(new_ce->name, ce->name, len);
2223                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2224                new_ce->ce_namelen = len;
2225                new_ce->ce_mode = ce->ce_mode;
2226                if (add_index_entry(istate, new_ce, 0))
2227                        return error("%s: cannot drop to stage #0",
2228                                     new_ce->name);
2229        }
2230        return unmerged;
2231}
2232
2233/*
2234 * Returns 1 if the path is an "other" path with respect to
2235 * the index; that is, the path is not mentioned in the index at all,
2236 * either as a file, a directory with some files in the index,
2237 * or as an unmerged entry.
2238 *
2239 * We helpfully remove a trailing "/" from directories so that
2240 * the output of read_directory can be used as-is.
2241 */
2242int index_name_is_other(const struct index_state *istate, const char *name,
2243                int namelen)
2244{
2245        int pos;
2246        if (namelen && name[namelen - 1] == '/')
2247                namelen--;
2248        pos = index_name_pos(istate, name, namelen);
2249        if (0 <= pos)
2250                return 0;       /* exact match */
2251        pos = -pos - 1;
2252        if (pos < istate->cache_nr) {
2253                struct cache_entry *ce = istate->cache[pos];
2254                if (ce_namelen(ce) == namelen &&
2255                    !memcmp(ce->name, name, namelen))
2256                        return 0; /* Yup, this one exists unmerged */
2257        }
2258        return 1;
2259}
2260
2261void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2262{
2263        int pos, len;
2264        unsigned long sz;
2265        enum object_type type;
2266        void *data;
2267
2268        len = strlen(path);
2269        pos = index_name_pos(istate, path, len);
2270        if (pos < 0) {
2271                /*
2272                 * We might be in the middle of a merge, in which
2273                 * case we would read stage #2 (ours).
2274                 */
2275                int i;
2276                for (i = -pos - 1;
2277                     (pos < 0 && i < istate->cache_nr &&
2278                      !strcmp(istate->cache[i]->name, path));
2279                     i++)
2280                        if (ce_stage(istate->cache[i]) == 2)
2281                                pos = i;
2282        }
2283        if (pos < 0)
2284                return NULL;
2285        data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2286        if (!data || type != OBJ_BLOB) {
2287                free(data);
2288                return NULL;
2289        }
2290        if (size)
2291                *size = sz;
2292        return data;
2293}
2294
2295void stat_validity_clear(struct stat_validity *sv)
2296{
2297        free(sv->sd);
2298        sv->sd = NULL;
2299}
2300
2301int stat_validity_check(struct stat_validity *sv, const char *path)
2302{
2303        struct stat st;
2304
2305        if (stat(path, &st) < 0)
2306                return sv->sd == NULL;
2307        if (!sv->sd)
2308                return 0;
2309        return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2310}
2311
2312void stat_validity_update(struct stat_validity *sv, int fd)
2313{
2314        struct stat st;
2315
2316        if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2317                stat_validity_clear(sv);
2318        else {
2319                if (!sv->sd)
2320                        sv->sd = xcalloc(1, sizeof(struct stat_data));
2321                fill_stat_data(sv->sd, &st);
2322        }
2323}