read-cache.con commit do not write null sha1s to on-disk index (4337b58)
   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 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, HASH_WRITE_OBJECT))
 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 struct pathspec *pathspec)
 710{
 711        return match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, NULL);
 712}
 713
 714/*
 715 * We fundamentally don't like some paths: we don't want
 716 * dot or dot-dot anywhere, and for obvious reasons don't
 717 * want to recurse into ".git" either.
 718 *
 719 * Also, we don't want double slashes or slashes at the
 720 * end that can make pathnames ambiguous.
 721 */
 722static int verify_dotfile(const char *rest)
 723{
 724        /*
 725         * The first character was '.', but that
 726         * has already been discarded, we now test
 727         * the rest.
 728         */
 729
 730        /* "." is not allowed */
 731        if (*rest == '\0' || is_dir_sep(*rest))
 732                return 0;
 733
 734        switch (*rest) {
 735        /*
 736         * ".git" followed by  NUL or slash is bad. This
 737         * shares the path end test with the ".." case.
 738         */
 739        case 'g':
 740                if (rest[1] != 'i')
 741                        break;
 742                if (rest[2] != 't')
 743                        break;
 744                rest += 2;
 745        /* fallthrough */
 746        case '.':
 747                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 748                        return 0;
 749        }
 750        return 1;
 751}
 752
 753int verify_path(const char *path)
 754{
 755        char c;
 756
 757        if (has_dos_drive_prefix(path))
 758                return 0;
 759
 760        goto inside;
 761        for (;;) {
 762                if (!c)
 763                        return 1;
 764                if (is_dir_sep(c)) {
 765inside:
 766                        c = *path++;
 767                        if ((c == '.' && !verify_dotfile(path)) ||
 768                            is_dir_sep(c) || c == '\0')
 769                                return 0;
 770                }
 771                c = *path++;
 772        }
 773}
 774
 775/*
 776 * Do we have another file that has the beginning components being a
 777 * proper superset of the name we're trying to add?
 778 */
 779static int has_file_name(struct index_state *istate,
 780                         const struct cache_entry *ce, int pos, int ok_to_replace)
 781{
 782        int retval = 0;
 783        int len = ce_namelen(ce);
 784        int stage = ce_stage(ce);
 785        const char *name = ce->name;
 786
 787        while (pos < istate->cache_nr) {
 788                struct cache_entry *p = istate->cache[pos++];
 789
 790                if (len >= ce_namelen(p))
 791                        break;
 792                if (memcmp(name, p->name, len))
 793                        break;
 794                if (ce_stage(p) != stage)
 795                        continue;
 796                if (p->name[len] != '/')
 797                        continue;
 798                if (p->ce_flags & CE_REMOVE)
 799                        continue;
 800                retval = -1;
 801                if (!ok_to_replace)
 802                        break;
 803                remove_index_entry_at(istate, --pos);
 804        }
 805        return retval;
 806}
 807
 808/*
 809 * Do we have another file with a pathname that is a proper
 810 * subset of the name we're trying to add?
 811 */
 812static int has_dir_name(struct index_state *istate,
 813                        const struct cache_entry *ce, int pos, int ok_to_replace)
 814{
 815        int retval = 0;
 816        int stage = ce_stage(ce);
 817        const char *name = ce->name;
 818        const char *slash = name + ce_namelen(ce);
 819
 820        for (;;) {
 821                int len;
 822
 823                for (;;) {
 824                        if (*--slash == '/')
 825                                break;
 826                        if (slash <= ce->name)
 827                                return retval;
 828                }
 829                len = slash - name;
 830
 831                pos = index_name_pos(istate, name, create_ce_flags(len, stage));
 832                if (pos >= 0) {
 833                        /*
 834                         * Found one, but not so fast.  This could
 835                         * be a marker that says "I was here, but
 836                         * I am being removed".  Such an entry is
 837                         * not a part of the resulting tree, and
 838                         * it is Ok to have a directory at the same
 839                         * path.
