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 24struct index_state the_index; 25 26static unsigned int hash_name(const char *name, int namelen) 27{ 28 unsigned int hash = 0x123; 29 30 do { 31 unsigned char c = *name++; 32 hash = hash*101 + c; 33 } while (--namelen); 34 return hash; 35} 36 37static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 38{ 39 void **pos; 40 unsigned int hash = hash_name(ce->name, ce_namelen(ce)); 41 42 istate->cache[nr] = ce; 43 pos = insert_hash(hash, ce, &istate->name_hash); 44 if (pos) { 45 ce->next = *pos; 46 *pos = ce; 47 } 48} 49 50/* 51 * We don't actually *remove* it, we can just mark it invalid so that 52 * we won't find it in lookups. 53 * 54 * Not only would we have to search the lists (simple enough), but 55 * we'd also have to rehash other hash buckets in case this makes the 56 * hash bucket empty (common). So it's much better to just mark 57 * it. 58 */ 59static void remove_hash_entry(struct index_state *istate, struct cache_entry *ce) 60{ 61 ce->ce_flags |= CE_UNHASHED; 62} 63 64static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 65{ 66 struct cache_entry *old = istate->cache[nr]; 67 68 if (ce != old) { 69 remove_hash_entry(istate, old); 70 set_index_entry(istate, nr, ce); 71 } 72 istate->cache_changed = 1; 73} 74 75int index_name_exists(struct index_state *istate, const char *name, int namelen) 76{ 77 unsigned int hash = hash_name(name, namelen); 78 struct cache_entry *ce = lookup_hash(hash, &istate->name_hash); 79 80 while (ce) { 81 if (!(ce->ce_flags & CE_UNHASHED)) { 82 if (!cache_name_compare(name, namelen, ce->name, ce->ce_flags)) 83 return 1; 84 } 85 ce = ce->next; 86 } 87 return 0; 88} 89 90/* 91 * This only updates the "non-critical" parts of the directory 92 * cache, ie the parts that aren't tracked by GIT, and only used 93 * to validate the cache. 94 */ 95void fill_stat_cache_info(struct cache_entry *ce, struct stat *st) 96{ 97 ce->ce_ctime = st->st_ctime; 98 ce->ce_mtime = st->st_mtime; 99 ce->ce_dev = st->st_dev; 100 ce->ce_ino = st->st_ino; 101 ce->ce_uid = st->st_uid; 102 ce->ce_gid = st->st_gid; 103 ce->ce_size = st->st_size; 104 105 if (assume_unchanged) 106 ce->ce_flags |= CE_VALID; 107 108 if (S_ISREG(st->st_mode)) 109 ce_mark_uptodate(ce); 110} 111 112static int ce_compare_data(struct cache_entry *ce, struct stat *st) 113{ 114 int match = -1; 115 int fd = open(ce->name, O_RDONLY); 116 117 if (fd >= 0) { 118 unsigned char sha1[20]; 119 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name)) 120 match = hashcmp(sha1, ce->sha1); 121 /* index_fd() closed the file descriptor already */ 122 } 123 return match; 124} 125 126static int ce_compare_link(struct cache_entry *ce, size_t expected_size) 127{ 128 int match = -1; 129 char *target; 130 void *buffer; 131 unsigned long size; 132 enum object_type type; 133 int len; 134 135 target = xmalloc(expected_size); 136 len = readlink(ce->name, target, expected_size); 137 if (len != expected_size) { 138 free(target); 139 return -1; 140 } 141 buffer = read_sha1_file(ce->sha1, &type, &size); 142 if (!buffer) { 143 free(target); 144 return -1; 145 } 146 if (size == expected_size) 147 match = memcmp(buffer, target, size); 148 free(buffer); 149 free(target); 150 return match; 151} 152 153static int ce_compare_gitlink(struct cache_entry *ce) 154{ 155 unsigned char sha1[20]; 156 157 /* 158 * We don't actually require that the .git directory 159 * under GITLINK directory be a valid git directory. It 160 * might even be missing (in case nobody populated that 161 * sub-project). 162 * 163 * If so, we consider it always to match. 164 */ 165 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0) 166 return 0; 167 return hashcmp(sha1, ce->sha1); 168} 169 170static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st) 171{ 172 switch (st->st_mode & S_IFMT) { 173 case S_IFREG: 174 if (ce_compare_data(ce, st)) 175 return DATA_CHANGED; 176 break; 177 case S_IFLNK: 178 if (ce_compare_link(ce, xsize_t(st->st_size))) 179 return DATA_CHANGED; 180 break; 181 case S_IFDIR: 182 if (S_ISGITLINK(ce->ce_mode)) 183 return 0; 184 default: 185 return TYPE_CHANGED; 186 } 187 return 0; 188} 189 190static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st) 191{ 192 unsigned int changed = 0; 193 194 if (ce->ce_flags & CE_REMOVE) 195 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; 196 197 switch (ce->ce_mode & S_IFMT) { 198 case S_IFREG: 199 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0; 200 /* We consider only the