read-cache.con commit Merge branch 'jk/format-person-part-buffer-limit' into maint (92ddfaa)
   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, OBJ_BLOB, ce->name, 0))
  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 ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 161{
 162        unsigned int changed = 0;
 163
 164        if (ce->ce_flags & CE_REMOVE)
 165                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 166
 167        switch (ce->ce_mode & S_IFMT) {
 168        case S_IFREG:
 169                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 170                /* We consider only the owner x bit to be relevant for
 171                 * "mode changes"
 172                 */
 173                if (trust_executable_bit &&
 174                    (0100 & (ce->ce_mode ^ st->st_mode)))
 175                        changed |= MODE_CHANGED;
 176                break;
 177        case S_IFLNK:
 178                if (!S_ISLNK(st->st_mode) &&
 179                    (has_symlinks || !S_ISREG(st->st_mode)))
 180                        changed |= TYPE_CHANGED;
 181                break;
 182        case S_IFGITLINK:
 183                /* We ignore most of the st_xxx fields for gitlinks */
 184                if (!S_ISDIR(st->st_mode))
 185                        changed |= TYPE_CHANGED;
 186                else if (ce_compare_gitlink(ce))
 187                        changed |= DATA_CHANGED;
 188                return changed;
 189        default:
 190                die("internal error: ce_mode is %o", ce->ce_mode);
 191        }
 192        if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
 193                changed |= MTIME_CHANGED;
 194        if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime)
 195                changed |= CTIME_CHANGED;
 196
 197#ifdef USE_NSEC
 198        if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
 199                changed |= MTIME_CHANGED;
 200        if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
 201                changed |= CTIME_CHANGED;
 202#endif
 203
 204        if (ce->ce_uid != (unsigned int) st->st_uid ||
 205            ce->ce_gid != (unsigned int) st->st_gid)
 206                changed |= OWNER_CHANGED;
 207        if (ce->ce_ino != (unsigned int) st->st_ino)
 208                changed |= INODE_CHANGED;
 209
 210#ifdef USE_STDEV
 211        /*
 212         * st_dev breaks on network filesystems where different
 213         * clients will have different views of what "device"
 214         * the filesystem is on
 215         */
 216        if (ce->ce_dev != (unsigned int) st->st_dev)
 217                changed |= INODE_CHANGED;
 218#endif
 219
 220        if (ce->ce_size != (unsigned int) st->st_size)
 221                changed |= DATA_CHANGED;
 222
 223        /* Racily smudged entry? */
 224        if (!ce->ce_size) {
 225                if (!is_empty_blob_sha1(ce->sha1))
 226                        changed |= DATA_CHANGED;
 227        }
 228
 229        return changed;
 230}
 231
 232static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
 233{
 234        return (!S_ISGITLINK(ce->ce_mode) &&
 235                istate->timestamp.sec &&
 236#ifdef USE_NSEC
 237                 /* nanosecond timestamped files can also be racy! */
 238                (istate->timestamp.sec < ce->ce_mtime.sec ||
 239                 (istate->timestamp.sec == ce->ce_mtime.sec &&
 240                  istate->timestamp.nsec <= ce->ce_mtime.nsec))
 241#else
 242                istate->timestamp.sec <= ce->ce_mtime.sec
 243#endif
 244                 );
 245}
 246
 247int ie_match_stat(const struct index_state *istate,
 248                  struct cache_entry *ce, struct stat *st,
 249                  unsigned int options)
 250{
 251        unsigned int changed;
 252        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 253        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 254        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 255
 256        /*
 257         * If it's marked as always valid in the index, it's
 258         * valid whatever the checked-out copy says.
 259         *
 260         * skip-worktree has the same effect with higher precedence
 261         */
 262        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 263                return 0;
 264        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 265                return 0;
 266
 267        /*
 268         * Intent-to-add entries have not been added, so the index entry
 269         * by definition never matches what is in the work tree until it
 270         * actually gets added.
 271         */
 272        if (ce->ce_flags & CE_INTENT_TO_ADD)
 273                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 274
 275        changed = ce_match_stat_basic(ce, st);
 276
 277        /*
 278         * Within 1 second of this sequence:
 279         *      echo xyzzy >file && git-update-index --add file
 280         * running this command:
 281         *      echo frotz >file
 282         * would give a falsely clean cache entry.  The mtime and
 283         * length match the cache, and other stat fields do not change.
 284         *
 285         * We could detect this at update-index time (the cache entry
 286         * being registered/updated records the same time as "now")
 287         * and delay the return from git-update-index, but that would
 288         * effectively mean we can make at most one commit per second,
 289         * which is not acceptable.  Instead, we check cache entries
 290         * whose mtime are the same as the index file timestamp more
 291         * carefully than others.
 292         */
 293        if (!changed && is_racy_timestamp(istate, ce)) {
 294                if (assume_racy_is_modified)
 295                        changed |= DATA_CHANGED;
 296                else
 297                        changed |= ce_modified_check_fs(ce, st);
 298        }
 299
 300        return changed;
 301}
 302
 303int ie_modified(const struct index_state *istate,
 304                struct cache_entry *ce, struct stat *st, unsigned int options)
 305{
 306        int changed, changed_fs;
 307
 308        changed = ie_match_stat(istate, ce, st, options);
 309        if (!changed)
 310                return 0;
 311        /*
 312         * If the mode or type has changed, there's no point in trying
 313         * to refresh the entry - it's not going to match
 314         */
 315        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 316                return changed;
 317
 318        /*
 319         * Immediately after read-tree or update-index --cacheinfo,
 320         * the length field is zero, as we have never even read the
 321         * lstat(2) information once, and we cannot trust DATA_CHANGED
 322         * returned by ie_match_stat() which in turn was returned by
 323         * ce_match_stat_basic() to signal that the filesize of the
 324         * blob changed.  We have to actually go to the filesystem to
 325         * see if the contents match, and if so, should answer "unchanged".
 326         *
 327         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 328         * already has checked the actual HEAD from the filesystem in the
 329         * subproject.  If ie_match_stat() already said it is different,
 330         * then we know it is.
