read-cache.c

   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
  12 /* Index extensions.
  13  *
  14  * The first letter should be 'A'..'Z' for extensions that are not
  15  * necessary for a correct operation (i.e. optimization data).
  16  * When new extensions are added that _needs_ to be understood in
  17  * order to correctly interpret the index file, pick character that
  18  * is outside the range, to cause the reader to abort.
  19  */
  20
  21 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  22 #define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  23
  24 struct index_state the_index;
  25
  26 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  27 {
  28         istate->cache[nr] = ce;
  29         add_name_hash(istate, ce);
  30 }
  31
  32 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  33 {
  34         struct cache_entry *old = istate->cache[nr];
  35
  36         remove_name_hash(old);
  37         set_index_entry(istate, nr, ce);
  38         istate->cache_changed = 1;
  39 }
  40
  41 /*
  42  * This only updates the "non-critical" parts of the directory
  43  * cache, ie the parts that aren't tracked by GIT, and only used
  44  * to validate the cache.
  45  */
  46 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
  47 {
  48         ce->ce_ctime = st->st_ctime;
  49         ce->ce_mtime = st->st_mtime;
  50         ce->ce_dev = st->st_dev;
  51         ce->ce_ino = st->st_ino;
  52         ce->ce_uid = st->st_uid;
  53         ce->ce_gid = st->st_gid;
  54         ce->ce_size = st->st_size;
  55
  56         if (assume_unchanged)
  57                 ce->ce_flags |= CE_VALID;
  58
  59         if (S_ISREG(st->st_mode))
  60                 ce_mark_uptodate(ce);
  61 }
  62
  63 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
  64 {
  65         int match = -1;
  66         int fd = open(ce->name, O_RDONLY);
  67
  68         if (fd >= 0) {
  69                 unsigned char sha1[20];
  70                 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
  71                         match = hashcmp(sha1, ce->sha1);
  72                 /* index_fd() closed the file descriptor already */
  73         }
  74         return match;
  75 }
  76
  77 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
  78 {
  79         int match = -1;
  80         char *target;
  81         void *buffer;
  82         unsigned long size;
  83         enum object_type type;
  84         int len;
  85
  86         target = xmalloc(expected_size);
  87         len = readlink(ce->name, target, expected_size);
  88         if (len != expected_size) {
  89                 free(target);
  90                 return -1;
  91         }
  92         buffer = read_sha1_file(ce->sha1, &type, &size);
  93         if (!buffer) {
  94                 free(target);
  95                 return -1;
  96         }
  97         if (size == expected_size)
  98                 match = memcmp(buffer, target, size);
  99         free(buffer);
 100         free(target);
 101         return match;
 102 }
 103
 104 static int ce_compare_gitlink(struct cache_entry *ce)
 105 {
 106         unsigned char sha1[20];
 107
 108         /*
 109          * We don't actually require that the .git directory
 110          * under GITLINK directory be a valid git directory. It
 111          * might even be missing (in case nobody populated that
 112          * sub-project).
 113          *
 114          * If so, we consider it always to match.
 115          */
 116         if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 117                 return 0;
 118         return hashcmp(sha1, ce->sha1);
 119 }
 120
 121 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
 122 {
 123         switch (st->st_mode & S_IFMT) {
 124         case S_IFREG:
 125                 if (ce_compare_data(ce, st))
 126                         return DATA_CHANGED;
 127                 break;
 128         case S_IFLNK:
 129                 if (ce_compare_link(ce, xsize_t(st->st_size)))
 130                         return DATA_CHANGED;
 131                 break;
 132         case S_IFDIR:
 133                 if (S_ISGITLINK(ce->ce_mode))
 134                         return 0;
 135         default:
 136                 return TYPE_CHANGED;
 137         }
 138         return 0;
 139 }
 140
 141 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 142 {
 143         unsigned int changed = 0;
 144
 145         if (ce->ce_flags & CE_REMOVE)
 146                 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 147
 148         switch (ce->ce_mode & S_IFMT) {
 149         case S_IFREG:
 150                 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 151                 /* We consider only the owner x bit to be relevant for
 152                  * "mode changes"
 153                  */
 154                 if (trust_executable_bit &&
 155                     (0100 & (ce->ce_mode ^ st->st_mode)))
 156                         changed |= MODE_CHANGED;
 157                 break;
 158         case S_IFLNK:
 159                 if (!S_ISLNK(st->st_mode) &&
 160                     (has_symlinks || !S_ISREG(st->st_mode)))
 161                         changed |= TYPE_CHANGED;
 162                 break;
 163         case S_IFGITLINK:
 164                 if (!S_ISDIR(st->st_mode))
 165                         changed |= TYPE_CHANGED;
 166                 else if (ce_compare_gitlink(ce))
 167                         changed |= DATA_CHANGED;
 168                 return changed;
 169         default:
 170                 die("internal error: ce_mode is %o", ce->ce_mode);
 171         }
 172         if (ce->ce_mtime != (unsigned int) st->st_mtime)
 173                 changed |= MTIME_CHANGED;
 174         if (ce->ce_ctime != (unsigned int) st->st_ctime)
 175                 changed |= CTIME_CHANGED;
 176
 177         if (ce->ce_uid != (unsigned int) st->st_uid ||
 178             ce->ce_gid != (unsigned int) st->st_gid)
 179                 changed |= OWNER_CHANGED;
 180         if (ce->ce_ino != (unsigned int) st->st_ino)
 181                 changed |= INODE_CHANGED;
 182
 183 #ifdef USE_STDEV
 184         /*
 185          * st_dev breaks on network filesystems where different
 186          * clients will have different views of what "device"
 187          * the filesystem is on
 188          */
 189         if (ce->ce_dev != (unsigned int) st->st_dev)
 190                 changed |= INODE_CHANGED;
 191 #endif
 192
 193         if (ce->ce_size != (unsigned int) st->st_size)
 194                 changed |= DATA_CHANGED;
 195
 196         return changed;
 197 }
 198
 199 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
 200 {
 201         return (istate->timestamp &&
 202                 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
 203 }
 204
 205 int ie_match_stat(const struct index_state *istate,
 206                   struct cache_entry *ce, struct stat *st,
 207                   unsigned int options)
 208 {
 209         unsigned int changed;
 210         int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 211         int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 212
 213         /*
 214          * If it's marked as always valid in the index, it's
 215          * valid whatever the checked-out copy says.
