read-cache.con commit Merge branch 'nd/maint-relative' into maint (0bc6180)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 */
   6#define NO_THE_INDEX_COMPATIBILITY_MACROS
   7#include "cache.h"
   8#include "cache-tree.h"
   9#include "refs.h"
  10#include "dir.h"
  11#include "tree.h"
  12#include "commit.h"
  13#include "blob.h"
  14#include "resolve-undo.h"
  15
  16static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really);
  17
  18/* Index extensions.
  19 *
  20 * The first letter should be 'A'..'Z' for extensions that are not
  21 * necessary for a correct operation (i.e. optimization data).
  22 * When new extensions are added that _needs_ to be understood in
  23 * order to correctly interpret the index file, pick character that
  24 * is outside the range, to cause the reader to abort.
  25 */
  26
  27#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  28#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  29#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  30
  31struct index_state the_index;
  32
  33static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  34{
  35        istate->cache[nr] = ce;
  36        add_name_hash(istate, ce);
  37}
  38
  39static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  40{
  41        struct cache_entry *old = istate->cache[nr];
  42
  43        remove_name_hash(old);
  44        set_index_entry(istate, nr, ce);
  45        istate->cache_changed = 1;
  46}
  47
  48void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  49{
  50        struct cache_entry *old = istate->cache[nr], *new;
  51        int namelen = strlen(new_name);
  52
  53        new = xmalloc(cache_entry_size(namelen));
  54        copy_cache_entry(new, old);
  55        new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
  56        new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
  57        memcpy(new->name, new_name, namelen + 1);
  58
  59        cache_tree_invalidate_path(istate->cache_tree, old->name);
  60        remove_index_entry_at(istate, nr);
  61        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  62}
  63
  64/*
  65 * This only updates the "non-critical" parts of the directory
  66 * cache, ie the parts that aren't tracked by GIT, and only used
  67 * to validate the cache.
  68 */
  69void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
  70{
  71        ce->ce_ctime.sec = (unsigned int)st->st_ctime;
  72        ce->ce_mtime.sec = (unsigned int)st->st_mtime;
  73        ce->ce_ctime.nsec = ST_CTIME_NSEC(*st);
  74        ce->ce_mtime.nsec = ST_MTIME_NSEC(*st);
  75        ce->ce_dev = st->st_dev;
  76        ce->ce_ino = st->st_ino;
  77        ce->ce_uid = st->st_uid;
  78        ce->ce_gid = st->st_gid;
  79        ce->ce_size = st->st_size;
  80
  81        if (assume_unchanged)
  82                ce->ce_flags |= CE_VALID;
  83
  84        if (S_ISREG(st->st_mode))
  85                ce_mark_uptodate(ce);
  86}
  87
  88static int ce_compare_data(struct cache_entry *ce, struct stat *st)
  89{
  90        int match = -1;
  91        int fd = open(ce->name, O_RDONLY);
  92
  93        if (fd >= 0) {
  94                unsigned char sha1[20];
  95                if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
  96                        match = hashcmp(sha1, ce->sha1);
  97                /* index_fd() closed the file descriptor already */
  98        }
  99        return match;
 100}
 101
 102static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
 103{
 104        int match = -1;
 105        void *buffer;
 106        unsigned long size;
 107        enum object_type type;
 108        struct strbuf sb = STRBUF_INIT;
 109
 110        if (strbuf_readlink(&sb, ce->name, expected_size))
 111                return -1;
 112
 113        buffer = read_sha1_file(ce->sha1, &type, &size);
 114        if (buffer) {
 115                if (size == sb.len)
 116                        match = memcmp(buffer, sb.buf, size);
 117                free(buffer);
 118        }
 119        strbuf_release(&sb);
 120        return match;
 121}
 122
 123static int ce_compare_gitlink(struct cache_entry *ce)
 124{
 125        unsigned char sha1[20];
 126
 127        /*
 128         * We don't actually require that the .git directory
 129         * under GITLINK directory be a valid git directory. It
 130         * might even be missing (in case nobody populated that
 131         * sub-project).
 132         *
 133         * If so, we consider it always to match.
 134         */
 135        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 136                return 0;
 137        return hashcmp(sha1, ce->sha1);
 138}
 139
 140static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
 141{
 142        switch (st->st_mode & S_IFMT) {
 143        case S_IFREG:
 144                if (ce_compare_data(ce, st))
 145                        return DATA_CHANGED;
 146                break;
 147        case S_IFLNK:
 148                if (ce_compare_link(ce, xsize_t(st->st_size)))
 149                        return DATA_CHANGED;
 150                break;
 151        case S_IFDIR:
 152                if (S_ISGITLINK(ce->ce_mode))
 153                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 154        default:
 155                return TYPE_CHANGED;
 156        }
 157        return 0;
 158}
 159
 160static int is_empty_blob_sha1(const unsigned char *sha1)
 161{
 162        static const unsigned char empty_blob_sha1[20] = {
 163                0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
 164                0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
 165        };
 166
 167        return !hashcmp(sha1, empty_blob_sha1);
 168}
 169
 170static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 171{
 172        unsigned int changed = 0;
 173
 174        if (ce->ce_flags & CE_REMOVE)
 175                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 176
 177        switch (ce->ce_mode & S_IFMT) {
 178        case S_IFREG:
 179                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 180                /* We consider only the owner x bit to be relevant for
 181                 * "mode changes"
 182                 */
 183                if (trust_executable_bit &&
 184                    (0100 & (ce->ce_mode ^ st->st_mode)))
 185                        changed |= MODE_CHANGED;
 186                break;
 187        case S_IFLNK:
 188                if (!S_ISLNK(st->st_mode) &&
 189                    (has_symlinks || !S_ISREG(st->st_mode)))
 190                        changed |= TYPE_CHANGED;
 191                break;
 192        case S_IFGITLINK:
 193                /* We ignore most of the st_xxx fields for gitlinks */
 194                if (!S_ISDIR(st->st_mode))
 195                        changed |= TYPE_CHANGED;
 196                else if (ce_compare_gitlink(ce))
 197                        changed |= DATA_CHANGED;
 198                return changed;
 199        default:
 200                die("internal error: ce_mode is %o", ce->ce_mode);
 201        }
 202        if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
 203                changed |= MTIME_CHANGED;
 204        if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime)
 205                changed |= CTIME_CHANGED;
 206
 207#ifdef USE_NSEC
 208        if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
 209                changed |= MTIME_CHANGED;
 210        if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
 211                changed |= CTIME_CHANGED;
 212#endif
 213
 214        if (ce->ce_uid != (unsigned int) st->st_uid ||
 215            ce->ce_gid != (unsigned int) st->st_gid)
 216                changed |= OWNER_CHANGED;
 217        if (ce->ce_ino != (unsigned int) st->st_ino)
 218                changed |= INODE_CHANGED;
 219
 220#ifdef USE_STDEV
 221        /*
 222         * st_dev breaks on network filesystems where different
 223         * clients will have different views of what "device"
 224         * the filesystem is on
 225         */
 226        if (ce->ce_dev != (unsigned int) st->st_dev)
 227                changed |= INODE_CHANGED;
 228#endif
 229
 230        if (ce->ce_size != (unsigned int) st->st_size)
 231                changed |= DATA_CHANGED;
 232
 233        /* Racily smudged entry? */
 234        if (!ce->ce_size) {
 235                if (!is_empty_blob_sha1(ce->sha1))
 236                        changed |= DATA_CHANGED;
 237        }
 238
 239        return changed;
 240}
 241
 242static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
 243{
 244        return (!S_ISGITLINK(ce->ce_mode) &&
 245                istate->timestamp.sec &&
 246#ifdef USE_NSEC
 247                 /* nanosecond timestamped files can also be racy! */
 248                (istate->timestamp.sec < ce->ce_mtime.sec ||
 249                 (istate->timestamp.sec == ce->ce_mtime.sec &&
 250                  istate->timestamp.nsec <= ce->ce_mtime.nsec))
 251#else
 252                istate->timestamp.sec <= ce->ce_mtime.sec
 253#endif
 254                 );
 255}
 256
 257int ie_match_stat(const struct index_state *istate,
 258                  struct cache_entry *ce, struct stat *st,
 259                  unsigned int options)
 260{
 261        unsigned int changed;
 262        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 263        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 264        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 265
 266        /*
 267         * If it's marked as always valid in the index, it's
 268         * valid whatever the checked-out copy says.
