read-cache.con commit Merge branch 'jk/patch-ids-no-merges' into maint (35ca3e5)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 */
   6#define NO_THE_INDEX_COMPATIBILITY_MACROS
   7#include "cache.h"
   8#include "tempfile.h"
   9#include "lockfile.h"
  10#include "cache-tree.h"
  11#include "refs.h"
  12#include "dir.h"
  13#include "tree.h"
  14#include "commit.h"
  15#include "blob.h"
  16#include "resolve-undo.h"
  17#include "strbuf.h"
  18#include "varint.h"
  19#include "split-index.h"
  20#include "utf8.h"
  21
  22/* Mask for the name length in ce_flags in the on-disk index */
  23
  24#define CE_NAMEMASK  (0x0fff)
  25
  26/* Index extensions.
  27 *
  28 * The first letter should be 'A'..'Z' for extensions that are not
  29 * necessary for a correct operation (i.e. optimization data).
  30 * When new extensions are added that _needs_ to be understood in
  31 * order to correctly interpret the index file, pick character that
  32 * is outside the range, to cause the reader to abort.
  33 */
  34
  35#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  36#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  37#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  38#define CACHE_EXT_LINK 0x6c696e6b         /* "link" */
  39#define CACHE_EXT_UNTRACKED 0x554E5452    /* "UNTR" */
  40
  41/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
  42#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
  43                 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
  44                 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
  45
  46struct index_state the_index;
  47static const char *alternate_index_output;
  48
  49static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  50{
  51        istate->cache[nr] = ce;
  52        add_name_hash(istate, ce);
  53}
  54
  55static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  56{
  57        struct cache_entry *old = istate->cache[nr];
  58
  59        replace_index_entry_in_base(istate, old, ce);
  60        remove_name_hash(istate, old);
  61        free(old);
  62        set_index_entry(istate, nr, ce);
  63        ce->ce_flags |= CE_UPDATE_IN_BASE;
  64        istate->cache_changed |= CE_ENTRY_CHANGED;
  65}
  66
  67void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  68{
  69        struct cache_entry *old = istate->cache[nr], *new;
  70        int namelen = strlen(new_name);
  71
  72        new = xmalloc(cache_entry_size(namelen));
  73        copy_cache_entry(new, old);
  74        new->ce_flags &= ~CE_HASHED;
  75        new->ce_namelen = namelen;
  76        new->index = 0;
  77        memcpy(new->name, new_name, namelen + 1);
  78
  79        cache_tree_invalidate_path(istate, old->name);
  80        untracked_cache_remove_from_index(istate, old->name);
  81        remove_index_entry_at(istate, nr);
  82        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  83}
  84
  85void fill_stat_data(struct stat_data *sd, struct stat *st)
  86{
  87        sd->sd_ctime.sec = (unsigned int)st->st_ctime;
  88        sd->sd_mtime.sec = (unsigned int)st->st_mtime;
  89        sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
  90        sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
  91        sd->sd_dev = st->st_dev;
  92        sd->sd_ino = st->st_ino;
  93        sd->sd_uid = st->st_uid;
  94        sd->sd_gid = st->st_gid;
  95        sd->sd_size = st->st_size;
  96}
  97
  98int match_stat_data(const struct stat_data *sd, struct stat *st)
  99{
 100        int changed = 0;
 101
 102        if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
 103                changed |= MTIME_CHANGED;
 104        if (trust_ctime && check_stat &&
 105            sd->sd_ctime.sec != (unsigned int)st->st_ctime)
 106                changed |= CTIME_CHANGED;
 107
 108#ifdef USE_NSEC
 109        if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
 110                changed |= MTIME_CHANGED;
 111        if (trust_ctime && check_stat &&
 112            sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
 113                changed |= CTIME_CHANGED;
 114#endif
 115
 116        if (check_stat) {
 117                if (sd->sd_uid != (unsigned int) st->st_uid ||
 118                        sd->sd_gid != (unsigned int) st->st_gid)
 119                        changed |= OWNER_CHANGED;
 120                if (sd->sd_ino != (unsigned int) st->st_ino)
 121                        changed |= INODE_CHANGED;
 122        }
 123
 124#ifdef USE_STDEV
 125        /*
 126         * st_dev breaks on network filesystems where different
 127         * clients will have different views of what "device"
 128         * the filesystem is on
 129         */
 130        if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
 131                        changed |= INODE_CHANGED;
 132#endif
 133
 134        if (sd->sd_size != (unsigned int) st->st_size)
 135                changed |= DATA_CHANGED;
 136
 137        return changed;
 138}
 139
 140/*
 141 * This only updates the "non-critical" parts of the directory
 142 * cache, ie the parts that aren't tracked by GIT, and only used
 143 * to validate the cache.
 144 */
 145void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 146{
 147        fill_stat_data(&ce->ce_stat_data, st);
 148
 149        if (assume_unchanged)
 150                ce->ce_flags |= CE_VALID;
 151
 152        if (S_ISREG(st->st_mode))
 153                ce_mark_uptodate(ce);
 154}
 155
 156static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
 157{
 158        int match = -1;
 159        int fd = open(ce->name, O_RDONLY);
 160
 161        if (fd >= 0) {
 162                unsigned char sha1[20];
 163                if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
 164                        match = hashcmp(sha1, ce->sha1);
 165                /* index_fd() closed the file descriptor already */
 166        }
 167        return match;
 168}
 169
 170static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
 171{
 172        int match = -1;
 173        void *buffer;
 174        unsigned long size;
 175        enum object_type type;
 176        struct strbuf sb = STRBUF_INIT;
 177
 178        if (strbuf_readlink(&sb, ce->name, expected_size))
 179                return -1;
 180
 181        buffer = read_sha1_file(ce->sha1, &type, &size);
 182        if (buffer) {
 183                if (size == sb.len)
 184                        match = memcmp(buffer, sb.buf, size);
 185                free(buffer);
 186        }
 187        strbuf_release(&sb);
 188        return match;
 189}
 190
 191static int ce_compare_gitlink(const struct cache_entry *ce)
 192{
 193        unsigned char sha1[20];
 194
 195        /*
 196         * We don't actually require that the .git directory
 197         * under GITLINK directory be a valid git directory. It
 198         * might even be missing (in case nobody populated that
 199         * sub-project).
 200         *
 201         * If so, we consider it always to match.
 202         */
 203        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 204                return 0;
 205        return hashcmp(sha1, ce->sha1);
 206}
 207
 208static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
 209{
 210        switch (st->st_mode & S_IFMT) {
 211        case S_IFREG:
 212                if (ce_compare_data(ce, st))
 213                        return DATA_CHANGED;
 214                break;
 215        case S_IFLNK:
 216                if (ce_compare_link(ce, xsize_t(st->st_size)))
 217                        return DATA_CHANGED;
 218                break;
 219        case S_IFDIR:
 220                if (S_ISGITLINK(ce->ce_mode))
 221                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 222        default:
 223                return TYPE_CHANGED;
 224        }
 225        return 0;
 226}
 227
 228static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
 229{
 230        unsigned int changed = 0;
 231
 232        if (ce->ce_flags & CE_REMOVE)
 233                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 234
 235        switch (ce->ce_mode & S_IFMT) {
 236        case S_IFREG:
 237                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 238                /* We consider only the owner x bit to be relevant for
 239                 * "mode changes"
 240                 */
 241                if (trust_executable_bit &&
 242                    (0100 & (ce->ce_mode ^ st->st_mode)))
 243                        changed |= MODE_CHANGED;
 244                break;
 245        case S_IFLNK:
 246                if (!S_ISLNK(st->st_mode) &&
 247                    (has_symlinks || !S_ISREG(st->st_mode)))
 248                        changed |= TYPE_CHANGED;
 249                break;
 250        case S_IFGITLINK:
 251                /* We ignore most of the st_xxx fields for gitlinks */
 252                if (!S_ISDIR(st->st_mode))
 253                        changed |= TYPE_CHANGED;
 254                else if (ce_compare_gitlink(ce))
 255                        changed |= DATA_CHANGED;
 256                return changed;
 257        default:
 258                die("internal error: ce_mode is %o", ce->ce_mode);
 259        }
 260
 261        changed |= match_stat_data(&ce->ce_stat_data, st);
 262
 263        /* Racily smudged entry? */
 264        if (!ce->ce_stat_data.sd_size) {
 265                if (!is_empty_blob_sha1(ce->sha1))
 266                        changed |= DATA_CHANGED;
 267        }
 268
 269        return changed;
 270}
 271
 272static int is_racy_stat(const struct index_state *istate,
 273                        const struct stat_data *sd)
 274{
 275        return (istate->timestamp.sec &&
 276#ifdef USE_NSEC
 277                 /* nanosecond timestamped files can also be racy! */
 278                (istate->timestamp.sec < sd->sd_mtime.sec ||
 279                 (istate->timestamp.sec == sd->sd_mtime.sec &&
 280                  istate->timestamp.nsec <= sd->sd_mtime.nsec))
 281#else
 282                istate->timestamp.sec <= sd->sd_mtime.sec
 283#endif
 284                );
 285}
 286
 287static int is_racy_timestamp(const struct index_state *istate,
 288                             const struct cache_entry *ce)
 289{
 290        return (!S_ISGITLINK(ce->ce_mode) &&
 291                is_racy_stat(istate, &ce->ce_stat_data));
 292}
 293
 294int match_stat_data_racy(const struct index_state *istate,
 295                         const struct stat_data *sd, struct stat *st)
 296{
 297        if (is_racy_stat(istate, sd))
 298                return MTIME_CHANGED;
 299        return match_stat_data(sd, st);
 300}
 301
 302int ie_match_stat(const struct index_state *istate,
 303                  const struct cache_entry *ce, struct stat *st,
 304                  unsigned int options)
 305{
 306        unsigned int changed;
 307        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 308        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 309        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 310
 311        /*
 312         * If it's marked as always valid in the index, it's
 313         * valid whatever the checked-out copy says.
