1/* 2 * GIT - The information manager from hell 3 * 4 * Copyright (C) Linus Torvalds, 2005 5 */ 6#define NO_THE_INDEX_COMPATIBILITY_MACROS 7#include "cache.h" 8#include "cache-tree.h" 9#include "refs.h" 10#include "dir.h" 11#include "tree.h" 12#include "commit.h" 13#include "diff.h" 14#include "diffcore.h" 15#include "revision.h" 16#include "blob.h" 17 18/* Index extensions. 19 * 20 * The first letter should be 'A'..'Z' for extensions that are not 21 * necessary for a correct operation (i.e. optimization data). 22 * When new extensions are added that _needs_ to be understood in 23 * order to correctly interpret the index file, pick character that 24 * is outside the range, to cause the reader to abort. 25 */ 26 27#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) ) 28#define CACHE_EXT_TREE 0x54524545 /* "TREE" */ 29 30struct index_state the_index; 31 32static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 33{ 34 istate->cache[nr] = ce; 35 add_name_hash(istate, ce); 36} 37 38static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 39{ 40 struct cache_entry *old = istate->cache[nr]; 41 42 remove_name_hash(old); 43 set_index_entry(istate, nr, ce); 44 istate->cache_changed = 1; 45} 46 47void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name) 48{ 49 struct cache_entry *old = istate->cache[nr], *new; 50 int namelen = strlen(new_name); 51 52 new = xmalloc(cache_entry_size(namelen)); 53 copy_cache_entry(new, old); 54 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK); 55 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen); 56 memcpy(new->name, new_name, namelen + 1); 57 58 cache_tree_invalidate_path(istate->cache_tree, old->name); 59 remove_index_entry_at(istate, nr); 60 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE); 61} 62 63/* 64 * This only updates the "non-critical" parts of the directory 65 * cache, ie the parts that aren't tracked by GIT, and only used 66 * to validate the cache. 67 */ 68void fill_stat_cache_info(struct cache_entry *ce, struct stat *st) 69{ 70 ce->ce_ctime = st->st_ctime; 71 ce->ce_mtime = st->st_mtime; 72 ce->ce_dev = st->st_dev; 73 ce->ce_ino = st->st_ino; 74 ce->ce_uid = st->st_uid; 75 ce->ce_gid = st->st_gid; 76 ce->ce_size = st->st_size; 77 78 if (assume_unchanged) 79 ce->ce_flags |= CE_VALID; 80 81 if (S_ISREG(st->st_mode)) 82 ce_mark_uptodate(ce); 83} 84 85static int ce_compare_data(struct cache_entry *ce, struct stat *st) 86{ 87 int match = -1; 88 int fd = open(ce->name, O_RDONLY); 89 90 if (fd >= 0) { 91 unsigned char sha1[20]; 92 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name)) 93 match = hashcmp(sha1, ce->sha1); 94 /* index_fd() closed the file descriptor already */ 95 } 96 return match; 97} 98 99static int ce_compare_link(struct cache_entry *ce, size_t expected_size) 100{ 101 int match = -1; 102 char *target; 103 void *buffer; 104 unsigned long size; 105 enum object_type type; 106 int len; 107 108 target = xmalloc(expected_size); 109 len = readlink(ce->name, target, expected_size); 110 if (len != expected_size) { 111 free(target); 112 return -1; 113 } 114 buffer = read_sha1_file(ce->sha1, &type, &size); 115 if (!buffer) { 116 free(target); 117 return -1; 118 } 119 if (size == expected_size) 120 match = memcmp(buffer, target, size); 121 free(buffer); 122 free(target); 123 return match; 124} 125 126static int ce_compare_gitlink(struct cache_entry *ce) 127{ 128 unsigned char sha1[20]; 129 130 /* 131 * We don't actually require that the .git directory 132 * under GITLINK directory be a valid git directory. It 133 * might even be missing (in case nobody populated that 134 * sub-project). 135 * 136 * If so, we consider it always to match. 137 */ 138 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0) 139 return 0; 140 return hashcmp(sha1, ce->sha1); 141} 142 143static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st) 144{ 145 switch (st->st_mode & S_IFMT) { 146 case S_IFREG: 147 if (ce_compare_data(ce, st)) 148 return DATA_CHANGED; 149 break; 150 case S_IFLNK: 151 if (ce_compare_link(ce, xsize_t(st->st_size))) 152 return DATA_CHANGED; 153 break; 154 case S_IFDIR: 155 if (S_ISGITLINK(ce->ce_mode)) 156 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0; 157 default: 158 return TYPE_CHANGED; 159 } 160 return 0; 161} 162 163int is_empty_blob_sha1(const unsigned char *sha1) 164{ 165 static const unsigned char empty_blob_sha1[20] = { 166 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b, 167 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91 168 }; 169 170 return !hashcmp(sha1, empty_blob_sha1); 171} 172 173static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st) 174{ 175 unsigned int changed = 0; 176 177 if (ce->ce_flags & CE_REMOVE) 178 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; 179 180 switch (ce->ce_mode & S_IFMT) { 181 case S_IFREG: 182 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0; 183 /* We consider only the owner x bit to be relevant for 184 * "mode changes" 185 */ 186 if (trust_executable_bit && 187 (0100 & (ce->ce_mode ^ st->st_mode))) 188 changed |= MODE_CHANGED; 189 break; 190 case S_IFLNK: 191 if (!S_ISLNK(st->st_mode) && 192 (has_symlinks || !S_ISREG(st->st_mode))) 193 changed |= TYPE_CHANGED; 194 break; 195 case S_IFGITLINK: 196 /* We ignore most of the st_xxx fields for gitlinks */ 197 if (!S_ISDIR(st->st_mode)) 198 changed |= TYPE_CHANGED; 199 else if (ce_compare_gitlink(ce)) 200 changed |= DATA_CHANGED; 201 return changed; 202 default: 203 die("internal error: ce_mode is %o", ce->ce_mode); 204 } 205 if (ce->ce_mtime != (unsigned int) st->st_mtime) 206 changed |= MTIME_CHANGED; 207 if (trust_ctime && ce->ce_ctime != (unsigned int) st->st_ctime) 208 changed |= CTIME_CHANGED; 209 210 if (ce->ce_uid != (unsigned int) st->st_uid || 211 ce->ce_gid != (unsigned int) st->st_gid) 212 changed |= OWNER_CHANGED; 