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 (!S_ISGITLINK(ce->ce_mode) && 202 istate->timestamp && 203 ((unsigned int)istate->timestamp) <= ce->ce_mtime); 204} 205 206int ie_match_stat(const struct index_state *istate, 207 struct cache_entry *ce, struct stat *st, 208 unsigned int options) 209{ 210 unsigned int changed; 211 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 212 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 213 214 /* 215 * If it's marked as always valid in the index, it's 216 * valid whatever the checked-out copy says. 217 */ 218 if (!ignore_valid && (ce->ce_flags & CE_VALID)) 219 return 0; 220 221 changed = ce_match_stat_basic(ce, st); 222 223 /* 224 * Within 1 second of this sequence: 225 * echo xyzzy >file && git-update-index --add file 226 * running this command: 227 * echo frotz >file 228 * would give a falsely clean cache entry. The mtime and 229 * length match the cache, and other stat fields do not change. 230 * 231 * We could detect this at update-index time (the cache entry 232 * being registered/updated records the same time as "now") 233 * and delay the return from git-update-index, but that would 234 * effectively mean we can make at most one commit per second, 235 * which is not acceptable. Instead, we check cache entries 236 * whose mtime are the same as the index file timestamp more 237 * carefully than others. 238 */ 239 if (!changed && is_racy_timestamp(istate, ce)) { 240 if (assume_racy_is_modified) 241 changed |= DATA_CHANGED; 242 else 243 changed |= ce_modified_check_fs(ce, st); 244 } 245 246 return changed; 247} 248 249int ie_modified(const struct index_state *istate, 250 struct cache_entry *ce, struct stat *st, unsigned int options) 251{ 252 int changed, changed_fs; 253 254 changed = ie_match_stat(istate, ce, st, options); 255 if (!changed) 256 return 0; 257 /* 258 * If the mode or type has changed, there's no point in trying 259 * to refresh the entry - it's not going to match 260 */ 261 if (changed & (MODE_CHANGED | TYPE_CHANGED)) 262 return changed; 263 264 /* Immediately after read-tree or update-index --cacheinfo, 265 * the length field is zero. For other cases the ce_size 266 * should match the SHA1 recorded in the index entry. 267 */ 268 if ((changed & DATA_CHANGED) && ce->ce_size != 0) 269 return changed; 270 271 changed_fs = ce_modified_check_fs(ce, st); 272 if (changed_fs) 273 return changed | changed_fs; 274 return 0; 275} 276 277int base_name_compare(const char *name1, int len1, int mode1, 278 const char *name2, int len2, int mode2) 279{ 280 unsigned char c1, c2; 281 int len = len1 < len2 ? len1 : len2; 282 int cmp; 283 284 cmp = memcmp(name1, name2, len); 285 if (cmp) 286 return cmp; 287 c1 = name1[len]; 288 c2 = name2[len]; 289 if (!c1 && S_ISDIR(mode1)) 290 c1 = '/'; 291 if (!c2 && S_ISDIR(mode2)) 292 c2 = '/'; 293 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0; 294} 295 296/* 297 * df_name_compare() is identical to base_name_compare(), except it 298 * compares conflicting directory/file entries as equal. Note that 299 * while a directory name compares as equal to a regular file, they 300 * then individually compare _differently_ to a filename that has 301 * a dot after the basename (because '\0' < '.' < '/'). 302 * 303 * This is used by routines that want to traverse the git namespace 304 * but then handle conflicting entries together when possible. 305 */ 306int df_name_compare(const char *name1, int len1, int mode1, 307 const char *name2, int len2, int mode2) 308{ 309 int len = len1 < len2 ? len1 : len2, cmp; 310 unsigned char c1, c2; 311 312 cmp = memcmp(name1, name2, len); 313 if (cmp) 314 return cmp; 315 /* Directories and files compare equal (same length, same name) */ 316 if (len1 == len2) 317 return 0; 318 c1 = name1[len]; 319 if (!c1 && S_ISDIR(mode1)) 320 c1 = '/'; 321 c2 = name2[len]; 322 if (!c2 && S_ISDIR(mode2)) 323 c2 = '/'; 324 if (c1 == '/' && !c2) 325 return 0; 326 if (c2 == '/' && !c1) 327 return 0; 328 return c1 - c2; 329} 330 331int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2) 332{ 333 int len1 = flags1 & CE_NAMEMASK; 334 