 269         *
 270         * skip-worktree has the same effect with higher precedence
 271         */
 272        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 273                return 0;
 274        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 275                return 0;
 276
 277        /*
 278         * Intent-to-add entries have not been added, so the index entry
 279         * by definition never matches what is in the work tree until it
 280         * actually gets added.
 281         */
 282        if (ce->ce_flags & CE_INTENT_TO_ADD)
 283                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 284
 285        changed = ce_match_stat_basic(ce, st);
 286
 287        /*
 288         * Within 1 second of this sequence:
 289         *      echo xyzzy >file && git-update-index --add file
 290         * running this command:
 291         *      echo frotz >file
 292         * would give a falsely clean cache entry.  The mtime and
 293         * length match the cache, and other stat fields do not change.
 294         *
 295         * We could detect this at update-index time (the cache entry
 296         * being registered/updated records the same time as "now")
 297         * and delay the return from git-update-index, but that would
 298         * effectively mean we can make at most one commit per second,
 299         * which is not acceptable.  Instead, we check cache entries
 300         * whose mtime are the same as the index file timestamp more
 301         * carefully than others.
 302         */
 303        if (!changed && is_racy_timestamp(istate, ce)) {
 304                if (assume_racy_is_modified)
 305                        changed |= DATA_CHANGED;
 306                else
 307                        changed |= ce_modified_check_fs(ce, st);
 308        }
 309
 310        return changed;
 311}
 312
 313int ie_modified(const struct index_state *istate,
 314                struct cache_entry *ce, struct stat *st, unsigned int options)
 315{
 316        int changed, changed_fs;
 317
 318        changed = ie_match_stat(istate, ce, st, options);
 319        if (!changed)
 320                return 0;
 321        /*
 322         * If the mode or type has changed, there's no point in trying
 323         * to refresh the entry - it's not going to match
 324         */
 325        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 326                return changed;
 327
 328        /*
 329         * Immediately after read-tree or update-index --cacheinfo,
 330         * the length field is zero, as we have never even read the
 331         * lstat(2) information once, and we cannot trust DATA_CHANGED
 332         * returned by ie_match_stat() which in turn was returned by
 333         * ce_match_stat_basic() to signal that the filesize of the
 334         * blob changed.  We have to actually go to the filesystem to
 335         * see if the contents match, and if so, should answer "unchanged".
 336         *
 337         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 338         * already has checked the actual HEAD from the filesystem in the
 339         * subproject.  If ie_match_stat() already said it is different,
 340         * then we know it is.
 341         */
 342        if ((changed & DATA_CHANGED) &&
 343            (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
 344                return changed;
 345
 346        changed_fs = ce_modified_check_fs(ce, st);
 347        if (changed_fs)
 348                return changed | changed_fs;
 349        return 0;
 350}
 351
 352int base_name_compare(const char *name1, int len1, int mode1,
 353                      const char *name2, int len2, int mode2)
 354{
 355        unsigned char c1, c2;
 356        int len = len1 < len2 ? len1 : len2;
 357        int cmp;
 358
 359        cmp = memcmp(name1, name2, len);
 360        if (cmp)
 361                return cmp;
 362        c1 = name1[len];
 363        c2 = name2[len];
 364        if (!c1 && S_ISDIR(mode1))
 365                c1 = '/';
 366        if (!c2 && S_ISDIR(mode2))
 367                c2 = '/';
 368        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 369}
 370
 371/*
 372 * df_name_compare() is identical to base_name_compare(), except it
 373 * compares conflicting directory/file entries as equal. Note that
 374 * while a directory name compares as equal to a regular file, they
 375 * then individually compare _differently_ to a filename that has
 376 * a dot after the basename (because '\0' < '.' < '/').
 377 *
 378 * This is used by routines that want to traverse the git namespace
 379 * but then handle conflicting entries together when possible.
