read-cache.con commit Merge branch 'tr/maint-apply-non-git-patch-parsefix' into maint (90360c7)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 */
   6#define NO_THE_INDEX_COMPATIBILITY_MACROS
   7#include "cache.h"
   8#include "cache-tree.h"
   9#include "refs.h"
  10#include "dir.h"
  11#include "tree.h"
  12#include "commit.h"
  13#include "blob.h"
  14#include "resolve-undo.h"
  15#include "strbuf.h"
  16#include "varint.h"
  17
  18static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really);
  19
  20/* Mask for the name length in ce_flags in the on-disk index */
  21
  22#define CE_NAMEMASK  (0x0fff)
  23
  24/* Index extensions.
  25 *
  26 * The first letter should be 'A'..'Z' for extensions that are not
  27 * necessary for a correct operation (i.e. optimization data).
  28 * When new extensions are added that _needs_ to be understood in
  29 * order to correctly interpret the index file, pick character that
  30 * is outside the range, to cause the reader to abort.
  31 */
  32
  33#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  34#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  35#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  36
  37struct index_state the_index;
  38
  39static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  40{
  41        istate->cache[nr] = ce;
  42        add_name_hash(istate, ce);
  43}
  44
  45static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  46{
  47        struct cache_entry *old = istate->cache[nr];
  48
  49        remove_name_hash(istate, old);
  50        set_index_entry(istate, nr, ce);
  51        istate->cache_changed = 1;
  52}
  53
  54void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  55{
  56        struct cache_entry *old = istate->cache[nr], *new;
  57        int namelen = strlen(new_name);
  58
  59        new = xmalloc(cache_entry_size(namelen));
  60        copy_cache_entry(new, old);
  61        new->ce_flags &= ~CE_STATE_MASK;
  62        new->ce_namelen = namelen;
  63        memcpy(new->name, new_name, namelen + 1);
  64
  65        cache_tree_invalidate_path(istate->cache_tree, old->name);
  66        remove_index_entry_at(istate, nr);
  67        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  68}
  69
  70/*
  71 * This only updates the "non-critical" parts of the directory
  72 * cache, ie the parts that aren't tracked by GIT, and only used
  73 * to validate the cache.
  74 */
  75void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
  76{
  77        ce->ce_ctime.sec = (unsigned int)st->st_ctime;
  78        ce->ce_mtime.sec = (unsigned int)st->st_mtime;
  79        ce->ce_ctime.nsec = ST_CTIME_NSEC(*st);
  80        ce->ce_mtime.nsec = ST_MTIME_NSEC(*st);
  81        ce->ce_dev = st->st_dev;
  82        ce->ce_ino = st->st_ino;
  83        ce->ce_uid = st->st_uid;
  84        ce->ce_gid = st->st_gid;
  85        ce->ce_size = st->st_size;
  86
  87        if (assume_unchanged)
  88                ce->ce_flags |= CE_VALID;
  89
  90        if (S_ISREG(st->st_mode))
  91                ce_mark_uptodate(ce);
  92}
  93
  94static int ce_compare_data(struct cache_entry *ce, struct stat *st)
  95{
  96        int match = -1;
  97        int fd = open(ce->name, O_RDONLY);
  98
  99        if (fd >= 0) {
 100                unsigned char sha1[20];
 101                if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
 102                        match = hashcmp(sha1, ce->sha1);
 103                /* index_fd() closed the file descriptor already */
 104        }
 105        return match;
 106}
 107
 108static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
 109{
 110        int match = -1;
 111        void *buffer;
 112        unsigned long size;
 113        enum object_type type;
 114        struct strbuf sb = STRBUF_INIT;
 115
 116        if (strbuf_readlink(&sb, ce->name, expected_size))
 117                return -1;
 118
 119        buffer = read_sha1_file(ce->sha1, &type, &size);
 120        if (buffer) {
 121                if (size == sb.len)
 122                        match = memcmp(buffer, sb.buf, size);
 123                free(buffer);
 124        }
 125        strbuf_release(&sb);
 126        return match;
 127}
 128
 129static int ce_compare_gitlink(struct cache_entry *ce)
 130{
 131        unsigned char sha1[20];
 132
 133        /*
 134         * We don't actually require that the .git directory
 135         * under GITLINK directory be a valid git directory. It
 136         * might even be missing (in case nobody populated that
 137         * sub-project).
 138         *
 139         * If so, we consider it always to match.
 140         */
 141        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 142                return 0;
 143        return hashcmp(sha1, ce->sha1);
 144}
 145
 146static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
 147{
 148        switch (st->st_mode & S_IFMT) {
 149        case S_IFREG:
 150                if (ce_compare_data(ce, st))
 151                        return DATA_CHANGED;
 152                break;
 153        case S_IFLNK:
 154                if (ce_compare_link(ce, xsize_t(st->st_size)))
 155                        return DATA_CHANGED;
 156                break;
 157        case S_IFDIR:
 158                if (S_ISGITLINK(ce->ce_mode))
 159                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 160        default:
 161                return TYPE_CHANGED;
 162        }
 163        return 0;
 164}
 165
 166static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 167{
 168        unsigned int changed = 0;
 169
 170        if (ce->ce_flags & CE_REMOVE)
 171                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 172
 173        switch (ce->ce_mode & S_IFMT) {
 174        case S_IFREG:
 175                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 176                /* We consider only the owner x bit to be relevant for
 177                 * "mode changes"
 178                 */
 179                if (trust_executable_bit &&
 180                    (0100 & (ce->ce_mode ^ st->st_mode)))
 181                        changed |= MODE_CHANGED;
 182                break;
 183        case S_IFLNK:
 184                if (!S_ISLNK(st->st_mode) &&
 185                    (has_symlinks || !S_ISREG(st->st_mode)))
 186                        changed |= TYPE_CHANGED;
 187                break;
 188        case S_IFGITLINK:
 189                /* We ignore most of the st_xxx fields for gitlinks */
 190                if (!S_ISDIR(st->st_mode))
 191                        changed |= TYPE_CHANGED;
 192                else if (ce_compare_gitlink(ce))
 193                        changed |= DATA_CHANGED;
 194                return changed;
 195        default:
 196                die("internal error: ce_mode is %o", ce->ce_mode);
 197        }
 198        if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
 199                changed |= MTIME_CHANGED;
 200        if (trust_ctime && check_stat &&
 201            ce->ce_ctime.sec != (unsigned int)st->st_ctime)
 202                changed |= CTIME_CHANGED;
 203
 204#ifdef USE_NSEC
 205        if (check_stat && ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
 206                changed |= MTIME_CHANGED;
 207        if (trust_ctime && check_stat &&
 208            ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
 209                changed |= CTIME_CHANGED;
 210#endif
 211
 212        if (check_stat) {
 213                if (ce->ce_uid != (unsigned int) st->st_uid ||
 214                        ce->ce_gid != (unsigned int) st->st_gid)
 215                        changed |= OWNER_CHANGED;
 216                if (ce->ce_ino != (unsigned int) st->st_ino)
 217                        changed |= INODE_CHANGED;
 218        }
 219
 220#ifdef USE_STDEV
 221        /*
 222         * st_dev breaks on network filesystems where different
 223         * clients will have different views of what "device"
 224         * the filesystem is on
 225         */
 226        if (check_stat && ce->ce_dev != (unsigned int) st->st_dev)
 227                        changed |= INODE_CHANGED;
 228#endif
 229
 230        if (ce->ce_size != (unsigned int) st->st_size)
 231                changed |= DATA_CHANGED;
 232
 233        /* Racily smudged entry? */
 234        if (!ce->ce_size) {
 235                if (!is_empty_blob_sha1(ce->sha1))
 236                        changed |= DATA_CHANGED;
 237        }
 238
 239        return changed;
 240}
 241
 242static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
 243{
 244        return (!S_ISGITLINK(ce->ce_mode) &&
 245                istate->timestamp.sec &&
 246#ifdef USE_NSEC
 247                 /* nanosecond timestamped files can also be racy! */
 248                (istate->timestamp.sec < ce->ce_mtime.sec ||
 249                 (istate->timestamp.sec == ce->ce_mtime.sec &&
 250                  istate->timestamp.nsec <= ce->ce_mtime.nsec))
 251#else
 252                istate->timestamp.sec <= ce->ce_mtime.sec
 253#endif
 254                 );
 255}
 256
 257int ie_match_stat(const struct index_state *istate,
 258                  struct cache_entry *ce, struct stat *st,
 259                  unsigned int options)
 260{
 261        unsigned int changed;
 262        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 263        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 264        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 265
 266        /*
 267         * If it's marked as always valid in the index, it's
 268         * valid whatever the checked-out copy says.