 314         *
 315         * skip-worktree has the same effect with higher precedence
 316         */
 317        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 318                return 0;
 319        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 320                return 0;
 321
 322        /*
 323         * Intent-to-add entries have not been added, so the index entry
 324         * by definition never matches what is in the work tree until it
 325         * actually gets added.
 326         */
 327        if (ce_intent_to_add(ce))
 328                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 329
 330        changed = ce_match_stat_basic(ce, st);
 331
 332        /*
 333         * Within 1 second of this sequence:
 334         *      echo xyzzy >file && git-update-index --add file
 335         * running this command:
 336         *      echo frotz >file
 337         * would give a falsely clean cache entry.  The mtime and
 338         * length match the cache, and other stat fields do not change.
 339         *
 340         * We could detect this at update-index time (the cache entry
 341         * being registered/updated records the same time as "now")
 342         * and delay the return from git-update-index, but that would
 343         * effectively mean we can make at most one commit per second,
 344         * which is not acceptable.  Instead, we check cache entries
 345         * whose mtime are the same as the index file timestamp more
 346         * carefully than others.
 347         */
 348        if (!changed && is_racy_timestamp(istate, ce)) {
 349                if (assume_racy_is_modified)
 350                        changed |= DATA_CHANGED;
 351                else
 352                        changed |= ce_modified_check_fs(ce, st);
 353        }
 354
 355        return changed;
 356}
 357
 358int ie_modified(const struct index_state *istate,
 359                const struct cache_entry *ce,
 360                struct stat *st, unsigned int options)
 361{
 362        int changed, changed_fs;
 363
 364        changed = ie_match_stat(istate, ce, st, options);
 365        if (!changed)
 366                return 0;
 367        /*
 368         * If the mode or type has changed, there's no point in trying
 369         * to refresh the entry - it's not going to match
 370         */
 371        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 372                return changed;
 373
 374        /*
 375         * Immediately after read-tree or update-index --cacheinfo,
 376         * the length field is zero, as we have never even read the
 377         * lstat(2) information once, and we cannot trust DATA_CHANGED
 378         * returned by ie_match_stat() which in turn was returned by
 379         * ce_match_stat_basic() to signal that the filesize of the
 380         * blob changed.  We have to actually go to the filesystem to
 381         * see if the contents match, and if so, should answer "unchanged".
 382         *
 383         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 384         * already has checked the actual HEAD from the filesystem in the
 385         * subproject.  If ie_match_stat() already said it is different,
 386         * then we know it is.
 387         */
 388        if ((changed & DATA_CHANGED) &&
 389            (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
 390                return changed;
 391
 392        changed_fs = ce_modified_check_fs(ce, st);
 393        if (changed_fs)
 394                return changed | changed_fs;
 395        return 0;
 396}
 397
 398int base_name_compare(const char *name1, int len1, int mode1,
 399                      const char *name2, int len2, int mode2)
 400{
 401        unsigned char c1, c2;
 402        int len = len1 < len2 ? len1 : len2;
 403        int cmp;
 404
 405        cmp = memcmp(name1, name2, len);
 406        if (cmp)
 407                return cmp;
 408        c1 = name1[len];
 409        c2 = name2[len];
 410        if (!c1 && S_ISDIR(mode1))
 411                c1 = '/';
 412        if (!c2 && S_ISDIR(mode2))
 413                c2 = '/';
 414        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 415}
 416
 417/*
 418 * df_name_compare() is identical to base_name_compare(), except it
 419 * compares conflicting directory/file entries as equal. Note that
 420 * while a directory name compares as equal to a regular file, they
 421 * then individually compare _differently_ to a filename that has
 422 * a dot after the basename (because '\0' < '.' < '/').
 423 *
 424 * This is used by routines that want to traverse the git namespace
 425 * but then handle conflicting entries together when possible.
 426 */
 427int df_name_compare(const char *name1, int len1, int mode1,
 428                    const char *name2, int len2, int mode2)
 429{
 430        int len = len1 < len2 ? len1 : len2, cmp;
 431        unsigned char c1, c2;
 432
 433        cmp = memcmp(name1, name2, len);
 434        if (cmp)
 435                return cmp;
 436        /* Directories and files compare equal (same length, same name) */
 437        if (len1 == len2)
 438                return 0;
 439        c1 = name1[len];
 440        if (!c1 && S_ISDIR(mode1))
 441                c1 = '/';
 442        c2 = name2[len];
 443        if (!c2 && S_ISDIR(mode2))
 444                c2 = '/';
 445        if (c1 == '/' && !c2)
 446                return 0;
 447        if (c2 == '/' && !c1)
 448                return 0;
 449        return c1 - c2;
 450}
 451
 452int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
 453{
 454        size_t min_len = (len1 < len2) ? len1 : len2;
 455        int cmp = memcmp(name1, name2, min_len);
 456        if (cmp)
 457                return cmp;
 458        if (len1 < len2)
 459                return -1;
 460        if (len1 > len2)
 461                return 1;
 462        return 0;
 463}
 464
 465int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 466{
 467        int cmp;
 468
 469        cmp = name_compare(name1, len1, name2, len2);
 470        if (cmp)
 471                return cmp;
 472
 473        if (stage1 < stage2)
 474                return -1;
 475        if (stage1 > stage2)
 476                return 1;
 477        return 0;
 478}
 479
 480static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 481{
 482        int first, last;
 483
 484        first = 0;
 485        last = istate->cache_nr;
 486        while (last > first) {
 487                int next = (last + first) >> 1;
 488                struct cache_entry *ce = istate->cache[next];
 489                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 490                if (!cmp)
 491                        return next;
 492                if (cmp < 0) {
 493                        last = next;
 494                        continue;
 495                }
 496                first = next+1;
 497        }
 498        return -first-1;
 499}
 500
 501int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 502{
 503        return index_name_stage_pos(istate, name, namelen, 0);
 504}
 505
 506/* Remove entry, return true if there are more entries to go.. */
 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->sha1, sha1);
 628}
 629
 630int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags, int force_mode)
 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        if (S_ISREG(st_mode) && force_mode)
 660                ce->ce_mode = create_ce_mode(force_mode);
 661        else if (trust_executable_bit && has_symlinks)
 662                ce->ce_mode = create_ce_mode(st_mode);
 663        else {
 664                /* If there is an existing entry, pick the mode bits and type
 665                 * from it, otherwise assume unexecutable regular file.
 666                 */
 667                struct cache_entry *ent;
 668                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 669
 670                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 671                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 672        }
 673
 674        /* When core.ignorecase=true, determine if a directory of the same name but differing
 675         * case already exists within the Git repository.  If it does, ensure the directory
 676         * case of the file being added to the repository matches (is folded into) the existing
 677         * entry's directory case.
