31cf0b503adb9c9b436a74df6ab11992d863a057
   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        /* When core.ignorecase=true, determine if a directory of the same name but differing
 612         * case already exists within the Git repository.  If it does, ensure the directory
 613         * case of the file being added to the repository matches (is folded into) the existing
 614         * entry's directory case.
 615         */
 616        if (ignore_case) {
 617                const char *startPtr = ce->name;
 618                const char *ptr = startPtr;
 619                while (*ptr) {
 620                        while (*ptr && *ptr != '/')
 621                                ++ptr;
 622                        if (*ptr == '/') {
 623                                struct cache_entry *foundce;
 624                                ++ptr;
 625                                foundce = index_name_exists(&the_index, ce->name, ptr - ce->name, ignore_case);
 626                                if (foundce) {
 627                                        memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
 628                                        startPtr = ptr;
 629                                }
 630                        }
 631                }
 632        }
 633
 634        alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 635        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 636                /* Nothing changed, really */
 637                free(ce);
 638                if (!S_ISGITLINK(alias->ce_mode))
 639                        ce_mark_uptodate(alias);
 640                alias->ce_flags |= CE_ADDED;
 641                return 0;
 642        }
 643        if (!intent_only) {
 644                if (index_path(ce->sha1, path, st, 1))
 645                        return error("unable to index file %s", path);
 646        } else
 647                record_intent_to_add(ce);
 648
 649        if (ignore_case && alias && different_name(ce, alias))
 650                ce = create_alias_ce(ce, alias);
 651        ce->ce_flags |= CE_ADDED;
 652
 653        /* It was suspected to be racily clean, but it turns out to be Ok */
 654        was_same = (alias &&
 655                    !ce_stage(alias) &&
 656                    !hashcmp(alias->sha1, ce->sha1) &&
 657                    ce->ce_mode == alias->ce_mode);
 658
 659        if (pretend)
 660                ;
 661        else if (add_index_entry(istate, ce, add_option))
 662                return error("unable to add %s to index",path);
 663        if (verbose && !was_same)
 664                printf("add '%s'\n", path);
 665        return 0;
 666}
 667
 668int add_file_to_index(struct index_state *istate, const char *path, int flags)
 669{
 670        struct stat st;
 671        if (lstat(path, &st))
 672                die_errno("unable to stat '%s'", path);
 673        return add_to_index(istate, path, &st, flags);
 674}
 675
 676struct cache_entry *make_cache_entry(unsigned int mode,
 677                const unsigned char *sha1, const char *path, int stage,
 678                int refresh)
 679{
 680        int size, len;
 681        struct cache_entry *ce;
 682
 683        if (!verify_path(path)) {
 684                error("Invalid path '%s'", path);
 685                return NULL;
 686        }
 687
 688        len = strlen(path);
 689        size = cache_entry_size(len);
 690        ce = xcalloc(1, size);
 691
 692        hashcpy(ce->sha1, sha1);
 693        memcpy(ce->name, path, len);
 694        ce->ce_flags = create_ce_flags(len, stage);
 695        ce->ce_mode = create_ce_mode(mode);
 696
 697        if (refresh)
 698                return refresh_cache_entry(ce, 0);
 699
 700        return ce;
 701}
 702
 703int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 704{
 705        int len = ce_namelen(a);
 706        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 707}
 708
 709int ce_path_match(const struct cache_entry *ce, const char **pathspec)
 710{
 711        const char *match, *name;
 712        int len;
 713
 714        if (!pathspec)
 715                return 1;
 716
 717        len = ce_namelen(ce);
 718        name = ce->name;
 719        while ((match = *pathspec++) != NULL) {
 720                int matchlen = strlen(match);
 721                if (matchlen > len)
 722                        continue;
 723                if (memcmp(name, match, matchlen))
 724                        continue;
 725                if (matchlen && name[matchlen-1] == '/')
 726                        return 1;
 727                if (name[matchlen] == '/' || !name[matchlen])
 728                        return 1;
 729                if (!matchlen)
 730                        return 1;
 731        }
 732        return 0;
 733}
 734
 735/*
 736 * We fundamentally don't like some paths: we don't want
 737 * dot or dot-dot anywhere, and for obvious reasons don't
 738 * want to recurse into ".git" either.
 739 *
 740 * Also, we don't want double slashes or slashes at the
 741 * end that can make pathnames ambiguous.
 742 */
 743static int verify_dotfile(const char *rest)
 744{
 745        /*
 746         * The first character was '.', but that
 747         * has already been discarded, we now test
 748         * the rest.