213 if (ce->ce_ino != (unsigned int) st->st_ino) 214 changed |= INODE_CHANGED; 215 216#ifdef USE_STDEV 217 /* 218 * st_dev breaks on network filesystems where different 219 * clients will have different views of what "device" 220 * the filesystem is on 221 */ 222 if (ce->ce_dev != (unsigned int) st->st_dev) 223 changed |= INODE_CHANGED; 224#endif 225 226 if (ce->ce_size != (unsigned int) st->st_size) 227 changed |= DATA_CHANGED; 228 229 /* Racily smudged entry? */ 230 if (!ce->ce_size) { 231 if (!is_empty_blob_sha1(ce->sha1)) 232 changed |= DATA_CHANGED; 233 } 234 235 return changed; 236} 237 238static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce) 239{ 240 return (!S_ISGITLINK(ce->ce_mode) && 241 istate->timestamp && 242 ((unsigned int)istate->timestamp) <= ce->ce_mtime); 243} 244 245int ie_match_stat(const struct index_state *istate, 246 struct cache_entry *ce, struct stat *st, 247 unsigned int options) 248{ 249 unsigned int changed; 250 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 251 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 252 253 /* 254 * If it's marked as always valid in the index, it's 255 * valid whatever the checked-out copy says. 256 */ 257 if (!ignore_valid && (ce->ce_flags & CE_VALID)) 258 return 0; 259 260 changed = ce_match_stat_basic(ce, st); 261 262 /* 263 * Within 1 second of this sequence: 264 * echo xyzzy >file && git-update-index --add file 265 * running this command: 266 * echo frotz >file 267 * would give a falsely clean cache entry. The mtime and 268 * length match the cache, and other stat fields do not change. 269 * 270 * We could detect this at update-index time (the cache entry 271 * being registered/updated records the same time as "now") 272 * and delay the return from git-update-index, but that would 273 * effectively mean we can make at most one commit per second, 274 * which is not acceptable. Instead, we check cache entries 275 * whose mtime are the same as the index file timestamp more 276 * carefully than others. 277 */ 278 if (!changed && is_racy_timestamp(istate, ce)) { 279 if (assume_racy_is_modified) 280 changed |= DATA_CHANGED; 281 else 282 changed |= ce_modified_check_fs(ce, st); 283 } 284 285 return changed; 286} 287 288int ie_modified(const struct index_state *istate, 289 struct cache_entry *ce, struct stat *st, unsigned int options) 290{ 291 int changed, changed_fs; 292 293 changed = ie_match_stat(istate, ce, st, options); 294 if (!changed) 295 return 0; 296 /* 297 * If the mode or type has changed, there's no point in trying 298 * to refresh the entry - it's not going to match 299 */ 300 if (changed & (MODE_CHANGED | TYPE_CHANGED)) 301 return changed; 302 303 /* 304 * Immediately after read-tree or update-index --cacheinfo, 305 * the length field is zero, as we have never even read the 306 * lstat(2) information once, and we cannot trust DATA_CHANGED 307 * returned by ie_match_stat() which in turn was returned by 308 * ce_match_stat_basic() to signal that the filesize of the 309 * blob changed. We have to actually go to the filesystem to 310 * see if the contents match, and if so, should answer "unchanged". 311 * 312 * The logic does not apply to gitlinks, as ce_match_stat_basic() 313 * already has checked the actual HEAD from the filesystem in the 314 * subproject. If ie_match_stat() already said it is different, 315 * then we know it is. 316 */ 317 if ((changed & DATA_CHANGED) && 318 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0)) 319 return changed; 320 321 changed_fs = ce_modified_check_fs(ce, st); 322 if (changed_fs) 323 return changed | changed_fs; 324 return 0; 325} 326 327int base_name_compare(const char *name1, int len1, int mode1, 328 const char *name2, int len2, int mode2) 329{ 330 unsigned char c1, c2; 331 int len = len1 < len2 ? len1 : len2; 332 int cmp; 333 334 cmp = memcmp(name1, name2, len); 335 if (cmp) 336 return cmp; 337 c1 = name1[len]; 338 c2 = name2[len]; 339 if (!c1 && S_ISDIR(mode1)) 340 c1 = '/'; 341 if (!c2 && S_ISDIR(mode2)) 342 c2 = '/'; 343 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0; 344} 345 346/* 347 * df_name_compare() is identical to base_name_compare(), except it 348 * compares conflicting directory/file entries as equal. Note that 349 * while a directory name compares as equal to a regular file, they 350 * then individually compare _differently_ to a filename that has 351 * a dot after the basename (because '\0' < '.' < '/'). 352 * 353 * This is used by routines that want to traverse the git namespace 354 * but then handle conflicting entries together when possible. 355 */ 356int df_name_compare(const char *name1, int len1, int mode1, 357 const char *name2, int len2, int mode2) 358{ 359 int len = len1 < len2 ? len1 : len2, cmp; 360 unsigned char c1, c2; 361 362 cmp = memcmp(name1, name2, len); 363 if (cmp) 364 return cmp; 365 /* Directories and files compare equal (same length, same name) */ 366 if (len1 == len2) 367 return 0; 368 c1 = name1[len]; 369 if (!c1 && S_ISDIR(mode1)) 370 c1 = '/'; 371 c2 = name2[len]; 372 if (!c2 && S_ISDIR(mode2)) 373 c2 = '/'; 374 if (c1 == '/' && !c2) 375 return 0; 376 if (c2 == '/' && !c1) 377 return 0; 378 return c1 - c2; 379} 380 381int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2) 382{ 383 int len1 = flags1 & CE_NAMEMASK; 384 int len2 = flags2 & CE_NAMEMASK; 385 int len = len1 < len2 ? len1 : len2; 386 int cmp; 387 388 cmp = memcmp(name1, name2, len); 389 if (cmp) 390 return cmp; 391 if (len1 < len2) 392 return -1; 393 if (len1 > len2) 394 return 1; 395 396 /* Compare stages */ 397 flags1 &= CE_STAGEMASK; 398 flags2 &= CE_STAGEMASK; 399 400 if (flags1 < flags2) 401 return -1; 402 if (flags1 > flags2) 403 return 1; 404 return 0; 405} 406 407int index_name_pos(const struct index_state *istate, const char *name, int namelen) 408{ 409 int first, last; 410 411 