 269         *
 270         * skip-worktree has the same effect with higher precedence
 271         */
 272        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 273                return 0;
 274        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 275                return 0;
 276
 277        /*
 278         * Intent-to-add entries have not been added, so the index entry
 279         * by definition never matches what is in the work tree until it
 280         * actually gets added.
 281         */
 282        if (ce->ce_flags & CE_INTENT_TO_ADD)
 283                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 284
 285        changed = ce_match_stat_basic(ce, st);
 286
 287        /*
 288         * Within 1 second of this sequence:
 289         *      echo xyzzy >file && git-update-index --add file
 290         * running this command:
 291         *      echo frotz >file
 292         * would give a falsely clean cache entry.  The mtime and
 293         * length match the cache, and other stat fields do not change.
 294         *
 295         * We could detect this at update-index time (the cache entry
 296         * being registered/updated records the same time as "now")
 297         * and delay the return from git-update-index, but that would
 298         * effectively mean we can make at most one commit per second,
 299         * which is not acceptable.  Instead, we check cache entries
 300         * whose mtime are the same as the index file timestamp more
 301         * carefully than others.
 302         */
 303        if (!changed && is_racy_timestamp(istate, ce)) {
 304                if (assume_racy_is_modified)
 305                        changed |= DATA_CHANGED;
 306                else
 307                        changed |= ce_modified_check_fs(ce, st);
 308        }
 309
 310        return changed;
 311}
 312
 313int ie_modified(const struct index_state *istate,
 314                struct cache_entry *ce, struct stat *st, unsigned int options)
 315{
 316        int changed, changed_fs;
 317
 318        changed = ie_match_stat(istate, ce, st, options);
 319        if (!changed)
 320                return 0;
 321        /*
 322         * If the mode or type has changed, there's no point in trying
 323         * to refresh the entry - it's not going to match
 324         */
 325        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 326                return changed;
 327
 328        /*
 329         * Immediately after read-tree or update-index --cacheinfo,
 330         * the length field is zero, as we have never even read the
 331         * lstat(2) information once, and we cannot trust DATA_CHANGED
 332         * returned by ie_match_stat() which in turn was returned by
 333         * ce_match_stat_basic() to signal that the filesize of the
 334         * blob changed.  We have to actually go to the filesystem to
 335         * see if the contents match, and if so, should answer "unchanged".
 336         *
 337         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 338         * already has checked the actual HEAD from the filesystem in the
 339         * subproject.  If ie_match_stat() already said it is different,
 340         * then we know it is.
 341         */
 342        if ((changed & DATA_CHANGED) &&
 343            (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
 344                return changed;
 345
 346        changed_fs = ce_modified_check_fs(ce, st);
 347        if (changed_fs)
 348                return changed | changed_fs;
 349        return 0;
 350}
 351
 352int base_name_compare(const char *name1, int len1, int mode1,
 353                      const char *name2, int len2, int mode2)
 354{
 355        unsigned char c1, c2;
 356        int len = len1 < len2 ? len1 : len2;
 357        int cmp;
 358
 359        cmp = memcmp(name1, name2, len);
 360        if (cmp)
 361                return cmp;
 362        c1 = name1[len];
 363        c2 = name2[len];
 364        if (!c1 && S_ISDIR(mode1))
 365                c1 = '/';
 366        if (!c2 && S_ISDIR(mode2))
 367                c2 = '/';
 368        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 369}
 370
 371/*
 372 * df_name_compare() is identical to base_name_compare(), except it
 373 * compares conflicting directory/file entries as equal. Note that
 374 * while a directory name compares as equal to a regular file, they
 375 * then individually compare _differently_ to a filename that has
 376 * a dot after the basename (because '\0' < '.' < '/').
 377 *
 378 * This is used by routines that want to traverse the git namespace
 379 * but then handle conflicting entries together when possible.
 380 */
 381int df_name_compare(const char *name1, int len1, int mode1,
 382                    const char *name2, int len2, int mode2)
 383{
 384        int len = len1 < len2 ? len1 : len2, cmp;
 385        unsigned char c1, c2;
 386
 387        cmp = memcmp(name1, name2, len);
 388        if (cmp)
 389                return cmp;
 390        /* Directories and files compare equal (same length, same name) */
 391        if (len1 == len2)
 392                return 0;
 393        c1 = name1[len];
 394        if (!c1 && S_ISDIR(mode1))
 395                c1 = '/';
 396        c2 = name2[len];
 397        if (!c2 && S_ISDIR(mode2))
 398                c2 = '/';
 399        if (c1 == '/' && !c2)
 400                return 0;
 401        if (c2 == '/' && !c1)
 402                return 0;
 403        return c1 - c2;
 404}
 405
 406int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 407{
 408        int len = len1 < len2 ? len1 : len2;
 409        int cmp;
 410
 411        cmp = memcmp(name1, name2, len);
 412        if (cmp)
 413                return cmp;
 414        if (len1 < len2)
 415                return -1;
 416        if (len1 > len2)
 417                return 1;
 418
 419        if (stage1 < stage2)
 420                return -1;
 421        if (stage1 > stage2)
 422                return 1;
 423        return 0;
 424}
 425
 426int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
 427{
 428        return cache_name_stage_compare(name1, len1, 0, name2, len2, 0);
 429}
 430
 431static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 432{
 433        int first, last;
 434
 435        first = 0;
 436        last = istate->cache_nr;
 437        while (last > first) {
 438                int next = (last + first) >> 1;
 439                struct cache_entry *ce = istate->cache[next];
 440                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 441                if (!cmp)
 442                        return next;
 443                if (cmp < 0) {
 444                        last = next;
 445                        continue;
 446                }
 447                first = next+1;
 448        }
 449        return -first-1;
 450}
 451
 452int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 453{
 454        return index_name_stage_pos(istate, name, namelen, 0);
 455}
 456
 457/* Remove entry, return true if there are more entries to go.. */
 458int remove_index_entry_at(struct index_state *istate, int pos)
 459{
 460        struct cache_entry *ce = istate->cache[pos];
 461
 462        record_resolve_undo(istate, ce);
 463        remove_name_hash(istate, ce);
 464        istate->cache_changed = 1;
 465        istate->cache_nr--;
 466        if (pos >= istate->cache_nr)
 467                return 0;
 468        memmove(istate->cache + pos,
 469                istate->cache + pos + 1,
 470                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 471        return 1;
 472}
 473
 474/*
 475 * Remove all cache ententries marked for removal, that is where
 476 * CE_REMOVE is set in ce_flags.  This is much more effective than
 477 * calling remove_index_entry_at() for each entry to be removed.