 380 */
 381int df_name_compare(const char *name1, int len1, int mode1,
 382                    const char *name2, int len2, int mode2)
 383{
 384        int len = len1 < len2 ? len1 : len2, cmp;
 385        unsigned char c1, c2;
 386
 387        cmp = memcmp(name1, name2, len);
 388        if (cmp)
 389                return cmp;
 390        /* Directories and files compare equal (same length, same name) */
 391        if (len1 == len2)
 392                return 0;
 393        c1 = name1[len];
 394        if (!c1 && S_ISDIR(mode1))
 395                c1 = '/';
 396        c2 = name2[len];
 397        if (!c2 && S_ISDIR(mode2))
 398                c2 = '/';
 399        if (c1 == '/' && !c2)
 400                return 0;
 401        if (c2 == '/' && !c1)
 402                return 0;
 403        return c1 - c2;
 404}
 405
 406int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 407{
 408        int len1 = flags1 & CE_NAMEMASK;
 409        int len2 = flags2 & CE_NAMEMASK;
 410        int len = len1 < len2 ? len1 : len2;
 411        int cmp;
 412
 413        cmp = memcmp(name1, name2, len);
 414        if (cmp)
 415                return cmp;
 416        if (len1 < len2)
 417                return -1;
 418        if (len1 > len2)
 419                return 1;
 420
 421        /* Compare stages  */
 422        flags1 &= CE_STAGEMASK;
 423        flags2 &= CE_STAGEMASK;
 424
 425        if (flags1 < flags2)
 426                return -1;
 427        if (flags1 > flags2)
 428                return 1;
 429        return 0;
 430}
 431
 432int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 433{
 434        int first, last;
 435
 436        first = 0;
 437        last = istate->cache_nr;
 438        while (last > first) {
 439                int next = (last + first) >> 1;
 440                struct cache_entry *ce = istate->cache[next];
 441                int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
 442                if (!cmp)
 443                        return next;
 444                if (cmp < 0) {
 445                        last = next;
 446                        continue;
 447                }
 448                first = next+1;
 449        }
 450        return -first-1;
 451}
 452
 453/* Remove entry, return true if there are more entries to go.. */
 454int remove_index_entry_at(struct index_state *istate, int pos)
 455{
 456        struct cache_entry *ce = istate->cache[pos];
 457
 458        record_resolve_undo(istate, ce);
 459        remove_name_hash(ce);
 460        istate->cache_changed = 1;
 461        istate->cache_nr--;
 462        if (pos >= istate->cache_nr)
 463                return 0;
 464        memmove(istate->cache + pos,
 465                istate->cache + pos + 1,
 466                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 467        return 1;
 468}
 469
 470/*
 471 * Remove all cache ententries marked for removal, that is where
 472 * CE_REMOVE is set in ce_flags.  This is much more effective than
 473 * calling remove_index_entry_at() for each entry to be removed.
 474 */
 475void remove_marked_cache_entries(struct index_state *istate)
 476{
 477        struct cache_entry **ce_array = istate->cache;
 478        unsigned int i, j;
 479
 480        for (i = j = 0; i < istate->cache_nr; i++) {
 481                if (ce_array[i]->ce_flags & CE_REMOVE)
 482                        remove_name_hash(ce_array[i]);
 483                else
 484                        ce_array[j++] = ce_array[i];
 485        }
 486        istate->cache_changed = 1;
 487        istate->cache_nr = j;
 488}
 489
 490int remove_file_from_index(struct index_state *istate, const char *path)
 491{
 492        int pos = index_name_pos(istate, path, strlen(path));
 493        if (pos < 0)
 494                pos = -pos-1;
 495        cache_tree_invalidate_path(istate->cache_tree, path);
 496        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 497                remove_index_entry_at(istate, pos);
 498        return 0;
 499}
 500
 501static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 502{
 503        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 504}
 505
 506static int index_name_pos_also_unmerged(struct index_state *istate,
 507        const char *path, int namelen)
 508{
 509        int pos = index_name_pos(istate, path, namelen);
 510        struct cache_entry *ce;
 511
 512        if (pos >= 0)
 513                return pos;
 514
 515        /* maybe unmerged? */
 516        pos = -1 - pos;
 517        if (pos >= istate->cache_nr ||
 518                        compare_name((ce = istate->cache[pos]), path, namelen))
 519                return -1;
 520
 521        /* order of preference: stage 2, 1, 3 */
 522        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 523                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 524                        !compare_name(ce, path, namelen))
 525                pos++;
 526        return pos;
 527}
 528
 529static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 530{
 531        int len = ce_namelen(ce);
 532        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 533}
 534
 535/*
 536 * If we add a filename that aliases in the cache, we will use the
 537 * name that we already have - but we don't want to update the same
 538 * alias twice, because that implies that there were actually two
 539 * different files with aliasing names!
 540 *
 541 * So we use the CE_ADDED flag to verify that the alias was an old
 542 * one before we accept it as
 543 */
 544static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
 545{
 546        int len;
 547        struct cache_entry *new;
 548
 549        if (alias->ce_flags & CE_ADDED)
 550                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 551
 552        /* Ok, create the new entry using the name of the existing alias */
 553        len = ce_namelen(alias);
 554        new = xcalloc(1, cache_entry_size(len));
 555        memcpy(new->name, alias->name, len);
 556        copy_cache_entry(new, ce);
 557        free(ce);
 558        return new;
 559}
 560
 561static void record_intent_to_add(struct cache_entry *ce)
 562{
 563        unsigned char sha1[20];
 564        if (write_sha1_file("", 0, blob_type, sha1))
 565                die("cannot create an empty blob in the object database");
 566        hashcpy(ce->sha1, sha1);
 567}
 568
 569int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 570{
 571        int size, namelen, was_same;
 572        mode_t st_mode = st->st_mode;
 573        struct cache_entry *ce, *alias;
 574        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 575        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 576        int pretend = flags & ADD_CACHE_PRETEND;
 577        int intent_only = flags & ADD_CACHE_INTENT;
 578        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 579                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 580
 581        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 582                return error("%s: can only add regular files, symbolic links or git-directories", path);
 583
 584        namelen = strlen(path);
 585        if (S_ISDIR(st_mode)) {
 586                while (namelen && path[namelen-1] == '/')
 587                        namelen--;
 588        }
 589        size = cache_entry_size(namelen);
 590        ce = xcalloc(1, size);
 591        memcpy(ce->name, path, namelen);
 592        ce->ce_flags = namelen;
 593        if (!intent_only)
 594                fill_stat_cache_info(ce, st);
 595        else
 596                ce->ce_flags |= CE_INTENT_TO_ADD;
 597
 598        if (trust_executable_bit && has_symlinks)
 599                ce->ce_mode = create_ce_mode(st_mode);
 600        else {
 601                /* If there is an existing entry, pick the mode bits and type
 602                 * from it, otherwise assume unexecutable regular file.
