read-cache.con commit Merge branch 'tk/sigchain-unnecessary-post-tempfile' (433cc7e)
   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 "utf8.h"
  21
  22static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
  23                                               unsigned int options);
  24
  25/* Mask for the name length in ce_flags in the on-disk index */
  26
  27#define CE_NAMEMASK  (0x0fff)
  28
  29/* Index extensions.
  30 *
  31 * The first letter should be 'A'..'Z' for extensions that are not
  32 * necessary for a correct operation (i.e. optimization data).
  33 * When new extensions are added that _needs_ to be understood in
  34 * order to correctly interpret the index file, pick character that
  35 * is outside the range, to cause the reader to abort.
  36 */
  37
  38#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  39#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  40#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  41#define CACHE_EXT_LINK 0x6c696e6b         /* "link" */
  42#define CACHE_EXT_UNTRACKED 0x554E5452    /* "UNTR" */
  43
  44/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
  45#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
  46                 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
  47                 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
  48
  49struct index_state the_index;
  50static const char *alternate_index_output;
  51
  52static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  53{
  54        istate->cache[nr] = ce;
  55        add_name_hash(istate, ce);
  56}
  57
  58static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  59{
  60        struct cache_entry *old = istate->cache[nr];
  61
  62        replace_index_entry_in_base(istate, old, ce);
  63        remove_name_hash(istate, old);
  64        free(old);
  65        set_index_entry(istate, nr, ce);
  66        ce->ce_flags |= CE_UPDATE_IN_BASE;
  67        istate->cache_changed |= CE_ENTRY_CHANGED;
  68}
  69
  70void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  71{
  72        struct cache_entry *old = istate->cache[nr], *new;
  73        int namelen = strlen(new_name);
  74
  75        new = xmalloc(cache_entry_size(namelen));
  76        copy_cache_entry(new, old);
  77        new->ce_flags &= ~CE_HASHED;
  78        new->ce_namelen = namelen;
  79        new->index = 0;
  80        memcpy(new->name, new_name, namelen + 1);
  81
  82        cache_tree_invalidate_path(istate, old->name);
  83        untracked_cache_remove_from_index(istate, old->name);
  84        remove_index_entry_at(istate, nr);
  85        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  86}
  87
  88void fill_stat_data(struct stat_data *sd, struct stat *st)
  89{
  90        sd->sd_ctime.sec = (unsigned int)st->st_ctime;
  91        sd->sd_mtime.sec = (unsigned int)st->st_mtime;
  92        sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
  93        sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
  94        sd->sd_dev = st->st_dev;
  95        sd->sd_ino = st->st_ino;
  96        sd->sd_uid = st->st_uid;
  97        sd->sd_gid = st->st_gid;
  98        sd->sd_size = st->st_size;
  99}
 100
 101int match_stat_data(const struct stat_data *sd, struct stat *st)
 102{
 103        int changed = 0;
 104
 105        if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
 106                changed |= MTIME_CHANGED;
 107        if (trust_ctime && check_stat &&
 108            sd->sd_ctime.sec != (unsigned int)st->st_ctime)
 109                changed |= CTIME_CHANGED;
 110
 111#ifdef USE_NSEC
 112        if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
 113                changed |= MTIME_CHANGED;
 114        if (trust_ctime && check_stat &&
 115            sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
 116                changed |= CTIME_CHANGED;
 117#endif
 118
 119        if (check_stat) {
 120                if (sd->sd_uid != (unsigned int) st->st_uid ||
 121                        sd->sd_gid != (unsigned int) st->st_gid)
 122                        changed |= OWNER_CHANGED;
 123                if (sd->sd_ino != (unsigned int) st->st_ino)
 124                        changed |= INODE_CHANGED;
 125        }
 126
 127#ifdef USE_STDEV
 128        /*
 129         * st_dev breaks on network filesystems where different
 130         * clients will have different views of what "device"
 131         * the filesystem is on
 132         */
 133        if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
 134                        changed |= INODE_CHANGED;
 135#endif
 136
 137        if (sd->sd_size != (unsigned int) st->st_size)
 138                changed |= DATA_CHANGED;
 139
 140        return changed;
 141}
 142
 143/*
 144 * This only updates the "non-critical" parts of the directory
 145 * cache, ie the parts that aren't tracked by GIT, and only used
 146 * to validate the cache.
 147 */
 148void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 149{
 150        fill_stat_data(&ce->ce_stat_data, st);
 151
 152        if (assume_unchanged)
 153                ce->ce_flags |= CE_VALID;
 154
 155        if (S_ISREG(st->st_mode))
 156                ce_mark_uptodate(ce);
 157}
 158
 159static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
 160{
 161        int match = -1;
 162        int fd = open(ce->name, O_RDONLY);
 163
 164        if (fd >= 0) {
 165                unsigned char sha1[20];
 166                if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
 167                        match = hashcmp(sha1, ce->sha1);
 168                /* index_fd() closed the file descriptor already */
 169        }
 170        return match;
 171}
 172
 173static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
 174{
 175        int match = -1;
 176        void *buffer;
 177        unsigned long size;
 178        enum object_type type;
 179        struct strbuf sb = STRBUF_INIT;
 180
 181        if (strbuf_readlink(&sb, ce->name, expected_size))
 182                return -1;
 183
 184        buffer = read_sha1_file(ce->sha1, &type, &size);
 185        if (buffer) {
 186                if (size == sb.len)
 187                        match = memcmp(buffer, sb.buf, size);
 188                free(buffer);
 189        }
 190        strbuf_release(&sb);
 191        return match;
 192}
 193
 194static int ce_compare_gitlink(const struct cache_entry *ce)
 195{
 196        unsigned char sha1[20];
 197
 198        /*
 199         * We don't actually require that the .git directory
 200         * under GITLINK directory be a valid git directory. It
 201         * might even be missing (in case nobody populated that
 202         * sub-project).
 203         *
 204         * If so, we consider it always to match.
 205         */
 206        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 207                return 0;
 208        return hashcmp(sha1, ce->sha1);
 209}
 210
 211static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
 212{
 213        switch (st->st_mode & S_IFMT) {
 214        case S_IFREG:
 215                if (ce_compare_data(ce, st))
 216                        return DATA_CHANGED;
 217                break;
 218        case S_IFLNK:
 219                if (ce_compare_link(ce, xsize_t(st->st_size)))
 220                        return DATA_CHANGED;
 221                break;
 222        case S_IFDIR:
 223                if (S_ISGITLINK(ce->ce_mode))
 224                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 225        default:
 226                return TYPE_CHANGED;
 227        }
 228        return 0;
 229}
 230
 231static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
 232{
 233        unsigned int changed = 0;
 234
 235        if (ce->ce_flags & CE_REMOVE)
 236                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 237
 238        switch (ce->ce_mode & S_IFMT) {
 239        case S_IFREG:
 240                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 241                /* We consider only the owner x bit to be relevant for
 242                 * "mode changes"
 243                 */
 244                if (trust_executable_bit &&
 245                    (0100 & (ce->ce_mode ^ st->st_mode)))
 246                        changed |= MODE_CHANGED;
 247                break;
 248        case S_IFLNK:
 249                if (!S_ISLNK(st->st_mode) &&
 250                    (has_symlinks || !S_ISREG(st->st_mode)))
 251                        changed |= TYPE_CHANGED;
 252                break;
 253        case S_IFGITLINK:
 254                /* We ignore most of the st_xxx fields for gitlinks */
 255                if (!S_ISDIR(st->st_mode))
 256                        changed |= TYPE_CHANGED;
 257                else if (ce_compare_gitlink(ce))
 258                        changed |= DATA_CHANGED;
 259                return changed;
 260        default:
 261                die("internal error: ce_mode is %o", ce->ce_mode);
 262        }
 263
 264        changed |= match_stat_data(&ce->ce_stat_data, st);
 265
 266        /* Racily smudged entry? */
 267        if (!ce->ce_stat_data.sd_size) {
 268                if (!is_empty_blob_sha1(ce->sha1))
 269                        changed |= DATA_CHANGED;
 270        }
 271
 272        return changed;
 273}
 274
 275static int is_racy_stat(const struct index_state *istate,
 276                        const struct stat_data *sd)
 277{
 278        return (istate->timestamp.sec &&
 279#ifdef USE_NSEC
 280                 /* nanosecond timestamped files can also be racy! */
 281                (istate->timestamp.sec < sd->sd_mtime.sec ||
 282                 (istate->timestamp.sec == sd->sd_mtime.sec &&
 283                  istate->timestamp.nsec <= sd->sd_mtime.nsec))
 284#else
 285                istate->timestamp.sec <= sd->sd_mtime.sec
 286#endif
 287                );
 288}
 289
 290static int is_racy_timestamp(const struct index_state *istate,
 291                             const struct cache_entry *ce)
 292{
 293        return (!S_ISGITLINK(ce->ce_mode) &&
 294                is_racy_stat(istate, &ce->ce_stat_data));
 295}
 296
 297int match_stat_data_racy(const struct index_state *istate,
 298                         const struct stat_data *sd, struct stat *st)
 299{
 300        if (is_racy_stat(istate, sd))
 301                return MTIME_CHANGED;
 302        return match_stat_data(sd, st);
 303}
 304
 305int ie_match_stat(const struct index_state *istate,
 306                  const struct cache_entry *ce, struct stat *st,
 307                  unsigned int options)
 308{
 309        unsigned int changed;
 310        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 311        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 312        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 313
 314        /*
 315         * If it's marked as always valid in the index, it's
 316         * valid whatever the checked-out copy says.