 478 */
 479void remove_marked_cache_entries(struct index_state *istate)
 480{
 481        struct cache_entry **ce_array = istate->cache;
 482        unsigned int i, j;
 483
 484        for (i = j = 0; i < istate->cache_nr; i++) {
 485                if (ce_array[i]->ce_flags & CE_REMOVE)
 486                        remove_name_hash(istate, ce_array[i]);
 487                else
 488                        ce_array[j++] = ce_array[i];
 489        }
 490        istate->cache_changed = 1;
 491        istate->cache_nr = j;
 492}
 493
 494int remove_file_from_index(struct index_state *istate, const char *path)
 495{
 496        int pos = index_name_pos(istate, path, strlen(path));
 497        if (pos < 0)
 498                pos = -pos-1;
 499        cache_tree_invalidate_path(istate->cache_tree, path);
 500        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 501                remove_index_entry_at(istate, pos);
 502        return 0;
 503}
 504
 505static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 506{
 507        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 508}
 509
 510static int index_name_pos_also_unmerged(struct index_state *istate,
 511        const char *path, int namelen)
 512{
 513        int pos = index_name_pos(istate, path, namelen);
 514        struct cache_entry *ce;
 515
 516        if (pos >= 0)
 517                return pos;
 518
 519        /* maybe unmerged? */
 520        pos = -1 - pos;
 521        if (pos >= istate->cache_nr ||
 522                        compare_name((ce = istate->cache[pos]), path, namelen))
 523                return -1;
 524
 525        /* order of preference: stage 2, 1, 3 */
 526        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 527                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 528                        !compare_name(ce, path, namelen))
 529                pos++;
 530        return pos;
 531}
 532
 533static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 534{
 535        int len = ce_namelen(ce);
 536        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 537}
 538
 539/*
 540 * If we add a filename that aliases in the cache, we will use the
 541 * name that we already have - but we don't want to update the same
 542 * alias twice, because that implies that there were actually two
 543 * different files with aliasing names!
 544 *
 545 * So we use the CE_ADDED flag to verify that the alias was an old
 546 * one before we accept it as
 547 */
 548static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
 549{
 550        int len;
 551        struct cache_entry *new;
 552
 553        if (alias->ce_flags & CE_ADDED)
 554                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 555
 556        /* Ok, create the new entry using the name of the existing alias */
 557        len = ce_namelen(alias);
 558        new = xcalloc(1, cache_entry_size(len));
 559        memcpy(new->name, alias->name, len);
 560        copy_cache_entry(new, ce);
 561        free(ce);
 562        return new;
 563}
 564
 565static void record_intent_to_add(struct cache_entry *ce)
 566{
 567        unsigned char sha1[20];
 568        if (write_sha1_file("", 0, blob_type, sha1))
 569                die("cannot create an empty blob in the object database");
 570        hashcpy(ce->sha1, sha1);
 571}
 572
 573int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 574{
 575        int size, namelen, was_same;
 576        mode_t st_mode = st->st_mode;
 577        struct cache_entry *ce, *alias;
 578        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 579        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 580        int pretend = flags & ADD_CACHE_PRETEND;
 581        int intent_only = flags & ADD_CACHE_INTENT;
 582        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 583                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 584
 585        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 586                return error("%s: can only add regular files, symbolic links or git-directories", path);
 587
 588        namelen = strlen(path);
 589        if (S_ISDIR(st_mode)) {
 590                while (namelen && path[namelen-1] == '/')
 591                        namelen--;
 592        }
 593        size = cache_entry_size(namelen);
 594        ce = xcalloc(1, size);
 595        memcpy(ce->name, path, namelen);
 596        ce->ce_namelen = namelen;
 597        if (!intent_only)
 598                fill_stat_cache_info(ce, st);
 599        else
 600                ce->ce_flags |= CE_INTENT_TO_ADD;
 601
 602        if (trust_executable_bit && has_symlinks)
 603                ce->ce_mode = create_ce_mode(st_mode);
 604        else {
 605                /* If there is an existing entry, pick the mode bits and type
 606                 * from it, otherwise assume unexecutable regular file.
 607                 */
 608                struct cache_entry *ent;
 609                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 610
 611                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 612                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 613        }
 614
 615        /* When core.ignorecase=true, determine if a directory of the same name but differing
 616         * case already exists within the Git repository.  If it does, ensure the directory
 617         * case of the file being added to the repository matches (is folded into) the existing
 618         * entry's directory case.
 619         */
 620        if (ignore_case) {
 621                const char *startPtr = ce->name;
 622                const char *ptr = startPtr;
 623                while (*ptr) {
 624                        while (*ptr && *ptr != '/')
 625                                ++ptr;
 626                        if (*ptr == '/') {
 627                                struct cache_entry *foundce;
 628                                ++ptr;
 629                                foundce = index_name_exists(istate, ce->name, ptr - ce->name, ignore_case);
 630                                if (foundce) {
 631                                        memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
 632                                        startPtr = ptr;
 633                                }
 634                        }
 635                }
 636        }
 637
 638        alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 639        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 640                /* Nothing changed, really */
 641                free(ce);
 642                if (!S_ISGITLINK(alias->ce_mode))
 643                        ce_mark_uptodate(alias);
 644                alias->ce_flags |= CE_ADDED;
 645                return 0;
 646        }
 647        if (!intent_only) {
 648                if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
 649                        return error("unable to index file %s", path);
 650        } else
 651                record_intent_to_add(ce);
 652
 653        if (ignore_case && alias && different_name(ce, alias))
 654                ce = create_alias_ce(ce, alias);
 655        ce->ce_flags |= CE_ADDED;
 656
 657        /* It was suspected to be racily clean, but it turns out to be Ok */
 658        was_same = (alias &&
 659                    !ce_stage(alias) &&
 660                    !hashcmp(alias->sha1, ce->sha1) &&
 661                    ce->ce_mode == alias->ce_mode);
 662
 663        if (pretend)
 664                ;
 665        else if (add_index_entry(istate, ce, add_option))
 666                return error("unable to add %s to index",path);
 667        if (verbose && !was_same)
 668                printf("add '%s'\n", path);
 669        return 0;
 670}
 671
 672int add_file_to_index(struct index_state *istate, const char *path, int flags)
 673{
 674        struct stat st;
 675        if (lstat(path, &st))
 676                die_errno("unable to stat '%s'", path);
 677        return add_to_index(istate, path, &st, flags);
 678}
 679
 680struct cache_entry *make_cache_entry(unsigned int mode,
 681                const unsigned char *sha1, const char *path, int stage,
 682                int refresh)
 683{
 684        int size, len;
 685        struct cache_entry *ce;
 686
 687        if (!verify_path(path)) {
 688                error("Invalid path '%s'", path);
 689                return NULL;
 690        }
 691
 692        len = strlen(path);
 693        size = cache_entry_size(len);
 694        ce = xcalloc(1, size);
 695
 696        hashcpy(ce->sha1, sha1);
 697        memcpy(ce->name, path, len);
 698        ce->ce_flags = create_ce_flags(stage);
 699        ce->ce_namelen = len;
 700        ce->ce_mode = create_ce_mode(mode);
 701
 702        if (refresh)
 703                return refresh_cache_entry(ce, 0);
 704
 705        return ce;
 706}
 707
 708int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 709{
 710        int len = ce_namelen(a);
 711        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 712}
 713
 714int ce_path_match(const struct cache_entry *ce, const struct pathspec *pathspec)
 715{
 716        return match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, NULL);
 717}
 718
 719/*
 720 * We fundamentally don't like some paths: we don't want
 721 * dot or dot-dot anywhere, and for obvious reasons don't
 722 * want to recurse into ".git" either.
 723 *
 724 * Also, we don't want double slashes or slashes at the
 725 * end that can make pathnames ambiguous.
 726 */
 727static int verify_dotfile(const char *rest)
 728{
 729        /*
 730         * The first character was '.', but that
 731         * has already been discarded, we now test
 732         * the rest.
