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 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 32static 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 */ 46void 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 63static 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 77static 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 104static 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 121static 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 141static 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 199static 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 205int 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 248int 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 276int 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 */ 305int 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 330int 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 356int 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.. */ 378int 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 393int 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 404static 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 409static 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 432int add_file_to_index(struct index_state *istate, const char *path, int verbose) 433{ 434 int size, namelen; 435 struct stat st; 436 struct cache_entry *ce, *alias; 437 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY; 438 439 if (lstat(path, &st)) 440 die("%s: unable to stat (%s)", path, strerror(errno)); 441 442 if (!S_ISREG(st.st_mode) && !S_ISLNK(st.st_mode) && !S_ISDIR(st.st_mode)) 443 die("%s: can only add regular files, symbolic links or git-directories", path); 444 445 namelen = strlen(path); 446 if (S_ISDIR(st.st_mode)) { 447 while (namelen && path[namelen-1] == '/') 448 namelen--; 449 } 450 size = cache_entry_size(namelen); 451 ce = xcalloc(1, size); 452 memcpy(ce->name, path, namelen); 453 ce->ce_flags = namelen; 454 fill_stat_cache_info(ce, &st); 455 456 if (trust_executable_bit && has_symlinks) 457 ce->ce_mode = create_ce_mode(st.st_mode); 458 else { 459 /* If there is an existing entry, pick the mode bits and type 460 * from it, otherwise assume unexecutable regular file. 461 */ 462 struct cache_entry *ent; 463 int pos = index_name_pos_also_unmerged(istate, path, namelen); 464 465 ent = (0 <= pos) ? istate->cache[pos] : NULL; 466 ce->ce_mode = ce_mode_from_stat(ent, st.st_mode); 467 } 468 469 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case); 470 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, &st, ce_option)) { 471 /* Nothing changed, really */ 472 free(ce); 473 ce_mark_uptodate(alias); 474 return 0; 475 } 476 477 if (index_path(ce->sha1, path, &st, 1)) 478 die("unable to index file %s", path); 479 if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE)) 480 die("unable to add %s to index",path); 481 if (verbose) 482 printf("add '%s'\n", path); 483 return 0; 484} 485 486struct cache_entry *make_cache_entry(unsigned int mode, 487 const unsigned char *sha1, const char *path, int stage, 488 int refresh) 489{ 490 int size, len; 491 struct cache_entry *ce; 492 493 if (!verify_path(path)) 494 return NULL; 495 496 len = strlen(path); 497 size = cache_entry_size(len); 498 ce = xcalloc(1, size); 499 500 hashcpy(ce->sha1, sha1); 501 memcpy(ce->name, path, len); 502 ce->ce_flags = create_ce_flags(len, stage); 503 ce->ce_mode = create_ce_mode(mode); 504 505 if (refresh) 506 return refresh_cache_entry(ce, 0); 507 508 return ce; 509} 510 511int ce_same_name(struct cache_entry *a, struct cache_entry *b) 512{ 513 int len = ce_namelen(a); 514 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 515} 516 517int ce_path_match(const