 216          */
 217         if (!ignore_valid && (ce->ce_flags & CE_VALID))
 218                 return 0;
 219
 220         changed = ce_match_stat_basic(ce, st);
 221
 222         /*
 223          * Within 1 second of this sequence:
 224          *      echo xyzzy >file && git-update-index --add file
 225          * running this command:
 226          *      echo frotz >file
 227          * would give a falsely clean cache entry.  The mtime and
 228          * length match the cache, and other stat fields do not change.
 229          *
 230          * We could detect this at update-index time (the cache entry
 231          * being registered/updated records the same time as "now")
 232          * and delay the return from git-update-index, but that would
 233          * effectively mean we can make at most one commit per second,
 234          * which is not acceptable.  Instead, we check cache entries
 235          * whose mtime are the same as the index file timestamp more
 236          * carefully than others.
 237          */
 238         if (!changed && is_racy_timestamp(istate, ce)) {
 239                 if (assume_racy_is_modified)
 240                         changed |= DATA_CHANGED;
 241                 else
 242                         changed |= ce_modified_check_fs(ce, st);
 243         }
 244
 245         return changed;
 246 }
 247
 248 int ie_modified(const struct index_state *istate,
 249                 struct cache_entry *ce, struct stat *st, unsigned int options)
 250 {
 251         int changed, changed_fs;
 252
 253         changed = ie_match_stat(istate, ce, st, options);
 254         if (!changed)
 255                 return 0;
 256         /*
 257          * If the mode or type has changed, there's no point in trying
 258          * to refresh the entry - it's not going to match
 259          */
 260         if (changed & (MODE_CHANGED | TYPE_CHANGED))
 261                 return changed;
 262
 263         /* Immediately after read-tree or update-index --cacheinfo,
 264          * the length field is zero.  For other cases the ce_size
 265          * should match the SHA1 recorded in the index entry.
 266          */
 267         if ((changed & DATA_CHANGED) && ce->ce_size != 0)
 268                 return changed;
 269
 270         changed_fs = ce_modified_check_fs(ce, st);
 271         if (changed_fs)
 272                 return changed | changed_fs;
 273         return 0;
 274 }
 275
 276 int base_name_compare(const char *name1, int len1, int mode1,
 277                       const char *name2, int len2, int mode2)
 278 {
 279         unsigned char c1, c2;
 280         int len = len1 < len2 ? len1 : len2;
 281         int cmp;
 282
 283         cmp = memcmp(name1, name2, len);
 284         if (cmp)
 285                 return cmp;
 286         c1 = name1[len];
 287         c2 = name2[len];
 288         if (!c1 && S_ISDIR(mode1))
 289                 c1 = '/';
 290         if (!c2 && S_ISDIR(mode2))
 291                 c2 = '/';
 292         return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 293 }
 294
 295 /*
 296  * df_name_compare() is identical to base_name_compare(), except it
 297  * compares conflicting directory/file entries as equal. Note that
 298  * while a directory name compares as equal to a regular file, they
 299  * then individually compare _differently_ to a filename that has
 300  * a dot after the basename (because '\0' < '.' < '/').
 301  *
 302  * This is used by routines that want to traverse the git namespace
 303  * but then handle conflicting entries together when possible.