owner x bit to be relevant for 201 * "mode changes" 202 */ 203 if (trust_executable_bit && 204 (0100 & (ce->ce_mode ^ st->st_mode))) 205 changed |= MODE_CHANGED; 206 break; 207 case S_IFLNK: 208 if (!S_ISLNK(st->st_mode) && 209 (has_symlinks || !S_ISREG(st->st_mode))) 210 changed |= TYPE_CHANGED; 211 break; 212 case S_IFGITLINK: 213 if (!S_ISDIR(st->st_mode)) 214 changed |= TYPE_CHANGED; 215 else if (ce_compare_gitlink(ce)) 216 changed |= DATA_CHANGED; 217 return changed; 218 default: 219 die("internal error: ce_mode is %o", ce->ce_mode); 220 } 221 if (ce->ce_mtime != (unsigned int) st->st_mtime) 222 changed |= MTIME_CHANGED; 223 if (ce->ce_ctime != (unsigned int) st->st_ctime) 224 changed |= CTIME_CHANGED; 225 226 if (ce->ce_uid != (unsigned int) st->st_uid || 227 ce->ce_gid != (unsigned int) st->st_gid) 228 changed |= OWNER_CHANGED; 229 if (ce->ce_ino != (unsigned int) st->st_ino) 230 changed |= INODE_CHANGED; 231 232#ifdef USE_STDEV 233 /* 234 * st_dev breaks on network filesystems where different 235 * clients will have different views of what "device" 236 * the filesystem is on 237 */ 238 if (ce->ce_dev != (unsigned int) st->st_dev) 239 changed |= INODE_CHANGED; 240#endif 241 242 if (ce->ce_size != (unsigned int) st->st_size) 243 changed |= DATA_CHANGED; 244 245 return changed; 246} 247 248static int is_racy_timestamp(struct index_state *istate, struct cache_entry *ce) 249{ 250 return (istate->timestamp && 251 ((unsigned int)istate->timestamp) <= ce->ce_mtime); 252} 253 254int ie_match_stat(struct index_state *istate, 255 struct cache_entry *ce, struct stat *st, 256 unsigned int options) 257{ 258 unsigned int changed; 259 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 260 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 261 262 /* 263 * If it's marked as always valid in the index, it's 264 * valid whatever the checked-out copy says. 265 */ 266 if (!ignore_valid && (ce->ce_flags & CE_VALID)) 267 return 0; 268 269 changed = ce_match_stat_basic(ce, st); 270 271 /* 272 * Within 1 second of this sequence: 273 * echo xyzzy >file && git-update-index --add file 274 * running this command: 275 * echo frotz >file 276 * would give a falsely clean cache entry. The mtime and 277 * length match the cache, and other stat fields do not change. 278 * 279 * We could detect this at update-index time (the cache entry 280 * being registered/updated records the same time as "now") 281 * and delay the return from git-update-index, but that would 282 * effectively mean we can make at most one commit per second, 283 * which is not acceptable. Instead, we check cache entries 284 * whose mtime are the same as the index file timestamp more 285 * carefully than others. 286 */ 287 if (!changed && is_racy_timestamp(istate, ce)) { 288 if (assume_racy_is_modified) 289 changed |= DATA_CHANGED; 290 else 291 changed |= ce_modified_check_fs(ce, st); 292 } 293 294 return changed; 295} 296 297int ie_modified(struct index_state *istate, 298 struct cache_entry *ce, struct stat *st, unsigned int options) 299{ 300 int changed, changed_fs; 301 302 changed = ie_match_stat(istate, ce, st, options); 303 if (!changed) 304 return 0; 305 /* 306 * If the mode or type has changed, there's no point in trying 307 * to refresh the entry - it's not going to match 308 */ 309 if (changed & (MODE_CHANGED | TYPE_CHANGED)) 310 return changed; 311 312 /* Immediately after read-tree or update-index --cacheinfo, 313 * the length field is zero. For other cases the ce_size 314 * should match the SHA1 recorded in the index entry. 315 */ 316 if ((changed & DATA_CHANGED) && ce->ce_size != 0) 317 return changed; 318 319 changed_fs = ce_modified_check_fs(ce, st); 320 if (changed_fs) 321 return changed | changed_fs; 322 return 0; 323} 324 325int base_name_compare(const char *name1, int len1, int mode1, 326 const char *name2, int len2, int mode2) 327{ 328 unsigned char c1, c2; 329 int len = len1 < len2 ? len1 : len2; 330 int cmp; 331 332 cmp = memcmp(name1, name2, len); 333 if (cmp) 334 return cmp; 335 c1 = name1[len]; 336 c2 = name2[len]; 337 if (!c1 && S_ISDIR(mode1)) 338 c1 = '/'; 339 if (!c2 && S_ISDIR(mode2)) 340 c2 = '/'; 341 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0; 342} 343 344int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2) 345{ 346 int len1 = flags1 & CE_NAMEMASK; 347 int len2 = flags2 & CE_NAMEMASK; 348 int len = len1 < len2 ? len1 : len2; 349 int cmp; 350 351 cmp = memcmp(name1, name2, len); 352 if (cmp) 353 return cmp; 354 if (len1 < len2) 355 