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 void *buffer; 103 unsigned long size; 104 enum object_type type; 105 struct strbuf sb = STRBUF_INIT; 106 107 if (strbuf_readlink(&sb, ce->name, expected_size)) 108 return -1; 109 110 buffer = read_sha1_file(ce->sha1, &type, &size); 111 if (buffer) { 112 if (size == sb.len) 113 match = memcmp(buffer, sb.buf, size); 114 free(buffer); 115 } 116 strbuf_release(&sb); 117 return match; 118} 119 120static int ce_compare_gitlink(struct cache_entry *ce) 121{ 122 unsigned char sha1[20]; 123 124 /* 125 * We don't actually require that the .git directory 126 * under GITLINK directory be a valid git directory. It 127 * might even be missing (in case nobody populated that 128 * sub-project). 129 * 130 * If so, we consider it always to match. 131 */ 132 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0) 133 return 0; 134 return hashcmp(sha1, ce->sha1); 135} 136 137static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st) 138{ 139 switch (st->st_mode & S_IFMT) { 140 case S_IFREG: 141 if (ce_compare_data(ce, st)) 142 return DATA_CHANGED; 143 break; 144 case S_IFLNK: 145 if (ce_compare_link(ce, xsize_t(st->st_size))) 146 return DATA_CHANGED; 147 break; 148 case S_IFDIR: 149 if (S_ISGITLINK(ce->ce_mode)) 150 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0; 151 default: 152 return TYPE_CHANGED; 153 } 154 return 0; 155} 156 157int is_empty_blob_sha1(const unsigned char *sha1) 158{ 159 static const unsigned char empty_blob_sha1[20] = { 160 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b, 161 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91 162 }; 163 164 return !hashcmp(sha1, empty_blob_sha1); 165} 166 167static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st) 168{ 169 unsigned int changed = 0; 170 171 if (ce->ce_flags & CE_REMOVE) 172 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; 173 174 switch (ce->ce_mode & S_IFMT) { 175 case S_IFREG: 176 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0; 177 /* We consider only the owner x bit to be relevant for 178 * "mode changes" 179 */ 180 if (trust_executable_bit && 181 (0100 & (ce->ce_mode ^ st->st_mode))) 182 changed |= MODE_CHANGED; 183 break; 184 case S_IFLNK: 185 if (!S_ISLNK(st->st_mode) && 186 (has_symlinks || !S_ISREG(st->st_mode))) 187 changed |= TYPE_CHANGED; 188 break; 189 case S_IFGITLINK: 190 /* We ignore most of the st_xxx fields for gitlinks */ 191 if (!S_ISDIR(st->st_mode)) 192 changed |= TYPE_CHANGED; 193 else if (ce_compare_gitlink(ce)) 194 changed |= DATA_CHANGED; 195 return changed; 196 default: 197 die("internal error: ce_mode is %o", ce->ce_mode); 198 } 199 if (ce->ce_mtime != (unsigned int) st->st_mtime) 200 changed |= MTIME_CHANGED; 201 if (trust_ctime && ce->ce_ctime != (unsigned int) st->st_ctime) 202 changed |= CTIME_CHANGED; 203 204 if (ce->ce_uid != (unsigned int) st->st_uid || 205 ce->ce_gid != (unsigned int) st->st_gid) 206 changed |= OWNER_CHANGED; 207 if (ce->ce_ino != (unsigned int) st->st_ino) 208 changed |= INODE_CHANGED; 209 210#ifdef USE_STDEV 211 /* 212 * st_dev breaks on network filesystems where different 213 * clients will have different views of what "device" 214 * the filesystem is on 215 */ 216 if (ce->ce_dev != (unsigned int) st->st_dev) 217 changed |= INODE_CHANGED; 218#endif 219 220 if (ce->ce_size != (unsigned int) st->st_size) 221 changed |= DATA_CHANGED; 222 223 /* Racily smudged entry? */ 224 if (!ce->ce_size) { 225 if (!is_empty_blob_sha1(ce->sha1)) 226 changed |= DATA_CHANGED; 227 } 228 229 return changed; 230} 231 232static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce) 233{ 234 return (!S_ISGITLINK(ce->ce_mode) && 235 istate->timestamp && 236 ((unsigned int)istate->timestamp) <= ce->ce_mtime); 237} 238 239int ie_match_stat(const struct index_state *istate, 240 struct cache_entry *ce, struct stat *st, 241 unsigned int options) 242{ 243 unsigned int changed; 244 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 245 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 246 247 /* 248 * If it's marked as always valid in the index, it's 249 * valid whatever the checked-out copy says. 250 */ 251 if (!ignore_valid && (ce->ce_flags & CE_VALID)) 252 return 0; 253 254 /* 255 * Intent-to-add entries have not been added, so the index entry 256 * by definition never matches what is in the work tree until it 257 * actually gets added. 258 */ 259 if (ce->ce_flags & CE_INTENT_TO_ADD) 260 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED; 261 262 changed = ce_match_stat_basic(ce, st); 263 264 /* 265 * Within 1 second of this sequence: 266 * echo xyzzy >file && git-update-index --add file 267 * running this command: 268 * echo frotz >file 269 * would give a falsely clean cache entry. The mtime and 270 * length match the cache, and other stat fields do not change. 271 * 272 * We could detect this at update-index time (the cache entry 273 * being registered/updated records the same time as "now") 274 * and delay the return from git-update-index, but that would 275 * effectively mean we can make at most one commit per second, 276 * which is not acceptable. Instead, we check cache entries 277 * whose mtime are the same as the index file timestamp more 278 * carefully than others. 