struct cache_entry *ce, const char **pathspec) 518{ 519 const char *match, *name; 520 int len; 521 522 if (!pathspec) 523 return 1; 524 525 len = ce_namelen(ce); 526 name = ce->name; 527 while ((match = *pathspec++) != NULL) { 528 int matchlen = strlen(match); 529 if (matchlen > len) 530 continue; 531 if (memcmp(name, match, matchlen)) 532 continue; 533 if (matchlen && name[matchlen-1] == '/') 534 return 1; 535 if (name[matchlen] == '/' || !name[matchlen]) 536 return 1; 537 if (!matchlen) 538 return 1; 539 } 540 return 0; 541} 542 543/* 544 * We fundamentally don't like some paths: we don't want 545 * dot or dot-dot anywhere, and for obvious reasons don't 546 * want to recurse into ".git" either. 547 * 548 * Also, we don't want double slashes or slashes at the 549 * end that can make pathnames ambiguous. 550 */ 551static int verify_dotfile(const char *rest) 552{ 553 /* 554 * The first character was '.', but that 555 * has already been discarded, we now test 556 * the rest. 557 */ 558 switch (*rest) { 559 /* "." is not allowed */ 560 case '\0': case '/': 561 return 0; 562 563 /* 564 * ".git" followed by NUL or slash is bad. This 565 * shares the path end test with the ".." case. 566 */ 567 case 'g': 568 if (rest[1] != 'i') 569 break; 570 if (rest[2] != 't') 571 break; 572 rest += 2; 573 /* fallthrough */ 574 case '.': 575 if (rest[1] == '\0' || rest[1] == '/') 576 return 0; 577 } 578 return 1; 579} 580 581int verify_path(const char *path) 582{ 583 char c; 584 585 goto inside; 586 for (;;) { 587 if (!c) 588 return 1; 589 if (c == '/') { 590inside: 591 c = *path++; 592 switch (c) { 593 default: 594 continue; 595 case '/': case '\0': 596 break; 597 case '.': 598 if (verify_dotfile(path)) 599 continue; 600 } 601 return 0; 602 } 603 c = *path++; 604 } 605} 606 607/* 608 * Do we have another file that has the beginning components being a 609 * proper superset of the name we're trying to add? 610 */ 611static int has_file_name(struct index_state *istate, 612 const struct cache_entry *ce, int pos, int ok_to_replace) 613{ 614 int retval = 0; 615 int len = ce_namelen(ce); 616 int stage = ce_stage(ce); 617 const char *name = ce->name; 618 619 while (pos < istate->cache_nr) { 620 struct cache_entry *p = istate->cache[pos++]; 621 622 if (len >= ce_namelen(p)) 623 break; 624 if (memcmp(name, p->name, len)) 625 break; 626 if (ce_stage(p) != stage) 627 continue; 628 if (p->name[len] != '/') 629 continue; 630 if (p->ce_flags & CE_REMOVE) 631 continue; 632 retval = -1; 633 if (!ok_to_replace) 634 break; 635 remove_index_entry_at(istate, --pos); 636 } 637 return retval; 638} 639 640/* 641 * Do we have another file with a pathname that is a proper 642 * subset of the name we're trying to add? 643 */ 644static int has_dir_name(struct index_state *istate, 645 const struct cache_entry *ce, int pos, int ok_to_replace) 646{ 647 int retval = 0; 648 int stage = ce_stage(ce); 649 const char *name = ce->name; 650 const char *slash = name + ce_namelen(ce); 651 652 for (;;) { 653 int len; 654 655 for (;;) { 656 if (*--slash == '/') 657 break; 658 if (slash <= ce->name) 659 return retval; 660 } 661 len = slash - name; 662 663 pos = index_name_pos(istate, name, create_ce_flags(len, stage)); 664 if (pos >= 0) { 665 /* 666 * Found one, but not so fast. This could 667 * be a marker that says "I was here, but 668 * I am being removed". Such an entry is 669 * not a part of the resulting tree, and 670 * it is Ok to have a directory at the same 671 * path. 672 */ 673 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) { 674 retval = -1; 675 if (!ok_to_replace) 676 break; 677 remove_index_entry_at(istate, pos); 678 continue; 679 } 680 } 681 else 682 pos = -pos-1; 683 684 /* 685 * Trivial