return -1; 356 if (len1 > len2) 357 return 1; 358 359 /* Compare stages */ 360 flags1 &= CE_STAGEMASK; 361 flags2 &= CE_STAGEMASK; 362 363 if (flags1 < flags2) 364 return -1; 365 if (flags1 > flags2) 366 return 1; 367 return 0; 368} 369 370int index_name_pos(struct index_state *istate, const char *name, int namelen) 371{ 372 int first, last; 373 374 first = 0; 375 last = istate->cache_nr; 376 while (last > first) { 377 int next = (last + first) >> 1; 378 struct cache_entry *ce = istate->cache[next]; 379 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags); 380 if (!cmp) 381 return next; 382 if (cmp < 0) { 383 last = next; 384 continue; 385 } 386 first = next+1; 387 } 388 return -first-1; 389} 390 391/* Remove entry, return true if there are more entries to go.. */ 392int remove_index_entry_at(struct index_state *istate, int pos) 393{ 394 struct cache_entry *ce = istate->cache[pos]; 395 396 remove_hash_entry(istate, ce); 397 istate->cache_changed = 1; 398 istate->cache_nr--; 399 if (pos >= istate->cache_nr) 400 return 0; 401 memmove(istate->cache + pos, 402 istate->cache + pos + 1, 403 (istate->cache_nr - pos) * sizeof(struct cache_entry *)); 404 return 1; 405} 406 407int remove_file_from_index(struct index_state *istate, const char *path) 408{ 409 int pos = index_name_pos(istate, path, strlen(path)); 410 if (pos < 0) 411 pos = -pos-1; 412 cache_tree_invalidate_path(istate->cache_tree, path); 413 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 414 remove_index_entry_at(istate, pos); 415 return 0; 416} 417 418static int compare_name(struct cache_entry *ce, const char *path, int namelen) 419{ 420 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); 421} 422 423static int index_name_pos_also_unmerged(struct index_state *istate, 424 const char *path, int namelen) 425{ 426 int pos = index_name_pos(istate, path, namelen); 427 struct cache_entry *ce; 428 429 if (pos >= 0) 430 return pos; 431 432 /* maybe unmerged? */ 433 pos = -1 - pos; 434 if (pos >= istate->cache_nr || 435 compare_name((ce = istate->cache[pos]), path, namelen)) 436 return -1; 437 438 /* order of preference: stage 2, 1, 3 */ 439 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && 440 ce_stage((ce = istate->cache[pos + 1])) == 2 && 441 !compare_name(ce, path, namelen)) 442 pos++; 443 return pos; 444} 445 446int add_file_to_index(struct index_state *istate, const char *path, int verbose) 447{ 448 int size, namelen, pos; 449 struct stat st; 450 struct cache_entry *ce; 451 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY; 452 453 if (lstat(path, &st)) 454 die("%s: unable to stat (%s)", path, strerror(errno)); 455 456 if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode)) 457 die("%s: can only add regular files, symbolic links or git-directories", path); 458 459 namelen = strlen(path); 460 if (S_ISDIR(st.st_mode)) { 461 while (namelen && path[namelen-1] == '/') 462 namelen--; 463 } 464 size = cache_entry_size(namelen); 465 ce = xcalloc(1, size); 466 memcpy(ce->name, path, namelen); 467 ce->ce_flags = namelen; 468 fill_stat_cache_info(ce, &st); 469 470 if (trust_executable_bit && has_symlinks) 471 ce->ce_mode = create_ce_mode(st.st_mode); 472 else { 473 /* If there is an existing entry, pick the mode bits and type 474 * from it, otherwise assume unexecutable regular file. 475 */ 476 struct cache_entry *ent; 477 int pos = index_name_pos_also_unmerged(istate, path, namelen); 478 479 ent = (0 <= pos) ? istate->cache[pos] : NULL; 480 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode); 481 } 482 483 pos = index_name_pos(istate, ce->name, namelen); 484 if (0 <= pos && 485 !ce_stage(istate->cache[pos]) && 486 !ie_match_stat(istate, istate->cache[pos], &st, ce_option)) { 487 /* Nothing changed, really */ 488 free(ce); 489 ce_mark_uptodate(istate->cache[pos]); 490 return 0; 491 } 492 493 if (index_path(ce->sha1, path, &st, 1)) 494 die("unable to index file %s", path); 495 if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE)) 496 die("unable to add %s to index",path); 497 if (verbose) 498 printf("add '%s'\n", path); 499 return 0; 500} 501 502struct cache_entry *make_cache_entry(unsigned int mode, 503 const unsigned char *sha1, const char *path, int stage, 504 int refresh) 505{ 506 int size, len; 507 struct cache_entry *ce; 508 509 if (!verify_path(path)) 510 return NULL; 511 512 len = strlen(path); 513 size = cache_entry_size(len); 514 ce = xcalloc(1, size); 515 516 hashcpy(ce->sha1, sha1); 517 memcpy(ce->name, path, len); 518 ce->ce_flags = create_ce_flags(len, stage); 519 ce->ce_mode = create_ce_mode(mode); 520 521 if (refresh) 522 return refresh_cache_entry(ce, 0); 523 524 return ce; 525} 526 527int ce_same_name(struct cache_entry *a, struct cache_entry *b) 528{ 529 int len = ce_namelen(a); 530 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 531} 532 533int ce_path_match(const struct cache_entry *ce, const char **pathspec) 534{ 535 const char *match, *name; 536 int len; 537 538 if (!pathspec) 539 return 1; 540 541 len = ce_namelen(ce); 542 name = ce->name; 543 while ((match = *pathspec++) != NULL) { 544 int matchlen = strlen(match); 545 if (matchlen > len) 546 continue; 547 if (memcmp(name, match, matchlen)) 548 continue; 549 if (matchlen && name[matchlen-1] == '/') 550 return 1; 551 if (name[matchlen] == '/' || !name[matchlen]) 552 return 1; 553 if (!matchlen) 554 return 1; 555 } 556 return 0; 557} 558 559/* 560 * We fundamentally don't like some paths: we don't want 561 * dot or dot-dot anywhere, and for obvious reasons don't 562 * want to recurse into ".git" either. 563 * 564 * Also, we don't want double slashes or slashes at the 565 * end that can make pathnames ambiguous. 566 */ 567static int verify_dotfile(const char *rest) 568{ 569 /* 570 * The first character was '.', but that 571 * has already been discarded, we now test 572 * the rest. 573 */ 574 switch (*rest) { 575 /* "." is not allowed */ 576 case '\0': case '/': 577 return 0; 578 579 /* 580 * ".git" followed by NUL or slash is bad. This 581 * shares the path end test with the ".." case. 582 */ 583 case 'g': 584 if (rest[1] != 'i') 585 break; 586 if (rest[2] != 't') 587 break; 588 rest += 2; 589 /* fallthrough */ 590 case '.': 591 if (rest[1] == '\0' || rest[1] == '/') 592 return 0; 593 } 594 return 1; 595} 596 597int verify_path(const char *path) 598{ 599 char c; 600 601 goto inside; 602 for (;;) { 603 if (!c) 604 return 1; 605 if (c == '/') { 606inside: 607 c = *path++; 608 switch (c) { 609 default: 610 continue; 611 case '/': case '\0': 612 break; 613 case '.': 614 if (verify_dotfile(path)) 615 continue; 616 } 617 return 0; 618 } 619 c = *path++; 620 } 621} 622 623/* 624 * Do we have another file that has the beginning components being a 625 * proper superset of the name we're trying to add? 626 */ 627static int has_file_name(struct index_state *istate, 628 const struct cache_entry *ce, int pos, int ok_to_replace) 629{ 630 int retval = 0; 631 int len = ce_namelen(ce); 632 int stage = ce_stage(ce); 633 const char *name = ce->name; 634 635 while (pos < istate->cache_nr) { 636 struct cache_entry *p = istate->cache[pos++]; 637 638 if (len >= ce_namelen(p)) 639 break; 640 if (memcmp(name, p->name, len)) 641 break; 642 if (ce_stage(p) != stage) 643 continue; 644 if (p->name[len] != '/') 645 continue; 646 if (p->ce_flags & CE_REMOVE) 647 continue; 648 retval = -1; 649 if (!ok_to_replace) 650 break; 651 remove_index_entry_at(istate, --pos); 652 } 653 return retval; 654} 655 656/* 657 * Do we have another file with a pathname that is a proper 658 * subset of the name we're trying to add? 659 */ 660static int has_dir_name(struct index_state *istate, 661 const struct cache_entry *ce, int pos, int ok_to_replace) 662{ 663 int retval = 0; 664 int stage = ce_stage(ce); 665 const char *name = ce->name; 666 const char *slash = name + ce_namelen(ce); 667 668 for (;;) { 669 int len; 670 671 for (;;) { 672 if (*--slash == '/') 673 break; 674 if (slash <= ce->name) 675 return retval; 676 } 677 len = slash - name; 678 679 pos = index_name_pos(istate, name, create_ce_flags(len, stage)); 680 if (pos >= 0) { 681 /* 682 * Found one, but not so fast. This could 683 * be a marker that says "I was here, but 684 * I am being removed". Such an entry is 685 * not a part of the resulting tree, and 686 * it is Ok to have a directory at the same 687 * path. 688 */ 689 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) { 690 retval = -1; 691 if (!ok_to_replace) 692 break; 693 remove_index_entry_at(istate, pos); 694 continue; 695 } 696 } 697 else 698 pos = -pos-1; 699 700 /* 701 * Trivial optimization: if we find an entry that 702 * already matches the sub-directory, then we know 703 * we're ok, and we can exit. 704 */ 705 while (pos < istate->cache_nr) { 706 struct cache_entry *p = istate->cache[pos]; 707 if ((ce_namelen(p) <= len) || 708 (p->name[len] != '/') || 709 memcmp(p->name, name, len)) 710 break; /* not our subdirectory */ 711 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE)) 712 /* 713 * p is at the same stage as our entry, and 714 * is a subdirectory of what we are looking 715 * at, so we cannot have conflicts at our 716 * level or anything shorter. 