 317         *
 318         * skip-worktree has the same effect with higher precedence
 319         */
 320        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 321                return 0;
 322        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 323                return 0;
 324
 325        /*
 326         * Intent-to-add entries have not been added, so the index entry
 327         * by definition never matches what is in the work tree until it
 328         * actually gets added.
 329         */
 330        if (ce->ce_flags & CE_INTENT_TO_ADD)
 331                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 332
 333        changed = ce_match_stat_basic(ce, st);
 334
 335        /*
 336         * Within 1 second of this sequence:
 337         *      echo xyzzy >file && git-update-index --add file
 338         * running this command:
 339         *      echo frotz >file
 340         * would give a falsely clean cache entry.  The mtime and
 341         * length match the cache, and other stat fields do not change.
 342         *
 343         * We could detect this at update-index time (the cache entry
 344         * being registered/updated records the same time as "now")
 345         * and delay the return from git-update-index, but that would
 346         * effectively mean we can make at most one commit per second,
 347         * which is not acceptable.  Instead, we check cache entries
 348         * whose mtime are the same as the index file timestamp more
 349         * carefully than others.
 350         */
 351        if (!changed && is_racy_timestamp(istate, ce)) {
 352                if (assume_racy_is_modified)
 353                        changed |= DATA_CHANGED;
 354                else
 355                        changed |= ce_modified_check_fs(ce, st);
 356        }
 357
 358        return changed;
 359}
 360
 361int ie_modified(const struct index_state *istate,
 362                const struct cache_entry *ce,
 363                struct stat *st, unsigned int options)
 364{
 365        int changed, changed_fs;
 366
 367        changed = ie_match_stat(istate, ce, st, options);
 368        if (!changed)
 369                return 0;
 370        /*
 371         * If the mode or type has changed, there's no point in trying
 372         * to refresh the entry - it's not going to match
 373         */
 374        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 375                return changed;
 376
 377        /*
 378         * Immediately after read-tree or update-index --cacheinfo,
 379         * the length field is zero, as we have never even read the
 380         * lstat(2) information once, and we cannot trust DATA_CHANGED
 381         * returned by ie_match_stat() which in turn was returned by
 382         * ce_match_stat_basic() to signal that the filesize of the
 383         * blob changed.  We have to actually go to the filesystem to
 384         * see if the contents match, and if so, should answer "unchanged".
 385         *
 386         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 387         * already has checked the actual HEAD from the filesystem in the
 388         * subproject.  If ie_match_stat() already said it is different,
 389         * then we know it is.
 390         */
 391        if ((changed & DATA_CHANGED) &&
 392            (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
 393                return changed;
 394
 395        changed_fs = ce_modified_check_fs(ce, st);
 396        if (changed_fs)
 397                return changed | changed_fs;
 398        return 0;
 399}
 400
 401int base_name_compare(const char *name1, int len1, int mode1,
 402                      const char *name2, int len2, int mode2)
 403{
 404        unsigned char c1, c2;
 405        int len = len1 < len2 ? len1 : len2;
 406        int cmp;
 407
 408        cmp = memcmp(name1, name2, len);
 409        if (cmp)
 410                return cmp;
 411        c1 = name1[len];
 412        c2 = name2[len];
 413        if (!c1 && S_ISDIR(mode1))
 414                c1 = '/';
 415        if (!c2 && S_ISDIR(mode2))
 416                c2 = '/';
 417        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 418}
 419
 420/*
 421 * df_name_compare() is identical to base_name_compare(), except it
 422 * compares conflicting directory/file entries as equal. Note that
 423 * while a directory name compares as equal to a regular file, they
 424 * then individually compare _differently_ to a filename that has
 425 * a dot after the basename (because '\0' < '.' < '/').
 426 *
 427 * This is used by routines that want to traverse the git namespace
 428 * but then handle conflicting entries together when possible.
 429 */
 430int df_name_compare(const char *name1, int len1, int mode1,
 431                    const char *name2, int len2, int mode2)
 432{
 433        int len = len1 < len2 ? len1 : len2, cmp;
 434        unsigned char c1, c2;
 435
 436        cmp = memcmp(name1, name2, len);
 437        if (cmp)
 438                return cmp;
 439        /* Directories and files compare equal (same length, same name) */
 440        if (len1 == len2)
 441                return 0;
 442        c1 = name1[len];
 443        if (!c1 && S_ISDIR(mode1))
 444                c1 = '/';
 445        c2 = name2[len];
 446        if (!c2 && S_ISDIR(mode2))
 447                c2 = '/';
 448        if (c1 == '/' && !c2)
 449                return 0;
 450        if (c2 == '/' && !c1)
 451                return 0;
 452        return c1 - c2;
 453}
 454
 455int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
 456{
 457        size_t min_len = (len1 < len2) ? len1 : len2;
 458        int cmp = memcmp(name1, name2, min_len);
 459        if (cmp)
 460                return cmp;
 461        if (len1 < len2)
 462                return -1;
 463        if (len1 > len2)
 464                return 1;
 465        return 0;
 466}
 467
 468int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 469{
 470        int cmp;
 471
 472        cmp = name_compare(name1, len1, name2, len2);
 473        if (cmp)
 474                return cmp;
 475
 476        if (stage1 < stage2)
 477                return -1;
 478        if (stage1 > stage2)
 479                return 1;
 480        return 0;
 481}
 482
 483static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 484{
 485        int first, last;
 486
 487        first = 0;
 488        last = istate->cache_nr;
 489        while (last > first) {
 490                int next = (last + first) >> 1;
 491                struct cache_entry *ce = istate->cache[next];
 492                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 493                if (!cmp)
 494                        return next;
 495                if (cmp < 0) {
 496                        last = next;
 497                        continue;
 498                }
 499                first = next+1;
 500        }
 501        return -first-1;
 502}
 503
 504int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 505{
 506        return index_name_stage_pos(istate, name, namelen, 0);
 507}
 508
 509/* Remove entry, return true if there are more entries to go.. */
 510int remove_index_entry_at(struct index_state *istate, int pos)
 511{
 512        struct cache_entry *ce = istate->cache[pos];
 513
 514        record_resolve_undo(istate, ce);
 515        remove_name_hash(istate, ce);
 516        save_or_free_index_entry(istate, ce);
 517        istate->cache_changed |= CE_ENTRY_REMOVED;
 518        istate->cache_nr--;
 519        if (pos >= istate->cache_nr)
 520                return 0;
 521        memmove(istate->cache + pos,
 522                istate->cache + pos + 1,
 523                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 524        return 1;
 525}
 526
 527/*
 528 * Remove all cache entries marked for removal, that is where
 529 * CE_REMOVE is set in ce_flags.  This is much more effective than
 530 * calling remove_index_entry_at() for each entry to be removed.
 531 */
 532void remove_marked_cache_entries(struct index_state *istate)
 533{
 534        struct cache_entry **ce_array = istate->cache;
 535        unsigned int i, j;
 536
 537        for (i = j = 0; i < istate->cache_nr; i++) {
 538                if (ce_array[i]->ce_flags & CE_REMOVE) {
 539                        remove_name_hash(istate, ce_array[i]);
 540                        save_or_free_index_entry(istate, ce_array[i]);
 541                }
 542                else
 543                        ce_array[j++] = ce_array[i];
 544        }
 545        if (j == istate->cache_nr)
 546                return;
 547        istate->cache_changed |= CE_ENTRY_REMOVED;
 548        istate->cache_nr = j;
 549}
 550
 551int remove_file_from_index(struct index_state *istate, const char *path)
 552{
 553        int pos = index_name_pos(istate, path, strlen(path));
 554        if (pos < 0)
 555                pos = -pos-1;
 556        cache_tree_invalidate_path(istate, path);
 557        untracked_cache_remove_from_index(istate, path);
 558        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 559                remove_index_entry_at(istate, pos);
 560        return 0;
 561}
 562
 563static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 564{
 565        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 566}
 567
 568static int index_name_pos_also_unmerged(struct index_state *istate,
 569        const char *path, int namelen)
 570{
 571        int pos = index_name_pos(istate, path, namelen);
 572        struct cache_entry *ce;
 573
 574        if (pos >= 0)
 575                return pos;
 576
 577        /* maybe unmerged? */
 578        pos = -1 - pos;
 579        if (pos >= istate->cache_nr ||
 580                        compare_name((ce = istate->cache[pos]), path, namelen))
 581                return -1;
 582
 583        /* order of preference: stage 2, 1, 3 */
 584        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 585                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 586                        !compare_name(ce, path, namelen))
 587                pos++;
 588        return pos;
 589}
 590
 591static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 592{
 593        int len = ce_namelen(ce);
 594        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 595}
 596
 597/*
 598 * If we add a filename that aliases in the cache, we will use the
 599 * name that we already have - but we don't want to update the same
 600 * alias twice, because that implies that there were actually two
 601 * different files with aliasing names!