 331         */
 332        if ((changed & DATA_CHANGED) &&
 333            (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
 334                return changed;
 335
 336        changed_fs = ce_modified_check_fs(ce, st);
 337        if (changed_fs)
 338                return changed | changed_fs;
 339        return 0;
 340}
 341
 342int base_name_compare(const char *name1, int len1, int mode1,
 343                      const char *name2, int len2, int mode2)
 344{
 345        unsigned char c1, c2;
 346        int len = len1 < len2 ? len1 : len2;
 347        int cmp;
 348
 349        cmp = memcmp(name1, name2, len);
 350        if (cmp)
 351                return cmp;
 352        c1 = name1[len];
 353        c2 = name2[len];
 354        if (!c1 && S_ISDIR(mode1))
 355                c1 = '/';
 356        if (!c2 && S_ISDIR(mode2))
 357                c2 = '/';
 358        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 359}
 360
 361/*
 362 * df_name_compare() is identical to base_name_compare(), except it
 363 * compares conflicting directory/file entries as equal. Note that
 364 * while a directory name compares as equal to a regular file, they
 365 * then individually compare _differently_ to a filename that has
 366 * a dot after the basename (because '\0' < '.' < '/').
 367 *
 368 * This is used by routines that want to traverse the git namespace
 369 * but then handle conflicting entries together when possible.
 370 */
 371int df_name_compare(const char *name1, int len1, int mode1,
 372                    const char *name2, int len2, int mode2)
 373{
 374        int len = len1 < len2 ? len1 : len2, cmp;
 375        unsigned char c1, c2;
 376
 377        cmp = memcmp(name1, name2, len);
 378        if (cmp)
 379                return cmp;
 380        /* Directories and files compare equal (same length, same name) */
 381        if (len1 == len2)
 382                return 0;
 383        c1 = name1[len];
 384        if (!c1 && S_ISDIR(mode1))
 385                c1 = '/';
 386        c2 = name2[len];
 387        if (!c2 && S_ISDIR(mode2))
 388                c2 = '/';
 389        if (c1 == '/' && !c2)
 390                return 0;
 391        if (c2 == '/' && !c1)
 392                return 0;
 393        return c1 - c2;
 394}
 395
 396int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 397{
 398        int len1 = flags1 & CE_NAMEMASK;
 399        int len2 = flags2 & CE_NAMEMASK;
 400        int len = len1 < len2 ? len1 : len2;
 401        int cmp;
 402
 403        cmp = memcmp(name1, name2, len);
 404        if (cmp)
 405                return cmp;
 406        if (len1 < len2)
 407                return -1;
 408        if (len1 > len2)
 409                return 1;
 410
 411        /* Compare stages  */
 412        flags1 &= CE_STAGEMASK;
 413        flags2 &= CE_STAGEMASK;
 414
 415        if (flags1 < flags2)
 416                return -1;
 417        if (flags1 > flags2)
 418                return 1;
 419        return 0;
 420}
 421
 422int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 423{
 424        int first, last;
 425
 426        first = 0;
 427        last = istate->cache_nr;
 428        while (last > first) {
 429                int next = (last + first) >> 1;
 430                struct cache_entry *ce = istate->cache[next];
 431                int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
 432                if (!cmp)
 433                        return next;
 434                if (cmp < 0) {
 435                        last = next;
 436                        continue;
 437                }
 438                first = next+1;
 439        }
 440        return -first-1;
 441}
 442
 443/* Remove entry, return true if there are more entries to go.. */
 444int remove_index_entry_at(struct index_state *istate, int pos)
 445{
 446        struct cache_entry *ce = istate->cache[pos];
 447
 448        record_resolve_undo(istate, ce);
 449        remove_name_hash(ce);
 450        istate->cache_changed = 1;
 451        istate->cache_nr--;
 452        if (pos >= istate->cache_nr)
 453                return 0;
 454        memmove(istate->cache + pos,
 455                istate->cache + pos + 1,
 456                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 457        return 1;
 458}
 459
 460/*
 461 * Remove all cache ententries marked for removal, that is where
 462 * CE_REMOVE is set in ce_flags.  This is much more effective than
 463 * calling remove_index_entry_at() for each entry to be removed.
 464 */
 465void remove_marked_cache_entries(struct index_state *istate)
 466{
 467        struct cache_entry **ce_array = istate->cache;
 468        unsigned int i, j;
 469
 470        for (i = j = 0; i < istate->cache_nr; i++) {
 471                if (ce_array[i]->ce_flags & CE_REMOVE)
 472                        remove_name_hash(ce_array[i]);
 473                else
 474                        ce_array[j++] = ce_array[i];
 475        }
 476        istate->cache_changed = 1;
 477        istate->cache_nr = j;
 478}
 479
 480int remove_file_from_index(struct index_state *istate, const char *path)
 481{
 482        int pos = index_name_pos(istate, path, strlen(path));
 483        if (pos < 0)
 484                pos = -pos-1;
 485        cache_tree_invalidate_path(istate->cache_tree, path);
 486        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 487                remove_index_entry_at(istate, pos);
 488        return 0;
 489}
 490
 491static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 492{
 493        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 494}
 495
 496static int index_name_pos_also_unmerged(struct index_state *istate,
 497        const char *path, int namelen)
 498{
 499        int pos = index_name_pos(istate, path, namelen);
 500        struct cache_entry *ce;
 501
 502        if (pos >= 0)
 503                return pos;
 504
 505        /* maybe unmerged? */
 506        pos = -1 - pos;
 507        if (pos >= istate->cache_nr ||
 508                        compare_name((ce = istate->cache[pos]), path, namelen))
 509                return -1;
 510
 511        /* order of preference: stage 2, 1, 3 */
 512        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 513                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 514                        !compare_name(ce, path, namelen))
 515                pos++;
 516        return pos;
 517}
 518
 519static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 520{
 521        int len = ce_namelen(ce);
 522        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 523}
 524
 525/*
 526 * If we add a filename that aliases in the cache, we will use the
 527 * name that we already have - but we don't want to update the same
 528 * alias twice, because that implies that there were actually two
 529 * different files with aliasing names!