 603                 */
 604                struct cache_entry *ent;
 605                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 606
 607                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 608                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 609        }
 610
 611        alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 612        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 613                /* Nothing changed, really */
 614                free(ce);
 615                if (!S_ISGITLINK(alias->ce_mode))
 616                        ce_mark_uptodate(alias);
 617                alias->ce_flags |= CE_ADDED;
 618                return 0;
 619        }
 620        if (!intent_only) {
 621                if (index_path(ce->sha1, path, st, 1))
 622                        return error("unable to index file %s", path);
 623        } else
 624                record_intent_to_add(ce);
 625
 626        if (ignore_case && alias && different_name(ce, alias))
 627                ce = create_alias_ce(ce, alias);
 628        ce->ce_flags |= CE_ADDED;
 629
 630        /* It was suspected to be racily clean, but it turns out to be Ok */
 631        was_same = (alias &&
 632                    !ce_stage(alias) &&
 633                    !hashcmp(alias->sha1, ce->sha1) &&
 634                    ce->ce_mode == alias->ce_mode);
 635
 636        if (pretend)
 637                ;
 638        else if (add_index_entry(istate, ce, add_option))
 639                return error("unable to add %s to index",path);
 640        if (verbose && !was_same)
 641                printf("add '%s'\n", path);
 642        return 0;
 643}
 644
 645int add_file_to_index(struct index_state *istate, const char *path, int flags)
 646{
 647        struct stat st;
 648        if (lstat(path, &st))
 649                die_errno("unable to stat '%s'", path);
 650        return add_to_index(istate, path, &st, flags);
 651}
 652
 653struct cache_entry *make_cache_entry(unsigned int mode,
 654                const unsigned char *sha1, const char *path, int stage,
 655                int refresh)
 656{
 657        int size, len;
 658        struct cache_entry *ce;
 659
 660        if (!verify_path(path)) {
 661                error("Invalid path '%s'", path);
 662                return NULL;
 663        }
 664
 665        len = strlen(path);
 666        size = cache_entry_size(len);
 667        ce = xcalloc(1, size);
 668
 669        hashcpy(ce->sha1, sha1);
 670        memcpy(ce->name, path, len);
 671        ce->ce_flags = create_ce_flags(len, stage);
 672        ce->ce_mode = create_ce_mode(mode);
 673
 674        if (refresh)
 675                return refresh_cache_entry(ce, 0);
 676
 677        return ce;
 678}
 679
 680int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 681{
 682        int len = ce_namelen(a);
 683        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 684}
 685
 686int ce_path_match(const struct cache_entry *ce, const char **pathspec)
 687{
 688        const char *match, *name;
 689        int len;
 690
 691        if (!pathspec)
 692                return 1;
 693
 694        len = ce_namelen(ce);
 695        name = ce->name;
 696        while ((match = *pathspec++) != NULL) {
 697                int matchlen = strlen(match);
 698                if (matchlen > len)
 699                        continue;
 700                if (memcmp(name, match, matchlen))
 701                        continue;
 702                if (matchlen && name[matchlen-1] == '/')
 703                        return 1;
 704                if (name[matchlen] == '/' || !name[matchlen])
 705                        return 1;
 706                if (!matchlen)
 707                        return 1;
 708        }
 709        return 0;
 710}
 711
 712/*
 713 * We fundamentally don't like some paths: we don't want
 714 * dot or dot-dot anywhere, and for obvious reasons don't
 715 * want to recurse into ".git" either.
 716 *
 717 * Also, we don't want double slashes or slashes at the
 718 * end that can make pathnames ambiguous.
 719 */
 720static int verify_dotfile(const char *rest)
 721{
 722        /*
 723         * The first character was '.', but that
 724         * has already been discarded, we now test
 725         * the rest.
 726         */
 727        switch (*rest) {
 728        /* "." is not allowed */
 729        case '\0': case '/':
 730                return 0;
 731
 732        /*
 733         * ".git" followed by  NUL or slash is bad. This
 734         * shares the path end test with the ".." case.
 735         */
 736        case 'g':
 737                if (rest[1] != 'i')
 738                        break;
 739                if (rest[2] != 't')
 740                        break;
 741                rest += 2;
 742        /* fallthrough */
 743        case '.':
 744                if (rest[1] == '\0' || rest[1] == '/')
 745                        return 0;
 746        }
 747        return 1;
 748}
 749
 750int verify_path(const char *path)
 751{
 752        char c;
 753
 754        goto inside;
 755        for (;;) {
 756                if (!c)
 757                        return 1;
 758                if (c == '/') {
 759inside:
 760                        c = *path++;
 761                        switch (c) {
 762                        default:
 763                                continue;
 764                        case '/': case '\0':
 765                                break;
 766                        case '.':
 767                                if (verify_dotfile(path))
 768                                        continue;
 769                        }
 770                        return 0;
 771                }
 772                c = *path++;
 773        }
 774}
 775
 776/*
 777 * Do we have another file that has the beginning components being a
 778 * proper superset of the name we're trying to add?
 779 */
 780static int has_file_name(struct index_state *istate,
 781                         const struct cache_entry *ce, int pos, int ok_to_replace)
 782{
 783        int retval = 0;
 784        int len = ce_namelen(ce);
 785        int stage = ce_stage(ce);
 786        const char *name = ce->name;
 787
 788        while (pos < istate->cache_nr) {
 789                struct cache_entry *p = istate->cache[pos++];
 790
 791                if (len >= ce_namelen(p))
 792                        break;
 793                if (memcmp(name, p->name, len))
 794                        break;
 795                if (ce_stage(p) != stage)
 796                        continue;
 797                if (p->name[len] != '/')
 798                        continue;
 799                if (p->ce_flags & CE_REMOVE)
 800                        continue;
 801                retval = -1;
 802                if (!ok_to_replace)
 803                        break;
 804                remove_index_entry_at(istate, --pos);
 805        }
 806        return retval;
 807}
 808
 809/*
 810 * Do we have another file with a pathname that is a proper
 811 * subset of the name we're trying to add?
 812 */
 813static int has_dir_name(struct index_state *istate,
 814                        const struct cache_entry *ce, int pos, int ok_to_replace)
 815{
 816        int retval = 0;
 817        int stage = ce_stage(ce);
 818        const char *name = ce->name;
 819        const char *slash = name + ce_namelen(ce);
 820
 821        for (;;) {
 822                int len;
 823
 824                for (;;) {
 825                        if (*--slash == '/')
 826                                break;
 827                        if (slash <= ce->name)
 828                                return retval;
 829                }
 830                len = slash - name;
 831
 832                pos = index_name_pos(istate, name, create_ce_flags(len, stage));
 833                if (pos >= 0) {
 834                        /*
 835                         * Found one, but not so fast.  This could
 836                         * be a marker that says "I was here, but
 837                         * I am being removed".  Such an entry is
 838                         * not a part of the resulting tree, and
 839                         * it is Ok to have a directory at the same
 840                         * path.
 841                         */
 842                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 843                                retval = -1;
 844                                if (!ok_to_replace)
 845                                        break;
 846                                remove_index_entry_at(istate, pos);
 847                                continue;
 848                        }
 849                }
 850                else
 851                        pos = -pos-1;
 852
 853                /*
 854                 * Trivial optimization: if we find an entry that
 855                 * already matches the sub-directory, then we know
 856                 * we're ok, and we can exit.