 678         */
 679        if (ignore_case) {
 680                adjust_dirname_case(istate, ce->name);
 681        }
 682
 683        alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 684        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 685                /* Nothing changed, really */
 686                if (!S_ISGITLINK(alias->ce_mode))
 687                        ce_mark_uptodate(alias);
 688                alias->ce_flags |= CE_ADDED;
 689
 690                free(ce);
 691                return 0;
 692        }
 693        if (!intent_only) {
 694                if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT)) {
 695                        free(ce);
 696                        return error("unable to index file %s", path);
 697                }
 698        } else
 699                set_object_name_for_intent_to_add_entry(ce);
 700
 701        if (ignore_case && alias && different_name(ce, alias))
 702                ce = create_alias_ce(istate, ce, alias);
 703        ce->ce_flags |= CE_ADDED;
 704
 705        /* It was suspected to be racily clean, but it turns out to be Ok */
 706        was_same = (alias &&
 707                    !ce_stage(alias) &&
 708                    !hashcmp(alias->sha1, ce->sha1) &&
 709                    ce->ce_mode == alias->ce_mode);
 710
 711        if (pretend)
 712                free(ce);
 713        else if (add_index_entry(istate, ce, add_option)) {
 714                free(ce);
 715                return error("unable to add %s to index", path);
 716        }
 717        if (verbose && !was_same)
 718                printf("add '%s'\n", path);
 719        return 0;
 720}
 721
 722int add_file_to_index(struct index_state *istate, const char *path,
 723        int flags, int force_mode)
 724{
 725        struct stat st;
 726        if (lstat(path, &st))
 727                die_errno("unable to stat '%s'", path);
 728        return add_to_index(istate, path, &st, flags, force_mode);
 729}
 730
 731struct cache_entry *make_cache_entry(unsigned int mode,
 732                const unsigned char *sha1, const char *path, int stage,
 733                unsigned int refresh_options)
 734{
 735        int size, len;
 736        struct cache_entry *ce, *ret;
 737
 738        if (!verify_path(path)) {
 739                error("Invalid path '%s'", path);
 740                return NULL;
 741        }
 742
 743        len = strlen(path);
 744        size = cache_entry_size(len);
 745        ce = xcalloc(1, size);
 746
 747        hashcpy(ce->sha1, sha1);
 748        memcpy(ce->name, path, len);
 749        ce->ce_flags = create_ce_flags(stage);
 750        ce->ce_namelen = len;
 751        ce->ce_mode = create_ce_mode(mode);
 752
 753        ret = refresh_cache_entry(ce, refresh_options);
 754        if (ret != ce)
 755                free(ce);
 756        return ret;
 757}
 758
 759int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
 760{
 761        int len = ce_namelen(a);
 762        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 763}
 764
 765/*
 766 * We fundamentally don't like some paths: we don't want
 767 * dot or dot-dot anywhere, and for obvious reasons don't
 768 * want to recurse into ".git" either.
 769 *
 770 * Also, we don't want double slashes or slashes at the
 771 * end that can make pathnames ambiguous.
 772 */
 773static int verify_dotfile(const char *rest)
 774{
 775        /*
 776         * The first character was '.', but that
 777         * has already been discarded, we now test
 778         * the rest.
 779         */
 780
 781        /* "." is not allowed */
 782        if (*rest == '\0' || is_dir_sep(*rest))
 783                return 0;
 784
 785        switch (*rest) {
 786        /*
 787         * ".git" followed by  NUL or slash is bad. This
 788         * shares the path end test with the ".." case.
 789         */
 790        case 'g':
 791        case 'G':
 792                if (rest[1] != 'i' && rest[1] != 'I')
 793                        break;
 794                if (rest[2] != 't' && rest[2] != 'T')
 795                        break;
 796                rest += 2;
 797        /* fallthrough */
 798        case '.':
 799                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 800                        return 0;
 801        }
 802        return 1;
 803}
 804
 805int verify_path(const char *path)
 806{
 807        char c;
 808
 809        if (has_dos_drive_prefix(path))
 810                return 0;
 811
 812        goto inside;
 813        for (;;) {
 814                if (!c)
 815                        return 1;
 816                if (is_dir_sep(c)) {
 817inside:
 818                        if (protect_hfs && is_hfs_dotgit(path))
 819                                return 0;
 820                        if (protect_ntfs && is_ntfs_dotgit(path))
 821                                return 0;
 822                        c = *path++;
 823                        if ((c == '.' && !verify_dotfile(path)) ||
 824                            is_dir_sep(c) || c == '\0')
 825                                return 0;
 826                }
 827                c = *path++;
 828        }
 829}
 830
 831/*
 832 * Do we have another file that has the beginning components being a
 833 * proper superset of the name we're trying to add?
 834 */
 835static int has_file_name(struct index_state *istate,
 836                         const struct cache_entry *ce, int pos, int ok_to_replace)
 837{
 838        int retval = 0;
 839        int len = ce_namelen(ce);
 840        int stage = ce_stage(ce);
 841        const char *name = ce->name;
 842
 843        while (pos < istate->cache_nr) {
 844                struct cache_entry *p = istate->cache[pos++];
 845
 846                if (len >= ce_namelen(p))
 847                        break;
 848                if (memcmp(name, p->name, len))
 849                        break;
 850                if (ce_stage(p) != stage)
 851                        continue;
 852                if (p->name[len] != '/')
 853                        continue;
 854                if (p->ce_flags & CE_REMOVE)
 855                        continue;
 856                retval = -1;
 857                if (!ok_to_replace)
 858                        break;
 859                remove_index_entry_at(istate, --pos);
 860        }
 861        return retval;
 862}
 863
 864/*
 865 * Do we have another file with a pathname that is a proper
 866 * subset of the name we're trying to add?
 867 */
 868static int has_dir_name(struct index_state *istate,
 869                        const struct cache_entry *ce, int pos, int ok_to_replace)
 870{
 871        int retval = 0;
 872        int stage = ce_stage(ce);
 873        const char *name = ce->name;
 874        const char *slash = name + ce_namelen(ce);
 875
 876        for (;;) {
 877                int len;
 878
 879                for (;;) {
 880                        if (*--slash == '/')
 881                                break;
 882                        if (slash <= ce->name)
 883                                return retval;
 884                }
 885                len = slash - name;
 886
 887                pos = index_name_stage_pos(istate, name, len, stage);
 888                if (pos >= 0) {
 889                        /*
 890                         * Found one, but not so fast.  This could
 891                         * be a marker that says "I was here, but
 892                         * I am being removed".  Such an entry is
 893                         * not a part of the resulting tree, and
 894                         * it is Ok to have a directory at the same
 895                         * path.
 896                         */
 897                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 898                                retval = -1;
 899                                if (!ok_to_replace)
 900                                        break;
 901                                remove_index_entry_at(istate, pos);
 902                                continue;
 903                        }
 904                }
 905                else
 906                        pos = -pos-1;
 907
 908                /*
 909                 * Trivial optimization: if we find an entry that
 910                 * already matches the sub-directory, then we know
 911                 * we're ok, and we can exit.
 912                 */
 913                while (pos < istate->cache_nr) {
 914                        struct cache_entry *p = istate->cache[pos];
 915                        if ((ce_namelen(p) <= len) ||
 916                            (p->name[len] != '/') ||
 917                            memcmp(p->name, name, len))
 918                                break; /* not our subdirectory */
 919                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 920                                /*
 921                                 * p is at the same stage as our entry, and
 922                                 * is a subdirectory of what we are looking
 923                                 * at, so we cannot have conflicts at our
 924                                 * level or anything shorter.
 925                                 */
 926                                return retval;
 927                        pos++;
 928                }
 929        }
 930        return retval;
 931}
 932
 933/* We may be in a situation where we already have path/file and path
 934 * is being added, or we already have path and path/file is being
 935 * added.  Either one would result in a nonsense tree that has path
 936 * twice when git-write-tree tries to write it out.  Prevent it.
 937 *
 938 * If ok-to-replace is specified, we remove the conflicting entries
 939 * from the cache so the caller should recompute the insert position.
 940 * When this happens, we return non-zero.
 941 */
 942static int check_file_directory_conflict(struct index_state *istate,
 943                                         const struct cache_entry *ce,
 944                                         int pos, int ok_to_replace)
 945{
 946        int retval;
 947
 948        /*
 949         * When ce is an "I am going away" entry, we allow it to be added
 950         */
 951        if (ce->ce_flags & CE_REMOVE)
 952                return 0;
 953
 954        /*
 955         * We check if the path is a sub-path of a subsequent pathname
 956         * first, since removing those will not change the position
 957         * in the array.
 958         */
 959        retval = has_file_name(istate, ce, pos, ok_to_replace);
 960
 961        /*
 962         * Then check if the path might have a clashing sub-directory
 963         * before it.