 530 *
 531 * So we use the CE_ADDED flag to verify that the alias was an old
 532 * one before we accept it as
 533 */
 534static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
 535{
 536        int len;
 537        struct cache_entry *new;
 538
 539        if (alias->ce_flags & CE_ADDED)
 540                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 541
 542        /* Ok, create the new entry using the name of the existing alias */
 543        len = ce_namelen(alias);
 544        new = xcalloc(1, cache_entry_size(len));
 545        memcpy(new->name, alias->name, len);
 546        copy_cache_entry(new, ce);
 547        free(ce);
 548        return new;
 549}
 550
 551static void record_intent_to_add(struct cache_entry *ce)
 552{
 553        unsigned char sha1[20];
 554        if (write_sha1_file("", 0, blob_type, sha1))
 555                die("cannot create an empty blob in the object database");
 556        hashcpy(ce->sha1, sha1);
 557}
 558
 559int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 560{
 561        int size, namelen, was_same;
 562        mode_t st_mode = st->st_mode;
 563        struct cache_entry *ce, *alias;
 564        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 565        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 566        int pretend = flags & ADD_CACHE_PRETEND;
 567        int intent_only = flags & ADD_CACHE_INTENT;
 568        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 569                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 570
 571        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 572                return error("%s: can only add regular files, symbolic links or git-directories", path);
 573
 574        namelen = strlen(path);
 575        if (S_ISDIR(st_mode)) {
 576                while (namelen && path[namelen-1] == '/')
 577                        namelen--;
 578        }
 579        size = cache_entry_size(namelen);
 580        ce = xcalloc(1, size);
 581        memcpy(ce->name, path, namelen);
 582        ce->ce_flags = namelen;
 583        if (!intent_only)
 584                fill_stat_cache_info(ce, st);
 585        else
 586                ce->ce_flags |= CE_INTENT_TO_ADD;
 587
 588        if (trust_executable_bit && has_symlinks)
 589                ce->ce_mode = create_ce_mode(st_mode);
 590        else {
 591                /* If there is an existing entry, pick the mode bits and type
 592                 * from it, otherwise assume unexecutable regular file.
 593                 */
 594                struct cache_entry *ent;
 595                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 596
 597                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 598                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 599        }
 600
 601        /* When core.ignorecase=true, determine if a directory of the same name but differing
 602         * case already exists within the Git repository.  If it does, ensure the directory
 603         * case of the file being added to the repository matches (is folded into) the existing
 604         * entry's directory case.
 605         */
 606        if (ignore_case) {
 607                const char *startPtr = ce->name;
 608                const char *ptr = startPtr;
 609                while (*ptr) {
 610                        while (*ptr && *ptr != '/')
 611                                ++ptr;
 612                        if (*ptr == '/') {
 613                                struct cache_entry *foundce;
 614                                ++ptr;
 615                                foundce = index_name_exists(&the_index, ce->name, ptr - ce->name, ignore_case);
 616                                if (foundce) {
 617                                        memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
 618                                        startPtr = ptr;
 619                                }
 620                        }
 621                }
 622        }
 623
 624        alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 625        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 626                /* Nothing changed, really */
 627                free(ce);
 628                if (!S_ISGITLINK(alias->ce_mode))
 629                        ce_mark_uptodate(alias);
 630                alias->ce_flags |= CE_ADDED;
 631                return 0;
 632        }
 633        if (!intent_only) {
 634                if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
 635                        return error("unable to index file %s", path);
 636        } else
 637                record_intent_to_add(ce);
 638
 639        if (ignore_case && alias && different_name(ce, alias))
 640                ce = create_alias_ce(ce, alias);
 641        ce->ce_flags |= CE_ADDED;
 642
 643        /* It was suspected to be racily clean, but it turns out to be Ok */
 644        was_same = (alias &&
 645                    !ce_stage(alias) &&
 646                    !hashcmp(alias->sha1, ce->sha1) &&
 647                    ce->ce_mode == alias->ce_mode);
 648
 649        if (pretend)
 650                ;
 651        else if (add_index_entry(istate, ce, add_option))
 652                return error("unable to add %s to index",path);
 653        if (verbose && !was_same)
 654                printf("add '%s'\n", path);
 655        return 0;
 656}
 657
 658int add_file_to_index(struct index_state *istate, const char *path, int flags)
 659{
 660        struct stat st;
 661        if (lstat(path, &st))
 662                die_errno("unable to stat '%s'", path);
 663        return add_to_index(istate, path, &st, flags);
 664}
 665
 666struct cache_entry *make_cache_entry(unsigned int mode,
 667                const unsigned char *sha1, const char *path, int stage,
 668                int refresh)
 669{
 670        int size, len;
 671        struct cache_entry *ce;
 672
 673        if (!verify_path(path)) {
 674                error("Invalid path '%s'", path);
 675                return NULL;
 676        }
 677
 678        len = strlen(path);
 679        size = cache_entry_size(len);
 680        ce = xcalloc(1, size);
 681
 682        hashcpy(ce->sha1, sha1);
 683        memcpy(ce->name, path, len);
 684        ce->ce_flags = create_ce_flags(len, stage);
 685        ce->ce_mode = create_ce_mode(mode);
 686
 687        if (refresh)
 688                return refresh_cache_entry(ce, 0);
 689
 690        return ce;
 691}
 692
 693int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 694{
 695        int len = ce_namelen(a);
 696        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 697}
 698
 699int ce_path_match(const struct cache_entry *ce, const struct pathspec *pathspec)
 700{
 701        return match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, NULL);
 702}
 703
 704/*
 705 * We fundamentally don't like some paths: we don't want
 706 * dot or dot-dot anywhere, and for obvious reasons don't
 707 * want to recurse into ".git" either.
 708 *
 709 * Also, we don't want double slashes or slashes at the
 710 * end that can make pathnames ambiguous.
 711 */
 712static int verify_dotfile(const char *rest)
 713{
 714        /*
 715         * The first character was '.', but that
 716         * has already been discarded, we now test
 717         * the rest.
 718         */
 719
 720        /* "." is not allowed */
 721        if (*rest == '\0' || is_dir_sep(*rest))
 722                return 0;
 723
 724        switch (*rest) {
 725        /*
 726         * ".git" followed by  NUL or slash is bad. This
 727         * shares the path end test with the ".." case.