 602 *
 603 * So we use the CE_ADDED flag to verify that the alias was an old
 604 * one before we accept it as
 605 */
 606static struct cache_entry *create_alias_ce(struct index_state *istate,
 607                                           struct cache_entry *ce,
 608                                           struct cache_entry *alias)
 609{
 610        int len;
 611        struct cache_entry *new;
 612
 613        if (alias->ce_flags & CE_ADDED)
 614                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 615
 616        /* Ok, create the new entry using the name of the existing alias */
 617        len = ce_namelen(alias);
 618        new = xcalloc(1, cache_entry_size(len));
 619        memcpy(new->name, alias->name, len);
 620        copy_cache_entry(new, ce);
 621        save_or_free_index_entry(istate, ce);
 622        return new;
 623}
 624
 625void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
 626{
 627        unsigned char sha1[20];
 628        if (write_sha1_file("", 0, blob_type, sha1))
 629                die("cannot create an empty blob in the object database");
 630        hashcpy(ce->sha1, sha1);
 631}
 632
 633int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 634{
 635        int size, namelen, was_same;
 636        mode_t st_mode = st->st_mode;
 637        struct cache_entry *ce, *alias;
 638        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 639        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 640        int pretend = flags & ADD_CACHE_PRETEND;
 641        int intent_only = flags & ADD_CACHE_INTENT;
 642        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 643                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 644
 645        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 646                return error("%s: can only add regular files, symbolic links or git-directories", path);
 647
 648        namelen = strlen(path);
 649        if (S_ISDIR(st_mode)) {
 650                while (namelen && path[namelen-1] == '/')
 651                        namelen--;
 652        }
 653        size = cache_entry_size(namelen);
 654        ce = xcalloc(1, size);
 655        memcpy(ce->name, path, namelen);
 656        ce->ce_namelen = namelen;
 657        if (!intent_only)
 658                fill_stat_cache_info(ce, st);
 659        else
 660                ce->ce_flags |= CE_INTENT_TO_ADD;
 661
 662        if (trust_executable_bit && has_symlinks)
 663                ce->ce_mode = create_ce_mode(st_mode);
 664        else {
 665                /* If there is an existing entry, pick the mode bits and type
 666                 * from it, otherwise assume unexecutable regular file.
 667                 */
 668                struct cache_entry *ent;
 669                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 670
 671                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 672                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 673        }
 674
 675        /* When core.ignorecase=true, determine if a directory of the same name but differing
 676         * case already exists within the Git repository.  If it does, ensure the directory
 677         * case of the file being added to the repository matches (is folded into) the existing
 678         * entry's directory case.
 679         */
 680        if (ignore_case) {
 681                const char *startPtr = ce->name;
 682                const char *ptr = startPtr;
 683                while (*ptr) {
 684                        while (*ptr && *ptr != '/')
 685                                ++ptr;
 686                        if (*ptr == '/') {
 687                                struct cache_entry *foundce;
 688                                ++ptr;
 689                                foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
 690                                if (foundce) {
 691                                        memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
 692                                        startPtr = ptr;
 693                                }
 694                        }
 695                }
 696        }
 697
 698        alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 699        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 700                /* Nothing changed, really */
 701                if (!S_ISGITLINK(alias->ce_mode))
 702                        ce_mark_uptodate(alias);
 703                alias->ce_flags |= CE_ADDED;
 704
 705                free(ce);
 706                return 0;
 707        }
 708        if (!intent_only) {
 709                if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT)) {
 710                        free(ce);
 711                        return error("unable to index file %s", path);
 712                }
 713        } else
 714                set_object_name_for_intent_to_add_entry(ce);
 715
 716        if (ignore_case && alias && different_name(ce, alias))
 717                ce = create_alias_ce(istate, ce, alias);
 718        ce->ce_flags |= CE_ADDED;
 719
 720        /* It was suspected to be racily clean, but it turns out to be Ok */
 721        was_same = (alias &&
 722                    !ce_stage(alias) &&
 723                    !hashcmp(alias->sha1, ce->sha1) &&
 724                    ce->ce_mode == alias->ce_mode);
 725
 726        if (pretend)
 727                free(ce);
 728        else if (add_index_entry(istate, ce, add_option)) {
 729                free(ce);
 730                return error("unable to add %s to index", path);
 731        }
 732        if (verbose && !was_same)
 733                printf("add '%s'\n", path);
 734        return 0;
 735}
 736
 737int add_file_to_index(struct index_state *istate, const char *path, int flags)
 738{
 739        struct stat st;
 740        if (lstat(path, &st))
 741                die_errno("unable to stat '%s'", path);
 742        return add_to_index(istate, path, &st, flags);
 743}
 744
 745struct cache_entry *make_cache_entry(unsigned int mode,
 746                const unsigned char *sha1, const char *path, int stage,
 747                unsigned int refresh_options)
 748{
 749        int size, len;
 750        struct cache_entry *ce, *ret;
 751
 752        if (!verify_path(path)) {
 753                error("Invalid path '%s'", path);
 754                return NULL;
 755        }
 756
 757        len = strlen(path);
 758        size = cache_entry_size(len);
 759        ce = xcalloc(1, size);
 760
 761        hashcpy(ce->sha1, sha1);
 762        memcpy(ce->name, path, len);
 763        ce->ce_flags = create_ce_flags(stage);
 764        ce->ce_namelen = len;
 765        ce->ce_mode = create_ce_mode(mode);
 766
 767        ret = refresh_cache_entry(ce, refresh_options);
 768        if (ret != ce)
 769                free(ce);
 770        return ret;
 771}
 772
 773int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
 774{
 775        int len = ce_namelen(a);
 776        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 777}
 778
 779/*
 780 * We fundamentally don't like some paths: we don't want
 781 * dot or dot-dot anywhere, and for obvious reasons don't
 782 * want to recurse into ".git" either.
 783 *
 784 * Also, we don't want double slashes or slashes at the
 785 * end that can make pathnames ambiguous.
 786 */
 787static int verify_dotfile(const char *rest)
 788{
 789        /*
 790         * The first character was '.', but that
 791         * has already been discarded, we now test
 792         * the rest.
 793         */
 794
 795        /* "." is not allowed */
 796        if (*rest == '\0' || is_dir_sep(*rest))
 797                return 0;
 798
 799        switch (*rest) {
 800        /*
 801         * ".git" followed by  NUL or slash is bad. This
 802         * shares the path end test with the ".." case.
 803         */
 804        case 'g':
 805        case 'G':
 806                if (rest[1] != 'i' && rest[1] != 'I')
 807                        break;
 808                if (rest[2] != 't' && rest[2] != 'T')
 809                        break;
 810                rest += 2;
 811        /* fallthrough */
 812        case '.':
 813                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 814                        return 0;
 815        }
 816        return 1;
 817}
 818
 819int verify_path(const char *path)
 820{
 821        char c;
 822
 823        if (has_dos_drive_prefix(path))
 824                return 0;
 825
 826        goto inside;
 827        for (;;) {
 828                if (!c)
 829                        return 1;
 830                if (is_dir_sep(c)) {
 831inside:
 832                        if (protect_hfs && is_hfs_dotgit(path))
 833                                return 0;
 834                        if (protect_ntfs && is_ntfs_dotgit(path))
 835                                return 0;
 836                        c = *path++;
 837                        if ((c == '.' && !verify_dotfile(path)) ||
 838                            is_dir_sep(c) || c == '\0')
 839                                return 0;
 840                }
 841                c = *path++;
 842        }
 843}
 844
 845/*
 846 * Do we have another file that has the beginning components being a
 847 * proper superset of the name we're trying to add?
 848 */
 849static int has_file_name(struct index_state *istate,
 850                         const struct cache_entry *ce, int pos, int ok_to_replace)
 851{
 852        int retval = 0;
 853        int len = ce_namelen(ce);
 854        int stage = ce_stage(ce);
 855        const char *name = ce->name;
 856
 857        while (pos < istate->cache_nr) {
 858                struct cache_entry *p = istate->cache[pos++];
 859
 860                if (len >= ce_namelen(p))
 861                        break;
 862                if (memcmp(name, p->name, len))
 863                        break;
 864                if (ce_stage(p) != stage)
 865                        continue;
 866                if (p->name[len] != '/')
 867                        continue;
 868                if (p->ce_flags & CE_REMOVE)
 869                        continue;
 870                retval = -1;
 871                if (!ok_to_replace)
 872                        break;
 873                remove_index_entry_at(istate, --pos);
 874        }
 875        return retval;
 876}
 877
 878/*
 879 * Do we have another file with a pathname that is a proper
 880 * subset of the name we're trying to add?
