read-cache.con commit Merge branch 'jc/em-dash-in-doc' (7316767)
   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 "tempfile.h"
   9#include "lockfile.h"
  10#include "cache-tree.h"
  11#include "refs.h"
  12#include "dir.h"
  13#include "tree.h"
  14#include "commit.h"
  15#include "blob.h"
  16#include "resolve-undo.h"
  17#include "strbuf.h"
  18#include "varint.h"
  19#include "split-index.h"
  20#include "sigchain.h"
  21#include "utf8.h"
  22
  23static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
  24                                               unsigned int options);
  25
  26/* Mask for the name length in ce_flags in the on-disk index */
  27
  28#define CE_NAMEMASK  (0x0fff)
  29
  30/* Index extensions.
  31 *
  32 * The first letter should be 'A'..'Z' for extensions that are not
  33 * necessary for a correct operation (i.e. optimization data).
  34 * When new extensions are added that _needs_ to be understood in
  35 * order to correctly interpret the index file, pick character that
  36 * is outside the range, to cause the reader to abort.
  37 */
  38
  39#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  40#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  41#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  42#define CACHE_EXT_LINK 0x6c696e6b         /* "link" */
  43#define CACHE_EXT_UNTRACKED 0x554E5452    /* "UNTR" */
  44
  45/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
  46#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
  47                 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
  48                 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
  49
  50struct index_state the_index;
  51static const char *alternate_index_output;
  52
  53static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  54{
  55        istate->cache[nr] = ce;
  56        add_name_hash(istate, ce);
  57}
  58
  59static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  60{
  61        struct cache_entry *old = istate->cache[nr];
  62
  63        replace_index_entry_in_base(istate, old, ce);
  64        remove_name_hash(istate, old);
  65        free(old);
  66        set_index_entry(istate, nr, ce);
  67        ce->ce_flags |= CE_UPDATE_IN_BASE;
  68        istate->cache_changed |= CE_ENTRY_CHANGED;
  69}
  70
  71void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  72{
  73        struct cache_entry *old = istate->cache[nr], *new;
  74        int namelen = strlen(new_name);
  75
  76        new = xmalloc(cache_entry_size(namelen));
  77        copy_cache_entry(new, old);
  78        new->ce_flags &= ~CE_HASHED;
  79        new->ce_namelen = namelen;
  80        new->index = 0;
  81        memcpy(new->name, new_name, namelen + 1);
  82
  83        cache_tree_invalidate_path(istate, old->name);
  84        untracked_cache_remove_from_index(istate, old->name);
  85        remove_index_entry_at(istate, nr);
  86        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  87}
  88
  89void fill_stat_data(struct stat_data *sd, struct stat *st)
  90{
  91        sd->sd_ctime.sec = (unsigned int)st->st_ctime;
  92        sd->sd_mtime.sec = (unsigned int)st->st_mtime;
  93        sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
  94        sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
  95        sd->sd_dev = st->st_dev;
  96        sd->sd_ino = st->st_ino;
  97        sd->sd_uid = st->st_uid;
  98        sd->sd_gid = st->st_gid;
  99        sd->sd_size = st->st_size;
 100}
 101
 102int match_stat_data(const struct stat_data *sd, struct stat *st)
 103{
 104        int changed = 0;
 105
 106        if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
 107                changed |= MTIME_CHANGED;
 108        if (trust_ctime && check_stat &&
 109            sd->sd_ctime.sec != (unsigned int)st->st_ctime)
 110                changed |= CTIME_CHANGED;
 111
 112#ifdef USE_NSEC
 113        if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
 114                changed |= MTIME_CHANGED;
 115        if (trust_ctime && check_stat &&
 116            sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
 117                changed |= CTIME_CHANGED;
 118#endif
 119
 120        if (check_stat) {
 121                if (sd->sd_uid != (unsigned int) st->st_uid ||
 122                        sd->sd_gid != (unsigned int) st->st_gid)
 123                        changed |= OWNER_CHANGED;
 124                if (sd->sd_ino != (unsigned int) st->st_ino)
 125                        changed |= INODE_CHANGED;
 126        }
 127
 128#ifdef USE_STDEV
 129        /*
 130         * st_dev breaks on network filesystems where different
 131         * clients will have different views of what "device"
 132         * the filesystem is on
 133         */
 134        if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
 135                        changed |= INODE_CHANGED;
 136#endif
 137
 138        if (sd->sd_size != (unsigned int) st->st_size)
 139                changed |= DATA_CHANGED;
 140
 141        return changed;
 142}
 143
 144/*
 145 * This only updates the "non-critical" parts of the directory
 146 * cache, ie the parts that aren't tracked by GIT, and only used
 147 * to validate the cache.
 148 */
 149void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 150{
 151        fill_stat_data(&ce->ce_stat_data, st);
 152
 153        if (assume_unchanged)
 154                ce->ce_flags |= CE_VALID;
 155
 156        if (S_ISREG(st->st_mode))
 157                ce_mark_uptodate(ce);
 158}
 159
 160static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
 161{
 162        int match = -1;
 163        int fd = open(ce->name, O_RDONLY);
 164
 165        if (fd >= 0) {
 166                unsigned char sha1[20];
 167                if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
 168                        match = hashcmp(sha1, ce->sha1);
 169                /* index_fd() closed the file descriptor already */
 170        }
 171        return match;
 172}
 173
 174static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
 175{
 176        int match = -1;
 177        void *buffer;
 178        unsigned long size;
 179        enum object_type type;
 180        struct strbuf sb = STRBUF_INIT;
 181
 182        if (strbuf_readlink(&sb, ce->name, expected_size))
 183                return -1;
 184
 185        buffer = read_sha1_file(ce->sha1, &type, &size);
 186        if (buffer) {
 187                if (size == sb.len)
 188                        match = memcmp(buffer, sb.buf, size);
 189                free(buffer);
 190        }
 191        strbuf_release(&sb);
 192        return match;
 193}
 194
 195static int ce_compare_gitlink(const struct cache_entry *ce)
 196{
 197        unsigned char sha1[20];
 198
 199        /*
 200         * We don't actually require that the .git directory
 201         * under GITLINK directory be a valid git directory. It
 202         * might even be missing (in case nobody populated that
 203         * sub-project).
 204         *
 205         * If so, we consider it always to match.
 206         */
 207        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 208                return 0;
 209        return hashcmp(sha1, ce->sha1);
 210}
 211
 212static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
 213{
 214        switch (st->st_mode & S_IFMT) {
 215        case S_IFREG:
 216                if (ce_compare_data(ce, st))
 217                        return DATA_CHANGED;
 218                break;
 219        case S_IFLNK:
 220                if (ce_compare_link(ce, xsize_t(st->st_size)))
 221                        return DATA_CHANGED;
 222                break;
 223        case S_IFDIR:
 224                if (S_ISGITLINK(ce->ce_mode))
 225                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 226        default:
 227                return TYPE_CHANGED;
 228        }
 229        return 0;
 230}
 231
 232static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
 233{
 234        unsigned int changed = 0;
 235
 236        if (ce->ce_flags & CE_REMOVE)
 237                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 238
 239        switch (ce->ce_mode & S_IFMT) {
 240        case S_IFREG:
 241                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 242                /* We consider only the owner x bit to be relevant for
 243                 * "mode changes"
 244                 */
 245                if (trust_executable_bit &&
 246                    (0100 & (ce->ce_mode ^ st->st_mode)))
 247                        changed |= MODE_CHANGED;
 248                break;
 249        case S_IFLNK:
 250                if (!S_ISLNK(st->st_mode) &&
 251                    (has_symlinks || !S_ISREG(st->st_mode)))
 252                        changed |= TYPE_CHANGED;
 253                break;
 254        case S_IFGITLINK:
 255                /* We ignore most of the st_xxx fields for gitlinks */
 256                if (!S_ISDIR(st->st_mode))
 257                        changed |= TYPE_CHANGED;
 258                else if (ce_compare_gitlink(ce))
 259                        changed |= DATA_CHANGED;
 260                return changed;
 261        default:
 262                die("internal error: ce_mode is %o", ce->ce_mode);
 263        }
 264
 265        changed |= match_stat_data(&ce->ce_stat_data, st);
 266
 267        /* Racily smudged entry? */
 268        if (!ce->ce_stat_data.sd_size) {
 269                if (!is_empty_blob_sha1(ce->sha1))
 270                        changed |= DATA_CHANGED;
 271        }
 272
 273        return changed;
 274}
 275
 276static int is_racy_stat(const struct index_state *istate,
 277                        const struct stat_data *sd)
 278{
 279        return (istate->timestamp.sec &&
 280#ifdef USE_NSEC
 281                 /* nanosecond timestamped files can also be racy! */
 282                (istate->timestamp.sec < sd->sd_mtime.sec ||
 283                 (istate->timestamp.sec == sd->sd_mtime.sec &&
 284                  istate->timestamp.nsec <= sd->sd_mtime.nsec))
 285#else
 286                istate->timestamp.sec <= sd->sd_mtime.sec
 287#endif
 288                );
 289}
 290
 291static int is_racy_timestamp(const struct index_state *istate,
 292                             const struct cache_entry *ce)
 293{
 294        return (!S_ISGITLINK(ce->ce_mode) &&
 295                is_racy_stat(istate, &ce->ce_stat_data));
 296}
 297
 298int match_stat_data_racy(const struct index_state *istate,
 299                         const struct stat_data *sd, struct stat *st)
 300{
 301        if (is_racy_stat(istate, sd))
 302                return MTIME_CHANGED;
 303        return match_stat_data(sd, st);
 304}
 305
 306int ie_match_stat(const struct index_state *istate,
 307                  const struct cache_entry *ce, struct stat *st,
 308                  unsigned int options)
 309{
 310        unsigned int changed;
 311        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 312        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 313        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 314
 315        /*
 316         * If it's marked as always valid in the index, it's
 317         * valid whatever the checked-out copy says.