717 */ 718 return retval; 719 pos++; 720 } 721 } 722 return retval; 723} 724 725/* We may be in a situation where we already have path/file and path 726 * is being added, or we already have path and path/file is being 727 * added. Either one would result in a nonsense tree that has path 728 * twice when git-write-tree tries to write it out. Prevent it. 729 * 730 * If ok-to-replace is specified, we remove the conflicting entries 731 * from the cache so the caller should recompute the insert position. 732 * When this happens, we return non-zero. 733 */ 734static int check_file_directory_conflict(struct index_state *istate, 735 const struct cache_entry *ce, 736 int pos, int ok_to_replace) 737{ 738 int retval; 739 740 /* 741 * When ce is an "I am going away" entry, we allow it to be added 742 */ 743 if (ce->ce_flags & CE_REMOVE) 744 return 0; 745 746 /* 747 * We check if the path is a sub-path of a subsequent pathname 748 * first, since removing those will not change the position 749 * in the array. 750 */ 751 retval = has_file_name(istate, ce, pos, ok_to_replace); 752 753 /* 754 * Then check if the path might have a clashing sub-directory 755 * before it. 756 */ 757 return retval + has_dir_name(istate, ce, pos, ok_to_replace); 758} 759 760static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option) 761{ 762 int pos; 763 int ok_to_add = option & ADD_CACHE_OK_TO_ADD; 764 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE; 765 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK; 766 767 cache_tree_invalidate_path(istate->cache_tree, ce->name); 768 pos = index_name_pos(istate, ce->name, ce->ce_flags); 769 770 /* existing match? Just replace it. */ 771 if (pos >= 0) { 772 replace_index_entry(istate, pos, ce); 773 return 0; 774 } 775 pos = -pos-1; 776 777 /* 778 * Inserting a merged entry ("stage 0") into the index 779 * will always replace all non-merged entries.. 780 */ 781 if (pos < istate->cache_nr && ce_stage(ce) == 0) { 782 while (ce_same_name(istate->cache[pos], ce)) { 783 ok_to_add = 1; 784 if (!remove_index_entry_at(istate, pos)) 785 break; 786 } 787 } 788 789 if (!ok_to_add) 790 return -1; 791 if (!verify_path(ce->name)) 792 return -1; 793 794 if (!skip_df_check && 795 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) { 796 if (!ok_to_replace) 797 return error("'%s' appears as both a file and as a directory", 798 ce->name); 799 pos = index_name_pos(istate, ce->name, ce->ce_flags); 800 pos = -pos-1; 801 } 802 return pos + 1; 803} 804 805int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option) 806{ 807 int pos; 808 809 if (option & ADD_CACHE_JUST_APPEND) 810 pos = istate->cache_nr; 811 else { 812 int ret; 813 ret = add_index_entry_with_check(istate, ce, option); 814 if (ret <= 0) 815 return ret; 816 pos = ret - 1; 817 } 818 819 /* Make sure the array is big enough .. */ 820 if (istate->cache_nr == istate->cache_alloc) { 821 istate->cache_alloc = alloc_nr(istate->cache_alloc); 822 istate->cache = xrealloc(istate->cache, 823 istate->cache_alloc * sizeof(struct cache_entry *)); 824 } 825 826 /* Add it in.. */ 827 istate->cache_nr++; 828 if (istate->cache_nr > pos + 1) 829 memmove(istate->cache + pos + 1, 830 istate->cache + pos, 831 (istate->cache_nr - pos - 1) * sizeof(ce)); 832 set_index_entry(istate, pos, ce); 833 istate->cache_changed = 1; 834 return 0; 835} 836 837/* 838 * "refresh" does not calculate a new sha1 file or bring the 839 * cache up-to-date for mode/content changes. But what it 840 * _does_ do is to "re-match" the stat information of a file 841 * with the cache, so that you can refresh the cache for a 842 * file that hasn't been changed but where the stat entry is 843 * out of date. 844 * 845 * For example, you'd want to do this after doing a "git-read-tree", 846 * to link up the stat cache details with the proper files. 847 */ 848static struct cache_entry *refresh_cache_ent(struct index_state *istate, 849 struct cache_entry *ce, 850 unsigned int options, int *err) 851{ 852 struct stat st; 853 struct cache_entry *updated; 854 int changed, size; 855 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 856 857 if (ce_uptodate(ce)) 858 return ce; 859 860 if (lstat(ce->name, &st) < 0) { 861 if (err) 862 *err = errno; 863 return NULL; 864 } 865 866 changed = ie_match_stat(istate, ce, &st, options); 867 if (!changed) { 868 /* 869 * The path is unchanged. If we were told to ignore 870 * valid bit, then we did the actual stat check and 871 * found that the entry is unmodified. If the entry 872 * is not marked VALID, this is the place to mark it 873 * valid again, under "assume unchanged" mode. 874 */ 875 if (ignore_valid && assume_unchanged && 876 !