 749         */
 750        switch (*rest) {
 751        /* "." is not allowed */
 752        case '\0': case '/':
 753                return 0;
 754
 755        /*
 756         * ".git" followed by  NUL or slash is bad. This
 757         * shares the path end test with the ".." case.
 758         */
 759        case 'g':
 760                if (rest[1] != 'i')
 761                        break;
 762                if (rest[2] != 't')
 763                        break;
 764                rest += 2;
 765        /* fallthrough */
 766        case '.':
 767                if (rest[1] == '\0' || rest[1] == '/')
 768                        return 0;
 769        }
 770        return 1;
 771}
 772
 773int verify_path(const char *path)
 774{
 775        char c;
 776
 777        if (has_dos_drive_prefix(path))
 778                return 0;
 779
 780        goto inside;
 781        for (;;) {
 782                if (!c)
 783                        return 1;
 784                if (is_dir_sep(c)) {
 785inside:
 786                        c = *path++;
 787                        switch (c) {
 788                        default:
 789                                continue;
 790                        case '/': case '\0':
 791                                break;
 792                        case '.':
 793                                if (verify_dotfile(path))
 794                                        continue;
 795                        }
 796                        return 0;
 797                }
 798                c = *path++;
 799        }
 800}
 801
 802/*
 803 * Do we have another file that has the beginning components being a
 804 * proper superset of the name we're trying to add?
 805 */
 806static int has_file_name(struct index_state *istate,
 807                         const struct cache_entry *ce, int pos, int ok_to_replace)
 808{
 809        int retval = 0;
 810        int len = ce_namelen(ce);
 811        int stage = ce_stage(ce);
 812        const char *name = ce->name;
 813
 814        while (pos < istate->cache_nr) {
 815                struct cache_entry *p = istate->cache[pos++];
 816
 817                if (len >= ce_namelen(p))
 818                        break;
 819                if (memcmp(name, p->name, len))
 820                        break;
 821                if (ce_stage(p) != stage)
 822                        continue;
 823                if (p->name[len] != '/')
 824                        continue;
 825                if (p->ce_flags & CE_REMOVE)
 826                        continue;
 827                retval = -1;
 828                if (!ok_to_replace)
 829                        break;
 830                remove_index_entry_at(istate, --pos);
 831        }
 832        return retval;
 833}
 834
 835/*
 836 * Do we have another file with a pathname that is a proper
 837 * subset of the name we're trying to add?
 838 */
 839static int has_dir_name(struct index_state *istate,
 840                        const struct cache_entry *ce, int pos, int ok_to_replace)
 841{
 842        int retval = 0;
 843        int stage = ce_stage(ce);
 844        const char *name = ce->name;
 845        const char *slash = name + ce_namelen(ce);
 846
 847        for (;;) {
 848                int len;
 849
 850                for (;;) {
 851                        if (*--slash == '/')
 852                                break;
 853                        if (slash <= ce->name)
 854                                return retval;
 855                }
 856                len = slash - name;
 857
 858                pos = index_name_pos(istate, name, create_ce_flags(len, stage));
 859                if (pos >= 0) {
 860                        /*
 861                         * Found one, but not so fast.  This could
 862                         * be a marker that says "I was here, but
 863                         * I am being removed".  Such an entry is
 864                         * not a part of the resulting tree, and
 865                         * it is Ok to have a directory at the same
 866                         * path.
 867                         */
 868                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 869                                retval = -1;
 870                                if (!ok_to_replace)
 871                                        break;
 872                                remove_index_entry_at(istate, pos);
 873                                continue;
 874                        }
 875                }
 876                else
 877                        pos = -pos-1;
 878
 879                /*
 880                 * Trivial optimization: if we find an entry that
 881                 * already matches the sub-directory, then we know
 882                 * we're ok, and we can exit.
 883                 */
 884                while (pos < istate->cache_nr) {
 885                        struct cache_entry *p = istate->cache[pos];
 886                        if ((ce_namelen(p) <= len) ||
 887                            (p->name[len] != '/') ||
 888                            memcmp(p->name, name, len))
 889                                break; /* not our subdirectory */
 890                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 891                                /*
 892                                 * p is at the same stage as our entry, and
 893                                 * is a subdirectory of what we are looking
 894                                 * at, so we cannot have conflicts at our
 895                                 * level or anything shorter.