 318         *
 319         * skip-worktree has the same effect with higher precedence
 320         */
 321        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 322                return 0;
 323        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 324                return 0;
 325
 326        /*
 327         * Intent-to-add entries have not been added, so the index entry
 328         * by definition never matches what is in the work tree until it
 329         * actually gets added.
 330         */
 331        if (ce->ce_flags & CE_INTENT_TO_ADD)
 332                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 333
 334        changed = ce_match_stat_basic(ce, st);
 335
 336        /*
 337         * Within 1 second of this sequence:
 338         *      echo xyzzy >file && git-update-index --add file
 339         * running this command:
 340         *      echo frotz >file
 341         * would give a falsely clean cache entry.  The mtime and
 342         * length match the cache, and other stat fields do not change.
 343         *
 344         * We could detect this at update-index time (the cache entry
 345         * being registered/updated records the same time as "now")
 346         * and delay the return from git-update-index, but that would
 347         * effectively mean we can make at most one commit per second,
 348         * which is not acceptable.  Instead, we check cache entries
 349         * whose mtime are the same as the index file timestamp more
 350         * carefully than others.
 351         */
 352        if (!changed && is_racy_timestamp(istate, ce)) {
 353                if (assume_racy_is_modified)
 354                        changed |= DATA_CHANGED;
 355                else
 356                        changed |= ce_modified_check_fs(ce, st);
 357        }
 358
 359        return changed;
 360}
 361
 362int ie_modified(const struct index_state *istate,
 363                const struct cache_entry *ce,
 364                struct stat *st, unsigned int options)
 365{
 366        int changed, changed_fs;
 367
 368        changed = ie_match_stat(istate, ce, st, options);
 369        if (!changed)
 370                return 0;
 371        /*
 372         * If the mode or type has changed, there's no point in trying
 373         * to refresh the entry - it's not going to match
 374         */
 375        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 376                return changed;
 377
 378        /*
 379         * Immediately after read-tree or update-index --cacheinfo,
 380         * the length field is zero, as we have never even read the
 381         * lstat(2) information once, and we cannot trust DATA_CHANGED
 382         * returned by ie_match_stat() which in turn was returned by
 383         * ce_match_stat_basic() to signal that the filesize of the
 384         * blob changed.  We have to actually go to the filesystem to
 385         * see if the contents match, and if so, should answer "unchanged".
 386         *
 387         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 388         * already has checked the actual HEAD from the filesystem in the
 389         * subproject.  If ie_match_stat() already said it is different,
 390         * then we know it is.
 391         */
 392        if ((changed & DATA_CHANGED) &&
 393            (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
 394                return changed;
 395
 396        changed_fs = ce_modified_check_fs(ce, st);
 397        if (changed_fs)
 398                return changed | changed_fs;
 399        return 0;
 400}
 401
 402int base_name_compare(const char *name1, int len1, int mode1,
 403                      const char *name2, int len2, int mode2)
 404{
 405        unsigned char c1, c2;
 406        int len = len1 < len2 ? len1 : len2;
 407        int cmp;
 408
 409        cmp = memcmp(name1, name2, len);
 410        if (cmp)
 411                return cmp;
 412        c1 = name1[len];
 413        c2 = name2[len];
 414        if (!c1 && S_ISDIR(mode1))
 415                c1 = '/';
 416        if (!c2 && S_ISDIR(mode2))
 417                c2 = '/';
 418        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 419}
 420
 421/*
 422 * df_name_compare() is identical to base_name_compare(), except it
 423 * compares conflicting directory/file entries as equal. Note that
 424 * while a directory name compares as equal to a regular file, they
 425 * then individually compare _differently_ to a filename that has
 426 * a dot after the basename (because '\0' < '.' < '/').
 427 *
 428 * This is used by routines that want to traverse the git namespace
 429 * but then handle conflicting entries together when possible.
 430 */
 431int df_name_compare(const char *name1, int len1, int mode1,
 432                    const char *name2, int len2, int mode2)
 433{
 434        int len = len1 < len2 ? len1 : len2, cmp;
 435        unsigned char c1, c2;
 436
 437        cmp = memcmp(name1, name2, len);
 438        if (cmp)
 439                return cmp;
 440        /* Directories and files compare equal (same length, same name) */
 441        if (len1 == len2)
 442                return 0;
 443        c1 = name1[len];
 444        if (!c1 && S_ISDIR(mode1))
 445                c1 = '/';
 446        c2 = name2[len];
 447        if (!c2 && S_ISDIR(mode2))
 448                c2 = '/';
 449        if (c1 == '/' && !c2)
 450                return 0;
 451        if (c2 == '/' && !c1)
 452                return 0;
 453        return c1 - c2;
 454}
 455
 456int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
 457{
 458        size_t min_len = (len1 < len2) ? len1 : len2;
 459        int cmp = memcmp(name1, name2, min_len);
 460        if (cmp)
 461                return cmp;
 462        if (len1 < len2)
 463                return -1;
 464        if (len1 > len2)
 465                return 1;
 466        return 0;
 467}
 468
 469int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 470{
 471        int cmp;
 472
 473        cmp = name_compare(name1, len1, name2, len2);
 474        if (cmp)
 475                return cmp;
 476
 477        if (stage1 < stage2)
 478                return -1;
 479        if (stage1 > stage2)
 480                return 1;
 481        return 0;
 482}
 483
 484static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 485{
 486        int first, last;
 487
 488        first = 0;
 489        last = istate->cache_nr;
 490        while (last > first) {
 491                int next = (last + first) >> 1;
 492                struct cache_entry *ce = istate->cache[next];
 493                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 494                if (!cmp)
 495                        return next;
 496                if (cmp < 0) {
 497                        last = next;
 498                        continue;
 499                }
 500                first = next+1;
 501        }
 502        return -first-1;
 503}
 504
 505int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 506{
 507        return index_name_stage_pos(istate, name, namelen, 0);
 508}
 509
 510/* Remove entry, return true if there are more entries to go.. */
 511int remove_index_entry_at(struct index_state *istate, int pos)
 512{
 513        struct cache_entry *ce = istate->cache[pos];
 514
 515        record_resolve_undo(istate, ce);
 516        remove_name_hash(istate, ce);
 517        save_or_free_index_entry(istate, ce);
 518        istate->cache_changed |= CE_ENTRY_REMOVED;
 519        istate->cache_nr--;
 520        if (pos >= istate->cache_nr)
 521                return 0;
 522        memmove(istate->cache + pos,
 523                istate->cache + pos + 1,
 524                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 525        return 1;
 526}
 527
 528/*
 529 * Remove all cache entries marked for removal, that is where
 530 * CE_REMOVE is set in ce_flags.  This is much more effective than
 531 * calling remove_index_entry_at() for each entry to be removed.
 532 */
 533void remove_marked_cache_entries(struct index_state *istate)
 534{
 535        struct cache_entry **ce_array = istate->cache;
 536        unsigned int i, j;
 537
 538        for (i = j = 0; i < istate->cache_nr; i++) {
 539                if (ce_array[i]->ce_flags & CE_REMOVE) {
 540                        remove_name_hash(istate, ce_array[i]);
 541                        save_or_free_index_entry(istate, ce_array[i]);
 542                }
 543                else
 544                        ce_array[j++] = ce_array[i];
 545        }
 546        if (j == istate->cache_nr)
 547                return;
 548        istate->cache_changed |= CE_ENTRY_REMOVED;
 549        istate->cache_nr = j;
 550}
 551
 552int remove_file_from_index(struct index_state *istate, const char *path)
 553{
 554        int pos = index_name_pos(istate, path, strlen(path));
 555        if (pos < 0)
 556                pos = -pos-1;
 557        cache_tree_invalidate_path(istate, path);
 558        untracked_cache_remove_from_index(istate, path);
 559        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 560                remove_index_entry_at(istate, pos);
 561        return 0;
 562}
 563
 564static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 565{
 566        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 567}
 568
 569static int index_name_pos_also_unmerged(struct index_state *istate,
 570        const char *path, int namelen)
 571{
 572        int pos = index_name_pos(istate, path, namelen);
 573        struct cache_entry *ce;
 574
 575        if (pos >= 0)
 576                return pos;
 577
 578        /* maybe unmerged? */
 579        pos = -1 - pos;
 580        if (pos >= istate->cache_nr ||
 581                        compare_name((ce = istate->cache[pos]), path, namelen))
 582                return -1;
 583
 584        /* order of preference: stage 2, 1, 3 */
 585        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 586                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 587                        !compare_name(ce, path, namelen))
 588                pos++;
 589        return pos;
 590}
 591
 592static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 593{
 594        int len = ce_namelen(ce);
 595        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 596}
 597
 598/*
 599 * If we add a filename that aliases in the cache, we will use the
 600 * name that we already have - but we don't want to update the same
 601 * alias twice, because that implies that there were actually two
 602 * different files with aliasing names!