 728         */
 729        case 'g':
 730                if (rest[1] != 'i')
 731                        break;
 732                if (rest[2] != 't')
 733                        break;
 734                rest += 2;
 735        /* fallthrough */
 736        case '.':
 737                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 738                        return 0;
 739        }
 740        return 1;
 741}
 742
 743int verify_path(const char *path)
 744{
 745        char c;
 746
 747        if (has_dos_drive_prefix(path))
 748                return 0;
 749
 750        goto inside;
 751        for (;;) {
 752                if (!c)
 753                        return 1;
 754                if (is_dir_sep(c)) {
 755inside:
 756                        c = *path++;
 757                        if ((c == '.' && !verify_dotfile(path)) ||
 758                            is_dir_sep(c) || c == '\0')
 759                                return 0;
 760                }
 761                c = *path++;
 762        }
 763}
 764
 765/*
 766 * Do we have another file that has the beginning components being a
 767 * proper superset of the name we're trying to add?
 768 */
 769static int has_file_name(struct index_state *istate,
 770                         const struct cache_entry *ce, int pos, int ok_to_replace)
 771{
 772        int retval = 0;
 773        int len = ce_namelen(ce);
 774        int stage = ce_stage(ce);
 775        const char *name = ce->name;
 776
 777        while (pos < istate->cache_nr) {
 778                struct cache_entry *p = istate->cache[pos++];
 779
 780                if (len >= ce_namelen(p))
 781                        break;
 782                if (memcmp(name, p->name, len))
 783                        break;
 784                if (ce_stage(p) != stage)
 785                        continue;
 786                if (p->name[len] != '/')
 787                        continue;
 788                if (p->ce_flags & CE_REMOVE)
 789                        continue;
 790                retval = -1;
 791                if (!ok_to_replace)
 792                        break;
 793                remove_index_entry_at(istate, --pos);
 794        }
 795        return retval;
 796}
 797
 798/*
 799 * Do we have another file with a pathname that is a proper
 800 * subset of the name we're trying to add?
 801 */
 802static int has_dir_name(struct index_state *istate,
 803                        const struct cache_entry *ce, int pos, int ok_to_replace)
 804{
 805        int retval = 0;
 806        int stage = ce_stage(ce);
 807        const char *name = ce->name;
 808        const char *slash = name + ce_namelen(ce);
 809
 810        for (;;) {
 811                int len;
 812
 813                for (;;) {
 814                        if (*--slash == '/')
 815                                break;
 816                        if (slash <= ce->name)
 817                                return retval;
 818                }
 819                len = slash - name;
 820
 821                pos = index_name_pos(istate, name, create_ce_flags(len, stage));
 822                if (pos >= 0) {
 823                        /*
 824                         * Found one, but not so fast.  This could
 825                         * be a marker that says "I was here, but
 826                         * I am being removed".  Such an entry is
 827                         * not a part of the resulting tree, and
 828                         * it is Ok to have a directory at the same
 829                         * path.
 830                         */
 831                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 832                                retval = -1;
 833                                if (!ok_to_replace)
 834                                        break;
 835                                remove_index_entry_at(istate, pos);
 836                                continue;
 837                        }
 838                }
 839                else
 840                        pos = -pos-1;
 841
 842                /*
 843                 * Trivial optimization: if we find an entry that
 844                 * already matches the sub-directory, then we know
 845                 * we're ok, and we can exit.
 846                 */
 847                while (pos < istate->cache_nr) {
 848                        struct cache_entry *p = istate->cache[pos];
 849                        if ((ce_namelen(p) <= len) ||
 850                            (p->name[len] != '/') ||
 851                            memcmp(p->name, name, len))
 852                                break; /* not our subdirectory */
 853                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 854                                /*
 855                                 * p is at the same stage as our entry, and
 856                                 * is a subdirectory of what we are looking
 857                                 * at, so we cannot have conflicts at our
 858                                 * level or anything shorter.
 859                                 */
 860                                return retval;
 861                        pos++;
 862                }
 863        }
 864        return retval;
 865}
 866
 867/* We may be in a situation where we already have path/file and path
 868 * is being added, or we already have path and path/file is being
 869 * added.  Either one would result in a nonsense tree that has path
 870 * twice when git-write-tree tries to write it out.  Prevent it.
 871 *
 872 * If ok-to-replace is specified, we remove the conflicting entries
 873 * from the cache so the caller should recompute the insert position.
 874 * When this happens, we return non-zero.
 875 */
 876static int check_file_directory_conflict(struct index_state *istate,
 877                                         const struct cache_entry *ce,
 878                                         int pos, int ok_to_replace)
 879{
 880        int retval;
 881
 882        /*
 883         * When ce is an "I am going away" entry, we allow it to be added
 884         */
 885        if (ce->ce_flags & CE_REMOVE)
 886                return 0;
 887
 888        /*
 889         * We check if the path is a sub-path of a subsequent pathname
 890         * first, since removing those will not change the position
 891         * in the array.
 892         */
 893        retval = has_file_name(istate, ce, pos, ok_to_replace);
 894
 895        /*
 896         * Then check if the path might have a clashing sub-directory
 897         * before it.
 898         */
 899        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 900}
 901
 902static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 903{
 904        int pos;
 905        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 906        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 907        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 908        int new_only = option & ADD_CACHE_NEW_ONLY;
 909
 910        cache_tree_invalidate_path(istate->cache_tree, ce->name);
 911        pos = index_name_pos(istate, ce->name, ce->ce_flags);
 912
 913        /* existing match? Just replace it. */
 914        if (pos >= 0) {
 915                if (!new_only)
 916                        replace_index_entry(istate, pos, ce);
 917                return 0;
 918        }
 919        pos = -pos-1;
 920
 921        /*
 922         * Inserting a merged entry ("stage 0") into the index
 923         * will always replace all non-merged entries..