 857                 */
 858                while (pos < istate->cache_nr) {
 859                        struct cache_entry *p = istate->cache[pos];
 860                        if ((ce_namelen(p) <= len) ||
 861                            (p->name[len] != '/') ||
 862                            memcmp(p->name, name, len))
 863                                break; /* not our subdirectory */
 864                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 865                                /*
 866                                 * p is at the same stage as our entry, and
 867                                 * is a subdirectory of what we are looking
 868                                 * at, so we cannot have conflicts at our
 869                                 * level or anything shorter.
 870                                 */
 871                                return retval;
 872                        pos++;
 873                }
 874        }
 875        return retval;
 876}
 877
 878/* We may be in a situation where we already have path/file and path
 879 * is being added, or we already have path and path/file is being
 880 * added.  Either one would result in a nonsense tree that has path
 881 * twice when git-write-tree tries to write it out.  Prevent it.
 882 *
 883 * If ok-to-replace is specified, we remove the conflicting entries
 884 * from the cache so the caller should recompute the insert position.
 885 * When this happens, we return non-zero.
 886 */
 887static int check_file_directory_conflict(struct index_state *istate,
 888                                         const struct cache_entry *ce,
 889                                         int pos, int ok_to_replace)
 890{
 891        int retval;
 892
 893        /*
 894         * When ce is an "I am going away" entry, we allow it to be added
 895         */
 896        if (ce->ce_flags & CE_REMOVE)
 897                return 0;
 898
 899        /*
 900         * We check if the path is a sub-path of a subsequent pathname
 901         * first, since removing those will not change the position
 902         * in the array.
 903         */
 904        retval = has_file_name(istate, ce, pos, ok_to_replace);
 905
 906        /*
 907         * Then check if the path might have a clashing sub-directory
 908         * before it.
 909         */
 910        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 911}
 912
 913static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 914{
 915        int pos;
 916        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 917        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 918        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 919        int new_only = option & ADD_CACHE_NEW_ONLY;
 920
 921        cache_tree_invalidate_path(istate->cache_tree, ce->name);
 922        pos = index_name_pos(istate, ce->name, ce->ce_flags);
 923
 924        /* existing match? Just replace it. */
 925        if (pos >= 0) {
 926                if (!new_only)
 927                        replace_index_entry(istate, pos, ce);
 928                return 0;
 929        }
 930        pos = -pos-1;
 931
 932        /*
 933         * Inserting a merged entry ("stage 0") into the index
 934         * will always replace all non-merged entries..
 935         */
 936        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 937                while (ce_same_name(istate->cache[pos], ce)) {
 938                        ok_to_add = 1;
 939                        if (!remove_index_entry_at(istate, pos))
 940                                break;
 941                }
 942        }
 943
 944        if (!ok_to_add)
 945                return -1;
 946        if (!verify_path(ce->name))
 947                return error("Invalid path '%s'", ce->name);
 948
 949        if (!skip_df_check &&
 950            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 951                if (!ok_to_replace)
 952                        return error("'%s' appears as both a file and as a directory",
 953                                     ce->name);
 954                pos = index_name_pos(istate, ce->name, ce->ce_flags);
 955                pos = -pos-1;
 956        }
 957        return pos + 1;
 958}
 959
 960int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 961{
 962        int pos;
 963
 964        if (option & ADD_CACHE_JUST_APPEND)
 965                pos = istate->cache_nr;
 966        else {
 967                int ret;
 968                ret = add_index_entry_with_check(istate, ce, option);
 969                if (ret <= 0)
 970                        return ret;
 971                pos = ret - 1;
 972        }
 973
 974        /* Make sure the array is big enough .. */
 975        if (istate->cache_nr == istate->cache_alloc) {
 976                istate->cache_alloc = alloc_nr(istate->cache_alloc);
 977                istate->cache = xrealloc(istate->cache,
 978                                        istate->cache_alloc * sizeof(struct cache_entry *));
 979        }
 980
 981        /* Add it in.. */
 982        istate->cache_nr++;
 983        if (istate->cache_nr > pos + 1)
 984                memmove(istate->cache + pos + 1,
 985                        istate->cache + pos,
 986                        (istate->cache_nr - pos - 1) * sizeof(ce));
 987        set_index_entry(istate, pos, ce);
 988        istate->cache_changed = 1;
 989        return 0;
 990}
 991
 992/*
 993 * "refresh" does not calculate a new sha1 file or bring the
 994 * cache up-to-date for mode/content changes. But what it
 995 * _does_ do is to "re-match" the stat information of a file
 996 * with the cache, so that you can refresh the cache for a
 997 * file that hasn't been changed but where the stat entry is
 998 * out of date.
 999 *
1000 * For example, you'd want to do this after doing a "git-read-tree",
1001 * to link up the stat cache details with the proper files.
1002 */
1003static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1004                                             struct cache_entry *ce,
1005                                             unsigned int options, int *err)
1006{
1007        struct stat st;
1008        struct cache_entry *updated;
1009        int changed, size;
1010        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1011        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1012
1013        if (ce_uptodate(ce))
1014                return ce;
1015
1016        /*
1017         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1018         * that the change to the work tree does not matter and told
1019         * us not to worry.
1020         */
1021        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1022                ce_mark_uptodate(ce);
1023                return ce;
1024        }
1025        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1026                ce_mark_uptodate(ce);
1027                return ce;
1028        }
1029
1030        if (lstat(ce->name, &st) < 0) {
1031                if (err)
1032                        *err = errno;
1033                return NULL;
1034        }
1035
1036        changed = ie_match_stat(istate, ce, &st, options);
1037        if (!changed) {
1038                /*
1039                 * The path is unchanged.  If we were told to ignore
1040                 * valid bit, then we did the actual stat check and
1041                 * found that the entry is unmodified.  If the entry
1042                 * is not marked VALID, this is the place to mark it
1043                 * valid again, under "assume unchanged" mode.
1044                 */
1045                if (ignore_valid && assume_unchanged &&
1046                    !(ce->ce_flags & CE_VALID))
1047                        ; /* mark this one VALID again */
1048                else {
1049                        /*
1050                         * We do not mark the index itself "modified"
1051                         * because CE_UPTODATE flag is in-core only;
1052                         * we are not going to write this change out.
1053                         */
1054                        if (!S_ISGITLINK(ce->ce_mode))
1055                                ce_mark_uptodate(ce);
1056                        return ce;
1057                }
1058        }
1059
1060        if (ie_modified(istate, ce, &st, options)) {
1061                if (err)
1062                        *err = EINVAL;
1063                return NULL;
1064        }
1065
1066        size = ce_size(ce);
1067        updated = xmalloc(size);
1068        memcpy(updated, ce, size);
1069        fill_stat_cache_info(updated, &st);
1070        /*
1071         * If ignore_valid is not set, we should leave CE_VALID bit
1072         * alone.  Otherwise, paths marked with --no-assume-unchanged
1073         * (i.e. things to be edited) will reacquire CE_VALID bit
1074         * automatically, which is not really what we want.