 603 *
 604 * So we use the CE_ADDED flag to verify that the alias was an old
 605 * one before we accept it as
 606 */
 607static struct cache_entry *create_alias_ce(struct index_state *istate,
 608                                           struct cache_entry *ce,
 609                                           struct cache_entry *alias)
 610{
 611        int len;
 612        struct cache_entry *new;
 613
 614        if (alias->ce_flags & CE_ADDED)
 615                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 616
 617        /* Ok, create the new entry using the name of the existing alias */
 618        len = ce_namelen(alias);
 619        new = xcalloc(1, cache_entry_size(len));
 620        memcpy(new->name, alias->name, len);
 621        copy_cache_entry(new, ce);
 622        save_or_free_index_entry(istate, ce);
 623        return new;
 624}
 625
 626void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
 627{
 628        unsigned char sha1[20];
 629        if (write_sha1_file("", 0, blob_type, sha1))
 630                die("cannot create an empty blob in the object database");
 631        hashcpy(ce->sha1, sha1);
 632}
 633
 634int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 635{
 636        int size, namelen, was_same;
 637        mode_t st_mode = st->st_mode;
 638        struct cache_entry *ce, *alias;
 639        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 640        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 641        int pretend = flags & ADD_CACHE_PRETEND;
 642        int intent_only = flags & ADD_CACHE_INTENT;
 643        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 644                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 645
 646        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 647                return error("%s: can only add regular files, symbolic links or git-directories", path);
 648
 649        namelen = strlen(path);
 650        if (S_ISDIR(st_mode)) {
 651                while (namelen && path[namelen-1] == '/')
 652                        namelen--;
 653        }
 654        size = cache_entry_size(namelen);
 655        ce = xcalloc(1, size);
 656        memcpy(ce->name, path, namelen);
 657        ce->ce_namelen = namelen;
 658        if (!intent_only)
 659                fill_stat_cache_info(ce, st);
 660        else
 661                ce->ce_flags |= CE_INTENT_TO_ADD;
 662
 663        if (trust_executable_bit && has_symlinks)
 664                ce->ce_mode = create_ce_mode(st_mode);
 665        else {
 666                /* If there is an existing entry, pick the mode bits and type
 667                 * from it, otherwise assume unexecutable regular file.
 668                 */
 669                struct cache_entry *ent;
 670                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 671
 672                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 673                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 674        }
 675
 676        /* When core.ignorecase=true, determine if a directory of the same name but differing
 677         * case already exists within the Git repository.  If it does, ensure the directory
 678         * case of the file being added to the repository matches (is folded into) the existing
 679         * entry's directory case.
 680         */
 681        if (ignore_case) {
 682                const char *startPtr = ce->name;
 683                const char *ptr = startPtr;
 684                while (*ptr) {
 685                        while (*ptr && *ptr != '/')
 686                                ++ptr;
 687                        if (*ptr == '/') {
 688                                struct cache_entry *foundce;
 689                                ++ptr;
 690                                foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
 691                                if (foundce) {
 692                                        memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
 693                                        startPtr = ptr;
 694                                }
 695                        }
 696                }
 697        }
 698
 699        alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 700        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 701                /* Nothing changed, really */
 702                if (!S_ISGITLINK(alias->ce_mode))
 703                        ce_mark_uptodate(alias);
 704                alias->ce_flags |= CE_ADDED;
 705
 706                free(ce);
 707                return 0;
 708        }
 709        if (!intent_only) {
 710                if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT)) {
 711                        free(ce);
 712                        return error("unable to index file %s", path);
 713                }
 714        } else
 715                set_object_name_for_intent_to_add_entry(ce);
 716
 717        if (ignore_case && alias && different_name(ce, alias))
 718                ce = create_alias_ce(istate, ce, alias);
 719        ce->ce_flags |= CE_ADDED;
 720
 721        /* It was suspected to be racily clean, but it turns out to be Ok */
 722        was_same = (alias &&
 723                    !ce_stage(alias) &&
 724                    !hashcmp(alias->sha1, ce->sha1) &&
 725                    ce->ce_mode == alias->ce_mode);
 726
 727        if (pretend)
 728                free(ce);
 729        else if (add_index_entry(istate, ce, add_option)) {
 730                free(ce);
 731                return error("unable to add %s to index", path);
 732        }
 733        if (verbose && !was_same)
 734                printf("add '%s'\n", path);
 735        return 0;
 736}
 737
 738int add_file_to_index(struct index_state *istate, const char *path, int flags)
 739{
 740        struct stat st;
 741        if (lstat(path, &st))
 742                die_errno("unable to stat '%s'", path);
 743        return add_to_index(istate, path, &st, flags);
 744}
 745
 746struct cache_entry *make_cache_entry(unsigned int mode,
 747                const unsigned char *sha1, const char *path, int stage,
 748                unsigned int refresh_options)
 749{
 750        int size, len;
 751        struct cache_entry *ce, *ret;
 752
 753        if (!verify_path(path)) {
 754                error("Invalid path '%s'", path);
 755                return NULL;
 756        }
 757
 758        len = strlen(path);
 759        size = cache_entry_size(len);
 760        ce = xcalloc(1, size);
 761
 762        hashcpy(ce->sha1, sha1);
 763        memcpy(ce->name, path, len);
 764        ce->ce_flags = create_ce_flags(stage);
 765        ce->ce_namelen = len;
 766        ce->ce_mode = create_ce_mode(mode);
 767
 768        ret = refresh_cache_entry(ce, refresh_options);
 769        if (ret != ce)
 770                free(ce);
 771        return ret;
 772}
 773
 774int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
 775{
 776        int len = ce_namelen(a);
 777        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 778}
 779
 780/*
 781 * We fundamentally don't like some paths: we don't want
 782 * dot or dot-dot anywhere, and for obvious reasons don't
 783 * want to recurse into ".git" either.
 784 *
 785 * Also, we don't want double slashes or slashes at the
 786 * end that can make pathnames ambiguous.
 787 */
 788static int verify_dotfile(const char *rest)
 789{
 790        /*
 791         * The first character was '.', but that
 792         * has already been discarded, we now test
 793         * the rest.
 794         */
 795
 796        /* "." is not allowed */
 797        if (*rest == '\0' || is_dir_sep(*rest))
 798                return 0;
 799
 800        switch (*rest) {
 801        /*
 802         * ".git" followed by  NUL or slash is bad. This
 803         * shares the path end test with the ".." case.
 804         */
 805        case 'g':
 806        case 'G':
 807                if (rest[1] != 'i' && rest[1] != 'I')
 808                        break;
 809                if (rest[2] != 't' && rest[2] != 'T')
 810                        break;
 811                rest += 2;
 812        /* fallthrough */
 813        case '.':
 814                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 815                        return 0;
 816        }
 817        return 1;
 818}
 819
 820int verify_path(const char *path)
 821{
 822        char c;
 823
 824        if (has_dos_drive_prefix(path))
 825                return 0;
 826
 827        goto inside;
 828        for (;;) {
 829                if (!c)
 830                        return 1;
 831                if (is_dir_sep(c)) {
 832inside:
 833                        if (protect_hfs && is_hfs_dotgit(path))
 834                                return 0;
 835                        if (protect_ntfs && is_ntfs_dotgit(path))
 836                                return 0;
 837                        c = *path++;
 838                        if ((c == '.' && !verify_dotfile(path)) ||
 839                            is_dir_sep(c) || c == '\0')
 840                                return 0;
 841                }
 842                c = *path++;
 843        }
 844}
 845
 846/*
 847 * Do we have another file that has the beginning components being a
 848 * proper superset of the name we're trying to add?
 849 */
 850static int has_file_name(struct index_state *istate,
 851                         const struct cache_entry *ce, int pos, int ok_to_replace)
 852{
 853        int retval = 0;
 854        int len = ce_namelen(ce);
 855        int stage = ce_stage(ce);
 856        const char *name = ce->name;
 857
 858        while (pos < istate->cache_nr) {
 859                struct cache_entry *p = istate->cache[pos++];
 860
 861                if (len >= ce_namelen(p))
 862                        break;
 863                if (memcmp(name, p->name, len))
 864                        break;
 865                if (ce_stage(p) != stage)
 866                        continue;
 867                if (p->name[len] != '/')
 868                        continue;
 869                if (p->ce_flags & CE_REMOVE)
 870                        continue;
 871                retval = -1;
 872                if (!ok_to_replace)
 873                        break;
 874                remove_index_entry_at(istate, --pos);
 875        }
 876        return retval;
 877}
 878
 879/*
 880 * Do we have another file with a pathname that is a proper
 881 * subset of the name we're trying to add?