279 */ 280 if (!changed && is_racy_timestamp(istate, ce)) { 281 if (assume_racy_is_modified) 282 changed |= DATA_CHANGED; 283 else 284 changed |= ce_modified_check_fs(ce, st); 285 } 286 287 return changed; 288} 289 290int ie_modified(const struct index_state *istate, 291 struct cache_entry *ce, struct stat *st, unsigned int options) 292{ 293 int changed, changed_fs; 294 295 changed = ie_match_stat(istate, ce, st, options); 296 if (!changed) 297 return 0; 298 /* 299 * If the mode or type has changed, there's no point in trying 300 * to refresh the entry - it's not going to match 301 */ 302 if (changed & (MODE_CHANGED | TYPE_CHANGED)) 303 return changed; 304 305 /* 306 * Immediately after read-tree or update-index --cacheinfo, 307 * the length field is zero, as we have never even read the 308 * lstat(2) information once, and we cannot trust DATA_CHANGED 309 * returned by ie_match_stat() which in turn was returned by 310 * ce_match_stat_basic() to signal that the filesize of the 311 * blob changed. We have to actually go to the filesystem to 312 * see if the contents match, and if so, should answer "unchanged". 313 * 314 * The logic does not apply to gitlinks, as ce_match_stat_basic() 315 * already has checked the actual HEAD from the filesystem in the 316 * subproject. If ie_match_stat() already said it is different, 317 * then we know it is. 318 */ 319 if ((changed & DATA_CHANGED) && 320 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0)) 321 return changed; 322 323 changed_fs = ce_modified_check_fs(ce, st); 324 if (changed_fs) 325 return changed | changed_fs; 326 return 0; 327} 328 329int base_name_compare(const char *name1, int len1, int mode1, 330 const char *name2, int len2, int mode2) 331{ 332 unsigned char c1, c2; 333 int len = len1 < len2 ? len1 : len2; 334 int cmp; 335 336 cmp = memcmp(name1, name2, len); 337 if (cmp) 338 return cmp; 339 c1 = name1[len]; 340 c2 = name2[len]; 341 if (!c1 && S_ISDIR(mode1)) 342 c1 = '/'; 343 if (!c2 && S_ISDIR(mode2)) 344 c2 = '/'; 345 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0; 346} 347 348/* 349 * df_name_compare() is identical to base_name_compare(), except it 350 * compares conflicting directory/file entries as equal. Note that 351 * while a directory name compares as equal to a regular file, they 352 * then individually compare _differently_ to a filename that has 353 * a dot after the basename (because '\0' < '.' < '/'). 354 * 355 * This is used by routines that want to traverse the git namespace 356 * but then handle conflicting entries together when possible. 357 */ 358int df_name_compare(const char *name1, int len1, int mode1, 359 const char *name2, int len2, int mode2) 360{ 361 int len = len1 < len2 ? len1 : len2, cmp; 362 unsigned char c1, c2; 363 364 cmp = memcmp(name1, name2, len); 365 if (cmp) 366 return cmp; 367 /* Directories and files compare equal (same length, same name) */ 368 if (len1 == len2) 369 return 0; 370 c1 = name1[len]; 371 if (!c1 && S_ISDIR(mode1)) 372 c1 = '/'; 373 c2 = name2[len]; 374 if (!c2 && S_ISDIR(mode2)) 375 c2 = '/'; 376 if (c1 == '/' && !c2) 377 return 0; 378 if (c2 == '/' && !c1) 379 return 0; 380 return c1 - c2; 381} 382 383int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2) 384{ 385 int len1 = flags1 & CE_NAMEMASK; 386 int len2 = flags2 & CE_NAMEMASK; 387 int len = len1 < len2 ? len1 : len2; 388 int cmp; 389 390 cmp = memcmp(name1, name2, len); 391 if (cmp) 392 return cmp; 393 if (len1 < len2) 394 return -1; 395 if (len1 > len2) 396 return 1; 397 398 /* Compare stages */ 399 flags1 &= CE_STAGEMASK; 400 flags2 &= CE_STAGEMASK; 401 402 if (flags1 < flags2) 403 return -1; 404 if (flags1 > flags2) 405 return 1; 406 return 0; 407} 408 409int index_name_pos(const struct index_state *istate, const char *name, int namelen) 410{ 411 int first, last; 412 413 first = 0; 414 last = istate->cache_nr; 415 while (last > first) { 416 int next = (last + first) >> 1; 417 struct cache_entry *ce = istate->cache[next]; 418 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags); 419 if (!cmp) 420 return next; 421 if (cmp < 0) { 422 last = next; 423 continue; 424 } 425 first = next+1; 426 } 427 return -first-1; 428} 429 430/* Remove entry, return true if there are more entries to go.. */ 431int remove_index_entry_at(struct index_state *istate, int pos) 432{ 433 struct cache_entry *ce = istate->cache[pos]; 434 435 remove_name_hash(ce); 436 istate->cache_changed = 1; 437 istate->cache_nr--; 438 if (pos >= istate->cache_nr) 439 return 0; 440 memmove(istate->cache + pos, 441 istate->cache + pos + 1, 442 (istate->cache_nr - pos) * sizeof(struct cache_entry *)); 443 return 1; 444} 445 446int remove_file_from_index(struct index_state *istate, const char *path) 447{ 448 int pos = index_name_pos(istate, path, strlen(path)); 449 if (pos < 0) 450 pos = -pos-1; 451 cache_tree_invalidate_path(istate->cache_tree, path); 452 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 453 remove_index_entry_at(istate, pos); 454 return 0; 455} 456 457static int compare_name(struct cache_entry *ce, const char *path, int namelen) 458{ 459 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); 460} 461 462static int index_name_pos_also_unmerged(struct index_state *istate, 463 const char *path, int namelen) 464{ 465 int pos = index_name_pos(istate, path, namelen); 466 struct cache_entry *ce; 467 468 if (pos >= 0) 469 return pos; 470 471 /* maybe unmerged? */ 472 pos = -1 - pos; 473 if (pos >= istate->cache_nr || 474 compare_name((ce = istate->cache[pos]), path, namelen)) 475 return -1; 476 477 /* order of preference: stage 2, 1, 3 */ 478 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && 479 ce_stage((ce = istate->cache[pos + 1])) == 2 && 480 !compare_name(ce, path, namelen)) 481 pos++; 482 return pos; 483} 484 485static int different_name(struct cache_entry *ce, struct cache_entry *alias) 486{ 487 int len = ce_namelen(ce); 488 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len); 489} 490 491/* 492 * If we add a filename that aliases in the cache, we will use the 493 * name that we already have - but we don't want to update the same 494 * alias twice, because that implies that there were actually two 495 * different files with aliasing names! 