 881 */
 882static int has_dir_name(struct index_state *istate,
 883                        const struct cache_entry *ce, int pos, int ok_to_replace)
 884{
 885        int retval = 0;
 886        int stage = ce_stage(ce);
 887        const char *name = ce->name;
 888        const char *slash = name + ce_namelen(ce);
 889
 890        for (;;) {
 891                int len;
 892
 893                for (;;) {
 894                        if (*--slash == '/')
 895                                break;
 896                        if (slash <= ce->name)
 897                                return retval;
 898                }
 899                len = slash - name;
 900
 901                pos = index_name_stage_pos(istate, name, len, stage);
 902                if (pos >= 0) {
 903                        /*
 904                         * Found one, but not so fast.  This could
 905                         * be a marker that says "I was here, but
 906                         * I am being removed".  Such an entry is
 907                         * not a part of the resulting tree, and
 908                         * it is Ok to have a directory at the same
 909                         * path.
 910                         */
 911                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 912                                retval = -1;
 913                                if (!ok_to_replace)
 914                                        break;
 915                                remove_index_entry_at(istate, pos);
 916                                continue;
 917                        }
 918                }
 919                else
 920                        pos = -pos-1;
 921
 922                /*
 923                 * Trivial optimization: if we find an entry that
 924                 * already matches the sub-directory, then we know
 925                 * we're ok, and we can exit.
 926                 */
 927                while (pos < istate->cache_nr) {
 928                        struct cache_entry *p = istate->cache[pos];
 929                        if ((ce_namelen(p) <= len) ||
 930                            (p->name[len] != '/') ||
 931                            memcmp(p->name, name, len))
 932                                break; /* not our subdirectory */
 933                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 934                                /*
 935                                 * p is at the same stage as our entry, and
 936                                 * is a subdirectory of what we are looking
 937                                 * at, so we cannot have conflicts at our
 938                                 * level or anything shorter.
 939                                 */
 940                                return retval;
 941                        pos++;
 942                }
 943        }
 944        return retval;
 945}
 946
 947/* We may be in a situation where we already have path/file and path
 948 * is being added, or we already have path and path/file is being
 949 * added.  Either one would result in a nonsense tree that has path
 950 * twice when git-write-tree tries to write it out.  Prevent it.
 951 *
 952 * If ok-to-replace is specified, we remove the conflicting entries
 953 * from the cache so the caller should recompute the insert position.
 954 * When this happens, we return non-zero.
 955 */
 956static int check_file_directory_conflict(struct index_state *istate,
 957                                         const struct cache_entry *ce,
 958                                         int pos, int ok_to_replace)
 959{
 960        int retval;
 961
 962        /*
 963         * When ce is an "I am going away" entry, we allow it to be added
 964         */
 965        if (ce->ce_flags & CE_REMOVE)
 966                return 0;
 967
 968        /*
 969         * We check if the path is a sub-path of a subsequent pathname
 970         * first, since removing those will not change the position
 971         * in the array.
 972         */
 973        retval = has_file_name(istate, ce, pos, ok_to_replace);
 974
 975        /*
 976         * Then check if the path might have a clashing sub-directory
 977         * before it.
 978         */
 979        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 980}
 981
 982static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 983{
 984        int pos;
 985        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 986        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 987        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 988        int new_only = option & ADD_CACHE_NEW_ONLY;
 989
 990        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
 991                cache_tree_invalidate_path(istate, ce->name);
 992        pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 993
 994        /* existing match? Just replace it. */
 995        if (pos >= 0) {
 996                if (!new_only)
 997                        replace_index_entry(istate, pos, ce);
 998                return 0;
 999        }
1000        pos = -pos-1;
1001
1002        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1003                untracked_cache_add_to_index(istate, ce->name);
1004
1005        /*
1006         * Inserting a merged entry ("stage 0") into the index
1007         * will always replace all non-merged entries..
1008         */
1009        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1010                while (ce_same_name(istate->cache[pos], ce)) {
1011                        ok_to_add = 1;
1012                        if (!remove_index_entry_at(istate, pos))
1013                                break;
1014                }
1015        }
1016
1017        if (!ok_to_add)
1018                return -1;
1019        if (!verify_path(ce->name))
1020                return error("Invalid path '%s'", ce->name);
1021
1022        if (!skip_df_check &&
1023            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1024                if (!ok_to_replace)
1025                        return error("'%s' appears as both a file and as a directory",
1026                                     ce->name);
1027                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1028                pos = -pos-1;
1029        }
1030        return pos + 1;
1031}
1032
1033int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1034{
1035        int pos;
1036
1037        if (option & ADD_CACHE_JUST_APPEND)
1038                pos = istate->cache_nr;
1039        else {
1040                int ret;
1041                ret = add_index_entry_with_check(istate, ce, option);
1042                if (ret <= 0)
1043                        return ret;
1044                pos = ret - 1;
1045        }
1046
1047        /* Make sure the array is big enough .. */
1048        ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1049
1050        /* Add it in.. */
1051        istate->cache_nr++;
1052        if (istate->cache_nr > pos + 1)
1053                memmove(istate->cache + pos + 1,
1054                        istate->cache + pos,
1055                        (istate->cache_nr - pos - 1) * sizeof(ce));
1056        set_index_entry(istate, pos, ce);
1057        istate->cache_changed |= CE_ENTRY_ADDED;
1058        return 0;
1059}
1060
1061/*
1062 * "refresh" does not calculate a new sha1 file or bring the
1063 * cache up-to-date for mode/content changes. But what it
1064 * _does_ do is to "re-match" the stat information of a file
1065 * with the cache, so that you can refresh the cache for a
1066 * file that hasn't been changed but where the stat entry is
1067 * out of date.
1068 *
1069 * For example, you'd want to do this after doing a "git-read-tree",
1070 * to link up the stat cache details with the proper files.
1071 */
1072static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1073                                             struct cache_entry *ce,
1074                                             unsigned int options, int *err,
1075                                             int *changed_ret)
1076{
1077        struct stat st;
1078        struct cache_entry *updated;
1079        int changed, size;
1080        int refresh = options & CE_MATCH_REFRESH;
1081        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1082        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1083        int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1084
1085        if (!refresh || ce_uptodate(ce))
1086                return ce;
1087
1088        /*
1089         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1090         * that the change to the work tree does not matter and told
1091         * us not to worry.
1092         */
1093        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1094                ce_mark_uptodate(ce);
1095                return ce;
1096        }
1097        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1098                ce_mark_uptodate(ce);
1099                return ce;
1100        }
1101
1102        if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1103                if (ignore_missing)
1104                        return ce;
1105                if (err)
1106                        *err = ENOENT;
1107                return NULL;
1108        }
1109
1110        if (lstat(ce->name, &st) < 0) {
1111                if (ignore_missing && errno == ENOENT)
1112                        return ce;
1113                if (err)
1114                        *err = errno;
1115                return NULL;
1116        }
1117
1118        changed = ie_match_stat(istate, ce, &st, options);
1119        if (changed_ret)
1120                *changed_ret = changed;
1121        if (!changed) {
1122                /*
1123                 * The path is unchanged.  If we were told to ignore
1124                 * valid bit, then we did the actual stat check and
1125                 * found that the entry is unmodified.  If the entry
1126                 * is not marked VALID, this is the place to mark it
1127                 * valid again, under "assume unchanged" mode.
1128                 */
1129                if (ignore_valid && assume_unchanged &&
1130                    !(ce->ce_flags & CE_VALID))
1131                        ; /* mark this one VALID again */
1132                else {
1133                        /*
1134                         * We do not mark the index itself "modified"
1135                         * because CE_UPTODATE flag is in-core only;
1136                         * we are not going to write this change out.
1137                         */
1138                        if (!S_ISGITLINK(ce->ce_mode))
1139                                ce_mark_uptodate(ce);
1140                        return ce;
1141                }
1142        }
1143
1144        if (ie_modified(istate, ce, &st, options)) {
1145                if (err)
1146                        *err = EINVAL;
1147                return NULL;
1148        }
1149
1150        size = ce_size(ce);
1151        updated = xmalloc(size);
1152        memcpy(updated, ce, size);
1153        fill_stat_cache_info(updated, &st);
1154        /*
1155         * If ignore_valid is not set, we should leave CE_VALID bit
1156         * alone.  Otherwise, paths marked with --no-assume-unchanged
1157         * (i.e. things to be edited) will reacquire CE_VALID bit
1158         * automatically, which is not really what we want.