(ce->ce_flags & CE_VALID)) 877 ; /* mark this one VALID again */ 878 else { 879 /* 880 * We do not mark the index itself "modified" 881 * because CE_UPTODATE flag is in-core only; 882 * we are not going to write this change out. 883 */ 884 ce_mark_uptodate(ce); 885 return ce; 886 } 887 } 888 889 if (ie_modified(istate, ce, &st, options)) { 890 if (err) 891 *err = EINVAL; 892 return NULL; 893 } 894 895 size = ce_size(ce); 896 updated = xmalloc(size); 897 memcpy(updated, ce, size); 898 fill_stat_cache_info(updated, &st); 899 /* 900 * If ignore_valid is not set, we should leave CE_VALID bit 901 * alone. Otherwise, paths marked with --no-assume-unchanged 902 * (i.e. things to be edited) will reacquire CE_VALID bit 903 * automatically, which is not really what we want. 904 */ 905 if (!ignore_valid && assume_unchanged && 906 !(ce->ce_flags & CE_VALID)) 907 updated->ce_flags &= ~CE_VALID; 908 909 return updated; 910} 911 912int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen) 913{ 914 int i; 915 int has_errors = 0; 916 int really = (flags & REFRESH_REALLY) != 0; 917 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0; 918 int quiet = (flags & REFRESH_QUIET) != 0; 919 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0; 920 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0; 921 922 for (i = 0; i < istate->cache_nr; i++) { 923 struct cache_entry *ce, *new; 924 int cache_errno = 0; 925 926 ce = istate->cache[i]; 927 if (ce_stage(ce)) { 928 while ((i < istate->cache_nr) && 929 ! strcmp(istate->cache[i]->name, ce->name)) 930 i++; 931 i--; 932 if (allow_unmerged) 933 continue; 934 printf("%s: needs merge\n", ce->name); 935 has_errors = 1; 936 continue; 937 } 938 939 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen)) 940 continue; 941 942 new = refresh_cache_ent(istate, ce, options, &cache_errno); 943 if (new == ce) 944 continue; 945 if (!new) { 946 if (not_new && cache_errno == ENOENT) 947 continue; 948 if (really && cache_errno == EINVAL) { 949 /* If we are doing --really-refresh that 950 * means the index is not valid anymore. 951 */ 952 ce->ce_flags &= ~CE_VALID; 953 istate->cache_changed = 1; 954 } 955 if (quiet) 956 continue; 957 printf("%s: needs update\n", ce->name); 958 has_errors = 1; 959 continue; 960 } 961 962 replace_index_entry(istate, i, new); 963 } 964 return has_errors; 965} 966 967struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really) 968{ 969 return refresh_cache_ent(&the_index, ce, really, NULL); 970} 971 972static int verify_hdr(struct cache_header *hdr, unsigned long size) 973{ 974 SHA_CTX c; 975 unsigned char sha1[20]; 976 977 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE)) 978 return error("bad signature"); 979 if (hdr->hdr_version != htonl(2)) 980 return error("bad index version"); 981 SHA1_Init(&c); 982 SHA1_Update(&c, hdr, size - 20); 983 SHA1_Final(sha1, &c); 984 if (hashcmp(sha1, (unsigned char *)hdr + size - 20)) 985 return error("bad index file sha1 signature"); 986 return 0; 987} 988 989static int read_index_extension(struct index_state *istate, 990 const char *ext, void *data, unsigned long sz) 991{ 992 switch (CACHE_EXT(ext)) { 993 case CACHE_EXT_TREE: 994 istate->cache_tree = cache_tree_read(data, sz); 995 break; 996 default: 997 if (*ext < 'A' || 'Z' < *ext) 998 return error("index uses %.4s extension, which we do not understand", 999 ext);1000 fprintf(stderr, "ignoring %.4s extension\n", ext);1001 break;1002 }1003 return 0;1004}10051006int read_index(struct index_state *istate)1007{1008 return read_index_from(istate, get_index_file());1009}10101011static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)1012{1013 size_t len;10141015 ce->ce_ctime = ntohl(ondisk->ctime.sec);1016 ce->ce_mtime = ntohl(ondisk->mtime.sec);1017 ce->ce_dev = ntohl(ondisk->dev);1018 ce->ce_ino = ntohl(ondisk->ino);1019 ce->ce_mode = ntohl(ondisk->mode);1020 ce->ce_uid = ntohl(ondisk->uid);1021 ce->ce_gid = ntohl(ondisk->gid);1022 ce->ce_size = ntohl(ondisk->size);1023 /* On-disk flags are just 16 bits */1024 ce->ce_flags = ntohs(ondisk->flags);1025 hashcpy(ce->sha1, ondisk->sha1);10261027 len = ce->ce_flags & CE_NAMEMASK;1028 if (len == CE_NAMEMASK)1029 len = strlen(ondisk->name);1030 /*1031 * NEEDSWORK: If the original index is crafted, this copy could1032 * go unchecked.1033 */1034 memcpy(ce->name, ondisk->name, len + 1);1035}10361037static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)1038{1039 long per_entry;10401041 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);10421043 /*1044 * Alignment can cause differences. This should be "alignof", but1045 * since that's a gcc'ism, just use the size of a pointer.1046 */1047 per_entry += sizeof(void *);1048 return ondisk_size + entries*per_entry;1049}10501051/* remember to discard_cache() before reading a different cache! */1052int read_index_from(struct index_state *istate, const char *path)1053{1054 int fd, i;1055 struct stat st;1056 unsigned long src_offset, dst_offset;1057 struct cache_header *hdr;1058 void *mmap;1059 size_t mmap_size;10601061 errno = EBUSY;1062 if (istate->alloc)1063 return istate->cache_nr;10641065 errno = ENOENT;1066 istate->timestamp = 0;1067 fd = open(path, O_RDONLY);1068 if (fd < 0) {1069 if (errno == ENOENT)1070 return 0;1071 die("index file open failed (%s)", strerror(errno));1072 }10731074 if (fstat(fd, &st))1075 die("cannot stat the open index (%s)", strerror(errno));10761077 errno = EINVAL;1078 mmap_size = xsize_t(st.st_size);1079 if (mmap_size < sizeof(struct cache_header) + 20)1080 die("index file smaller than expected");10811082 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);1083 close(fd);1084 if (mmap == MAP_FAILED)1085 die("unable to map index file");10861087 hdr = mmap;1088 if (verify_hdr(hdr, mmap_size) < 0)1089 goto unmap;10901091 istate->cache_nr = ntohl(hdr->hdr_entries);1092 istate->cache_alloc = alloc_nr(istate->cache_nr);1093 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));10941095 /*1096 * The disk format is actually larger than the in-memory format,1097 * due to space for nsec etc, so even though the in-memory one1098 * has room for a few more flags, we can allocate using the same1099 * index size1100 */1101 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));11021103 src_offset = sizeof(*hdr);1104 dst_offset = 0;1105 for (i = 0; i < istate->cache_nr; i++) {1106 struct ondisk_cache_entry *disk_ce;1107 struct cache_entry *ce;11081109 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);1110 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);1111 convert_from_disk(disk_ce, ce);1112 set_index_entry(istate, i, ce);11131114 src_offset += ondisk_ce_size(ce);1115 dst_offset += ce_size(ce);1116 }1117 istate->timestamp = st.st_mtime;1118 while (src_offset <= mmap_size - 20 - 8) {1119 /* After an array of active_nr index entries,1120 * there can be arbitrary number of extended1121 * sections, each of which is prefixed with1122 * extension name (4-byte) and section length1123 * in 4-byte network byte order.1124 */1125 unsigned long extsize;1126 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);1127 extsize = ntohl(extsize);1128 if (read_index_extension(istate,1129 (const char *) mmap + src_offset,1130 (char *) mmap + src_offset + 8,1131 extsize) < 0)1132 goto unmap;1133 src_offset += 8;1134 src_offset += extsize;1135 }1136 munmap(mmap, mmap_size);1137 return istate->cache_nr;11381139unmap:1140 munmap(mmap, mmap_size);1141 errno = EINVAL;1142 die("index file corrupt");1143}11441145int discard_index(struct index_state *istate)1146{1147 istate->cache_nr = 0;1148 istate->cache_changed = 0;1149 istate->timestamp = 0;1150 free_hash(&istate->name_hash);1151 cache_tree_free(&(istate->cache_tree));1152 free(istate->alloc);1153 istate->alloc = NULL;11541155 /* no need to throw away allocated active_cache */1156 return 0;1157}11581159#define WRITE_BUFFER_SIZE 81921160static unsigned char write_buffer[WRITE_BUFFER_SIZE];1161static unsigned long write_buffer_len;11621163static int ce_write_flush(SHA_CTX *context, int fd)1164{1165 unsigned int buffered = write_buffer_len;1166 if (buffered) {1167 SHA1_Update(context, write_buffer, buffered);1168 if (write_in_full(fd, write_buffer, buffered) != buffered)1169 return -1;1170 write_buffer_len = 0;1171 }1172 return 0;1173}11741175static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)1176{1177 while (len) {1178 unsigned int buffered = write_buffer_len;1179 unsigned int partial = WRITE_BUFFER_SIZE - buffered;1180 if (partial > len)1181 partial = len;1182 memcpy(write_buffer + buffered, data, partial);1183 buffered += partial;1184 if (buffered == WRITE_BUFFER_SIZE) {1185 write_buffer_len = buffered;1186 if (ce_write_flush(context, fd))1187 