 896                                 */
 897                                return retval;
 898                        pos++;
 899                }
 900        }
 901        return retval;
 902}
 903
 904/* We may be in a situation where we already have path/file and path
 905 * is being added, or we already have path and path/file is being
 906 * added.  Either one would result in a nonsense tree that has path
 907 * twice when git-write-tree tries to write it out.  Prevent it.
 908 *
 909 * If ok-to-replace is specified, we remove the conflicting entries
 910 * from the cache so the caller should recompute the insert position.
 911 * When this happens, we return non-zero.
 912 */
 913static int check_file_directory_conflict(struct index_state *istate,
 914                                         const struct cache_entry *ce,
 915                                         int pos, int ok_to_replace)
 916{
 917        int retval;
 918
 919        /*
 920         * When ce is an "I am going away" entry, we allow it to be added
 921         */
 922        if (ce->ce_flags & CE_REMOVE)
 923                return 0;
 924
 925        /*
 926         * We check if the path is a sub-path of a subsequent pathname
 927         * first, since removing those will not change the position
 928         * in the array.
 929         */
 930        retval = has_file_name(istate, ce, pos, ok_to_replace);
 931
 932        /*
 933         * Then check if the path might have a clashing sub-directory
 934         * before it.
 935         */
 936        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 937}
 938
 939static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 940{
 941        int pos;
 942        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 943        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 944        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 945        int new_only = option & ADD_CACHE_NEW_ONLY;
 946
 947        cache_tree_invalidate_path(istate->cache_tree, ce->name);
 948        pos = index_name_pos(istate, ce->name, ce->ce_flags);
 949
 950        /* existing match? Just replace it. */
 951        if (pos >= 0) {
 952                if (!new_only)
 953                        replace_index_entry(istate, pos, ce);
 954                return 0;
 955        }
 956        pos = -pos-1;
 957
 958        /*
 959         * Inserting a merged entry ("stage 0") into the index
 960         * will always replace all non-merged entries..
 961         */
 962        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 963                while (ce_same_name(istate->cache[pos], ce)) {
 964                        ok_to_add = 1;
 965                        if (!remove_index_entry_at(istate, pos))
 966                                break;
 967                }
 968        }
 969
 970        if (!ok_to_add)
 971                return -1;
 972        if (!verify_path(ce->name))
 973                return error("Invalid path '%s'", ce->name);
 974
 975        if (!skip_df_check &&
 976            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 977                if (!ok_to_replace)
 978                        return error("'%s' appears as both a file and as a directory",
 979                                     ce->name);
 980                pos = index_name_pos(istate, ce->name, ce->ce_flags);
 981                pos = -pos-1;
 982        }
 983        return pos + 1;
 984}
 985
 986int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 987{
 988        int pos;
 989
 990        if (option & ADD_CACHE_JUST_APPEND)
 991                pos = istate->cache_nr;
 992        else {
 993                int ret;
 994                ret = add_index_entry_with_check(istate, ce, option);
 995                if (ret <= 0)
 996                        return ret;
 997                pos = ret - 1;
 998        }
 999
1000        /* Make sure the array is big enough .. */
1001        if (istate->cache_nr == istate->cache_alloc) {
1002                istate->cache_alloc = alloc_nr(istate->cache_alloc);
1003                istate->cache = xrealloc(istate->cache,
1004                                        istate->cache_alloc * sizeof(struct cache_entry *));
1005        }
1006
1007        /* Add it in.. */
1008        istate->cache_nr++;
1009        if (istate->cache_nr > pos + 1)
1010                memmove(istate->cache + pos + 1,
1011                        istate->cache + pos,
1012                        (istate->cache_nr - pos - 1) * sizeof(ce));
1013        set_index_entry(istate, pos, ce);
1014        istate->cache_changed = 1;
1015        return 0;
1016}
1017
1018/*
1019 * "refresh" does not calculate a new sha1 file or bring the
1020 * cache up-to-date for mode/content changes. But what it
1021 * _does_ do is to "re-match" the stat information of a file
1022 * with the cache, so that you can refresh the cache for a
1023 * file that hasn't been changed but where the stat entry is
1024 * out of date.
1025 *
1026 * For example, you'd want to do this after doing a "git-read-tree",
1027 * to link up the stat cache details with the proper files.
1028 */
1029static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1030                                             struct cache_entry *ce,
1031                                             unsigned int options, int *err)
1032{
1033        struct stat st;
1034        struct cache_entry *updated;
1035        int changed, size;
1036        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1037        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1038
1039        if (ce_uptodate(ce))
1040                return ce;
1041
1042        /*
1043         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1044         * that the change to the work tree does not matter and told
1045         * us not to worry.