 964         */
 965        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 966}
 967
 968static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 969{
 970        int pos;
 971        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 972        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 973        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 974        int new_only = option & ADD_CACHE_NEW_ONLY;
 975
 976        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
 977                cache_tree_invalidate_path(istate, ce->name);
 978        pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 979
 980        /* existing match? Just replace it. */
 981        if (pos >= 0) {
 982                if (!new_only)
 983                        replace_index_entry(istate, pos, ce);
 984                return 0;
 985        }
 986        pos = -pos-1;
 987
 988        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
 989                untracked_cache_add_to_index(istate, ce->name);
 990
 991        /*
 992         * Inserting a merged entry ("stage 0") into the index
 993         * will always replace all non-merged entries..
 994         */
 995        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 996                while (ce_same_name(istate->cache[pos], ce)) {
 997                        ok_to_add = 1;
 998                        if (!remove_index_entry_at(istate, pos))
 999                                break;
1000                }
1001        }
1002
1003        if (!ok_to_add)
1004                return -1;
1005        if (!verify_path(ce->name))
1006                return error("Invalid path '%s'", ce->name);
1007
1008        if (!skip_df_check &&
1009            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1010                if (!ok_to_replace)
1011                        return error("'%s' appears as both a file and as a directory",
1012                                     ce->name);
1013                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1014                pos = -pos-1;
1015        }
1016        return pos + 1;
1017}
1018
1019int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1020{
1021        int pos;
1022
1023        if (option & ADD_CACHE_JUST_APPEND)
1024                pos = istate->cache_nr;
1025        else {
1026                int ret;
1027                ret = add_index_entry_with_check(istate, ce, option);
1028                if (ret <= 0)
1029                        return ret;
1030                pos = ret - 1;
1031        }
1032
1033        /* Make sure the array is big enough .. */
1034        ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1035
1036        /* Add it in.. */
1037        istate->cache_nr++;
1038        if (istate->cache_nr > pos + 1)
1039                memmove(istate->cache + pos + 1,
1040                        istate->cache + pos,
1041                        (istate->cache_nr - pos - 1) * sizeof(ce));
1042        set_index_entry(istate, pos, ce);
1043        istate->cache_changed |= CE_ENTRY_ADDED;
1044        return 0;
1045}
1046
1047/*
1048 * "refresh" does not calculate a new sha1 file or bring the
1049 * cache up-to-date for mode/content changes. But what it
1050 * _does_ do is to "re-match" the stat information of a file
1051 * with the cache, so that you can refresh the cache for a
1052 * file that hasn't been changed but where the stat entry is
1053 * out of date.
1054 *
1055 * For example, you'd want to do this after doing a "git-read-tree",
1056 * to link up the stat cache details with the proper files.
1057 */
1058static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1059                                             struct cache_entry *ce,
1060                                             unsigned int options, int *err,
1061                                             int *changed_ret)
1062{
1063        struct stat st;
1064        struct cache_entry *updated;
1065        int changed, size;
1066        int refresh = options & CE_MATCH_REFRESH;
1067        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1068        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1069        int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1070
1071        if (!refresh || ce_uptodate(ce))
1072                return ce;
1073
1074        /*
1075         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1076         * that the change to the work tree does not matter and told
1077         * us not to worry.
1078         */
1079        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1080                ce_mark_uptodate(ce);
1081                return ce;
1082        }
1083        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1084                ce_mark_uptodate(ce);
1085                return ce;
1086        }
1087
1088        if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1089                if (ignore_missing)
1090                        return ce;
1091                if (err)
1092                        *err = ENOENT;
1093                return NULL;
1094        }
1095
1096        if (lstat(ce->name, &st) < 0) {
1097                if (ignore_missing && errno == ENOENT)
1098                        return ce;
1099                if (err)
1100                        *err = errno;
1101                return NULL;
1102        }
1103
1104        changed = ie_match_stat(istate, ce, &st, options);
1105        if (changed_ret)
1106                *changed_ret = changed;
1107        if (!changed) {
1108                /*
1109                 * The path is unchanged.  If we were told to ignore
1110                 * valid bit, then we did the actual stat check and
1111                 * found that the entry is unmodified.  If the entry
1112                 * is not marked VALID, this is the place to mark it
1113                 * valid again, under "assume unchanged" mode.
1114                 */
1115                if (ignore_valid && assume_unchanged &&
1116                    !(ce->ce_flags & CE_VALID))
1117                        ; /* mark this one VALID again */
1118                else {
1119                        /*
1120                         * We do not mark the index itself "modified"
1121                         * because CE_UPTODATE flag is in-core only;
1122                         * we are not going to write this change out.
1123                         */
1124                        if (!S_ISGITLINK(ce->ce_mode))
1125                                ce_mark_uptodate(ce);
1126                        return ce;
1127                }
1128        }
1129
1130        if (ie_modified(istate, ce, &st, options)) {
1131                if (err)
1132                        *err = EINVAL;
1133                return NULL;
1134        }
1135
1136        size = ce_size(ce);
1137        updated = xmalloc(size);
1138        memcpy(updated, ce, size);
1139        fill_stat_cache_info(updated, &st);
1140        /*
1141         * If ignore_valid is not set, we should leave CE_VALID bit
1142         * alone.  Otherwise, paths marked with --no-assume-unchanged
1143         * (i.e. things to be edited) will reacquire CE_VALID bit
1144         * automatically, which is not really what we want.
1145         */
1146        if (!ignore_valid && assume_unchanged &&
1147            !(ce->ce_flags & CE_VALID))
1148                updated->ce_flags &= ~CE_VALID;
1149
1150        /* istate->cache_changed is updated in the caller */
1151        return updated;
1152}
1153
1154static void show_file(const char * fmt, const char * name, int in_porcelain,
1155                      int * first, const char *header_msg)
1156{
1157        if (in_porcelain && *first && header_msg) {
1158                printf("%s\n", header_msg);
1159                *first = 0;
1160        }
1161        printf(fmt, name);
1162}
1163
1164int refresh_index(struct index_state *istate, unsigned int flags,
1165                  const struct pathspec *pathspec,
1166                  char *seen, const char *header_msg)
1167{
1168        int i;
1169        int has_errors = 0;
1170        int really = (flags & REFRESH_REALLY) != 0;
1171        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1172        int quiet = (flags & REFRESH_QUIET) != 0;
1173        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1174        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1175        int first = 1;
1176        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1177        unsigned int options = (CE_MATCH_REFRESH |
1178                                (really ? CE_MATCH_IGNORE_VALID : 0) |
1179                                (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1180        const char *modified_fmt;
1181        const char *deleted_fmt;
1182        const char *typechange_fmt;
1183        const char *added_fmt;
1184        const char *unmerged_fmt;
1185
1186        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1187        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1188        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1189        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1190        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1191        for (i = 0; i < istate->cache_nr; i++) {
1192                struct cache_entry *ce, *new;
1193                int cache_errno = 0;
1194                int changed = 0;
1195                int filtered = 0;
1196
1197                ce = istate->cache[i];
1198                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1199                        continue;
1200
1201                if (pathspec && !ce_path_match(ce, pathspec, seen))
1202                        filtered = 1;
1203
1204                if (ce_stage(ce)) {
1205                        while ((i < istate->cache_nr) &&
1206                               ! strcmp(istate->cache[i]->name, ce->name))
1207                                i++;
1208                        i--;
1209                        if (allow_unmerged)
1210                                continue;
1211                        if (!filtered)
1212                                show_file(unmerged_fmt, ce->name, in_porcelain,
1213                                          &first, header_msg);
1214                        has_errors = 1;
1215                        continue;
1216                }
1217
1218                if (filtered)
1219                        continue;
1220
1221                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1222                if (new == ce)
1223                        continue;
1224                if (!new) {
1225                        const char *fmt;
1226
1227                        if (really && cache_errno == EINVAL) {
1228                                /* If we are doing --really-refresh that
1229                                 * means the index is not valid anymore.