first = 0; 412 last = istate->cache_nr; 413 while (last > first) { 414 int next = (last + first) >> 1; 415 struct cache_entry *ce = istate->cache[next]; 416 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags); 417 if (!cmp) 418 return next; 419 if (cmp < 0) { 420 last = next; 421 continue; 422 } 423 first = next+1; 424 } 425 return -first-1; 426} 427 428/* Remove entry, return true if there are more entries to go.. */ 429int remove_index_entry_at(struct index_state *istate, int pos) 430{ 431 struct cache_entry *ce = istate->cache[pos]; 432 433 remove_name_hash(ce); 434 istate->cache_changed = 1; 435 istate->cache_nr--; 436 if (pos >= istate->cache_nr) 437 return 0; 438 memmove(istate->cache + pos, 439 istate->cache + pos + 1, 440 (istate->cache_nr - pos) * sizeof(struct cache_entry *)); 441 return 1; 442} 443 444int remove_file_from_index(struct index_state *istate, const char *path) 445{ 446 int pos = index_name_pos(istate, path, strlen(path)); 447 if (pos < 0) 448 pos = -pos-1; 449 cache_tree_invalidate_path(istate->cache_tree, path); 450 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 451 remove_index_entry_at(istate, pos); 452 return 0; 453} 454 455static int compare_name(struct cache_entry *ce, const char *path, int namelen) 456{ 457 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); 458} 459 460static int index_name_pos_also_unmerged(struct index_state *istate, 461 const char *path, int namelen) 462{ 463 int pos = index_name_pos(istate, path, namelen); 464 struct cache_entry *ce; 465 466 if (pos >= 0) 467 return pos; 468 469 /* maybe unmerged? */ 470 pos = -1 - pos; 471 if (pos >= istate->cache_nr || 472 compare_name((ce = istate->cache[pos]), path, namelen)) 473 return -1; 474 475 /* order of preference: stage 2, 1, 3 */ 476 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && 477 ce_stage((ce = istate->cache[pos + 1])) == 2 && 478 !compare_name(ce, path, namelen)) 479 pos++; 480 return pos; 481} 482 483static int different_name(struct cache_entry *ce, struct cache_entry *alias) 484{ 485 int len = ce_namelen(ce); 486 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len); 487} 488 489/* 490 * If we add a filename that aliases in the cache, we will use the 491 * name that we already have - but we don't want to update the same 492 * alias twice, because that implies that there were actually two 493 * different files with aliasing names! 494 * 495 * So we use the CE_ADDED flag to verify that the alias was an old 496 * one before we accept it as 497 */ 498static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias) 499{ 500 int len; 501 struct cache_entry *new; 502 503 if (alias->ce_flags & CE_ADDED) 504 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); 505 506 /* Ok, create the new entry using the name of the existing alias */ 507 len = ce_namelen(alias); 508 new = xcalloc(1, cache_entry_size(len)); 509 memcpy(new->name, alias->name, len); 510 copy_cache_entry(new, ce); 511 free(ce); 512 return new; 513} 514 515static void record_intent_to_add(struct cache_entry *ce) 516{ 517 unsigned char sha1[20]; 518 if (write_sha1_file("", 0, blob_type, sha1)) 519 die("cannot create an empty blob in the object database"); 520 hashcpy(ce->sha1, sha1); 521} 522 523int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags) 524{ 525 int size, namelen, was_same; 526 mode_t st_mode = st->st_mode; 527 struct cache_entry *ce, *alias; 528 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY; 529 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); 530 int pretend = flags & ADD_CACHE_PRETEND; 531 int intent_only = flags & ADD_CACHE_INTENT; 532 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| 533 (intent_only ? ADD_CACHE_NEW_ONLY : 0)); 534 535 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 536 return error("%s: can only add regular files, symbolic links or git-directories", path); 537 538 namelen = strlen(path); 539 if (S_ISDIR(st_mode)) { 540 while (namelen && path[namelen-1] == '/') 541 namelen--; 542 } 543 size = cache_entry_size(namelen); 544 ce = xcalloc(1, size); 545 memcpy(ce->name, path, namelen); 546 ce->ce_flags = namelen; 547 if (!intent_only) 548 fill_stat_cache_info(ce, st); 549 else 550 ce->ce_flags |= CE_INTENT_TO_ADD; 551 552 if (trust_executable_bit && has_symlinks) 553 ce->ce_mode = create_ce_mode(st_mode); 554 else { 555 /* If there is an existing entry, pick the mode bits and type 556 * from it, otherwise assume unexecutable regular file. 557 */ 558 struct cache_entry *ent; 559 int pos = index_name_pos_also_unmerged(istate, path, namelen); 560 561 ent = (0 <= pos) ? istate->cache[pos] : NULL; 562 ce->ce_mode = ce_mode_from_stat(ent, st_mode); 563 } 564 565 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case); 566 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) { 567 /* Nothing changed, really */ 568 free(ce); 569 ce_mark_uptodate(alias); 570 alias->ce_flags |= CE_ADDED; 571 return 0; 572 } 573 if (!intent_only) { 574 if (index_path(ce->sha1, path, st, 1)) 575 return error("unable to index file %s", path); 576 } else 577 record_intent_to_add(ce); 578 579 if (ignore_case && alias && different_name(ce, alias)) 580 ce = create_alias_ce(ce, alias); 581 ce->ce_flags |= CE_ADDED; 582 583 /* It was suspected to be racily clean, but it turns out to be Ok */ 584 was_same = (alias && 585 !ce_stage(alias) && 586 !hashcmp(alias->sha1, ce->sha1) && 587 ce->ce_mode == alias->ce_mode); 588 589 if (pretend) 590 ; 591 else if (add_index_entry(istate, ce, add_option)) 592 return error("unable to add %s to index",path); 593 if (verbose && !was_same) 594 printf("add '%s'\n", path); 595 return 0; 596} 597 598int add_file_to_index(struct index_state *istate, const char *path, int flags) 599{ 600 struct stat st; 601 if (lstat(path, &st)) 602 die("%s: unable to stat (%s)", path, strerror(errno)); 