optimization: if we find an entry that 686 * already matches the sub-directory, then we know 687 * we're ok, and we can exit. 688 */ 689 while (pos < istate->cache_nr) { 690 struct cache_entry *p = istate->cache[pos]; 691 if ((ce_namelen(p) <= len) || 692 (p->name[len] != '/') || 693 memcmp(p->name, name, len)) 694 break; /* not our subdirectory */ 695 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE)) 696 /* 697 * p is at the same stage as our entry, and 698 * is a subdirectory of what we are looking 699 * at, so we cannot have conflicts at our 700 * level or anything shorter. 701 */ 702 return retval; 703 pos++; 704 } 705 } 706 return retval; 707} 708 709/* We may be in a situation where we already have path/file and path 710 * is being added, or we already have path and path/file is being 711 * added. Either one would result in a nonsense tree that has path 712 * twice when git-write-tree tries to write it out. Prevent it. 713 * 714 * If ok-to-replace is specified, we remove the conflicting entries 715 * from the cache so the caller should recompute the insert position. 716 * When this happens, we return non-zero. 717 */ 718static int check_file_directory_conflict(struct index_state *istate, 719 const struct cache_entry *ce, 720 int pos, int ok_to_replace) 721{ 722 int retval; 723 724 /* 725 * When ce is an "I am going away" entry, we allow it to be added 726 */ 727 if (ce->ce_flags & CE_REMOVE) 728 return 0; 729 730 /* 731 * We check if the path is a sub-path of a subsequent pathname 732 * first, since removing those will not change the position 733 * in the array. 734 */ 735 retval = has_file_name(istate, ce, pos, ok_to_replace); 736 737 /* 738 * Then check if the path might have a clashing sub-directory 739 * before it. 740 */ 741 return retval + has_dir_name(istate, ce, pos, ok_to_replace); 742} 743 744static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option) 745{ 746 int pos; 747 int ok_to_add = option & ADD_CACHE_OK_TO_ADD; 748 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE; 749 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK; 750 751 cache_tree_invalidate_path(istate->cache_tree, ce->name); 752 pos = index_name_pos(istate, ce->name, ce->ce_flags); 753 754 /* existing match? Just replace it. */ 755 if (pos >= 0) { 756 replace_index_entry(istate, pos, ce); 757 return 0; 758 } 759 pos = -pos-1; 760 761 /* 762 * Inserting a merged entry ("stage 0") into the index 763 * will always replace all non-merged entries.. 764 */ 765 if (pos < istate->cache_nr && ce_stage(ce) == 0) { 766 while (ce_same_name(istate->cache[pos], ce)) { 767 ok_to_add = 1; 768 if (!remove_index_entry_at(istate, pos)) 769 break; 770 } 771 } 772 773 if (!ok_to_add) 774 return -1; 775 if (!verify_path(ce->name)) 776 return -1; 777 778 if (!skip_df_check && 779 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) { 780 if (!ok_to_replace) 781 return error("'%s' appears as both a file and as a directory", 782 ce->name); 783 pos = index_name_pos(istate, ce->name, ce->ce_flags); 784 pos = -pos-1; 785 } 786 return pos + 1; 787} 788 789int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option) 790{ 791 int pos; 792 793 if (option & ADD_CACHE_JUST_APPEND) 794 pos = istate->cache_nr; 795 else { 796 int ret; 797 ret = add_index_entry_with_check(istate, ce, option); 798 if (ret <= 0) 799 return ret; 800 pos = ret - 1; 801 } 802 803 /* Make sure the array is big enough .. */ 804 if (istate->cache_nr == istate->cache_alloc) { 805 istate->cache_alloc = alloc_nr(istate->cache_alloc); 806 istate->cache = xrealloc(istate->cache, 807 istate->cache_alloc * sizeof(struct cache_entry *)); 808 } 809 810 /* Add it in.. */ 811 istate->cache_nr++; 812 if (istate->cache_nr > pos + 1) 813 