 882 */
 883static int has_dir_name(struct index_state *istate,
 884                        const struct cache_entry *ce, int pos, int ok_to_replace)
 885{
 886        int retval = 0;
 887        int stage = ce_stage(ce);
 888        const char *name = ce->name;
 889        const char *slash = name + ce_namelen(ce);
 890
 891        for (;;) {
 892                int len;
 893
 894                for (;;) {
 895                        if (*--slash == '/')
 896                                break;
 897                        if (slash <= ce->name)
 898                                return retval;
 899                }
 900                len = slash - name;
 901
 902                pos = index_name_stage_pos(istate, name, len, stage);
 903                if (pos >= 0) {
 904                        /*
 905                         * Found one, but not so fast.  This could
 906                         * be a marker that says "I was here, but
 907                         * I am being removed".  Such an entry is
 908                         * not a part of the resulting tree, and
 909                         * it is Ok to have a directory at the same
 910                         * path.
 911                         */
 912                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 913                                retval = -1;
 914                                if (!ok_to_replace)
 915                                        break;
 916                                remove_index_entry_at(istate, pos);
 917                                continue;
 918                        }
 919                }
 920                else
 921                        pos = -pos-1;
 922
 923                /*
 924                 * Trivial optimization: if we find an entry that
 925                 * already matches the sub-directory, then we know
 926                 * we're ok, and we can exit.
 927                 */
 928                while (pos < istate->cache_nr) {
 929                        struct cache_entry *p = istate->cache[pos];
 930                        if ((ce_namelen(p) <= len) ||
 931                            (p->name[len] != '/') ||
 932                            memcmp(p->name, name, len))
 933                                break; /* not our subdirectory */
 934                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 935                                /*
 936                                 * p is at the same stage as our entry, and
 937                                 * is a subdirectory of what we are looking
 938                                 * at, so we cannot have conflicts at our
 939                                 * level or anything shorter.
 940                                 */
 941                                return retval;
 942                        pos++;
 943                }
 944        }
 945        return retval;
 946}
 947
 948/* We may be in a situation where we already have path/file and path
 949 * is being added, or we already have path and path/file is being
 950 * added.  Either one would result in a nonsense tree that has path
 951 * twice when git-write-tree tries to write it out.  Prevent it.
 952 *
 953 * If ok-to-replace is specified, we remove the conflicting entries
 954 * from the cache so the caller should recompute the insert position.
 955 * When this happens, we return non-zero.
 956 */
 957static int check_file_directory_conflict(struct index_state *istate,
 958                                         const struct cache_entry *ce,
 959                                         int pos, int ok_to_replace)
 960{
 961        int retval;
 962
 963        /*
 964         * When ce is an "I am going away" entry, we allow it to be added
 965         */
 966        if (ce->ce_flags & CE_REMOVE)
 967                return 0;
 968
 969        /*
 970         * We check if the path is a sub-path of a subsequent pathname
 971         * first, since removing those will not change the position
 972         * in the array.
 973         */
 974        retval = has_file_name(istate, ce, pos, ok_to_replace);
 975
 976        /*
 977         * Then check if the path might have a clashing sub-directory
 978         * before it.
 979         */
 980        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 981}
 982
 983static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 984{
 985        int pos;
 986        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 987        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 988        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 989        int new_only = option & ADD_CACHE_NEW_ONLY;
 990
 991        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
 992                cache_tree_invalidate_path(istate, ce->name);
 993        pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 994
 995        /* existing match? Just replace it. */
 996        if (pos >= 0) {
 997                if (!new_only)
 998                        replace_index_entry(istate, pos, ce);
 999                return 0;
1000        }
1001        pos = -pos-1;
1002
1003        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1004                untracked_cache_add_to_index(istate, ce->name);
1005
1006        /*
1007         * Inserting a merged entry ("stage 0") into the index
1008         * will always replace all non-merged entries..
1009         */
1010        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1011                while (ce_same_name(istate->cache[pos], ce)) {
1012                        ok_to_add = 1;
1013                        if (!remove_index_entry_at(istate, pos))
1014                                break;
1015                }
1016        }
1017
1018        if (!ok_to_add)
1019                return -1;
1020        if (!verify_path(ce->name))
1021                return error("Invalid path '%s'", ce->name);
1022
1023        if (!skip_df_check &&
1024            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1025                if (!ok_to_replace)
1026                        return error("'%s' appears as both a file and as a directory",
1027                                     ce->name);
1028                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1029                pos = -pos-1;
1030        }
1031        return pos + 1;
1032}
1033
1034int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1035{
1036        int pos;
1037
1038        if (option & ADD_CACHE_JUST_APPEND)
1039                pos = istate->cache_nr;
1040        else {
1041                int ret;
1042                ret = add_index_entry_with_check(istate, ce, option);
1043                if (ret <= 0)
1044                        return ret;
1045                pos = ret - 1;
1046        }
1047
1048        /* Make sure the array is big enough .. */
1049        ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1050
1051        /* Add it in.. */
1052        istate->cache_nr++;
1053        if (istate->cache_nr > pos + 1)
1054                memmove(istate->cache + pos + 1,
1055                        istate->cache + pos,
1056                        (istate->cache_nr - pos - 1) * sizeof(ce));
1057        set_index_entry(istate, pos, ce);
1058        istate->cache_changed |= CE_ENTRY_ADDED;
1059        return 0;
1060}
1061
1062/*
1063 * "refresh" does not calculate a new sha1 file or bring the
1064 * cache up-to-date for mode/content changes. But what it
1065 * _does_ do is to "re-match" the stat information of a file
1066 * with the cache, so that you can refresh the cache for a
1067 * file that hasn't been changed but where the stat entry is
1068 * out of date.
1069 *
1070 * For example, you'd want to do this after doing a "git-read-tree",
1071 * to link up the stat cache details with the proper files.
1072 */
1073static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1074                                             struct cache_entry *ce,
1075                                             unsigned int options, int *err,
1076                                             int *changed_ret)
1077{
1078        struct stat st;
1079        struct cache_entry *updated;
1080        int changed, size;
1081        int refresh = options & CE_MATCH_REFRESH;
1082        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1083        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1084        int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1085
1086        if (!refresh || ce_uptodate(ce))
1087                return ce;
1088
1089        /*
1090         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1091         * that the change to the work tree does not matter and told
1092         * us not to worry.
1093         */
1094        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1095                ce_mark_uptodate(ce);
1096                return ce;
1097        }
1098        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1099                ce_mark_uptodate(ce);
1100                return ce;
1101        }
1102
1103        if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1104                if (ignore_missing)
1105                        return ce;
1106                if (err)
1107                        *err = ENOENT;
1108                return NULL;
1109        }
1110
1111        if (lstat(ce->name, &st) < 0) {
1112                if (ignore_missing && errno == ENOENT)
1113                        return ce;
1114                if (err)
1115                        *err = errno;
1116                return NULL;
1117        }
1118
1119        changed = ie_match_stat(istate, ce, &st, options);
1120        if (changed_ret)
1121                *changed_ret = changed;
1122        if (!changed) {
1123                /*
1124                 * The path is unchanged.  If we were told to ignore
1125                 * valid bit, then we did the actual stat check and
1126                 * found that the entry is unmodified.  If the entry
1127                 * is not marked VALID, this is the place to mark it
1128                 * valid again, under "assume unchanged" mode.
1129                 */
1130                if (ignore_valid && assume_unchanged &&
1131                    !(ce->ce_flags & CE_VALID))
1132                        ; /* mark this one VALID again */
1133                else {
1134                        /*
1135                         * We do not mark the index itself "modified"
1136                         * because CE_UPTODATE flag is in-core only;
1137                         * we are not going to write this change out.
1138                         */
1139                        if (!S_ISGITLINK(ce->ce_mode))
1140                                ce_mark_uptodate(ce);
1141                        return ce;
1142                }
1143        }
1144
1145        if (ie_modified(istate, ce, &st, options)) {
1146                if (err)
1147                        *err = EINVAL;
1148                return NULL;
1149        }
1150
1151        size = ce_size(ce);
1152        updated = xmalloc(size);
1153        memcpy(updated, ce, size);
1154        fill_stat_cache_info(updated, &st);
1155        /*
1156         * If ignore_valid is not set, we should leave CE_VALID bit
1157         * alone.  Otherwise, paths marked with --no-assume-unchanged
1158         * (i.e. things to be edited) will reacquire CE_VALID bit
1159         * automatically, which is not really what we want.