 304  */
 305 int df_name_compare(const char *name1, int len1, int mode1,
 306                     const char *name2, int len2, int mode2)
 307 {
 308         int len = len1 < len2 ? len1 : len2, cmp;
 309         unsigned char c1, c2;
 310
 311         cmp = memcmp(name1, name2, len);
 312         if (cmp)
 313                 return cmp;
 314         /* Directories and files compare equal (same length, same name) */
 315         if (len1 == len2)
 316                 return 0;
 317         c1 = name1[len];
 318         if (!c1 && S_ISDIR(mode1))
 319                 c1 = '/';
 320         c2 = name2[len];
 321         if (!c2 && S_ISDIR(mode2))
 322                 c2 = '/';
 323         if (c1 == '/' && !c2)
 324                 return 0;
 325         if (c2 == '/' && !c1)
 326                 return 0;
 327         return c1 - c2;
 328 }
 329
 330 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 331 {
 332         int len1 = flags1 & CE_NAMEMASK;
 333         int len2 = flags2 & CE_NAMEMASK;
 334         int len = len1 < len2 ? len1 : len2;
 335         int cmp;
 336
 337         cmp = memcmp(name1, name2, len);
 338         if (cmp)
 339                 return cmp;
 340         if (len1 < len2)
 341                 return -1;
 342         if (len1 > len2)
 343                 return 1;
 344
 345         /* Compare stages  */
 346         flags1 &= CE_STAGEMASK;
 347         flags2 &= CE_STAGEMASK;
 348
 349         if (flags1 < flags2)
 350                 return -1;
 351         if (flags1 > flags2)
 352                 return 1;
 353         return 0;
 354 }
 355
 356 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 357 {
 358         int first, last;
 359
 360         first = 0;
 361         last = istate->cache_nr;
 362         while (last > first) {
 363                 int next = (last + first) >> 1;
 364                 struct cache_entry *ce = istate->cache[next];
 365                 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
 366                 if (!cmp)
 367                         return next;
 368                 if (cmp < 0) {
 369                         last = next;
 370                         continue;
 371                 }
 372                 first = next+1;
 373         }
 374         return -first-1;
 375 }
 376
 377 /* Remove entry, return true if there are more entries to go.. */
 378 int remove_index_entry_at(struct index_state *istate, int pos)
 379 {
 380         struct cache_entry *ce = istate->cache[pos];
 381
 382         remove_name_hash(ce);
 383         istate->cache_changed = 1;
 384         istate->cache_nr--;
 385         if (pos >= istate->cache_nr)
 386                 return 0;
 387         memmove(istate->cache + pos,
 388                 istate->cache + pos + 1,
 389                 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 390         return 1;
 391 }
 392
 393 int remove_file_from_index(struct index_state *istate, const char *path)
 394 {
 395         int pos = index_name_pos(istate, path, strlen(path));
 396         if (pos < 0)
 397                 pos = -pos-1;
 398         cache_tree_invalidate_path(istate->cache_tree, path);
 399         while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 400                 remove_index_entry_at(istate, pos);
 401         return 0;
 402 }
 403
 404 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 405 {
 406         return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 407 }
 408
 409 static int index_name_pos_also_unmerged(struct index_state *istate,
 410         const char *path, int namelen)
 411 {
 412         int pos = index_name_pos(istate, path, namelen);
 413         struct cache_entry *ce;
 414
 415         if (pos >= 0)
 416                 return pos;
 417
 418         /* maybe unmerged? */
 419         pos = -1 - pos;
 420         if (pos >= istate->cache_nr ||
 421                         compare_name((ce = istate->cache[pos]), path, namelen))
 422                 return -1;
 423
 424         /* order of preference: stage 2, 1, 3 */
 425         if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 426                         ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 427                         !compare_name(ce, path, namelen))
 428                 pos++;
 429         return pos;
 430 }
 431
 432 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 433 {
 434         int len = ce_namelen(ce);
 435         return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 436 }
 437
 438 /*
 439  * If we add a filename that aliases in the cache, we will use the
 440  * name that we already have - but we don't want to update the same
 441  * alias twice, because that implies that there were actually two
 442  * different files with aliasing names!
 443  *
 444  * So we use the CE_ADDED flag to verify that the alias was an old
 445  * one before we accept it as
 446  */
 447 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
 448 {
 449         int len;
 450         struct cache_entry *new;
 451
 452         if (alias->ce_flags & CE_ADDED)
 453                 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 454
 455         /* Ok, create the new entry using the name of the existing alias */
 456         len = ce_namelen(alias);
 457         new = xcalloc(1, cache_entry_size(len));
 458         memcpy(new->name, alias->name, len);
 459         copy_cache_entry(new, ce);
 460         free(ce);
 461         return new;
 462 }
 463
 464 int add_file_to_index(struct index_state *istate, const char *path, int verbose)
 465 {
 466         int size, namelen;
 467         struct stat st;
 468         struct cache_entry *ce, *alias;
 469         unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
 470
 471         if (lstat(path, &st))
 472                 die("%s: unable to stat (%s)", path, strerror(errno));
 473
 474         if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode))
 475                 die("%s: can only add regular files, symbolic links or git-directories", path);
 476
 477         namelen = strlen(path);
 478         if (S_ISDIR(st.st_mode)) {
 479                 while (namelen && path[namelen-1] == '/')
 480                         namelen--;
 481         }
 482         size = cache_entry_size(namelen);
 483         ce = xcalloc(1, size);
 484         memcpy(ce->name, path, namelen);
 485         ce->ce_flags = namelen;
 486         fill_stat_cache_info(ce, &st);
 487
 488         if (trust_executable_bit && has_symlinks)
 489                 ce->ce_mode = create_ce_mode(st.st_mode);
 490         else {
 491                 /* If there is an existing entry, pick the mode bits and type
 492                  * from it, otherwise assume unexecutable regular file.