return -1;1188 buffered = 0;1189 }1190 write_buffer_len = buffered;1191 len -= partial;1192 data = (char *) data + partial;1193 }1194 return 0;1195}11961197static int write_index_ext_header(SHA_CTX *context, int fd,1198 unsigned int ext, unsigned int sz)1199{1200 ext = htonl(ext);1201 sz = htonl(sz);1202 return ((ce_write(context, fd, &ext, 4) < 0) ||1203 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;1204}12051206static int ce_flush(SHA_CTX *context, int fd)1207{1208 unsigned int left = write_buffer_len;12091210 if (left) {1211 write_buffer_len = 0;1212 SHA1_Update(context, write_buffer, left);1213 }12141215 /* Flush first if not enough space for SHA1 signature */1216 if (left + 20 > WRITE_BUFFER_SIZE) {1217 if (write_in_full(fd, write_buffer, left) != left)1218 return -1;1219 left = 0;1220 }12211222 /* Append the SHA1 signature at the end */1223 SHA1_Final(write_buffer + left, context);1224 left += 20;1225 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;1226}12271228static void ce_smudge_racily_clean_entry(struct cache_entry *ce)1229{1230 /*1231 * The only thing we care about in this function is to smudge the1232 * falsely clean entry due to touch-update-touch race, so we leave1233 * everything else as they are. We are called for entries whose1234 * ce_mtime match the index file mtime.1235 */1236 struct stat st;12371238 if (lstat(ce->name, &st) < 0)1239 return;1240 if (ce_match_stat_basic(ce, &st))1241 return;1242 if (ce_modified_check_fs(ce, &st)) {1243 /* This is "racily clean"; smudge it. Note that this1244 * is a tricky code. At first glance, it may appear1245 * that it can break with this sequence:1246 *1247 * $ echo xyzzy >frotz1248 * $ git-update-index --add frotz1249 * $ : >frotz1250 * $ sleep 31251 * $ echo filfre >nitfol1252 * $ git-update-index --add nitfol1253 *1254 * but it does not. When the second update-index runs,1255 * it notices that the entry "frotz" has the same timestamp1256 * as index, and if we were to smudge it by resetting its1257 * size to zero here, then the object name recorded1258 * in index is the 6-byte file but the cached stat information1259 * becomes zero --- which would then match what we would1260 * obtain from the filesystem next time we stat("frotz").1261 *1262 * However, the second update-index, before calling1263 * this function, notices that the cached size is 61264 * bytes and what is on the filesystem is an empty1265 * file, and never calls us, so the cached size information1266 * for "frotz" stays 6 which does not match the filesystem.1267 */1268 ce->ce_size = 0;1269 }1270}12711272static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)1273{1274 int size = ondisk_ce_size(ce);1275 struct ondisk_cache_entry *ondisk = xcalloc(1, size);12761277 ondisk->ctime.sec = htonl(ce->ce_ctime);1278 ondisk->ctime.nsec = 0;1279 ondisk->mtime.sec = htonl(ce->ce_mtime);1280 ondisk->mtime.nsec = 0;1281 ondisk->dev = htonl(ce->ce_dev);1282 ondisk->ino = htonl(ce->ce_ino);1283 ondisk->mode = htonl(ce->ce_mode);1284 ondisk->uid = htonl(ce->ce_uid);1285 ondisk->gid = htonl(ce->ce_gid);1286 ondisk->size = htonl(ce->ce_size);1287 hashcpy(ondisk->sha1, ce->sha1);1288 ondisk->flags = htons(ce->ce_flags);1289 memcpy(ondisk->name, ce->name, ce_namelen(ce));12901291 return ce_write(c, fd, ondisk, size);1292}12931294int write_index(struct index_state *istate, int newfd)1295{1296 SHA_CTX c;1297 struct cache_header hdr;1298 int i, err, removed;1299 struct cache_entry **cache = istate->cache;1300 int entries = istate->cache_nr;13011302 for (i = removed = 0; i < entries; i++)1303 if (cache[i]->ce_flags & CE_REMOVE)1304 removed++;13051306 hdr.hdr_signature = htonl(CACHE_SIGNATURE);1307 hdr.hdr_version = htonl(2);1308 hdr.hdr_entries = htonl(entries - removed);13091310 SHA1_Init(&c);1311 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)1312 return -1;13131314 for (i = 0; i < entries; i++) {1315 struct cache_entry *ce = cache[i];1316 if (ce->ce_flags & CE_REMOVE)1317 continue;1318 if (is_racy_timestamp(istate, ce))1319 ce_smudge_racily_clean_entry(ce);1320 if (ce_write_entry(&c, newfd, ce) < 0)1321 return -1;1322 }13231324 /* Write extension data here */1325 if (istate->cache_tree) {1326 struct strbuf sb;13271328 strbuf_init(&sb, 0);1329 cache_tree_write(&sb, istate->cache_tree);1330 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 01331 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1332 strbuf_release(&sb);1333 if (err)1334 return -1;1335 }1336 return ce_flush(&c, newfd);1337}