1160         */
1161        if (!ignore_valid && assume_unchanged &&
1162            !(ce->ce_flags & CE_VALID))
1163                updated->ce_flags &= ~CE_VALID;
1164
1165        /* istate->cache_changed is updated in the caller */
1166        return updated;
1167}
1168
1169static void show_file(const char * fmt, const char * name, int in_porcelain,
1170                      int * first, const char *header_msg)
1171{
1172        if (in_porcelain && *first && header_msg) {
1173                printf("%s\n", header_msg);
1174                *first = 0;
1175        }
1176        printf(fmt, name);
1177}
1178
1179int refresh_index(struct index_state *istate, unsigned int flags,
1180                  const struct pathspec *pathspec,
1181                  char *seen, const char *header_msg)
1182{
1183        int i;
1184        int has_errors = 0;
1185        int really = (flags & REFRESH_REALLY) != 0;
1186        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1187        int quiet = (flags & REFRESH_QUIET) != 0;
1188        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1189        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1190        int first = 1;
1191        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1192        unsigned int options = (CE_MATCH_REFRESH |
1193                                (really ? CE_MATCH_IGNORE_VALID : 0) |
1194                                (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1195        const char *modified_fmt;
1196        const char *deleted_fmt;
1197        const char *typechange_fmt;
1198        const char *added_fmt;
1199        const char *unmerged_fmt;
1200
1201        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1202        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1203        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1204        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1205        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1206        for (i = 0; i < istate->cache_nr; i++) {
1207                struct cache_entry *ce, *new;
1208                int cache_errno = 0;
1209                int changed = 0;
1210                int filtered = 0;
1211
1212                ce = istate->cache[i];
1213                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1214                        continue;
1215
1216                if (pathspec && !ce_path_match(ce, pathspec, seen))
1217                        filtered = 1;
1218
1219                if (ce_stage(ce)) {
1220                        while ((i < istate->cache_nr) &&
1221                               ! strcmp(istate->cache[i]->name, ce->name))
1222                                i++;
1223                        i--;
1224                        if (allow_unmerged)
1225                                continue;
1226                        if (!filtered)
1227                                show_file(unmerged_fmt, ce->name, in_porcelain,
1228                                          &first, header_msg);
1229                        has_errors = 1;
1230                        continue;
1231                }
1232
1233                if (filtered)
1234                        continue;
1235
1236                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1237                if (new == ce)
1238                        continue;
1239                if (!new) {
1240                        const char *fmt;
1241
1242                        if (really && cache_errno == EINVAL) {
1243                                /* If we are doing --really-refresh that
1244                                 * means the index is not valid anymore.
1245                                 */
1246                                ce->ce_flags &= ~CE_VALID;
1247                                ce->ce_flags |= CE_UPDATE_IN_BASE;
1248                                istate->cache_changed |= CE_ENTRY_CHANGED;
1249                        }
1250                        if (quiet)
1251                                continue;
1252
1253                        if (cache_errno == ENOENT)
1254                                fmt = deleted_fmt;
1255                        else if (ce->ce_flags & CE_INTENT_TO_ADD)
1256                                fmt = added_fmt; /* must be before other checks */
1257                        else if (changed & TYPE_CHANGED)
1258                                fmt = typechange_fmt;
1259                        else
1260                                fmt = modified_fmt;
1261                        show_file(fmt,
1262                                  ce->name, in_porcelain, &first, header_msg);
1263                        has_errors = 1;
1264                        continue;
1265                }
1266
1267                replace_index_entry(istate, i, new);
1268        }
1269        return has_errors;
1270}
1271
1272static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1273                                               unsigned int options)
1274{
1275        return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1276}
1277
1278
1279/*****************************************************************
1280 * Index File I/O
1281 *****************************************************************/
1282
1283#define INDEX_FORMAT_DEFAULT 3
1284
1285static unsigned int get_index_format_default(void)
1286{
1287        char *envversion = getenv("GIT_INDEX_VERSION");
1288        char *endp;
1289        int value;
1290        unsigned int version = INDEX_FORMAT_DEFAULT;
1291
1292        if (!envversion) {
1293                if (!git_config_get_int("index.version", &value))
1294                        version = value;
1295                if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1296                        warning(_("index.version set, but the value is invalid.\n"
1297                                  "Using version %i"), INDEX_FORMAT_DEFAULT);
1298                        return INDEX_FORMAT_DEFAULT;
1299                }
1300                return version;
1301        }
1302
1303        version = strtoul(envversion, &endp, 10);
1304        if (*endp ||
1305            version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1306                warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1307                          "Using version %i"), INDEX_FORMAT_DEFAULT);
1308                version = INDEX_FORMAT_DEFAULT;
1309        }
1310        return version;
1311}
1312
1313/*
1314 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1315 * Again - this is just a (very strong in practice) heuristic that
1316 * the inode hasn't changed.
1317 *
1318 * We save the fields in big-endian order to allow using the
1319 * index file over NFS transparently.
1320 */
1321struct ondisk_cache_entry {
1322        struct cache_time ctime;
1323        struct cache_time mtime;
1324        uint32_t dev;
1325        uint32_t ino;
1326        uint32_t mode;
1327        uint32_t uid;
1328        uint32_t gid;
1329        uint32_t size;
1330        unsigned char sha1[20];
1331        uint16_t flags;
1332        char name[FLEX_ARRAY]; /* more */
1333};
1334
1335/*
1336 * This struct is used when CE_EXTENDED bit is 1
1337 * The struct must match ondisk_cache_entry exactly from
1338 * ctime till flags
1339 */
1340struct ondisk_cache_entry_extended {
1341        struct cache_time ctime;
1342        struct cache_time mtime;
1343        uint32_t dev;
1344        uint32_t ino;
1345        uint32_t mode;
1346        uint32_t uid;
1347        uint32_t gid;
1348        uint32_t size;
1349        unsigned char sha1[20];
1350        uint16_t flags;
1351        uint16_t flags2;
1352        char name[FLEX_ARRAY]; /* more */
1353};
1354
1355/* These are only used for v3 or lower */
1356#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1357#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1358#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1359#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1360                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1361                            ondisk_cache_entry_size(ce_namelen(ce)))
1362
1363static int verify_hdr(struct cache_header *hdr, unsigned long size)
1364{
1365        git_SHA_CTX c;
1366        unsigned char sha1[20];
1367        int hdr_version;
1368
1369        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1370                return error("bad signature");
1371        hdr_version = ntohl(hdr->hdr_version);
1372        if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1373                return error("bad index version %d", hdr_version);
1374        git_SHA1_Init(&c);
1375        git_SHA1_Update(&c, hdr, size - 20);
1376        git_SHA1_Final(sha1, &c);
1377        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1378                return error("bad index file sha1 signature");
1379        return 0;
1380}
1381
1382static int read_index_extension(struct index_state *istate,
1383                                const char *ext, void *data, unsigned long sz)
1384{
1385        switch (CACHE_EXT(ext)) {
1386        case CACHE_EXT_TREE:
1387                istate->cache_tree = cache_tree_read(data, sz);
1388                break;
1389        case CACHE_EXT_RESOLVE_UNDO:
1390                istate->resolve_undo = resolve_undo_read(data, sz);
1391                break;
1392        case CACHE_EXT_LINK:
1393                if (read_link_extension(istate, data, sz))
1394                        return -1;
1395                break;
1396        case CACHE_EXT_UNTRACKED:
1397                istate->untracked = read_untracked_extension(data, sz);
1398                break;
1399        default:
1400                if (*ext < 'A' || 'Z' < *ext)
1401                        return error("index uses %.4s extension, which we do not understand",
1402                                     ext);
1403                fprintf(stderr, "ignoring %.4s extension\n", ext);
1404                break;
1405        }
1406        return 0;
1407}
1408
1409int hold_locked_index(struct lock_file *lk, int die_on_error)
1410{
1411        return hold_lock_file_for_update(lk, get_index_file(),
1412                                         die_on_error
1413                                         ? LOCK_DIE_ON_ERROR
1414                                         : 0);
1415}
1416
1417int read_index(struct index_state *istate)
1418{
1419        return read_index_from(istate, get_index_file());
1420}
1421
1422static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1423                                                   unsigned int flags,
1424                                                   const char *name,
1425                                                   size_t len)
1426{
1427        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1428
1429        ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1430        ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1431        ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1432        ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1433        ce->ce_stat_data.sd_dev   = get_be32(&ondisk->dev);
1434        ce->ce_stat_data.sd_ino   = get_be32(&ondisk->ino);
1435        ce->ce_mode  = get_be32(&ondisk->mode);
1436        ce->ce_stat_data.sd_uid   = get_be32(&ondisk->uid);
1437        ce->ce_stat_data.sd_gid   = get_be32(&ondisk->gid);
1438        ce->ce_stat_data.sd_size  = get_be32(&ondisk->size);
1439        ce->ce_flags = flags & ~CE_NAMEMASK;
1440        ce->ce_namelen = len;
1441        ce->index = 0;
1442        hashcpy(ce->sha1, ondisk->sha1);
1443        memcpy(ce->name, name, len);
1444        ce->name[len] = '\0';
1445        return ce;
1446}
1447
1448/*
1449 * Adjacent cache entries tend to share the leading paths, so it makes
1450 * sense to only store the differences in later entries.  In the v4
1451 * on-disk format of the index, each on-disk cache entry stores the
1452 * number of bytes to be stripped from the end of the previous name,
1453 * and the bytes to append to the result, to come up with its name.