603 return add_to_index(istate, path, &st, flags); 604} 605 606struct cache_entry *make_cache_entry(unsigned int mode, 607 const unsigned char *sha1, const char *path, int stage, 608 int refresh) 609{ 610 int size, len; 611 struct cache_entry *ce; 612 613 if (!verify_path(path)) { 614 error("Invalid path '%s'", path); 615 return NULL; 616 } 617 618 len = strlen(path); 619 size = cache_entry_size(len); 620 ce = xcalloc(1, size); 621 622 hashcpy(ce->sha1, sha1); 623 memcpy(ce->name, path, len); 624 ce->ce_flags = create_ce_flags(len, stage); 625 ce->ce_mode = create_ce_mode(mode); 626 627 if (refresh) 628 return refresh_cache_entry(ce, 0); 629 630 return ce; 631} 632 633int ce_same_name(struct cache_entry *a, struct cache_entry *b) 634{ 635 int len = ce_namelen(a); 636 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 637} 638 639int ce_path_match(const struct cache_entry *ce, const char **pathspec) 640{ 641 const char *match, *name; 642 int len; 643 644 if (!pathspec) 645 return 1; 646 647 len = ce_namelen(ce); 648 name = ce->name; 649 while ((match = *pathspec++) != NULL) { 650 int matchlen = strlen(match); 651 if (matchlen > len) 652 continue; 653 if (memcmp(name, match, matchlen)) 654 continue; 655 if (matchlen && name[matchlen-1] == '/') 656 return 1; 657 if (name[matchlen] == '/' || !name[matchlen]) 658 return 1; 659 if (!matchlen) 660 return 1; 661 } 662 return 0; 663} 664 665/* 666 * We fundamentally don't like some paths: we don't want 667 * dot or dot-dot anywhere, and for obvious reasons don't 668 * want to recurse into ".git" either. 669 * 670 * Also, we don't want double slashes or slashes at the 671 * end that can make pathnames ambiguous. 672 */ 673static int verify_dotfile(const char *rest) 674{ 675 /* 676 * The first character was '.', but that 677 * has already been discarded, we now test 678 * the rest. 679 */ 680 switch (*rest) { 681 /* "." is not allowed */ 682 case '\0': case '/': 683 return 0; 684 685 /* 686 * ".git" followed by NUL or slash is bad. This 687 * shares the path end test with the ".." case. 688 */ 689 case 'g': 690 if (rest[1] != 'i') 691 break; 692 if (rest[2] != 't') 693 break; 694 rest += 2; 695 /* fallthrough */ 696 case '.': 697 if (rest[1] == '\0' || rest[1] == '/') 698 return 0; 699 } 700 return 1; 701} 702 703int verify_path(const char *path) 704{ 705 char c; 706 707 goto inside; 708 for (;;) { 709 if (!c) 710 return 1; 711 if (c == '/') { 712inside: 713 c = *path++; 714 switch (c) { 715 default: 716 continue; 717 case '/': case '\0': 718 break; 719 case '.': 720 if (verify_dotfile(path)) 721 continue; 722 } 723 return 0; 724 } 725 c = *path++; 726 } 727} 728 729/* 730 * Do we have another file that has the beginning components being a 731 * proper superset of the name we're trying to add? 732 */ 733static int has_file_name(struct index_state *istate, 734 const struct cache_entry *ce, int pos, int ok_to_replace) 735{ 736 int retval = 0; 737 int len = ce_namelen(ce); 738 int stage = ce_stage(ce); 739 const char *name = ce->name; 740 741 while (pos < istate->cache_nr) { 742 struct cache_entry *p = istate->cache[pos++]; 743 744 if (len >= ce_namelen(p)) 745 break; 746 if (memcmp(name, p->name, len)) 747 break; 748 if (ce_stage(p) != stage) 749 continue; 750 if (p->name[len] != '/') 751 continue; 752 if (p->ce_flags & CE_REMOVE) 753 continue; 754 retval = -1; 755 if (!ok_to_replace) 756 break; 757 remove_index_entry_at(istate, --pos); 758 } 759 return retval; 760} 761 762/* 763 * Do we have another file with a pathname that is a proper 764 * subset of the name we're trying to add? 765 */ 766static int has_dir_name(struct index_state *istate, 767 const struct cache_entry *ce, int pos, int ok_to_replace) 768{ 769 int retval = 0; 770 int stage = ce_stage(ce); 771 const char *name = ce->name; 772 const char *slash = name + ce_namelen(ce); 773 774 for (;;) { 775 int len; 776 777 for (;;) { 778 if (*--slash == '/') 779 break; 780 if (slash <= ce->name) 781 return retval; 782 } 783 len = slash - name; 784 785 pos = index_name_pos(istate, name, create_ce_flags(len, stage)); 786 if (pos >= 0) { 787 /* 788 * Found one, but not so fast. This could 789 * be a marker that says "I was here, but 790 * I am being removed". Such an entry is 791 * not a part of the resulting tree, and 792 * it is Ok to have a directory at the same 793 * path. 794 */ 795 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) { 796 retval = -1; 797 if (!ok_to_replace) 798 break; 799 remove_index_entry_at(istate, pos); 800 continue; 801 } 802 } 803 else 804 pos = -pos-1; 805 806 /* 807 * Trivial optimization: if we find an entry that 808 * already matches the sub-directory, then we know 809 * we're ok, and we can exit. 810 */ 811 while (pos < istate->cache_nr) { 812 struct cache_entry *p = istate->cache[pos]; 813 if ((ce_namelen(p) <= len) || 814 (p->name[len] != '/') || 815 memcmp(p->name, name, len)) 816 break; /* not our subdirectory */ 817 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE)) 818 /* 819 * p is at the same stage as our entry, and 820 * is a subdirectory of what we are looking 821 * at, so we cannot have conflicts at our 822 * level or anything shorter. 823 */ 824 return retval; 825 pos++; 826 } 827 } 828 return retval; 829} 830 831/* We may be in a situation where we already have path/file and path 832 * is being added, or we already have path and path/file is being 833 * added. Either one would result in a nonsense tree that has path 834 * twice when git-write-tree tries to write it out. Prevent it. 835 * 836 * If ok-to-replace is specified, we remove the conflicting entries 837 * from the cache so the caller should recompute the insert position. 