1159         */
1160        if (!ignore_valid && assume_unchanged &&
1161            !(ce->ce_flags & CE_VALID))
1162                updated->ce_flags &= ~CE_VALID;
1163
1164        /* istate->cache_changed is updated in the caller */
1165        return updated;
1166}
1167
1168static void show_file(const char * fmt, const char * name, int in_porcelain,
1169                      int * first, const char *header_msg)
1170{
1171        if (in_porcelain && *first && header_msg) {
1172                printf("%s\n", header_msg);
1173                *first = 0;
1174        }
1175        printf(fmt, name);
1176}
1177
1178int refresh_index(struct index_state *istate, unsigned int flags,
1179                  const struct pathspec *pathspec,
1180                  char *seen, const char *header_msg)
1181{
1182        int i;
1183        int has_errors = 0;
1184        int really = (flags & REFRESH_REALLY) != 0;
1185        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1186        int quiet = (flags & REFRESH_QUIET) != 0;
1187        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1188        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1189        int first = 1;
1190        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1191        unsigned int options = (CE_MATCH_REFRESH |
1192                                (really ? CE_MATCH_IGNORE_VALID : 0) |
1193                                (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1194        const char *modified_fmt;
1195        const char *deleted_fmt;
1196        const char *typechange_fmt;
1197        const char *added_fmt;
1198        const char *unmerged_fmt;
1199
1200        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1201        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1202        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1203        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1204        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1205        for (i = 0; i < istate->cache_nr; i++) {
1206                struct cache_entry *ce, *new;
1207                int cache_errno = 0;
1208                int changed = 0;
1209                int filtered = 0;
1210
1211                ce = istate->cache[i];
1212                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1213                        continue;
1214
1215                if (pathspec && !ce_path_match(ce, pathspec, seen))
1216                        filtered = 1;
1217
1218                if (ce_stage(ce)) {
1219                        while ((i < istate->cache_nr) &&
1220                               ! strcmp(istate->cache[i]->name, ce->name))
1221                                i++;
1222                        i--;
1223                        if (allow_unmerged)
1224                                continue;
1225                        if (!filtered)
1226                                show_file(unmerged_fmt, ce->name, in_porcelain,
1227                                          &first, header_msg);
1228                        has_errors = 1;
1229                        continue;
1230                }
1231
1232                if (filtered)
1233                        continue;
1234
1235                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1236                if (new == ce)
1237                        continue;
1238                if (!new) {
1239                        const char *fmt;
1240
1241                        if (really && cache_errno == EINVAL) {
1242                                /* If we are doing --really-refresh that
1243                                 * means the index is not valid anymore.
1244                                 */
1245                                ce->ce_flags &= ~CE_VALID;
1246                                ce->ce_flags |= CE_UPDATE_IN_BASE;
1247                                istate->cache_changed |= CE_ENTRY_CHANGED;
1248                        }
1249                        if (quiet)
1250                                continue;
1251
1252                        if (cache_errno == ENOENT)
1253                                fmt = deleted_fmt;
1254                        else if (ce->ce_flags & CE_INTENT_TO_ADD)
1255                                fmt = added_fmt; /* must be before other checks */
1256                        else if (changed & TYPE_CHANGED)
1257                                fmt = typechange_fmt;
1258                        else
1259                                fmt = modified_fmt;
1260                        show_file(fmt,
1261                                  ce->name, in_porcelain, &first, header_msg);
1262                        has_errors = 1;
1263                        continue;
1264                }
1265
1266                replace_index_entry(istate, i, new);
1267        }
1268        return has_errors;
1269}
1270
1271static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1272                                               unsigned int options)
1273{
1274        return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1275}
1276
1277
1278/*****************************************************************
1279 * Index File I/O
1280 *****************************************************************/
1281
1282#define INDEX_FORMAT_DEFAULT 3
1283
1284static unsigned int get_index_format_default(void)
1285{
1286        char *envversion = getenv("GIT_INDEX_VERSION");
1287        char *endp;
1288        int value;
1289        unsigned int version = INDEX_FORMAT_DEFAULT;
1290
1291        if (!envversion) {
1292                if (!git_config_get_int("index.version", &value))
1293                        version = value;
1294                if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1295                        warning(_("index.version set, but the value is invalid.\n"
1296                                  "Using version %i"), INDEX_FORMAT_DEFAULT);
1297                        return INDEX_FORMAT_DEFAULT;
1298                }
1299                return version;
1300        }
1301
1302        version = strtoul(envversion, &endp, 10);
1303        if (*endp ||
1304            version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1305                warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1306                          "Using version %i"), INDEX_FORMAT_DEFAULT);
1307                version = INDEX_FORMAT_DEFAULT;
1308        }
1309        return version;
1310}
1311
1312/*
1313 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1314 * Again - this is just a (very strong in practice) heuristic that
1315 * the inode hasn't changed.
1316 *
1317 * We save the fields in big-endian order to allow using the
1318 * index file over NFS transparently.
1319 */
1320struct ondisk_cache_entry {
1321        struct cache_time ctime;
1322        struct cache_time mtime;
1323        uint32_t dev;
1324        uint32_t ino;
1325        uint32_t mode;
1326        uint32_t uid;
1327        uint32_t gid;
1328        uint32_t size;
1329        unsigned char sha1[20];
1330        uint16_t flags;
1331        char name[FLEX_ARRAY]; /* more */
1332};
1333
1334/*
1335 * This struct is used when CE_EXTENDED bit is 1
1336 * The struct must match ondisk_cache_entry exactly from
1337 * ctime till flags
1338 */
1339struct ondisk_cache_entry_extended {
1340        struct cache_time ctime;
1341        struct cache_time mtime;
1342        uint32_t dev;
1343        uint32_t ino;
1344        uint32_t mode;
1345        uint32_t uid;
1346        uint32_t gid;
1347        uint32_t size;
1348        unsigned char sha1[20];
1349        uint16_t flags;
1350        uint16_t flags2;
1351        char name[FLEX_ARRAY]; /* more */
1352};
1353
1354/* These are only used for v3 or lower */
1355#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1356#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1357#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1358#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1359                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1360                            ondisk_cache_entry_size(ce_namelen(ce)))
1361
1362static int verify_hdr(struct cache_header *hdr, unsigned long size)
1363{
1364        git_SHA_CTX c;
1365        unsigned char sha1[20];
1366        int hdr_version;
1367
1368        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1369                return error("bad signature");
1370        hdr_version = ntohl(hdr->hdr_version);
1371        if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1372                return error("bad index version %d", hdr_version);
1373        git_SHA1_Init(&c);
1374        git_SHA1_Update(&c, hdr, size - 20);
1375        git_SHA1_Final(sha1, &c);
1376        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1377                return error("bad index file sha1 signature");
1378        return 0;
1379}
1380
1381static int read_index_extension(struct index_state *istate,
1382                                const char *ext, void *data, unsigned long sz)
1383{
1384        switch (CACHE_EXT(ext)) {
1385        case CACHE_EXT_TREE:
1386                istate->cache_tree = cache_tree_read(data, sz);
1387                break;
1388        case CACHE_EXT_RESOLVE_UNDO:
1389                istate->resolve_undo = resolve_undo_read(data, sz);
1390                break;
1391        case CACHE_EXT_LINK:
1392                if (read_link_extension(istate, data, sz))
1393                        return -1;
1394                break;
1395        case CACHE_EXT_UNTRACKED:
1396                istate->untracked = read_untracked_extension(data, sz);
1397                break;
1398        default:
1399                if (*ext < 'A' || 'Z' < *ext)
1400                        return error("index uses %.4s extension, which we do not understand",
1401                                     ext);
1402                fprintf(stderr, "ignoring %.4s extension\n", ext);
1403                break;
1404        }
1405        return 0;
1406}
1407
1408int hold_locked_index(struct lock_file *lk, int die_on_error)
1409{
1410        return hold_lock_file_for_update(lk, get_index_file(),
1411                                         die_on_error
1412                                         ? LOCK_DIE_ON_ERROR
1413                                         : 0);
1414}
1415
1416int read_index(struct index_state *istate)
1417{
1418        return read_index_from(istate, get_index_file());
1419}
1420
1421static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1422                                                   unsigned int flags,
1423                                                   const char *name,
1424                                                   size_t len)
1425{
1426        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1427
1428        ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1429        ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1430        ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1431        ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1432        ce->ce_stat_data.sd_dev   = get_be32(&ondisk->dev);
1433        ce->ce_stat_data.sd_ino   = get_be32(&ondisk->ino);
1434        ce->ce_mode  = get_be32(&ondisk->mode);
1435        ce->ce_stat_data.sd_uid   = get_be32(&ondisk->uid);
1436        ce->ce_stat_data.sd_gid   = get_be32(&ondisk->gid);
1437        ce->ce_stat_data.sd_size  = get_be32(&ondisk->size);
1438        ce->ce_flags = flags & ~CE_NAMEMASK;
1439        ce->ce_namelen = len;
1440        ce->index = 0;
1441        hashcpy(ce->sha1, ondisk->sha1);
1442        memcpy(ce->name, name, len);
1443        ce->name[len] = '\0';
1444        return ce;
1445}
1446
1447/*
1448 * Adjacent cache entries tend to share the leading paths, so it makes
1449 * sense to only store the differences in later entries.  In the v4
1450 * on-disk format of the index, each on-disk cache entry stores the
1451 * number of bytes to be stripped from the end of the previous name,
1452 * and the bytes to append to the result, to come up with its name.