1046         */
1047        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1048                ce_mark_uptodate(ce);
1049                return ce;
1050        }
1051        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1052                ce_mark_uptodate(ce);
1053                return ce;
1054        }
1055
1056        if (lstat(ce->name, &st) < 0) {
1057                if (err)
1058                        *err = errno;
1059                return NULL;
1060        }
1061
1062        changed = ie_match_stat(istate, ce, &st, options);
1063        if (!changed) {
1064                /*
1065                 * The path is unchanged.  If we were told to ignore
1066                 * valid bit, then we did the actual stat check and
1067                 * found that the entry is unmodified.  If the entry
1068                 * is not marked VALID, this is the place to mark it
1069                 * valid again, under "assume unchanged" mode.
1070                 */
1071                if (ignore_valid && assume_unchanged &&
1072                    !(ce->ce_flags & CE_VALID))
1073                        ; /* mark this one VALID again */
1074                else {
1075                        /*
1076                         * We do not mark the index itself "modified"
1077                         * because CE_UPTODATE flag is in-core only;
1078                         * we are not going to write this change out.
1079                         */
1080                        if (!S_ISGITLINK(ce->ce_mode))
1081                                ce_mark_uptodate(ce);
1082                        return ce;
1083                }
1084        }
1085
1086        if (ie_modified(istate, ce, &st, options)) {
1087                if (err)
1088                        *err = EINVAL;
1089                return NULL;
1090        }
1091
1092        size = ce_size(ce);
1093        updated = xmalloc(size);
1094        memcpy(updated, ce, size);
1095        fill_stat_cache_info(updated, &st);
1096        /*
1097         * If ignore_valid is not set, we should leave CE_VALID bit
1098         * alone.  Otherwise, paths marked with --no-assume-unchanged
1099         * (i.e. things to be edited) will reacquire CE_VALID bit
1100         * automatically, which is not really what we want.
1101         */
1102        if (!ignore_valid && assume_unchanged &&
1103            !(ce->ce_flags & CE_VALID))
1104                updated->ce_flags &= ~CE_VALID;
1105
1106        return updated;
1107}
1108
1109static void show_file(const char * fmt, const char * name, int in_porcelain,
1110                      int * first, char *header_msg)
1111{
1112        if (in_porcelain && *first && header_msg) {
1113                printf("%s\n", header_msg);
1114                *first=0;
1115        }
1116        printf(fmt, name);
1117}
1118
1119int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1120                  char *seen, char *header_msg)
1121{
1122        int i;
1123        int has_errors = 0;
1124        int really = (flags & REFRESH_REALLY) != 0;
1125        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1126        int quiet = (flags & REFRESH_QUIET) != 0;
1127        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1128        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1129        int first = 1;
1130        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1131        unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1132        const char *needs_update_fmt;
1133        const char *needs_merge_fmt;
1134
1135        needs_update_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1136        needs_merge_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1137        for (i = 0; i < istate->cache_nr; i++) {
1138                struct cache_entry *ce, *new;
1139                int cache_errno = 0;
1140
1141                ce = istate->cache[i];
1142                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1143                        continue;
1144
1145                if (ce_stage(ce)) {
1146                        while ((i < istate->cache_nr) &&
1147                               ! strcmp(istate->cache[i]->name, ce->name))
1148                                i++;
1149                        i--;
1150                        if (allow_unmerged)
1151                                continue;
1152                        show_file(needs_merge_fmt, ce->name, in_porcelain, &first, header_msg);
1153                        has_errors = 1;
1154                        continue;
1155                }
1156
1157                if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1158                        continue;
1159
1160                new = refresh_cache_ent(istate, ce, options, &cache_errno);
1161                if (new == ce)
1162                        continue;
1163                if (!new) {
1164                        if (not_new && cache_errno == ENOENT)
1165                                continue;
1166                        if (really && cache_errno == EINVAL) {
1167                                /* If we are doing --really-refresh that
1168                                 * means the index is not valid anymore.