838 * When this happens, we return non-zero. 839 */ 840static int check_file_directory_conflict(struct index_state *istate, 841 const struct cache_entry *ce, 842 int pos, int ok_to_replace) 843{ 844 int retval; 845 846 /* 847 * When ce is an "I am going away" entry, we allow it to be added 848 */ 849 if (ce->ce_flags & CE_REMOVE) 850 return 0; 851 852 /* 853 * We check if the path is a sub-path of a subsequent pathname 854 * first, since removing those will not change the position 855 * in the array. 856 */ 857 retval = has_file_name(istate, ce, pos, ok_to_replace); 858 859 /* 860 * Then check if the path might have a clashing sub-directory 861 * before it. 862 */ 863 return retval + has_dir_name(istate, ce, pos, ok_to_replace); 864} 865 866static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option) 867{ 868 int pos; 869 int ok_to_add = option & ADD_CACHE_OK_TO_ADD; 870 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE; 871 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK; 872 int new_only = option & ADD_CACHE_NEW_ONLY; 873 874 cache_tree_invalidate_path(istate->cache_tree, ce->name); 875 pos = index_name_pos(istate, ce->name, ce->ce_flags); 876 877 /* existing match? Just replace it. */ 878 if (pos >= 0) { 879 if (!new_only) 880 replace_index_entry(istate, pos, ce); 881 return 0; 882 } 883 pos = -pos-1; 884 885 /* 886 * Inserting a merged entry ("stage 0") into the index 887 * will always replace all non-merged entries.. 888 */ 889 if (pos < istate->cache_nr && ce_stage(ce) == 0) { 890 while (ce_same_name(istate->cache[pos], ce)) { 891 ok_to_add = 1; 892 if (!remove_index_entry_at(istate, pos)) 893 break; 894 } 895 } 896 897 if (!ok_to_add) 898 return -1; 899 if (!verify_path(ce->name)) 900 return error("Invalid path '%s'", ce->name); 901 902 if (!skip_df_check && 903 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) { 904 if (!ok_to_replace) 905 return error("'%s' appears as both a file and as a directory", 906 ce->name); 907 pos = index_name_pos(istate, ce->name, ce->ce_flags); 908 pos = -pos-1; 909 } 910 return pos + 1; 911} 912 913int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option) 914{ 915 int pos; 916 917 if (option & ADD_CACHE_JUST_APPEND) 918 pos = istate->cache_nr; 919 else { 920 int ret; 921 ret = add_index_entry_with_check(istate, ce, option); 922 if (ret <= 0) 923 return ret; 924 pos = ret - 1; 925 } 926 927 /* Make sure the array is big enough .. */ 928 if (istate->cache_nr == istate->cache_alloc) { 929 istate->cache_alloc = alloc_nr(istate->cache_alloc); 930 istate->cache = xrealloc(istate->cache, 931 istate->cache_alloc * sizeof(struct cache_entry *)); 932 } 933 934 /* Add it in.. */ 935 istate->cache_nr++; 936 if (istate->cache_nr > pos + 1) 937 memmove(istate->cache + pos + 1, 938 istate->cache + pos, 939 (istate->cache_nr - pos - 1) * sizeof(ce)); 940 set_index_entry(istate, pos, ce); 941 istate->cache_changed = 1; 942 return 0; 943} 944 945/* 946 * "refresh" does not calculate a new sha1 file or bring the 947 * cache up-to-date for mode/content changes. But what it 948 * _does_ do is to "re-match" the stat information of a file 949 * with the cache, so that you can refresh the cache for a 950 * file that hasn't been changed but where the stat entry is 951 * out of date. 952 * 953 * For example, you'd want to do this after doing a "git-read-tree", 954 * to link up the stat cache details with the proper files. 955 */ 956static struct cache_entry *refresh_cache_ent(struct index_state *istate, 957 struct cache_entry *ce, 958 unsigned int options, int *err) 959{ 960 struct stat st; 961 struct cache_entry *updated; 962 int changed, size; 963 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 964 965 if (ce_uptodate(ce)) 966 return ce; 967 968 /* 969 * CE_VALID means the user promised us that the change to 970 * the work tree does not matter and told us not to worry. 971 */ 972 if (!ignore_valid && (ce->ce_flags & CE_VALID)) { 973 ce_mark_uptodate(ce); 974 return ce; 975 } 976 977 if (lstat(ce->name, &st) < 0) { 978 if (err) 979 *err = errno; 980 return NULL; 981 } 982 983 changed = ie_match_stat(istate, ce, &st, options); 984 if (!changed) { 985 /* 986 * The path is unchanged. If we were told to ignore 987 * valid bit, then we did the actual stat check and 988 * found that the entry is unmodified. If the entry 989 * is not marked VALID, this is the place to mark it 990 * valid again, under "assume unchanged" mode. 991 */ 992 if (ignore_valid && assume_unchanged && 993 !(ce->ce_flags & CE_VALID)) 994 ; /* mark this one VALID again */ 995 else { 996 /* 997 * We do not mark the index itself "modified" 998 * because CE_UPTODATE flag is in-core only; 999 * we are not going to write this change out.1000 */1001 ce_mark_uptodate(ce);1002 return ce;1003 }1004 }10051006 if (ie_modified(istate, ce, &st, options)) {1007 if (err)1008 *err = EINVAL;1009 return NULL;1010 }10111012 size = ce_size(ce);1013 updated = xmalloc(size);1014 memcpy(updated, ce, size);1015 fill_stat_cache_info(updated, &st);1016 /*1017 * If ignore_valid is not set, we should leave CE_VALID bit1018 * alone. Otherwise, paths marked with --no-assume-unchanged1019 * (i.e. things to be edited) will reacquire CE_VALID bit1020 * automatically, which is not really what we want.1021 */1022 if (!ignore_valid && assume_unchanged &&1023 !