 493                  */
 494                 struct cache_entry *ent;
 495                 int pos = index_name_pos_also_unmerged(istate, path, namelen);
 496
 497                 ent = (0 <= pos) ? istate->cache[pos] : NULL;
 498                 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode);
 499         }
 500
 501         alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 502         if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, &st, ce_option)) {
 503                 /* Nothing changed, really */
 504                 free(ce);
 505                 ce_mark_uptodate(alias);
 506                 alias->ce_flags |= CE_ADDED;
 507                 return 0;
 508         }
 509         if (index_path(ce->sha1, path, &st, 1))
 510                 die("unable to index file %s", path);
 511         if (ignore_case && alias && different_name(ce, alias))
 512                 ce = create_alias_ce(ce, alias);
 513         ce->ce_flags |= CE_ADDED;
 514         if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE))
 515                 die("unable to add %s to index",path);
 516         if (verbose)
 517                 printf("add '%s'\n", path);
 518         return 0;
 519 }
 520
 521 struct cache_entry *make_cache_entry(unsigned int mode,
 522                 const unsigned char *sha1, const char *path, int stage,
 523                 int refresh)
 524 {
 525         int size, len;
 526         struct cache_entry *ce;
 527
 528         if (!verify_path(path))
 529                 return NULL;
 530
 531         len = strlen(path);
 532         size = cache_entry_size(len);
 533         ce = xcalloc(1, size);
 534
 535         hashcpy(ce->sha1, sha1);
 536         memcpy(ce->name, path, len);
 537         ce->ce_flags = create_ce_flags(len, stage);
 538         ce->ce_mode = create_ce_mode(mode);
 539
 540         if (refresh)
 541                 return refresh_cache_entry(ce, 0);
 542
 543         return ce;
 544 }
 545
 546 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 547 {
 548         int len = ce_namelen(a);
 549         return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 550 }
 551
 552 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
 553 {
 554         const char *match, *name;
 555         int len;
 556
 557         if (!pathspec)
 558                 return 1;
 559
 560         len = ce_namelen(ce);
 561         name = ce->name;
 562         while ((match = *pathspec++) != NULL) {
 563                 int matchlen = strlen(match);
 564                 if (matchlen > len)
 565                         continue;
 566                 if (memcmp(name, match, matchlen))
 567                         continue;
 568                 if (matchlen && name[matchlen-1] == '/')
 569                         return 1;
 570                 if (name[matchlen] == '/' || !name[matchlen])
 571                         return 1;
 572                 if (!matchlen)
 573                         return 1;
 574         }
 575         return 0;
 576 }
 577
 578 /*
 579  * We fundamentally don't like some paths: we don't want
 580  * dot or dot-dot anywhere, and for obvious reasons don't
 581  * want to recurse into ".git" either.
 582  *
 583  * Also, we don't want double slashes or slashes at the
 584  * end that can make pathnames ambiguous.
 585  */
 586 static int verify_dotfile(const char *rest)
 587 {
 588         /*
 589          * The first character was '.', but that
 590          * has already been discarded, we now test
 591          * the rest.
 592          */
 593         switch (*rest) {
 594         /* "." is not allowed */
 595         case '\0': case '/':
 596                 return 0;
 597
 598         /*
 599          * ".git" followed by  NUL or slash is bad. This
 600          * shares the path end test with the ".." case.
 601          */
 602         case 'g':
 603                 if (rest[1] != 'i')
 604                         break;
 605                 if (rest[2] != 't')
 606                         break;
 607                 rest += 2;
 608         /* fallthrough */
 609         case '.':
 610                 if (rest[1] == '\0' || rest[1] == '/')
 611                         return 0;
 612         }
 613         return 1;
 614 }
 615
 616 int verify_path(const char *path)
 617 {
 618         char c;
 619
 620         goto inside;
 621         for (;;) {
 622                 if (!c)
 623                         return 1;
 624                 if (c == '/') {
 625 inside:
 626                         c = *path++;
 627                         switch (c) {
 628                         default:
 629                                 continue;
 630                         case '/': case '\0':
 631                                 break;
 632                         case '.':
 633                                 if (verify_dotfile(path))
 634                                         continue;
 635                         }
 636                         return 0;
 637                 }
 638                 c = *path++;
 639         }
 640 }
 641
 642 /*
 643  * Do we have another file that has the beginning components being a
 644  * proper superset of the name we're trying to add?
 645  */
 646 static int has_file_name(struct index_state *istate,
 647                          const struct cache_entry *ce, int pos, int ok_to_replace)
 648 {
 649         int retval = 0;
 650         int len = ce_namelen(ce);
 651         int stage = ce_stage(ce);
 652         const char *name = ce->name;
 653
 654         while (pos < istate->cache_nr) {
 655                 struct cache_entry *p = istate->cache[pos++];
 656
 657                 if (len >= ce_namelen(p))
 658                         break;
 659                 if (memcmp(name, p->name, len))
 660                         break;
 661                 if (ce_stage(p) != stage)
 662                         continue;
 663                 if (p->name[len] != '/')
 664                         continue;
 665                 if (p->ce_flags & CE_REMOVE)
 666                         continue;
 667                 retval = -1;
 668                 if (!ok_to_replace)
 669                         break;
 670                 remove_index_entry_at(istate, --pos);
 671         }
 672         return retval;
 673 }
 674
 675 /*
 676  * Do we have another file with a pathname that is a proper
 677  * subset of the name we're trying to add?
 678  */
 679 static int has_dir_name(struct index_state *istate,
 680                         const struct cache_entry *ce, int pos, int ok_to_replace)
 681 {
 682         int retval = 0;
 683         int stage = ce_stage(ce);
 684         const char *name = ce->name;
 685         const char *slash = name + ce_namelen(ce);
 686
 687         for (;;) {
 688                 int len;
 689
 690                 for (;;) {
 691                         if (*--slash == '/')
 692                                 break;
 693                         if (slash <= ce->name)
 694                                 return retval;
 695                 }
 696                 len = slash - name;
 697
 698                 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
 699                 if (pos >= 0) {
 700                         /*
 701                          * Found one, but not so fast.  This could
 702                          * be a marker that says "I was here, but
 703                          * I am being removed".  Such an entry is
 704                          * not a part of the resulting tree, and
 705                          * it is Ok to have a directory at the same
 706                          * path.
 707                          */
 708                         if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 709                                 retval = -1;
 710                                 if (!ok_to_replace)
 711                                         break;
 712                                 remove_index_entry_at(istate, pos);
 713                                 continue;
 714                         }
 715                 }
 716                 else
 717                         pos = -pos-1;
 718
 719                 /*
 720                  * Trivial optimization: if we find an entry that
 721                  * already matches the sub-directory, then we know
 722                  * we're ok, and we can exit.