1075         */
1076        if (!ignore_valid && assume_unchanged &&
1077            !(ce->ce_flags & CE_VALID))
1078                updated->ce_flags &= ~CE_VALID;
1079
1080        return updated;
1081}
1082
1083static void show_file(const char * fmt, const char * name, int in_porcelain,
1084                      int * first, char *header_msg)
1085{
1086        if (in_porcelain && *first && header_msg) {
1087                printf("%s\n", header_msg);
1088                *first=0;
1089        }
1090        printf(fmt, name);
1091}
1092
1093int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1094                  char *seen, char *header_msg)
1095{
1096        int i;
1097        int has_errors = 0;
1098        int really = (flags & REFRESH_REALLY) != 0;
1099        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1100        int quiet = (flags & REFRESH_QUIET) != 0;
1101        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1102        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1103        int first = 1;
1104        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1105        unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1106        const char *needs_update_fmt;
1107        const char *needs_merge_fmt;
1108
1109        needs_update_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1110        needs_merge_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1111        for (i = 0; i < istate->cache_nr; i++) {
1112                struct cache_entry *ce, *new;
1113                int cache_errno = 0;
1114
1115                ce = istate->cache[i];
1116                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1117                        continue;
1118
1119                if (ce_stage(ce)) {
1120                        while ((i < istate->cache_nr) &&
1121                               ! strcmp(istate->cache[i]->name, ce->name))
1122                                i++;
1123                        i--;
1124                        if (allow_unmerged)
1125                                continue;
1126                        show_file(needs_merge_fmt, ce->name, in_porcelain, &first, header_msg);
1127                        has_errors = 1;
1128                        continue;
1129                }
1130
1131                if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1132                        continue;
1133
1134                new = refresh_cache_ent(istate, ce, options, &cache_errno);
1135                if (new == ce)
1136                        continue;
1137                if (!new) {
1138                        if (not_new && cache_errno == ENOENT)
1139                                continue;
1140                        if (really && cache_errno == EINVAL) {
1141                                /* If we are doing --really-refresh that
1142                                 * means the index is not valid anymore.
1143                                 */
1144                                ce->ce_flags &= ~CE_VALID;
1145                                istate->cache_changed = 1;
1146                        }
1147                        if (quiet)
1148                                continue;
1149                        show_file(needs_update_fmt, ce->name, in_porcelain, &first, header_msg);
1150                        has_errors = 1;
1151                        continue;
1152                }
1153
1154                replace_index_entry(istate, i, new);
1155        }
1156        return has_errors;
1157}
1158
1159static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1160{
1161        return refresh_cache_ent(&the_index, ce, really, NULL);
1162}
1163
1164static int verify_hdr(struct cache_header *hdr, unsigned long size)
1165{
1166        git_SHA_CTX c;
1167        unsigned char sha1[20];
1168
1169        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1170                return error("bad signature");
1171        if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
1172                return error("bad index version");
1173        git_SHA1_Init(&c);
1174        git_SHA1_Update(&c, hdr, size - 20);
1175        git_SHA1_Final(sha1, &c);
1176        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1177                return error("bad index file sha1 signature");
1178        return 0;
1179}
1180
1181static int read_index_extension(struct index_state *istate,
1182                                const char *ext, void *data, unsigned long sz)
1183{
1184        switch (CACHE_EXT(ext)) {
1185        case CACHE_EXT_TREE:
1186                istate->cache_tree = cache_tree_read(data, sz);
1187                break;
1188        case CACHE_EXT_RESOLVE_UNDO:
1189                istate->resolve_undo = resolve_undo_read(data, sz);
1190                break;
1191        default:
1192                if (*ext < 'A' || 'Z' < *ext)
1193                        return error("index uses %.4s extension, which we do not understand",
1194                                     ext);
1195                fprintf(stderr, "ignoring %.4s extension\n", ext);
1196                break;
1197        }
1198        return 0;
1199}
1200
1201int read_index(struct index_state *istate)
1202{
1203        return read_index_from(istate, get_index_file());
1204}
1205
1206static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1207{
1208        size_t len;
1209        const char *name;
1210
1211        ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);
1212        ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);
1213        ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);
1214        ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);
1215        ce->ce_dev   = ntohl(ondisk->dev);
1216        ce->ce_ino   = ntohl(ondisk->ino);
1217        ce->ce_mode  = ntohl(ondisk->mode);
1218        ce->ce_uid   = ntohl(ondisk->uid);
1219        ce->ce_gid   = ntohl(ondisk->gid);
1220        ce->ce_size  = ntohl(ondisk->size);
1221        /* On-disk flags are just 16 bits */
1222        ce->ce_flags = ntohs(ondisk->flags);
1223
1224        hashcpy(ce->sha1, ondisk->sha1);
1225
1226        len = ce->ce_flags & CE_NAMEMASK;
1227
1228        if (ce->ce_flags & CE_EXTENDED) {
1229                struct ondisk_cache_entry_extended *ondisk2;
1230                int extended_flags;
1231                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1232                extended_flags = ntohs(ondisk2->flags2) << 16;
1233                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1234                if (extended_flags & ~CE_EXTENDED_FLAGS)
1235                        die("Unknown index entry format %08x", extended_flags);
1236                ce->ce_flags |= extended_flags;
1237                name = ondisk2->name;
1238        }
1239        else
1240                name = ondisk->name;
1241
1242        if (len == CE_NAMEMASK)
1243                len = strlen(name);
1244        /*
1245         * NEEDSWORK: If the original index is crafted, this copy could
1246         * go unchecked.
1247         */
1248        memcpy(ce->name, name, len + 1);
1249}
1250
1251static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1252{
1253        long per_entry;
1254
1255        per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1256
1257        /*
1258         * Alignment can cause differences. This should be "alignof", but
1259         * since that's a gcc'ism, just use the size of a pointer.
