read-cache.con commit Merge branch 'bw/repo-object' (85ce4a6)
   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 "config.h"
   9#include "tempfile.h"
  10#include "lockfile.h"
  11#include "cache-tree.h"
  12#include "refs.h"
  13#include "dir.h"
  14#include "tree.h"
  15#include "commit.h"
  16#include "blob.h"
  17#include "resolve-undo.h"
  18#include "strbuf.h"
  19#include "varint.h"
  20#include "split-index.h"
  21#include "utf8.h"
  22
  23/* Mask for the name length in ce_flags in the on-disk index */
  24
  25#define CE_NAMEMASK  (0x0fff)
  26
  27/* Index extensions.
  28 *
  29 * The first letter should be 'A'..'Z' for extensions that are not
  30 * necessary for a correct operation (i.e. optimization data).
  31 * When new extensions are added that _needs_ to be understood in
  32 * order to correctly interpret the index file, pick character that
  33 * is outside the range, to cause the reader to abort.
  34 */
  35
  36#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  37#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  38#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  39#define CACHE_EXT_LINK 0x6c696e6b         /* "link" */
  40#define CACHE_EXT_UNTRACKED 0x554E5452    /* "UNTR" */
  41
  42/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
  43#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
  44                 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
  45                 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
  46
  47struct index_state the_index;
  48static const char *alternate_index_output;
  49
  50static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  51{
  52        istate->cache[nr] = ce;
  53        add_name_hash(istate, ce);
  54}
  55
  56static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  57{
  58        struct cache_entry *old = istate->cache[nr];
  59
  60        replace_index_entry_in_base(istate, old, ce);
  61        remove_name_hash(istate, old);
  62        free(old);
  63        set_index_entry(istate, nr, ce);
  64        ce->ce_flags |= CE_UPDATE_IN_BASE;
  65        istate->cache_changed |= CE_ENTRY_CHANGED;
  66}
  67
  68void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  69{
  70        struct cache_entry *old = istate->cache[nr], *new;
  71        int namelen = strlen(new_name);
  72
  73        new = xmalloc(cache_entry_size(namelen));
  74        copy_cache_entry(new, old);
  75        new->ce_flags &= ~CE_HASHED;
  76        new->ce_namelen = namelen;
  77        new->index = 0;
  78        memcpy(new->name, new_name, namelen + 1);
  79
  80        cache_tree_invalidate_path(istate, old->name);
  81        untracked_cache_remove_from_index(istate, old->name);
  82        remove_index_entry_at(istate, nr);
  83        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  84}
  85
  86void fill_stat_data(struct stat_data *sd, struct stat *st)
  87{
  88        sd->sd_ctime.sec = (unsigned int)st->st_ctime;
  89        sd->sd_mtime.sec = (unsigned int)st->st_mtime;
  90        sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
  91        sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
  92        sd->sd_dev = st->st_dev;
  93        sd->sd_ino = st->st_ino;
  94        sd->sd_uid = st->st_uid;
  95        sd->sd_gid = st->st_gid;
  96        sd->sd_size = st->st_size;
  97}
  98
  99int match_stat_data(const struct stat_data *sd, struct stat *st)
 100{
 101        int changed = 0;
 102
 103        if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
 104                changed |= MTIME_CHANGED;
 105        if (trust_ctime && check_stat &&
 106            sd->sd_ctime.sec != (unsigned int)st->st_ctime)
 107                changed |= CTIME_CHANGED;
 108
 109#ifdef USE_NSEC
 110        if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
 111                changed |= MTIME_CHANGED;
 112        if (trust_ctime && check_stat &&
 113            sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
 114                changed |= CTIME_CHANGED;
 115#endif
 116
 117        if (check_stat) {
 118                if (sd->sd_uid != (unsigned int) st->st_uid ||
 119                        sd->sd_gid != (unsigned int) st->st_gid)
 120                        changed |= OWNER_CHANGED;
 121                if (sd->sd_ino != (unsigned int) st->st_ino)
 122                        changed |= INODE_CHANGED;
 123        }
 124
 125#ifdef USE_STDEV
 126        /*
 127         * st_dev breaks on network filesystems where different
 128         * clients will have different views of what "device"
 129         * the filesystem is on
 130         */
 131        if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
 132                        changed |= INODE_CHANGED;
 133#endif
 134
 135        if (sd->sd_size != (unsigned int) st->st_size)
 136                changed |= DATA_CHANGED;
 137
 138        return changed;
 139}
 140
 141/*
 142 * This only updates the "non-critical" parts of the directory
 143 * cache, ie the parts that aren't tracked by GIT, and only used
 144 * to validate the cache.
 145 */
 146void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 147{
 148        fill_stat_data(&ce->ce_stat_data, st);
 149
 150        if (assume_unchanged)
 151                ce->ce_flags |= CE_VALID;
 152
 153        if (S_ISREG(st->st_mode))
 154                ce_mark_uptodate(ce);
 155}
 156
 157static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
 158{
 159        int match = -1;
 160        int fd = git_open_cloexec(ce->name, O_RDONLY);
 161
 162        if (fd >= 0) {
 163                unsigned char sha1[20];
 164                if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
 165                        match = hashcmp(sha1, ce->oid.hash);
 166                /* index_fd() closed the file descriptor already */
 167        }
 168        return match;
 169}
 170
 171static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
 172{
 173        int match = -1;
 174        void *buffer;
 175        unsigned long size;
 176        enum object_type type;
 177        struct strbuf sb = STRBUF_INIT;
 178
 179        if (strbuf_readlink(&sb, ce->name, expected_size))
 180                return -1;
 181
 182        buffer = read_sha1_file(ce->oid.hash, &type, &size);
 183        if (buffer) {
 184                if (size == sb.len)
 185                        match = memcmp(buffer, sb.buf, size);
 186                free(buffer);
 187        }
 188        strbuf_release(&sb);
 189        return match;
 190}
 191
 192static int ce_compare_gitlink(const struct cache_entry *ce)
 193{
 194        unsigned char sha1[20];
 195
 196        /*
 197         * We don't actually require that the .git directory
 198         * under GITLINK directory be a valid git directory. It
 199         * might even be missing (in case nobody populated that
 200         * sub-project).
 201         *
 202         * If so, we consider it always to match.
 203         */
 204        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 205                return 0;
 206        return hashcmp(sha1, ce->oid.hash);
 207}
 208
 209static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
 210{
 211        switch (st->st_mode & S_IFMT) {
 212        case S_IFREG:
 213                if (ce_compare_data(ce, st))
 214                        return DATA_CHANGED;
 215                break;
 216        case S_IFLNK:
 217                if (ce_compare_link(ce, xsize_t(st->st_size)))
 218                        return DATA_CHANGED;
 219                break;
 220        case S_IFDIR:
 221                if (S_ISGITLINK(ce->ce_mode))
 222                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 223        default:
 224                return TYPE_CHANGED;
 225        }
 226        return 0;
 227}
 228
 229static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
 230{
 231        unsigned int changed = 0;
 232
 233        if (ce->ce_flags & CE_REMOVE)
 234                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 235
 236        switch (ce->ce_mode & S_IFMT) {
 237        case S_IFREG:
 238                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 239                /* We consider only the owner x bit to be relevant for
 240                 * "mode changes"
 241                 */
 242                if (trust_executable_bit &&
 243                    (0100 & (ce->ce_mode ^ st->st_mode)))
 244                        changed |= MODE_CHANGED;
 245                break;
 246        case S_IFLNK:
 247                if (!S_ISLNK(st->st_mode) &&
 248                    (has_symlinks || !S_ISREG(st->st_mode)))
 249                        changed |= TYPE_CHANGED;
 250                break;
 251        case S_IFGITLINK:
 252                /* We ignore most of the st_xxx fields for gitlinks */
 253                if (!S_ISDIR(st->st_mode))
 254                        changed |= TYPE_CHANGED;
 255                else if (ce_compare_gitlink(ce))
 256                        changed |= DATA_CHANGED;
 257                return changed;
 258        default:
 259                die("internal error: ce_mode is %o", ce->ce_mode);
 260        }
 261
 262        changed |= match_stat_data(&ce->ce_stat_data, st);
 263
 264        /* Racily smudged entry? */
 265        if (!ce->ce_stat_data.sd_size) {
 266                if (!is_empty_blob_sha1(ce->oid.hash))
 267                        changed |= DATA_CHANGED;
 268        }
 269
 270        return changed;
 271}
 272
 273static int is_racy_stat(const struct index_state *istate,
 274                        const struct stat_data *sd)
 275{
 276        return (istate->timestamp.sec &&
 277#ifdef USE_NSEC
 278                 /* nanosecond timestamped files can also be racy! */
 279                (istate->timestamp.sec < sd->sd_mtime.sec ||
 280                 (istate->timestamp.sec == sd->sd_mtime.sec &&
 281                  istate->timestamp.nsec <= sd->sd_mtime.nsec))
 282#else
 283                istate->timestamp.sec <= sd->sd_mtime.sec
 284#endif
 285                );
 286}
 287
 288static int is_racy_timestamp(const struct index_state *istate,
 289                             const struct cache_entry *ce)
 290{
 291        return (!S_ISGITLINK(ce->ce_mode) &&
 292                is_racy_stat(istate, &ce->ce_stat_data));
 293}
 294
 295int match_stat_data_racy(const struct index_state *istate,
 296                         const struct stat_data *sd, struct stat *st)
 297{
 298        if (is_racy_stat(istate, sd))
 299                return MTIME_CHANGED;
 300        return match_stat_data(sd, st);
 301}
 302
 303int ie_match_stat(const struct index_state *istate,
 304                  const struct cache_entry *ce, struct stat *st,
 305                  unsigned int options)
 306{
 307        unsigned int changed;
 308        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 309        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 310        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 311
 312        /*
 313         * If it's marked as always valid in the index, it's
 314         * valid whatever the checked-out copy says.
 315         *
 316         * skip-worktree has the same effect with higher precedence
 317         */
 318        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 319                return 0;
 320        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 321                return 0;
 322
 323        /*
 324         * Intent-to-add entries have not been added, so the index entry
 325         * by definition never matches what is in the work tree until it
 326         * actually gets added.
 327         */
 328        if (ce_intent_to_add(ce))
 329                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 330
 331        changed = ce_match_stat_basic(ce, st);
 332
 333        /*
 334         * Within 1 second of this sequence:
 335         *      echo xyzzy >file && git-update-index --add file
 336         * running this command:
 337         *      echo frotz >file
 338         * would give a falsely clean cache entry.  The mtime and
 339         * length match the cache, and other stat fields do not change.