 840                         */
 841                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 842                                retval = -1;
 843                                if (!ok_to_replace)
 844                                        break;
 845                                remove_index_entry_at(istate, pos);
 846                                continue;
 847                        }
 848                }
 849                else
 850                        pos = -pos-1;
 851
 852                /*
 853                 * Trivial optimization: if we find an entry that
 854                 * already matches the sub-directory, then we know
 855                 * we're ok, and we can exit.
 856                 */
 857                while (pos < istate->cache_nr) {
 858                        struct cache_entry *p = istate->cache[pos];
 859                        if ((ce_namelen(p) <= len) ||
 860                            (p->name[len] != '/') ||
 861                            memcmp(p->name, name, len))
 862                                break; /* not our subdirectory */
 863                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 864                                /*
 865                                 * p is at the same stage as our entry, and
 866                                 * is a subdirectory of what we are looking
 867                                 * at, so we cannot have conflicts at our
 868                                 * level or anything shorter.
 869                                 */
 870                                return retval;
 871                        pos++;
 872                }
 873        }
 874        return retval;
 875}
 876
 877/* We may be in a situation where we already have path/file and path
 878 * is being added, or we already have path and path/file is being
 879 * added.  Either one would result in a nonsense tree that has path
 880 * twice when git-write-tree tries to write it out.  Prevent it.
 881 *
 882 * If ok-to-replace is specified, we remove the conflicting entries
 883 * from the cache so the caller should recompute the insert position.
 884 * When this happens, we return non-zero.
 885 */
 886static int check_file_directory_conflict(struct index_state *istate,
 887                                         const struct cache_entry *ce,
 888                                         int pos, int ok_to_replace)
 889{
 890        int retval;
 891
 892        /*
 893         * When ce is an "I am going away" entry, we allow it to be added
 894         */
 895        if (ce->ce_flags & CE_REMOVE)
 896                return 0;
 897
 898        /*
 899         * We check if the path is a sub-path of a subsequent pathname
 900         * first, since removing those will not change the position
 901         * in the array.
 902         */
 903        retval = has_file_name(istate, ce, pos, ok_to_replace);
 904
 905        /*
 906         * Then check if the path might have a clashing sub-directory
 907         * before it.
 908         */
 909        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 910}
 911
 912static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 913{
 914        int pos;
 915        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 916        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 917        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 918        int new_only = option & ADD_CACHE_NEW_ONLY;
 919
 920        cache_tree_invalidate_path(istate->cache_tree, ce->name);
 921        pos = index_name_pos(istate, ce->name, ce->ce_flags);
 922
 923        /* existing match? Just replace it. */
 924        if (pos >= 0) {
 925                if (!new_only)
 926                        replace_index_entry(istate, pos, ce);
 927                return 0;
 928        }
 929        pos = -pos-1;
 930
 931        /*
 932         * Inserting a merged entry ("stage 0") into the index
 933         * will always replace all non-merged entries..
 934         */
 935        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 936                while (ce_same_name(istate->cache[pos], ce)) {
 937                        ok_to_add = 1;
 938                        if (!remove_index_entry_at(istate, pos))
 939                                break;
 940                }
 941        }
 942
 943        if (!ok_to_add)
 944                return -1;
 945        if (!verify_path(ce->name))
 946                return error("Invalid path '%s'", ce->name);
 947
 948        if (!skip_df_check &&
 949            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 950                if (!ok_to_replace)
 951                        return error("'%s' appears as both a file and as a directory",
 952                                     ce->name);
 953                pos = index_name_pos(istate, ce->name, ce->ce_flags);
 954                pos = -pos-1;
 955        }
 956        return pos + 1;
 957}
 958
 959int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 960{
 961        int pos;
 962
 963        if (option & ADD_CACHE_JUST_APPEND)
 964                pos = istate->cache_nr;
 965        else {
 966                int ret;
 967                ret = add_index_entry_with_check(istate, ce, option);
 968                if (ret <= 0)
 969                        return ret;
 970                pos = ret - 1;
 971        }
 972
 973        /* Make sure the array is big enough .. */
 974        if (istate->cache_nr == istate->cache_alloc) {
 975                istate->cache_alloc = alloc_nr(istate->cache_alloc);
 976                istate->cache = xrealloc(istate->cache,
 977                                        istate->cache_alloc * sizeof(struct cache_entry *));
 978        }
 979
 980        /* Add it in.. */
 981        istate->cache_nr++;
 982        if (istate->cache_nr > pos + 1)
 983                memmove(istate->cache + pos + 1,
 984                        istate->cache + pos,
 985                        (istate->cache_nr - pos - 1) * sizeof(ce));
 986        set_index_entry(istate, pos, ce);
 987        istate->cache_changed = 1;
 988        return 0;
 989}
 990
 991/*
 992 * "refresh" does not calculate a new sha1 file or bring the
 993 * cache up-to-date for mode/content changes. But what it
 994 * _does_ do is to "re-match" the stat information of a file
 995 * with the cache, so that you can refresh the cache for a
 996 * file that hasn't been changed but where the stat entry is
 997 * out of date.