 733         */
 734
 735        /* "." is not allowed */
 736        if (*rest == '\0' || is_dir_sep(*rest))
 737                return 0;
 738
 739        switch (*rest) {
 740        /*
 741         * ".git" followed by  NUL or slash is bad. This
 742         * shares the path end test with the ".." case.
 743         */
 744        case 'g':
 745                if (rest[1] != 'i')
 746                        break;
 747                if (rest[2] != 't')
 748                        break;
 749                rest += 2;
 750        /* fallthrough */
 751        case '.':
 752                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 753                        return 0;
 754        }
 755        return 1;
 756}
 757
 758int verify_path(const char *path)
 759{
 760        char c;
 761
 762        if (has_dos_drive_prefix(path))
 763                return 0;
 764
 765        goto inside;
 766        for (;;) {
 767                if (!c)
 768                        return 1;
 769                if (is_dir_sep(c)) {
 770inside:
 771                        c = *path++;
 772                        if ((c == '.' && !verify_dotfile(path)) ||
 773                            is_dir_sep(c) || c == '\0')
 774                                return 0;
 775                }
 776                c = *path++;
 777        }
 778}
 779
 780/*
 781 * Do we have another file that has the beginning components being a
 782 * proper superset of the name we're trying to add?
 783 */
 784static int has_file_name(struct index_state *istate,
 785                         const struct cache_entry *ce, int pos, int ok_to_replace)
 786{
 787        int retval = 0;
 788        int len = ce_namelen(ce);
 789        int stage = ce_stage(ce);
 790        const char *name = ce->name;
 791
 792        while (pos < istate->cache_nr) {
 793                struct cache_entry *p = istate->cache[pos++];
 794
 795                if (len >= ce_namelen(p))
 796                        break;
 797                if (memcmp(name, p->name, len))
 798                        break;
 799                if (ce_stage(p) != stage)
 800                        continue;
 801                if (p->name[len] != '/')
 802                        continue;
 803                if (p->ce_flags & CE_REMOVE)
 804                        continue;
 805                retval = -1;
 806                if (!ok_to_replace)
 807                        break;
 808                remove_index_entry_at(istate, --pos);
 809        }
 810        return retval;
 811}
 812
 813/*
 814 * Do we have another file with a pathname that is a proper
 815 * subset of the name we're trying to add?
 816 */
 817static int has_dir_name(struct index_state *istate,
 818                        const struct cache_entry *ce, int pos, int ok_to_replace)
 819{
 820        int retval = 0;
 821        int stage = ce_stage(ce);
 822        const char *name = ce->name;
 823        const char *slash = name + ce_namelen(ce);
 824
 825        for (;;) {
 826                int len;
 827
 828                for (;;) {
 829                        if (*--slash == '/')
 830                                break;
 831                        if (slash <= ce->name)
 832                                return retval;
 833                }
 834                len = slash - name;
 835
 836                pos = index_name_stage_pos(istate, name, len, stage);
 837                if (pos >= 0) {
 838                        /*
 839                         * Found one, but not so fast.  This could
 840                         * be a marker that says "I was here, but
 841                         * I am being removed".  Such an entry is
 842                         * not a part of the resulting tree, and
 843                         * it is Ok to have a directory at the same
 844                         * path.
 845                         */
 846                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 847                                retval = -1;
 848                                if (!ok_to_replace)
 849                                        break;
 850                                remove_index_entry_at(istate, pos);
 851                                continue;
 852                        }
 853                }
 854                else
 855                        pos = -pos-1;
 856
 857                /*
 858                 * Trivial optimization: if we find an entry that
 859                 * already matches the sub-directory, then we know
 860                 * we're ok, and we can exit.
 861                 */
 862                while (pos < istate->cache_nr) {
 863                        struct cache_entry *p = istate->cache[pos];
 864                        if ((ce_namelen(p) <= len) ||
 865                            (p->name[len] != '/') ||
 866                            memcmp(p->name, name, len))
 867                                break; /* not our subdirectory */
 868                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 869                                /*
 870                                 * p is at the same stage as our entry, and
 871                                 * is a subdirectory of what we are looking
 872                                 * at, so we cannot have conflicts at our
 873                                 * level or anything shorter.
 874                                 */
 875                                return retval;
 876                        pos++;
 877                }
 878        }
 879        return retval;
 880}
 881
 882/* We may be in a situation where we already have path/file and path
 883 * is being added, or we already have path and path/file is being
 884 * added.  Either one would result in a nonsense tree that has path
 885 * twice when git-write-tree tries to write it out.  Prevent it.
 886 *
 887 * If ok-to-replace is specified, we remove the conflicting entries
 888 * from the cache so the caller should recompute the insert position.
 889 * When this happens, we return non-zero.
 890 */
 891static int check_file_directory_conflict(struct index_state *istate,
 892                                         const struct cache_entry *ce,
 893                                         int pos, int ok_to_replace)
 894{
 895        int retval;
 896
 897        /*
 898         * When ce is an "I am going away" entry, we allow it to be added
 899         */
 900        if (ce->ce_flags & CE_REMOVE)
 901                return 0;
 902
 903        /*
 904         * We check if the path is a sub-path of a subsequent pathname
 905         * first, since removing those will not change the position
 906         * in the array.
 907         */
 908        retval = has_file_name(istate, ce, pos, ok_to_replace);
 909
 910        /*
 911         * Then check if the path might have a clashing sub-directory
 912         * before it.
 913         */
 914        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 915}
 916
 917static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 918{
 919        int pos;
 920        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 921        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 922        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 923        int new_only = option & ADD_CACHE_NEW_ONLY;
 924
 925        cache_tree_invalidate_path(istate->cache_tree, ce->name);
 926        pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 927
 928        /* existing match? Just replace it. */
 929        if (pos >= 0) {
 930                if (!new_only)
 931                        replace_index_entry(istate, pos, ce);
 932                return 0;
 933        }
 934        pos = -pos-1;
 935
 936        /*
 937         * Inserting a merged entry ("stage 0") into the index
 938         * will always replace all non-merged entries..
 939         */
 940        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 941                while (ce_same_name(istate->cache[pos], ce)) {
 942                        ok_to_add = 1;
 943                        if (!remove_index_entry_at(istate, pos))
 944                                break;
 945                }
 946        }
 947
 948        if (!ok_to_add)
 949                return -1;
 950        if (!verify_path(ce->name))
 951                return error("Invalid path '%s'", ce->name);
 952
 953        if (!skip_df_check &&
 954            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 955                if (!ok_to_replace)
 956                        return error("'%s' appears as both a file and as a directory",
 957                                     ce->name);
 958                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 959                pos = -pos-1;
 960        }
 961        return pos + 1;
 962}
 963
 964int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 965{
 966        int pos;
 967
 968        if (option & ADD_CACHE_JUST_APPEND)
 969                pos = istate->cache_nr;
 970        else {
 971                int ret;
 972                ret = add_index_entry_with_check(istate, ce, option);
 973                if (ret <= 0)
 974                        return ret;
 975                pos = ret - 1;
 976        }
 977
 978        /* Make sure the array is big enough .. */
 979        if (istate->cache_nr == istate->cache_alloc) {
 980                istate->cache_alloc = alloc_nr(istate->cache_alloc);
 981                istate->cache = xrealloc(istate->cache,
 982                                        istate->cache_alloc * sizeof(*istate->cache));
 983        }
 984
 985        /* Add it in.. */
 986        istate->cache_nr++;
 987        if (istate->cache_nr > pos + 1)
 988                memmove(istate->cache + pos + 1,
 989                        istate->cache + pos,
 990                        (istate->cache_nr - pos - 1) * sizeof(ce));
 991        set_index_entry(istate, pos, ce);
 992        istate->cache_changed = 1;
 993        return 0;
 994}
 995
 996/*
 997 * "refresh" does not calculate a new sha1 file or bring the
 998 * cache up-to-date for mode/content changes. But what it
 999 * _does_ do is to "re-match" the stat information of a file
1000 * with the cache, so that you can refresh the cache for a
1001 * file that hasn't been changed but where the stat entry is
1002 * out of date.