1260         */
1261        per_entry += sizeof(void *);
1262        return ondisk_size + entries*per_entry;
1263}
1264
1265/* remember to discard_cache() before reading a different cache! */
1266int read_index_from(struct index_state *istate, const char *path)
1267{
1268        int fd, i;
1269        struct stat st;
1270        unsigned long src_offset, dst_offset;
1271        struct cache_header *hdr;
1272        void *mmap;
1273        size_t mmap_size;
1274
1275        errno = EBUSY;
1276        if (istate->initialized)
1277                return istate->cache_nr;
1278
1279        errno = ENOENT;
1280        istate->timestamp.sec = 0;
1281        istate->timestamp.nsec = 0;
1282        fd = open(path, O_RDONLY);
1283        if (fd < 0) {
1284                if (errno == ENOENT)
1285                        return 0;
1286                die_errno("index file open failed");
1287        }
1288
1289        if (fstat(fd, &st))
1290                die_errno("cannot stat the open index");
1291
1292        errno = EINVAL;
1293        mmap_size = xsize_t(st.st_size);
1294        if (mmap_size < sizeof(struct cache_header) + 20)
1295                die("index file smaller than expected");
1296
1297        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1298        close(fd);
1299        if (mmap == MAP_FAILED)
1300                die_errno("unable to map index file");
1301
1302        hdr = mmap;
1303        if (verify_hdr(hdr, mmap_size) < 0)
1304                goto unmap;
1305
1306        istate->cache_nr = ntohl(hdr->hdr_entries);
1307        istate->cache_alloc = alloc_nr(istate->cache_nr);
1308        istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1309
1310        /*
1311         * The disk format is actually larger than the in-memory format,
1312         * due to space for nsec etc, so even though the in-memory one
1313         * has room for a few  more flags, we can allocate using the same
1314         * index size
1315         */
1316        istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1317        istate->initialized = 1;
1318
1319        src_offset = sizeof(*hdr);
1320        dst_offset = 0;
1321        for (i = 0; i < istate->cache_nr; i++) {
1322                struct ondisk_cache_entry *disk_ce;
1323                struct cache_entry *ce;
1324
1325                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1326                ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1327                convert_from_disk(disk_ce, ce);
1328                set_index_entry(istate, i, ce);
1329
1330                src_offset += ondisk_ce_size(ce);
1331                dst_offset += ce_size(ce);
1332        }
1333        istate->timestamp.sec = st.st_mtime;
1334        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1335
1336        while (src_offset <= mmap_size - 20 - 8) {
1337                /* After an array of active_nr index entries,
1338                 * there can be arbitrary number of extended
1339                 * sections, each of which is prefixed with
1340                 * extension name (4-byte) and section length
1341                 * in 4-byte network byte order.
1342                 */
1343                uint32_t extsize;
1344                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1345                extsize = ntohl(extsize);
1346                if (read_index_extension(istate,
1347                                         (const char *) mmap + src_offset,
1348                                         (char *) mmap + src_offset + 8,
1349                                         extsize) < 0)
1350                        goto unmap;
1351                src_offset += 8;
1352                src_offset += extsize;
1353        }
1354        munmap(mmap, mmap_size);
1355        return istate->cache_nr;
1356
1357unmap:
1358        munmap(mmap, mmap_size);
1359        errno = EINVAL;
1360        die("index file corrupt");
1361}
1362
1363int is_index_unborn(struct index_state *istate)
1364{
1365        return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);
1366}
1367
1368int discard_index(struct index_state *istate)
1369{
1370        resolve_undo_clear_index(istate);
1371        istate->cache_nr = 0;
1372        istate->cache_changed = 0;
1373        istate->timestamp.sec = 0;
1374        istate->timestamp.nsec = 0;
1375        istate->name_hash_initialized = 0;
1376        free_hash(&istate->name_hash);
1377        cache_tree_free(&(istate->cache_tree));
1378        free(istate->alloc);
1379        istate->alloc = NULL;
1380        istate->initialized = 0;
1381
1382        /* no need to throw away allocated active_cache */
1383        return 0;
1384}
1385
1386int unmerged_index(const struct index_state *istate)
1387{
1388        int i;
1389        for (i = 0; i < istate->cache_nr; i++) {
1390                if (ce_stage(istate->cache[i]))
1391                        return 1;
1392        }
1393        return 0;
1394}
1395
1396#define WRITE_BUFFER_SIZE 8192
1397static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1398static unsigned long write_buffer_len;
1399
1400static int ce_write_flush(git_SHA_CTX *context, int fd)
1401{
1402        unsigned int buffered = write_buffer_len;
1403        if (buffered) {
1404                git_SHA1_Update(context, write_buffer, buffered);
1405                if (write_in_full(fd, write_buffer, buffered) != buffered)
1406                        return -1;
1407                write_buffer_len = 0;
1408        }
1409        return 0;
1410}
1411
1412static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1413{
1414        while (len) {
1415                unsigned int buffered = write_buffer_len;
1416                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1417                if (partial > len)
1418                        partial = len;
1419                memcpy(write_buffer + buffered, data, partial);
1420                buffered += partial;
1421                if (buffered == WRITE_BUFFER_SIZE) {
1422                        write_buffer_len = buffered;
1423                        if (ce_write_flush(context, fd))
1424                                return -1;
1425                        buffered = 0;
1426                }
1427                write_buffer_len = buffered;
1428                len -= partial;
1429                data = (char *) data + partial;
1430        }
1431        return 0;
1432}
1433
1434static int write_index_ext_header(git_SHA_CTX *context, int fd,
1435                                  unsigned int ext, unsigned int sz)
1436{
1437        ext = htonl(ext);
1438        sz = htonl(sz);
1439        return ((ce_write(context, fd, &ext, 4) < 0) ||
1440                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1441}
1442
1443static int ce_flush(git_SHA_CTX *context, int fd)
1444{
1445        unsigned int left = write_buffer_len;
1446
1447        if (left) {
1448                write_buffer_len = 0;
1449                git_SHA1_Update(context, write_buffer, left);
1450        }
1451
1452        /* Flush first if not enough space for SHA1 signature */
1453        if (left + 20 > WRITE_BUFFER_SIZE) {
1454                if (write_in_full(fd, write_buffer, left) != left)
1455                        return -1;
1456                left = 0;
1457        }
1458
1459        /* Append the SHA1 signature at the end */
1460        git_SHA1_Final(write_buffer + left, context);
1461        left += 20;
1462        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1463}
1464
1465static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1466{
1467        /*
1468         * The only thing we care about in this function is to smudge the
1469         * falsely clean entry due to touch-update-touch race, so we leave
1470         * everything else as they are.  We are called for entries whose
1471         * ce_mtime match the index file mtime.
1472         *
1473         * Note that this actually does not do much for gitlinks, for
1474         * which ce_match_stat_basic() always goes to the actual
1475         * contents.  The caller checks with is_racy_timestamp() which
1476         * always says "no" for gitlinks, so we are not called for them ;-)
1477         */
1478        struct stat st;
1479
1480        if (lstat(ce->name, &st) < 0)
1481                return;
1482        if (ce_match_stat_basic(ce, &st))
1483                return;
1484        if (ce_modified_check_fs(ce, &st)) {
1485                /* This is "racily clean"; smudge it.  Note that this
1486                 * is a tricky code.  At first glance, it may appear
1487                 * that it can break with this sequence:
1488                 *
1489                 * $ echo xyzzy >frotz
1490                 * $ git-update-index --add frotz
1491                 * $ : >frotz
1492                 * $ sleep 3
1493                 * $ echo filfre >nitfol
1494                 * $ git-update-index --add nitfol
1495                 *
1496                 * but it does not.  When the second update-index runs,
1497                 * it notices that the entry "frotz" has the same timestamp
1498                 * as index, and if we were to smudge it by resetting its
1499                 * size to zero here, then the object name recorded
1500                 * in index is the 6-byte file but the cached stat information
1501                 * becomes zero --- which would then match what we would
1502                 * obtain from the filesystem next time we stat("frotz").