memmove(istate->cache + pos + 1, 814 istate->cache + pos, 815 (istate->cache_nr - pos - 1) * sizeof(ce)); 816 set_index_entry(istate, pos, ce); 817 istate->cache_changed = 1; 818 return 0; 819} 820 821/* 822 * "refresh" does not calculate a new sha1 file or bring the 823 * cache up-to-date for mode/content changes. But what it 824 * _does_ do is to "re-match" the stat information of a file 825 * with the cache, so that you can refresh the cache for a 826 * file that hasn't been changed but where the stat entry is 827 * out of date. 828 * 829 * For example, you'd want to do this after doing a "git-read-tree", 830 * to link up the stat cache details with the proper files. 831 */ 832static struct cache_entry *refresh_cache_ent(struct index_state *istate, 833 struct cache_entry *ce, 834 unsigned int options, int *err) 835{ 836 struct stat st; 837 struct cache_entry *updated; 838 int changed, size; 839 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 840 841 if (ce_uptodate(ce)) 842 return ce; 843 844 if (lstat(ce->name, &st) < 0) { 845 if (err) 846 *err = errno; 847 return NULL; 848 } 849 850 changed = ie_match_stat(istate, ce, &st, options); 851 if (!changed) { 852 /* 853 * The path is unchanged. If we were told to ignore 854 * valid bit, then we did the actual stat check and 855 * found that the entry is unmodified. If the entry 856 * is not marked VALID, this is the place to mark it 857 * valid again, under "assume unchanged" mode. 858 */ 859 if (ignore_valid && assume_unchanged && 860 !(ce->ce_flags & CE_VALID)) 861 ; /* mark this one VALID again */ 862 else { 863 /* 864 * We do not mark the index itself "modified" 865 * because CE_UPTODATE flag is in-core only; 866 * we are not going to write this change out. 867 */ 868 ce_mark_uptodate(ce); 869 return ce; 870 } 871 } 872 873 if (ie_modified(istate, ce, &st, options)) { 874 if (err) 875 *err = EINVAL; 876 return NULL; 877 } 878 879 size = ce_size(ce); 880 updated = xmalloc(size); 881 memcpy(updated, ce, size); 882 fill_stat_cache_info(updated, &st); 883 /* 884 * If ignore_valid is not set, we should leave CE_VALID bit 885 * alone. Otherwise, paths marked with --no-assume-unchanged 886 * (i.e. things to be edited) will reacquire CE_VALID bit 887 * automatically, which is not really what we want. 888 */ 889 if (!ignore_valid && assume_unchanged && 890 !(ce->ce_flags & CE_VALID)) 891 updated->ce_flags &= ~CE_VALID; 892 893 return updated; 894} 895 896int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen) 897{ 898 int i; 899 int has_errors = 0; 900 int really = (flags & REFRESH_REALLY) != 0; 901 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0; 902 int quiet = (flags & REFRESH_QUIET) != 0; 903 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0; 904 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0; 905 906 for (i = 0; i < istate->cache_nr; i++) { 907 struct cache_entry *ce, *new; 908 int cache_errno = 0; 909 910 ce = istate->cache[i]; 911 if (ce_stage(ce)) { 912 while ((i < istate->cache_nr) && 913 ! strcmp(istate->cache[i]->name, ce->name)) 914 i++; 915 i--; 916 if (allow_unmerged) 917 continue; 918 printf("%s: needs merge\n", ce->name); 919 has_errors = 1; 920 continue; 921 } 922 923 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen)) 924 continue; 925 926 new = refresh_cache_ent(istate, ce, options, &cache_errno); 927 if (new == ce) 928 continue; 929 if (!new) { 930 if (not_new && cache_errno == ENOENT) 931 continue; 932 if (really && cache_errno == EINVAL) { 933 /* If we are doing --really-refresh that 934 * means the index is not valid anymore. 