int len2 = flags2 & CE_NAMEMASK; 335 int len = len1 < len2 ? len1 : len2; 336 int cmp; 337 338 cmp = memcmp(name1, name2, len); 339 if (cmp) 340 return cmp; 341 if (len1 < len2) 342 return -1; 343 if (len1 > len2) 344 return 1; 345 346 /* Compare stages */ 347 flags1 &= CE_STAGEMASK; 348 flags2 &= CE_STAGEMASK; 349 350 if (flags1 < flags2) 351 return -1; 352 if (flags1 > flags2) 353 return 1; 354 return 0; 355} 356 357int index_name_pos(const struct index_state *istate, const char *name, int namelen) 358{ 359 int first, last; 360 361 first = 0; 362 last = istate->cache_nr; 363 while (last > first) { 364 int next = (last + first) >> 1; 365 struct cache_entry *ce = istate->cache[next]; 366 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags); 367 if (!cmp) 368 return next; 369 if (cmp < 0) { 370 last = next; 371 continue; 372 } 373 first = next+1; 374 } 375 return -first-1; 376} 377 378/* Remove entry, return true if there are more entries to go.. */ 379int remove_index_entry_at(struct index_state *istate, int pos) 380{ 381 struct cache_entry *ce = istate->cache[pos]; 382 383 remove_name_hash(ce); 384 istate->cache_changed = 1; 385 istate->cache_nr--; 386 if (pos >= istate->cache_nr) 387 return 0; 388 memmove(istate->cache + pos, 389 istate->cache + pos + 1, 390 (istate->cache_nr - pos) * sizeof(struct cache_entry *)); 391 return 1; 392} 393 394int remove_file_from_index(struct index_state *istate, const char *path) 395{ 396 int pos = index_name_pos(istate, path, strlen(path)); 397 if (pos < 0) 398 pos = -pos-1; 399 cache_tree_invalidate_path(istate->cache_tree, path); 400 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 401 remove_index_entry_at(istate, pos); 402 return 0; 403} 404 405static int compare_name(struct cache_entry *ce, const char *path, int namelen) 406{ 407 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); 408} 409 410static int index_name_pos_also_unmerged(struct index_state *istate, 411 const char *path, int namelen) 412{ 413 int pos = index_name_pos(istate, path, namelen); 414 struct cache_entry *ce; 415 416 if (pos >= 0) 417 return pos; 418 419 /* maybe unmerged? */ 420 pos = -1 - pos; 421 if (pos >= istate->cache_nr || 422 compare_name((ce = istate->cache[pos]), path, namelen)) 423 return -1; 424 425 /* order of preference: stage 2, 1, 3 */ 426 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && 427 ce_stage((ce = istate->cache[pos + 1])) == 2 && 428 !compare_name(ce, path, namelen)) 429 pos++; 430 return pos; 431} 432 433static int different_name(struct cache_entry *ce, struct cache_entry *alias) 434{ 435 int len = ce_namelen(ce); 436 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len); 437} 438 439/* 440 * If we add a filename that aliases in the cache, we will use the 441 * name that we already have - but we don't want to update the same 442 * alias twice, because that implies that there were actually two 443 * different files with aliasing names! 444 * 445 * So we use the CE_ADDED flag to verify that the alias was an old 446 * one before we accept it as 447 */ 448static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias) 449{ 450 int len; 451 struct cache_entry *new; 452 453 if (alias->ce_flags & CE_ADDED) 454 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); 455 456 /* Ok, create the new entry using the name of the existing alias */ 457 len = ce_namelen(alias); 458 new = xcalloc(1, cache_entry_size(len)); 459 memcpy(new->name, alias->name, len); 460 copy_cache_entry(new, ce); 461 free(ce); 462 return new; 463} 464 465int add_to_index(struct index_state *istate, const char *path, struct stat *st, int verbose) 466{ 467 int size, namelen; 468 mode_t st_mode = st->st_mode; 469 struct cache_entry *ce, *alias; 470 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY; 471 472 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 473 die("%s: can only add regular files, symbolic links or git-directories", path); 474 475 namelen = strlen(path); 476 if (S_ISDIR(st_mode)) { 477 while (namelen && path[namelen-1] == '/') 478 namelen--; 479 } 480 size = cache_entry_size(namelen); 481 ce = xcalloc(1, size); 482 memcpy(ce->name, path, namelen); 483 ce->ce_flags = namelen; 484 fill_stat_cache_info(ce, st); 485 486 if (trust_executable_bit && has_symlinks) 487 ce->ce_mode = create_ce_mode(st_mode); 488 else { 489 /* If there is an existing entry, pick the mode bits and type 490 * from it, otherwise assume unexecutable regular file. 491 */ 492 struct cache_entry *ent; 493 int pos = index_name_pos_also_unmerged(istate, path, namelen); 494 495 ent = (0 <= pos) ? istate->cache[pos] : NULL; 496 ce->ce_mode = ce_mode_from_stat(ent, st_mode); 497 } 498 499 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case); 500 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) { 501 /* Nothing changed, really */ 502 free(ce); 503 ce_mark_uptodate(alias); 504 alias->ce_flags |= CE_ADDED; 505 return 0; 506 } 507 if (index_path(ce->sha1, path, st, 1)) 508 die("unable to index file %s", path); 509 if (ignore_case && alias && different_name(ce, alias)) 510 ce = create_alias_ce(ce, alias); 511 ce->ce_flags |= CE_ADDED; 512 if (add_index_entry(istate, ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE)) 513 die("unable to add %s to index",path); 514 if (verbose) 515 printf("add '%s'\n", path); 516 return 0; 517} 518 519int add_file_to_index(struct index_state *istate, const char *path, int verbose) 520{ 521 struct stat st; 522 if (lstat(path, &st)) 523 die("%s: unable to stat (%s)", path, strerror(errno)); 524 return add_to_index(istate, path, &st, verbose); 525} 526 527struct cache_entry *make_cache_entry(unsigned int mode, 528 const unsigned char *sha1, const char *path, int stage, 529 int refresh) 530{ 531 int size, len; 532 struct cache_entry *ce; 533 534 if (!verify_path(path)) 535 return NULL; 536 537 len = strlen(path); 538 size = cache_entry_size(len); 539 ce = xcalloc(1, size); 540 541 hashcpy(ce->sha1, sha1); 542 memcpy(ce->name, path, len); 543 ce->ce_flags = create_ce_flags(len, stage); 544 ce->ce_mode = create_ce_mode(mode); 545 546 if (refresh) 547 return refresh_cache_entry(ce, 0); 548 549 return ce; 550} 551 552int ce_same_name(struct cache_entry *a, struct cache_entry *b) 553{ 554 int len = ce_namelen(a); 555 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 556} 557 558int ce_path_match(const struct cache_entry *ce, const char **pathspec) 559{ 560 const char *match, *name; 561 int len; 562 563 if (!pathspec) 564 return 1; 565 566 len = ce_namelen(ce); 567 name = ce->name; 568 while ((match = *pathspec++) != NULL) { 569 int matchlen = strlen(match); 570 if (matchlen > len) 571 continue; 572 if (memcmp(name, match, matchlen)) 573 continue; 574 if (matchlen && name[matchlen-1] == '/') 575 return 1; 576 if (name[matchlen] == '/' || !name[matchlen]) 577 return 1; 578 if (!matchlen) 579 return 1; 580 } 581 return 0; 582} 583 584/* 585 * We fundamentally don't like some paths: we don't want 586 * dot or dot-dot anywhere, and for obvious reasons don't 587 * want to recurse into ".git" either. 588 * 589 * Also, we don't want double slashes or slashes at the 590 * end that can make pathnames ambiguous. 591 */ 592static int verify_dotfile(const char *rest) 593{ 594 /* 595 * The first character was '.', but that 596 * has already been discarded, we now test 597 * the rest. 598 */ 599 switch (*rest) { 600 /* "." is not allowed */ 601 case '\0': case '/': 602 return 0; 603 604 /* 605 * ".git" followed by NUL or slash is bad. This 606 * shares the path end test with the ".." case. 607 */ 608 case 'g': 609 if (rest[1] != 'i') 610 break; 611 if (rest[2] != 't') 612 break; 613 rest += 2; 614 /* fallthrough */ 615 case '.': 616 if (rest[1] == '\0' || rest[1] == '/') 617 return 0; 618 } 619 return 1; 620} 621 622int verify_path(const char *path) 623{ 624 char c; 625 626 goto inside; 627 for (;;) { 628 if (!c) 629 return 1; 630 if (c == '/') { 631inside: 632 c = *path++; 633 switch (c) { 634 default: 635 continue; 636 case '/': case '\0': 637 break; 638 case '.': 639 if (verify_dotfile(path)) 640 continue; 641 } 642 return 0; 643 } 644 c = *path++; 645 } 646} 647 648/* 649 * Do we have another file that has the beginning components