1169                                 */
1170                                ce->ce_flags &= ~CE_VALID;
1171                                istate->cache_changed = 1;
1172                        }
1173                        if (quiet)
1174                                continue;
1175                        show_file(needs_update_fmt, ce->name, in_porcelain, &first, header_msg);
1176                        has_errors = 1;
1177                        continue;
1178                }
1179
1180                replace_index_entry(istate, i, new);
1181        }
1182        return has_errors;
1183}
1184
1185static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1186{
1187        return refresh_cache_ent(&the_index, ce, really, NULL);
1188}
1189
1190static int verify_hdr(struct cache_header *hdr, unsigned long size)
1191{
1192        git_SHA_CTX c;
1193        unsigned char sha1[20];
1194
1195        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1196                return error("bad signature");
1197        if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
1198                return error("bad index version");
1199        git_SHA1_Init(&c);
1200        git_SHA1_Update(&c, hdr, size - 20);
1201        git_SHA1_Final(sha1, &c);
1202        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1203                return error("bad index file sha1 signature");
1204        return 0;
1205}
1206
1207static int read_index_extension(struct index_state *istate,
1208                                const char *ext, void *data, unsigned long sz)
1209{
1210        switch (CACHE_EXT(ext)) {
1211        case CACHE_EXT_TREE:
1212                istate->cache_tree = cache_tree_read(data, sz);
1213                break;
1214        case CACHE_EXT_RESOLVE_UNDO:
1215                istate->resolve_undo = resolve_undo_read(data, sz);
1216                break;
1217        default:
1218                if (*ext < 'A' || 'Z' < *ext)
1219                        return error("index uses %.4s extension, which we do not understand",
1220                                     ext);
1221                fprintf(stderr, "ignoring %.4s extension\n", ext);
1222                break;
1223        }
1224        return 0;
1225}
1226
1227int read_index(struct index_state *istate)
1228{
1229        return read_index_from(istate, get_index_file());
1230}
1231
1232static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1233{
1234        size_t len;
1235        const char *name;
1236
1237        ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);
1238        ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);
1239        ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);
1240        ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);
1241        ce->ce_dev   = ntohl(ondisk->dev);
1242        ce->ce_ino   = ntohl(ondisk->ino);
1243        ce->ce_mode  = ntohl(ondisk->mode);
1244        ce->ce_uid   = ntohl(ondisk->uid);
1245        ce->ce_gid   = ntohl(ondisk->gid);
1246        ce->ce_size  = ntohl(ondisk->size);
1247        /* On-disk flags are just 16 bits */
1248        ce->ce_flags = ntohs(ondisk->flags);
1249
1250        hashcpy(ce->sha1, ondisk->sha1);
1251
1252        len = ce->ce_flags & CE_NAMEMASK;
1253
1254        if (ce->ce_flags & CE_EXTENDED) {
1255                struct ondisk_cache_entry_extended *ondisk2;
1256                int extended_flags;
1257                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1258                extended_flags = ntohs(ondisk2->flags2) << 16;
1259                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1260                if (extended_flags & ~CE_EXTENDED_FLAGS)
1261                        die("Unknown index entry format %08x", extended_flags);
1262                ce->ce_flags |= extended_flags;
1263                name = ondisk2->name;
1264        }
1265        else
1266                name = ondisk->name;
1267
1268        if (len == CE_NAMEMASK)
1269                len = strlen(name);
1270        /*
1271         * NEEDSWORK: If the original index is crafted, this copy could
1272         * go unchecked.
1273         */
1274        memcpy(ce->name, name, len + 1);
1275}
1276
1277static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1278{
1279        long per_entry;
1280
1281        per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1282
1283        /*
1284         * Alignment can cause differences. This should be "alignof", but
1285         * since that's a gcc'ism, just use the size of a pointer.