1454 */
1455static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1456{
1457        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1458        size_t len = decode_varint(&cp);
1459
1460        if (name->len < len)
1461                die("malformed name field in the index");
1462        strbuf_remove(name, name->len - len, len);
1463        for (ep = cp; *ep; ep++)
1464                ; /* find the end */
1465        strbuf_add(name, cp, ep - cp);
1466        return (const char *)ep + 1 - cp_;
1467}
1468
1469static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1470                                            unsigned long *ent_size,
1471                                            struct strbuf *previous_name)
1472{
1473        struct cache_entry *ce;
1474        size_t len;
1475        const char *name;
1476        unsigned int flags;
1477
1478        /* On-disk flags are just 16 bits */
1479        flags = get_be16(&ondisk->flags);
1480        len = flags & CE_NAMEMASK;
1481
1482        if (flags & CE_EXTENDED) {
1483                struct ondisk_cache_entry_extended *ondisk2;
1484                int extended_flags;
1485                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1486                extended_flags = get_be16(&ondisk2->flags2) << 16;
1487                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1488                if (extended_flags & ~CE_EXTENDED_FLAGS)
1489                        die("Unknown index entry format %08x", extended_flags);
1490                flags |= extended_flags;
1491                name = ondisk2->name;
1492        }
1493        else
1494                name = ondisk->name;
1495
1496        if (!previous_name) {
1497                /* v3 and earlier */
1498                if (len == CE_NAMEMASK)
1499                        len = strlen(name);
1500                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1501
1502                *ent_size = ondisk_ce_size(ce);
1503        } else {
1504                unsigned long consumed;
1505                consumed = expand_name_field(previous_name, name);
1506                ce = cache_entry_from_ondisk(ondisk, flags,
1507                                             previous_name->buf,
1508                                             previous_name->len);
1509
1510                *ent_size = (name - ((char *)ondisk)) + consumed;
1511        }
1512        return ce;
1513}
1514
1515static void check_ce_order(struct index_state *istate)
1516{
1517        unsigned int i;
1518
1519        for (i = 1; i < istate->cache_nr; i++) {
1520                struct cache_entry *ce = istate->cache[i - 1];
1521                struct cache_entry *next_ce = istate->cache[i];
1522                int name_compare = strcmp(ce->name, next_ce->name);
1523
1524                if (0 < name_compare)
1525                        die("unordered stage entries in index");
1526                if (!name_compare) {
1527                        if (!ce_stage(ce))
1528                                die("multiple stage entries for merged file '%s'",
1529                                    ce->name);
1530                        if (ce_stage(ce) > ce_stage(next_ce))
1531                                die("unordered stage entries for '%s'",
1532                                    ce->name);
1533                }
1534        }
1535}
1536
1537/* remember to discard_cache() before reading a different cache! */
1538int do_read_index(struct index_state *istate, const char *path, int must_exist)
1539{
1540        int fd, i;
1541        struct stat st;
1542        unsigned long src_offset;
1543        struct cache_header *hdr;
1544        void *mmap;
1545        size_t mmap_size;
1546        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1547
1548        if (istate->initialized)
1549                return istate->cache_nr;
1550
1551        istate->timestamp.sec = 0;
1552        istate->timestamp.nsec = 0;
1553        fd = open(path, O_RDONLY);
1554        if (fd < 0) {
1555                if (!must_exist && errno == ENOENT)
1556                        return 0;
1557                die_errno("%s: index file open failed", path);
1558        }
1559
1560        if (fstat(fd, &st))
1561                die_errno("cannot stat the open index");
1562
1563        mmap_size = xsize_t(st.st_size);
1564        if (mmap_size < sizeof(struct cache_header) + 20)
1565                die("index file smaller than expected");
1566
1567        mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1568        if (mmap == MAP_FAILED)
1569                die_errno("unable to map index file");
1570        close(fd);
1571
1572        hdr = mmap;
1573        if (verify_hdr(hdr, mmap_size) < 0)
1574                goto unmap;
1575
1576        hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1577        istate->version = ntohl(hdr->hdr_version);
1578        istate->cache_nr = ntohl(hdr->hdr_entries);
1579        istate->cache_alloc = alloc_nr(istate->cache_nr);
1580        istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1581        istate->initialized = 1;
1582
1583        if (istate->version == 4)
1584                previous_name = &previous_name_buf;
1585        else
1586                previous_name = NULL;
1587
1588        src_offset = sizeof(*hdr);
1589        for (i = 0; i < istate->cache_nr; i++) {
1590                struct ondisk_cache_entry *disk_ce;
1591                struct cache_entry *ce;
1592                unsigned long consumed;
1593
1594                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1595                ce = create_from_disk(disk_ce, &consumed, previous_name);
1596                set_index_entry(istate, i, ce);
1597
1598                src_offset += consumed;
1599        }
1600        strbuf_release(&previous_name_buf);
1601        istate->timestamp.sec = st.st_mtime;
1602        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1603
1604        while (src_offset <= mmap_size - 20 - 8) {
1605                /* After an array of active_nr index entries,
1606                 * there can be arbitrary number of extended
1607                 * sections, each of which is prefixed with
1608                 * extension name (4-byte) and section length
1609                 * in 4-byte network byte order.
1610                 */
1611                uint32_t extsize;
1612                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1613                extsize = ntohl(extsize);
1614                if (read_index_extension(istate,
1615                                         (const char *) mmap + src_offset,
1616                                         (char *) mmap + src_offset + 8,
1617                                         extsize) < 0)
1618                        goto unmap;
1619                src_offset += 8;
1620                src_offset += extsize;
1621        }
1622        munmap(mmap, mmap_size);
1623        return istate->cache_nr;
1624
1625unmap:
1626        munmap(mmap, mmap_size);
1627        die("index file corrupt");
1628}
1629
1630int read_index_from(struct index_state *istate, const char *path)
1631{
1632        struct split_index *split_index;
1633        int ret;
1634
1635        /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1636        if (istate->initialized)
1637                return istate->cache_nr;
1638
1639        ret = do_read_index(istate, path, 0);
1640        split_index = istate->split_index;
1641        if (!split_index || is_null_sha1(split_index->base_sha1)) {
1642                check_ce_order(istate);
1643                return ret;
1644        }
1645
1646        if (split_index->base)
1647                discard_index(split_index->base);
1648        else
1649                split_index->base = xcalloc(1, sizeof(*split_index->base));
1650        ret = do_read_index(split_index->base,
1651                            git_path("sharedindex.%s",
1652                                     sha1_to_hex(split_index->base_sha1)), 1);
1653        if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1654                die("broken index, expect %s in %s, got %s",
1655                    sha1_to_hex(split_index->base_sha1),
1656                    git_path("sharedindex.%s",
1657                             sha1_to_hex(split_index->base_sha1)),
1658                    sha1_to_hex(split_index->base->sha1));
1659        merge_base_index(istate);
1660        check_ce_order(istate);
1661        return ret;
1662}
1663
1664int is_index_unborn(struct index_state *istate)
1665{
1666        return (!istate->cache_nr && !istate->timestamp.sec);
1667}
1668
1669int discard_index(struct index_state *istate)
1670{
1671        int i;
1672
1673        for (i = 0; i < istate->cache_nr; i++) {
1674                if (istate->cache[i]->index &&
1675                    istate->split_index &&
1676                    istate->split_index->base &&
1677                    istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1678                    istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1679                        continue;
1680                free(istate->cache[i]);
1681        }
1682        resolve_undo_clear_index(istate);
1683        istate->cache_nr = 0;
1684        istate->cache_changed = 0;
1685        istate->timestamp.sec = 0;
1686        istate->timestamp.nsec = 0;
1687        free_name_hash(istate);
1688        cache_tree_free(&(istate->cache_tree));
1689        istate->initialized = 0;
1690        free(istate->cache);
1691        istate->cache = NULL;
1692        istate->cache_alloc = 0;
1693        discard_split_index(istate);
1694        free_untracked_cache(istate->untracked);
1695        istate->untracked = NULL;
1696        return 0;
1697}
1698
1699int unmerged_index(const struct index_state *istate)
1700{
1701        int i;
1702        for (i = 0; i < istate->cache_nr; i++) {
1703                if (ce_stage(istate->cache[i]))
1704                        return 1;
1705        }
1706        return 0;
1707}
1708
1709#define WRITE_BUFFER_SIZE 8192
1710static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1711static unsigned long write_buffer_len;
1712
1713static int ce_write_flush(git_SHA_CTX *context, int fd)
1714{
1715        unsigned int buffered = write_buffer_len;
1716        if (buffered) {
1717                git_SHA1_Update(context, write_buffer, buffered);
1718                if (write_in_full(fd, write_buffer, buffered) != buffered)
1719                        return -1;
1720                write_buffer_len = 0;
1721        }
1722        return 0;
1723}
1724
1725static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1726{
1727        while (len) {
1728                unsigned int buffered = write_buffer_len;
1729                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1730                if (partial > len)
1731                        partial = len;
1732                memcpy(write_buffer + buffered, data, partial);
1733                buffered += partial;
1734                if (buffered == WRITE_BUFFER_SIZE) {
1735                        write_buffer_len = buffered;
1736                        if (ce_write_flush(context, fd))
1737                                return -1;
1738                        buffered = 0;
1739                }
1740                write_buffer_len = buffered;
1741                len -= partial;
1742                data = (char *) data + partial;
1743        }
1744        return 0;
1745}
1746
1747static int write_index_ext_header(git_SHA_CTX *context, int fd,
1748                                  unsigned int ext, unsigned int sz)
1749{
1750        ext = htonl(ext);
1751        sz = htonl(sz);
1752        return ((ce_write(context, fd, &ext, 4) < 0) ||
1753                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1754}
1755
1756static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1757{
1758        unsigned int left = write_buffer_len;
1759
1760        if (left) {
1761                write_buffer_len = 0;
1762                git_SHA1_Update(context, write_buffer, left);
1763        }
1764
1765        /* Flush first if not enough space for SHA1 signature */
1766        if (left + 20 > WRITE_BUFFER_SIZE) {
1767                if (write_in_full(fd, write_buffer, left) != left)
1768                        return -1;
1769                left = 0;
1770        }
1771
1772        /* Append the SHA1 signature at the end */
1773        git_SHA1_Final(write_buffer + left, context);
1774        hashcpy(sha1, write_buffer + left);
1775        left += 20;
1776        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1777}
1778
1779static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1780{
1781        /*
1782         * The only thing we care about in this function is to smudge the
1783         * falsely clean entry due to touch-update-touch race, so we leave
1784         * everything else as they are.  We are called for entries whose
1785         * ce_stat_data.sd_mtime match the index file mtime.