 340         *
 341         * We could detect this at update-index time (the cache entry
 342         * being registered/updated records the same time as "now")
 343         * and delay the return from git-update-index, but that would
 344         * effectively mean we can make at most one commit per second,
 345         * which is not acceptable.  Instead, we check cache entries
 346         * whose mtime are the same as the index file timestamp more
 347         * carefully than others.
 348         */
 349        if (!changed && is_racy_timestamp(istate, ce)) {
 350                if (assume_racy_is_modified)
 351                        changed |= DATA_CHANGED;
 352                else
 353                        changed |= ce_modified_check_fs(ce, st);
 354        }
 355
 356        return changed;
 357}
 358
 359int ie_modified(const struct index_state *istate,
 360                const struct cache_entry *ce,
 361                struct stat *st, unsigned int options)
 362{
 363        int changed, changed_fs;
 364
 365        changed = ie_match_stat(istate, ce, st, options);
 366        if (!changed)
 367                return 0;
 368        /*
 369         * If the mode or type has changed, there's no point in trying
 370         * to refresh the entry - it's not going to match
 371         */
 372        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 373                return changed;
 374
 375        /*
 376         * Immediately after read-tree or update-index --cacheinfo,
 377         * the length field is zero, as we have never even read the
 378         * lstat(2) information once, and we cannot trust DATA_CHANGED
 379         * returned by ie_match_stat() which in turn was returned by
 380         * ce_match_stat_basic() to signal that the filesize of the
 381         * blob changed.  We have to actually go to the filesystem to
 382         * see if the contents match, and if so, should answer "unchanged".
 383         *
 384         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 385         * already has checked the actual HEAD from the filesystem in the
 386         * subproject.  If ie_match_stat() already said it is different,
 387         * then we know it is.
 388         */
 389        if ((changed & DATA_CHANGED) &&
 390            (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
 391                return changed;
 392
 393        changed_fs = ce_modified_check_fs(ce, st);
 394        if (changed_fs)
 395                return changed | changed_fs;
 396        return 0;
 397}
 398
 399int base_name_compare(const char *name1, int len1, int mode1,
 400                      const char *name2, int len2, int mode2)
 401{
 402        unsigned char c1, c2;
 403        int len = len1 < len2 ? len1 : len2;
 404        int cmp;
 405
 406        cmp = memcmp(name1, name2, len);
 407        if (cmp)
 408                return cmp;
 409        c1 = name1[len];
 410        c2 = name2[len];
 411        if (!c1 && S_ISDIR(mode1))
 412                c1 = '/';
 413        if (!c2 && S_ISDIR(mode2))
 414                c2 = '/';
 415        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 416}
 417
 418/*
 419 * df_name_compare() is identical to base_name_compare(), except it
 420 * compares conflicting directory/file entries as equal. Note that
 421 * while a directory name compares as equal to a regular file, they
 422 * then individually compare _differently_ to a filename that has
 423 * a dot after the basename (because '\0' < '.' < '/').
 424 *
 425 * This is used by routines that want to traverse the git namespace
 426 * but then handle conflicting entries together when possible.
 427 */
 428int df_name_compare(const char *name1, int len1, int mode1,
 429                    const char *name2, int len2, int mode2)
 430{
 431        int len = len1 < len2 ? len1 : len2, cmp;
 432        unsigned char c1, c2;
 433
 434        cmp = memcmp(name1, name2, len);
 435        if (cmp)
 436                return cmp;
 437        /* Directories and files compare equal (same length, same name) */
 438        if (len1 == len2)
 439                return 0;
 440        c1 = name1[len];
 441        if (!c1 && S_ISDIR(mode1))
 442                c1 = '/';
 443        c2 = name2[len];
 444        if (!c2 && S_ISDIR(mode2))
 445                c2 = '/';
 446        if (c1 == '/' && !c2)
 447                return 0;
 448        if (c2 == '/' && !c1)
 449                return 0;
 450        return c1 - c2;
 451}
 452
 453int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
 454{
 455        size_t min_len = (len1 < len2) ? len1 : len2;
 456        int cmp = memcmp(name1, name2, min_len);
 457        if (cmp)
 458                return cmp;
 459        if (len1 < len2)
 460                return -1;
 461        if (len1 > len2)
 462                return 1;
 463        return 0;
 464}
 465
 466int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 467{
 468        int cmp;
 469
 470        cmp = name_compare(name1, len1, name2, len2);
 471        if (cmp)
 472                return cmp;
 473
 474        if (stage1 < stage2)
 475                return -1;
 476        if (stage1 > stage2)
 477                return 1;
 478        return 0;
 479}
 480
 481static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 482{
 483        int first, last;
 484
 485        first = 0;
 486        last = istate->cache_nr;
 487        while (last > first) {
 488                int next = (last + first) >> 1;
 489                struct cache_entry *ce = istate->cache[next];
 490                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 491                if (!cmp)
 492                        return next;
 493                if (cmp < 0) {
 494                        last = next;
 495                        continue;
 496                }
 497                first = next+1;
 498        }
 499        return -first-1;
 500}
 501
 502int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 503{
 504        return index_name_stage_pos(istate, name, namelen, 0);
 505}
 506
 507int remove_index_entry_at(struct index_state *istate, int pos)
 508{
 509        struct cache_entry *ce = istate->cache[pos];
 510
 511        record_resolve_undo(istate, ce);
 512        remove_name_hash(istate, ce);
 513        save_or_free_index_entry(istate, ce);
 514        istate->cache_changed |= CE_ENTRY_REMOVED;
 515        istate->cache_nr--;
 516        if (pos >= istate->cache_nr)
 517                return 0;
 518        memmove(istate->cache + pos,
 519                istate->cache + pos + 1,
 520                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 521        return 1;
 522}
 523
 524/*
 525 * Remove all cache entries marked for removal, that is where
 526 * CE_REMOVE is set in ce_flags.  This is much more effective than
 527 * calling remove_index_entry_at() for each entry to be removed.
 528 */
 529void remove_marked_cache_entries(struct index_state *istate)
 530{
 531        struct cache_entry **ce_array = istate->cache;
 532        unsigned int i, j;
 533
 534        for (i = j = 0; i < istate->cache_nr; i++) {
 535                if (ce_array[i]->ce_flags & CE_REMOVE) {
 536                        remove_name_hash(istate, ce_array[i]);
 537                        save_or_free_index_entry(istate, ce_array[i]);
 538                }
 539                else
 540                        ce_array[j++] = ce_array[i];
 541        }
 542        if (j == istate->cache_nr)
 543                return;
 544        istate->cache_changed |= CE_ENTRY_REMOVED;
 545        istate->cache_nr = j;
 546}
 547
 548int remove_file_from_index(struct index_state *istate, const char *path)
 549{
 550        int pos = index_name_pos(istate, path, strlen(path));
 551        if (pos < 0)
 552                pos = -pos-1;
 553        cache_tree_invalidate_path(istate, path);
 554        untracked_cache_remove_from_index(istate, path);
 555        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 556                remove_index_entry_at(istate, pos);
 557        return 0;
 558}
 559
 560static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 561{
 562        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 563}
 564
 565static int index_name_pos_also_unmerged(struct index_state *istate,
 566        const char *path, int namelen)
 567{
 568        int pos = index_name_pos(istate, path, namelen);
 569        struct cache_entry *ce;
 570
 571        if (pos >= 0)
 572                return pos;
 573
 574        /* maybe unmerged? */
 575        pos = -1 - pos;
 576        if (pos >= istate->cache_nr ||
 577                        compare_name((ce = istate->cache[pos]), path, namelen))
 578                return -1;
 579
 580        /* order of preference: stage 2, 1, 3 */
 581        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 582                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 583                        !compare_name(ce, path, namelen))
 584                pos++;
 585        return pos;
 586}
 587
 588static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 589{
 590        int len = ce_namelen(ce);
 591        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 592}
 593
 594/*
 595 * If we add a filename that aliases in the cache, we will use the
 596 * name that we already have - but we don't want to update the same
 597 * alias twice, because that implies that there were actually two
 598 * different files with aliasing names!
 599 *
 600 * So we use the CE_ADDED flag to verify that the alias was an old
 601 * one before we accept it as
 602 */
 603static struct cache_entry *create_alias_ce(struct index_state *istate,
 604                                           struct cache_entry *ce,
 605                                           struct cache_entry *alias)
 606{
 607        int len;
 608        struct cache_entry *new;
 609
 610        if (alias->ce_flags & CE_ADDED)
 611                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 612
 613        /* Ok, create the new entry using the name of the existing alias */
 614        len = ce_namelen(alias);
 615        new = xcalloc(1, cache_entry_size(len));
 616        memcpy(new->name, alias->name, len);
 617        copy_cache_entry(new, ce);
 618        save_or_free_index_entry(istate, ce);
 619        return new;
 620}
 621
 622void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
 623{
 624        unsigned char sha1[20];
 625        if (write_sha1_file("", 0, blob_type, sha1))
 626                die("cannot create an empty blob in the object database");
 627        hashcpy(ce->oid.hash, sha1);
 628}
 629
 630int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 631{
 632        int size, namelen, was_same;
 633        mode_t st_mode = st->st_mode;
 634        struct cache_entry *ce, *alias;
 635        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 636        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 637        int pretend = flags & ADD_CACHE_PRETEND;
 638        int intent_only = flags & ADD_CACHE_INTENT;
 639        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 640                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 641
 642        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 643                return error("%s: can only add regular files, symbolic links or git-directories", path);
 644
 645        namelen = strlen(path);
 646        if (S_ISDIR(st_mode)) {
 647                while (namelen && path[namelen-1] == '/')
 648                        namelen--;
 649        }
 650        size = cache_entry_size(namelen);
 651        ce = xcalloc(1, size);
 652        memcpy(ce->name, path, namelen);
 653        ce->ce_namelen = namelen;
 654        if (!intent_only)
 655                fill_stat_cache_info(ce, st);
 656        else
 657                ce->ce_flags |= CE_INTENT_TO_ADD;
 658
 659
 660        if (trust_executable_bit && has_symlinks) {
 661                ce->ce_mode = create_ce_mode(st_mode);
 662        } else {
 663                /* If there is an existing entry, pick the mode bits and type
 664                 * from it, otherwise assume unexecutable regular file.