1003 *
1004 * For example, you'd want to do this after doing a "git-read-tree",
1005 * to link up the stat cache details with the proper files.
1006 */
1007static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1008                                             struct cache_entry *ce,
1009                                             unsigned int options, int *err,
1010                                             int *changed_ret)
1011{
1012        struct stat st;
1013        struct cache_entry *updated;
1014        int changed, size;
1015        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1016        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1017
1018        if (ce_uptodate(ce))
1019                return ce;
1020
1021        /*
1022         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1023         * that the change to the work tree does not matter and told
1024         * us not to worry.
1025         */
1026        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1027                ce_mark_uptodate(ce);
1028                return ce;
1029        }
1030        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1031                ce_mark_uptodate(ce);
1032                return ce;
1033        }
1034
1035        if (lstat(ce->name, &st) < 0) {
1036                if (err)
1037                        *err = errno;
1038                return NULL;
1039        }
1040
1041        changed = ie_match_stat(istate, ce, &st, options);
1042        if (changed_ret)
1043                *changed_ret = changed;
1044        if (!changed) {
1045                /*
1046                 * The path is unchanged.  If we were told to ignore
1047                 * valid bit, then we did the actual stat check and
1048                 * found that the entry is unmodified.  If the entry
1049                 * is not marked VALID, this is the place to mark it
1050                 * valid again, under "assume unchanged" mode.
1051                 */
1052                if (ignore_valid && assume_unchanged &&
1053                    !(ce->ce_flags & CE_VALID))
1054                        ; /* mark this one VALID again */
1055                else {
1056                        /*
1057                         * We do not mark the index itself "modified"
1058                         * because CE_UPTODATE flag is in-core only;
1059                         * we are not going to write this change out.
1060                         */
1061                        if (!S_ISGITLINK(ce->ce_mode))
1062                                ce_mark_uptodate(ce);
1063                        return ce;
1064                }
1065        }
1066
1067        if (ie_modified(istate, ce, &st, options)) {
1068                if (err)
1069                        *err = EINVAL;
1070                return NULL;
1071        }
1072
1073        size = ce_size(ce);
1074        updated = xmalloc(size);
1075        memcpy(updated, ce, size);
1076        fill_stat_cache_info(updated, &st);
1077        /*
1078         * If ignore_valid is not set, we should leave CE_VALID bit
1079         * alone.  Otherwise, paths marked with --no-assume-unchanged
1080         * (i.e. things to be edited) will reacquire CE_VALID bit
1081         * automatically, which is not really what we want.
1082         */
1083        if (!ignore_valid && assume_unchanged &&
1084            !(ce->ce_flags & CE_VALID))
1085                updated->ce_flags &= ~CE_VALID;
1086
1087        return updated;
1088}
1089
1090static void show_file(const char * fmt, const char * name, int in_porcelain,
1091                      int * first, const char *header_msg)
1092{
1093        if (in_porcelain && *first && header_msg) {
1094                printf("%s\n", header_msg);
1095                *first = 0;
1096        }
1097        printf(fmt, name);
1098}
1099
1100int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1101                  char *seen, const char *header_msg)
1102{
1103        int i;
1104        int has_errors = 0;
1105        int really = (flags & REFRESH_REALLY) != 0;
1106        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1107        int quiet = (flags & REFRESH_QUIET) != 0;
1108        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1109        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1110        int first = 1;
1111        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1112        unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1113        const char *modified_fmt;
1114        const char *deleted_fmt;
1115        const char *typechange_fmt;
1116        const char *added_fmt;
1117        const char *unmerged_fmt;
1118
1119        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1120        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1121        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1122        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1123        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1124        for (i = 0; i < istate->cache_nr; i++) {
1125                struct cache_entry *ce, *new;
1126                int cache_errno = 0;
1127                int changed = 0;
1128                int filtered = 0;
1129
1130                ce = istate->cache[i];
1131                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1132                        continue;
1133
1134                if (pathspec &&
1135                    !match_pathspec(pathspec, ce->name, ce_namelen(ce), 0, seen))
1136                        filtered = 1;
1137
1138                if (ce_stage(ce)) {
1139                        while ((i < istate->cache_nr) &&
1140                               ! strcmp(istate->cache[i]->name, ce->name))
1141                                i++;
1142                        i--;
1143                        if (allow_unmerged)
1144                                continue;
1145                        if (!filtered)
1146                                show_file(unmerged_fmt, ce->name, in_porcelain,
1147                                          &first, header_msg);
1148                        has_errors = 1;
1149                        continue;
1150                }
1151
1152                if (filtered)
1153                        continue;
1154
1155                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1156                if (new == ce)
1157                        continue;
1158                if (!new) {
1159                        const char *fmt;
1160
1161                        if (not_new && cache_errno == ENOENT)
1162                                continue;
1163                        if (really && cache_errno == EINVAL) {
1164                                /* If we are doing --really-refresh that
1165                                 * means the index is not valid anymore.
1166                                 */
1167                                ce->ce_flags &= ~CE_VALID;
1168                                istate->cache_changed = 1;
1169                        }
1170                        if (quiet)
1171                                continue;
1172
1173                        if (cache_errno == ENOENT)
1174                                fmt = deleted_fmt;
1175                        else if (ce->ce_flags & CE_INTENT_TO_ADD)
1176                                fmt = added_fmt; /* must be before other checks */
1177                        else if (changed & TYPE_CHANGED)
1178                                fmt = typechange_fmt;
1179                        else
1180                                fmt = modified_fmt;
1181                        show_file(fmt,
1182                                  ce->name, in_porcelain, &first, header_msg);
1183                        has_errors = 1;
1184                        continue;
1185                }
1186
1187                replace_index_entry(istate, i, new);
1188        }
1189        return has_errors;
1190}
1191
1192static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1193{
1194        return refresh_cache_ent(&the_index, ce, really, NULL, NULL);
1195}
1196
1197
1198/*****************************************************************
1199 * Index File I/O
1200 *****************************************************************/
1201
1202#define INDEX_FORMAT_DEFAULT 3
1203
1204/*
1205 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1206 * Again - this is just a (very strong in practice) heuristic that
1207 * the inode hasn't changed.
1208 *
1209 * We save the fields in big-endian order to allow using the
1210 * index file over NFS transparently.