 924         */
 925        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 926                while (ce_same_name(istate->cache[pos], ce)) {
 927                        ok_to_add = 1;
 928                        if (!remove_index_entry_at(istate, pos))
 929                                break;
 930                }
 931        }
 932
 933        if (!ok_to_add)
 934                return -1;
 935        if (!verify_path(ce->name))
 936                return error("Invalid path '%s'", ce->name);
 937
 938        if (!skip_df_check &&
 939            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 940                if (!ok_to_replace)
 941                        return error("'%s' appears as both a file and as a directory",
 942                                     ce->name);
 943                pos = index_name_pos(istate, ce->name, ce->ce_flags);
 944                pos = -pos-1;
 945        }
 946        return pos + 1;
 947}
 948
 949int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 950{
 951        int pos;
 952
 953        if (option & ADD_CACHE_JUST_APPEND)
 954                pos = istate->cache_nr;
 955        else {
 956                int ret;
 957                ret = add_index_entry_with_check(istate, ce, option);
 958                if (ret <= 0)
 959                        return ret;
 960                pos = ret - 1;
 961        }
 962
 963        /* Make sure the array is big enough .. */
 964        if (istate->cache_nr == istate->cache_alloc) {
 965                istate->cache_alloc = alloc_nr(istate->cache_alloc);
 966                istate->cache = xrealloc(istate->cache,
 967                                        istate->cache_alloc * sizeof(struct cache_entry *));
 968        }
 969
 970        /* Add it in.. */
 971        istate->cache_nr++;
 972        if (istate->cache_nr > pos + 1)
 973                memmove(istate->cache + pos + 1,
 974                        istate->cache + pos,
 975                        (istate->cache_nr - pos - 1) * sizeof(ce));
 976        set_index_entry(istate, pos, ce);
 977        istate->cache_changed = 1;
 978        return 0;
 979}
 980
 981/*
 982 * "refresh" does not calculate a new sha1 file or bring the
 983 * cache up-to-date for mode/content changes. But what it
 984 * _does_ do is to "re-match" the stat information of a file
 985 * with the cache, so that you can refresh the cache for a
 986 * file that hasn't been changed but where the stat entry is
 987 * out of date.
 988 *
 989 * For example, you'd want to do this after doing a "git-read-tree",
 990 * to link up the stat cache details with the proper files.
 991 */
 992static struct cache_entry *refresh_cache_ent(struct index_state *istate,
 993                                             struct cache_entry *ce,
 994                                             unsigned int options, int *err,
 995                                             int *changed_ret)
 996{
 997        struct stat st;
 998        struct cache_entry *updated;
 999        int changed, size;
1000        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1001        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1002
1003        if (ce_uptodate(ce))
1004                return ce;
1005
1006        /*
1007         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1008         * that the change to the work tree does not matter and told
1009         * us not to worry.
1010         */
1011        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1012                ce_mark_uptodate(ce);
1013                return ce;
1014        }
1015        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1016                ce_mark_uptodate(ce);
1017                return ce;
1018        }
1019
1020        if (lstat(ce->name, &st) < 0) {
1021                if (err)
1022                        *err = errno;
1023                return NULL;
1024        }
1025
1026        changed = ie_match_stat(istate, ce, &st, options);
1027        if (changed_ret)
1028                *changed_ret = changed;
1029        if (!changed) {
1030                /*
1031                 * The path is unchanged.  If we were told to ignore
1032                 * valid bit, then we did the actual stat check and
1033                 * found that the entry is unmodified.  If the entry
1034                 * is not marked VALID, this is the place to mark it
1035                 * valid again, under "assume unchanged" mode.
1036                 */
1037                if (ignore_valid && assume_unchanged &&
1038                    !(ce->ce_flags & CE_VALID))
1039                        ; /* mark this one VALID again */
1040                else {
1041                        /*
1042                         * We do not mark the index itself "modified"
1043                         * because CE_UPTODATE flag is in-core only;
1044                         * we are not going to write this change out.
1045                         */
1046                        if (!S_ISGITLINK(ce->ce_mode))
1047                                ce_mark_uptodate(ce);
1048                        return ce;
1049                }
1050        }
1051
1052        if (ie_modified(istate, ce, &st, options)) {
1053                if (err)
1054                        *err = EINVAL;
1055                return NULL;
1056        }
1057
1058        size = ce_size(ce);
1059        updated = xmalloc(size);
1060        memcpy(updated, ce, size);
1061        fill_stat_cache_info(updated, &st);
1062        /*
1063         * If ignore_valid is not set, we should leave CE_VALID bit
1064         * alone.  Otherwise, paths marked with --no-assume-unchanged
1065         * (i.e. things to be edited) will reacquire CE_VALID bit
1066         * automatically, which is not really what we want.
1067         */
1068        if (!ignore_valid && assume_unchanged &&
1069            !(ce->ce_flags & CE_VALID))
1070                updated->ce_flags &= ~CE_VALID;
1071
1072        return updated;
1073}
1074
1075static void show_file(const char * fmt, const char * name, int in_porcelain,
1076                      int * first, const char *header_msg)
1077{
1078        if (in_porcelain && *first && header_msg) {
1079                printf("%s\n", header_msg);
1080                *first = 0;
1081        }
1082        printf(fmt, name);
1083}
1084
1085int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1086                  char *seen, const char *header_msg)
1087{
1088        int i;
1089        int has_errors = 0;
1090        int really = (flags & REFRESH_REALLY) != 0;
1091        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1092        int quiet = (flags & REFRESH_QUIET) != 0;
1093        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1094        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1095        int first = 1;
1096        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1097        unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1098        const char *modified_fmt;
1099        const char *deleted_fmt;
1100        const char *typechange_fmt;
1101        const char *added_fmt;
1102        const char *unmerged_fmt;
1103
1104        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1105        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1106        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1107        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1108        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1109        for (i = 0; i < istate->cache_nr; i++) {
1110                struct cache_entry *ce, *new;
1111                int cache_errno = 0;
1112                int changed = 0;
1113                int filtered = 0;
1114
1115                ce = istate->cache[i];
1116                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1117                        continue;
1118
1119                if (pathspec &&
1120                    !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1121                        filtered = 1;
1122
1123                if (ce_stage(ce)) {
1124                        while ((i < istate->cache_nr) &&
1125                               ! strcmp(istate->cache[i]->name, ce->name))
1126                                i++;
1127                        i--;
1128                        if (allow_unmerged)
1129                                continue;
1130                        if (!filtered)
1131                                show_file(unmerged_fmt, ce->name, in_porcelain,
1132                                          &first, header_msg);
1133                        has_errors = 1;
1134                        continue;
1135                }
1136
1137                if (filtered)
1138                        continue;
1139
1140                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1141                if (new == ce)
1142                        continue;
1143                if (!new) {
1144                        const char *fmt;
1145
1146                        if (not_new && cache_errno == ENOENT)
1147                                continue;
1148                        if (really && cache_errno == EINVAL) {
1149                                /* If we are doing --really-refresh that
1150                                 * means the index is not valid anymore.