 998 *
 999 * For example, you'd want to do this after doing a "git-read-tree",
1000 * to link up the stat cache details with the proper files.
1001 */
1002static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1003                                             struct cache_entry *ce,
1004                                             unsigned int options, int *err,
1005                                             int *changed_ret)
1006{
1007        struct stat st;
1008        struct cache_entry *updated;
1009        int changed, size;
1010        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1011        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1012
1013        if (ce_uptodate(ce))
1014                return ce;
1015
1016        /*
1017         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1018         * that the change to the work tree does not matter and told
1019         * us not to worry.
1020         */
1021        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1022                ce_mark_uptodate(ce);
1023                return ce;
1024        }
1025        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1026                ce_mark_uptodate(ce);
1027                return ce;
1028        }
1029
1030        if (lstat(ce->name, &st) < 0) {
1031                if (err)
1032                        *err = errno;
1033                return NULL;
1034        }
1035
1036        changed = ie_match_stat(istate, ce, &st, options);
1037        if (changed_ret)
1038                *changed_ret = changed;
1039        if (!changed) {
1040                /*
1041                 * The path is unchanged.  If we were told to ignore
1042                 * valid bit, then we did the actual stat check and
1043                 * found that the entry is unmodified.  If the entry
1044                 * is not marked VALID, this is the place to mark it
1045                 * valid again, under "assume unchanged" mode.
1046                 */
1047                if (ignore_valid && assume_unchanged &&
1048                    !(ce->ce_flags & CE_VALID))
1049                        ; /* mark this one VALID again */
1050                else {
1051                        /*
1052                         * We do not mark the index itself "modified"
1053                         * because CE_UPTODATE flag is in-core only;
1054                         * we are not going to write this change out.
1055                         */
1056                        if (!S_ISGITLINK(ce->ce_mode))
1057                                ce_mark_uptodate(ce);
1058                        return ce;
1059                }
1060        }
1061
1062        if (ie_modified(istate, ce, &st, options)) {
1063                if (err)
1064                        *err = EINVAL;
1065                return NULL;
1066        }
1067
1068        size = ce_size(ce);
1069        updated = xmalloc(size);
1070        memcpy(updated, ce, size);
1071        fill_stat_cache_info(updated, &st);
1072        /*
1073         * If ignore_valid is not set, we should leave CE_VALID bit
1074         * alone.  Otherwise, paths marked with --no-assume-unchanged
1075         * (i.e. things to be edited) will reacquire CE_VALID bit
1076         * automatically, which is not really what we want.