 665                 */
 666                struct cache_entry *ent;
 667                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 668
 669                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 670                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 671        }
 672
 673        /* When core.ignorecase=true, determine if a directory of the same name but differing
 674         * case already exists within the Git repository.  If it does, ensure the directory
 675         * case of the file being added to the repository matches (is folded into) the existing
 676         * entry's directory case.
 677         */
 678        if (ignore_case) {
 679                adjust_dirname_case(istate, ce->name);
 680        }
 681
 682        alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 683        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 684                /* Nothing changed, really */
 685                if (!S_ISGITLINK(alias->ce_mode))
 686                        ce_mark_uptodate(alias);
 687                alias->ce_flags |= CE_ADDED;
 688
 689                free(ce);
 690                return 0;
 691        }
 692        if (!intent_only) {
 693                if (index_path(ce->oid.hash, path, st, HASH_WRITE_OBJECT)) {
 694                        free(ce);
 695                        return error("unable to index file %s", path);
 696                }
 697        } else
 698                set_object_name_for_intent_to_add_entry(ce);
 699
 700        if (ignore_case && alias && different_name(ce, alias))
 701                ce = create_alias_ce(istate, ce, alias);
 702        ce->ce_flags |= CE_ADDED;
 703
 704        /* It was suspected to be racily clean, but it turns out to be Ok */
 705        was_same = (alias &&
 706                    !ce_stage(alias) &&
 707                    !oidcmp(&alias->oid, &ce->oid) &&
 708                    ce->ce_mode == alias->ce_mode);
 709
 710        if (pretend)
 711                free(ce);
 712        else if (add_index_entry(istate, ce, add_option)) {
 713                free(ce);
 714                return error("unable to add %s to index", path);
 715        }
 716        if (verbose && !was_same)
 717                printf("add '%s'\n", path);
 718        return 0;
 719}
 720
 721int add_file_to_index(struct index_state *istate, const char *path, int flags)
 722{
 723        struct stat st;
 724        if (lstat(path, &st))
 725                die_errno("unable to stat '%s'", path);
 726        return add_to_index(istate, path, &st, flags);
 727}
 728
 729struct cache_entry *make_cache_entry(unsigned int mode,
 730                const unsigned char *sha1, const char *path, int stage,
 731                unsigned int refresh_options)
 732{
 733        int size, len;
 734        struct cache_entry *ce, *ret;
 735
 736        if (!verify_path(path)) {
 737                error("Invalid path '%s'", path);
 738                return NULL;
 739        }
 740
 741        len = strlen(path);
 742        size = cache_entry_size(len);
 743        ce = xcalloc(1, size);
 744
 745        hashcpy(ce->oid.hash, sha1);
 746        memcpy(ce->name, path, len);
 747        ce->ce_flags = create_ce_flags(stage);
 748        ce->ce_namelen = len;
 749        ce->ce_mode = create_ce_mode(mode);
 750
 751        ret = refresh_cache_entry(ce, refresh_options);
 752        if (ret != ce)
 753                free(ce);
 754        return ret;
 755}
 756
 757/*
 758 * Chmod an index entry with either +x or -x.
 759 *
 760 * Returns -1 if the chmod for the particular cache entry failed (if it's
 761 * not a regular file), -2 if an invalid flip argument is passed in, 0
 762 * otherwise.
 763 */
 764int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
 765                      char flip)
 766{
 767        if (!S_ISREG(ce->ce_mode))
 768                return -1;
 769        switch (flip) {
 770        case '+':
 771                ce->ce_mode |= 0111;
 772                break;
 773        case '-':
 774                ce->ce_mode &= ~0111;
 775                break;
 776        default:
 777                return -2;
 778        }
 779        cache_tree_invalidate_path(istate, ce->name);
 780        ce->ce_flags |= CE_UPDATE_IN_BASE;
 781        istate->cache_changed |= CE_ENTRY_CHANGED;
 782
 783        return 0;
 784}
 785
 786int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
 787{
 788        int len = ce_namelen(a);
 789        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 790}
 791
 792/*
 793 * We fundamentally don't like some paths: we don't want
 794 * dot or dot-dot anywhere, and for obvious reasons don't
 795 * want to recurse into ".git" either.
 796 *
 797 * Also, we don't want double slashes or slashes at the
 798 * end that can make pathnames ambiguous.
 799 */
 800static int verify_dotfile(const char *rest)
 801{
 802        /*
 803         * The first character was '.', but that
 804         * has already been discarded, we now test
 805         * the rest.
 806         */
 807
 808        /* "." is not allowed */
 809        if (*rest == '\0' || is_dir_sep(*rest))
 810                return 0;
 811
 812        switch (*rest) {
 813        /*
 814         * ".git" followed by  NUL or slash is bad. This
 815         * shares the path end test with the ".." case.
 816         */
 817        case 'g':
 818        case 'G':
 819                if (rest[1] != 'i' && rest[1] != 'I')
 820                        break;
 821                if (rest[2] != 't' && rest[2] != 'T')
 822                        break;
 823                rest += 2;
 824        /* fallthrough */
 825        case '.':
 826                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 827                        return 0;
 828        }
 829        return 1;
 830}
 831
 832int verify_path(const char *path)
 833{
 834        char c;
 835
 836        if (has_dos_drive_prefix(path))
 837                return 0;
 838
 839        goto inside;
 840        for (;;) {
 841                if (!c)
 842                        return 1;
 843                if (is_dir_sep(c)) {
 844inside:
 845                        if (protect_hfs && is_hfs_dotgit(path))
 846                                return 0;
 847                        if (protect_ntfs && is_ntfs_dotgit(path))
 848                                return 0;
 849                        c = *path++;
 850                        if ((c == '.' && !verify_dotfile(path)) ||
 851                            is_dir_sep(c) || c == '\0')
 852                                return 0;
 853                }
 854                c = *path++;
 855        }
 856}
 857
 858/*
 859 * Do we have another file that has the beginning components being a
 860 * proper superset of the name we're trying to add?
 861 */
 862static int has_file_name(struct index_state *istate,
 863                         const struct cache_entry *ce, int pos, int ok_to_replace)
 864{
 865        int retval = 0;
 866        int len = ce_namelen(ce);
 867        int stage = ce_stage(ce);
 868        const char *name = ce->name;
 869
 870        while (pos < istate->cache_nr) {
 871                struct cache_entry *p = istate->cache[pos++];
 872
 873                if (len >= ce_namelen(p))
 874                        break;
 875                if (memcmp(name, p->name, len))
 876                        break;
 877                if (ce_stage(p) != stage)
 878                        continue;
 879                if (p->name[len] != '/')
 880                        continue;
 881                if (p->ce_flags & CE_REMOVE)
 882                        continue;
 883                retval = -1;
 884                if (!ok_to_replace)
 885                        break;
 886                remove_index_entry_at(istate, --pos);
 887        }
 888        return retval;
 889}
 890
 891
 892/*
 893 * Like strcmp(), but also return the offset of the first change.
 894 * If strings are equal, return the length.
 895 */
 896int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
 897{
 898        size_t k;
 899
 900        if (!first_change)
 901                return strcmp(s1, s2);
 902
 903        for (k = 0; s1[k] == s2[k]; k++)
 904                if (s1[k] == '\0')
 905                        break;
 906
 907        *first_change = k;
 908        return (unsigned char)s1[k] - (unsigned char)s2[k];
 909}
 910
 911/*
 912 * Do we have another file with a pathname that is a proper
 913 * subset of the name we're trying to add?
 914 *
 915 * That is, is there another file in the index with a path
 916 * that matches a sub-directory in the given entry?
 917 */
 918static int has_dir_name(struct index_state *istate,
 919                        const struct cache_entry *ce, int pos, int ok_to_replace)
 920{
 921        int retval = 0;
 922        int stage = ce_stage(ce);
 923        const char *name = ce->name;
 924        const char *slash = name + ce_namelen(ce);
 925        size_t len_eq_last;
 926        int cmp_last = 0;
 927
 928        /*
 929         * We are frequently called during an iteration on a sorted
 930         * list of pathnames and while building a new index.  Therefore,
 931         * there is a high probability that this entry will eventually
 932         * be appended to the index, rather than inserted in the middle.
 933         * If we can confirm that, we can avoid binary searches on the
 934         * components of the pathname.
 935         *
 936         * Compare the entry's full path with the last path in the index.
 937         */
 938        if (istate->cache_nr > 0) {
 939                cmp_last = strcmp_offset(name,
 940                        istate->cache[istate->cache_nr - 1]->name,
 941                        &len_eq_last);
 942                if (cmp_last > 0) {
 943                        if (len_eq_last == 0) {
 944                                /*
 945                                 * The entry sorts AFTER the last one in the
 946                                 * index and their paths have no common prefix,
 947                                 * so there cannot be a F/D conflict.
 948                                 */
 949                                return retval;
 950                        } else {
 951                                /*
 952                                 * The entry sorts AFTER the last one in the
 953                                 * index, but has a common prefix.  Fall through
 954                                 * to the loop below to disect the entry's path
 955                                 * and see where the difference is.
 956                                 */
 957                        }
 958                } else if (cmp_last == 0) {
 959                        /*
 960                         * The entry exactly matches the last one in the
 961                         * index, but because of multiple stage and CE_REMOVE
 962                         * items, we fall through and let the regular search
 963                         * code handle it.
 964                         */
 965                }
 966        }
 967
 968        for (;;) {
 969                size_t len;
 970
 971                for (;;) {
 972                        if (*--slash == '/')
 973                                break;
 974                        if (slash <= ce->name)
 975                                return retval;
 976                }
 977                len = slash - name;
 978
 979                if (cmp_last > 0) {
 980                        /*
 981                         * (len + 1) is a directory boundary (including
 982                         * the trailing slash).  And since the loop is
 983                         * decrementing "slash", the first iteration is
 984                         * the longest directory prefix; subsequent
 985                         * iterations consider parent directories.
 986                         */
 987
 988                        if (len + 1 <= len_eq_last) {
 989                                /*
 990                                 * The directory prefix (including the trailing
 991                                 * slash) also appears as a prefix in the last
 992                                 * entry, so the remainder cannot collide (because
 993                                 * strcmp said the whole path was greater).