1786         *
1787         * Note that this actually does not do much for gitlinks, for
1788         * which ce_match_stat_basic() always goes to the actual
1789         * contents.  The caller checks with is_racy_timestamp() which
1790         * always says "no" for gitlinks, so we are not called for them ;-)
1791         */
1792        struct stat st;
1793
1794        if (lstat(ce->name, &st) < 0)
1795                return;
1796        if (ce_match_stat_basic(ce, &st))
1797                return;
1798        if (ce_modified_check_fs(ce, &st)) {
1799                /* This is "racily clean"; smudge it.  Note that this
1800                 * is a tricky code.  At first glance, it may appear
1801                 * that it can break with this sequence:
1802                 *
1803                 * $ echo xyzzy >frotz
1804                 * $ git-update-index --add frotz
1805                 * $ : >frotz
1806                 * $ sleep 3
1807                 * $ echo filfre >nitfol
1808                 * $ git-update-index --add nitfol
1809                 *
1810                 * but it does not.  When the second update-index runs,
1811                 * it notices that the entry "frotz" has the same timestamp
1812                 * as index, and if we were to smudge it by resetting its
1813                 * size to zero here, then the object name recorded
1814                 * in index is the 6-byte file but the cached stat information
1815                 * becomes zero --- which would then match what we would
1816                 * obtain from the filesystem next time we stat("frotz").
1817                 *
1818                 * However, the second update-index, before calling
1819                 * this function, notices that the cached size is 6
1820                 * bytes and what is on the filesystem is an empty
1821                 * file, and never calls us, so the cached size information
1822                 * for "frotz" stays 6 which does not match the filesystem.
1823                 */
1824                ce->ce_stat_data.sd_size = 0;
1825        }
1826}
1827
1828/* Copy miscellaneous fields but not the name */
1829static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1830                                       struct cache_entry *ce)
1831{
1832        short flags;
1833
1834        ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1835        ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1836        ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1837        ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1838        ondisk->dev  = htonl(ce->ce_stat_data.sd_dev);
1839        ondisk->ino  = htonl(ce->ce_stat_data.sd_ino);
1840        ondisk->mode = htonl(ce->ce_mode);
1841        ondisk->uid  = htonl(ce->ce_stat_data.sd_uid);
1842        ondisk->gid  = htonl(ce->ce_stat_data.sd_gid);
1843        ondisk->size = htonl(ce->ce_stat_data.sd_size);
1844        hashcpy(ondisk->sha1, ce->sha1);
1845
1846        flags = ce->ce_flags & ~CE_NAMEMASK;
1847        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1848        ondisk->flags = htons(flags);
1849        if (ce->ce_flags & CE_EXTENDED) {
1850                struct ondisk_cache_entry_extended *ondisk2;
1851                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1852                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1853                return ondisk2->name;
1854        }
1855        else {
1856                return ondisk->name;
1857        }
1858}
1859
1860static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1861                          struct strbuf *previous_name)
1862{
1863        int size;
1864        struct ondisk_cache_entry *ondisk;
1865        int saved_namelen = saved_namelen; /* compiler workaround */
1866        char *name;
1867        int result;
1868
1869        if (ce->ce_flags & CE_STRIP_NAME) {
1870                saved_namelen = ce_namelen(ce);
1871                ce->ce_namelen = 0;
1872        }
1873
1874        if (!previous_name) {
1875                size = ondisk_ce_size(ce);
1876                ondisk = xcalloc(1, size);
1877                name = copy_cache_entry_to_ondisk(ondisk, ce);
1878                memcpy(name, ce->name, ce_namelen(ce));
1879        } else {
1880                int common, to_remove, prefix_size;
1881                unsigned char to_remove_vi[16];
1882                for (common = 0;
1883                     (ce->name[common] &&
1884                      common < previous_name->len &&
1885                      ce->name[common] == previous_name->buf[common]);
1886                     common++)
1887                        ; /* still matching */
1888                to_remove = previous_name->len - common;
1889                prefix_size = encode_varint(to_remove, to_remove_vi);
1890
1891                if (ce->ce_flags & CE_EXTENDED)
1892                        size = offsetof(struct ondisk_cache_entry_extended, name);
1893                else
1894                        size = offsetof(struct ondisk_cache_entry, name);
1895                size += prefix_size + (ce_namelen(ce) - common + 1);
1896
1897                ondisk = xcalloc(1, size);
1898                name = copy_cache_entry_to_ondisk(ondisk, ce);
1899                memcpy(name, to_remove_vi, prefix_size);
1900                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1901
1902                strbuf_splice(previous_name, common, to_remove,
1903                              ce->name + common, ce_namelen(ce) - common);
1904        }
1905        if (ce->ce_flags & CE_STRIP_NAME) {
1906                ce->ce_namelen = saved_namelen;
1907                ce->ce_flags &= ~CE_STRIP_NAME;
1908        }
1909
1910        result = ce_write(c, fd, ondisk, size);
1911        free(ondisk);
1912        return result;
1913}
1914
1915/*
1916 * This function verifies if index_state has the correct sha1 of the
1917 * index file.  Don't die if we have any other failure, just return 0.
1918 */
1919static int verify_index_from(const struct index_state *istate, const char *path)
1920{
1921        int fd;
1922        ssize_t n;
1923        struct stat st;
1924        unsigned char sha1[20];
1925
1926        if (!istate->initialized)
1927                return 0;
1928
1929        fd = open(path, O_RDONLY);
1930        if (fd < 0)
1931                return 0;
1932
1933        if (fstat(fd, &st))
1934                goto out;
1935
1936        if (st.st_size < sizeof(struct cache_header) + 20)
1937                goto out;
1938
1939        n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1940        if (n != 20)
1941                goto out;
1942
1943        if (hashcmp(istate->sha1, sha1))
1944                goto out;
1945
1946        close(fd);
1947        return 1;
1948
1949out:
1950        close(fd);
1951        return 0;
1952}
1953
1954static int verify_index(const struct index_state *istate)
1955{
1956        return verify_index_from(istate, get_index_file());
1957}
1958
1959static int has_racy_timestamp(struct index_state *istate)
1960{
1961        int entries = istate->cache_nr;
1962        int i;
1963
1964        for (i = 0; i < entries; i++) {
1965                struct cache_entry *ce = istate->cache[i];
1966                if (is_racy_timestamp(istate, ce))
1967                        return 1;
1968        }
1969        return 0;
1970}
1971
1972/*
1973 * Opportunistically update the index but do not complain if we can't
1974 */
1975void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1976{
1977        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1978            verify_index(istate) &&
1979            write_locked_index(istate, lockfile, COMMIT_LOCK))
1980                rollback_lock_file(lockfile);
1981}
1982
1983static int do_write_index(struct index_state *istate, int newfd,
1984                          int strip_extensions)
1985{
1986        git_SHA_CTX c;
1987        struct cache_header hdr;
1988        int i, err, removed, extended, hdr_version;
1989        struct cache_entry **cache = istate->cache;
1990        int entries = istate->cache_nr;
1991        struct stat st;
1992        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1993
1994        for (i = removed = extended = 0; i < entries; i++) {
1995                if (cache[i]->ce_flags & CE_REMOVE)
1996                        removed++;
1997
1998                /* reduce extended entries if possible */
1999                cache[i]->ce_flags &= ~CE_EXTENDED;
2000                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2001                        extended++;
2002                        cache[i]->ce_flags |= CE_EXTENDED;
2003                }
2004        }
2005
2006        if (!istate->version) {
2007                istate->version = get_index_format_default();
2008                if (getenv("GIT_TEST_SPLIT_INDEX"))
2009                        init_split_index(istate);
2010        }
2011
2012        /* demote version 3 to version 2 when the latter suffices */
2013        if (istate->version == 3 || istate->version == 2)
2014                istate->version = extended ? 3 : 2;
2015
2016        hdr_version = istate->version;
2017
2018        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2019        hdr.hdr_version = htonl(hdr_version);
2020        hdr.hdr_entries = htonl(entries - removed);
2021
2022        git_SHA1_Init(&c);
2023        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2024                return -1;
2025
2026        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2027        for (i = 0; i < entries; i++) {
2028                struct cache_entry *ce = cache[i];
2029                if (ce->ce_flags & CE_REMOVE)
2030                        continue;
2031                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2032                        ce_smudge_racily_clean_entry(ce);
2033                if (is_null_sha1(ce->sha1)) {
2034                        static const char msg[] = "cache entry has null sha1: %s";
2035                        static int allow = -1;
2036
2037                        if (allow < 0)
2038                                allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2039                        if (allow)
2040                                warning(msg, ce->name);
2041                        else
2042                                return error(msg, ce->name);