1230                                 */
1231                                ce->ce_flags &= ~CE_VALID;
1232                                ce->ce_flags |= CE_UPDATE_IN_BASE;
1233                                istate->cache_changed |= CE_ENTRY_CHANGED;
1234                        }
1235                        if (quiet)
1236                                continue;
1237
1238                        if (cache_errno == ENOENT)
1239                                fmt = deleted_fmt;
1240                        else if (ce_intent_to_add(ce))
1241                                fmt = added_fmt; /* must be before other checks */
1242                        else if (changed & TYPE_CHANGED)
1243                                fmt = typechange_fmt;
1244                        else
1245                                fmt = modified_fmt;
1246                        show_file(fmt,
1247                                  ce->name, in_porcelain, &first, header_msg);
1248                        has_errors = 1;
1249                        continue;
1250                }
1251
1252                replace_index_entry(istate, i, new);
1253        }
1254        return has_errors;
1255}
1256
1257struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1258                                               unsigned int options)
1259{
1260        return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1261}
1262
1263
1264/*****************************************************************
1265 * Index File I/O
1266 *****************************************************************/
1267
1268#define INDEX_FORMAT_DEFAULT 3
1269
1270static unsigned int get_index_format_default(void)
1271{
1272        char *envversion = getenv("GIT_INDEX_VERSION");
1273        char *endp;
1274        int value;
1275        unsigned int version = INDEX_FORMAT_DEFAULT;
1276
1277        if (!envversion) {
1278                if (!git_config_get_int("index.version", &value))
1279                        version = value;
1280                if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1281                        warning(_("index.version set, but the value is invalid.\n"
1282                                  "Using version %i"), INDEX_FORMAT_DEFAULT);
1283                        return INDEX_FORMAT_DEFAULT;
1284                }
1285                return version;
1286        }
1287
1288        version = strtoul(envversion, &endp, 10);
1289        if (*endp ||
1290            version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1291                warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1292                          "Using version %i"), INDEX_FORMAT_DEFAULT);
1293                version = INDEX_FORMAT_DEFAULT;
1294        }
1295        return version;
1296}
1297
1298/*
1299 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1300 * Again - this is just a (very strong in practice) heuristic that
1301 * the inode hasn't changed.
1302 *
1303 * We save the fields in big-endian order to allow using the
1304 * index file over NFS transparently.
1305 */
1306struct ondisk_cache_entry {
1307        struct cache_time ctime;
1308        struct cache_time mtime;
1309        uint32_t dev;
1310        uint32_t ino;
1311        uint32_t mode;
1312        uint32_t uid;
1313        uint32_t gid;
1314        uint32_t size;
1315        unsigned char sha1[20];
1316        uint16_t flags;
1317        char name[FLEX_ARRAY]; /* more */
1318};
1319
1320/*
1321 * This struct is used when CE_EXTENDED bit is 1
1322 * The struct must match ondisk_cache_entry exactly from
1323 * ctime till flags
1324 */
1325struct ondisk_cache_entry_extended {
1326        struct cache_time ctime;
1327        struct cache_time mtime;
1328        uint32_t dev;
1329        uint32_t ino;
1330        uint32_t mode;
1331        uint32_t uid;
1332        uint32_t gid;
1333        uint32_t size;
1334        unsigned char sha1[20];
1335        uint16_t flags;
1336        uint16_t flags2;
1337        char name[FLEX_ARRAY]; /* more */
1338};
1339
1340/* These are only used for v3 or lower */
1341#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1342#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1343#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1344#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1345                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1346                            ondisk_cache_entry_size(ce_namelen(ce)))
1347
1348static int verify_hdr(struct cache_header *hdr, unsigned long size)
1349{
1350        git_SHA_CTX c;
1351        unsigned char sha1[20];
1352        int hdr_version;
1353
1354        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1355                return error("bad signature");
1356        hdr_version = ntohl(hdr->hdr_version);
1357        if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1358                return error("bad index version %d", hdr_version);
1359        git_SHA1_Init(&c);
1360        git_SHA1_Update(&c, hdr, size - 20);
1361        git_SHA1_Final(sha1, &c);
1362        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1363                return error("bad index file sha1 signature");
1364        return 0;
1365}
1366
1367static int read_index_extension(struct index_state *istate,
1368                                const char *ext, void *data, unsigned long sz)
1369{
1370        switch (CACHE_EXT(ext)) {
1371        case CACHE_EXT_TREE:
1372                istate->cache_tree = cache_tree_read(data, sz);
1373                break;
1374        case CACHE_EXT_RESOLVE_UNDO:
1375                istate->resolve_undo = resolve_undo_read(data, sz);
1376                break;
1377        case CACHE_EXT_LINK:
1378                if (read_link_extension(istate, data, sz))
1379                        return -1;
1380                break;
1381        case CACHE_EXT_UNTRACKED:
1382                istate->untracked = read_untracked_extension(data, sz);
1383                break;
1384        default:
1385                if (*ext < 'A' || 'Z' < *ext)
1386                        return error("index uses %.4s extension, which we do not understand",
1387                                     ext);
1388                fprintf(stderr, "ignoring %.4s extension\n", ext);
1389                break;
1390        }
1391        return 0;
1392}
1393
1394int hold_locked_index(struct lock_file *lk, int die_on_error)
1395{
1396        return hold_lock_file_for_update(lk, get_index_file(),
1397                                         die_on_error
1398                                         ? LOCK_DIE_ON_ERROR
1399                                         : 0);
1400}
1401
1402int read_index(struct index_state *istate)
1403{
1404        return read_index_from(istate, get_index_file());
1405}
1406
1407static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1408                                                   unsigned int flags,
1409                                                   const char *name,
1410                                                   size_t len)
1411{
1412        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1413
1414        ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1415        ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1416        ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1417        ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1418        ce->ce_stat_data.sd_dev   = get_be32(&ondisk->dev);
1419        ce->ce_stat_data.sd_ino   = get_be32(&ondisk->ino);
1420        ce->ce_mode  = get_be32(&ondisk->mode);
1421        ce->ce_stat_data.sd_uid   = get_be32(&ondisk->uid);
1422        ce->ce_stat_data.sd_gid   = get_be32(&ondisk->gid);
1423        ce->ce_stat_data.sd_size  = get_be32(&ondisk->size);
1424        ce->ce_flags = flags & ~CE_NAMEMASK;
1425        ce->ce_namelen = len;
1426        ce->index = 0;
1427        hashcpy(ce->sha1, ondisk->sha1);
1428        memcpy(ce->name, name, len);
1429        ce->name[len] = '\0';
1430        return ce;
1431}
1432
1433/*
1434 * Adjacent cache entries tend to share the leading paths, so it makes
1435 * sense to only store the differences in later entries.  In the v4
1436 * on-disk format of the index, each on-disk cache entry stores the
1437 * number of bytes to be stripped from the end of the previous name,
1438 * and the bytes to append to the result, to come up with its name.
1439 */
1440static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1441{
1442        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1443        size_t len = decode_varint(&cp);
1444
1445        if (name->len < len)
1446                die("malformed name field in the index");
1447        strbuf_remove(name, name->len - len, len);
1448        for (ep = cp; *ep; ep++)
1449                ; /* find the end */
1450        strbuf_add(name, cp, ep - cp);
1451        return (const char *)ep + 1 - cp_;
1452}
1453
1454static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1455                                            unsigned long *ent_size,
1456                                            struct strbuf *previous_name)
1457{
1458        struct cache_entry *ce;
1459        size_t len;
1460        const char *name;
1461        unsigned int flags;
1462
1463        /* On-disk flags are just 16 bits */
1464        flags = get_be16(&ondisk->flags);
1465        len = flags & CE_NAMEMASK;
1466
1467        if (flags & CE_EXTENDED) {
1468                struct ondisk_cache_entry_extended *ondisk2;
1469                int extended_flags;
1470                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1471                extended_flags = get_be16(&ondisk2->flags2) << 16;
1472                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1473                if (extended_flags & ~CE_EXTENDED_FLAGS)
1474                        die("Unknown index entry format %08x", extended_flags);
1475                flags |= extended_flags;
1476                name = ondisk2->name;
1477        }
1478        else
1479                name = ondisk->name;
1480
1481        if (!previous_name) {
1482                /* v3 and earlier */
1483                if (len == CE_NAMEMASK)
1484                        len = strlen(name);
1485                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1486
1487                *ent_size = ondisk_ce_size(ce);
1488        } else {
1489                unsigned long consumed;
1490                consumed = expand_name_field(previous_name, name);
1491                ce = cache_entry_from_ondisk(ondisk, flags,
1492                                             previous_name->buf,
1493                                             previous_name->len);
1494
1495                *ent_size = (name - ((char *)ondisk)) + consumed;
1496        }
1497        return ce;
1498}
1499
1500static void check_ce_order(struct index_state *istate)
1501{
1502        unsigned int i;
1503
1504        for (i = 1; i < istate->cache_nr; i++) {
1505                struct cache_entry *ce = istate->cache[i - 1];