 994                                 *
 995                                 * EQ: last: xxx/A
 996                                 *     this: xxx/B
 997                                 *
 998                                 * LT: last: xxx/file_A
 999                                 *     this: xxx/file_B
1000                                 */
1001                                return retval;
1002                        }
1003
1004                        if (len > len_eq_last) {
1005                                /*
1006                                 * This part of the directory prefix (excluding
1007                                 * the trailing slash) is longer than the known
1008                                 * equal portions, so this sub-directory cannot
1009                                 * collide with a file.
1010                                 *
1011                                 * GT: last: xxxA
1012                                 *     this: xxxB/file
1013                                 */
1014                                return retval;
1015                        }
1016
1017                        if (istate->cache_nr > 0 &&
1018                                ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1019                                /*
1020                                 * The directory prefix lines up with part of
1021                                 * a longer file or directory name, but sorts
1022                                 * after it, so this sub-directory cannot
1023                                 * collide with a file.
1024                                 *
1025                                 * last: xxx/yy-file (because '-' sorts before '/')
1026                                 * this: xxx/yy/abc
1027                                 */
1028                                return retval;
1029                        }
1030
1031                        /*
1032                         * This is a possible collision. Fall through and
1033                         * let the regular search code handle it.
1034                         *
1035                         * last: xxx
1036                         * this: xxx/file
1037                         */
1038                }
1039
1040                pos = index_name_stage_pos(istate, name, len, stage);
1041                if (pos >= 0) {
1042                        /*
1043                         * Found one, but not so fast.  This could
1044                         * be a marker that says "I was here, but
1045                         * I am being removed".  Such an entry is
1046                         * not a part of the resulting tree, and
1047                         * it is Ok to have a directory at the same
1048                         * path.
1049                         */
1050                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1051                                retval = -1;
1052                                if (!ok_to_replace)
1053                                        break;
1054                                remove_index_entry_at(istate, pos);
1055                                continue;
1056                        }
1057                }
1058                else
1059                        pos = -pos-1;
1060
1061                /*
1062                 * Trivial optimization: if we find an entry that
1063                 * already matches the sub-directory, then we know
1064                 * we're ok, and we can exit.
1065                 */
1066                while (pos < istate->cache_nr) {
1067                        struct cache_entry *p = istate->cache[pos];
1068                        if ((ce_namelen(p) <= len) ||
1069                            (p->name[len] != '/') ||
1070                            memcmp(p->name, name, len))
1071                                break; /* not our subdirectory */
1072                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1073                                /*
1074                                 * p is at the same stage as our entry, and
1075                                 * is a subdirectory of what we are looking
1076                                 * at, so we cannot have conflicts at our
1077                                 * level or anything shorter.
1078                                 */
1079                                return retval;
1080                        pos++;
1081                }
1082        }
1083        return retval;
1084}
1085
1086/* We may be in a situation where we already have path/file and path
1087 * is being added, or we already have path and path/file is being
1088 * added.  Either one would result in a nonsense tree that has path
1089 * twice when git-write-tree tries to write it out.  Prevent it.
1090 *
1091 * If ok-to-replace is specified, we remove the conflicting entries
1092 * from the cache so the caller should recompute the insert position.
1093 * When this happens, we return non-zero.
1094 */
1095static int check_file_directory_conflict(struct index_state *istate,
1096                                         const struct cache_entry *ce,
1097                                         int pos, int ok_to_replace)
1098{
1099        int retval;
1100
1101        /*
1102         * When ce is an "I am going away" entry, we allow it to be added
1103         */
1104        if (ce->ce_flags & CE_REMOVE)
1105                return 0;
1106
1107        /*
1108         * We check if the path is a sub-path of a subsequent pathname
1109         * first, since removing those will not change the position
1110         * in the array.
1111         */
1112        retval = has_file_name(istate, ce, pos, ok_to_replace);
1113
1114        /*
1115         * Then check if the path might have a clashing sub-directory
1116         * before it.
1117         */
1118        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1119}
1120
1121static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1122{
1123        int pos;
1124        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1125        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1126        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1127        int new_only = option & ADD_CACHE_NEW_ONLY;
1128
1129        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1130                cache_tree_invalidate_path(istate, ce->name);
1131
1132        /*
1133         * If this entry's path sorts after the last entry in the index,
1134         * we can avoid searching for it.
1135         */
1136        if (istate->cache_nr > 0 &&
1137                strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1138                pos = -istate->cache_nr - 1;
1139        else
1140                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1141
1142        /* existing match? Just replace it. */
1143        if (pos >= 0) {
1144                if (!new_only)
1145                        replace_index_entry(istate, pos, ce);
1146                return 0;
1147        }
1148        pos = -pos-1;
1149
1150        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1151                untracked_cache_add_to_index(istate, ce->name);
1152
1153        /*
1154         * Inserting a merged entry ("stage 0") into the index
1155         * will always replace all non-merged entries..
1156         */
1157        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1158                while (ce_same_name(istate->cache[pos], ce)) {
1159                        ok_to_add = 1;
1160                        if (!remove_index_entry_at(istate, pos))
1161                                break;
1162                }
1163        }
1164
1165        if (!ok_to_add)
1166                return -1;
1167        if (!verify_path(ce->name))
1168                return error("Invalid path '%s'", ce->name);
1169
1170        if (!skip_df_check &&
1171            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1172                if (!ok_to_replace)
1173                        return error("'%s' appears as both a file and as a directory",
1174                                     ce->name);
1175                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1176                pos = -pos-1;
1177        }
1178        return pos + 1;
1179}
1180
1181int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1182{
1183        int pos;
1184
1185        if (option & ADD_CACHE_JUST_APPEND)
1186                pos = istate->cache_nr;
1187        else {
1188                int ret;
1189                ret = add_index_entry_with_check(istate, ce, option);
1190                if (ret <= 0)
1191                        return ret;
1192                pos = ret - 1;
1193        }
1194
1195        /* Make sure the array is big enough .. */
1196        ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1197
1198        /* Add it in.. */
1199        istate->cache_nr++;
1200        if (istate->cache_nr > pos + 1)
1201                memmove(istate->cache + pos + 1,
1202                        istate->cache + pos,
1203                        (istate->cache_nr - pos - 1) * sizeof(ce));
1204        set_index_entry(istate, pos, ce);
1205        istate->cache_changed |= CE_ENTRY_ADDED;
1206        return 0;
1207}
1208
1209/*
1210 * "refresh" does not calculate a new sha1 file or bring the
1211 * cache up-to-date for mode/content changes. But what it
1212 * _does_ do is to "re-match" the stat information of a file
1213 * with the cache, so that you can refresh the cache for a
1214 * file that hasn't been changed but where the stat entry is
1215 * out of date.
1216 *
1217 * For example, you'd want to do this after doing a "git-read-tree",
1218 * to link up the stat cache details with the proper files.
1219 */
1220static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1221                                             struct cache_entry *ce,
1222                                             unsigned int options, int *err,
1223                                             int *changed_ret)
1224{
1225        struct stat st;
1226        struct cache_entry *updated;
1227        int changed, size;
1228        int refresh = options & CE_MATCH_REFRESH;
1229        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1230        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1231        int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1232
1233        if (!refresh || ce_uptodate(ce))
1234                return ce;
1235
1236        /*
1237         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1238         * that the change to the work tree does not matter and told
1239         * us not to worry.
1240         */
1241        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1242                ce_mark_uptodate(ce);
1243                return ce;
1244        }
1245        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1246                ce_mark_uptodate(ce);
1247                return ce;
1248        }
1249
1250        if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1251                if (ignore_missing)
1252                        return ce;
1253                if (err)
1254                        *err = ENOENT;
1255                return NULL;
1256        }
1257
1258        if (lstat(ce->name, &st) < 0) {
1259                if (ignore_missing && errno == ENOENT)
1260                        return ce;
1261                if (err)
1262                        *err = errno;
1263                return NULL;
1264        }
1265
1266        changed = ie_match_stat(istate, ce, &st, options);
1267        if (changed_ret)
1268                *changed_ret = changed;
1269        if (!changed) {
1270                /*
1271                 * The path is unchanged.  If we were told to ignore
1272                 * valid bit, then we did the actual stat check and
1273                 * found that the entry is unmodified.  If the entry
1274                 * is not marked VALID, this is the place to mark it
1275                 * valid again, under "assume unchanged" mode.
1276                 */
1277                if (ignore_valid && assume_unchanged &&
1278                    !(ce->ce_flags & CE_VALID))
1279                        ; /* mark this one VALID again */
1280                else {
1281                        /*
1282                         * We do not mark the index itself "modified"
1283                         * because CE_UPTODATE flag is in-core only;
1284                         * we are not going to write this change out.
1285                         */
1286                        if (!S_ISGITLINK(ce->ce_mode))
1287                                ce_mark_uptodate(ce);
1288                        return ce;
1289                }
1290        }
1291
1292        if (ie_modified(istate, ce, &st, options)) {
1293                if (err)
1294                        *err = EINVAL;
1295                return NULL;
1296        }
1297
1298        size = ce_size(ce);
1299        updated = xmalloc(size);
1300        memcpy(updated, ce, size);
1301        fill_stat_cache_info(updated, &st);
1302        /*
1303         * If ignore_valid is not set, we should leave CE_VALID bit
1304         * alone.  Otherwise, paths marked with --no-assume-unchanged
1305         * (i.e. things to be edited) will reacquire CE_VALID bit
1306         * automatically, which is not really what we want.