1211 */
1212struct ondisk_cache_entry {
1213        struct cache_time ctime;
1214        struct cache_time mtime;
1215        unsigned int dev;
1216        unsigned int ino;
1217        unsigned int mode;
1218        unsigned int uid;
1219        unsigned int gid;
1220        unsigned int size;
1221        unsigned char sha1[20];
1222        unsigned short flags;
1223        char name[FLEX_ARRAY]; /* more */
1224};
1225
1226/*
1227 * This struct is used when CE_EXTENDED bit is 1
1228 * The struct must match ondisk_cache_entry exactly from
1229 * ctime till flags
1230 */
1231struct ondisk_cache_entry_extended {
1232        struct cache_time ctime;
1233        struct cache_time mtime;
1234        unsigned int dev;
1235        unsigned int ino;
1236        unsigned int mode;
1237        unsigned int uid;
1238        unsigned int gid;
1239        unsigned int size;
1240        unsigned char sha1[20];
1241        unsigned short flags;
1242        unsigned short flags2;
1243        char name[FLEX_ARRAY]; /* more */
1244};
1245
1246/* These are only used for v3 or lower */
1247#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1248#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1249#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1250#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1251                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1252                            ondisk_cache_entry_size(ce_namelen(ce)))
1253
1254static int verify_hdr(struct cache_header *hdr, unsigned long size)
1255{
1256        git_SHA_CTX c;
1257        unsigned char sha1[20];
1258        int hdr_version;
1259
1260        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1261                return error("bad signature");
1262        hdr_version = ntohl(hdr->hdr_version);
1263        if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1264                return error("bad index version %d", hdr_version);
1265        git_SHA1_Init(&c);
1266        git_SHA1_Update(&c, hdr, size - 20);
1267        git_SHA1_Final(sha1, &c);
1268        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1269                return error("bad index file sha1 signature");
1270        return 0;
1271}
1272
1273static int read_index_extension(struct index_state *istate,
1274                                const char *ext, void *data, unsigned long sz)
1275{
1276        switch (CACHE_EXT(ext)) {
1277        case CACHE_EXT_TREE:
1278                istate->cache_tree = cache_tree_read(data, sz);
1279                break;
1280        case CACHE_EXT_RESOLVE_UNDO:
1281                istate->resolve_undo = resolve_undo_read(data, sz);
1282                break;
1283        default:
1284                if (*ext < 'A' || 'Z' < *ext)
1285                        return error("index uses %.4s extension, which we do not understand",
1286                                     ext);
1287                fprintf(stderr, "ignoring %.4s extension\n", ext);
1288                break;
1289        }
1290        return 0;
1291}
1292
1293int read_index(struct index_state *istate)
1294{
1295        return read_index_from(istate, get_index_file());
1296}
1297
1298#ifndef NEEDS_ALIGNED_ACCESS
1299#define ntoh_s(var) ntohs(var)
1300#define ntoh_l(var) ntohl(var)
1301#else
1302static inline uint16_t ntoh_s_force_align(void *p)
1303{
1304        uint16_t x;
1305        memcpy(&x, p, sizeof(x));
1306        return ntohs(x);
1307}
1308static inline uint32_t ntoh_l_force_align(void *p)
1309{
1310        uint32_t x;
1311        memcpy(&x, p, sizeof(x));
1312        return ntohl(x);
1313}
1314#define ntoh_s(var) ntoh_s_force_align(&(var))
1315#define ntoh_l(var) ntoh_l_force_align(&(var))
1316#endif
1317
1318static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1319                                                   unsigned int flags,
1320                                                   const char *name,
1321                                                   size_t len)
1322{
1323        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1324
1325        ce->ce_ctime.sec = ntoh_l(ondisk->ctime.sec);
1326        ce->ce_mtime.sec = ntoh_l(ondisk->mtime.sec);
1327        ce->ce_ctime.nsec = ntoh_l(ondisk->ctime.nsec);
1328        ce->ce_mtime.nsec = ntoh_l(ondisk->mtime.nsec);
1329        ce->ce_dev   = ntoh_l(ondisk->dev);
1330        ce->ce_ino   = ntoh_l(ondisk->ino);
1331        ce->ce_mode  = ntoh_l(ondisk->mode);
1332        ce->ce_uid   = ntoh_l(ondisk->uid);
1333        ce->ce_gid   = ntoh_l(ondisk->gid);
1334        ce->ce_size  = ntoh_l(ondisk->size);
1335        ce->ce_flags = flags & ~CE_NAMEMASK;
1336        ce->ce_namelen = len;
1337        hashcpy(ce->sha1, ondisk->sha1);
1338        memcpy(ce->name, name, len);
1339        ce->name[len] = '\0';
1340        return ce;
1341}
1342
1343/*
1344 * Adjacent cache entries tend to share the leading paths, so it makes
1345 * sense to only store the differences in later entries.  In the v4
1346 * on-disk format of the index, each on-disk cache entry stores the
1347 * number of bytes to be stripped from the end of the previous name,
1348 * and the bytes to append to the result, to come up with its name.
1349 */
1350static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1351{
1352        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1353        size_t len = decode_varint(&cp);
1354
1355        if (name->len < len)
1356                die("malformed name field in the index");
1357        strbuf_remove(name, name->len - len, len);
1358        for (ep = cp; *ep; ep++)
1359                ; /* find the end */
1360        strbuf_add(name, cp, ep - cp);
1361        return (const char *)ep + 1 - cp_;
1362}
1363
1364static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1365                                            unsigned long *ent_size,
1366                                            struct strbuf *previous_name)
1367{
1368        struct cache_entry *ce;
1369        size_t len;
1370        const char *name;
1371        unsigned int flags;
1372
1373        /* On-disk flags are just 16 bits */
1374        flags = ntoh_s(ondisk->flags);
1375        len = flags & CE_NAMEMASK;
1376
1377        if (flags & CE_EXTENDED) {
1378                struct ondisk_cache_entry_extended *ondisk2;
1379                int extended_flags;
1380                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1381                extended_flags = ntoh_s(ondisk2->flags2) << 16;
1382                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1383                if (extended_flags & ~CE_EXTENDED_FLAGS)
1384                        die("Unknown index entry format %08x", extended_flags);
1385                flags |= extended_flags;
1386                name = ondisk2->name;
1387        }
1388        else
1389                name = ondisk->name;
1390
1391        if (!previous_name) {
1392                /* v3 and earlier */
1393                if (len == CE_NAMEMASK)
1394                        len = strlen(name);
1395                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1396
1397                *ent_size = ondisk_ce_size(ce);
1398        } else {
1399                unsigned long consumed;
1400                consumed = expand_name_field(previous_name, name);
1401                ce = cache_entry_from_ondisk(ondisk, flags,
1402                                             previous_name->buf,
1403                                             previous_name->len);
1404
1405                *ent_size = (name - ((char *)ondisk)) + consumed;
1406        }
1407        return ce;
1408}
1409
1410/* remember to discard_cache() before reading a different cache! */
1411int read_index_from(struct index_state *istate, const char *path)
1412{
1413        int fd, i;
1414        struct stat st;
1415        unsigned long src_offset;
1416        struct cache_header *hdr;
1417        void *mmap;
1418        size_t mmap_size;
1419        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1420
1421        if (istate->initialized)
1422                return istate->cache_nr;
1423
1424        istate->timestamp.sec = 0;
1425        istate->timestamp.nsec = 0;
1426        fd = open(path, O_RDONLY);
1427        if (fd < 0) {
1428                if (errno == ENOENT)
1429                        return 0;
1430                die_errno("index file open failed");
1431        }
1432
1433        if (fstat(fd, &st))
1434                die_errno("cannot stat the open index");
1435
1436        mmap_size = xsize_t(st.st_size);
1437        if (mmap_size < sizeof(struct cache_header) + 20)
1438                die("index file smaller than expected");
1439
1440        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1441        if (mmap == MAP_FAILED)
1442                die_errno("unable to map index file");
1443        close(fd);
1444
1445        hdr = mmap;
1446        if (verify_hdr(hdr, mmap_size) < 0)
1447                goto unmap;
1448
1449        istate->version = ntohl(hdr->hdr_version);
1450        istate->cache_nr = ntohl(hdr->hdr_entries);
1451        istate->cache_alloc = alloc_nr(istate->cache_nr);
1452        istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1453        istate->initialized = 1;
1454
1455        if (istate->version == 4)
1456                previous_name = &previous_name_buf;
1457        else
1458                previous_name = NULL;
1459
1460        src_offset = sizeof(*hdr);
1461        for (i = 0; i < istate->cache_nr; i++) {
1462                struct ondisk_cache_entry *disk_ce;
1463                struct cache_entry *ce;
1464                unsigned long consumed;
1465
1466                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1467                ce = create_from_disk(disk_ce, &consumed, previous_name);
1468                set_index_entry(istate, i, ce);
1469
1470                src_offset += consumed;
1471        }
1472        strbuf_release(&previous_name_buf);
1473        istate->timestamp.sec = st.st_mtime;
1474        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1475
1476        while (src_offset <= mmap_size - 20 - 8) {
1477                /* After an array of active_nr index entries,
1478                 * there can be arbitrary number of extended
1479                 * sections, each of which is prefixed with
1480                 * extension name (4-byte) and section length
1481                 * in 4-byte network byte order.