being a 650 * proper superset of the name we're trying to add? 651 */ 652static int has_file_name(struct index_state *istate, 653 const struct cache_entry *ce, int pos, int ok_to_replace) 654{ 655 int retval = 0; 656 int len = ce_namelen(ce); 657 int stage = ce_stage(ce); 658 const char *name = ce->name; 659 660 while (pos < istate->cache_nr) { 661 struct cache_entry *p = istate->cache[pos++]; 662 663 if (len >= ce_namelen(p)) 664 break; 665 if (memcmp(name, p->name, len)) 666 break; 667 if (ce_stage(p) != stage) 668 continue; 669 if (p->name[len] != '/') 670 continue; 671 if (p->ce_flags & CE_REMOVE) 672 continue; 673 retval = -1; 674 if (!ok_to_replace) 675 break; 676 remove_index_entry_at(istate, --pos); 677 } 678 return retval; 679} 680 681/* 682 * Do we have another file with a pathname that is a proper 683 * subset of the name we're trying to add? 684 */ 685static int has_dir_name(struct index_state *istate, 686 const struct cache_entry *ce, int pos, int ok_to_replace) 687{ 688 int retval = 0; 689 int stage = ce_stage(ce); 690 const char *name = ce->name; 691 const char *slash = name + ce_namelen(ce); 692 693 for (;;) { 694 int len; 695 696 for (;;) { 697 if (*--slash == '/') 698 break; 699 if (slash <= ce->name) 700 return retval; 701 } 702 len = slash - name; 703 704 pos = index_name_pos(istate, name, create_ce_flags(len, stage)); 705 if (pos >= 0) { 706 /* 707 * Found one, but not so fast. This could 708 * be a marker that says "I was here, but 709 * I am being removed". Such an entry is 710 * not a part of the resulting tree, and 711 * it is Ok to have a directory at the same 712 * path. 713 */ 714 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) { 715 retval = -1; 716 if (!ok_to_replace) 717 break; 718 remove_index_entry_at(istate, pos); 719 continue; 720 } 721 } 722 else 723 pos = -pos-1; 724 725 /* 726 * Trivial optimization: if we find an entry that 727 * already matches the sub-directory, then we know 728 * we're ok, and we can exit. 729 */ 730 while (pos < istate->cache_nr) { 731 struct cache_entry *p = istate->cache[pos]; 732 if ((ce_namelen(p) <= len) || 733 (p->name[len] != '/') || 734 memcmp(p->name, name, len)) 735 break; /* not our subdirectory */ 736 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE)) 737 /* 738 * p is at the same stage as our entry, and 739 * is a subdirectory of what we are looking 740 * at, so we cannot have conflicts at our 741 * level or anything shorter. 742 */ 743 return retval; 744 pos++; 745 } 746 } 747 return retval; 748} 749 750/* We may be in a situation where we already have path/file and path 751 * is being added, or we already have path and path/file is being 752 * added. Either one would result in a nonsense tree that has path 753 * twice when git-write-tree tries to write it out. Prevent it. 754 * 755 * If ok-to-replace is specified, we remove the conflicting entries 756 * from the cache so the caller should recompute the insert position. 757 * When this happens, we return non-zero. 758 */ 759static int check_file_directory_conflict(struct index_state *istate, 760 const struct cache_entry *ce, 761 int pos, int ok_to_replace) 762{ 763 int retval; 764 765 /* 766 * When ce is an "I am going away" entry, we allow it to be added 767 */ 768 if (ce->ce_flags & CE_REMOVE) 769 return 0; 770 771 /* 772 * We check if the path is a sub-path of a subsequent pathname 773 * first, since removing those will not change the position 774 * in the array. 775 */ 776 retval = has_file_name(istate, ce, pos, ok_to_replace); 777 778 /* 779 * Then check if the path might have a clashing sub-directory 780 * before it. 781 */ 782 return retval + has_dir_name(istate, ce, pos, ok_to_replace); 783} 784 785static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option) 786{ 787 int pos; 788 int ok_to_add = option & ADD_CACHE_OK_TO_ADD; 789 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE; 790 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK; 791 792 cache_tree_invalidate_path(istate->cache_tree, ce->name); 793 pos = index_name_pos(istate, ce->name, ce->ce_flags); 794 795 /* existing match? Just replace it. */ 796 if (pos >= 0) { 797 replace_index_entry(istate, pos, ce); 798 return 0; 799 } 800 pos = -pos-1; 801 802 /* 803 * Inserting a merged entry ("stage 0") into the index 804 * will always replace all non-merged entries.. 