1307         */
1308        if (!ignore_valid && assume_unchanged &&
1309            !(ce->ce_flags & CE_VALID))
1310                updated->ce_flags &= ~CE_VALID;
1311
1312        /* istate->cache_changed is updated in the caller */
1313        return updated;
1314}
1315
1316static void show_file(const char * fmt, const char * name, int in_porcelain,
1317                      int * first, const char *header_msg)
1318{
1319        if (in_porcelain && *first && header_msg) {
1320                printf("%s\n", header_msg);
1321                *first = 0;
1322        }
1323        printf(fmt, name);
1324}
1325
1326int refresh_index(struct index_state *istate, unsigned int flags,
1327                  const struct pathspec *pathspec,
1328                  char *seen, const char *header_msg)
1329{
1330        int i;
1331        int has_errors = 0;
1332        int really = (flags & REFRESH_REALLY) != 0;
1333        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1334        int quiet = (flags & REFRESH_QUIET) != 0;
1335        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1336        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1337        int first = 1;
1338        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1339        unsigned int options = (CE_MATCH_REFRESH |
1340                                (really ? CE_MATCH_IGNORE_VALID : 0) |
1341                                (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1342        const char *modified_fmt;
1343        const char *deleted_fmt;
1344        const char *typechange_fmt;
1345        const char *added_fmt;
1346        const char *unmerged_fmt;
1347
1348        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1349        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1350        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1351        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1352        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1353        for (i = 0; i < istate->cache_nr; i++) {
1354                struct cache_entry *ce, *new;
1355                int cache_errno = 0;
1356                int changed = 0;
1357                int filtered = 0;
1358
1359                ce = istate->cache[i];
1360                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1361                        continue;
1362
1363                if (pathspec && !ce_path_match(ce, pathspec, seen))
1364                        filtered = 1;
1365
1366                if (ce_stage(ce)) {
1367                        while ((i < istate->cache_nr) &&
1368                               ! strcmp(istate->cache[i]->name, ce->name))
1369                                i++;
1370                        i--;
1371                        if (allow_unmerged)
1372                                continue;
1373                        if (!filtered)
1374                                show_file(unmerged_fmt, ce->name, in_porcelain,
1375                                          &first, header_msg);
1376                        has_errors = 1;
1377                        continue;
1378                }
1379
1380                if (filtered)
1381                        continue;
1382
1383                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1384                if (new == ce)
1385                        continue;
1386                if (!new) {
1387                        const char *fmt;
1388
1389                        if (really && cache_errno == EINVAL) {
1390                                /* If we are doing --really-refresh that
1391                                 * means the index is not valid anymore.
1392                                 */
1393                                ce->ce_flags &= ~CE_VALID;
1394                                ce->ce_flags |= CE_UPDATE_IN_BASE;
1395                                istate->cache_changed |= CE_ENTRY_CHANGED;
1396                        }
1397                        if (quiet)
1398                                continue;
1399
1400                        if (cache_errno == ENOENT)
1401                                fmt = deleted_fmt;
1402                        else if (ce_intent_to_add(ce))
1403                                fmt = added_fmt; /* must be before other checks */
1404                        else if (changed & TYPE_CHANGED)
1405                                fmt = typechange_fmt;
1406                        else
1407                                fmt = modified_fmt;
1408                        show_file(fmt,
1409                                  ce->name, in_porcelain, &first, header_msg);
1410                        has_errors = 1;
1411                        continue;
1412                }
1413
1414                replace_index_entry(istate, i, new);
1415        }
1416        return has_errors;
1417}
1418
1419struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1420                                               unsigned int options)
1421{
1422        return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1423}
1424
1425
1426/*****************************************************************
1427 * Index File I/O
1428 *****************************************************************/
1429
1430#define INDEX_FORMAT_DEFAULT 3
1431
1432static unsigned int get_index_format_default(void)
1433{
1434        char *envversion = getenv("GIT_INDEX_VERSION");
1435        char *endp;
1436        int value;
1437        unsigned int version = INDEX_FORMAT_DEFAULT;
1438
1439        if (!envversion) {
1440                if (!git_config_get_int("index.version", &value))
1441                        version = value;
1442                if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1443                        warning(_("index.version set, but the value is invalid.\n"
1444                                  "Using version %i"), INDEX_FORMAT_DEFAULT);
1445                        return INDEX_FORMAT_DEFAULT;
1446                }
1447                return version;
1448        }
1449
1450        version = strtoul(envversion, &endp, 10);
1451        if (*endp ||
1452            version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1453                warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1454                          "Using version %i"), INDEX_FORMAT_DEFAULT);
1455                version = INDEX_FORMAT_DEFAULT;
1456        }
1457        return version;
1458}
1459
1460/*
1461 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1462 * Again - this is just a (very strong in practice) heuristic that
1463 * the inode hasn't changed.
1464 *
1465 * We save the fields in big-endian order to allow using the
1466 * index file over NFS transparently.
1467 */
1468struct ondisk_cache_entry {
1469        struct cache_time ctime;
1470        struct cache_time mtime;
1471        uint32_t dev;
1472        uint32_t ino;
1473        uint32_t mode;
1474        uint32_t uid;
1475        uint32_t gid;
1476        uint32_t size;
1477        unsigned char sha1[20];
1478        uint16_t flags;
1479        char name[FLEX_ARRAY]; /* more */
1480};
1481
1482/*
1483 * This struct is used when CE_EXTENDED bit is 1
1484 * The struct must match ondisk_cache_entry exactly from
1485 * ctime till flags
1486 */
1487struct ondisk_cache_entry_extended {
1488        struct cache_time ctime;
1489        struct cache_time mtime;
1490        uint32_t dev;
1491        uint32_t ino;
1492        uint32_t mode;
1493        uint32_t uid;
1494        uint32_t gid;
1495        uint32_t size;
1496        unsigned char sha1[20];
1497        uint16_t flags;
1498        uint16_t flags2;
1499        char name[FLEX_ARRAY]; /* more */
1500};
1501
1502/* These are only used for v3 or lower */
1503#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1504#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1505#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1506#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1507                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1508                            ondisk_cache_entry_size(ce_namelen(ce)))
1509
1510/* Allow fsck to force verification of the index checksum. */
1511int verify_index_checksum;
1512
1513static int verify_hdr(struct cache_header *hdr, unsigned long size)
1514{
1515        git_SHA_CTX c;
1516        unsigned char sha1[20];
1517        int hdr_version;
1518
1519        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1520                return error("bad signature");
1521        hdr_version = ntohl(hdr->hdr_version);
1522        if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1523                return error("bad index version %d", hdr_version);
1524
1525        if (!verify_index_checksum)
1526                return 0;
1527
1528        git_SHA1_Init(&c);
1529        git_SHA1_Update(&c, hdr, size - 20);
1530        git_SHA1_Final(sha1, &c);
1531        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1532                return error("bad index file sha1 signature");
1533        return 0;
1534}
1535
1536static int read_index_extension(struct index_state *istate,
1537                                const char *ext, void *data, unsigned long sz)
1538{
1539        switch (CACHE_EXT(ext)) {
1540        case CACHE_EXT_TREE:
1541                istate->cache_tree = cache_tree_read(data, sz);
1542                break;
1543        case CACHE_EXT_RESOLVE_UNDO:
1544                istate->resolve_undo = resolve_undo_read(data, sz);
1545                break;
1546        case CACHE_EXT_LINK:
1547                if (read_link_extension(istate, data, sz))
1548                        return -1;
1549                break;
1550        case CACHE_EXT_UNTRACKED:
1551                istate->untracked = read_untracked_extension(data, sz);
1552                break;
1553        default:
1554                if (*ext < 'A' || 'Z' < *ext)
1555                        return error("index uses %.4s extension, which we do not understand",
1556                                     ext);
1557                fprintf(stderr, "ignoring %.4s extension\n", ext);
1558                break;
1559        }
1560        return 0;
1561}
1562
1563int hold_locked_index(struct lock_file *lk, int lock_flags)
1564{
1565        return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1566}
1567
1568int read_index(struct index_state *istate)
1569{
1570        return read_index_from(istate, get_index_file());
1571}
1572
1573static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1574                                                   unsigned int flags,
1575                                                   const char *name,
1576                                                   size_t len)
1577{
1578        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1579
1580        ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1581        ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1582        ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1583        ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1584        ce->ce_stat_data.sd_dev   = get_be32(&ondisk->dev);
1585        ce->ce_stat_data.sd_ino   = get_be32(&ondisk->ino);
1586        ce->ce_mode  = get_be32(&ondisk->mode);
1587        ce->ce_stat_data.sd_uid   = get_be32(&ondisk->uid);
1588        ce->ce_stat_data.sd_gid   = get_be32(&ondisk->gid);
1589        ce->ce_stat_data.sd_size  = get_be32(&ondisk->size);
1590        ce->ce_flags = flags & ~CE_NAMEMASK;
1591        ce->ce_namelen = len;
1592        ce->index = 0;
1593        hashcpy(ce->oid.hash, ondisk->sha1);
1594        memcpy(ce->name, name, len);
1595        ce->name[len] = '\0';
1596        return ce;
1597}
1598
1599/*
1600 * Adjacent cache entries tend to share the leading paths, so it makes
1601 * sense to only store the differences in later entries.  In the v4
1602 * on-disk format of the index, each on-disk cache entry stores the
1603 * number of bytes to be stripped from the end of the previous name,
1604 * and the bytes to append to the result, to come up with its name.