1453 */
1454static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1455{
1456        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1457        size_t len = decode_varint(&cp);
1458
1459        if (name->len < len)
1460                die("malformed name field in the index");
1461        strbuf_remove(name, name->len - len, len);
1462        for (ep = cp; *ep; ep++)
1463                ; /* find the end */
1464        strbuf_add(name, cp, ep - cp);
1465        return (const char *)ep + 1 - cp_;
1466}
1467
1468static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1469                                            unsigned long *ent_size,
1470                                            struct strbuf *previous_name)
1471{
1472        struct cache_entry *ce;
1473        size_t len;
1474        const char *name;
1475        unsigned int flags;
1476
1477        /* On-disk flags are just 16 bits */
1478        flags = get_be16(&ondisk->flags);
1479        len = flags & CE_NAMEMASK;
1480
1481        if (flags & CE_EXTENDED) {
1482                struct ondisk_cache_entry_extended *ondisk2;
1483                int extended_flags;
1484                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1485                extended_flags = get_be16(&ondisk2->flags2) << 16;
1486                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1487                if (extended_flags & ~CE_EXTENDED_FLAGS)
1488                        die("Unknown index entry format %08x", extended_flags);
1489                flags |= extended_flags;
1490                name = ondisk2->name;
1491        }
1492        else
1493                name = ondisk->name;
1494
1495        if (!previous_name) {
1496                /* v3 and earlier */
1497                if (len == CE_NAMEMASK)
1498                        len = strlen(name);
1499                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1500
1501                *ent_size = ondisk_ce_size(ce);
1502        } else {
1503                unsigned long consumed;
1504                consumed = expand_name_field(previous_name, name);
1505                ce = cache_entry_from_ondisk(ondisk, flags,
1506                                             previous_name->buf,
1507                                             previous_name->len);
1508
1509                *ent_size = (name - ((char *)ondisk)) + consumed;
1510        }
1511        return ce;
1512}
1513
1514static void check_ce_order(struct index_state *istate)
1515{
1516        unsigned int i;
1517
1518        for (i = 1; i < istate->cache_nr; i++) {
1519                struct cache_entry *ce = istate->cache[i - 1];
1520                struct cache_entry *next_ce = istate->cache[i];
1521                int name_compare = strcmp(ce->name, next_ce->name);
1522
1523                if (0 < name_compare)
1524                        die("unordered stage entries in index");
1525                if (!name_compare) {
1526                        if (!ce_stage(ce))
1527                                die("multiple stage entries for merged file '%s'",
1528                                    ce->name);
1529                        if (ce_stage(ce) > ce_stage(next_ce))
1530                                die("unordered stage entries for '%s'",
1531                                    ce->name);
1532                }
1533        }
1534}
1535
1536/* remember to discard_cache() before reading a different cache! */
1537int do_read_index(struct index_state *istate, const char *path, int must_exist)
1538{
1539        int fd, i;
1540        struct stat st;
1541        unsigned long src_offset;
1542        struct cache_header *hdr;
1543        void *mmap;
1544        size_t mmap_size;
1545        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1546
1547        if (istate->initialized)
1548                return istate->cache_nr;
1549
1550        istate->timestamp.sec = 0;
1551        istate->timestamp.nsec = 0;
1552        fd = open(path, O_RDONLY);
1553        if (fd < 0) {
1554                if (!must_exist && errno == ENOENT)
1555                        return 0;
1556                die_errno("%s: index file open failed", path);
1557        }
1558
1559        if (fstat(fd, &st))
1560                die_errno("cannot stat the open index");
1561
1562        mmap_size = xsize_t(st.st_size);
1563        if (mmap_size < sizeof(struct cache_header) + 20)
1564                die("index file smaller than expected");
1565
1566        mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1567        if (mmap == MAP_FAILED)
1568                die_errno("unable to map index file");
1569        close(fd);
1570
1571        hdr = mmap;
1572        if (verify_hdr(hdr, mmap_size) < 0)
1573                goto unmap;
1574
1575        hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1576        istate->version = ntohl(hdr->hdr_version);
1577        istate->cache_nr = ntohl(hdr->hdr_entries);
1578        istate->cache_alloc = alloc_nr(istate->cache_nr);
1579        istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1580        istate->initialized = 1;
1581
1582        if (istate->version == 4)
1583                previous_name = &previous_name_buf;
1584        else
1585                previous_name = NULL;
1586
1587        src_offset = sizeof(*hdr);
1588        for (i = 0; i < istate->cache_nr; i++) {
1589                struct ondisk_cache_entry *disk_ce;
1590                struct cache_entry *ce;
1591                unsigned long consumed;
1592
1593                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1594                ce = create_from_disk(disk_ce, &consumed, previous_name);
1595                set_index_entry(istate, i, ce);
1596
1597                src_offset += consumed;
1598        }
1599        strbuf_release(&previous_name_buf);
1600        istate->timestamp.sec = st.st_mtime;
1601        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1602
1603        while (src_offset <= mmap_size - 20 - 8) {
1604                /* After an array of active_nr index entries,
1605                 * there can be arbitrary number of extended
1606                 * sections, each of which is prefixed with
1607                 * extension name (4-byte) and section length
1608                 * in 4-byte network byte order.
1609                 */
1610                uint32_t extsize;
1611                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1612                extsize = ntohl(extsize);
1613                if (read_index_extension(istate,
1614                                         (const char *) mmap + src_offset,
1615                                         (char *) mmap + src_offset + 8,
1616                                         extsize) < 0)
1617                        goto unmap;
1618                src_offset += 8;
1619                src_offset += extsize;
1620        }
1621        munmap(mmap, mmap_size);
1622        return istate->cache_nr;
1623
1624unmap:
1625        munmap(mmap, mmap_size);
1626        die("index file corrupt");
1627}
1628
1629int read_index_from(struct index_state *istate, const char *path)
1630{
1631        struct split_index *split_index;
1632        int ret;
1633
1634        /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1635        if (istate->initialized)
1636                return istate->cache_nr;
1637
1638        ret = do_read_index(istate, path, 0);
1639        split_index = istate->split_index;
1640        if (!split_index || is_null_sha1(split_index->base_sha1)) {
1641                check_ce_order(istate);
1642                return ret;
1643        }
1644
1645        if (split_index->base)
1646                discard_index(split_index->base);
1647        else
1648                split_index->base = xcalloc(1, sizeof(*split_index->base));
1649        ret = do_read_index(split_index->base,
1650                            git_path("sharedindex.%s",
1651                                     sha1_to_hex(split_index->base_sha1)), 1);
1652        if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1653                die("broken index, expect %s in %s, got %s",
1654                    sha1_to_hex(split_index->base_sha1),
1655                    git_path("sharedindex.%s",
1656                             sha1_to_hex(split_index->base_sha1)),
1657                    sha1_to_hex(split_index->base->sha1));
1658        merge_base_index(istate);
1659        check_ce_order(istate);
1660        return ret;
1661}
1662
1663int is_index_unborn(struct index_state *istate)
1664{
1665        return (!istate->cache_nr && !istate->timestamp.sec);
1666}
1667
1668int discard_index(struct index_state *istate)
1669{
1670        int i;
1671
1672        for (i = 0; i < istate->cache_nr; i++) {
1673                if (istate->cache[i]->index &&
1674                    istate->split_index &&
1675                    istate->split_index->base &&
1676                    istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1677                    istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1678                        continue;
1679                free(istate->cache[i]);
1680        }
1681        resolve_undo_clear_index(istate);
1682        istate->cache_nr = 0;
1683        istate->cache_changed = 0;
1684        istate->timestamp.sec = 0;
1685        istate->timestamp.nsec = 0;
1686        free_name_hash(istate);
1687        cache_tree_free(&(istate->cache_tree));
1688        istate->initialized = 0;
1689        free(istate->cache);
1690        istate->cache = NULL;
1691        istate->cache_alloc = 0;
1692        discard_split_index(istate);
1693        free_untracked_cache(istate->untracked);
1694        istate->untracked = NULL;
1695        return 0;
1696}
1697
1698int unmerged_index(const struct index_state *istate)
1699{
1700        int i;
1701        for (i = 0; i < istate->cache_nr; i++) {
1702                if (ce_stage(istate->cache[i]))
1703                        return 1;
1704        }
1705        return 0;
1706}
1707
1708#define WRITE_BUFFER_SIZE 8192
1709static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1710static unsigned long write_buffer_len;
1711
1712static int ce_write_flush(git_SHA_CTX *context, int fd)
1713{
1714        unsigned int buffered = write_buffer_len;
1715        if (buffered) {
1716                git_SHA1_Update(context, write_buffer, buffered);
1717                if (write_in_full(fd, write_buffer, buffered) != buffered)
1718                        return -1;
1719                write_buffer_len = 0;
1720        }
1721        return 0;
1722}
1723
1724static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1725{
1726        while (len) {
1727                unsigned int buffered = write_buffer_len;
1728                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1729                if (partial > len)
1730                        partial = len;
1731                memcpy(write_buffer + buffered, data, partial);
1732                buffered += partial;
1733                if (buffered == WRITE_BUFFER_SIZE) {
1734                        write_buffer_len = buffered;
1735                        if (ce_write_flush(context, fd))
1736                                return -1;
1737                        buffered = 0;
1738                }
1739                write_buffer_len = buffered;
1740                len -= partial;
1741                data = (char *) data + partial;
1742        }
1743        return 0;
1744}
1745
1746static int write_index_ext_header(git_SHA_CTX *context, int fd,
1747                                  unsigned int ext, unsigned int sz)
1748{
1749        ext = htonl(ext);
1750        sz = htonl(sz);
1751        return ((ce_write(context, fd, &ext, 4) < 0) ||
1752                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1753}
1754
1755static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1756{
1757        unsigned int left = write_buffer_len;
1758
1759        if (left) {
1760                write_buffer_len = 0;
1761                git_SHA1_Update(context, write_buffer, left);
1762        }
1763
1764        /* Flush first if not enough space for SHA1 signature */
1765        if (left + 20 > WRITE_BUFFER_SIZE) {
1766                if (write_in_full(fd, write_buffer, left) != left)
1767                        return -1;
1768                left = 0;
1769        }
1770
1771        /* Append the SHA1 signature at the end */
1772        git_SHA1_Final(write_buffer + left, context);
1773        hashcpy(sha1, write_buffer + left);
1774        left += 20;
1775        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1776}
1777
1778static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1779{
1780        /*
1781         * The only thing we care about in this function is to smudge the
1782         * falsely clean entry due to touch-update-touch race, so we leave
1783         * everything else as they are.  We are called for entries whose
1784         * ce_stat_data.sd_mtime match the index file mtime.