805 */ 806 if (pos < istate->cache_nr && ce_stage(ce) == 0) { 807 while (ce_same_name(istate->cache[pos], ce)) { 808 ok_to_add = 1; 809 if (!remove_index_entry_at(istate, pos)) 810 break; 811 } 812 } 813 814 if (!ok_to_add) 815 return -1; 816 if (!verify_path(ce->name)) 817 return -1; 818 819 if (!skip_df_check && 820 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) { 821 if (!ok_to_replace) 822 return error("'%s' appears as both a file and as a directory", 823 ce->name); 824 pos = index_name_pos(istate, ce->name, ce->ce_flags); 825 pos = -pos-1; 826 } 827 return pos + 1; 828} 829 830int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option) 831{ 832 int pos; 833 834 if (option & ADD_CACHE_JUST_APPEND) 835 pos = istate->cache_nr; 836 else { 837 int ret; 838 ret = add_index_entry_with_check(istate, ce, option); 839 if (ret <= 0) 840 return ret; 841 pos = ret - 1; 842 } 843 844 /* Make sure the array is big enough .. */ 845 if (istate->cache_nr == istate->cache_alloc) { 846 istate->cache_alloc = alloc_nr(istate->cache_alloc); 847 istate->cache = xrealloc(istate->cache, 848 istate->cache_alloc * sizeof(struct cache_entry *)); 849 } 850 851 /* Add it in.. */ 852 istate->cache_nr++; 853 if (istate->cache_nr > pos + 1) 854 memmove(istate->cache + pos + 1, 855 istate->cache + pos, 856 (istate->cache_nr - pos - 1) * sizeof(ce)); 857 set_index_entry(istate, pos, ce); 858 istate->cache_changed = 1; 859 return 0; 860} 861 862/* 863 * "refresh" does not calculate a new sha1 file or bring the 864 * cache up-to-date for mode/content changes. But what it 865 * _does_ do is to "re-match" the stat information of a file 866 * with the cache, so that you can refresh the cache for a 867 * file that hasn't been changed but where the stat entry is 868 * out of date. 869 * 870 * For example, you'd want to do this after doing a "git-read-tree", 871 * to link up the stat cache details with the proper files. 872 */ 873static struct cache_entry *refresh_cache_ent(struct index_state *istate, 874 struct cache_entry *ce, 875 unsigned int options, int *err) 876{ 877 struct stat st; 878 struct cache_entry *updated; 879 int changed, size; 880 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 881 882 if (ce_uptodate(ce)) 883 return ce; 884 885 if (lstat(ce->name, &st) < 0) { 886 if (err) 887 *err = errno; 888 return NULL; 889 } 890 891 changed = ie_match_stat(istate, ce, &st, options); 892 if (!changed) { 893 /* 894 * The path is unchanged. If we were told to ignore 895 * valid bit, then we did the actual stat check and 896 * found that the entry is unmodified. If the entry 897 * is not marked VALID, this is the place to mark it 898 * valid again, under "assume unchanged" mode. 899 */ 900 if (ignore_valid && assume_unchanged && 901 !(ce->ce_flags & CE_VALID)) 902 ; /* mark this one VALID again */ 903 else { 904 /* 905 * We do not mark the index itself "modified" 906 * because CE_UPTODATE flag is in-core only; 907 * we are not going to write this change out. 908 */ 909 ce_mark_uptodate(ce); 910 return ce; 911 } 912 } 913 914 if (ie_modified(istate, ce, &st, options)) { 915 if (err) 916 *err = EINVAL; 917 return NULL; 918 } 919 920 size = ce_size(ce); 921 updated = xmalloc(size); 922 memcpy(updated, ce, size); 923 fill_stat_cache_info(updated, &st); 924 /* 925 * If ignore_valid is not set, we should leave CE_VALID bit 926 * alone. Otherwise, paths marked with --no-assume-unchanged 927 * (i.e. things to be edited) will reacquire CE_VALID bit 928 * automatically, which is not really what we want. 929 */ 930 if (!ignore_valid && assume_unchanged && 931 !