1605 */
1606static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1607{
1608        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1609        size_t len = decode_varint(&cp);
1610
1611        if (name->len < len)
1612                die("malformed name field in the index");
1613        strbuf_remove(name, name->len - len, len);
1614        for (ep = cp; *ep; ep++)
1615                ; /* find the end */
1616        strbuf_add(name, cp, ep - cp);
1617        return (const char *)ep + 1 - cp_;
1618}
1619
1620static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1621                                            unsigned long *ent_size,
1622                                            struct strbuf *previous_name)
1623{
1624        struct cache_entry *ce;
1625        size_t len;
1626        const char *name;
1627        unsigned int flags;
1628
1629        /* On-disk flags are just 16 bits */
1630        flags = get_be16(&ondisk->flags);
1631        len = flags & CE_NAMEMASK;
1632
1633        if (flags & CE_EXTENDED) {
1634                struct ondisk_cache_entry_extended *ondisk2;
1635                int extended_flags;
1636                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1637                extended_flags = get_be16(&ondisk2->flags2) << 16;
1638                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1639                if (extended_flags & ~CE_EXTENDED_FLAGS)
1640                        die("Unknown index entry format %08x", extended_flags);
1641                flags |= extended_flags;
1642                name = ondisk2->name;
1643        }
1644        else
1645                name = ondisk->name;
1646
1647        if (!previous_name) {
1648                /* v3 and earlier */
1649                if (len == CE_NAMEMASK)
1650                        len = strlen(name);
1651                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1652
1653                *ent_size = ondisk_ce_size(ce);
1654        } else {
1655                unsigned long consumed;
1656                consumed = expand_name_field(previous_name, name);
1657                ce = cache_entry_from_ondisk(ondisk, flags,
1658                                             previous_name->buf,
1659                                             previous_name->len);
1660
1661                *ent_size = (name - ((char *)ondisk)) + consumed;
1662        }
1663        return ce;
1664}
1665
1666static void check_ce_order(struct index_state *istate)
1667{
1668        unsigned int i;
1669
1670        for (i = 1; i < istate->cache_nr; i++) {
1671                struct cache_entry *ce = istate->cache[i - 1];
1672                struct cache_entry *next_ce = istate->cache[i];
1673                int name_compare = strcmp(ce->name, next_ce->name);
1674
1675                if (0 < name_compare)
1676                        die("unordered stage entries in index");
1677                if (!name_compare) {
1678                        if (!ce_stage(ce))
1679                                die("multiple stage entries for merged file '%s'",
1680                                    ce->name);
1681                        if (ce_stage(ce) > ce_stage(next_ce))
1682                                die("unordered stage entries for '%s'",
1683                                    ce->name);
1684                }
1685        }
1686}
1687
1688static void tweak_untracked_cache(struct index_state *istate)
1689{
1690        switch (git_config_get_untracked_cache()) {
1691        case -1: /* keep: do nothing */
1692                break;
1693        case 0: /* false */
1694                remove_untracked_cache(istate);
1695                break;
1696        case 1: /* true */
1697                add_untracked_cache(istate);
1698                break;
1699        default: /* unknown value: do nothing */
1700                break;
1701        }
1702}
1703
1704static void tweak_split_index(struct index_state *istate)
1705{
1706        switch (git_config_get_split_index()) {
1707        case -1: /* unset: do nothing */
1708                break;
1709        case 0: /* false */
1710                remove_split_index(istate);
1711                break;
1712        case 1: /* true */
1713                add_split_index(istate);
1714                break;
1715        default: /* unknown value: do nothing */
1716                break;
1717        }
1718}
1719
1720static void post_read_index_from(struct index_state *istate)
1721{
1722        check_ce_order(istate);
1723        tweak_untracked_cache(istate);
1724        tweak_split_index(istate);
1725}
1726
1727/* remember to discard_cache() before reading a different cache! */
1728int do_read_index(struct index_state *istate, const char *path, int must_exist)
1729{
1730        int fd, i;
1731        struct stat st;
1732        unsigned long src_offset;
1733        struct cache_header *hdr;
1734        void *mmap;
1735        size_t mmap_size;
1736        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1737
1738        if (istate->initialized)
1739                return istate->cache_nr;
1740
1741        istate->timestamp.sec = 0;
1742        istate->timestamp.nsec = 0;
1743        fd = open(path, O_RDONLY);
1744        if (fd < 0) {
1745                if (!must_exist && errno == ENOENT)
1746                        return 0;
1747                die_errno("%s: index file open failed", path);
1748        }
1749
1750        if (fstat(fd, &st))
1751                die_errno("cannot stat the open index");
1752
1753        mmap_size = xsize_t(st.st_size);
1754        if (mmap_size < sizeof(struct cache_header) + 20)
1755                die("index file smaller than expected");
1756
1757        mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1758        if (mmap == MAP_FAILED)
1759                die_errno("unable to map index file");
1760        close(fd);
1761
1762        hdr = mmap;
1763        if (verify_hdr(hdr, mmap_size) < 0)
1764                goto unmap;
1765
1766        hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1767        istate->version = ntohl(hdr->hdr_version);
1768        istate->cache_nr = ntohl(hdr->hdr_entries);
1769        istate->cache_alloc = alloc_nr(istate->cache_nr);
1770        istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1771        istate->initialized = 1;
1772
1773        if (istate->version == 4)
1774                previous_name = &previous_name_buf;
1775        else
1776                previous_name = NULL;
1777
1778        src_offset = sizeof(*hdr);
1779        for (i = 0; i < istate->cache_nr; i++) {
1780                struct ondisk_cache_entry *disk_ce;
1781                struct cache_entry *ce;
1782                unsigned long consumed;
1783
1784                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1785                ce = create_from_disk(disk_ce, &consumed, previous_name);
1786                set_index_entry(istate, i, ce);
1787
1788                src_offset += consumed;
1789        }
1790        strbuf_release(&previous_name_buf);
1791        istate->timestamp.sec = st.st_mtime;
1792        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1793
1794        while (src_offset <= mmap_size - 20 - 8) {
1795                /* After an array of active_nr index entries,
1796                 * there can be arbitrary number of extended
1797                 * sections, each of which is prefixed with
1798                 * extension name (4-byte) and section length
1799                 * in 4-byte network byte order.
1800                 */
1801                uint32_t extsize;
1802                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1803                extsize = ntohl(extsize);
1804                if (read_index_extension(istate,
1805                                         (const char *) mmap + src_offset,
1806                                         (char *) mmap + src_offset + 8,
1807                                         extsize) < 0)
1808                        goto unmap;
1809                src_offset += 8;
1810                src_offset += extsize;
1811        }
1812        munmap(mmap, mmap_size);
1813        return istate->cache_nr;
1814
1815unmap:
1816        munmap(mmap, mmap_size);
1817        die("index file corrupt");
1818}
1819
1820/*
1821 * Signal that the shared index is used by updating its mtime.
1822 *
1823 * This way, shared index can be removed if they have not been used
1824 * for some time.
1825 */
1826static void freshen_shared_index(char *base_sha1_hex, int warn)
1827{
1828        char *shared_index = git_pathdup("sharedindex.%s", base_sha1_hex);
1829        if (!check_and_freshen_file(shared_index, 1) && warn)
1830                warning("could not freshen shared index '%s'", shared_index);
1831        free(shared_index);
1832}
1833
1834int read_index_from(struct index_state *istate, const char *path)
1835{
1836        struct split_index *split_index;
1837        int ret;
1838        char *base_sha1_hex;
1839        const char *base_path;
1840
1841        /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1842        if (istate->initialized)
1843                return istate->cache_nr;
1844
1845        ret = do_read_index(istate, path, 0);
1846
1847        split_index = istate->split_index;
1848        if (!split_index || is_null_sha1(split_index->base_sha1)) {
1849                post_read_index_from(istate);
1850                return ret;
1851        }
1852
1853        if (split_index->base)
1854                discard_index(split_index->base);
1855        else
1856                split_index->base = xcalloc(1, sizeof(*split_index->base));
1857
1858        base_sha1_hex = sha1_to_hex(split_index->base_sha1);
1859        base_path = git_path("sharedindex.%s", base_sha1_hex);
1860        ret = do_read_index(split_index->base, base_path, 1);
1861        if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1862                die("broken index, expect %s in %s, got %s",
1863                    base_sha1_hex, base_path,
1864                    sha1_to_hex(split_index->base->sha1));
1865
1866        freshen_shared_index(base_sha1_hex, 0);
1867        merge_base_index(istate);
1868        post_read_index_from(istate);
1869        return ret;
1870}
1871
1872int is_index_unborn(struct index_state *istate)
1873{
1874        return (!istate->cache_nr && !istate->timestamp.sec);
1875}
1876
1877int discard_index(struct index_state *istate)
1878{
1879        int i;
1880
1881        for (i = 0; i < istate->cache_nr; i++) {
1882                if (istate->cache[i]->index &&
1883                    istate->split_index &&
1884                    istate->split_index->base &&
1885                    istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1886                    istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1887                        continue;
1888                free(istate->cache[i]);
1889        }
1890        resolve_undo_clear_index(istate);
1891        istate->cache_nr = 0;
1892        istate->cache_changed = 0;
1893        istate->timestamp.sec = 0;
1894        istate->timestamp.nsec = 0;
1895        free_name_hash(istate);
1896        cache_tree_free(&(istate->cache_tree));
1897        istate->initialized = 0;
1898        FREE_AND_NULL(istate->cache);
1899        istate->cache_alloc = 0;
1900        discard_split_index(istate);
1901        free_untracked_cache(istate->untracked);
1902        istate->untracked = NULL;
1903        return 0;
1904}
1905
1906int unmerged_index(const struct index_state *istate)
1907{
1908        int i;
1909        for (i = 0; i < istate->cache_nr; i++) {
1910                if (ce_stage(istate->cache[i]))
1911                        return 1;
1912        }
1913        return 0;
1914}
1915
1916#define WRITE_BUFFER_SIZE 8192
1917static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1918static unsigned long write_buffer_len;
1919
1920static int ce_write_flush(git_SHA_CTX *context, int fd)
1921{
1922        unsigned int buffered = write_buffer_len;
1923        if (buffered) {
1924                git_SHA1_Update(context, write_buffer, buffered);
1925                if (write_in_full(fd, write_buffer, buffered) != buffered)
1926                        return -1;
1927                write_buffer_len = 0;
1928        }
1929        return 0;
1930}
1931
1932static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1933{
1934        while (len) {
1935                unsigned int buffered = write_buffer_len;
1936                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1937                if (partial > len)
1938                        partial = len;
1939                memcpy(write_buffer + buffered, data, partial);
1940                buffered += partial;
1941                if (buffered == WRITE_BUFFER_SIZE) {
1942                        write_buffer_len = buffered;
1943                        if (ce_write_flush(context, fd))
1944                                return -1;
1945                        buffered = 0;
1946                }
1947                write_buffer_len = buffered;
1948                len -= partial;
1949                data = (char *) data + partial;
1950        }
1951        return 0;
1952}
1953
1954static int write_index_ext_header(git_SHA_CTX *context, int fd,
1955                                  unsigned int ext, unsigned int sz)
1956{
1957        ext = htonl(ext);
1958        sz = htonl(sz);
1959        return ((ce_write(context, fd, &ext, 4) < 0) ||
1960                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1961}
1962
1963static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1964{
1965        unsigned int left = write_buffer_len;
1966
1967        if (left) {
1968                write_buffer_len = 0;
1969                git_SHA1_Update(context, write_buffer, left);
1970        }
1971
1972        /* Flush first if not enough space for SHA1 signature */
1973        if (left + 20 > WRITE_BUFFER_SIZE) {
1974                if (write_in_full(fd, write_buffer, left) != left)
1975                        return -1;
1976                left = 0;
1977        }
1978
1979        /* Append the SHA1 signature at the end */
1980        git_SHA1_Final(write_buffer + left, context);
1981        hashcpy(sha1, write_buffer + left);
1982        left += 20;
1983        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1984}
1985
1986static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1987{
1988        /*
1989         * The only thing we care about in this function is to smudge the
1990         * falsely clean entry due to touch-update-touch race, so we leave
1991         * everything else as they are.  We are called for entries whose
1992         * ce_stat_data.sd_mtime match the index file mtime.