(ce->ce_flags & CE_VALID))1024 updated->ce_flags &= ~CE_VALID;10251026 return updated;1027}10281029int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)1030{1031 int i;1032 int has_errors = 0;1033 int really = (flags & REFRESH_REALLY) != 0;1034 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;1035 int quiet = (flags & REFRESH_QUIET) != 0;1036 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;1037 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;1038 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;1039 const char *needs_update_message;10401041 needs_update_message = ((flags & REFRESH_SAY_CHANGED)1042 ? "locally modified" : "needs update");1043 for (i = 0; i < istate->cache_nr; i++) {1044 struct cache_entry *ce, *new;1045 int cache_errno = 0;10461047 ce = istate->cache[i];1048 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))1049 continue;10501051 if (ce_stage(ce)) {1052 while ((i < istate->cache_nr) &&1053 ! strcmp(istate->cache[i]->name, ce->name))1054 i++;1055 i--;1056 if (allow_unmerged)1057 continue;1058 printf("%s: needs merge\n", ce->name);1059 has_errors = 1;1060 continue;1061 }10621063 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))1064 continue;10651066 new = refresh_cache_ent(istate, ce, options, &cache_errno);1067 if (new == ce)1068 continue;1069 if (!new) {1070 if (not_new && cache_errno == ENOENT)1071 continue;1072 if (really && cache_errno == EINVAL) {1073 /* If we are doing --really-refresh that1074 * means the index is not valid anymore.1075 */1076 ce->ce_flags &= ~CE_VALID;1077 istate->cache_changed = 1;1078 }1079 if (quiet)1080 continue;1081 printf("%s: %s\n", ce->name, needs_update_message);1082 has_errors = 1;1083 continue;1084 }10851086 replace_index_entry(istate, i, new);1087 }1088 return has_errors;1089}10901091struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)1092{1093 return refresh_cache_ent(&the_index, ce, really, NULL);1094}10951096static int verify_hdr(struct cache_header *hdr, unsigned long size)1097{1098 git_SHA_CTX c;1099 unsigned char sha1[20];11001101 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))1102 return error("bad signature");1103 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))1104 return error("bad index version");1105 git_SHA1_Init(&c);1106 git_SHA1_Update(&c, hdr, size - 20);1107 git_SHA1_Final(sha1, &c);1108 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))1109 return error("bad index file sha1 signature");1110 return 0;1111}11121113static int read_index_extension(struct index_state *istate,1114 const char *ext, void *data, unsigned long sz)1115{1116 switch (CACHE_EXT(ext)) {1117 case CACHE_EXT_TREE:1118 istate->cache_tree = cache_tree_read(data, sz);1119 break;1120 default:1121 if (*ext < 'A' || 'Z' < *ext)1122 return error("index uses %.4s extension, which we do not understand",1123 ext);1124 fprintf(stderr, "ignoring %.4s extension\n", ext);1125 break;1126 }1127 return 0;1128}11291130int read_index(struct index_state *istate)1131{1132 return read_index_from(istate, get_index_file());1133}11341135static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)1136{1137 size_t len;1138 const char *name;11391140 ce->ce_ctime = ntohl(ondisk->ctime.sec);1141 ce->ce_mtime = ntohl(ondisk->mtime.sec);1142 ce->ce_dev = ntohl(ondisk->dev);1143 ce->ce_ino = ntohl(ondisk->ino);1144 ce->ce_mode = ntohl(ondisk->mode);1145 ce->ce_uid = ntohl(ondisk->uid);1146 ce->ce_gid = ntohl(ondisk->gid);1147 ce->ce_size = ntohl(ondisk->size);1148 /* On-disk flags are just 16 bits */1149 ce->ce_flags = ntohs(ondisk->flags);11501151 hashcpy(ce->sha1, ondisk->sha1);11521153 len = ce->ce_flags & CE_NAMEMASK;11541155 if (ce->ce_flags & CE_EXTENDED) {1156 struct ondisk_cache_entry_extended *ondisk2;1157 int extended_flags;1158 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1159 extended_flags = ntohs(ondisk2->flags2) << 16;1160 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1161 if (extended_flags & ~CE_EXTENDED_FLAGS)1162 die("Unknown index entry format %08x", extended_flags);1163 ce->ce_flags |= extended_flags;1164 name = ondisk2->name;1165 }1166 else1167 name = ondisk->name;11681169 if (len == CE_NAMEMASK)1170 len = strlen(name);1171 /*1172 * NEEDSWORK: If the original index is crafted, this copy could1173 * go unchecked.1174 */1175 memcpy(ce->name, name, len + 1);1176}11771178static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)1179{1180 long per_entry;11811182 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);11831184 /*1185 * Alignment can cause differences. This should be "alignof", but1186 * since that's a gcc'ism, just use the size of a pointer.1187 */1188 per_entry += sizeof(void *);1189 return ondisk_size + entries*per_entry;1190}11911192/* remember to discard_cache() before reading a different cache! */1193int read_index_from(struct index_state *istate, const char *path)1194{1195 int fd, i;1196 struct stat st;1197 unsigned long src_offset, dst_offset;1198 struct cache_header *hdr;1199 void *mmap;1200 size_t mmap_size;12011202 errno = EBUSY;1203 if (istate->initialized)1204 return istate->cache_nr;12051206 errno = ENOENT;1207 istate->timestamp = 0;1208 fd = open(path, O_RDONLY);1209 if (fd < 0) {1210 if (errno == ENOENT)1211 return 0;1212 die("index file open failed (%s)", strerror(errno));1213 }12141215 if (fstat(fd, &st))1216 die("cannot stat the open index (%s)", strerror(errno));12171218 errno = EINVAL;1219 mmap_size = xsize_t(st.st_size);1220 if (mmap_size < sizeof(struct cache_header) + 20)1221 die("index file smaller than expected");12221223 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);1224 close(fd);1225 if (mmap == MAP_FAILED)1226 die("unable to map index file");12271228 hdr = mmap;1229 if (verify_hdr(hdr, mmap_size) < 0)1230 goto unmap;12311232 istate->cache_nr = ntohl(hdr->hdr_entries);1233 istate->cache_alloc = alloc_nr(istate->cache_nr);1234 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));12351236 /*1237 * The disk format is actually larger than the in-memory format,1238 * due to space for nsec etc, so even though the in-memory one1239 * has room for a few more flags, we can allocate using the same1240 * index size1241 */1242 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));1243 istate->initialized = 1;12441245 src_offset = sizeof(*hdr);1246 dst_offset = 0;1247 for (i = 0; i < istate->cache_nr; i++) {1248 struct ondisk_cache_entry *disk_ce;1249 struct cache_entry *ce;12501251 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);1252 