 723                  */
 724                 while (pos < istate->cache_nr) {
 725                         struct cache_entry *p = istate->cache[pos];
 726                         if ((ce_namelen(p) <= len) ||
 727                             (p->name[len] != '/') ||
 728                             memcmp(p->name, name, len))
 729                                 break; /* not our subdirectory */
 730                         if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 731                                 /*
 732                                  * p is at the same stage as our entry, and
 733                                  * is a subdirectory of what we are looking
 734                                  * at, so we cannot have conflicts at our
 735                                  * level or anything shorter.
 736                                  */
 737                                 return retval;
 738                         pos++;
 739                 }
 740         }
 741         return retval;
 742 }
 743
 744 /* We may be in a situation where we already have path/file and path
 745  * is being added, or we already have path and path/file is being
 746  * added.  Either one would result in a nonsense tree that has path
 747  * twice when git-write-tree tries to write it out.  Prevent it.
 748  *
 749  * If ok-to-replace is specified, we remove the conflicting entries
 750  * from the cache so the caller should recompute the insert position.
 751  * When this happens, we return non-zero.
 752  */
 753 static int check_file_directory_conflict(struct index_state *istate,
 754                                          const struct cache_entry *ce,
 755                                          int pos, int ok_to_replace)
 756 {
 757         int retval;
 758
 759         /*
 760          * When ce is an "I am going away" entry, we allow it to be added
 761          */
 762         if (ce->ce_flags & CE_REMOVE)
 763                 return 0;
 764
 765         /*
 766          * We check if the path is a sub-path of a subsequent pathname
 767          * first, since removing those will not change the position
 768          * in the array.
 769          */
 770         retval = has_file_name(istate, ce, pos, ok_to_replace);
 771
 772         /*
 773          * Then check if the path might have a clashing sub-directory
 774          * before it.
 775          */
 776         return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 777 }
 778
 779 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 780 {
 781         int pos;
 782         int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 783         int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 784         int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 785
 786         cache_tree_invalidate_path(istate->cache_tree, ce->name);
 787         pos = index_name_pos(istate, ce->name, ce->ce_flags);
 788
 789         /* existing match? Just replace it. */
 790         if (pos >= 0) {
 791                 replace_index_entry(istate, pos, ce);
 792                 return 0;
 793         }
 794         pos = -pos-1;
 795
 796         /*
 797          * Inserting a merged entry ("stage 0") into the index
 798          * will always replace all non-merged entries..
 799          */
 800         if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 801                 while (ce_same_name(istate->cache[pos], ce)) {
 802                         ok_to_add = 1;
 803                         if (!remove_index_entry_at(istate, pos))
 804                                 break;
 805                 }
 806         }
 807
 808         if (!ok_to_add)
 809                 return -1;
 810         if (!verify_path(ce->name))
 811                 return -1;
 812
 813         if (!skip_df_check &&
 814             check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 815                 if (!ok_to_replace)
 816                         return error("'%s' appears as both a file and as a directory",
 817                                      ce->name);
 818                 pos = index_name_pos(istate, ce->name, ce->ce_flags);
 819                 pos = -pos-1;
 820         }
 821         return pos + 1;
 822 }
 823
 824 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 825 {
 826         int pos;
 827
 828         if (option & ADD_CACHE_JUST_APPEND)
 829                 pos = istate->cache_nr;
 830         else {
 831                 int ret;
 832                 ret = add_index_entry_with_check(istate, ce, option);
 833                 if (ret <= 0)
 834                         return ret;
 835                 pos = ret - 1;
 836         }
 837
 838         /* Make sure the array is big enough .. */
 839         if (istate->cache_nr == istate->cache_alloc) {
 840                 istate->cache_alloc = alloc_nr(istate->cache_alloc);
 841                 istate->cache = xrealloc(istate->cache,
 842                                         istate->cache_alloc * sizeof(struct cache_entry *));
 843         }
 844
 845         /* Add it in.. */
 846         istate->cache_nr++;
 847         if (istate->cache_nr > pos + 1)
 848                 memmove(istate->cache + pos + 1,
 849                         istate->cache + pos,
 850                         (istate->cache_nr - pos - 1) * sizeof(ce));
 851         set_index_entry(istate, pos, ce);
 852         istate->cache_changed = 1;
 853         return 0;
 854 }
 855
 856 /*
 857  * "refresh" does not calculate a new sha1 file or bring the
 858  * cache up-to-date for mode/content changes. But what it
 859  * _does_ do is to "re-match" the stat information of a file
 860  * with the cache, so that you can refresh the cache for a
 861  * file that hasn't been changed but where the stat entry is
 862  * out of date.
 863  *
 864  * For example, you'd want to do this after doing a "git-read-tree",
 865  * to link up the stat cache details with the proper files.
 866  */
 867 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
 868                                              struct cache_entry *ce,
 869                                              unsigned int options, int *err)
 870 {
 871         struct stat st;
 872         struct cache_entry *updated;
 873         int changed, size;
 874         int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 875
 876         if (ce_uptodate(ce))
 877                 return ce;
 878
 879         if (lstat(ce->name, &st) < 0) {
 880                 if (err)
 881                         *err = errno;
 882                 return NULL;
 883         }
 884
 885         changed = ie_match_stat(istate, ce, &st, options);
 886         if (!changed) {
 887                 /*
 888                  * The path is unchanged.  If we were told to ignore
 889                  * valid bit, then we did the actual stat check and
 890                  * found that the entry is unmodified.  If the entry
 891                  * is not marked VALID, this is the place to mark it
 892                  * valid again, under "assume unchanged" mode.