1286         */
1287        per_entry += sizeof(void *);
1288        return ondisk_size + entries*per_entry;
1289}
1290
1291/* remember to discard_cache() before reading a different cache! */
1292int read_index_from(struct index_state *istate, const char *path)
1293{
1294        int fd, i;
1295        struct stat st;
1296        unsigned long src_offset, dst_offset;
1297        struct cache_header *hdr;
1298        void *mmap;
1299        size_t mmap_size;
1300
1301        errno = EBUSY;
1302        if (istate->initialized)
1303                return istate->cache_nr;
1304
1305        errno = ENOENT;
1306        istate->timestamp.sec = 0;
1307        istate->timestamp.nsec = 0;
1308        fd = open(path, O_RDONLY);
1309        if (fd < 0) {
1310                if (errno == ENOENT)
1311                        return 0;
1312                die_errno("index file open failed");
1313        }
1314
1315        if (fstat(fd, &st))
1316                die_errno("cannot stat the open index");
1317
1318        errno = EINVAL;
1319        mmap_size = xsize_t(st.st_size);
1320        if (mmap_size < sizeof(struct cache_header) + 20)
1321                die("index file smaller than expected");
1322
1323        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1324        close(fd);
1325        if (mmap == MAP_FAILED)
1326                die_errno("unable to map index file");
1327
1328        hdr = mmap;
1329        if (verify_hdr(hdr, mmap_size) < 0)
1330                goto unmap;
1331
1332        istate->cache_nr = ntohl(hdr->hdr_entries);
1333        istate->cache_alloc = alloc_nr(istate->cache_nr);
1334        istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1335
1336        /*
1337         * The disk format is actually larger than the in-memory format,
1338         * due to space for nsec etc, so even though the in-memory one
1339         * has room for a few  more flags, we can allocate using the same
1340         * index size
1341         */
1342        istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1343        istate->initialized = 1;
1344
1345        src_offset = sizeof(*hdr);
1346        dst_offset = 0;
1347        for (i = 0; i < istate->cache_nr; i++) {
1348                struct ondisk_cache_entry *disk_ce;
1349                struct cache_entry *ce;
1350
1351                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1352                ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1353                convert_from_disk(disk_ce, ce);
1354                set_index_entry(istate, i, ce);
1355
1356                src_offset += ondisk_ce_size(ce);
1357                dst_offset += ce_size(ce);
1358        }
1359        istate->timestamp.sec = st.st_mtime;
1360        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1361
1362        while (src_offset <= mmap_size - 20 - 8) {
1363                /* After an array of active_nr index entries,
1364                 * there can be arbitrary number of extended
1365                 * sections, each of which is prefixed with
1366                 * extension name (4-byte) and section length
1367                 * in 4-byte network byte order.
1368                 */
1369                uint32_t extsize;
1370                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1371                extsize = ntohl(extsize);
1372                if (read_index_extension(istate,
1373                                         (const char *) mmap + src_offset,
1374                                         (char *) mmap + src_offset + 8,
1375                                         extsize) < 0)
1376                        goto unmap;
1377                src_offset += 8;
1378                src_offset += extsize;
1379        }
1380        munmap(mmap, mmap_size);
1381        return istate->cache_nr;
1382
1383unmap:
1384        munmap(mmap, mmap_size);
1385        errno = EINVAL;
1386        die("index file corrupt");
1387}
1388
1389int is_index_unborn(struct index_state *istate)
1390{
1391        return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);
1392}
1393
1394int discard_index(struct index_state *istate)
1395{
1396        resolve_undo_clear_index(istate);
1397        istate->cache_nr = 0;
1398        istate->cache_changed = 0;
1399        istate->timestamp.sec = 0;
1400        istate->timestamp.nsec = 0;
1401        istate->name_hash_initialized = 0;
1402        free_hash(&istate->name_hash);
1403        cache_tree_free(&(istate->cache_tree));
1404        free(istate->alloc);
1405        istate->alloc = NULL;
1406        istate->initialized = 0;
1407
1408        /* no need to throw away allocated active_cache */
1409        return 0;
1410}
1411
1412int unmerged_index(const struct index_state *istate)
1413{
1414        int i;
1415        for (i = 0; i < istate->cache_nr; i++) {
1416                if (ce_stage(istate->cache[i]))
1417                        return 1;
1418        }
1419        return 0;
1420}
1421
1422#define WRITE_BUFFER_SIZE 8192
1423static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1424static unsigned long write_buffer_len;
1425
1426static int ce_write_flush(git_SHA_CTX *context, int fd)
1427{
1428        unsigned int buffered = write_buffer_len;
1429        if (buffered) {
1430                git_SHA1_Update(context, write_buffer, buffered);
1431                if (write_in_full(fd, write_buffer, buffered) != buffered)
1432                        return -1;
1433                write_buffer_len = 0;
1434        }
1435        return 0;
1436}
1437
1438static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1439{
1440        while (len) {
1441                unsigned int buffered = write_buffer_len;
1442                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1443                if (partial > len)
1444                        partial = len;
1445                memcpy(write_buffer + buffered, data, partial);
1446                buffered += partial;
1447                if (buffered == WRITE_BUFFER_SIZE) {
1448                        write_buffer_len = buffered;
1449                        if (ce_write_flush(context, fd))
1450                                return -1;
1451                        buffered = 0;
1452                }
1453                write_buffer_len = buffered;
1454                len -= partial;
1455                data = (char *) data + partial;
1456        }
1457        return 0;
1458}
1459
1460static int write_index_ext_header(git_SHA_CTX *context, int fd,
1461                                  unsigned int ext, unsigned int sz)
1462{
1463        ext = htonl(ext);
1464        sz = htonl(sz);
1465        return ((ce_write(context, fd, &ext, 4) < 0) ||
1466                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1467}
1468
1469static int ce_flush(git_SHA_CTX *context, int fd)
1470{
1471        unsigned int left = write_buffer_len;
1472
1473        if (left) {
1474                write_buffer_len = 0;
1475                git_SHA1_Update(context, write_buffer, left);
1476        }
1477
1478        /* Flush first if not enough space for SHA1 signature */
1479        if (left + 20 > WRITE_BUFFER_SIZE) {
1480                if (write_in_full(fd, write_buffer, left) != left)
1481                        return -1;
1482                left = 0;
1483        }
1484
1485        /* Append the SHA1 signature at the end */
1486        git_SHA1_Final(write_buffer + left, context);
1487        left += 20;
1488        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1489}
1490
1491static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1492{
1493        /*
1494         * The only thing we care about in this function is to smudge the
1495         * falsely clean entry due to touch-update-touch race, so we leave
1496         * everything else as they are.  We are called for entries whose
1497         * ce_mtime match the index file mtime.