2043                }
2044                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2045                        return -1;
2046        }
2047        strbuf_release(&previous_name_buf);
2048
2049        /* Write extension data here */
2050        if (!strip_extensions && istate->split_index) {
2051                struct strbuf sb = STRBUF_INIT;
2052
2053                err = write_link_extension(&sb, istate) < 0 ||
2054                        write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2055                                               sb.len) < 0 ||
2056                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2057                strbuf_release(&sb);
2058                if (err)
2059                        return -1;
2060        }
2061        if (!strip_extensions && istate->cache_tree) {
2062                struct strbuf sb = STRBUF_INIT;
2063
2064                cache_tree_write(&sb, istate->cache_tree);
2065                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2066                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2067                strbuf_release(&sb);
2068                if (err)
2069                        return -1;
2070        }
2071        if (!strip_extensions && istate->resolve_undo) {
2072                struct strbuf sb = STRBUF_INIT;
2073
2074                resolve_undo_write(&sb, istate->resolve_undo);
2075                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2076                                             sb.len) < 0
2077                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2078                strbuf_release(&sb);
2079                if (err)
2080                        return -1;
2081        }
2082        if (!strip_extensions && istate->untracked) {
2083                struct strbuf sb = STRBUF_INIT;
2084
2085                write_untracked_extension(&sb, istate->untracked);
2086                err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2087                                             sb.len) < 0 ||
2088                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2089                strbuf_release(&sb);
2090                if (err)
2091                        return -1;
2092        }
2093
2094        if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2095                return -1;
2096        istate->timestamp.sec = (unsigned int)st.st_mtime;
2097        istate->timestamp.nsec = ST_MTIME_NSEC(st);
2098        return 0;
2099}
2100
2101void set_alternate_index_output(const char *name)
2102{
2103        alternate_index_output = name;
2104}
2105
2106static int commit_locked_index(struct lock_file *lk)
2107{
2108        if (alternate_index_output)
2109                return commit_lock_file_to(lk, alternate_index_output);
2110        else
2111                return commit_lock_file(lk);
2112}
2113
2114static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2115                                 unsigned flags)
2116{
2117        int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
2118        if (ret)
2119                return ret;
2120        assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2121               (COMMIT_LOCK | CLOSE_LOCK));
2122        if (flags & COMMIT_LOCK)
2123                return commit_locked_index(lock);
2124        else if (flags & CLOSE_LOCK)
2125                return close_lock_file(lock);
2126        else
2127                return ret;
2128}
2129
2130static int write_split_index(struct index_state *istate,
2131                             struct lock_file *lock,
2132                             unsigned flags)
2133{
2134        int ret;
2135        prepare_to_write_split_index(istate);
2136        ret = do_write_locked_index(istate, lock, flags);
2137        finish_writing_split_index(istate);
2138        return ret;
2139}
2140
2141static struct tempfile temporary_sharedindex;
2142
2143static int write_shared_index(struct index_state *istate,
2144                              struct lock_file *lock, unsigned flags)
2145{
2146        struct split_index *si = istate->split_index;
2147        int fd, ret;
2148
2149        fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2150        if (fd < 0) {
2151                hashclr(si->base_sha1);
2152                return do_write_locked_index(istate, lock, flags);
2153        }
2154        move_cache_to_base_index(istate);
2155        ret = do_write_index(si->base, fd, 1);
2156        if (ret) {
2157                delete_tempfile(&temporary_sharedindex);
2158                return ret;
2159        }
2160        ret = rename_tempfile(&temporary_sharedindex,
2161                              git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2162        if (!ret)
2163                hashcpy(si->base_sha1, si->base->sha1);
2164        return ret;
2165}
2166
2167int write_locked_index(struct index_state *istate, struct lock_file *lock,
2168                       unsigned flags)
2169{
2170        struct split_index *si = istate->split_index;
2171
2172        if (!si || alternate_index_output ||
2173            (istate->cache_changed & ~EXTMASK)) {
2174                if (si)
2175                        hashclr(si->base_sha1);
2176                return do_write_locked_index(istate, lock, flags);
2177        }
2178
2179        if (getenv("GIT_TEST_SPLIT_INDEX")) {
2180                int v = si->base_sha1[0];
2181                if ((v & 15) < 6)
2182                        istate->cache_changed |= SPLIT_INDEX_ORDERED;
2183        }
2184        if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2185                int ret = write_shared_index(istate, lock, flags);
2186                if (ret)
2187                        return ret;
2188        }
2189
2190        return write_split_index(istate, lock, flags);
2191}
2192
2193/*
2194 * Read the index file that is potentially unmerged into given
2195 * index_state, dropping any unmerged entries.  Returns true if
2196 * the index is unmerged.  Callers who want to refuse to work
2197 * from an unmerged state can call this and check its return value,
2198 * instead of calling read_cache().
2199 */
2200int read_index_unmerged(struct index_state *istate)
2201{
2202        int i;
2203        int unmerged = 0;
2204
2205        read_index(istate);
2206        for (i = 0; i < istate->cache_nr; i++) {
2207                struct cache_entry *ce = istate->cache[i];
2208                struct cache_entry *new_ce;
2209                int size, len;
2210
2211                if (!ce_stage(ce))
2212                        continue;
2213                unmerged = 1;
2214                len = ce_namelen(ce);
2215                size = cache_entry_size(len);
2216                new_ce = xcalloc(1, size);
2217                memcpy(new_ce->name, ce->name, len);
2218                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2219                new_ce->ce_namelen = len;
2220                new_ce->ce_mode = ce->ce_mode;
2221                if (add_index_entry(istate, new_ce, 0))
2222                        return error("%s: cannot drop to stage #0",
2223                                     new_ce->name);
2224        }
2225        return unmerged;
2226}
2227
2228/*
2229 * Returns 1 if the path is an "other" path with respect to
2230 * the index; that is, the path is not mentioned in the index at all,
2231 * either as a file, a directory with some files in the index,
2232 * or as an unmerged entry.
2233 *
2234 * We helpfully remove a trailing "/" from directories so that
2235 * the output of read_directory can be used as-is.
2236 */
2237int index_name_is_other(const struct index_state *istate, const char *name,
2238                int namelen)
2239{
2240        int pos;
2241        if (namelen && name[namelen - 1] == '/')
2242                namelen--;
2243        pos = index_name_pos(istate, name, namelen);
2244        if (0 <= pos)
2245                return 0;       /* exact match */
2246        pos = -pos - 1;
2247        if (pos < istate->cache_nr) {
2248                struct cache_entry *ce = istate->cache[pos];
2249                if (ce_namelen(ce) == namelen &&
2250                    !memcmp(ce->name, name, namelen))
2251                        return 0; /* Yup, this one exists unmerged */
2252        }
2253        return 1;
2254}
2255
2256void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2257{
2258        int pos, len;
2259        unsigned long sz;
2260        enum object_type type;
2261        void *data;
2262
2263        len = strlen(path);
2264        pos = index_name_pos(istate, path, len);
2265        if (pos < 0) {
2266                /*
2267                 * We might be in the middle of a merge, in which
2268                 * case we would read stage #2 (ours).
2269                 */
2270                int i;
2271                for (i = -pos - 1;
2272                     (pos < 0 && i < istate->cache_nr &&
2273                      !strcmp(istate->cache[i]->name, path));
2274                     i++)
2275                        if (ce_stage(istate->cache[i]) == 2)
2276                                pos = i;
2277        }
2278        if (pos < 0)
2279                return NULL;
2280        data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2281        if (!data || type != OBJ_BLOB) {
2282                free(data);
2283                return NULL;
2284        }
2285        if (size)
2286                *size = sz;
2287        return data;
2288}
2289
2290void stat_validity_clear(struct stat_validity *sv)
2291{
2292        free(sv->sd);
2293        sv->sd = NULL;
2294}
2295
2296int stat_validity_check(struct stat_validity *sv, const char *path)
2297{
2298        struct stat st;
2299
2300        if (stat(path, &st) < 0)
2301                return sv->sd == NULL;
2302        if (!sv->sd)
2303                return 0;
2304        return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2305}
2306
2307void stat_validity_update(struct stat_validity *sv, int fd)
2308{
2309        struct stat st;
2310
2311        if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2312                stat_validity_clear(sv);
2313        else {
2314                if (!sv->sd)
2315                        sv->sd = xcalloc(1, sizeof(struct stat_data));
2316                fill_stat_data(sv->sd, &st);
2317        }
2318}