(ce->ce_flags & CE_VALID)) 932 updated->ce_flags &= ~CE_VALID; 933 934 return updated; 935} 936 937int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen) 938{ 939 int i; 940 int has_errors = 0; 941 int really = (flags & REFRESH_REALLY) != 0; 942 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0; 943 int quiet = (flags & REFRESH_QUIET) != 0; 944 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0; 945 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0; 946 947 for (i = 0; i < istate->cache_nr; i++) { 948 struct cache_entry *ce, *new; 949 int cache_errno = 0; 950 951 ce = istate->cache[i]; 952 if (ce_stage(ce)) { 953 while ((i < istate->cache_nr) && 954 ! strcmp(istate->cache[i]->name, ce->name)) 955 i++; 956 i--; 957 if (allow_unmerged) 958 continue; 959 printf("%s: needs merge\n", ce->name); 960 has_errors = 1; 961 continue; 962 } 963 964 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen)) 965 continue; 966 967 new = refresh_cache_ent(istate, ce, options, &cache_errno); 968 if (new == ce) 969 continue; 970 if (!new) { 971 if (not_new && cache_errno == ENOENT) 972 continue; 973 if (really && cache_errno == EINVAL) { 974 /* If we are doing --really-refresh that 975 * means the index is not valid anymore. 976 */ 977 ce->ce_flags &= ~CE_VALID; 978 istate->cache_changed = 1; 979 } 980 if (quiet) 981 continue; 982 printf("%s: needs update\n", ce->name); 983 has_errors = 1; 984 continue; 985 } 986 987 replace_index_entry(istate, i, new); 988 } 989 return has_errors; 990} 991 992struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really) 993{ 994 return refresh_cache_ent(&the_index, ce, really, NULL); 995} 996 997static int verify_hdr(struct cache_header *hdr, unsigned long size) 998{ 999 SHA_CTX c;1000 unsigned char sha1[20];10011002 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))1003 return error("bad signature");1004 if (hdr->hdr_version != htonl(2))1005 return error("bad index version");1006 SHA1_Init(&c);1007 SHA1_Update(&c, hdr, size - 20);1008 SHA1_Final(sha1, &c);1009 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))1010 return error("bad index file sha1 signature");1011 return 0;1012}10131014static int read_index_extension(struct index_state *istate,1015 const char *ext, void *data, unsigned long sz)1016{1017 switch (CACHE_EXT(ext)) {1018 case CACHE_EXT_TREE:1019 istate->cache_tree = cache_tree_read(data, sz);1020 break;1021 default:1022 if (*ext < 'A' || 'Z' < *ext)1023 return error("index uses %.4s extension, which we do not understand",1024 ext);1025 fprintf(stderr, "ignoring %.4s extension\n", ext);1026 break;1027 }1028 return 0;1029}10301031int read_index(struct index_state *istate)1032{1033 return read_index_from(istate, get_index_file());1034}10351036static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)1037{1038 size_t len;10391040 ce->ce_ctime = ntohl(ondisk->ctime.sec);1041 ce->ce_mtime = ntohl(ondisk->mtime.sec);1042 ce->ce_dev = ntohl(ondisk->dev);1043 ce->ce_ino = ntohl(ondisk->ino);1044 ce->ce_mode = ntohl(ondisk->mode);1045 ce->ce_uid = ntohl(ondisk->uid);1046 ce->ce_gid = ntohl(ondisk->gid);1047 ce->ce_size = ntohl(ondisk->size);1048 /* On-disk flags are just 16 bits */1049 ce->ce_flags = ntohs(ondisk->flags);1050 hashcpy(ce->sha1, ondisk->sha1);10511052 len = ce->ce_flags & CE_NAMEMASK;1053 if (len == CE_NAMEMASK)1054 len = strlen(ondisk->name);1055 /*1056 * NEEDSWORK: If the original index is crafted, this copy could1057 * go unchecked.1058 */1059 memcpy(ce->name, ondisk->name, len + 1);1060}10611062static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)1063{1064 long per_entry;10651066 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);10671068 /*1069 * Alignment can cause differences. This should be "alignof", but1070 * since that's a gcc'ism, just use the size of a pointer.1071 */1072 per_entry += sizeof(void *);1073 return ondisk_size + entries*per_entry;1074}10751076/* remember to discard_cache() before reading a different cache! */1077int read_index_from(struct index_state *istate, const char *path)1078{1079 int fd, i;1080 struct stat st;1081 unsigned long src_offset, dst_offset;1082 struct cache_header *hdr;1083 void *mmap;1084 size_t mmap_size;10851086 errno = EBUSY;1087 if (istate->alloc)1088 return istate->cache_nr;10891090 errno = ENOENT;1091 istate->timestamp = 0;1092 fd = open(path, O_RDONLY);1093 if (fd < 0) {1094 