 893                  */
 894                 if (ignore_valid && assume_unchanged &&
 895                     !(ce->ce_flags & CE_VALID))
 896                         ; /* mark this one VALID again */
 897                 else {
 898                         /*
 899                          * We do not mark the index itself "modified"
 900                          * because CE_UPTODATE flag is in-core only;
 901                          * we are not going to write this change out.
 902                          */
 903                         ce_mark_uptodate(ce);
 904                         return ce;
 905                 }
 906         }
 907
 908         if (ie_modified(istate, ce, &st, options)) {
 909                 if (err)
 910                         *err = EINVAL;
 911                 return NULL;
 912         }
 913
 914         size = ce_size(ce);
 915         updated = xmalloc(size);
 916         memcpy(updated, ce, size);
 917         fill_stat_cache_info(updated, &st);
 918         /*
 919          * If ignore_valid is not set, we should leave CE_VALID bit
 920          * alone.  Otherwise, paths marked with --no-assume-unchanged
 921          * (i.e. things to be edited) will reacquire CE_VALID bit
 922          * automatically, which is not really what we want.
 923          */
 924         if (!ignore_valid && assume_unchanged &&
 925             !(ce->ce_flags & CE_VALID))
 926                 updated->ce_flags &= ~CE_VALID;
 927
 928         return updated;
 929 }
 930
 931 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
 932 {
 933         int i;
 934         int has_errors = 0;
 935         int really = (flags & REFRESH_REALLY) != 0;
 936         int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
 937         int quiet = (flags & REFRESH_QUIET) != 0;
 938         int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
 939         unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
 940
 941         for (i = 0; i < istate->cache_nr; i++) {
 942                 struct cache_entry *ce, *new;
 943                 int cache_errno = 0;
 944
 945                 ce = istate->cache[i];
 946                 if (ce_stage(ce)) {
 947                         while ((i < istate->cache_nr) &&
 948                                ! strcmp(istate->cache[i]->name, ce->name))
 949                                 i++;
 950                         i--;
 951                         if (allow_unmerged)
 952                                 continue;
 953                         printf("%s: needs merge\n", ce->name);
 954                         has_errors = 1;
 955                         continue;
 956                 }
 957
 958                 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
 959                         continue;
 960
 961                 new = refresh_cache_ent(istate, ce, options, &cache_errno);
 962                 if (new == ce)
 963                         continue;
 964                 if (!new) {
 965                         if (not_new && cache_errno == ENOENT)
 966                                 continue;
 967                         if (really && cache_errno == EINVAL) {
 968                                 /* If we are doing --really-refresh that
 969                                  * means the index is not valid anymore.
 970                                  */
 971                                 ce->ce_flags &= ~CE_VALID;
 972                                 istate->cache_changed = 1;
 973                         }
 974                         if (quiet)
 975                                 continue;
 976                         printf("%s: needs update\n", ce->name);
 977                         has_errors = 1;
 978                         continue;
 979                 }
 980
 981                 replace_index_entry(istate, i, new);
 982         }
 983         return has_errors;
 984 }
 985
 986 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
 987 {
 988         return refresh_cache_ent(&the_index, ce, really, NULL);
 989 }
 990
 991 static int verify_hdr(struct cache_header *hdr, unsigned long size)
 992 {
 993         SHA_CTX c;
 994         unsigned char sha1[20];
 995
 996         if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
 997                 return error("bad signature");
 998         if (hdr->hdr_version != htonl(2))
 999                 return error("bad index version");
1000         SHA1_Init(&c);
1001         SHA1_Update(&c, hdr, size - 20);
1002         SHA1_Final(sha1, &c);
1003         if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1004                 return error("bad index file sha1 signature");
1005         return 0;
1006 }
1007
1008 static int read_index_extension(struct index_state *istate,
1009                                 const char *ext, void *data, unsigned long sz)
1010 {
1011         switch (CACHE_EXT(ext)) {
1012         case CACHE_EXT_TREE:
1013                 istate->cache_tree = cache_tree_read(data, sz);
1014                 break;
1015         default:
1016                 if (*ext < 'A' || 'Z' < *ext)
1017                         return error("index uses %.4s extension, which we do not understand",
1018                                      ext);
1019                 fprintf(stderr, "ignoring %.4s extension\n", ext);
1020                 break;
1021         }
1022         return 0;
1023 }
1024
1025 int read_index(struct index_state *istate)
1026 {
1027         return read_index_from(istate, get_index_file());
1028 }
1029
1030 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1031 {
1032         size_t len;
1033
1034         ce->ce_ctime = ntohl(ondisk->ctime.sec);
1035         ce->ce_mtime = ntohl(ondisk->mtime.sec);
1036         ce->ce_dev   = ntohl(ondisk->dev);
1037         ce->ce_ino   = ntohl(ondisk->ino);
1038         ce->ce_mode  = ntohl(ondisk->mode);
1039         ce->ce_uid   = ntohl(ondisk->uid);
1040         ce->ce_gid   = ntohl(ondisk->gid);
1041         ce->ce_size  = ntohl(ondisk->size);
1042         /* On-disk flags are just 16 bits */
1043         ce->ce_flags = ntohs(ondisk->flags);
1044         hashcpy(ce->sha1, ondisk->sha1);
1045
1046         len = ce->ce_flags & CE_NAMEMASK;
1047         if (len == CE_NAMEMASK)
1048                 len = strlen(ondisk->name);
1049         /*
1050          * NEEDSWORK: If the original index is crafted, this copy could
1051          * go unchecked.