1785         *
1786         * Note that this actually does not do much for gitlinks, for
1787         * which ce_match_stat_basic() always goes to the actual
1788         * contents.  The caller checks with is_racy_timestamp() which
1789         * always says "no" for gitlinks, so we are not called for them ;-)
1790         */
1791        struct stat st;
1792
1793        if (lstat(ce->name, &st) < 0)
1794                return;
1795        if (ce_match_stat_basic(ce, &st))
1796                return;
1797        if (ce_modified_check_fs(ce, &st)) {
1798                /* This is "racily clean"; smudge it.  Note that this
1799                 * is a tricky code.  At first glance, it may appear
1800                 * that it can break with this sequence:
1801                 *
1802                 * $ echo xyzzy >frotz
1803                 * $ git-update-index --add frotz
1804                 * $ : >frotz
1805                 * $ sleep 3
1806                 * $ echo filfre >nitfol
1807                 * $ git-update-index --add nitfol
1808                 *
1809                 * but it does not.  When the second update-index runs,
1810                 * it notices that the entry "frotz" has the same timestamp
1811                 * as index, and if we were to smudge it by resetting its
1812                 * size to zero here, then the object name recorded
1813                 * in index is the 6-byte file but the cached stat information
1814                 * becomes zero --- which would then match what we would
1815                 * obtain from the filesystem next time we stat("frotz").
1816                 *
1817                 * However, the second update-index, before calling
1818                 * this function, notices that the cached size is 6
1819                 * bytes and what is on the filesystem is an empty
1820                 * file, and never calls us, so the cached size information
1821                 * for "frotz" stays 6 which does not match the filesystem.
1822                 */
1823                ce->ce_stat_data.sd_size = 0;
1824        }
1825}
1826
1827/* Copy miscellaneous fields but not the name */
1828static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1829                                       struct cache_entry *ce)
1830{
1831        short flags;
1832
1833        ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1834        ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1835        ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1836        ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1837        ondisk->dev  = htonl(ce->ce_stat_data.sd_dev);
1838        ondisk->ino  = htonl(ce->ce_stat_data.sd_ino);
1839        ondisk->mode = htonl(ce->ce_mode);
1840        ondisk->uid  = htonl(ce->ce_stat_data.sd_uid);
1841        ondisk->gid  = htonl(ce->ce_stat_data.sd_gid);
1842        ondisk->size = htonl(ce->ce_stat_data.sd_size);
1843        hashcpy(ondisk->sha1, ce->sha1);
1844
1845        flags = ce->ce_flags & ~CE_NAMEMASK;
1846        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1847        ondisk->flags = htons(flags);
1848        if (ce->ce_flags & CE_EXTENDED) {
1849                struct ondisk_cache_entry_extended *ondisk2;
1850                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1851                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1852                return ondisk2->name;
1853        }
1854        else {
1855                return ondisk->name;
1856        }
1857}
1858
1859static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1860                          struct strbuf *previous_name)
1861{
1862        int size;
1863        struct ondisk_cache_entry *ondisk;
1864        int saved_namelen = saved_namelen; /* compiler workaround */
1865        char *name;
1866        int result;
1867
1868        if (ce->ce_flags & CE_STRIP_NAME) {
1869                saved_namelen = ce_namelen(ce);
1870                ce->ce_namelen = 0;
1871        }
1872
1873        if (!previous_name) {
1874                size = ondisk_ce_size(ce);
1875                ondisk = xcalloc(1, size);
1876                name = copy_cache_entry_to_ondisk(ondisk, ce);
1877                memcpy(name, ce->name, ce_namelen(ce));
1878        } else {
1879                int common, to_remove, prefix_size;
1880                unsigned char to_remove_vi[16];
1881                for (common = 0;
1882                     (ce->name[common] &&
1883                      common < previous_name->len &&
1884                      ce->name[common] == previous_name->buf[common]);
1885                     common++)
1886                        ; /* still matching */
1887                to_remove = previous_name->len - common;
1888                prefix_size = encode_varint(to_remove, to_remove_vi);
1889
1890                if (ce->ce_flags & CE_EXTENDED)
1891                        size = offsetof(struct ondisk_cache_entry_extended, name);
1892                else
1893                        size = offsetof(struct ondisk_cache_entry, name);
1894                size += prefix_size + (ce_namelen(ce) - common + 1);
1895
1896                ondisk = xcalloc(1, size);
1897                name = copy_cache_entry_to_ondisk(ondisk, ce);
1898                memcpy(name, to_remove_vi, prefix_size);
1899                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1900
1901                strbuf_splice(previous_name, common, to_remove,
1902                              ce->name + common, ce_namelen(ce) - common);
1903        }
1904        if (ce->ce_flags & CE_STRIP_NAME) {
1905                ce->ce_namelen = saved_namelen;
1906                ce->ce_flags &= ~CE_STRIP_NAME;
1907        }
1908
1909        result = ce_write(c, fd, ondisk, size);
1910        free(ondisk);
1911        return result;
1912}
1913
1914/*
1915 * This function verifies if index_state has the correct sha1 of the
1916 * index file.  Don't die if we have any other failure, just return 0.
1917 */
1918static int verify_index_from(const struct index_state *istate, const char *path)
1919{
1920        int fd;
1921        ssize_t n;
1922        struct stat st;
1923        unsigned char sha1[20];
1924
1925        if (!istate->initialized)
1926                return 0;
1927
1928        fd = open(path, O_RDONLY);
1929        if (fd < 0)
1930                return 0;
1931
1932        if (fstat(fd, &st))
1933                goto out;
1934
1935        if (st.st_size < sizeof(struct cache_header) + 20)
1936                goto out;
1937
1938        n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1939        if (n != 20)
1940                goto out;
1941
1942        if (hashcmp(istate->sha1, sha1))
1943                goto out;
1944
1945        close(fd);
1946        return 1;
1947
1948out:
1949        close(fd);
1950        return 0;
1951}
1952
1953static int verify_index(const struct index_state *istate)
1954{
1955        return verify_index_from(istate, get_index_file());
1956}
1957
1958static int has_racy_timestamp(struct index_state *istate)
1959{
1960        int entries = istate->cache_nr;
1961        int i;
1962
1963        for (i = 0; i < entries; i++) {
1964                struct cache_entry *ce = istate->cache[i];
1965                if (is_racy_timestamp(istate, ce))
1966                        return 1;
1967        }
1968        return 0;
1969}
1970
1971/*
1972 * Opportunistically update the index but do not complain if we can't
1973 */
1974void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1975{
1976        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1977            verify_index(istate) &&
1978            write_locked_index(istate, lockfile, COMMIT_LOCK))
1979                rollback_lock_file(lockfile);
1980}
1981
1982static int do_write_index(struct index_state *istate, int newfd,
1983                          int strip_extensions)
1984{
1985        git_SHA_CTX c;
1986        struct cache_header hdr;
1987        int i, err, removed, extended, hdr_version;
1988        struct cache_entry **cache = istate->cache;
1989        int entries = istate->cache_nr;
1990        struct stat st;
1991        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1992
1993        for (i = removed = extended = 0; i < entries; i++) {
1994                if (cache[i]->ce_flags & CE_REMOVE)
1995                        removed++;
1996
1997                /* reduce extended entries if possible */
1998                cache[i]->ce_flags &= ~CE_EXTENDED;
1999                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2000                        extended++;
2001                        cache[i]->ce_flags |= CE_EXTENDED;
2002                }
2003        }
2004
2005        if (!istate->version) {
2006                istate->version = get_index_format_default();
2007                if (getenv("GIT_TEST_SPLIT_INDEX"))
2008                        init_split_index(istate);
2009        }
2010
2011        /* demote version 3 to version 2 when the latter suffices */
2012        if (istate->version == 3 || istate->version == 2)
2013                istate->version = extended ? 3 : 2;
2014
2015        hdr_version = istate->version;
2016
2017        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2018        hdr.hdr_version = htonl(hdr_version);
2019        hdr.hdr_entries = htonl(entries - removed);
2020
2021        git_SHA1_Init(&c);
2022        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2023                return -1;
2024
2025        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2026        for (i = 0; i < entries; i++) {
2027                struct cache_entry *ce = cache[i];
2028                if (ce->ce_flags & CE_REMOVE)
2029                        continue;
2030                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2031                        ce_smudge_racily_clean_entry(ce);
2032                if (is_null_sha1(ce->sha1)) {
2033                        static const char msg[] = "cache entry has null sha1: %s";
2034                        static int allow = -1;
2035
2036                        if (allow < 0)
2037                                allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2038                        if (allow)
2039                                warning(msg, ce->name);
2040                        else
2041                                return error(msg, ce->name);