1077         */
1078        if (!ignore_valid && assume_unchanged &&
1079            !(ce->ce_flags & CE_VALID))
1080                updated->ce_flags &= ~CE_VALID;
1081
1082        return updated;
1083}
1084
1085static void show_file(const char * fmt, const char * name, int in_porcelain,
1086                      int * first, const char *header_msg)
1087{
1088        if (in_porcelain && *first && header_msg) {
1089                printf("%s\n", header_msg);
1090                *first = 0;
1091        }
1092        printf(fmt, name);
1093}
1094
1095int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1096                  char *seen, const char *header_msg)
1097{
1098        int i;
1099        int has_errors = 0;
1100        int really = (flags & REFRESH_REALLY) != 0;
1101        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1102        int quiet = (flags & REFRESH_QUIET) != 0;
1103        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1104        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1105        int first = 1;
1106        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1107        unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1108        const char *modified_fmt;
1109        const char *deleted_fmt;
1110        const char *typechange_fmt;
1111        const char *added_fmt;
1112        const char *unmerged_fmt;
1113
1114        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1115        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1116        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1117        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1118        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1119        for (i = 0; i < istate->cache_nr; i++) {
1120                struct cache_entry *ce, *new;
1121                int cache_errno = 0;
1122                int changed = 0;
1123                int filtered = 0;
1124
1125                ce = istate->cache[i];
1126                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1127                        continue;
1128
1129                if (pathspec &&
1130                    !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1131                        filtered = 1;
1132
1133                if (ce_stage(ce)) {
1134                        while ((i < istate->cache_nr) &&
1135                               ! strcmp(istate->cache[i]->name, ce->name))
1136                                i++;
1137                        i--;
1138                        if (allow_unmerged)
1139                                continue;
1140                        if (!filtered)
1141                                show_file(unmerged_fmt, ce->name, in_porcelain,
1142                                          &first, header_msg);
1143                        has_errors = 1;
1144                        continue;
1145                }
1146
1147                if (filtered)
1148                        continue;
1149
1150                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1151                if (new == ce)
1152                        continue;
1153                if (!new) {
1154                        const char *fmt;
1155
1156                        if (not_new && cache_errno == ENOENT)
1157                                continue;
1158                        if (really && cache_errno == EINVAL) {
1159                                /* If we are doing --really-refresh that
1160                                 * means the index is not valid anymore.
1161                                 */
1162                                ce->ce_flags &= ~CE_VALID;
1163                                istate->cache_changed = 1;
1164                        }
1165                        if (quiet)
1166                                continue;
1167
1168                        if (cache_errno == ENOENT)
1169                                fmt = deleted_fmt;
1170                        else if (ce->ce_flags & CE_INTENT_TO_ADD)
1171                                fmt = added_fmt; /* must be before other checks */
1172                        else if (changed & TYPE_CHANGED)
1173                                fmt = typechange_fmt;
1174                        else
1175                                fmt = modified_fmt;
1176                        show_file(fmt,
1177                                  ce->name, in_porcelain, &first, header_msg);
1178                        has_errors = 1;
1179                        continue;
1180                }
1181
1182                replace_index_entry(istate, i, new);
1183        }
1184        return has_errors;
1185}
1186
1187static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1188{
1189        return refresh_cache_ent(&the_index, ce, really, NULL, NULL);
1190}
1191
1192static int verify_hdr(struct cache_header *hdr, unsigned long size)
1193{
1194        git_SHA_CTX c;
1195        unsigned char sha1[20];
1196
1197        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1198                return error("bad signature");
1199        if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))
1200                return error("bad index version");
1201        git_SHA1_Init(&c);
1202        git_SHA1_Update(&c, hdr, size - 20);
1203        git_SHA1_Final(sha1, &c);
1204        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1205                return error("bad index file sha1 signature");