935 */ 936 ce->ce_flags &= ~CE_VALID; 937 istate->cache_changed = 1; 938 } 939 if (quiet) 940 continue; 941 printf("%s: needs update\n", ce->name); 942 has_errors = 1; 943 continue; 944 } 945 946 replace_index_entry(istate, i, new); 947 } 948 return has_errors; 949} 950 951struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really) 952{ 953 return refresh_cache_ent(&the_index, ce, really, NULL); 954} 955 956static int verify_hdr(struct cache_header *hdr, unsigned long size) 957{ 958 SHA_CTX c; 959 unsigned char sha1[20]; 960 961 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE)) 962 return error("bad signature"); 963 if (hdr->hdr_version != htonl(2)) 964 return error("bad index version"); 965 SHA1_Init(&c); 966 SHA1_Update(&c, hdr, size - 20); 967 SHA1_Final(sha1, &c); 968 if (hashcmp(sha1, (unsigned char *)hdr + size - 20)) 969 return error("bad index file sha1 signature"); 970 return 0; 971} 972 973static int read_index_extension(struct index_state *istate, 974 const char *ext, void *data, unsigned long sz) 975{ 976 switch (CACHE_EXT(ext)) { 977 case CACHE_EXT_TREE: 978 istate->cache_tree = cache_tree_read(data, sz); 979 break; 980 default: 981 if (*ext < 'A' || 'Z' < *ext) 982 return error("index uses %.4s extension, which we do not understand", 983 ext); 984 fprintf(stderr, "ignoring %.4s extension\n", ext); 985 break; 986 } 987 return 0; 988} 989 990int read_index(struct index_state *istate) 991{ 992 return read_index_from(istate, get_index_file()); 993} 994 995static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce) 996{ 997 size_t len; 998 999 ce->ce_ctime = ntohl(ondisk->ctime.sec);1000 ce->ce_mtime = ntohl(ondisk->mtime.sec);1001 ce->ce_dev = ntohl(ondisk->dev);1002 ce->ce_ino = ntohl(ondisk->ino);1003 ce->ce_mode = ntohl(ondisk->mode);1004 ce->ce_uid = ntohl(ondisk->uid);1005 ce->ce_gid = ntohl(ondisk->gid);1006 ce->ce_size = ntohl(ondisk->size);1007 /* On-disk flags are just 16 bits */1008 ce->ce_flags = ntohs(ondisk->flags);1009 hashcpy(ce->sha1, ondisk->sha1);10101011 len = ce->ce_flags & CE_NAMEMASK;1012 if (len == CE_NAMEMASK)1013 len = strlen(ondisk->name);1014 /*1015 * NEEDSWORK: If the original index is crafted, this copy could1016 * go unchecked.1017 */1018 memcpy(ce->name, ondisk->name, len + 1);1019}10201021static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)1022{1023 long per_entry;10241025 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);10261027 /*1028 * Alignment can cause differences. This should be "alignof", but1029 * since that's a gcc'ism, just use the size of a pointer.1030 */1031 per_entry += sizeof(void *);1032 return ondisk_size + entries*per_entry;1033}10341035/* remember to discard_cache() before reading a different cache! */1036int read_index_from(struct index_state *istate, const char *path)1037{1038 int fd, i;1039 struct stat st;1040 unsigned long src_offset, dst_offset;1041 struct cache_header *hdr;1042 void *mmap;1043 size_t mmap_size;10441045 errno = EBUSY;1046 if (istate->alloc)1047 return istate->cache_nr;10481049 errno = ENOENT;1050 istate->timestamp = 0;1051 fd = open(path, O_RDONLY);1052 if (fd < 0) {1053 if (errno == ENOENT)1054 return 0;1055 die("index file open failed (%s)", strerror(errno));1056 }10571058 if (fstat(fd, &st))1059 die("cannot stat the open index (%s)", strerror(errno));10601061 errno = EINVAL;1062 mmap_size = xsize_t(st.st_size);1063 if (mmap_size < sizeof(struct cache_header) + 20)1064 die("index file smaller than expected");10651066 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);1067 close(fd);1068 if (mmap == MAP_FAILED)1069 die("unable to map index file");10701071 hdr = mmap;1072 if (verify_hdr(hdr, mmap_size) < 0)1073 goto unmap;10741075 istate->cache_nr = ntohl(hdr->hdr_entries);1076 istate->cache_alloc = alloc_nr(istate->cache_nr);1077 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));10781079 /*1080 * The disk format is actually larger than the in-memory format,1081 * due to space for nsec etc, so even though the in-memory one1082 * has room for a few more flags, we can allocate using the same1083 * index size1084 */1085 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));10861087 src_offset = sizeof(*hdr);1088 dst_offset = 0;1089 for (i = 0; i < istate->cache_nr; i++) {1090 struct ondisk_cache_entry *disk_ce;1091 struct cache_entry *ce;10921093 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);1094 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);1095 convert_from_disk(disk_ce, ce);1096 set_index_entry(istate, i, ce);10971098 src_offset += ondisk_ce_size(ce);1099 dst_offset += ce_size(ce);1100 }1101 istate->timestamp = st.st_mtime;1102 while (src_offset <= mmap_size - 20 - 8) {1103 /* After an array of active_nr index entries,1104 * there can be arbitrary number of extended1105 * sections, each of which is prefixed with1106 * extension name (4-byte) and section length1107 * in 4-byte network byte order.1108 */1109 unsigned long extsize;1110 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);1111 extsize = ntohl(extsize);1112 if (read_index_extension(istate,1113 (const char *) mmap + src_offset,1114 (char *) mmap + src_offset + 8,1115 extsize) < 0)1116 goto unmap;1117 src_offset += 8;1118 src_offset += extsize;1119 }1120 munmap(mmap, mmap_size);1121 return istate->cache_nr;11221123unmap:1124 munmap(mmap, mmap_size);1125 errno = EINVAL;1126 die("index file corrupt");1127}11281129int discard_index(struct index_state *istate)1130{1131 istate->cache_nr = 0;1132 istate->cache_changed = 0;1133 istate->timestamp = 0;1134 free_hash(&istate->name_hash);1135 cache_tree_free(&(istate->cache_tree));1136 free(istate->alloc);1137 istate->alloc = NULL;11381139 /* no need to throw away allocated active_cache */1140 return 0;1141}11421143int unmerged_index(const struct index_state *istate)1144{1145 int i;1146 for (i = 0; i < istate->cache_nr; i++) {1147 if (ce_stage(istate->cache[i]))1148 return 1;1149 }1150 return 0;1151}11521153#define WRITE_BUFFER_SIZE 81921154static unsigned char write_buffer[WRITE_BUFFER_SIZE];1155static unsigned long write_buffer_len;11561157static int ce_write_flush(SHA_CTX *context, int fd)1158{1159 unsigned int buffered = write_buffer_len;1160 if (buffered) {1161 SHA1_Update(context, write_buffer, buffered);1162 if (write_in_full(fd, write_buffer, buffered) != buffered)1163 return -1;1164 write_buffer_len = 0;1165 }1166 return 0;1167}11681169static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)1170{1171 while (len) {1172 unsigned int buffered = write_buffer_len;1173 unsigned int partial = WRITE_BUFFER_SIZE - buffered;1174 if (partial > len)1175 partial = len;1176 memcpy(write_buffer + buffered, data, partial);1177 buffered += partial;1178 if (buffered == WRITE_BUFFER_SIZE) {1179 write_buffer_len = buffered;1180 if (ce_write_flush(context, fd))1181 return -1;1182 buffered = 0;1183 }1184 write_buffer_len = buffered;1185 len -= partial;1186 data = (char *) data + partial;1187 }1188 return 0;1189}11901191static int write_index_ext_header(SHA_CTX *context, int fd,1192 unsigned int ext, unsigned int sz)1193{1194 ext = htonl(ext);1195 sz = htonl(sz);1196 return ((ce_write(context, fd, &ext, 4) < 0) ||1197 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;1198}11991200static int ce_flush(SHA_CTX *context, int fd)1201{1202 unsigned int left = write_buffer_len;12031204 if (left) {1205 write_buffer_len = 0;1206 SHA1_Update(context, write_buffer, left);1207 }12081209 /* Flush first if not enough space for SHA1 signature */1210 if (left + 20 > WRITE_BUFFER_SIZE) {1211 if (write_in_full(fd, write_buffer, left) != left)1212 