1503                 *
1504                 * However, the second update-index, before calling
1505                 * this function, notices that the cached size is 6
1506                 * bytes and what is on the filesystem is an empty
1507                 * file, and never calls us, so the cached size information
1508                 * for "frotz" stays 6 which does not match the filesystem.
1509                 */
1510                ce->ce_size = 0;
1511        }
1512}
1513
1514static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
1515{
1516        int size = ondisk_ce_size(ce);
1517        struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1518        char *name;
1519        int result;
1520
1521        ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1522        ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1523        ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1524        ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1525        ondisk->dev  = htonl(ce->ce_dev);
1526        ondisk->ino  = htonl(ce->ce_ino);
1527        ondisk->mode = htonl(ce->ce_mode);
1528        ondisk->uid  = htonl(ce->ce_uid);
1529        ondisk->gid  = htonl(ce->ce_gid);
1530        ondisk->size = htonl(ce->ce_size);
1531        hashcpy(ondisk->sha1, ce->sha1);
1532        ondisk->flags = htons(ce->ce_flags);
1533        if (ce->ce_flags & CE_EXTENDED) {
1534                struct ondisk_cache_entry_extended *ondisk2;
1535                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1536                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1537                name = ondisk2->name;
1538        }
1539        else
1540                name = ondisk->name;
1541        memcpy(name, ce->name, ce_namelen(ce));
1542
1543        result = ce_write(c, fd, ondisk, size);
1544        free(ondisk);
1545        return result;
1546}
1547
1548int write_index(struct index_state *istate, int newfd)
1549{
1550        git_SHA_CTX c;
1551        struct cache_header hdr;
1552        int i, err, removed, extended;
1553        struct cache_entry **cache = istate->cache;
1554        int entries = istate->cache_nr;
1555        struct stat st;
1556
1557        for (i = removed = extended = 0; i < entries; i++) {
1558                if (cache[i]->ce_flags & CE_REMOVE)
1559                        removed++;
1560
1561                /* reduce extended entries if possible */
1562                cache[i]->ce_flags &= ~CE_EXTENDED;
1563                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1564                        extended++;
1565                        cache[i]->ce_flags |= CE_EXTENDED;
1566                }
1567        }
1568
1569        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1570        /* for extended format, increase version so older git won't try to read it */
1571        hdr.hdr_version = htonl(extended ? 3 : 2);
1572        hdr.hdr_entries = htonl(entries - removed);
1573
1574        git_SHA1_Init(&c);
1575        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1576                return -1;
1577
1578        for (i = 0; i < entries; i++) {
1579                struct cache_entry *ce = cache[i];
1580                if (ce->ce_flags & CE_REMOVE)
1581                        continue;
1582                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1583                        ce_smudge_racily_clean_entry(ce);
1584                if (ce_write_entry(&c, newfd, ce) < 0)
1585                        return -1;
1586        }
1587
1588        /* Write extension data here */
1589        if (istate->cache_tree) {
1590                struct strbuf sb = STRBUF_INIT;
1591
1592                cache_tree_write(&sb, istate->cache_tree);
1593                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1594                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1595                strbuf_release(&sb);
1596                if (err)
1597                        return -1;
1598        }
1599        if (istate->resolve_undo) {
1600                struct strbuf sb = STRBUF_INIT;
1601
1602                resolve_undo_write(&sb, istate->resolve_undo);
1603                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1604                                             sb.len) < 0
1605                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1606                strbuf_release(&sb);
1607                if (err)
1608                        return -1;
1609        }
1610
1611        if (ce_flush(&c, newfd) || fstat(newfd, &st))
1612                return -1;
1613        istate->timestamp.sec = (unsigned int)st.st_mtime;
1614        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1615        return 0;
1616}
1617
1618/*
1619 * Read the index file that is potentially unmerged into given
1620 * index_state, dropping any unmerged entries.  Returns true if
1621 * the index is unmerged.  Callers who want to refuse to work
1622 * from an unmerged state can call this and check its return value,
1623 * instead of calling read_cache().
1624 */
1625int read_index_unmerged(struct index_state *istate)
1626{
1627        int i;
1628        int unmerged = 0;
1629
1630        read_index(istate);
1631        for (i = 0; i < istate->cache_nr; i++) {
1632                struct cache_entry *ce = istate->cache[i];
1633                struct cache_entry *new_ce;
1634                int size, len;
1635
1636                if (!ce_stage(ce))
1637                        continue;
1638                unmerged = 1;
1639                len = strlen(ce->name);
1640                size = cache_entry_size(len);
1641                new_ce = xcalloc(1, size);
1642                memcpy(new_ce->name, ce->name, len);
1643                new_ce->ce_flags = create_ce_flags(len, 0) | CE_CONFLICTED;
1644                new_ce->ce_mode = ce->ce_mode;
1645                if (add_index_entry(istate, new_ce, 0))
1646                        return error("%s: cannot drop to stage #0",
1647                                     ce->name);
1648                i = index_name_pos(istate, new_ce->name, len);
1649        }
1650        return unmerged;
1651}
1652
1653/*
1654 * Returns 1 if the path is an "other" path with respect to
1655 * the index; that is, the path is not mentioned in the index at all,
1656 * either as a file, a directory with some files in the index,
1657 * or as an unmerged entry.
1658 *
1659 * We helpfully remove a trailing "/" from directories so that
1660 * the output of read_directory can be used as-is.
1661 */
1662int index_name_is_other(const struct index_state *istate, const char *name,
1663                int namelen)
1664{
1665        int pos;
1666        if (namelen && name[namelen - 1] == '/')
1667                namelen--;
1668        pos = index_name_pos(istate, name, namelen);
1669        if (0 <= pos)
1670                return 0;       /* exact match */
1671        pos = -pos - 1;
1672        if (pos < istate->cache_nr) {
1673                struct cache_entry *ce = istate->cache[pos];
1674                if (ce_namelen(ce) == namelen &&
1675                    !memcmp(ce->name, name, namelen))
1676                        return 0; /* Yup, this one exists unmerged */
1677        }
1678        return 1;
1679}