1206        return 0;
1207}
1208
1209static int read_index_extension(struct index_state *istate,
1210                                const char *ext, void *data, unsigned long sz)
1211{
1212        switch (CACHE_EXT(ext)) {
1213        case CACHE_EXT_TREE:
1214                istate->cache_tree = cache_tree_read(data, sz);
1215                break;
1216        case CACHE_EXT_RESOLVE_UNDO:
1217                istate->resolve_undo = resolve_undo_read(data, sz);
1218                break;
1219        default:
1220                if (*ext < 'A' || 'Z' < *ext)
1221                        return error("index uses %.4s extension, which we do not understand",
1222                                     ext);
1223                fprintf(stderr, "ignoring %.4s extension\n", ext);
1224                break;
1225        }
1226        return 0;
1227}
1228
1229int read_index(struct index_state *istate)
1230{
1231        return read_index_from(istate, get_index_file());
1232}
1233
1234static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk)
1235{
1236        struct cache_entry *ce;
1237        size_t len;
1238        const char *name;
1239        unsigned int flags;
1240
1241        /* On-disk flags are just 16 bits */
1242        flags = ntohs(ondisk->flags);
1243        len = flags & CE_NAMEMASK;
1244
1245        if (flags & CE_EXTENDED) {
1246                struct ondisk_cache_entry_extended *ondisk2;
1247                int extended_flags;
1248                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1249                extended_flags = ntohs(ondisk2->flags2) << 16;
1250                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1251                if (extended_flags & ~CE_EXTENDED_FLAGS)
1252                        die("Unknown index entry format %08x", extended_flags);
1253                flags |= extended_flags;
1254                name = ondisk2->name;
1255        }
1256        else
1257                name = ondisk->name;
1258
1259        if (len == CE_NAMEMASK)
1260                len = strlen(name);
1261
1262        ce = xmalloc(cache_entry_size(len));
1263
1264        ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);
1265        ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);
1266        ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);
1267        ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);
1268        ce->ce_dev   = ntohl(ondisk->dev);
1269        ce->ce_ino   = ntohl(ondisk->ino);
1270        ce->ce_mode  = ntohl(ondisk->mode);
1271        ce->ce_uid   = ntohl(ondisk->uid);
1272        ce->ce_gid   = ntohl(ondisk->gid);
1273        ce->ce_size  = ntohl(ondisk->size);
1274        ce->ce_flags = flags;
1275
1276        hashcpy(ce->sha1, ondisk->sha1);
1277
1278        memcpy(ce->name, name, len);
1279        ce->name[len] = '\0';
1280        return ce;
1281}
1282
1283/* remember to discard_cache() before reading a different cache! */
1284int read_index_from(struct index_state *istate, const char *path)
1285{
1286        int fd, i;
1287        struct stat st;
1288        unsigned long src_offset;
1289        struct cache_header *hdr;
1290        void *mmap;
1291        size_t mmap_size;
1292
1293        errno = EBUSY;
1294        if (istate->initialized)
1295                return istate->cache_nr;
1296
1297        errno = ENOENT;
1298        istate->timestamp.sec = 0;
1299        istate->timestamp.nsec = 0;
1300        fd = open(path, O_RDONLY);
1301        if (fd < 0) {
1302                if (errno == ENOENT)
1303                        return 0;
1304                die_errno("index file open failed");
1305        }
1306
1307        if (fstat(fd, &st))
1308                die_errno("cannot stat the open index");
1309
1310        errno = EINVAL;
1311        mmap_size = xsize_t(st.st_size);
1312        if (mmap_size < sizeof(struct cache_header) + 20)
1313                die("index file smaller than expected");
1314
1315        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1316        close(fd);
1317        if (mmap == MAP_FAILED)
1318                die_errno("unable to map index file");
1319
1320        hdr = mmap;
1321        if (verify_hdr(hdr, mmap_size) < 0)
1322                goto unmap;
1323
1324        istate->cache_nr = ntohl(hdr->hdr_entries);
1325        istate->cache_alloc = alloc_nr(istate->cache_nr);
1326        istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1327        istate->initialized = 1;
1328
1329        src_offset = sizeof(*hdr);
1330        for (i = 0; i < istate->cache_nr; i++) {
1331                struct ondisk_cache_entry *disk_ce;
1332                struct cache_entry *ce;
1333
1334                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1335                ce = create_from_disk(disk_ce);
1336                set_index_entry(istate, i, ce);
1337
1338                src_offset += ondisk_ce_size(ce);
1339        }
1340        istate->timestamp.sec = st.st_mtime;
1341        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1342
1343        while (src_offset <= mmap_size - 20 - 8) {
1344                /* After an array of active_nr index entries,
1345                 * there can be arbitrary number of extended
1346                 * sections, each of which is prefixed with
1347                 * extension name (4-byte) and section length
1348                 * in 4-byte network byte order.