2042                }
2043                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2044                        return -1;
2045        }
2046        strbuf_release(&previous_name_buf);
2047
2048        /* Write extension data here */
2049        if (!strip_extensions && istate->split_index) {
2050                struct strbuf sb = STRBUF_INIT;
2051
2052                err = write_link_extension(&sb, istate) < 0 ||
2053                        write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2054                                               sb.len) < 0 ||
2055                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2056                strbuf_release(&sb);
2057                if (err)
2058                        return -1;
2059        }
2060        if (!strip_extensions && istate->cache_tree) {
2061                struct strbuf sb = STRBUF_INIT;
2062
2063                cache_tree_write(&sb, istate->cache_tree);
2064                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2065                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2066                strbuf_release(&sb);
2067                if (err)
2068                        return -1;
2069        }
2070        if (!strip_extensions && istate->resolve_undo) {
2071                struct strbuf sb = STRBUF_INIT;
2072
2073                resolve_undo_write(&sb, istate->resolve_undo);
2074                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2075                                             sb.len) < 0
2076                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2077                strbuf_release(&sb);
2078                if (err)
2079                        return -1;
2080        }
2081        if (!strip_extensions && istate->untracked) {
2082                struct strbuf sb = STRBUF_INIT;
2083
2084                write_untracked_extension(&sb, istate->untracked);
2085                err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2086                                             sb.len) < 0 ||
2087                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2088                strbuf_release(&sb);
2089                if (err)
2090                        return -1;
2091        }
2092
2093        if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2094                return -1;
2095        istate->timestamp.sec = (unsigned int)st.st_mtime;
2096        istate->timestamp.nsec = ST_MTIME_NSEC(st);
2097        return 0;
2098}
2099
2100void set_alternate_index_output(const char *name)
2101{
2102        alternate_index_output = name;
2103}
2104
2105static int commit_locked_index(struct lock_file *lk)
2106{
2107        if (alternate_index_output)
2108                return commit_lock_file_to(lk, alternate_index_output);
2109        else
2110                return commit_lock_file(lk);
2111}
2112
2113static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2114                                 unsigned flags)
2115{
2116        int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
2117        if (ret)
2118                return ret;
2119        assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2120               (COMMIT_LOCK | CLOSE_LOCK));
2121        if (flags & COMMIT_LOCK)
2122                return commit_locked_index(lock);
2123        else if (flags & CLOSE_LOCK)
2124                return close_lock_file(lock);
2125        else
2126                return ret;
2127}
2128
2129static int write_split_index(struct index_state *istate,
2130                             struct lock_file *lock,
2131                             unsigned flags)
2132{
2133        int ret;
2134        prepare_to_write_split_index(istate);
2135        ret = do_write_locked_index(istate, lock, flags);
2136        finish_writing_split_index(istate);
2137        return ret;
2138}
2139
2140static struct tempfile temporary_sharedindex;
2141
2142static int write_shared_index(struct index_state *istate,
2143                              struct lock_file *lock, unsigned flags)
2144{
2145        struct split_index *si = istate->split_index;
2146        int fd, ret;
2147
2148        fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2149        if (fd < 0) {
2150                hashclr(si->base_sha1);
2151                return do_write_locked_index(istate, lock, flags);
2152        }
2153        move_cache_to_base_index(istate);
2154        ret = do_write_index(si->base, fd, 1);
2155        if (ret) {
2156                delete_tempfile(&temporary_sharedindex);
2157                return ret;
2158        }
2159        ret = rename_tempfile(&temporary_sharedindex,
2160                              git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2161        if (!ret)
2162                hashcpy(si->base_sha1, si->base->sha1);
2163        return ret;
2164}
2165
2166int write_locked_index(struct index_state *istate, struct lock_file *lock,
2167                       unsigned flags)
2168{
2169        struct split_index *si = istate->split_index;
2170
2171        if (!si || alternate_index_output ||
2172            (istate->cache_changed & ~EXTMASK)) {
2173                if (si)
2174                        hashclr(si->base_sha1);
2175                return do_write_locked_index(istate, lock, flags);
2176        }
2177
2178        if (getenv("GIT_TEST_SPLIT_INDEX")) {
2179                int v = si->base_sha1[0];
2180                if ((v & 15) < 6)
2181                        istate->cache_changed |= SPLIT_INDEX_ORDERED;
2182        }
2183        if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2184                int ret = write_shared_index(istate, lock, flags);
2185                if (ret)
2186                        return ret;
2187        }
2188
2189        return write_split_index(istate, lock, flags);
2190}
2191
2192/*
2193 * Read the index file that is potentially unmerged into given
2194 * index_state, dropping any unmerged entries.  Returns true if
2195 * the index is unmerged.  Callers who want to refuse to work
2196 * from an unmerged state can call this and check its return value,
2197 * instead of calling read_cache().
2198 */
2199int read_index_unmerged(struct index_state *istate)
2200{
2201        int i;
2202        int unmerged = 0;
2203
2204        read_index(istate);
2205        for (i = 0; i < istate->cache_nr; i++) {
2206                struct cache_entry *ce = istate->cache[i];
2207                struct cache_entry *new_ce;
2208                int size, len;
2209
2210                if (!ce_stage(ce))
2211                        continue;
2212                unmerged = 1;
2213                len = ce_namelen(ce);
2214                size = cache_entry_size(len);
2215                new_ce = xcalloc(1, size);
2216                memcpy(new_ce->name, ce->name, len);
2217                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2218                new_ce->ce_namelen = len;
2219                new_ce->ce_mode = ce->ce_mode;
2220                if (add_index_entry(istate, new_ce, 0))
2221                        return error("%s: cannot drop to stage #0",
2222                                     new_ce->name);
2223        }
2224        return unmerged;
2225}
2226
2227/*
2228 * Returns 1 if the path is an "other" path with respect to
2229 * the index; that is, the path is not mentioned in the index at all,
2230 * either as a file, a directory with some files in the index,
2231 * or as an unmerged entry.
2232 *
2233 * We helpfully remove a trailing "/" from directories so that
2234 * the output of read_directory can be used as-is.
2235 */
2236int index_name_is_other(const struct index_state *istate, const char *name,
2237                int namelen)
2238{
2239        int pos;
2240        if (namelen && name[namelen - 1] == '/')
2241                namelen--;
2242        pos = index_name_pos(istate, name, namelen);
2243        if (0 <= pos)
2244                return 0;       /* exact match */
2245        pos = -pos - 1;
2246        if (pos < istate->cache_nr) {
2247                struct cache_entry *ce = istate->cache[pos];
2248                if (ce_namelen(ce) == namelen &&
2249                    !memcmp(ce->name, name, namelen))
2250                        return 0; /* Yup, this one exists unmerged */
2251        }
2252        return 1;
2253}
2254
2255void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2256{
2257        int pos, len;
2258        unsigned long sz;
2259        enum object_type type;
2260        void *data;
2261
2262        len = strlen(path);
2263        pos = index_name_pos(istate, path, len);
2264        if (pos < 0) {
2265                /*
2266                 * We might be in the middle of a merge, in which
2267                 * case we would read stage #2 (ours).
2268                 */
2269                int i;
2270                for (i = -pos - 1;
2271                     (pos < 0 && i < istate->cache_nr &&
2272                      !strcmp(istate->cache[i]->name, path));
2273                     i++)
2274                        if (ce_stage(istate->cache[i]) == 2)
2275                                pos = i;
2276        }
2277        if (pos < 0)
2278                return NULL;
2279        data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2280        if (!data || type != OBJ_BLOB) {
2281                free(data);
2282                return NULL;
2283        }
2284        if (size)
2285                *size = sz;
2286        return data;
2287}
2288
2289void stat_validity_clear(struct stat_validity *sv)
2290{
2291        free(sv->sd);
2292        sv->sd = NULL;
2293}
2294
2295int stat_validity_check(struct stat_validity *sv, const char *path)
2296{
2297        struct stat st;
2298
2299        if (stat(path, &st) < 0)
2300                return sv->sd == NULL;
2301        if (!sv->sd)
2302                return 0;
2303        return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2304}
2305
2306void stat_validity_update(struct stat_validity *sv, int fd)
2307{
2308        struct stat st;
2309
2310        if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2311                stat_validity_clear(sv);
2312        else {
2313                if (!sv->sd)
2314                        sv->sd = xcalloc(1, sizeof(struct stat_data));
2315                fill_stat_data(sv->sd, &st);
2316        }
2317}