if (errno == ENOENT)1095 return 0;1096 die("index file open failed (%s)", strerror(errno));1097 }10981099 if (fstat(fd, &st))1100 die("cannot stat the open index (%s)", strerror(errno));11011102 errno = EINVAL;1103 mmap_size = xsize_t(st.st_size);1104 if (mmap_size < sizeof(struct cache_header) + 20)1105 die("index file smaller than expected");11061107 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);1108 close(fd);1109 if (mmap == MAP_FAILED)1110 die("unable to map index file");11111112 hdr = mmap;1113 if (verify_hdr(hdr, mmap_size) < 0)1114 goto unmap;11151116 istate->cache_nr = ntohl(hdr->hdr_entries);1117 istate->cache_alloc = alloc_nr(istate->cache_nr);1118 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));11191120 /*1121 * The disk format is actually larger than the in-memory format,1122 * due to space for nsec etc, so even though the in-memory one1123 * has room for a few more flags, we can allocate using the same1124 * index size1125 */1126 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));11271128 src_offset = sizeof(*hdr);1129 dst_offset = 0;1130 for (i = 0; i < istate->cache_nr; i++) {1131 struct ondisk_cache_entry *disk_ce;1132 struct cache_entry *ce;11331134 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);1135 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);1136 convert_from_disk(disk_ce, ce);1137 set_index_entry(istate, i, ce);11381139 src_offset += ondisk_ce_size(ce);1140 dst_offset += ce_size(ce);1141 }1142 istate->timestamp = st.st_mtime;1143 while (src_offset <= mmap_size - 20 - 8) {1144 /* After an array of active_nr index entries,1145 * there can be arbitrary number of extended1146 * sections, each of which is prefixed with1147 * extension name (4-byte) and section length1148 * in 4-byte network byte order.1149 */1150 unsigned long extsize;1151 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);1152 extsize = ntohl(extsize);1153 if (read_index_extension(istate,1154 (const char *) mmap + src_offset,1155 (char *) mmap + src_offset + 8,1156 extsize) < 0)1157 goto unmap;1158 src_offset += 8;1159 src_offset += extsize;1160 }1161 munmap(mmap, mmap_size);1162 return istate->cache_nr;11631164unmap:1165 munmap(mmap, mmap_size);1166 errno = EINVAL;1167 die("index file corrupt");1168}11691170int discard_index(struct index_state *istate)1171{1172 istate->cache_nr = 0;1173 istate->cache_changed = 0;1174 istate->timestamp = 0;1175 free_hash(&istate->name_hash);1176 cache_tree_free(&(istate->cache_tree));1177 free(istate->alloc);1178 istate->alloc = NULL;11791180 /* no need to throw away allocated active_cache */1181 return 0;1182}11831184int unmerged_index(const struct index_state *istate)1185{1186 int i;1187 for (i = 0; i < istate->cache_nr; i++) {1188 if (ce_stage(istate->cache[i]))1189 return 1;1190 }1191 return 0;1192}11931194#define WRITE_BUFFER_SIZE 81921195static unsigned char write_buffer[WRITE_BUFFER_SIZE];1196static unsigned long write_buffer_len;11971198static int ce_write_flush(SHA_CTX *context, int fd)1199{1200 unsigned int buffered = write_buffer_len;1201 if (buffered) {1202 SHA1_Update(context, write_buffer, buffered);1203 if (write_in_full(fd, write_buffer, buffered) != buffered)1204 return -1;1205 write_buffer_len = 0;1206 }1207 return 0;1208}12091210static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)1211{1212 while (len) {1213 unsigned int buffered = write_buffer_len;1214 unsigned int partial = WRITE_BUFFER_SIZE - buffered;1215 if (partial > len)1216 partial = len;1217 memcpy(write_buffer + buffered, data, partial);1218 buffered += partial;1219 if (buffered == WRITE_BUFFER_SIZE) {1220 write_buffer_len = buffered;1221 if (ce_write_flush(context, fd))1222 return -1;1223 buffered = 0;1224 }1225 write_buffer_len = buffered;1226 len -= partial;1227 data = (char *) data + partial;1228 }1229 return 0;1230}12311232static int write_index_ext_header(SHA_CTX *context, int fd,1233 unsigned int ext, unsigned int sz)1234{1235 ext = htonl(ext);1236 sz = htonl(sz);1237 return ((ce_write(context, fd, &ext, 4) < 0) ||1238 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;1239}12401241static