1349                 */
1350                uint32_t extsize;
1351                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1352                extsize = ntohl(extsize);
1353                if (read_index_extension(istate,
1354                                         (const char *) mmap + src_offset,
1355                                         (char *) mmap + src_offset + 8,
1356                                         extsize) < 0)
1357                        goto unmap;
1358                src_offset += 8;
1359                src_offset += extsize;
1360        }
1361        munmap(mmap, mmap_size);
1362        return istate->cache_nr;
1363
1364unmap:
1365        munmap(mmap, mmap_size);
1366        errno = EINVAL;
1367        die("index file corrupt");
1368}
1369
1370int is_index_unborn(struct index_state *istate)
1371{
1372        return (!istate->cache_nr && !istate->timestamp.sec);
1373}
1374
1375int discard_index(struct index_state *istate)
1376{
1377        int i;
1378
1379        for (i = 0; i < istate->cache_nr; i++)
1380                free(istate->cache[i]);
1381        resolve_undo_clear_index(istate);
1382        istate->cache_nr = 0;
1383        istate->cache_changed = 0;
1384        istate->timestamp.sec = 0;
1385        istate->timestamp.nsec = 0;
1386        istate->name_hash_initialized = 0;
1387        free_hash(&istate->name_hash);
1388        cache_tree_free(&(istate->cache_tree));
1389        istate->initialized = 0;
1390
1391        /* no need to throw away allocated active_cache */
1392        return 0;
1393}
1394
1395int unmerged_index(const struct index_state *istate)
1396{
1397        int i;
1398        for (i = 0; i < istate->cache_nr; i++) {
1399                if (ce_stage(istate->cache[i]))
1400                        return 1;
1401        }
1402        return 0;
1403}
1404
1405#define WRITE_BUFFER_SIZE 8192
1406static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1407static unsigned long write_buffer_len;
1408
1409static int ce_write_flush(git_SHA_CTX *context, int fd)
1410{
1411        unsigned int buffered = write_buffer_len;
1412        if (buffered) {
1413                git_SHA1_Update(context, write_buffer, buffered);
1414                if (write_in_full(fd, write_buffer, buffered) != buffered)
1415                        return -1;
1416                write_buffer_len = 0;
1417        }
1418        return 0;
1419}
1420
1421static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1422{
1423        while (len) {
1424                unsigned int buffered = write_buffer_len;
1425                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1426                if (partial > len)
1427                        partial = len;
1428                memcpy(write_buffer + buffered, data, partial);
1429                buffered += partial;
1430                if (buffered == WRITE_BUFFER_SIZE) {
1431                        write_buffer_len = buffered;
1432                        if (ce_write_flush(context, fd))
1433                                return -1;
1434                        buffered = 0;
1435                }
1436                write_buffer_len = buffered;
1437                len -= partial;
1438                data = (char *) data + partial;
1439        }
1440        return 0;
1441}
1442
1443static int write_index_ext_header(git_SHA_CTX *context, int fd,
1444                                  unsigned int ext, unsigned int sz)
1445{
1446        ext = htonl(ext);
1447        sz = htonl(sz);
1448        return ((ce_write(context, fd, &ext, 4) < 0) ||
1449                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1450}
1451
1452static int ce_flush(git_SHA_CTX *context, int fd)
1453{
1454        unsigned int left = write_buffer_len;
1455
1456        if (left) {
1457                write_buffer_len = 0;
1458                git_SHA1_Update(context, write_buffer, left);
1459        }
1460
1461        /* Flush first if not enough space for SHA1 signature */
1462        if (left + 20 > WRITE_BUFFER_SIZE) {
1463                if (write_in_full(fd, write_buffer, left) != left)
1464                        return -1;
1465                left = 0;
1466        }
1467
1468        /* Append the SHA1 signature at the end */
1469        git_SHA1_Final(write_buffer + left, context);
1470        left += 20;
1471        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1472}
1473
1474static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1475{
1476        /*
1477         * The only thing we care about in this function is to smudge the
1478         * falsely clean entry due to touch-update-touch race, so we leave
1479         * everything else as they are.  We are called for entries whose
1480         * ce_mtime match the index file mtime.