1052          */
1053         memcpy(ce->name, ondisk->name, len + 1);
1054 }
1055
1056 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1057 {
1058         long per_entry;
1059
1060         per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1061
1062         /*
1063          * Alignment can cause differences. This should be "alignof", but
1064          * since that's a gcc'ism, just use the size of a pointer.
1065          */
1066         per_entry += sizeof(void *);
1067         return ondisk_size + entries*per_entry;
1068 }
1069
1070 /* remember to discard_cache() before reading a different cache! */
1071 int read_index_from(struct index_state *istate, const char *path)
1072 {
1073         int fd, i;
1074         struct stat st;
1075         unsigned long src_offset, dst_offset;
1076         struct cache_header *hdr;
1077         void *mmap;
1078         size_t mmap_size;
1079
1080         errno = EBUSY;
1081         if (istate->alloc)
1082                 return istate->cache_nr;
1083
1084         errno = ENOENT;
1085         istate->timestamp = 0;
1086         fd = open(path, O_RDONLY);
1087         if (fd < 0) {
1088                 if (errno == ENOENT)
1089                         return 0;
1090                 die("index file open failed (%s)", strerror(errno));
1091         }
1092
1093         if (fstat(fd, &st))
1094                 die("cannot stat the open index (%s)", strerror(errno));
1095
1096         errno = EINVAL;
1097         mmap_size = xsize_t(st.st_size);
1098         if (mmap_size < sizeof(struct cache_header) + 20)
1099                 die("index file smaller than expected");
1100
1101         mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1102         close(fd);
1103         if (mmap == MAP_FAILED)
1104                 die("unable to map index file");
1105
1106         hdr = mmap;
1107         if (verify_hdr(hdr, mmap_size) < 0)
1108                 goto unmap;
1109
1110         istate->cache_nr = ntohl(hdr->hdr_entries);
1111         istate->cache_alloc = alloc_nr(istate->cache_nr);
1112         istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1113
1114         /*
1115          * The disk format is actually larger than the in-memory format,
1116          * due to space for nsec etc, so even though the in-memory one
1117          * has room for a few  more flags, we can allocate using the same
1118          * index size
1119          */
1120         istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1121
1122         src_offset = sizeof(*hdr);
1123         dst_offset = 0;
1124         for (i = 0; i < istate->cache_nr; i++) {
1125                 struct ondisk_cache_entry *disk_ce;
1126                 struct cache_entry *ce;
1127
1128                 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1129                 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1130                 convert_from_disk(disk_ce, ce);
1131                 set_index_entry(istate, i, ce);
1132
1133                 src_offset += ondisk_ce_size(ce);
1134                 dst_offset += ce_size(ce);
1135         }
1136         istate->timestamp = st.st_mtime;
1137         while (src_offset <= mmap_size - 20 - 8) {
1138                 /* After an array of active_nr index entries,
1139                  * there can be arbitrary number of extended
1140                  * sections, each of which is prefixed with
1141                  * extension name (4-byte) and section length
1142                  * in 4-byte network byte order.
1143                  */
1144                 unsigned long extsize;
1145                 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1146                 extsize = ntohl(extsize);
1147                 if (read_index_extension(istate,
1148                                          (const char *) mmap + src_offset,
1149                                          (char *) mmap + src_offset + 8,
1150                                          extsize) < 0)
1151                         goto unmap;
1152                 src_offset += 8;
1153                 src_offset += extsize;
1154         }
1155         munmap(mmap, mmap_size);
1156         return istate->cache_nr;
1157
1158 unmap:
1159         munmap(mmap, mmap_size);
1160         errno = EINVAL;
1161         die("index file corrupt");
1162 }
1163
1164 int discard_index(struct index_state *istate)
1165 {
1166         istate->cache_nr = 0;
1167         istate->cache_changed = 0;
1168         istate->timestamp = 0;
1169         free_hash(&istate->name_hash);
1170         cache_tree_free(&(istate->cache_tree));
1171         free(istate->alloc);
1172         istate->alloc = NULL;
1173
1174         /* no need to throw away allocated active_cache */
1175         return 0;
1176 }
1177
1178 int unmerged_index(const struct index_state *istate)
1179 {
1180         int i;
1181         for (i = 0; i < istate->cache_nr; i++) {
1182                 if (ce_stage(istate->cache[i]))
1183                         return 1;
1184         }
1185         return 0;
1186 }
1187
1188 #define WRITE_BUFFER_SIZE 8192
1189 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1190 static unsigned long write_buffer_len;
1191
1192 static int ce_write_flush(SHA_CTX *context, int fd)
1193 {
1194         unsigned int buffered = write_buffer_len;
1195         if (buffered) {
1196                 SHA1_Update(context, write_buffer, buffered);
1197                 if (write_in_full(fd, write_buffer, buffered) != buffered)
1198                         return -1;
1199                 write_buffer_len = 0;
1200         }
1201         return 0;
1202 }
1203
1204 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1205 {
1206         while (len) {
1207                 unsigned int buffered = write_buffer_len;
1208                 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1209                 if (partial > len)
1210                         partial = len;
1211                 memcpy(write_buffer + buffered, data, partial);
1212                 buffered += partial;
1213                 if (buffered == WRITE_BUFFER_SIZE) {
1214                         write_buffer_len = buffered;
1215                         if (ce_write_flush(context, fd))
1216                                 return -1;
1217                         buffered = 0;
1218                 }
1219                 write_buffer_len = buffered;
1220                 len -= partial;
1221                 data = (char *) data + partial;
1222         }
1223         return 0;
1224 }
1225
1226 static int write_index_ext_header(SHA_CTX *context, int fd,
1227                                   unsigned int ext, unsigned int sz)
1228 {
1229         ext = htonl(ext);
1230         sz = htonl(sz);
1231         return ((ce_write(context, fd, &ext, 4) < 0) ||
1232                 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1233 }
1234
1235 static int ce_flush(SHA_CTX *context, int fd)
1236 {
1237         unsigned int left = write_buffer_len;
1238
1239         if (left) {
1240                 write_buffer_len = 0;
1241                 SHA1_Update(context, write_buffer, left);
1242         }
1243
1244         /* Flush first if not enough space for SHA1 signature */
1245         if (left + 20 > WRITE_BUFFER_SIZE) {
1246                 if (write_in_full(fd, write_buffer, left) != left)
1247                         return -1;
1248                 left = 0;
1249         }
1250
1251         /* Append the SHA1 signature at the end */
1252         SHA1_Final(write_buffer + left, context);
1253         left += 20;
1254         return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1255 }
1256
1257 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1258 {
1259         /*
1260          * The only thing we care about in this function is to smudge the
1261          * falsely clean entry due to touch-update-touch race, so we leave
1262          * everything else as they are.  We are called for entries whose
1263          * ce_mtime match the index file mtime.