1506                struct cache_entry *next_ce = istate->cache[i];
1507                int name_compare = strcmp(ce->name, next_ce->name);
1508
1509                if (0 < name_compare)
1510                        die("unordered stage entries in index");
1511                if (!name_compare) {
1512                        if (!ce_stage(ce))
1513                                die("multiple stage entries for merged file '%s'",
1514                                    ce->name);
1515                        if (ce_stage(ce) > ce_stage(next_ce))
1516                                die("unordered stage entries for '%s'",
1517                                    ce->name);
1518                }
1519        }
1520}
1521
1522static void tweak_untracked_cache(struct index_state *istate)
1523{
1524        switch (git_config_get_untracked_cache()) {
1525        case -1: /* keep: do nothing */
1526                break;
1527        case 0: /* false */
1528                remove_untracked_cache(istate);
1529                break;
1530        case 1: /* true */
1531                add_untracked_cache(istate);
1532                break;
1533        default: /* unknown value: do nothing */
1534                break;
1535        }
1536}
1537
1538static void post_read_index_from(struct index_state *istate)
1539{
1540        check_ce_order(istate);
1541        tweak_untracked_cache(istate);
1542}
1543
1544/* remember to discard_cache() before reading a different cache! */
1545int do_read_index(struct index_state *istate, const char *path, int must_exist)
1546{
1547        int fd, i;
1548        struct stat st;
1549        unsigned long src_offset;
1550        struct cache_header *hdr;
1551        void *mmap;
1552        size_t mmap_size;
1553        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1554
1555        if (istate->initialized)
1556                return istate->cache_nr;
1557
1558        istate->timestamp.sec = 0;
1559        istate->timestamp.nsec = 0;
1560        fd = open(path, O_RDONLY);
1561        if (fd < 0) {
1562                if (!must_exist && errno == ENOENT)
1563                        return 0;
1564                die_errno("%s: index file open failed", path);
1565        }
1566
1567        if (fstat(fd, &st))
1568                die_errno("cannot stat the open index");
1569
1570        mmap_size = xsize_t(st.st_size);
1571        if (mmap_size < sizeof(struct cache_header) + 20)
1572                die("index file smaller than expected");
1573
1574        mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1575        if (mmap == MAP_FAILED)
1576                die_errno("unable to map index file");
1577        close(fd);
1578
1579        hdr = mmap;
1580        if (verify_hdr(hdr, mmap_size) < 0)
1581                goto unmap;
1582
1583        hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1584        istate->version = ntohl(hdr->hdr_version);
1585        istate->cache_nr = ntohl(hdr->hdr_entries);
1586        istate->cache_alloc = alloc_nr(istate->cache_nr);
1587        istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1588        istate->initialized = 1;
1589
1590        if (istate->version == 4)
1591                previous_name = &previous_name_buf;
1592        else
1593                previous_name = NULL;
1594
1595        src_offset = sizeof(*hdr);
1596        for (i = 0; i < istate->cache_nr; i++) {
1597                struct ondisk_cache_entry *disk_ce;
1598                struct cache_entry *ce;
1599                unsigned long consumed;
1600
1601                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1602                ce = create_from_disk(disk_ce, &consumed, previous_name);
1603                set_index_entry(istate, i, ce);
1604
1605                src_offset += consumed;
1606        }
1607        strbuf_release(&previous_name_buf);
1608        istate->timestamp.sec = st.st_mtime;
1609        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1610
1611        while (src_offset <= mmap_size - 20 - 8) {
1612                /* After an array of active_nr index entries,
1613                 * there can be arbitrary number of extended
1614                 * sections, each of which is prefixed with
1615                 * extension name (4-byte) and section length
1616                 * in 4-byte network byte order.
1617                 */
1618                uint32_t extsize;
1619                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1620                extsize = ntohl(extsize);
1621                if (read_index_extension(istate,
1622                                         (const char *) mmap + src_offset,
1623                                         (char *) mmap + src_offset + 8,
1624                                         extsize) < 0)
1625                        goto unmap;
1626                src_offset += 8;
1627                src_offset += extsize;
1628        }
1629        munmap(mmap, mmap_size);
1630        return istate->cache_nr;
1631
1632unmap:
1633        munmap(mmap, mmap_size);
1634        die("index file corrupt");
1635}
1636
1637int read_index_from(struct index_state *istate, const char *path)
1638{
1639        struct split_index *split_index;
1640        int ret;
1641
1642        /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1643        if (istate->initialized)
1644                return istate->cache_nr;
1645
1646        ret = do_read_index(istate, path, 0);
1647
1648        split_index = istate->split_index;
1649        if (!split_index || is_null_sha1(split_index->base_sha1)) {
1650                post_read_index_from(istate);
1651                return ret;
1652        }
1653
1654        if (split_index->base)
1655                discard_index(split_index->base);
1656        else
1657                split_index->base = xcalloc(1, sizeof(*split_index->base));
1658        ret = do_read_index(split_index->base,
1659                            git_path("sharedindex.%s",
1660                                     sha1_to_hex(split_index->base_sha1)), 1);
1661        if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1662                die("broken index, expect %s in %s, got %s",
1663                    sha1_to_hex(split_index->base_sha1),
1664                    git_path("sharedindex.%s",
1665                             sha1_to_hex(split_index->base_sha1)),
1666                    sha1_to_hex(split_index->base->sha1));
1667        merge_base_index(istate);
1668        post_read_index_from(istate);
1669        return ret;
1670}
1671
1672int is_index_unborn(struct index_state *istate)
1673{
1674        return (!istate->cache_nr && !istate->timestamp.sec);
1675}
1676
1677int discard_index(struct index_state *istate)
1678{
1679        int i;
1680
1681        for (i = 0; i < istate->cache_nr; i++) {
1682                if (istate->cache[i]->index &&
1683                    istate->split_index &&
1684                    istate->split_index->base &&
1685                    istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1686                    istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1687                        continue;
1688                free(istate->cache[i]);
1689        }
1690        resolve_undo_clear_index(istate);
1691        istate->cache_nr = 0;
1692        istate->cache_changed = 0;
1693        istate->timestamp.sec = 0;
1694        istate->timestamp.nsec = 0;
1695        free_name_hash(istate);
1696        cache_tree_free(&(istate->cache_tree));
1697        istate->initialized = 0;
1698        free(istate->cache);
1699        istate->cache = NULL;
1700        istate->cache_alloc = 0;
1701        discard_split_index(istate);
1702        free_untracked_cache(istate->untracked);
1703        istate->untracked = NULL;
1704        return 0;
1705}
1706
1707int unmerged_index(const struct index_state *istate)
1708{
1709        int i;
1710        for (i = 0; i < istate->cache_nr; i++) {
1711                if (ce_stage(istate->cache[i]))
1712                        return 1;
1713        }
1714        return 0;
1715}
1716
1717#define WRITE_BUFFER_SIZE 8192
1718static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1719static unsigned long write_buffer_len;
1720
1721static int ce_write_flush(git_SHA_CTX *context, int fd)
1722{
1723        unsigned int buffered = write_buffer_len;
1724        if (buffered) {
1725                git_SHA1_Update(context, write_buffer, buffered);
1726                if (write_in_full(fd, write_buffer, buffered) != buffered)
1727                        return -1;
1728                write_buffer_len = 0;
1729        }
1730        return 0;
1731}
1732
1733static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1734{
1735        while (len) {
1736                unsigned int buffered = write_buffer_len;
1737                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1738                if (partial > len)
1739                        partial = len;
1740                memcpy(write_buffer + buffered, data, partial);
1741                buffered += partial;
1742                if (buffered == WRITE_BUFFER_SIZE) {
1743                        write_buffer_len = buffered;
1744                        if (ce_write_flush(context, fd))
1745                                return -1;
1746                        buffered = 0;
1747                }
1748                write_buffer_len = buffered;
1749                len -= partial;
1750                data = (char *) data + partial;
1751        }
1752        return 0;
1753}
1754
1755static int write_index_ext_header(git_SHA_CTX *context, int fd,
1756                                  unsigned int ext, unsigned int sz)
1757{
1758        ext = htonl(ext);
1759        sz = htonl(sz);
1760        return ((ce_write(context, fd, &ext, 4) < 0) ||
1761                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1762}
1763
1764static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1765{
1766        unsigned int left = write_buffer_len;
1767
1768        if (left) {
1769                write_buffer_len = 0;
1770                git_SHA1_Update(context, write_buffer, left);
1771        }
1772
1773        /* Flush first if not enough space for SHA1 signature */
1774        if (left + 20 > WRITE_BUFFER_SIZE) {
1775                if (write_in_full(fd, write_buffer, left) != left)
1776                        return -1;
1777                left = 0;
1778        }
1779
1780        /* Append the SHA1 signature at the end */
1781        git_SHA1_Final(write_buffer + left, context);
1782        hashcpy(sha1, write_buffer + left);
1783        left += 20;
1784        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1785}
1786
1787static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1788{
1789        /*
1790         * The only thing we care about in this function is to smudge the
1791         * falsely clean entry due to touch-update-touch race, so we leave
1792         * everything else as they are.  We are called for entries whose
1793         * ce_stat_data.sd_mtime match the index file mtime.