1481         *
1482         * Note that this actually does not do much for gitlinks, for
1483         * which ce_match_stat_basic() always goes to the actual
1484         * contents.  The caller checks with is_racy_timestamp() which
1485         * always says "no" for gitlinks, so we are not called for them ;-)
1486         */
1487        struct stat st;
1488
1489        if (lstat(ce->name, &st) < 0)
1490                return;
1491        if (ce_match_stat_basic(ce, &st))
1492                return;
1493        if (ce_modified_check_fs(ce, &st)) {
1494                /* This is "racily clean"; smudge it.  Note that this
1495                 * is a tricky code.  At first glance, it may appear
1496                 * that it can break with this sequence:
1497                 *
1498                 * $ echo xyzzy >frotz
1499                 * $ git-update-index --add frotz
1500                 * $ : >frotz
1501                 * $ sleep 3
1502                 * $ echo filfre >nitfol
1503                 * $ git-update-index --add nitfol
1504                 *
1505                 * but it does not.  When the second update-index runs,
1506                 * it notices that the entry "frotz" has the same timestamp
1507                 * as index, and if we were to smudge it by resetting its
1508                 * size to zero here, then the object name recorded
1509                 * in index is the 6-byte file but the cached stat information
1510                 * becomes zero --- which would then match what we would
1511                 * obtain from the filesystem next time we stat("frotz").
1512                 *
1513                 * However, the second update-index, before calling
1514                 * this function, notices that the cached size is 6
1515                 * bytes and what is on the filesystem is an empty
1516                 * file, and never calls us, so the cached size information
1517                 * for "frotz" stays 6 which does not match the filesystem.
1518                 */
1519                ce->ce_size = 0;
1520        }
1521}
1522
1523static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
1524{
1525        int size = ondisk_ce_size(ce);
1526        struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1527        char *name;
1528        int result;
1529
1530        ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1531        ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1532        ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1533        ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1534        ondisk->dev  = htonl(ce->ce_dev);
1535        ondisk->ino  = htonl(ce->ce_ino);
1536        ondisk->mode = htonl(ce->ce_mode);
1537        ondisk->uid  = htonl(ce->ce_uid);
1538        ondisk->gid  = htonl(ce->ce_gid);
1539        ondisk->size = htonl(ce->ce_size);
1540        hashcpy(ondisk->sha1, ce->sha1);
1541        ondisk->flags = htons(ce->ce_flags);
1542        if (ce->ce_flags & CE_EXTENDED) {
1543                struct ondisk_cache_entry_extended *ondisk2;
1544                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1545                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1546                name = ondisk2->name;
1547        }
1548        else
1549                name = ondisk->name;
1550        memcpy(name, ce->name, ce_namelen(ce));
1551
1552        result = ce_write(c, fd, ondisk, size);
1553        free(ondisk);
1554        return result;
1555}
1556
1557static int has_racy_timestamp(struct index_state *istate)
1558{
1559        int entries = istate->cache_nr;
1560        int i;
1561
1562        for (i = 0; i < entries; i++) {
1563                struct cache_entry *ce = istate->cache[i];
1564                if (is_racy_timestamp(istate, ce))
1565                        return 1;
1566        }
1567        return 0;
1568}
1569
1570/*
1571 * Opportunisticly update the index but do not complain if we can't
1572 */
1573void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1574{
1575        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1576            !write_index(istate, lockfile->fd))
1577                commit_locked_index(lockfile);
1578        else
1579                rollback_lock_file(lockfile);
1580}
1581
1582int write_index(struct index_state *istate, int newfd)
1583{
1584        git_SHA_CTX c;
1585        struct cache_header hdr;
1586        int i, err, removed, extended;
1587        struct cache_entry **cache = istate->cache;
1588        int entries = istate->cache_nr;
1589        struct stat st;
1590
1591        for (i = removed = extended = 0; i < entries; i++) {
1592                if (cache[i]->ce_flags & CE_REMOVE)
1593                        removed++;
1594
1595                /* reduce extended entries if possible */
1596                cache[i]->ce_flags &= ~CE_EXTENDED;