496 * 497 * So we use the CE_ADDED flag to verify that the alias was an old 498 * one before we accept it as 499 */ 500static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias) 501{ 502 int len; 503 struct cache_entry *new; 504 505 if (alias->ce_flags & CE_ADDED) 506 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); 507 508 /* Ok, create the new entry using the name of the existing alias */ 509 len = ce_namelen(alias); 510 new = xcalloc(1, cache_entry_size(len)); 511 memcpy(new->name, alias->name, len); 512 copy_cache_entry(new, ce); 513 free(ce); 514 return new; 515} 516 517static void record_intent_to_add(struct cache_entry *ce) 518{ 519 unsigned char sha1[20]; 520 if (write_sha1_file("", 0, blob_type, sha1)) 521 die("cannot create an empty blob in the object database"); 522 hashcpy(ce->sha1, sha1); 523} 524 525int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags) 526{ 527 int size, namelen, was_same; 528 mode_t st_mode = st->st_mode; 529 struct cache_entry *ce, *alias; 530 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY; 531 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); 532 int pretend = flags & ADD_CACHE_PRETEND; 533 int intent_only = flags & ADD_CACHE_INTENT; 534 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| 535 (intent_only ? ADD_CACHE_NEW_ONLY : 0)); 536 537 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 538 return error("%s: can only add regular files, symbolic links or git-directories", path); 539 540 namelen = strlen(path); 541 if (S_ISDIR(st_mode)) { 542 while (namelen && path[namelen-1] == '/') 543 namelen--; 544 } 545 size = cache_entry_size(namelen); 546 ce = xcalloc(1, size); 547 memcpy(ce->name, path, namelen); 548 ce->ce_flags = namelen; 549 if (!intent_only) 550 fill_stat_cache_info(ce, st); 551 else 552 ce->ce_flags |= CE_INTENT_TO_ADD; 553 554 if (trust_executable_bit && has_symlinks) 555 ce->ce_mode = create_ce_mode(st_mode); 556 else { 557 /* If there is an existing entry, pick the mode bits and type 558 * from it, otherwise assume unexecutable regular file. 559 */ 560 struct cache_entry *ent; 561 int pos = index_name_pos_also_unmerged(istate, path, namelen); 562 563 ent = (0 <= pos) ? istate->cache[pos] : NULL; 564 ce->ce_mode = ce_mode_from_stat(ent, st_mode); 565 } 566 567 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case); 568 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) { 569 /* Nothing changed, really */ 570 free(ce); 571 ce_mark_uptodate(alias); 572 alias->ce_flags |= CE_ADDED; 573 return 0; 574 } 575 if (!intent_only) { 576 if (index_path(ce->sha1, path, st, 1)) 577 return error("unable to index file %s", path); 578 } else 579 record_intent_to_add(ce); 580 581 if (ignore_case && alias && different_name(ce, alias)) 582 ce = create_alias_ce(ce, alias); 583 ce->ce_flags |= CE_ADDED; 584 585 /* It was suspected to be racily clean, but it turns out to be Ok */ 586 was_same = (alias && 587 !ce_stage(alias) && 588 !hashcmp(alias->sha1, ce->sha1) && 589 ce->ce_mode == alias->ce_mode); 590 591 if (pretend) 592 ; 593 else if (add_index_entry(istate, ce, add_option)) 594 return error("unable to add %s to index",path); 595 if (verbose && !was_same) 596 printf("add '%s'\n", path); 597 return 0; 598} 599 600int add_file_to_index(struct index_state *istate, const char *path, int flags) 601{ 602 struct stat st; 603 if (lstat(path, &st)) 604 die("%s: unable to stat (%s)", path, strerror(errno)); 605 return add_to_index(istate, path, &st, flags); 606} 607 608struct cache_entry *make_cache_entry(unsigned int mode, 609 const unsigned char *sha1, const char *path, int stage, 610 int refresh) 611{ 612 int size, len; 613 struct cache_entry *ce; 614 615 if (!verify_path(path)) { 616 error("Invalid path '%s'", path); 617 return NULL; 618 } 619 620 len = strlen(path); 621 size = cache_entry_size(len); 622 ce = xcalloc(1, size); 623 624 hashcpy(ce->sha1, sha1); 625 memcpy(ce->name, path, len); 626 ce->ce_flags = create_ce_flags(len, stage); 627 ce->ce_mode = create_ce_mode(mode); 628 629 if (refresh) 630 return refresh_cache_entry(ce, 0); 631 632 return ce; 633} 634 635int ce_same_name(struct cache_entry *a, struct cache_entry *b) 636{ 637 int len = ce_namelen(a); 638 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 639} 640 641int ce_path_match(const struct cache_entry *ce, const char **pathspec) 642{ 643 const char *match, *name; 644 int len; 645 646 if (!pathspec) 647 return 1; 648 649 len = ce_namelen(ce); 650 name = ce->name; 651 while ((match = *pathspec++) != NULL) { 652 int matchlen = strlen(match); 653 if (matchlen > len) 654 continue; 655 if (memcmp(name, match, matchlen)) 656 continue; 657 if (matchlen && name[matchlen-1] == '/') 658 return 1; 659 if (name[matchlen] == '/' || !name[matchlen]) 660 return 1; 661 if (!matchlen) 662 return 1; 663 } 664 return 0; 665} 666 667/* 668 * We fundamentally don't like some paths: we don't want 669 * dot or dot-dot anywhere, and for obvious reasons don't 670 * want to recurse into ".git" either. 