int ce_flush(SHA_CTX *context, int fd)1242{1243 unsigned int left = write_buffer_len;12441245 if (left) {1246 write_buffer_len = 0;1247 SHA1_Update(context, write_buffer, left);1248 }12491250 /* Flush first if not enough space for SHA1 signature */1251 if (left + 20 > WRITE_BUFFER_SIZE) {1252 if (write_in_full(fd, write_buffer, left) != left)1253 return -1;1254 left = 0;1255 }12561257 /* Append the SHA1 signature at the end */1258 SHA1_Final(write_buffer + left, context);1259 left += 20;1260 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;1261}12621263static void ce_smudge_racily_clean_entry(struct cache_entry *ce)1264{1265 /*1266 * The only thing we care about in this function is to smudge the1267 * falsely clean entry due to touch-update-touch race, so we leave1268 * everything else as they are. We are called for entries whose1269 * ce_mtime match the index file mtime.1270 */1271 struct stat st;12721273 if (lstat(ce->name, &st) < 0)1274 return;1275 if (ce_match_stat_basic(ce, &st))1276 return;1277 if (ce_modified_check_fs(ce, &st)) {1278 /* This is "racily clean"; smudge it. Note that this1279 * is a tricky code. At first glance, it may appear1280 * that it can break with this sequence:1281 *1282 * $ echo xyzzy >frotz1283 * $ git-update-index --add frotz1284 * $ : >frotz1285 * $ sleep 31286 * $ echo filfre >nitfol1287 * $ git-update-index --add nitfol1288 *1289 * but it does not. When the second update-index runs,1290 * it notices that the entry "frotz" has the same timestamp1291 * as index, and if we were to smudge it by resetting its1292 * size to zero here, then the object name recorded1293 * in index is the 6-byte file but the cached stat information1294 * becomes zero --- which would then match what we would1295 * obtain from the filesystem next time we stat("frotz").1296 *1297 * However, the second update-index, before calling1298 * this function, notices that the cached size is 61299 * bytes and what is on the filesystem is an empty1300 * file, and never calls us, so the cached size information1301 * for "frotz" stays 6 which does not match the filesystem.1302 */1303 ce->ce_size = 0;1304 }1305}13061307static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)1308{1309 int size = ondisk_ce_size(ce);1310 struct ondisk_cache_entry *ondisk = xcalloc(1, size);13111312 ondisk->ctime.sec = htonl(ce->ce_ctime);1313 ondisk->ctime.nsec = 0;1314 ondisk->mtime.sec = htonl(ce->ce_mtime);1315 ondisk->mtime.nsec = 0;1316 ondisk->dev = htonl(ce->ce_dev);1317 ondisk->ino = htonl(ce->ce_ino);1318 ondisk->mode = htonl(ce->ce_mode);1319 ondisk->uid = htonl(ce->ce_uid);1320 ondisk->gid = htonl(ce->ce_gid);1321 ondisk->size = htonl(ce->ce_size);1322 hashcpy(ondisk->sha1, ce->sha1);1323 ondisk->flags = htons(ce->ce_flags);1324 memcpy(ondisk->name, ce->name, ce_namelen(ce));13251326 return ce_write(c, fd, ondisk, size);1327}13281329int write_index(const struct index_state *istate, int newfd)1330{1331 SHA_CTX c;1332 struct cache_header hdr;1333 int i, err, removed;1334 struct cache_entry **cache = istate->cache;1335 int entries = istate->cache_nr;13361337 for (i = removed = 0; i < entries; i++)1338 if (cache[i]->ce_flags & CE_REMOVE)1339 removed++;13401341 hdr.hdr_signature = htonl(CACHE_SIGNATURE);1342 hdr.hdr_version = htonl(2);1343 hdr.hdr_entries = htonl(entries - removed);13441345 SHA1_Init(&c);1346 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)1347 return -1;13481349 for (i = 0; i < entries; i++) {1350 struct cache_entry *ce = cache[i];1351 if (ce->ce_flags & CE_REMOVE)1352 continue;1353 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))1354 ce_smudge_racily_clean_entry(ce);1355 if (ce_write_entry(&c, newfd, ce) < 0)1356 return -1;1357 }13581359 /* Write extension data here */1360 if (istate->cache_tree) {1361 struct strbuf sb;13621363 strbuf_init(&sb, 0);1364 cache_tree_write(&sb, istate->cache_tree);1365 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 01366 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1367 strbuf_release(&sb);1368 if (err)1369 return -1;1370 }1371 return ce_flush(&c, newfd);1372}