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);1253 convert_from_disk(disk_ce, ce);1254 set_index_entry(istate, i, ce);12551256 src_offset += ondisk_ce_size(ce);1257 dst_offset += ce_size(ce);1258 }1259 istate->timestamp = st.st_mtime;1260 while (src_offset <= mmap_size - 20 - 8) {1261 /* After an array of active_nr index entries,1262 * there can be arbitrary number of extended1263 * sections, each of which is prefixed with1264 * extension name (4-byte) and section length1265 * in 4-byte network byte order.1266 */1267 unsigned long extsize;1268 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);1269 extsize = ntohl(extsize);1270 if (read_index_extension(istate,1271 (const char *) mmap + src_offset,1272 (char *) mmap + src_offset + 8,1273 extsize) < 0)1274 goto unmap;1275 src_offset += 8;1276 src_offset += extsize;1277 }1278 munmap(mmap, mmap_size);1279 return istate->cache_nr;12801281unmap:1282 munmap(mmap, mmap_size);1283 errno = EINVAL;1284 die("index file corrupt");1285}12861287int is_index_unborn(struct index_state *istate)1288{1289 return (!istate->cache_nr && !istate->alloc && !istate->timestamp);1290}12911292int discard_index(struct index_state *istate)1293{1294 istate->cache_nr = 0;1295 istate->cache_changed = 0;1296 istate->timestamp = 0;1297 istate->name_hash_initialized = 0;1298 free_hash(&istate->name_hash);1299 cache_tree_free(&(istate->cache_tree));1300 free(istate->alloc);1301 istate->alloc = NULL;1302 istate->initialized = 0;13031304 /* no need to throw away allocated active_cache */1305 return 0;1306}13071308int unmerged_index(const struct index_state *istate)1309{1310 int i;1311 for (i = 0; i < istate->cache_nr; i++) {1312 if (ce_stage(istate->cache[i]))1313 return 1;1314 }1315 return 0;1316}13171318#define WRITE_BUFFER_SIZE 81921319static unsigned char write_buffer[WRITE_BUFFER_SIZE];1320static unsigned long write_buffer_len;13211322static int ce_write_flush(git_SHA_CTX *context, int fd)1323{1324 unsigned int buffered = write_buffer_len;1325 if (buffered) {1326 git_SHA1_Update(context, write_buffer, buffered);1327 if (write_in_full(fd, write_buffer, buffered) != buffered)1328 return -1;1329 write_buffer_len = 0;1330 }1331 return 0;1332}13331334static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)1335{1336 while (len) {1337 unsigned int buffered = write_buffer_len;1338 unsigned int partial = WRITE_BUFFER_SIZE - buffered;1339 if (partial > len)1340 partial = len;1341 memcpy(write_buffer + buffered, data, partial);1342 buffered += partial;1343 if (buffered == WRITE_BUFFER_SIZE) {1344 write_buffer_len = buffered;1345 if (ce_write_flush(context, fd))1346 return -1;1347 buffered = 0;1348 }1349 write_buffer_len = buffered;1350 len -= partial;1351 data = (char *) data + partial;1352 }1353 return 0;1354}13551356static int write_index_ext_header(git_SHA_CTX *context, int fd,1357 unsigned int ext, unsigned int sz)1358{1359 ext = htonl(ext);1360 sz = htonl(sz);1361 return ((ce_write(context, fd, &ext, 4) < 0) ||1362 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;1363}13641365static int ce_flush(git_SHA_CTX *context, int fd)1366{1367 unsigned int left = write_buffer_len;13681369 if (left) {1370 write_buffer_len = 0;1371 git_SHA1_Update(context, write_buffer, left);1372 }13731374 /* Flush first if not enough space for SHA1 signature */1375 if (left + 20 > WRITE_BUFFER_SIZE) {1376 if (write_in_full(fd, write_buffer, left) != left)1377 return -1;1378 left = 0;1379 }13801381 /* Append the SHA1 signature at the end */1382 git_SHA1_Final(write_buffer + left, context);1383 left += 20;1384 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;1385}13861387static void ce_smudge_racily_clean_entry(struct cache_entry *ce)1388{1389 /*1390 * The only thing we care about in this function is to smudge the1391 * falsely clean entry due to touch-update-touch race, so we leave1392 * everything else as they are. We are called for entries whose1393 * ce_mtime match the index file mtime.1394 *1395 * Note that this actually does not do much for gitlinks, for1396 * which ce_match_stat_basic() always goes to the actual1397 * contents. The caller checks with is_racy_timestamp() which1398 * always says "no" for gitlinks, so we are not called for them ;-)1399 */1400 struct stat st;14011402 if (lstat(ce->name, &st) < 0)1403 return;1404 if (ce_match_stat_basic(ce, &st))1405 return;1406 if (ce_modified_check_fs(ce, &st)) {1407 /* This is "racily clean"; smudge it. Note that this1408 * is a tricky code. At first glance, it may appear1409 * that it can break with this sequence:1410 *1411 * $ echo xyzzy >frotz1412 * $ git-update-index --add frotz1413 * $ : >frotz1414 * $ sleep 31415 * $ echo filfre >nitfol1416 * $ git-update-index --add nitfol1417 *1418 * but it does not. When the second update-index runs,1419 * it notices that the entry "frotz" has the same timestamp1420 * as index, and if we were to smudge it by resetting its1421 * size to zero here, then the object name recorded1422 * in index is the 6-byte file but the cached stat information1423 * becomes zero --- which would then match what we would1424 * obtain from the filesystem next time we stat("frotz").1425 *1426 * However, the second update-index, before calling1427 * this function, notices that the cached size is 61428 * bytes and what is on the filesystem is an empty1429 * file, and never calls us, so the cached size information1430 * for "frotz" stays 6 which does not match the filesystem.1431 */1432 ce->ce_size = 0;1433 }1434}14351436static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)1437{1438 int size = ondisk_ce_size(ce);1439 struct ondisk_cache_entry *ondisk = xcalloc(1, size);1440 char *name;14411442 ondisk->ctime.sec = htonl(ce->ce_ctime);1443 ondisk->ctime.nsec = 0;1444 ondisk->mtime.sec = htonl(ce->ce_mtime);1445 ondisk->mtime.nsec = 0;1446 