1794         *
1795         * Note that this actually does not do much for gitlinks, for
1796         * which ce_match_stat_basic() always goes to the actual
1797         * contents.  The caller checks with is_racy_timestamp() which
1798         * always says "no" for gitlinks, so we are not called for them ;-)
1799         */
1800        struct stat st;
1801
1802        if (lstat(ce->name, &st) < 0)
1803                return;
1804        if (ce_match_stat_basic(ce, &st))
1805                return;
1806        if (ce_modified_check_fs(ce, &st)) {
1807                /* This is "racily clean"; smudge it.  Note that this
1808                 * is a tricky code.  At first glance, it may appear
1809                 * that it can break with this sequence:
1810                 *
1811                 * $ echo xyzzy >frotz
1812                 * $ git-update-index --add frotz
1813                 * $ : >frotz
1814                 * $ sleep 3
1815                 * $ echo filfre >nitfol
1816                 * $ git-update-index --add nitfol
1817                 *
1818                 * but it does not.  When the second update-index runs,
1819                 * it notices that the entry "frotz" has the same timestamp
1820                 * as index, and if we were to smudge it by resetting its
1821                 * size to zero here, then the object name recorded
1822                 * in index is the 6-byte file but the cached stat information
1823                 * becomes zero --- which would then match what we would
1824                 * obtain from the filesystem next time we stat("frotz").
1825                 *
1826                 * However, the second update-index, before calling
1827                 * this function, notices that the cached size is 6
1828                 * bytes and what is on the filesystem is an empty
1829                 * file, and never calls us, so the cached size information
1830                 * for "frotz" stays 6 which does not match the filesystem.
1831                 */
1832                ce->ce_stat_data.sd_size = 0;
1833        }
1834}
1835
1836/* Copy miscellaneous fields but not the name */
1837static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1838                                       struct cache_entry *ce)
1839{
1840        short flags;
1841
1842        ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1843        ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1844        ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1845        ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1846        ondisk->dev  = htonl(ce->ce_stat_data.sd_dev);
1847        ondisk->ino  = htonl(ce->ce_stat_data.sd_ino);
1848        ondisk->mode = htonl(ce->ce_mode);
1849        ondisk->uid  = htonl(ce->ce_stat_data.sd_uid);
1850        ondisk->gid  = htonl(ce->ce_stat_data.sd_gid);
1851        ondisk->size = htonl(ce->ce_stat_data.sd_size);
1852        hashcpy(ondisk->sha1, ce->sha1);
1853
1854        flags = ce->ce_flags & ~CE_NAMEMASK;
1855        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1856        ondisk->flags = htons(flags);
1857        if (ce->ce_flags & CE_EXTENDED) {
1858                struct ondisk_cache_entry_extended *ondisk2;
1859                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1860                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1861                return ondisk2->name;
1862        }
1863        else {
1864                return ondisk->name;
1865        }
1866}
1867
1868static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1869                          struct strbuf *previous_name)
1870{
1871        int size;
1872        struct ondisk_cache_entry *ondisk;
1873        int saved_namelen = saved_namelen; /* compiler workaround */
1874        char *name;
1875        int result;
1876
1877        if (ce->ce_flags & CE_STRIP_NAME) {
1878                saved_namelen = ce_namelen(ce);
1879                ce->ce_namelen = 0;
1880        }
1881
1882        if (!previous_name) {
1883                size = ondisk_ce_size(ce);
1884                ondisk = xcalloc(1, size);
1885                name = copy_cache_entry_to_ondisk(ondisk, ce);
1886                memcpy(name, ce->name, ce_namelen(ce));
1887        } else {
1888                int common, to_remove, prefix_size;
1889                unsigned char to_remove_vi[16];
1890                for (common = 0;
1891                     (ce->name[common] &&
1892                      common < previous_name->len &&
1893                      ce->name[common] == previous_name->buf[common]);
1894                     common++)
1895                        ; /* still matching */
1896                to_remove = previous_name->len - common;
1897                prefix_size = encode_varint(to_remove, to_remove_vi);
1898
1899                if (ce->ce_flags & CE_EXTENDED)
1900                        size = offsetof(struct ondisk_cache_entry_extended, name);
1901                else
1902                        size = offsetof(struct ondisk_cache_entry, name);
1903                size += prefix_size + (ce_namelen(ce) - common + 1);
1904
1905                ondisk = xcalloc(1, size);
1906                name = copy_cache_entry_to_ondisk(ondisk, ce);
1907                memcpy(name, to_remove_vi, prefix_size);
1908                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1909
1910                strbuf_splice(previous_name, common, to_remove,
1911                              ce->name + common, ce_namelen(ce) - common);
1912        }
1913        if (ce->ce_flags & CE_STRIP_NAME) {
1914                ce->ce_namelen = saved_namelen;
1915                ce->ce_flags &= ~CE_STRIP_NAME;
1916        }
1917
1918        result = ce_write(c, fd, ondisk, size);
1919        free(ondisk);
1920        return result;
1921}
1922
1923/*
1924 * This function verifies if index_state has the correct sha1 of the
1925 * index file.  Don't die if we have any other failure, just return 0.
1926 */
1927static int verify_index_from(const struct index_state *istate, const char *path)
1928{
1929        int fd;
1930        ssize_t n;
1931        struct stat st;
1932        unsigned char sha1[20];
1933
1934        if (!istate->initialized)
1935                return 0;
1936
1937        fd = open(path, O_RDONLY);
1938        if (fd < 0)
1939                return 0;
1940
1941        if (fstat(fd, &st))
1942                goto out;
1943
1944        if (st.st_size < sizeof(struct cache_header) + 20)
1945                goto out;
1946
1947        n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1948        if (n != 20)
1949                goto out;
1950
1951        if (hashcmp(istate->sha1, sha1))
1952                goto out;
1953
1954        close(fd);
1955        return 1;
1956
1957out:
1958        close(fd);
1959        return 0;
1960}
1961
1962static int verify_index(const struct index_state *istate)
1963{
1964        return verify_index_from(istate, get_index_file());
1965}
1966
1967static int has_racy_timestamp(struct index_state *istate)
1968{
1969        int entries = istate->cache_nr;
1970        int i;
1971
1972        for (i = 0; i < entries; i++) {
1973                struct cache_entry *ce = istate->cache[i];
1974                if (is_racy_timestamp(istate, ce))
1975                        return 1;
1976        }
1977        return 0;
1978}
1979
1980/*
1981 * Opportunistically update the index but do not complain if we can't
1982 */
1983void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1984{
1985        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1986            verify_index(istate) &&
1987            write_locked_index(istate, lockfile, COMMIT_LOCK))
1988                rollback_lock_file(lockfile);
1989}
1990
1991static int do_write_index(struct index_state *istate, int newfd,
1992                          int strip_extensions)
1993{
1994        git_SHA_CTX c;
1995        struct cache_header hdr;
1996        int i, err, removed, extended, hdr_version;
1997        struct cache_entry **cache = istate->cache;
1998        int entries = istate->cache_nr;
1999        struct stat st;
2000        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2001
2002        for (i = removed = extended = 0; i < entries; i++) {
2003                if (cache[i]->ce_flags & CE_REMOVE)
2004                        removed++;
2005
2006                /* reduce extended entries if possible */
2007                cache[i]->ce_flags &= ~CE_EXTENDED;
2008                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2009                        extended++;
2010                        cache[i]->ce_flags |= CE_EXTENDED;
2011                }
2012        }
2013
2014        if (!istate->version) {
2015                istate->version = get_index_format_default();
2016                if (getenv("GIT_TEST_SPLIT_INDEX"))
2017                        init_split_index(istate);
2018        }
2019
2020        /* demote version 3 to version 2 when the latter suffices */
2021        if (istate->version == 3 || istate->version == 2)
2022                istate->version = extended ? 3 : 2;
2023
2024        hdr_version = istate->version;
2025
2026        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2027        hdr.hdr_version = htonl(hdr_version);
2028        hdr.hdr_entries = htonl(entries - removed);
2029
2030        git_SHA1_Init(&c);
2031        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2032                return -1;
2033
2034        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2035        for (i = 0; i < entries; i++) {
2036                struct cache_entry *ce = cache[i];
2037                if (ce->ce_flags & CE_REMOVE)
2038                        continue;
2039                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2040                        ce_smudge_racily_clean_entry(ce);
2041                if (is_null_sha1(ce->sha1)) {
2042                        static const char msg[] = "cache entry has null sha1: %s";
2043                        static int allow = -1;
2044
2045                        if (allow < 0)
2046                                allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2047                        if (allow)
2048                                warning(msg, ce->name);
2049                        else
2050                                return error(msg, ce->name);