1498         *
1499         * Note that this actually does not do much for gitlinks, for
1500         * which ce_match_stat_basic() always goes to the actual
1501         * contents.  The caller checks with is_racy_timestamp() which
1502         * always says "no" for gitlinks, so we are not called for them ;-)
1503         */
1504        struct stat st;
1505
1506        if (lstat(ce->name, &st) < 0)
1507                return;
1508        if (ce_match_stat_basic(ce, &st))
1509                return;
1510        if (ce_modified_check_fs(ce, &st)) {
1511                /* This is "racily clean"; smudge it.  Note that this
1512                 * is a tricky code.  At first glance, it may appear
1513                 * that it can break with this sequence:
1514                 *
1515                 * $ echo xyzzy >frotz
1516                 * $ git-update-index --add frotz
1517                 * $ : >frotz
1518                 * $ sleep 3
1519                 * $ echo filfre >nitfol
1520                 * $ git-update-index --add nitfol
1521                 *
1522                 * but it does not.  When the second update-index runs,
1523                 * it notices that the entry "frotz" has the same timestamp
1524                 * as index, and if we were to smudge it by resetting its
1525                 * size to zero here, then the object name recorded
1526                 * in index is the 6-byte file but the cached stat information
1527                 * becomes zero --- which would then match what we would
1528                 * obtain from the filesystem next time we stat("frotz").
1529                 *
1530                 * However, the second update-index, before calling
1531                 * this function, notices that the cached size is 6
1532                 * bytes and what is on the filesystem is an empty
1533                 * file, and never calls us, so the cached size information
1534                 * for "frotz" stays 6 which does not match the filesystem.
1535                 */
1536                ce->ce_size = 0;
1537        }
1538}
1539
1540static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
1541{
1542        int size = ondisk_ce_size(ce);
1543        struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1544        char *name;
1545        int result;
1546
1547        ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1548        ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1549        ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1550        ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1551        ondisk->dev  = htonl(ce->ce_dev);
1552        ondisk->ino  = htonl(ce->ce_ino);
1553        ondisk->mode = htonl(ce->ce_mode);
1554        ondisk->uid  = htonl(ce->ce_uid);
1555        ondisk->gid  = htonl(ce->ce_gid);
1556        ondisk->size = htonl(ce->ce_size);
1557        hashcpy(ondisk->sha1, ce->sha1);
1558        ondisk->flags = htons(ce->ce_flags);
1559        if (ce->ce_flags & CE_EXTENDED) {
1560                struct ondisk_cache_entry_extended *ondisk2;
1561                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1562                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1563                name = ondisk2->name;
1564        }
1565        else
1566                name = ondisk->name;
1567        memcpy(name, ce->name, ce_namelen(ce));
1568
1569        result = ce_write(c, fd, ondisk, size);
1570        free(ondisk);
1571        return result;
1572}
1573
1574static int has_racy_timestamp(struct index_state *istate)
1575{
1576        int entries = istate->cache_nr;
1577        int i;
1578
1579        for (i = 0; i < entries; i++) {
1580                struct cache_entry *ce = istate->cache[i];
1581                if (is_racy_timestamp(istate, ce))
1582                        return 1;
1583        }
1584        return 0;
1585}
1586
1587/*
1588 * Opportunisticly update the index but do not complain if we can't
1589 */
1590void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1591{
1592        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1593            !write_index(istate, lockfile->fd))
1594                commit_locked_index(lockfile);
1595        else
1596                rollback_lock_file(lockfile);
1597}
1598
1599int write_index(struct index_state *istate, int newfd)
1600{
1601        git_SHA_CTX c;
1602        struct cache_header hdr;
1603        int i, err, removed, extended;
1604        struct cache_entry **cache = istate->cache;
1605        int entries = istate->cache_nr;
1606        struct stat st;
1607
1608        for (i = removed = extended = 0; i < entries; i++) {
1609                if (cache[i]->ce_flags & CE_REMOVE)
1610                        removed++;
1611
1612                /* reduce extended entries if possible */