ondisk->dev = htonl(ce->ce_dev);1447 ondisk->ino = htonl(ce->ce_ino);1448 ondisk->mode = htonl(ce->ce_mode);1449 ondisk->uid = htonl(ce->ce_uid);1450 ondisk->gid = htonl(ce->ce_gid);1451 ondisk->size = htonl(ce->ce_size);1452 hashcpy(ondisk->sha1, ce->sha1);1453 ondisk->flags = htons(ce->ce_flags);1454 if (ce->ce_flags & CE_EXTENDED) {1455 struct ondisk_cache_entry_extended *ondisk2;1456 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1457 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);1458 name = ondisk2->name;1459 }1460 else1461 name = ondisk->name;1462 memcpy(name, ce->name, ce_namelen(ce));14631464 return ce_write(c, fd, ondisk, size);1465}14661467int write_index(const struct index_state *istate, int newfd)1468{1469 git_SHA_CTX c;1470 struct cache_header hdr;1471 int i, err, removed, extended;1472 struct cache_entry **cache = istate->cache;1473 int entries = istate->cache_nr;14741475 for (i = removed = extended = 0; i < entries; i++) {1476 if (cache[i]->ce_flags & CE_REMOVE)1477 removed++;14781479 /* reduce extended entries if possible */1480 cache[i]->ce_flags &= ~CE_EXTENDED;1481 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {1482 extended++;1483 cache[i]->ce_flags |= CE_EXTENDED;1484 }1485 }14861487 hdr.hdr_signature = htonl(CACHE_SIGNATURE);1488 /* for extended format, increase version so older git won't try to read it */1489 hdr.hdr_version = htonl(extended ? 3 : 2);1490 hdr.hdr_entries = htonl(entries - removed);14911492 git_SHA1_Init(&c);1493 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)1494 return -1;14951496 for (i = 0; i < entries; i++) {1497 struct cache_entry *ce = cache[i];1498 if (ce->ce_flags & CE_REMOVE)1499 continue;1500 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))1501 ce_smudge_racily_clean_entry(ce);1502 if (ce_write_entry(&c, newfd, ce) < 0)1503 return -1;1504 }15051506 /* Write extension data here */1507 if (istate->cache_tree) {1508 struct strbuf sb = STRBUF_INIT;15091510 cache_tree_write(&sb, istate->cache_tree);1511 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 01512 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1513 strbuf_release(&sb);1514 if (err)1515 return -1;1516 }1517 return ce_flush(&c, newfd);1518}15191520/*1521 * Read the index file that is potentially unmerged into given1522 * index_state, dropping any unmerged entries. Returns true is1523 * the index is unmerged. Callers who want to refuse to work1524 * from an unmerged state can call this and check its return value,1525 * instead of calling read_cache().1526 */1527int read_index_unmerged(struct index_state *istate)1528{1529 int i;1530 int unmerged = 0;15311532 read_index(istate);1533 for (i = 0; i < istate->cache_nr; i++) {1534 struct cache_entry *ce = istate->cache[i];1535 struct cache_entry *new_ce;1536 int size, len;15371538 if (!ce_stage(ce))1539 continue;1540 unmerged = 1;1541 len = strlen(ce->name);1542 size = cache_entry_size(len);1543 new_ce = xcalloc(1, size);1544 hashcpy(new_ce->sha1, ce->sha1);1545 memcpy(new_ce->name, ce->name, len);1546 new_ce->ce_flags = create_ce_flags(len, 0);1547 new_ce->ce_mode = ce->ce_mode;1548 if (add_index_entry(istate, new_ce, 0))1549 return error("%s: cannot drop to stage #0",1550 ce->name);1551 i = index_name_pos(istate, new_ce->name, len);1552 }1553 return unmerged;1554}15551556struct update_callback_data1557{1558 int flags;1559 int add_errors;1560};15611562static void update_callback(struct diff_queue_struct *q,1563 struct diff_options *opt, void *cbdata)1564{1565 int i;1566 struct update_callback_data *data = cbdata;15671568 for (i = 0; i < q->nr; i++) {1569 struct diff_filepair *p = q->queue[i];1570 const char *path = p->one->path;1571 switch (p->status) {1572 default:1573 die("unexpected diff status %c", p->status);1574 case DIFF_STATUS_UNMERGED:1575 case DIFF_STATUS_MODIFIED:1576 case DIFF_STATUS_TYPE_CHANGED:1577 if (add_file_to_index(&the_index, path, data->flags)) {1578 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))1579 die("updating files failed");1580 data->add_errors++;1581 }1582 break;1583 case DIFF_STATUS_DELETED:1584 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)1585 break;1586 if (!(data->flags & ADD_CACHE_PRETEND))1587 remove_file_from_index(&the_index, path);1588 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))1589 printf("remove '%s'\n", path);1590 break;1591 }1592 }1593}15941595int add_files_to_cache(const char *prefix, const char **pathspec, int flags)1596{1597 struct update_callback_data data;1598 struct rev_info rev;1599 init_revisions(&rev, prefix);1600 setup_revisions(0, NULL, &rev, NULL);1601 rev.prune_data = pathspec;1602 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;1603 rev.diffopt.format_callback = update_callback;1604 data.flags = flags;1605 data.add_errors = 0;1606 rev.diffopt.format_callback_data = &data;1607 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);1608 return !!data.add_errors;1609}16101611/*1612 * Returns 1 if the path is an "other" path with respect to1613 * the index; that is, the path is not mentioned in the index at all,1614 * either as a file, a directory with some files in the index,1615 * or as an unmerged entry.1616 *1617 * We helpfully remove a trailing "/" from directories so that1618 * the output of read_directory can be used as-is.1619 */1620int index_name_is_other(const struct index_state *istate, const char *name,1621 int namelen)1622{1623 int pos;1624 if (namelen && name[namelen - 1] == '/')1625 namelen--;1626 pos = index_name_pos(istate, name, namelen);1627 if (0 <= pos)1628 return 0; /* exact match */1629 pos = -pos - 1;1630 if (pos < istate->cache_nr) {1631 struct cache_entry *ce = istate->cache[pos];1632 if (ce_namelen(ce) == namelen &&1633 !memcmp(ce->name, name, namelen))1634 return 0; /* Yup, this one exists unmerged */1635 }1636 return 1;1637}