1151                                 */
1152                                ce->ce_flags &= ~CE_VALID;
1153                                istate->cache_changed = 1;
1154                        }
1155                        if (quiet)
1156                                continue;
1157
1158                        if (cache_errno == ENOENT)
1159                                fmt = deleted_fmt;
1160                        else if (ce->ce_flags & CE_INTENT_TO_ADD)
1161                                fmt = added_fmt; /* must be before other checks */
1162                        else if (changed & TYPE_CHANGED)
1163                                fmt = typechange_fmt;
1164                        else
1165                                fmt = modified_fmt;
1166                        show_file(fmt,
1167                                  ce->name, in_porcelain, &first, header_msg);
1168                        has_errors = 1;
1169                        continue;
1170                }
1171
1172                replace_index_entry(istate, i, new);
1173        }
1174        return has_errors;
1175}
1176
1177static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1178{
1179        return refresh_cache_ent(&the_index, ce, really, NULL, NULL);
1180}
1181
1182static int verify_hdr(struct cache_header *hdr, unsigned long size)
1183{
1184        git_SHA_CTX c;
1185        unsigned char sha1[20];
1186
1187        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1188                return error("bad signature");
1189        if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
1190                return error("bad index version");
1191        git_SHA1_Init(&c);
1192        git_SHA1_Update(&c, hdr, size - 20);
1193        git_SHA1_Final(sha1, &c);
1194        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1195                return error("bad index file sha1 signature");
1196        return 0;
1197}
1198
1199static int read_index_extension(struct index_state *istate,
1200                                const char *ext, void *data, unsigned long sz)
1201{
1202        switch (CACHE_EXT(ext)) {
1203        case CACHE_EXT_TREE:
1204                istate->cache_tree = cache_tree_read(data, sz);
1205                break;
1206        case CACHE_EXT_RESOLVE_UNDO:
1207                istate->resolve_undo = resolve_undo_read(data, sz);
1208                break;
1209        default:
1210                if (*ext < 'A' || 'Z' < *ext)
1211                        return error("index uses %.4s extension, which we do not understand",
1212                                     ext);
1213                fprintf(stderr, "ignoring %.4s extension\n", ext);
1214                break;
1215        }
1216        return 0;
1217}
1218
1219int read_index(struct index_state *istate)
1220{
1221        return read_index_from(istate, get_index_file());
1222}
1223
1224static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk)
1225{
1226        struct cache_entry *ce;
1227        size_t len;
1228        const char *name;
1229        unsigned int flags;
1230
1231        /* On-disk flags are just 16 bits */
1232        flags = ntohs(ondisk->flags);
1233        len = flags & CE_NAMEMASK;
1234
1235        if (flags & CE_EXTENDED) {
1236                struct ondisk_cache_entry_extended *ondisk2;
1237                int extended_flags;
1238                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1239                extended_flags = ntohs(ondisk2->flags2) << 16;
1240                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1241                if (extended_flags & ~CE_EXTENDED_FLAGS)
1242                        die("Unknown index entry format %08x", extended_flags);
1243                flags |= extended_flags;
1244                name = ondisk2->name;
1245        }
1246        else
1247                name = ondisk->name;
1248
1249        if (len == CE_NAMEMASK)
1250                len = strlen(name);
1251
1252        ce = xmalloc(cache_entry_size(len));
1253
1254        ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);
1255        ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);
1256        ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);
1257        ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);
1258        ce->ce_dev   = ntohl(ondisk->dev);
1259        ce->ce_ino   = ntohl(ondisk->ino);
1260        ce->ce_mode  = ntohl(ondisk->mode);
1261        ce->ce_uid   = ntohl(ondisk->uid);
1262        ce->ce_gid   = ntohl(ondisk->gid);
1263        ce->ce_size  = ntohl(ondisk->size);
1264        ce->ce_flags = flags;
1265
1266        hashcpy(ce->sha1, ondisk->sha1);
1267
1268        memcpy(ce->name, name, len);
1269        ce->name[len] = '\0';
1270        return ce;
1271}
1272
1273/* remember to discard_cache() before reading a different cache! */
1274int read_index_from(struct index_state *istate, const char *path)
1275{
1276        int fd, i;
1277        struct stat st;
1278        unsigned long src_offset;
1279        struct cache_header *hdr;
1280        void *mmap;
1281        size_t mmap_size;
1282
1283        errno = EBUSY;
1284        if (istate->initialized)
1285                return istate->cache_nr;
1286
1287        errno = ENOENT;
1288        istate->timestamp.sec = 0;
1289        istate->timestamp.nsec = 0;
1290        fd = open(path, O_RDONLY);
1291        if (fd < 0) {
1292                if (errno == ENOENT)
1293                        return 0;
1294                die_errno("index file open failed");
1295        }
1296
1297        if (fstat(fd, &st))
1298                die_errno("cannot stat the open index");
1299
1300        errno = EINVAL;
1301        mmap_size = xsize_t(st.st_size);
1302        if (mmap_size < sizeof(struct cache_header) + 20)
1303                die("index file smaller than expected");
1304
1305        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1306        close(fd);
1307        if (mmap == MAP_FAILED)
1308                die_errno("unable to map index file");
1309
1310        hdr = mmap;
1311        if (verify_hdr(hdr, mmap_size) < 0)
1312                goto unmap;
1313
1314        istate->cache_nr = ntohl(hdr->hdr_entries);
1315        istate->cache_alloc = alloc_nr(istate->cache_nr);
1316        istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1317        istate->initialized = 1;
1318
1319        src_offset = sizeof(*hdr);
1320        for (i = 0; i < istate->cache_nr; i++) {
1321                struct ondisk_cache_entry *disk_ce;
1322                struct cache_entry *ce;
1323
1324                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1325                ce = create_from_disk(disk_ce);
1326                set_index_entry(istate, i, ce);
1327
1328                src_offset += ondisk_ce_size(ce);
1329        }
1330        istate->timestamp.sec = st.st_mtime;
1331        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1332
1333        while (src_offset <= mmap_size - 20 - 8) {
1334                /* After an array of active_nr index entries,
1335                 * there can be arbitrary number of extended
1336                 * sections, each of which is prefixed with
1337                 * extension name (4-byte) and section length
1338                 * in 4-byte network byte order.