return -1;1213 left = 0;1214 }12151216 /* Append the SHA1 signature at the end */1217 SHA1_Final(write_buffer + left, context);1218 left += 20;1219 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;1220}12211222static void ce_smudge_racily_clean_entry(struct cache_entry *ce)1223{1224 /*1225 * The only thing we care about in this function is to smudge the1226 * falsely clean entry due to touch-update-touch race, so we leave1227 * everything else as they are. We are called for entries whose1228 * ce_mtime match the index file mtime.1229 */1230 struct stat st;12311232 if (lstat(ce->name, &st) < 0)1233 return;1234 if (ce_match_stat_basic(ce, &st))1235 return;1236 if (ce_modified_check_fs(ce, &st)) {1237 /* This is "racily clean"; smudge it. Note that this1238 * is a tricky code. At first glance, it may appear1239 * that it can break with this sequence:1240 *1241 * $ echo xyzzy >frotz1242 * $ git-update-index --add frotz1243 * $ : >frotz1244 * $ sleep 31245 * $ echo filfre >nitfol1246 * $ git-update-index --add nitfol1247 *1248 * but it does not. When the second update-index runs,1249 * it notices that the entry "frotz" has the same timestamp1250 * as index, and if we were to smudge it by resetting its1251 * size to zero here, then the object name recorded1252 * in index is the 6-byte file but the cached stat information1253 * becomes zero --- which would then match what we would1254 * obtain from the filesystem next time we stat("frotz").1255 *1256 * However, the second update-index, before calling1257 * this function, notices that the cached size is 61258 * bytes and what is on the filesystem is an empty1259 * file, and never calls us, so the cached size information1260 * for "frotz" stays 6 which does not match the filesystem.1261 */1262 ce->ce_size = 0;1263 }1264}12651266static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)1267{1268 int size = ondisk_ce_size(ce);1269 struct ondisk_cache_entry *ondisk = xcalloc(1, size);12701271 ondisk->ctime.sec = htonl(ce->ce_ctime);1272 ondisk->ctime.nsec = 0;1273 ondisk->mtime.sec = htonl(ce->ce_mtime);1274 ondisk->mtime.nsec = 0;1275 ondisk->dev = htonl(ce->ce_dev);1276 ondisk->ino = htonl(ce->ce_ino);1277 ondisk->mode = htonl(ce->ce_mode);1278 ondisk->uid = htonl(ce->ce_uid);1279 ondisk->gid = htonl(ce->ce_gid);1280 ondisk->size = htonl(ce->ce_size);1281 hashcpy(ondisk->sha1, ce->sha1);1282 ondisk->flags = htons(ce->ce_flags);1283 memcpy(ondisk->name, ce->name, ce_namelen(ce));12841285 return ce_write(c, fd, ondisk, size);1286}12871288int write_index(const struct index_state *istate, int newfd)1289{1290 SHA_CTX c;1291 struct cache_header hdr;1292 int i, err, removed;1293 struct cache_entry **cache = istate->cache;1294 int entries = istate->cache_nr;12951296 for (i = removed = 0; i < entries; i++)1297 if (cache[i]->ce_flags & CE_REMOVE)1298 removed++;12991300 hdr.hdr_signature = htonl(CACHE_SIGNATURE);1301 hdr.hdr_version = htonl(2);1302 hdr.hdr_entries = htonl(entries - removed);13031304 SHA1_Init(&c);1305 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)1306 return -1;13071308 for (i = 0; i < entries; i++) {1309 struct cache_entry *ce = cache[i];1310 if (ce->ce_flags & CE_REMOVE)1311 continue;1312 if (is_racy_timestamp(istate, ce))1313 ce_smudge_racily_clean_entry(ce);1314 if (ce_write_entry(&c, newfd, ce) < 0)1315 return -1;1316 }13171318 /* Write extension data here */1319 if (istate->cache_tree) {1320 struct strbuf sb;13211322 strbuf_init(&sb, 0);1323 cache_tree_write(&sb, istate->cache_tree);1324 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 01325 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1326 strbuf_release(&sb);1327 if (err)1328 return -1;1329 }1330 return ce_flush(&c, newfd);1331}