671 * 672 * Also, we don't want double slashes or slashes at the 673 * end that can make pathnames ambiguous. 674 */ 675static int verify_dotfile(const char *rest) 676{ 677 /* 678 * The first character was '.', but that 679 * has already been discarded, we now test 680 * the rest. 681 */ 682 switch (*rest) { 683 /* "." is not allowed */ 684 case '\0': case '/': 685 return 0; 686 687 /* 688 * ".git" followed by NUL or slash is bad. This 689 * shares the path end test with the ".." case. 690 */ 691 case 'g': 692 if (rest[1] != 'i') 693 break; 694 if (rest[2] != 't') 695 break; 696 rest += 2; 697 /* fallthrough */ 698 case '.': 699 if (rest[1] == '\0' || rest[1] == '/') 700 return 0; 701 } 702 return 1; 703} 704 705int verify_path(const char *path) 706{ 707 char c; 708 709 goto inside; 710 for (;;) { 711 if (!c) 712 return 1; 713 if (c == '/') { 714inside: 715 c = *path++; 716 switch (c) { 717 default: 718 continue; 719 case '/': case '\0': 720 break; 721 case '.': 722 if (verify_dotfile(path)) 723 continue; 724 } 725 return 0; 726 } 727 c = *path++; 728 } 729} 730 731/* 732 * Do we have another file that has the beginning components being a 733 * proper superset of the name we're trying to add? 734 */ 735static int has_file_name(struct index_state *istate, 736 const struct cache_entry *ce, int pos, int ok_to_replace) 737{ 738 int retval = 0; 739 int len = ce_namelen(ce); 740 int stage = ce_stage(ce); 741 const char *name = ce->name; 742 743 while (pos < istate->cache_nr) { 744 struct cache_entry *p = istate->cache[pos++]; 745 746 if (len >= ce_namelen(p)) 747 break; 748 if (memcmp(name, p->name, len)) 749 break; 750 if (ce_stage(p) != stage) 751 continue; 752 if (p->name[len] != '/') 753 continue; 754 if (p->ce_flags & CE_REMOVE) 755 continue; 756 retval = -1; 757 if (!ok_to_replace) 758 break; 759 remove_index_entry_at(istate, --pos); 760 } 761 return retval; 762} 763 764/* 765 * Do we have another file with a pathname that is a proper 766 * subset of the name we're trying to add? 767 */ 768static int has_dir_name(struct index_state *istate, 769 const struct cache_entry *ce, int pos, int ok_to_replace) 770{ 771 int retval = 0; 772 int stage = ce_stage(ce); 773 const char *name = ce->name; 774 const char *slash = name + ce_namelen(ce); 775 776 for (;;) { 777 int len; 778 779 for (;;) { 780 if (*--slash == '/') 781 break; 782 if (slash <= ce->name) 783 return retval; 784 } 785 len = slash - name; 786 787 pos = index_name_pos(istate, name, create_ce_flags(len, stage)); 788 if (pos >= 0) { 789 /* 790 * Found one, but not so fast. This could 791 * be a marker that says "I was here, but 792 * I am being removed". Such an entry is 793 * not a part of the resulting tree, and 794 * it is Ok to have a directory at the same 795 * path. 796 */ 797 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) { 798 retval = -1; 799 if (!ok_to_replace) 800 break; 801 remove_index_entry_at(istate, pos); 802 continue; 803 } 804 } 805 else 806 pos = -pos-1; 807 808 /* 809 * Trivial optimization: if we find an entry that 810 * already matches the sub-directory, then we know 811 * we're ok, and we can exit. 812 */ 813 while (pos < istate->cache_nr) { 814 struct cache_entry *p = istate->cache[pos]; 815 if ((ce_namelen(p) <= len) || 816 (p->name[len] != '/') || 817 memcmp(p->name, name, len)) 818 break; /* not our subdirectory */ 819 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE)) 820 /* 821 * p is at the same stage as our entry, and 822 * is a subdirectory of what we are looking 823 * at, so we cannot have conflicts at our 824 * level or anything shorter. 825 */ 826 return retval; 827 pos++; 828 } 829 } 830 return retval; 831} 832 833/* We may be in a situation where we already have path/file and path 834 * is being added, or we already have path and path/file is being 835 * added. Either one would result in a nonsense tree that has path 836 * twice when git-write-tree tries to write it out. Prevent it. 837 * 838 * If ok-to-replace is specified, we remove the conflicting entries 839 * from the cache so the caller should recompute the insert position. 840 * When this happens, we return non-zero. 841 */ 842static int check_file_directory_conflict(struct index_state *istate, 843 const struct cache_entry *ce, 844 int pos, int ok_to_replace) 845{ 846 int retval; 847 848 /* 849 * When ce is an "I am going away" entry, we allow it to be added 850 */ 851 if (ce->ce_flags & CE_REMOVE) 852 return 0; 853 854 /* 855 * We check if the path is a sub-path of a subsequent pathname 856 * first, since removing those will not change the position 857 * in the array. 858 */ 859 retval = has_file_name(istate, ce, pos, ok_to_replace); 860 861 /* 862 * Then check if the path might have a clashing sub-directory 863 * before it. 864 */ 865 return retval + has_dir_name(istate, ce, pos, ok_to_replace); 866} 867 868static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option) 869{ 870 int pos; 871 int ok_to_add = option & ADD_CACHE_OK_TO_ADD; 872 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE; 873 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK; 874 int new_only = option & ADD_CACHE_NEW_ONLY; 875 876 cache_tree_invalidate_path(istate->cache_tree, ce->name); 877 pos = index_name_pos(istate, ce->name, ce->ce_flags); 878 879 /* existing match? Just replace it. */ 880 if (pos >= 0) { 881 if (!new_only) 882 replace_index_entry(istate, pos, ce); 883 return 0; 884 } 885 pos = -pos-1; 886 887 /* 888 * Inserting a merged entry ("stage 0") into the index 889 * will always replace all non-merged entries.. 