1993         *
1994         * Note that this actually does not do much for gitlinks, for
1995         * which ce_match_stat_basic() always goes to the actual
1996         * contents.  The caller checks with is_racy_timestamp() which
1997         * always says "no" for gitlinks, so we are not called for them ;-)
1998         */
1999        struct stat st;
2000
2001        if (lstat(ce->name, &st) < 0)
2002                return;
2003        if (ce_match_stat_basic(ce, &st))
2004                return;
2005        if (ce_modified_check_fs(ce, &st)) {
2006                /* This is "racily clean"; smudge it.  Note that this
2007                 * is a tricky code.  At first glance, it may appear
2008                 * that it can break with this sequence:
2009                 *
2010                 * $ echo xyzzy >frotz
2011                 * $ git-update-index --add frotz
2012                 * $ : >frotz
2013                 * $ sleep 3
2014                 * $ echo filfre >nitfol
2015                 * $ git-update-index --add nitfol
2016                 *
2017                 * but it does not.  When the second update-index runs,
2018                 * it notices that the entry "frotz" has the same timestamp
2019                 * as index, and if we were to smudge it by resetting its
2020                 * size to zero here, then the object name recorded
2021                 * in index is the 6-byte file but the cached stat information
2022                 * becomes zero --- which would then match what we would
2023                 * obtain from the filesystem next time we stat("frotz").
2024                 *
2025                 * However, the second update-index, before calling
2026                 * this function, notices that the cached size is 6
2027                 * bytes and what is on the filesystem is an empty
2028                 * file, and never calls us, so the cached size information
2029                 * for "frotz" stays 6 which does not match the filesystem.
2030                 */
2031                ce->ce_stat_data.sd_size = 0;
2032        }
2033}
2034
2035/* Copy miscellaneous fields but not the name */
2036static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2037                                       struct cache_entry *ce)
2038{
2039        short flags;
2040
2041        ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2042        ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2043        ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2044        ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2045        ondisk->dev  = htonl(ce->ce_stat_data.sd_dev);
2046        ondisk->ino  = htonl(ce->ce_stat_data.sd_ino);
2047        ondisk->mode = htonl(ce->ce_mode);
2048        ondisk->uid  = htonl(ce->ce_stat_data.sd_uid);
2049        ondisk->gid  = htonl(ce->ce_stat_data.sd_gid);
2050        ondisk->size = htonl(ce->ce_stat_data.sd_size);
2051        hashcpy(ondisk->sha1, ce->oid.hash);
2052
2053        flags = ce->ce_flags & ~CE_NAMEMASK;
2054        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2055        ondisk->flags = htons(flags);
2056        if (ce->ce_flags & CE_EXTENDED) {
2057                struct ondisk_cache_entry_extended *ondisk2;
2058                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2059                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2060                return ondisk2->name;
2061        }
2062        else {
2063                return ondisk->name;
2064        }
2065}
2066
2067static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
2068                          struct strbuf *previous_name)
2069{
2070        int size;
2071        struct ondisk_cache_entry *ondisk;
2072        int saved_namelen = saved_namelen; /* compiler workaround */
2073        char *name;
2074        int result;
2075
2076        if (ce->ce_flags & CE_STRIP_NAME) {
2077                saved_namelen = ce_namelen(ce);
2078                ce->ce_namelen = 0;
2079        }
2080
2081        if (!previous_name) {
2082                size = ondisk_ce_size(ce);
2083                ondisk = xcalloc(1, size);
2084                name = copy_cache_entry_to_ondisk(ondisk, ce);
2085                memcpy(name, ce->name, ce_namelen(ce));
2086        } else {
2087                int common, to_remove, prefix_size;
2088                unsigned char to_remove_vi[16];
2089                for (common = 0;
2090                     (ce->name[common] &&
2091                      common < previous_name->len &&
2092                      ce->name[common] == previous_name->buf[common]);
2093                     common++)
2094                        ; /* still matching */
2095                to_remove = previous_name->len - common;
2096                prefix_size = encode_varint(to_remove, to_remove_vi);
2097
2098                if (ce->ce_flags & CE_EXTENDED)
2099                        size = offsetof(struct ondisk_cache_entry_extended, name);
2100                else
2101                        size = offsetof(struct ondisk_cache_entry, name);
2102                size += prefix_size + (ce_namelen(ce) - common + 1);
2103
2104                ondisk = xcalloc(1, size);
2105                name = copy_cache_entry_to_ondisk(ondisk, ce);
2106                memcpy(name, to_remove_vi, prefix_size);
2107                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
2108
2109                strbuf_splice(previous_name, common, to_remove,
2110                              ce->name + common, ce_namelen(ce) - common);
2111        }
2112        if (ce->ce_flags & CE_STRIP_NAME) {
2113                ce->ce_namelen = saved_namelen;
2114                ce->ce_flags &= ~CE_STRIP_NAME;
2115        }
2116
2117        result = ce_write(c, fd, ondisk, size);
2118        free(ondisk);
2119        return result;
2120}
2121
2122/*
2123 * This function verifies if index_state has the correct sha1 of the
2124 * index file.  Don't die if we have any other failure, just return 0.
2125 */
2126static int verify_index_from(const struct index_state *istate, const char *path)
2127{
2128        int fd;
2129        ssize_t n;
2130        struct stat st;
2131        unsigned char sha1[20];
2132
2133        if (!istate->initialized)
2134                return 0;
2135
2136        fd = open(path, O_RDONLY);
2137        if (fd < 0)
2138                return 0;
2139
2140        if (fstat(fd, &st))
2141                goto out;
2142
2143        if (st.st_size < sizeof(struct cache_header) + 20)
2144                goto out;
2145
2146        n = pread_in_full(fd, sha1, 20, st.st_size - 20);
2147        if (n != 20)
2148                goto out;
2149
2150        if (hashcmp(istate->sha1, sha1))
2151                goto out;
2152
2153        close(fd);
2154        return 1;
2155
2156out:
2157        close(fd);
2158        return 0;
2159}
2160
2161static int verify_index(const struct index_state *istate)
2162{
2163        return verify_index_from(istate, get_index_file());
2164}
2165
2166static int has_racy_timestamp(struct index_state *istate)
2167{
2168        int entries = istate->cache_nr;
2169        int i;
2170
2171        for (i = 0; i < entries; i++) {
2172                struct cache_entry *ce = istate->cache[i];
2173                if (is_racy_timestamp(istate, ce))
2174                        return 1;
2175        }
2176        return 0;
2177}
2178
2179/*
2180 * Opportunistically update the index but do not complain if we can't
2181 */
2182void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2183{
2184        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2185            verify_index(istate) &&
2186            write_locked_index(istate, lockfile, COMMIT_LOCK))
2187                rollback_lock_file(lockfile);
2188}
2189
2190static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2191                          int strip_extensions)
2192{
2193        int newfd = tempfile->fd;
2194        git_SHA_CTX c;
2195        struct cache_header hdr;
2196        int i, err, removed, extended, hdr_version;
2197        struct cache_entry **cache = istate->cache;
2198        int entries = istate->cache_nr;
2199        struct stat st;
2200        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2201        int drop_cache_tree = 0;
2202
2203        for (i = removed = extended = 0; i < entries; i++) {
2204                if (cache[i]->ce_flags & CE_REMOVE)
2205                        removed++;
2206
2207                /* reduce extended entries if possible */
2208                cache[i]->ce_flags &= ~CE_EXTENDED;
2209                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2210                        extended++;
2211                        cache[i]->ce_flags |= CE_EXTENDED;
2212                }
2213        }
2214
2215        if (!istate->version) {
2216                istate->version = get_index_format_default();
2217                if (getenv("GIT_TEST_SPLIT_INDEX"))
2218                        init_split_index(istate);
2219        }
2220
2221        /* demote version 3 to version 2 when the latter suffices */
2222        if (istate->version == 3 || istate->version == 2)
2223                istate->version = extended ? 3 : 2;
2224
2225        hdr_version = istate->version;
2226
2227        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2228        hdr.hdr_version = htonl(hdr_version);
2229        hdr.hdr_entries = htonl(entries - removed);
2230
2231        git_SHA1_Init(&c);
2232        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2233                return -1;
2234
2235        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2236        for (i = 0; i < entries; i++) {
2237                struct cache_entry *ce = cache[i];
2238                if (ce->ce_flags & CE_REMOVE)
2239                        continue;
2240                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2241                        ce_smudge_racily_clean_entry(ce);
2242                if (is_null_oid(&ce->oid)) {
2243                        static const char msg[] = "cache entry has null sha1: %s";
2244                        static int allow = -1;
2245
2246                        if (allow < 0)
2247                                allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2248                        if (allow)
2249                                warning(msg, ce->name);
2250                        else
2251                                return error(msg, ce->name);
2252
2253                        drop_cache_tree = 1;
2254                }
2255                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2256                        return -1;
2257        }
2258        strbuf_release(&previous_name_buf);
2259
2260        /* Write extension data here */
2261        if (!strip_extensions && istate->split_index) {
2262                struct strbuf sb = STRBUF_INIT;
2263
2264                err = write_link_extension(&sb, istate) < 0 ||
2265                        write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2266                                               sb.len) < 0 ||
2267                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2268                strbuf_release(&sb);
2269                if (err)
2270                        return -1;
2271        }
2272        if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2273                struct strbuf sb = STRBUF_INIT;
2274
2275                cache_tree_write(&sb, istate->cache_tree);
2276                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2277                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2278                strbuf_release(&sb);
2279                if (err)
2280                        return -1;
2281        }
2282        if (!strip_extensions && istate->resolve_undo) {
2283                struct strbuf sb = STRBUF_INIT;
2284
2285                resolve_undo_write(&sb, istate->resolve_undo);
2286                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2287                                             sb.len) < 0
2288                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2289                strbuf_release(&sb);
2290                if (err)
2291                        return -1;
2292        }
2293        if (!strip_extensions && istate->untracked) {
2294                struct strbuf sb = STRBUF_INIT;
2295
2296                write_untracked_extension(&sb, istate->untracked);
2297                err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2298                                             sb.len) < 0 ||