1339                 */
1340                uint32_t extsize;
1341                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1342                extsize = ntohl(extsize);
1343                if (read_index_extension(istate,
1344                                         (const char *) mmap + src_offset,
1345                                         (char *) mmap + src_offset + 8,
1346                                         extsize) < 0)
1347                        goto unmap;
1348                src_offset += 8;
1349                src_offset += extsize;
1350        }
1351        munmap(mmap, mmap_size);
1352        return istate->cache_nr;
1353
1354unmap:
1355        munmap(mmap, mmap_size);
1356        errno = EINVAL;
1357        die("index file corrupt");
1358}
1359
1360int is_index_unborn(struct index_state *istate)
1361{
1362        return (!istate->cache_nr && !istate->timestamp.sec);
1363}
1364
1365int discard_index(struct index_state *istate)
1366{
1367        int i;
1368
1369        for (i = 0; i < istate->cache_nr; i++)
1370                free(istate->cache[i]);
1371        resolve_undo_clear_index(istate);
1372        istate->cache_nr = 0;
1373        istate->cache_changed = 0;
1374        istate->timestamp.sec = 0;
1375        istate->timestamp.nsec = 0;
1376        istate->name_hash_initialized = 0;
1377        free_hash(&istate->name_hash);
1378        cache_tree_free(&(istate->cache_tree));
1379        istate->initialized = 0;
1380
1381        /* no need to throw away allocated active_cache */
1382        return 0;
1383}
1384
1385int unmerged_index(const struct index_state *istate)
1386{
1387        int i;
1388        for (i = 0; i < istate->cache_nr; i++) {
1389                if (ce_stage(istate->cache[i]))
1390                        return 1;
1391        }
1392        return 0;
1393}
1394
1395#define WRITE_BUFFER_SIZE 8192
1396static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1397static unsigned long write_buffer_len;
1398
1399static int ce_write_flush(git_SHA_CTX *context, int fd)
1400{
1401        unsigned int buffered = write_buffer_len;
1402        if (buffered) {
1403                git_SHA1_Update(context, write_buffer, buffered);
1404                if (write_in_full(fd, write_buffer, buffered) != buffered)
1405                        return -1;
1406                write_buffer_len = 0;
1407        }
1408        return 0;
1409}
1410
1411static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1412{
1413        while (len) {
1414                unsigned int buffered = write_buffer_len;
1415                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1416                if (partial > len)
1417                        partial = len;
1418                memcpy(write_buffer + buffered, data, partial);
1419                buffered += partial;
1420                if (buffered == WRITE_BUFFER_SIZE) {
1421                        write_buffer_len = buffered;
1422                        if (ce_write_flush(context, fd))
1423                                return -1;
1424                        buffered = 0;
1425                }
1426                write_buffer_len = buffered;
1427                len -= partial;
1428                data = (char *) data + partial;
1429        }
1430        return 0;
1431}
1432
1433static int write_index_ext_header(git_SHA_CTX *context, int fd,
1434                                  unsigned int ext, unsigned int sz)
1435{
1436        ext = htonl(ext);
1437        sz = htonl(sz);
1438        return ((ce_write(context, fd, &ext, 4) < 0) ||
1439                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1440}
1441
1442static int ce_flush(git_SHA_CTX *context, int fd)
1443{
1444        unsigned int left = write_buffer_len;
1445
1446        if (left) {
1447                write_buffer_len = 0;
1448                git_SHA1_Update(context, write_buffer, left);
1449        }
1450
1451        /* Flush first if not enough space for SHA1 signature */
1452        if (left + 20 > WRITE_BUFFER_SIZE) {
1453                if (write_in_full(fd, write_buffer, left) != left)
1454                        return -1;
1455                left = 0;
1456        }
1457
1458        /* Append the SHA1 signature at the end */
1459        git_SHA1_Final(write_buffer + left, context);
1460        left += 20;
1461        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1462}
1463
1464static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1465{
1466        /*
1467         * The only thing we care about in this function is to smudge the
1468         * falsely clean entry due to touch-update-touch race, so we leave
1469         * everything else as they are.  We are called for entries whose
1470         * ce_mtime match the index file mtime.
1471         *
1472         * Note that this actually does not do much for gitlinks, for
1473         * which ce_match_stat_basic() always goes to the actual
1474         * contents.  The caller checks with is_racy_timestamp() which
1475         * always says "no" for gitlinks, so we are not called for them ;-)
1476         */
1477        struct stat st;
1478
1479        if (lstat(ce->name, &st) < 0)
1480                return;
1481        if (ce_match_stat_basic(ce, &st))
1482                return;
1483        if (ce_modified_check_fs(ce, &st)) {
1484                /* This is "racily clean"; smudge it.  Note that this
1485                 * is a tricky code.  At first glance, it may appear
1486                 * that it can break with this sequence:
1487                 *
1488                 * $ echo xyzzy >frotz
1489                 * $ git-update-index --add frotz
1490                 * $ : >frotz
1491                 * $ sleep 3
1492                 * $ echo filfre >nitfol
1493                 * $ git-update-index --add nitfol
1494                 *
1495                 * but it does not.  When the second update-index runs,
1496                 * it notices that the entry "frotz" has the same timestamp
1497                 * as index, and if we were to smudge it by resetting its
1498                 * size to zero here, then the object name recorded
1499                 * in index is the 6-byte file but the cached stat information
1500                 * becomes zero --- which would then match what we would
1501                 * obtain from the filesystem next time we stat("frotz").
1502                 *
1503                 * However, the second update-index, before calling
1504                 * this function, notices that the cached size is 6
1505                 * bytes and what is on the filesystem is an empty
1506                 * file, and never calls us, so the cached size information
1507                 * for "frotz" stays 6 which does not match the filesystem.