890 */ 891 if (pos < istate->cache_nr && ce_stage(ce) == 0) { 892 while (ce_same_name(istate->cache[pos], ce)) { 893 ok_to_add = 1; 894 if (!remove_index_entry_at(istate, pos)) 895 break; 896 } 897 } 898 899 if (!ok_to_add) 900 return -1; 901 if (!verify_path(ce->name)) 902 return error("Invalid path '%s'", ce->name); 903 904 if (!skip_df_check && 905 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) { 906 if (!ok_to_replace) 907 return error("'%s' appears as both a file and as a directory", 908 ce->name); 909 pos = index_name_pos(istate, ce->name, ce->ce_flags); 910 pos = -pos-1; 911 } 912 return pos + 1; 913} 914 915int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option) 916{ 917 int pos; 918 919 if (option & ADD_CACHE_JUST_APPEND) 920 pos = istate->cache_nr; 921 else { 922 int ret; 923 ret = add_index_entry_with_check(istate, ce, option); 924 if (ret <= 0) 925 return ret; 926 pos = ret - 1; 927 } 928 929 /* Make sure the array is big enough .. */ 930 if (istate->cache_nr == istate->cache_alloc) { 931 istate->cache_alloc = alloc_nr(istate->cache_alloc); 932 istate->cache = xrealloc(istate->cache, 933 istate->cache_alloc * sizeof(struct cache_entry *)); 934 } 935 936 /* Add it in.. */ 937 istate->cache_nr++; 938 if (istate->cache_nr > pos + 1) 939 memmove(istate->cache + pos + 1, 940 istate->cache + pos, 941 (istate->cache_nr - pos - 1) * sizeof(ce)); 942 set_index_entry(istate, pos, ce); 943 istate->cache_changed = 1; 944 return 0; 945} 946 947/* 948 * "refresh" does not calculate a new sha1 file or bring the 949 * cache up-to-date for mode/content changes. But what it 950 * _does_ do is to "re-match" the stat information of a file 951 * with the cache, so that you can refresh the cache for a 952 * file that hasn't been changed but where the stat entry is 953 * out of date. 954 * 955 * For example, you'd want to do this after doing a "git-read-tree", 956 * to link up the stat cache details with the proper files. 957 */ 958static struct cache_entry *refresh_cache_ent(struct index_state *istate, 959 struct cache_entry *ce, 960 unsigned int options, int *err) 961{ 962 struct stat st; 963 struct cache_entry *updated; 964 int changed, size; 965 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 966 967 if (ce_uptodate(ce)) 968 return ce; 969 970 /* 971 * CE_VALID means the user promised us that the change to 972 * the work tree does not matter and told us not to worry. 973 */ 974 if (!ignore_valid && (ce->ce_flags & CE_VALID)) { 975 ce_mark_uptodate(ce); 976 return ce; 977 } 978 979 if (lstat(ce->name, &st) < 0) { 980 if (err) 981 *err = errno; 982 return NULL; 983 } 984 985 changed = ie_match_stat(istate, ce, &st, options); 986 if (!changed) { 987 /* 988 * The path is unchanged. If we were told to ignore 989 * valid bit, then we did the actual stat check and 990 * found that the entry is unmodified. If the entry 991 * is not marked VALID, this is the place to mark it 992 * valid again, under "assume unchanged" mode. 993 */ 994 if (ignore_valid && assume_unchanged && 995 !(ce->ce_flags & CE_VALID)) 996 ; /* mark this one VALID again */ 997 else { 998 /* 999 * We do not mark the index itself "modified"1000 * because CE_UPTODATE flag is in-core only;1001 * we are not going to write this change out.1002 */1003 ce_mark_uptodate(ce);1004 return ce;1005 }1006 }10071008 if (ie_modified(istate, ce, &st, options)) {1009 if (err)1010 *err = EINVAL;1011 return NULL;1012 }10131014 size = ce_size(ce);1015 updated = xmalloc(size);1016 memcpy(updated, ce, size);1017 fill_stat_cache_info(updated, &st);1018 /*1019 * If ignore_valid is not set, we should leave CE_VALID bit1020 * alone. Otherwise, paths marked with --no-assume-unchanged1021 * (i.e. things to be edited) will reacquire CE_VALID bit1022 * automatically, which is not really what we want.1023 */1024 if (!ignore_valid && assume_unchanged &&1025 !(ce->ce_flags & CE_VALID))1026 updated->ce_flags &= ~CE_VALID;10271028 return updated;1029}10301031int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)1032{1033 int i;1034 int has_errors = 0;1035 int really = (flags & REFRESH_REALLY) != 0;1036 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;1037 int quiet = (flags & REFRESH_QUIET) != 0;1038 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;1039 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;1040 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;1041 const char *needs_update_message;10421043 needs_update_message = ((flags & REFRESH_SAY_CHANGED)1044 ? "locally modified" : "needs update");1045 for (i = 0; i < istate->cache_nr; i++) {1046 struct cache_entry *ce, *new;1047 int cache_errno = 0;10481049 ce = istate->cache[i];1050 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))1051 continue;10521053 if (ce_stage(ce)) {1054 while ((i < istate->cache_nr) &&1055 ! strcmp(istate->cache[i]->name, ce->name))1056 i++;1057 i--;1058 if (allow_unmerged)1059 continue;1060 printf("%s: needs merge\n", ce->name);1061 has_errors = 1;1062 continue;1063 }10641065 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))1066 continue;10671068 new = refresh_cache_ent(istate, ce, options, &cache_errno);1069 if (new == ce)1070 continue;1071 if (!new) {1072 if (not_new && cache_errno == ENOENT)1073 continue;1074 if (really && cache_errno == EINVAL) {1075 /* If we are doing --really-refresh that1076 * means the index is not valid anymore.1077 */1078 ce->ce_flags &= ~CE_VALID;1079 istate->cache_changed = 1;1080 }1081 if (quiet)1082 continue;1083 printf("%s: %s\n", ce->name, needs_update_message);1084 has_errors = 1;1085 continue;1086 }10871088 replace_index_entry(istate, i, new);1089 }1090 return has_errors;1091}10921093struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)1094{1095 return refresh_cache_ent(&the_index, ce, really, NULL);1096}10971098static int verify_hdr(struct cache_header *hdr, unsigned long size)1099{1100 git_SHA_CTX c;1101 unsigned char sha1[20];11021103 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))1104 return error("bad signature");1105 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))1106 return error("bad index version");1107 git_SHA1_Init(&c);1108 git_SHA1_Update(&c, hdr, size - 20);1109 git_SHA1_Final(sha1, &c);1110 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))1111 return error("bad index file sha1 signature");1112 return 0;1113}11141115static int read_index_extension(struct index_state *istate,1116 const char *ext, void *data, unsigned long sz)1117{1118 switch (CACHE_EXT(ext)) {1119 case CACHE_EXT_TREE:1120 istate->cache_tree = cache_tree_read(data, sz);1121 break;1122 default:1123 if (*ext < 'A' || 'Z' < *ext)1124 return error("index uses %.4s extension, which we do not understand",1125 ext);1126 fprintf(stderr, "ignoring %.4s extension\n", ext);1127 break;1128 }1129 return 0;1130}11311132int read_index(struct index_state *istate)1133{1134 return read_index_from(istate, get_index_file());1135}11361137static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)1138{1139 size_t len;1140 const char *name;11411142 ce->ce_ctime = ntohl(ondisk->ctime.sec);1143 ce->ce_mtime = ntohl(ondisk->mtime.sec);1144 ce->ce_dev = ntohl(ondisk->dev);1145 ce->ce_ino = ntohl(ondisk->ino);1146 ce->ce_mode = ntohl(ondisk->mode);1147 ce->ce_uid = ntohl(ondisk->uid);1148 ce->ce_gid = ntohl(ondisk->gid);1149 ce->ce_size = ntohl(ondisk->size);1150 /* On-disk flags are just 16 bits */1151 ce->ce_flags = ntohs(ondisk->flags);11521153 hashcpy(ce->sha1, ondisk->sha1);11541155 len = ce->ce_flags & CE_NAMEMASK;11561157 if (ce->ce_flags & CE_EXTENDED) {1158 struct ondisk_cache_entry_extended *ondisk2;1159 int extended_flags;1160 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1161 extended_flags = ntohs(ondisk2->flags2) << 16;1162 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1163 if (extended_flags & ~CE_EXTENDED_FLAGS)1164 die("Unknown index entry format %08x", extended_flags);1165 ce->ce_flags |= extended_flags;1166 name = ondisk2->name;1167 }1168 else1169 name = ondisk->name;11701171 if (len == CE_NAMEMASK)1172 len = strlen(name);1173 /*1174 * NEEDSWORK: If the original index is crafted, this copy could1175 * go unchecked.1176 */1177 memcpy(ce->name, name, len + 1);1178}11791180static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)1181{1182 long per_entry;11831184 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);11851186 /*1187 * Alignment can cause differences. This should be "alignof", but1188 * since that's a gcc'ism, just use the size of a pointer.1189 */1190 per_entry += sizeof(void *);1191 return ondisk_size + entries*per_entry;1192}11931194/* remember to discard_cache() before reading a different cache! */1195int read_index_from(struct index_state *istate, const char *path)1196{1197 int fd, i;1198 struct stat st;1199 unsigned long src_offset, dst_offset;1200 struct cache_header *hdr;1201 void *mmap;1202 size_t mmap_size;12031204 errno = EBUSY;1205 if (istate->initialized)1206 return istate->cache_nr;12071208 errno = ENOENT;1209 istate->timestamp = 0;1210 fd = open(path, O_RDONLY);1211 if (fd < 0) {1212 if (errno == ENOENT)1213 return 0;1214 die("index file open failed (%s)", strerror(errno));1215 }12161217 if (fstat(fd, &st))1218 die("cannot stat the open index (%s)", strerror(errno));12191220 errno = EINVAL;1221 mmap_size = xsize_t(st.st_size);1222 if (mmap_size < sizeof(struct cache_header) + 20)1223 die("index file smaller than expected");12241225 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);1226 close(fd);1227 if (mmap == MAP_FAILED)1228 die("unable to map index file");12291230 hdr = mmap;1231 if (verify_hdr(hdr, mmap_size) < 0)1232 goto unmap;12331234 istate->cache_nr = ntohl(hdr->hdr_entries);1235 istate->cache_alloc = alloc_nr(istate->cache_nr);1236 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));12371238 /*1239 * The disk format is actually larger than the in-memory format,1240 * due to space for nsec etc, so even though the in-memory one1241 * has room for a few more flags, we can allocate using the same1242 * index size1243 */1244 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));1245 istate->initialized = 1;12461247 src_offset = sizeof(*hdr);1248 dst_offset = 0;1249 for (i = 0; i < istate->cache_nr; i++) {1250 struct ondisk_cache_entry *disk_ce;1251 struct cache_entry *ce;12521253 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);1254 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);1255 convert_from_disk(disk_ce, ce);1256 set_index_entry(istate, i, ce);12571258 src_offset += ondisk_ce_size(ce);1259 dst_offset += ce_size(ce);1260 }1261 istate->timestamp = st.st_mtime;1262 while (src_offset <= mmap_size - 20 - 8) {1263 /* After an array of active_nr index entries,1264 * there can be arbitrary number of extended1265 * sections, each of which is prefixed with1266 * extension name (4-byte) and section length1267 * in 4-byte network byte order.1268 */1269 unsigned long extsize;1270 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);1271 extsize = ntohl(extsize);1272 