2299                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2300                strbuf_release(&sb);
2301                if (err)
2302                        return -1;
2303        }
2304
2305        if (ce_flush(&c, newfd, istate->sha1))
2306                return -1;
2307        if (close_tempfile(tempfile))
2308                return error(_("could not close '%s'"), tempfile->filename.buf);
2309        if (stat(tempfile->filename.buf, &st))
2310                return -1;
2311        istate->timestamp.sec = (unsigned int)st.st_mtime;
2312        istate->timestamp.nsec = ST_MTIME_NSEC(st);
2313        return 0;
2314}
2315
2316void set_alternate_index_output(const char *name)
2317{
2318        alternate_index_output = name;
2319}
2320
2321static int commit_locked_index(struct lock_file *lk)
2322{
2323        if (alternate_index_output)
2324                return commit_lock_file_to(lk, alternate_index_output);
2325        else
2326                return commit_lock_file(lk);
2327}
2328
2329static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2330                                 unsigned flags)
2331{
2332        int ret = do_write_index(istate, &lock->tempfile, 0);
2333        if (ret)
2334                return ret;
2335        assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2336               (COMMIT_LOCK | CLOSE_LOCK));
2337        if (flags & COMMIT_LOCK)
2338                return commit_locked_index(lock);
2339        else if (flags & CLOSE_LOCK)
2340                return close_lock_file(lock);
2341        else
2342                return ret;
2343}
2344
2345static int write_split_index(struct index_state *istate,
2346                             struct lock_file *lock,
2347                             unsigned flags)
2348{
2349        int ret;
2350        prepare_to_write_split_index(istate);
2351        ret = do_write_locked_index(istate, lock, flags);
2352        finish_writing_split_index(istate);
2353        return ret;
2354}
2355
2356static const char *shared_index_expire = "2.weeks.ago";
2357
2358static unsigned long get_shared_index_expire_date(void)
2359{
2360        static unsigned long shared_index_expire_date;
2361        static int shared_index_expire_date_prepared;
2362
2363        if (!shared_index_expire_date_prepared) {
2364                git_config_get_expiry("splitindex.sharedindexexpire",
2365                                      &shared_index_expire);
2366                shared_index_expire_date = approxidate(shared_index_expire);
2367                shared_index_expire_date_prepared = 1;
2368        }
2369
2370        return shared_index_expire_date;
2371}
2372
2373static int should_delete_shared_index(const char *shared_index_path)
2374{
2375        struct stat st;
2376        unsigned long expiration;
2377
2378        /* Check timestamp */
2379        expiration = get_shared_index_expire_date();
2380        if (!expiration)
2381                return 0;
2382        if (stat(shared_index_path, &st))
2383                return error_errno(_("could not stat '%s'"), shared_index_path);
2384        if (st.st_mtime > expiration)
2385                return 0;
2386
2387        return 1;
2388}
2389
2390static int clean_shared_index_files(const char *current_hex)
2391{
2392        struct dirent *de;
2393        DIR *dir = opendir(get_git_dir());
2394
2395        if (!dir)
2396                return error_errno(_("unable to open git dir: %s"), get_git_dir());
2397
2398        while ((de = readdir(dir)) != NULL) {
2399                const char *sha1_hex;
2400                const char *shared_index_path;
2401                if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2402                        continue;
2403                if (!strcmp(sha1_hex, current_hex))
2404                        continue;
2405                shared_index_path = git_path("%s", de->d_name);
2406                if (should_delete_shared_index(shared_index_path) > 0 &&
2407                    unlink(shared_index_path))
2408                        warning_errno(_("unable to unlink: %s"), shared_index_path);
2409        }
2410        closedir(dir);
2411
2412        return 0;
2413}
2414
2415static struct tempfile temporary_sharedindex;
2416
2417static int write_shared_index(struct index_state *istate,
2418                              struct lock_file *lock, unsigned flags)
2419{
2420        struct split_index *si = istate->split_index;
2421        int fd, ret;
2422
2423        fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2424        if (fd < 0) {
2425                hashclr(si->base_sha1);
2426                return do_write_locked_index(istate, lock, flags);
2427        }
2428        move_cache_to_base_index(istate);
2429        ret = do_write_index(si->base, &temporary_sharedindex, 1);
2430        if (ret) {
2431                delete_tempfile(&temporary_sharedindex);
2432                return ret;
2433        }
2434        ret = rename_tempfile(&temporary_sharedindex,
2435                              git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2436        if (!ret) {
2437                hashcpy(si->base_sha1, si->base->sha1);
2438                clean_shared_index_files(sha1_to_hex(si->base->sha1));
2439        }
2440
2441        return ret;
2442}
2443
2444static const int default_max_percent_split_change = 20;
2445
2446static int too_many_not_shared_entries(struct index_state *istate)
2447{
2448        int i, not_shared = 0;
2449        int max_split = git_config_get_max_percent_split_change();
2450
2451        switch (max_split) {
2452        case -1:
2453                /* not or badly configured: use the default value */
2454                max_split = default_max_percent_split_change;
2455                break;
2456        case 0:
2457                return 1; /* 0% means always write a new shared index */
2458        case 100:
2459                return 0; /* 100% means never write a new shared index */
2460        default:
2461                break; /* just use the configured value */
2462        }
2463
2464        /* Count not shared entries */
2465        for (i = 0; i < istate->cache_nr; i++) {
2466                struct cache_entry *ce = istate->cache[i];
2467                if (!ce->index)
2468                        not_shared++;
2469        }
2470
2471        return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2472}
2473
2474int write_locked_index(struct index_state *istate, struct lock_file *lock,
2475                       unsigned flags)
2476{
2477        int new_shared_index, ret;
2478        struct split_index *si = istate->split_index;
2479
2480        if (!si || alternate_index_output ||
2481            (istate->cache_changed & ~EXTMASK)) {
2482                if (si)
2483                        hashclr(si->base_sha1);
2484                return do_write_locked_index(istate, lock, flags);
2485        }
2486
2487        if (getenv("GIT_TEST_SPLIT_INDEX")) {
2488                int v = si->base_sha1[0];
2489                if ((v & 15) < 6)
2490                        istate->cache_changed |= SPLIT_INDEX_ORDERED;
2491        }
2492        if (too_many_not_shared_entries(istate))
2493                istate->cache_changed |= SPLIT_INDEX_ORDERED;
2494
2495        new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2496
2497        if (new_shared_index) {
2498                ret = write_shared_index(istate, lock, flags);
2499                if (ret)
2500                        return ret;
2501        }
2502
2503        ret = write_split_index(istate, lock, flags);
2504
2505        /* Freshen the shared index only if the split-index was written */
2506        if (!ret && !new_shared_index)
2507                freshen_shared_index(sha1_to_hex(si->base_sha1), 1);
2508
2509        return ret;
2510}
2511
2512/*
2513 * Read the index file that is potentially unmerged into given
2514 * index_state, dropping any unmerged entries.  Returns true if
2515 * the index is unmerged.  Callers who want to refuse to work
2516 * from an unmerged state can call this and check its return value,
2517 * instead of calling read_cache().
2518 */
2519int read_index_unmerged(struct index_state *istate)
2520{
2521        int i;
2522        int unmerged = 0;
2523
2524        read_index(istate);
2525        for (i = 0; i < istate->cache_nr; i++) {
2526                struct cache_entry *ce = istate->cache[i];
2527                struct cache_entry *new_ce;
2528                int size, len;
2529
2530                if (!ce_stage(ce))
2531                        continue;
2532                unmerged = 1;
2533                len = ce_namelen(ce);
2534                size = cache_entry_size(len);
2535                new_ce = xcalloc(1, size);
2536                memcpy(new_ce->name, ce->name, len);
2537                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2538                new_ce->ce_namelen = len;
2539                new_ce->ce_mode = ce->ce_mode;
2540                if (add_index_entry(istate, new_ce, 0))
2541                        return error("%s: cannot drop to stage #0",
2542                                     new_ce->name);
2543        }
2544        return unmerged;
2545}
2546
2547/*
2548 * Returns 1 if the path is an "other" path with respect to
2549 * the index; that is, the path is not mentioned in the index at all,
2550 * either as a file, a directory with some files in the index,
2551 * or as an unmerged entry.
2552 *
2553 * We helpfully remove a trailing "/" from directories so that
2554 * the output of read_directory can be used as-is.
2555 */
2556int index_name_is_other(const struct index_state *istate, const char *name,
2557                int namelen)
2558{
2559        int pos;
2560        if (namelen && name[namelen - 1] == '/')
2561                namelen--;
2562        pos = index_name_pos(istate, name, namelen);
2563        if (0 <= pos)
2564                return 0;       /* exact match */
2565        pos = -pos - 1;
2566        if (pos < istate->cache_nr) {
2567                struct cache_entry *ce = istate->cache[pos];
2568                if (ce_namelen(ce) == namelen &&
2569                    !memcmp(ce->name, name, namelen))
2570                        return 0; /* Yup, this one exists unmerged */
2571        }
2572        return 1;
2573}
2574
2575void *read_blob_data_from_index(const struct index_state *istate,
2576                                const char *path, unsigned long *size)
2577{
2578        int pos, len;
2579        unsigned long sz;
2580        enum object_type type;
2581        void *data;
2582
2583        len = strlen(path);
2584        pos = index_name_pos(istate, path, len);
2585        if (pos < 0) {
2586                /*
2587                 * We might be in the middle of a merge, in which
2588                 * case we would read stage #2 (ours).
2589                 */
2590                int i;
2591                for (i = -pos - 1;
2592                     (pos < 0 && i < istate->cache_nr &&
2593                      !strcmp(istate->cache[i]->name, path));
2594                     i++)
2595                        if (ce_stage(istate->cache[i]) == 2)
2596                                pos = i;
2597        }
2598        if (pos < 0)
2599                return NULL;
2600        data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2601        if (!data || type != OBJ_BLOB) {
2602                free(data);
2603                return NULL;
2604        }
2605        if (size)
2606                *size = sz;
2607        return data;
2608}
2609
2610void stat_validity_clear(struct stat_validity *sv)
2611{
2612        FREE_AND_NULL(sv->sd);
2613}
2614
2615int stat_validity_check(struct stat_validity *sv, const char *path)
2616{
2617        struct stat st;
2618
2619        if (stat(path, &st) < 0)
2620                return sv->sd == NULL;
2621        if (!sv->sd)
2622                return 0;
2623        return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2624}
2625
2626void stat_validity_update(struct stat_validity *sv, int fd)
2627{
2628        struct stat st;
2629
2630        if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2631                stat_validity_clear(sv);
2632        else {
2633                if (!sv->sd)
2634                        sv->sd = xcalloc(1, sizeof(struct stat_data));
2635                fill_stat_data(sv->sd, &st);
2636        }
2637}
2638
2639void move_index_extensions(struct index_state *dst, struct index_state *src)
2640{
2641        dst->untracked = src->untracked;
2642        src->untracked = NULL;
2643}