2051                }
2052                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2053                        return -1;
2054        }
2055        strbuf_release(&previous_name_buf);
2056
2057        /* Write extension data here */
2058        if (!strip_extensions && istate->split_index) {
2059                struct strbuf sb = STRBUF_INIT;
2060
2061                err = write_link_extension(&sb, istate) < 0 ||
2062                        write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2063                                               sb.len) < 0 ||
2064                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2065                strbuf_release(&sb);
2066                if (err)
2067                        return -1;
2068        }
2069        if (!strip_extensions && istate->cache_tree) {
2070                struct strbuf sb = STRBUF_INIT;
2071
2072                cache_tree_write(&sb, istate->cache_tree);
2073                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2074                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2075                strbuf_release(&sb);
2076                if (err)
2077                        return -1;
2078        }
2079        if (!strip_extensions && istate->resolve_undo) {
2080                struct strbuf sb = STRBUF_INIT;
2081
2082                resolve_undo_write(&sb, istate->resolve_undo);
2083                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2084                                             sb.len) < 0
2085                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2086                strbuf_release(&sb);
2087                if (err)
2088                        return -1;
2089        }
2090        if (!strip_extensions && istate->untracked) {
2091                struct strbuf sb = STRBUF_INIT;
2092
2093                write_untracked_extension(&sb, istate->untracked);
2094                err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2095                                             sb.len) < 0 ||
2096                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2097                strbuf_release(&sb);
2098                if (err)
2099                        return -1;
2100        }
2101
2102        if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2103                return -1;
2104        istate->timestamp.sec = (unsigned int)st.st_mtime;
2105        istate->timestamp.nsec = ST_MTIME_NSEC(st);
2106        return 0;
2107}
2108
2109void set_alternate_index_output(const char *name)
2110{
2111        alternate_index_output = name;
2112}
2113
2114static int commit_locked_index(struct lock_file *lk)
2115{
2116        if (alternate_index_output)
2117                return commit_lock_file_to(lk, alternate_index_output);
2118        else
2119                return commit_lock_file(lk);
2120}
2121
2122static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2123                                 unsigned flags)
2124{
2125        int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
2126        if (ret)
2127                return ret;
2128        assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2129               (COMMIT_LOCK | CLOSE_LOCK));
2130        if (flags & COMMIT_LOCK)
2131                return commit_locked_index(lock);
2132        else if (flags & CLOSE_LOCK)
2133                return close_lock_file(lock);
2134        else
2135                return ret;
2136}
2137
2138static int write_split_index(struct index_state *istate,
2139                             struct lock_file *lock,
2140                             unsigned flags)
2141{
2142        int ret;
2143        prepare_to_write_split_index(istate);
2144        ret = do_write_locked_index(istate, lock, flags);
2145        finish_writing_split_index(istate);
2146        return ret;
2147}
2148
2149static struct tempfile temporary_sharedindex;
2150
2151static int write_shared_index(struct index_state *istate,
2152                              struct lock_file *lock, unsigned flags)
2153{
2154        struct split_index *si = istate->split_index;
2155        int fd, ret;
2156
2157        fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2158        if (fd < 0) {
2159                hashclr(si->base_sha1);
2160                return do_write_locked_index(istate, lock, flags);
2161        }
2162        move_cache_to_base_index(istate);
2163        ret = do_write_index(si->base, fd, 1);
2164        if (ret) {
2165                delete_tempfile(&temporary_sharedindex);
2166                return ret;
2167        }
2168        ret = rename_tempfile(&temporary_sharedindex,
2169                              git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2170        if (!ret)
2171                hashcpy(si->base_sha1, si->base->sha1);
2172        return ret;
2173}
2174
2175int write_locked_index(struct index_state *istate, struct lock_file *lock,
2176                       unsigned flags)
2177{
2178        struct split_index *si = istate->split_index;
2179
2180        if (!si || alternate_index_output ||
2181            (istate->cache_changed & ~EXTMASK)) {
2182                if (si)
2183                        hashclr(si->base_sha1);
2184                return do_write_locked_index(istate, lock, flags);
2185        }
2186
2187        if (getenv("GIT_TEST_SPLIT_INDEX")) {
2188                int v = si->base_sha1[0];
2189                if ((v & 15) < 6)
2190                        istate->cache_changed |= SPLIT_INDEX_ORDERED;
2191        }
2192        if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2193                int ret = write_shared_index(istate, lock, flags);
2194                if (ret)
2195                        return ret;
2196        }
2197
2198        return write_split_index(istate, lock, flags);
2199}
2200
2201/*
2202 * Read the index file that is potentially unmerged into given
2203 * index_state, dropping any unmerged entries.  Returns true if
2204 * the index is unmerged.  Callers who want to refuse to work
2205 * from an unmerged state can call this and check its return value,
2206 * instead of calling read_cache().
2207 */
2208int read_index_unmerged(struct index_state *istate)
2209{
2210        int i;
2211        int unmerged = 0;
2212
2213        read_index(istate);
2214        for (i = 0; i < istate->cache_nr; i++) {
2215                struct cache_entry *ce = istate->cache[i];
2216                struct cache_entry *new_ce;
2217                int size, len;
2218
2219                if (!ce_stage(ce))
2220                        continue;
2221                unmerged = 1;
2222                len = ce_namelen(ce);
2223                size = cache_entry_size(len);
2224                new_ce = xcalloc(1, size);
2225                memcpy(new_ce->name, ce->name, len);
2226                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2227                new_ce->ce_namelen = len;
2228                new_ce->ce_mode = ce->ce_mode;
2229                if (add_index_entry(istate, new_ce, 0))
2230                        return error("%s: cannot drop to stage #0",
2231                                     new_ce->name);
2232        }
2233        return unmerged;
2234}
2235
2236/*
2237 * Returns 1 if the path is an "other" path with respect to
2238 * the index; that is, the path is not mentioned in the index at all,
2239 * either as a file, a directory with some files in the index,
2240 * or as an unmerged entry.
2241 *
2242 * We helpfully remove a trailing "/" from directories so that
2243 * the output of read_directory can be used as-is.
2244 */
2245int index_name_is_other(const struct index_state *istate, const char *name,
2246                int namelen)
2247{
2248        int pos;
2249        if (namelen && name[namelen - 1] == '/')
2250                namelen--;
2251        pos = index_name_pos(istate, name, namelen);
2252        if (0 <= pos)
2253                return 0;       /* exact match */
2254        pos = -pos - 1;
2255        if (pos < istate->cache_nr) {
2256                struct cache_entry *ce = istate->cache[pos];
2257                if (ce_namelen(ce) == namelen &&
2258                    !memcmp(ce->name, name, namelen))
2259                        return 0; /* Yup, this one exists unmerged */
2260        }
2261        return 1;
2262}
2263
2264void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2265{
2266        int pos, len;
2267        unsigned long sz;
2268        enum object_type type;
2269        void *data;
2270
2271        len = strlen(path);
2272        pos = index_name_pos(istate, path, len);
2273        if (pos < 0) {
2274                /*
2275                 * We might be in the middle of a merge, in which
2276                 * case we would read stage #2 (ours).
2277                 */
2278                int i;
2279                for (i = -pos - 1;
2280                     (pos < 0 && i < istate->cache_nr &&
2281                      !strcmp(istate->cache[i]->name, path));
2282                     i++)
2283                        if (ce_stage(istate->cache[i]) == 2)
2284                                pos = i;
2285        }
2286        if (pos < 0)
2287                return NULL;
2288        data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2289        if (!data || type != OBJ_BLOB) {
2290                free(data);
2291                return NULL;
2292        }
2293        if (size)
2294                *size = sz;
2295        return data;
2296}
2297
2298void stat_validity_clear(struct stat_validity *sv)
2299{
2300        free(sv->sd);
2301        sv->sd = NULL;
2302}
2303
2304int stat_validity_check(struct stat_validity *sv, const char *path)
2305{
2306        struct stat st;
2307
2308        if (stat(path, &st) < 0)
2309                return sv->sd == NULL;
2310        if (!sv->sd)
2311                return 0;
2312        return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2313}
2314
2315void stat_validity_update(struct stat_validity *sv, int fd)
2316{
2317        struct stat st;
2318
2319        if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2320                stat_validity_clear(sv);
2321        else {
2322                if (!sv->sd)
2323                        sv->sd = xcalloc(1, sizeof(struct stat_data));
2324                fill_stat_data(sv->sd, &st);
2325        }
2326}