1482                 */
1483                uint32_t extsize;
1484                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1485                extsize = ntohl(extsize);
1486                if (read_index_extension(istate,
1487                                         (const char *) mmap + src_offset,
1488                                         (char *) mmap + src_offset + 8,
1489                                         extsize) < 0)
1490                        goto unmap;
1491                src_offset += 8;
1492                src_offset += extsize;
1493        }
1494        munmap(mmap, mmap_size);
1495        return istate->cache_nr;
1496
1497unmap:
1498        munmap(mmap, mmap_size);
1499        die("index file corrupt");
1500}
1501
1502int is_index_unborn(struct index_state *istate)
1503{
1504        return (!istate->cache_nr && !istate->timestamp.sec);
1505}
1506
1507int discard_index(struct index_state *istate)
1508{
1509        int i;
1510
1511        for (i = 0; i < istate->cache_nr; i++)
1512                free(istate->cache[i]);
1513        resolve_undo_clear_index(istate);
1514        istate->cache_nr = 0;
1515        istate->cache_changed = 0;
1516        istate->timestamp.sec = 0;
1517        istate->timestamp.nsec = 0;
1518        free_name_hash(istate);
1519        cache_tree_free(&(istate->cache_tree));
1520        istate->initialized = 0;
1521
1522        /* no need to throw away allocated active_cache */
1523        return 0;
1524}
1525
1526int unmerged_index(const struct index_state *istate)
1527{
1528        int i;
1529        for (i = 0; i < istate->cache_nr; i++) {
1530                if (ce_stage(istate->cache[i]))
1531                        return 1;
1532        }
1533        return 0;
1534}
1535
1536#define WRITE_BUFFER_SIZE 8192
1537static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1538static unsigned long write_buffer_len;
1539
1540static int ce_write_flush(git_SHA_CTX *context, int fd)
1541{
1542        unsigned int buffered = write_buffer_len;
1543        if (buffered) {
1544                git_SHA1_Update(context, write_buffer, buffered);
1545                if (write_in_full(fd, write_buffer, buffered) != buffered)
1546                        return -1;
1547                write_buffer_len = 0;
1548        }
1549        return 0;
1550}
1551
1552static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1553{
1554        while (len) {
1555                unsigned int buffered = write_buffer_len;
1556                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1557                if (partial > len)
1558                        partial = len;
1559                memcpy(write_buffer + buffered, data, partial);
1560                buffered += partial;
1561                if (buffered == WRITE_BUFFER_SIZE) {
1562                        write_buffer_len = buffered;
1563                        if (ce_write_flush(context, fd))
1564                                return -1;
1565                        buffered = 0;
1566                }
1567                write_buffer_len = buffered;
1568                len -= partial;
1569                data = (char *) data + partial;
1570        }
1571        return 0;
1572}
1573
1574static int write_index_ext_header(git_SHA_CTX *context, int fd,
1575                                  unsigned int ext, unsigned int sz)
1576{
1577        ext = htonl(ext);
1578        sz = htonl(sz);
1579        return ((ce_write(context, fd, &ext, 4) < 0) ||
1580                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1581}
1582
1583static int ce_flush(git_SHA_CTX *context, int fd)
1584{
1585        unsigned int left = write_buffer_len;
1586
1587        if (left) {
1588                write_buffer_len = 0;
1589                git_SHA1_Update(context, write_buffer, left);
1590        }
1591
1592        /* Flush first if not enough space for SHA1 signature */
1593        if (left + 20 > WRITE_BUFFER_SIZE) {
1594                if (write_in_full(fd, write_buffer, left) != left)
1595                        return -1;
1596                left = 0;
1597        }
1598
1599        /* Append the SHA1 signature at the end */
1600        git_SHA1_Final(write_buffer + left, context);
1601        left += 20;
1602        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1603}
1604
1605static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1606{
1607        /*
1608         * The only thing we care about in this function is to smudge the
1609         * falsely clean entry due to touch-update-touch race, so we leave
1610         * everything else as they are.  We are called for entries whose
1611         * ce_mtime match the index file mtime.
1612         *
1613         * Note that this actually does not do much for gitlinks, for
1614         * which ce_match_stat_basic() always goes to the actual
1615         * contents.  The caller checks with is_racy_timestamp() which
1616         * always says "no" for gitlinks, so we are not called for them ;-)
1617         */
1618        struct stat st;
1619
1620        if (lstat(ce->name, &st) < 0)
1621                return;
1622        if (ce_match_stat_basic(ce, &st))
1623                return;
1624        if (ce_modified_check_fs(ce, &st)) {
1625                /* This is "racily clean"; smudge it.  Note that this
1626                 * is a tricky code.  At first glance, it may appear
1627                 * that it can break with this sequence:
1628                 *
1629                 * $ echo xyzzy >frotz
1630                 * $ git-update-index --add frotz
1631                 * $ : >frotz
1632                 * $ sleep 3
1633                 * $ echo filfre >nitfol
1634                 * $ git-update-index --add nitfol
1635                 *
1636                 * but it does not.  When the second update-index runs,
1637                 * it notices that the entry "frotz" has the same timestamp
1638                 * as index, and if we were to smudge it by resetting its
1639                 * size to zero here, then the object name recorded
1640                 * in index is the 6-byte file but the cached stat information
1641                 * becomes zero --- which would then match what we would
1642                 * obtain from the filesystem next time we stat("frotz").
1643                 *
1644                 * However, the second update-index, before calling
1645                 * this function, notices that the cached size is 6
1646                 * bytes and what is on the filesystem is an empty
1647                 * file, and never calls us, so the cached size information
1648                 * for "frotz" stays 6 which does not match the filesystem.