if (read_index_extension(istate,1273 (const char *) mmap + src_offset,1274 (char *) mmap + src_offset + 8,1275 extsize) < 0)1276 goto unmap;1277 src_offset += 8;1278 src_offset += extsize;1279 }1280 munmap(mmap, mmap_size);1281 return istate->cache_nr;12821283unmap:1284 munmap(mmap, mmap_size);1285 errno = EINVAL;1286 die("index file corrupt");1287}12881289int is_index_unborn(struct index_state *istate)1290{1291 return (!istate->cache_nr && !istate->alloc && !istate->timestamp);1292}12931294int discard_index(struct index_state *istate)1295{1296 istate->cache_nr = 0;1297 istate->cache_changed = 0;1298 istate->timestamp = 0;1299 istate->name_hash_initialized = 0;1300 free_hash(&istate->name_hash);1301 cache_tree_free(&(istate->cache_tree));1302 free(istate->alloc);1303 istate->alloc = NULL;1304 istate->initialized = 0;13051306 /* no need to throw away allocated active_cache */1307 return 0;1308}13091310int unmerged_index(const struct index_state *istate)1311{1312 int i;1313 for (i = 0; i < istate->cache_nr; i++) {1314 if (ce_stage(istate->cache[i]))1315 return 1;1316 }1317 return 0;1318}13191320#define WRITE_BUFFER_SIZE 81921321static unsigned char write_buffer[WRITE_BUFFER_SIZE];1322static unsigned long write_buffer_len;13231324static int ce_write_flush(git_SHA_CTX *context, int fd)1325{1326 unsigned int buffered = write_buffer_len;1327 if (buffered) {1328 git_SHA1_Update(context, write_buffer, buffered);1329 if (write_in_full(fd, write_buffer, buffered) != buffered)1330 return -1;1331 write_buffer_len = 0;1332 }1333 return 0;1334}13351336static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)1337{1338 while (len) {1339 unsigned int buffered = write_buffer_len;1340 unsigned int partial = WRITE_BUFFER_SIZE - buffered;1341 if (partial > len)1342 partial = len;1343 memcpy(write_buffer + buffered, data, partial);1344 buffered += partial;1345 if (buffered == WRITE_BUFFER_SIZE) {1346 write_buffer_len = buffered;1347 if (ce_write_flush(context, fd))1348 return -1;1349 buffered = 0;1350 }1351 write_buffer_len = buffered;1352 len -= partial;1353 data = (char *) data + partial;1354 }1355 return 0;1356}13571358static int write_index_ext_header(git_SHA_CTX *context, int fd,1359 unsigned int ext, unsigned int sz)1360{1361 ext = htonl(ext);1362 sz = htonl(sz);1363 return ((ce_write(context, fd, &ext, 4) < 0) ||1364 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;1365}13661367static int ce_flush(git_SHA_CTX *context, int fd)1368{1369 unsigned int left = write_buffer_len;13701371 if (left) {1372 write_buffer_len = 0;1373 git_SHA1_Update(context, write_buffer, left);1374 }13751376 /* Flush first if not enough space for SHA1 signature */1377 if (left + 20 > WRITE_BUFFER_SIZE) {1378 if (write_in_full(fd, write_buffer, left) != left)1379 return -1;1380 left = 0;1381 }13821383 /* Append the SHA1 signature at the end */1384 git_SHA1_Final(write_buffer + left, context);1385 left += 20;1386 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;1387}13881389static void ce_smudge_racily_clean_entry(struct cache_entry *ce)1390{1391 /*1392 * The only thing we care about in this function is to smudge the1393 * falsely clean entry due to touch-update-touch race, so we leave1394 * everything else as they are. We are called for entries whose1395 * ce_mtime match the index file mtime.1396 *1397 * Note that this actually does not do much for gitlinks, for1398 * which ce_match_stat_basic() always goes to the actual1399 * contents. The caller checks with is_racy_timestamp() which1400 * always says "no" for gitlinks, so we are not called for them ;-)1401 */1402 struct stat st;14031404 if (lstat(ce->name, &st) < 0)1405 return;1406 if (ce_match_stat_basic(ce, &st))1407 return;1408 if (ce_modified_check_fs(ce, &st)) {1409 /* This is "racily clean"; smudge it. Note that this1410 * is a tricky code. At first glance, it may appear1411 * that it can break with this sequence:1412 *1413 * $ echo xyzzy >frotz1414 * $ git-update-index --add frotz1415 * $ : >frotz1416 * $ sleep 31417 * $ echo filfre >nitfol1418 * $ git-update-index --add nitfol1419 *1420 * but it does not. When the second update-index runs,1421 * it notices that the entry "frotz" has the same timestamp1422 * as index, and if we were to smudge it by resetting its1423 * size to zero here, then the object name recorded1424 * in index is the 6-byte file but the cached stat information1425 * becomes zero --- which would then match what we would1426 * obtain from the filesystem next time we stat("frotz").1427 *1428 * However, the second update-index, before calling1429 * this function, notices that the cached size is 61430 * bytes and what is on the filesystem is an empty1431 * file, and never calls us, so the cached size information1432 * for "frotz" stays 6 which does not match the filesystem.1433 */1434 ce->ce_size = 0;1435 }1436}14371438static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)1439{1440 int size = ondisk_ce_size(ce);1441 struct ondisk_cache_entry *ondisk = xcalloc(1, size);1442 char *name;14431444 ondisk->ctime.sec = htonl(ce->ce_ctime);1445 ondisk->ctime.nsec = 0;1446 ondisk->mtime.sec = htonl(ce->ce_mtime);1447 ondisk->mtime.nsec = 0;1448 ondisk->dev = htonl(ce->ce_dev);1449 ondisk->ino = htonl(ce->ce_ino);1450 ondisk->mode = htonl(ce->ce_mode);1451 ondisk->uid = htonl(ce->ce_uid);1452 ondisk->gid = htonl(ce->ce_gid);1453 ondisk->size = htonl(ce->ce_size);1454 hashcpy(ondisk->sha1, ce->sha1);1455 ondisk->flags = htons(ce->ce_flags);1456 if (ce->ce_flags & CE_EXTENDED) {1457 struct ondisk_cache_entry_extended *ondisk2;1458 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1459 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);1460 name = ondisk2->name;1461 }1462 else1463 name = ondisk->name;1464 memcpy(name, ce->name, ce_namelen(ce));14651466 return ce_write(c, fd, ondisk, size);1467}14681469int write_index(const struct index_state *istate, int newfd)1470{1471 git_SHA_CTX c;1472 struct cache_header hdr;1473 int i, err, removed, extended;1474 struct cache_entry **cache = istate->cache;1475 int entries = istate->cache_nr;14761477 for (i = removed = extended = 0; i < entries; i++) {1478 if (cache[i]->ce_flags & CE_REMOVE)1479 removed++;14801481 /* reduce extended entries if possible */1482 cache[i]->ce_flags &= ~CE_EXTENDED;1483 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {1484 extended++;1485 cache[i]->ce_flags |= CE_EXTENDED;1486 }1487 }14881489 hdr.hdr_signature = htonl(CACHE_SIGNATURE);1490 /* for extended format, increase version so older git won't try to read it */1491 hdr.hdr_version = htonl(extended ? 3 : 2);1492 hdr.hdr_entries = htonl(entries - removed);14931494 git_SHA1_Init(&c);1495 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)1496 return -1;14971498 for (i = 0; i < entries; i++) {1499 struct cache_entry *ce = cache[i];1500 if (ce->ce_flags & CE_REMOVE)1501 continue;1502 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))1503 ce_smudge_racily_clean_entry(ce);1504 if (ce_write_entry(&c, newfd, ce) < 0)1505 return -1;1506 }15071508 /* Write extension data here */1509 if (istate->cache_tree) {1510 struct strbuf sb = STRBUF_INIT;15111512 cache_tree_write(&sb, istate->cache_tree);1513 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 01514 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1515 strbuf_release(&sb);1516 if (err)1517 return -1;1518 }1519 return ce_flush(&c, newfd);1520}15211522/*1523 * Read the index file that is potentially unmerged into given1524 * index_state, dropping any unmerged entries. Returns true if1525 * the index is unmerged. Callers who want to refuse to work1526 * from an unmerged state can call this and check its return value,1527 * instead of calling read_cache().1528 */1529int read_index_unmerged(struct index_state *istate)1530{1531 int i;1532 int unmerged = 0;15331534 read_index(istate);1535 for (i = 0; i < istate->cache_nr; i++) {1536 struct cache_entry *ce = istate->cache[i];1537 struct cache_entry *new_ce;1538 int size, len;15391540 if (!ce_stage(ce))1541 continue;1542 unmerged = 1;1543 len = strlen(ce->name);1544 size = cache_entry_size(len);1545 new_ce = xcalloc(1, size);1546 hashcpy(new_ce->sha1, ce->sha1);1547 memcpy(new_ce->name, ce->name, len);1548 new_ce->ce_flags = create_ce_flags(len, 0);1549 new_ce->ce_mode = ce->ce_mode;1550 if (add_index_entry(istate, new_ce, 0))1551 return error("%s: cannot drop to stage #0",1552 ce->name);1553 i = index_name_pos(istate, new_ce->name, len);1554 }1555 return unmerged;1556}15571558struct update_callback_data1559{1560 int flags;1561 int add_errors;1562};15631564static void update_callback(struct diff_queue_struct *q,1565 struct diff_options *opt, void *cbdata)1566{1567 int i;1568 struct update_callback_data *data = cbdata;15691570 for (i = 0; i < q->nr; i++) {1571 struct diff_filepair *p = q->queue[i];1572 const char *path = p->one->path;1573 switch (p->status) {1574 default:1575 die("unexpected diff status %c", p->status);1576 case DIFF_STATUS_UNMERGED:1577 case DIFF_STATUS_MODIFIED:1578 case DIFF_STATUS_TYPE_CHANGED:1579 if (add_file_to_index(&the_index, path, data->flags)) {1580 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))1581 die("updating files failed");1582 data->add_errors++;1583 }1584 break;1585 case DIFF_STATUS_DELETED:1586 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)1587 break;1588 if (!(data->flags & ADD_CACHE_PRETEND))1589 remove_file_from_index(&the_index, path);1590 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))1591 printf("remove '%s'\n", path);1592 break;1593 }1594 }1595}15961597int add_files_to_cache(const char *prefix, const char **pathspec, int flags)1598{1599 struct update_callback_data data;1600 struct rev_info rev;1601 init_revisions(&rev, prefix);1602 setup_revisions(0, NULL, &rev, NULL);1603 rev.prune_data = pathspec;1604 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;1605 rev.diffopt.format_callback = update_callback;1606 data.flags = flags;1607 data.add_errors = 0;1608 rev.diffopt.format_callback_data = &data;1609 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);1610 return !!data.add_errors;1611}16121613/*1614 * Returns 1 if the path is an "other" path with respect to1615 * the index; that is, the path is not mentioned in the index at all,1616 * either as a file, a directory with some files in the index,1617 * or as an unmerged entry.1618 *1619 * We helpfully remove a trailing "/" from directories so that1620 * the output of read_directory can be used as-is.1621 */1622int index_name_is_other(const struct index_state *istate, const char *name,1623 int namelen)1624{1625 int pos;1626 if (namelen && name[namelen - 1] == '/')1627 namelen--;1628 pos = index_name_pos(istate, name, namelen);1629 if (0 <= pos)1630 return 0; /* exact match */1631 pos = -pos - 1;1632 if (pos < istate->cache_nr) {1633 struct cache_entry *ce = istate->cache[pos];1634 if (ce_namelen(ce) == namelen &&1635 !memcmp(ce->name, name, namelen))1636 return 0; /* Yup, this one exists unmerged */1637 }1638 return 1;1639}