1613                cache[i]->ce_flags &= ~CE_EXTENDED;
1614                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1615                        extended++;
1616                        cache[i]->ce_flags |= CE_EXTENDED;
1617                }
1618        }
1619
1620        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1621        /* for extended format, increase version so older git won't try to read it */
1622        hdr.hdr_version = htonl(extended ? 3 : 2);
1623        hdr.hdr_entries = htonl(entries - removed);
1624
1625        git_SHA1_Init(&c);
1626        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1627                return -1;
1628
1629        for (i = 0; i < entries; i++) {
1630                struct cache_entry *ce = cache[i];
1631                if (ce->ce_flags & CE_REMOVE)
1632                        continue;
1633                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1634                        ce_smudge_racily_clean_entry(ce);
1635                if (ce_write_entry(&c, newfd, ce) < 0)
1636                        return -1;
1637        }
1638
1639        /* Write extension data here */
1640        if (istate->cache_tree) {
1641                struct strbuf sb = STRBUF_INIT;
1642
1643                cache_tree_write(&sb, istate->cache_tree);
1644                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1645                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1646                strbuf_release(&sb);
1647                if (err)
1648                        return -1;
1649        }
1650        if (istate->resolve_undo) {
1651                struct strbuf sb = STRBUF_INIT;
1652
1653                resolve_undo_write(&sb, istate->resolve_undo);
1654                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1655                                             sb.len) < 0
1656                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1657                strbuf_release(&sb);
1658                if (err)
1659                        return -1;
1660        }
1661
1662        if (ce_flush(&c, newfd) || fstat(newfd, &st))
1663                return -1;
1664        istate->timestamp.sec = (unsigned int)st.st_mtime;
1665        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1666        return 0;
1667}
1668
1669/*
1670 * Read the index file that is potentially unmerged into given
1671 * index_state, dropping any unmerged entries.  Returns true if
1672 * the index is unmerged.  Callers who want to refuse to work
1673 * from an unmerged state can call this and check its return value,
1674 * instead of calling read_cache().
1675 */
1676int read_index_unmerged(struct index_state *istate)
1677{
1678        int i;
1679        int unmerged = 0;
1680
1681        read_index(istate);
1682        for (i = 0; i < istate->cache_nr; i++) {
1683                struct cache_entry *ce = istate->cache[i];
1684                struct cache_entry *new_ce;
1685                int size, len;
1686
1687                if (!ce_stage(ce))
1688                        continue;
1689                unmerged = 1;
1690                len = strlen(ce->name);
1691                size = cache_entry_size(len);
1692                new_ce = xcalloc(1, size);
1693                memcpy(new_ce->name, ce->name, len);
1694                new_ce->ce_flags = create_ce_flags(len, 0) | CE_CONFLICTED;
1695                new_ce->ce_mode = ce->ce_mode;
1696                if (add_index_entry(istate, new_ce, 0))
1697                        return error("%s: cannot drop to stage #0",
1698                                     ce->name);
1699                i = index_name_pos(istate, new_ce->name, len);
1700        }
1701        return unmerged;
1702}
1703
1704/*
1705 * Returns 1 if the path is an "other" path with respect to
1706 * the index; that is, the path is not mentioned in the index at all,
1707 * either as a file, a directory with some files in the index,
1708 * or as an unmerged entry.
1709 *
1710 * We helpfully remove a trailing "/" from directories so that
1711 * the output of read_directory can be used as-is.
1712 */
1713int index_name_is_other(const struct index_state *istate, const char *name,
1714                int namelen)
1715{
1716        int pos;
1717        if (namelen && name[namelen - 1] == '/')
1718                namelen--;
1719        pos = index_name_pos(istate, name, namelen);
1720        if (0 <= pos)
1721                return 0;       /* exact match */
1722        pos = -pos - 1;
1723        if (pos < istate->cache_nr) {
1724                struct cache_entry *ce = istate->cache[pos];
1725                if (ce_namelen(ce) == namelen &&
1726                    !memcmp(ce->name, name, namelen))
1727                        return 0; /* Yup, this one exists unmerged */
1728        }
1729        return 1;
1730}