1649                 */
1650                ce->ce_size = 0;
1651        }
1652}
1653
1654/* Copy miscellaneous fields but not the name */
1655static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1656                                       struct cache_entry *ce)
1657{
1658        short flags;
1659
1660        ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1661        ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1662        ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1663        ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1664        ondisk->dev  = htonl(ce->ce_dev);
1665        ondisk->ino  = htonl(ce->ce_ino);
1666        ondisk->mode = htonl(ce->ce_mode);
1667        ondisk->uid  = htonl(ce->ce_uid);
1668        ondisk->gid  = htonl(ce->ce_gid);
1669        ondisk->size = htonl(ce->ce_size);
1670        hashcpy(ondisk->sha1, ce->sha1);
1671
1672        flags = ce->ce_flags;
1673        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1674        ondisk->flags = htons(flags);
1675        if (ce->ce_flags & CE_EXTENDED) {
1676                struct ondisk_cache_entry_extended *ondisk2;
1677                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1678                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1679                return ondisk2->name;
1680        }
1681        else {
1682                return ondisk->name;
1683        }
1684}
1685
1686static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1687                          struct strbuf *previous_name)
1688{
1689        int size;
1690        struct ondisk_cache_entry *ondisk;
1691        char *name;
1692        int result;
1693
1694        if (!previous_name) {
1695                size = ondisk_ce_size(ce);
1696                ondisk = xcalloc(1, size);
1697                name = copy_cache_entry_to_ondisk(ondisk, ce);
1698                memcpy(name, ce->name, ce_namelen(ce));
1699        } else {
1700                int common, to_remove, prefix_size;
1701                unsigned char to_remove_vi[16];
1702                for (common = 0;
1703                     (ce->name[common] &&
1704                      common < previous_name->len &&
1705                      ce->name[common] == previous_name->buf[common]);
1706                     common++)
1707                        ; /* still matching */
1708                to_remove = previous_name->len - common;
1709                prefix_size = encode_varint(to_remove, to_remove_vi);
1710
1711                if (ce->ce_flags & CE_EXTENDED)
1712                        size = offsetof(struct ondisk_cache_entry_extended, name);
1713                else
1714                        size = offsetof(struct ondisk_cache_entry, name);
1715                size += prefix_size + (ce_namelen(ce) - common + 1);
1716
1717                ondisk = xcalloc(1, size);
1718                name = copy_cache_entry_to_ondisk(ondisk, ce);
1719                memcpy(name, to_remove_vi, prefix_size);
1720                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1721
1722                strbuf_splice(previous_name, common, to_remove,
1723                              ce->name + common, ce_namelen(ce) - common);
1724        }
1725
1726        result = ce_write(c, fd, ondisk, size);
1727        free(ondisk);
1728        return result;
1729}
1730
1731static int has_racy_timestamp(struct index_state *istate)
1732{
1733        int entries = istate->cache_nr;
1734        int i;
1735
1736        for (i = 0; i < entries; i++) {
1737                struct cache_entry *ce = istate->cache[i];
1738                if (is_racy_timestamp(istate, ce))
1739                        return 1;
1740        }
1741        return 0;
1742}
1743
1744/*
1745 * Opportunisticly update the index but do not complain if we can't
1746 */
1747void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1748{
1749        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1750            !write_index(istate, lockfile->fd))
1751                commit_locked_index(lockfile);
1752        else
1753                rollback_lock_file(lockfile);
1754}
1755
1756int write_index(struct index_state *istate, int newfd)
1757{
1758        git_SHA_CTX c;
1759        struct cache_header hdr;
1760        int i, err, removed, extended, hdr_version;
1761        struct cache_entry **cache = istate->cache;
1762        int entries = istate->cache_nr;
1763        struct stat st;
1764        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1765
1766        for (i = removed = extended = 0; i < entries; i++) {
1767                if (cache[i]->ce_flags & CE_REMOVE)
1768                        removed++;
1769
1770                /* reduce extended entries if possible */
1771                cache[i]->ce_flags &= ~CE_EXTENDED;
1772                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1773                        extended++;
1774                        cache[i]->ce_flags |= CE_EXTENDED;
1775                }
1776        }
1777
1778        if (!istate->version)
1779                istate->version = INDEX_FORMAT_DEFAULT;
1780
1781        /* demote version 3 to version 2 when the latter suffices */
1782        if (istate->version == 3 || istate->version == 2)
1783                istate->version = extended ? 3 : 2;
1784
1785        hdr_version = istate->version;
1786
1787        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1788        hdr.hdr_version = htonl(hdr_version);
1789        hdr.hdr_entries = htonl(entries - removed);
1790
1791        git_SHA1_Init(&c);
1792        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1793                return -1;
1794
1795        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1796        for (i = 0; i < entries; i++) {
1797                struct cache_entry *ce = cache[i];
1798                if (ce->ce_flags & CE_REMOVE)
1799                        continue;
1800                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1801                        ce_smudge_racily_clean_entry(ce);
1802                if (is_null_sha1(ce->sha1))
1803                        return error("cache entry has null sha1: %s", ce->name);
1804                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1805                        return -1;
1806        }
1807        strbuf_release(&previous_name_buf);
1808
1809        /* Write extension data here */
1810        if (istate->cache_tree) {
1811                struct strbuf sb = STRBUF_INIT;
1812
1813                cache_tree_write(&sb, istate->cache_tree);
1814                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1815                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1816                strbuf_release(&sb);
1817                if (err)
1818                        return -1;
1819        }
1820        if (istate->resolve_undo) {
1821                struct strbuf sb = STRBUF_INIT;
1822
1823                resolve_undo_write(&sb, istate->resolve_undo);
1824                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1825                                             sb.len) < 0
1826                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1827                strbuf_release(&sb);
1828                if (err)
1829                        return -1;
1830        }
1831
1832        if (ce_flush(&c, newfd) || fstat(newfd, &st))
1833                return -1;
1834        istate->timestamp.sec = (unsigned int)st.st_mtime;
1835        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1836        return 0;
1837}
1838
1839/*
1840 * Read the index file that is potentially unmerged into given
1841 * index_state, dropping any unmerged entries.  Returns true if
1842 * the index is unmerged.  Callers who want to refuse to work
1843 * from an unmerged state can call this and check its return value,
1844 * instead of calling read_cache().
1845 */
1846int read_index_unmerged(struct index_state *istate)
1847{
1848        int i;
1849        int unmerged = 0;
1850
1851        read_index(istate);
1852        for (i = 0; i < istate->cache_nr; i++) {
1853                struct cache_entry *ce = istate->cache[i];
1854                struct cache_entry *new_ce;
1855                int size, len;
1856
1857                if (!ce_stage(ce))
1858                        continue;
1859                unmerged = 1;
1860                len = ce_namelen(ce);
1861                size = cache_entry_size(len);
1862                new_ce = xcalloc(1, size);
1863                memcpy(new_ce->name, ce->name, len);
1864                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
1865                new_ce->ce_namelen = len;
1866                new_ce->ce_mode = ce->ce_mode;
1867                if (add_index_entry(istate, new_ce, 0))
1868                        return error("%s: cannot drop to stage #0",
1869                                     ce->name);
1870                i = index_name_pos(istate, new_ce->name, len);
1871        }
1872        return unmerged;
1873}
1874
1875/*
1876 * Returns 1 if the path is an "other" path with respect to
1877 * the index; that is, the path is not mentioned in the index at all,
1878 * either as a file, a directory with some files in the index,
1879 * or as an unmerged entry.
1880 *
1881 * We helpfully remove a trailing "/" from directories so that
1882 * the output of read_directory can be used as-is.
1883 */
1884int index_name_is_other(const struct index_state *istate, const char *name,
1885                int namelen)
1886{
1887        int pos;
1888        if (namelen && name[namelen - 1] == '/')
1889                namelen--;
1890        pos = index_name_pos(istate, name, namelen);
1891        if (0 <= pos)
1892                return 0;       /* exact match */
1893        pos = -pos - 1;
1894        if (pos < istate->cache_nr) {
1895                struct cache_entry *ce = istate->cache[pos];
1896                if (ce_namelen(ce) == namelen &&
1897                    !memcmp(ce->name, name, namelen))
1898                        return 0; /* Yup, this one exists unmerged */
1899        }
1900        return 1;
1901}
1902
1903void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
1904{
1905        int pos, len;
1906        unsigned long sz;
1907        enum object_type type;
1908        void *data;
1909
1910        len = strlen(path);
1911        pos = index_name_pos(istate, path, len);
1912        if (pos < 0) {
1913                /*
1914                 * We might be in the middle of a merge, in which
1915                 * case we would read stage #2 (ours).
1916                 */
1917                int i;
1918                for (i = -pos - 1;
1919                     (pos < 0 && i < istate->cache_nr &&
1920                      !strcmp(istate->cache[i]->name, path));
1921                     i++)
1922                        if (ce_stage(istate->cache[i]) == 2)
1923                                pos = i;
1924        }
1925        if (pos < 0)
1926                return NULL;
1927        data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
1928        if (!data || type != OBJ_BLOB) {
1929                free(data);
1930                return NULL;
1931        }
1932        if (size)
1933                *size = sz;
1934        return data;
1935}