1264          */
1265         struct stat st;
1266
1267         if (lstat(ce->name, &st) < 0)
1268                 return;
1269         if (ce_match_stat_basic(ce, &st))
1270                 return;
1271         if (ce_modified_check_fs(ce, &st)) {
1272                 /* This is "racily clean"; smudge it.  Note that this
1273                  * is a tricky code.  At first glance, it may appear
1274                  * that it can break with this sequence:
1275                  *
1276                  * $ echo xyzzy >frotz
1277                  * $ git-update-index --add frotz
1278                  * $ : >frotz
1279                  * $ sleep 3
1280                  * $ echo filfre >nitfol
1281                  * $ git-update-index --add nitfol
1282                  *
1283                  * but it does not.  When the second update-index runs,
1284                  * it notices that the entry "frotz" has the same timestamp
1285                  * as index, and if we were to smudge it by resetting its
1286                  * size to zero here, then the object name recorded
1287                  * in index is the 6-byte file but the cached stat information
1288                  * becomes zero --- which would then match what we would
1289                  * obtain from the filesystem next time we stat("frotz").
1290                  *
1291                  * However, the second update-index, before calling
1292                  * this function, notices that the cached size is 6
1293                  * bytes and what is on the filesystem is an empty
1294                  * file, and never calls us, so the cached size information
1295                  * for "frotz" stays 6 which does not match the filesystem.
1296                  */
1297                 ce->ce_size = 0;
1298         }
1299 }
1300
1301 static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1302 {
1303         int size = ondisk_ce_size(ce);
1304         struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1305
1306         ondisk->ctime.sec = htonl(ce->ce_ctime);
1307         ondisk->ctime.nsec = 0;
1308         ondisk->mtime.sec = htonl(ce->ce_mtime);
1309         ondisk->mtime.nsec = 0;
1310         ondisk->dev  = htonl(ce->ce_dev);
1311         ondisk->ino  = htonl(ce->ce_ino);
1312         ondisk->mode = htonl(ce->ce_mode);
1313         ondisk->uid  = htonl(ce->ce_uid);
1314         ondisk->gid  = htonl(ce->ce_gid);
1315         ondisk->size = htonl(ce->ce_size);
1316         hashcpy(ondisk->sha1, ce->sha1);
1317         ondisk->flags = htons(ce->ce_flags);
1318         memcpy(ondisk->name, ce->name, ce_namelen(ce));
1319
1320         return ce_write(c, fd, ondisk, size);
1321 }
1322
1323 int write_index(const struct index_state *istate, int newfd)
1324 {
1325         SHA_CTX c;
1326         struct cache_header hdr;
1327         int i, err, removed;
1328         struct cache_entry **cache = istate->cache;
1329         int entries = istate->cache_nr;
1330
1331         for (i = removed = 0; i < entries; i++)
1332                 if (cache[i]->ce_flags & CE_REMOVE)
1333                         removed++;
1334
1335         hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1336         hdr.hdr_version = htonl(2);
1337         hdr.hdr_entries = htonl(entries - removed);
1338
1339         SHA1_Init(&c);
1340         if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1341                 return -1;
1342
1343         for (i = 0; i < entries; i++) {
1344                 struct cache_entry *ce = cache[i];
1345                 if (ce->ce_flags & CE_REMOVE)
1346                         continue;
1347                 if (is_racy_timestamp(istate, ce))
1348                         ce_smudge_racily_clean_entry(ce);
1349                 if (ce_write_entry(&c, newfd, ce) < 0)
1350                         return -1;
1351         }
1352
1353         /* Write extension data here */
1354         if (istate->cache_tree) {
1355                 struct strbuf sb;
1356
1357                 strbuf_init(&sb, 0);
1358                 cache_tree_write(&sb, istate->cache_tree);
1359                 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1360                         || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1361                 strbuf_release(&sb);
1362                 if (err)
1363                         return -1;
1364         }
1365         return ce_flush(&c, newfd);
1366 }