1508                 */
1509                ce->ce_size = 0;
1510        }
1511}
1512
1513static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
1514{
1515        int size = ondisk_ce_size(ce);
1516        struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1517        char *name;
1518        int result;
1519
1520        ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1521        ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1522        ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1523        ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1524        ondisk->dev  = htonl(ce->ce_dev);
1525        ondisk->ino  = htonl(ce->ce_ino);
1526        ondisk->mode = htonl(ce->ce_mode);
1527        ondisk->uid  = htonl(ce->ce_uid);
1528        ondisk->gid  = htonl(ce->ce_gid);
1529        ondisk->size = htonl(ce->ce_size);
1530        hashcpy(ondisk->sha1, ce->sha1);
1531        ondisk->flags = htons(ce->ce_flags);
1532        if (ce->ce_flags & CE_EXTENDED) {
1533                struct ondisk_cache_entry_extended *ondisk2;
1534                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1535                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1536                name = ondisk2->name;
1537        }
1538        else
1539                name = ondisk->name;
1540        memcpy(name, ce->name, ce_namelen(ce));
1541
1542        result = ce_write(c, fd, ondisk, size);
1543        free(ondisk);
1544        return result;
1545}
1546
1547static int has_racy_timestamp(struct index_state *istate)
1548{
1549        int entries = istate->cache_nr;
1550        int i;
1551
1552        for (i = 0; i < entries; i++) {
1553                struct cache_entry *ce = istate->cache[i];
1554                if (is_racy_timestamp(istate, ce))
1555                        return 1;
1556        }
1557        return 0;
1558}
1559
1560/*
1561 * Opportunisticly update the index but do not complain if we can't
1562 */
1563void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1564{
1565        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1566            !write_index(istate, lockfile->fd))
1567                commit_locked_index(lockfile);
1568        else
1569                rollback_lock_file(lockfile);
1570}
1571
1572int write_index(struct index_state *istate, int newfd)
1573{
1574        git_SHA_CTX c;
1575        struct cache_header hdr;
1576        int i, err, removed, extended;
1577        struct cache_entry **cache = istate->cache;
1578        int entries = istate->cache_nr;
1579        struct stat st;
1580
1581        for (i = removed = extended = 0; i < entries; i++) {
1582                if (cache[i]->ce_flags & CE_REMOVE)
1583                        removed++;
1584
1585                /* reduce extended entries if possible */
1586                cache[i]->ce_flags &= ~CE_EXTENDED;
1587                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1588                        extended++;
1589                        cache[i]->ce_flags |= CE_EXTENDED;
1590                }
1591        }
1592
1593        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1594        /* for extended format, increase version so older git won't try to read it */
1595        hdr.hdr_version = htonl(extended ? 3 : 2);
1596        hdr.hdr_entries = htonl(entries - removed);
1597
1598        git_SHA1_Init(&c);
1599        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1600                return -1;
1601
1602        for (i = 0; i < entries; i++) {
1603                struct cache_entry *ce = cache[i];
1604                if (ce->ce_flags & CE_REMOVE)
1605                        continue;
1606                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1607                        ce_smudge_racily_clean_entry(ce);
1608                if (ce_write_entry(&c, newfd, ce) < 0)
1609                        return -1;
1610        }
1611
1612        /* Write extension data here */
1613        if (istate->cache_tree) {
1614                struct strbuf sb = STRBUF_INIT;
1615
1616                cache_tree_write(&sb, istate->cache_tree);
1617                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1618                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1619                strbuf_release(&sb);
1620                if (err)
1621                        return -1;
1622        }
1623        if (istate->resolve_undo) {
1624                struct strbuf sb = STRBUF_INIT;
1625
1626                resolve_undo_write(&sb, istate->resolve_undo);
1627                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1628                                             sb.len) < 0
1629                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1630                strbuf_release(&sb);
1631                if (err)
1632                        return -1;
1633        }
1634
1635        if (ce_flush(&c, newfd) || fstat(newfd, &st))
1636                return -1;
1637        istate->timestamp.sec = (unsigned int)st.st_mtime;
1638        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1639        return 0;
1640}
1641
1642/*
1643 * Read the index file that is potentially unmerged into given
1644 * index_state, dropping any unmerged entries.  Returns true if
1645 * the index is unmerged.  Callers who want to refuse to work
1646 * from an unmerged state can call this and check its return value,
1647 * instead of calling read_cache().
1648 */
1649int read_index_unmerged(struct index_state *istate)
1650{
1651        int i;
1652        int unmerged = 0;
1653
1654        read_index(istate);
1655        for (i = 0; i < istate->cache_nr; i++) {
1656                struct cache_entry *ce = istate->cache[i];
1657                struct cache_entry *new_ce;
1658                int size, len;
1659
1660                if (!ce_stage(ce))
1661                        continue;
1662                unmerged = 1;
1663                len = strlen(ce->name);
1664                size = cache_entry_size(len);
1665                new_ce = xcalloc(1, size);
1666                memcpy(new_ce->name, ce->name, len);
1667                new_ce->ce_flags = create_ce_flags(len, 0) | CE_CONFLICTED;
1668                new_ce->ce_mode = ce->ce_mode;
1669                if (add_index_entry(istate, new_ce, 0))
1670                        return error("%s: cannot drop to stage #0",
1671                                     ce->name);
1672                i = index_name_pos(istate, new_ce->name, len);
1673        }
1674        return unmerged;
1675}
1676
1677/*
1678 * Returns 1 if the path is an "other" path with respect to
1679 * the index; that is, the path is not mentioned in the index at all,
1680 * either as a file, a directory with some files in the index,
1681 * or as an unmerged entry.
1682 *
1683 * We helpfully remove a trailing "/" from directories so that
1684 * the output of read_directory can be used as-is.
1685 */
1686int index_name_is_other(const struct index_state *istate, const char *name,
1687                int namelen)
1688{
1689        int pos;
1690        if (namelen && name[namelen - 1] == '/')
1691                namelen--;
1692        pos = index_name_pos(istate, name, namelen);
1693        if (0 <= pos)
1694                return 0;       /* exact match */
1695        pos = -pos - 1;
1696        if (pos < istate->cache_nr) {
1697                struct cache_entry *ce = istate->cache[pos];
1698                if (ce_namelen(ce) == namelen &&
1699                    !memcmp(ce->name, name, namelen))
1700                        return 0; /* Yup, this one exists unmerged */
1701        }
1702        return 1;
1703}