1597                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1598                        extended++;
1599                        cache[i]->ce_flags |= CE_EXTENDED;
1600                }
1601        }
1602
1603        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1604        /* for extended format, increase version so older git won't try to read it */
1605        hdr.hdr_version = htonl(extended ? 3 : 2);
1606        hdr.hdr_entries = htonl(entries - removed);
1607
1608        git_SHA1_Init(&c);
1609        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1610                return -1;
1611
1612        for (i = 0; i < entries; i++) {
1613                struct cache_entry *ce = cache[i];
1614                if (ce->ce_flags & CE_REMOVE)
1615                        continue;
1616                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1617                        ce_smudge_racily_clean_entry(ce);
1618                if (is_null_sha1(ce->sha1))
1619                        return error("cache entry has null sha1: %s", ce->name);
1620                if (ce_write_entry(&c, newfd, ce) < 0)
1621                        return -1;
1622        }
1623
1624        /* Write extension data here */
1625        if (istate->cache_tree) {
1626                struct strbuf sb = STRBUF_INIT;
1627
1628                cache_tree_write(&sb, istate->cache_tree);
1629                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1630                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1631                strbuf_release(&sb);
1632                if (err)
1633                        return -1;
1634        }
1635        if (istate->resolve_undo) {
1636                struct strbuf sb = STRBUF_INIT;
1637
1638                resolve_undo_write(&sb, istate->resolve_undo);
1639                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1640                                             sb.len) < 0
1641                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1642                strbuf_release(&sb);
1643                if (err)
1644                        return -1;
1645        }
1646
1647        if (ce_flush(&c, newfd) || fstat(newfd, &st))
1648                return -1;
1649        istate->timestamp.sec = (unsigned int)st.st_mtime;
1650        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1651        return 0;
1652}
1653
1654/*
1655 * Read the index file that is potentially unmerged into given
1656 * index_state, dropping any unmerged entries.  Returns true if
1657 * the index is unmerged.  Callers who want to refuse to work
1658 * from an unmerged state can call this and check its return value,
1659 * instead of calling read_cache().
1660 */
1661int read_index_unmerged(struct index_state *istate)
1662{
1663        int i;
1664        int unmerged = 0;
1665
1666        read_index(istate);
1667        for (i = 0; i < istate->cache_nr; i++) {
1668                struct cache_entry *ce = istate->cache[i];
1669                struct cache_entry *new_ce;
1670                int size, len;
1671
1672                if (!ce_stage(ce))
1673                        continue;
1674                unmerged = 1;
1675                len = strlen(ce->name);
1676                size = cache_entry_size(len);
1677                new_ce = xcalloc(1, size);
1678                memcpy(new_ce->name, ce->name, len);
1679                new_ce->ce_flags = create_ce_flags(len, 0) | CE_CONFLICTED;
1680                new_ce->ce_mode = ce->ce_mode;
1681                if (add_index_entry(istate, new_ce, 0))
1682                        return error("%s: cannot drop to stage #0",
1683                                     ce->name);
1684                i = index_name_pos(istate, new_ce->name, len);
1685        }
1686        return unmerged;
1687}
1688
1689/*
1690 * Returns 1 if the path is an "other" path with respect to
1691 * the index; that is, the path is not mentioned in the index at all,
1692 * either as a file, a directory with some files in the index,
1693 * or as an unmerged entry.
1694 *
1695 * We helpfully remove a trailing "/" from directories so that
1696 * the output of read_directory can be used as-is.
1697 */
1698int index_name_is_other(const struct index_state *istate, const char *name,
1699                int namelen)
1700{
1701        int pos;
1702        if (namelen && name[namelen - 1] == '/')
1703                namelen--;
1704        pos = index_name_pos(istate, name, namelen);
1705        if (0 <= pos)
1706                return 0;       /* exact match */
1707        pos = -pos - 1;
1708        if (pos < istate->cache_nr) {
1709                struct cache_entry *ce = istate->cache[pos];
1710                if (ce_namelen(ce) == namelen &&
1711                    !memcmp(ce->name, name, namelen))
1712                        return 0; /* Yup, this one exists unmerged */
1713        }
1714        return 1;
1715}