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} 445 446/* 447 * Remove all cache ententries marked for removal, that is where 448 * CE_REMOVE is set in ce_flags. This is much more effective than 449 * calling remove_index_entry_at() for each entry to be removed. 450 */ 451void remove_marked_cache_entries(struct index_state *istate) 452{ 453 struct cache_entry **ce_array = istate->cache; 454 unsigned int i, j; 455 456 for (i = j = 0; i < istate->cache_nr; i++) { 457 if (ce_array[i]->ce_flags & CE_REMOVE) 458 remove_name_hash(ce_array[i]); 459 else 460 ce_array[j++] = ce_array[i]; 461 } 462 istate->cache_changed = 1; 463 istate->cache_nr = j; 464} 465 466int remove_file_from_index(struct index_state *istate, const char *path) 467{ 468 int pos = index_name_pos(istate, path, strlen(path)); 469 if (pos < 0) 470 pos = -pos-1; 471 cache_tree_invalidate_path(istate->cache_tree, path); 472 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 473 remove_index_entry_at(istate, pos); 474 return 0; 475} 476 477static int compare_name(struct cache_entry *ce, const char *path, int namelen) 478{ 479 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); 480} 481 482static int index_name_pos_also_unmerged(struct index_state *istate, 483 const char *path, int namelen) 484{ 485 int pos = index_name_pos(istate, path, namelen); 486 struct cache_entry *ce; 487 488 if (pos >= 0) 489 return pos; 490 491 /* maybe unmerged? */ 492 pos = -1 - pos; 493 if (pos >= istate->cache_nr || 494 compare_name((ce = istate->cache[pos]), path, namelen)) 495 return -1; 496 497 /* order of preference: stage 2, 1, 3 */ 498 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && 499 ce_stage((ce = istate->cache[pos + 1])) == 2 && 500 !compare_name(ce, path, namelen)) 501 pos++; 502 return pos; 503} 504 505static int different_name(struct cache_entry *ce, struct cache_entry *alias) 506{ 507 int len = ce_namelen(ce); 508 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len); 509} 510 511/* 512 * If we add a filename that aliases in the cache, we will use the 513 * name that we already have - but we don't want to update the same 514 * alias twice, because that implies that there were actually two 515 * different files with aliasing names! 516 * 517 * So we use the CE_ADDED flag to verify that the alias was an old 518 * one before we accept it as 519 */ 520static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias) 521{ 522 int len; 523 struct cache_entry *new; 524 525 if (alias->ce_flags & CE_ADDED) 526 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); 527 528 /* Ok, create the new entry using the name of the existing alias */ 529 len = ce_namelen(alias); 530 new = xcalloc(1, cache_entry_size(len)); 531 memcpy(new->name, alias->name, len); 532 copy_cache_entry(new, ce); 533 free(ce); 534 return new; 535} 536 537static void record_intent_to_add(struct cache_entry *ce) 538{ 539 unsigned char sha1[20]; 540 if (write_sha1_file("", 0, blob_type, sha1)) 541 die("cannot create an empty blob in the object database"); 542 hashcpy(ce->sha1, sha1); 543} 544 545int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags) 546{ 547 int size, namelen, was_same; 548 mode_t st_mode = st->st_mode; 549 struct cache_entry *ce, *alias; 550 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY; 551 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); 552 int pretend = flags & ADD_CACHE_PRETEND; 553 int intent_only = flags & ADD_CACHE_INTENT; 554 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| 555 (intent_only ? ADD_CACHE_NEW_ONLY : 0)); 556 557 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 558 return error("%s: can only add regular files, symbolic links or git-directories", path); 559 560 namelen = strlen(path); 561 if (S_ISDIR(st_mode)) { 562 while (namelen && path[namelen-1] == '/') 563 namelen--; 564 } 565 size = cache_entry_size(namelen); 566 ce = xcalloc(1, size); 567 memcpy(ce->name, path, namelen); 568 ce->ce_flags = namelen; 569 if (!intent_only) 570 fill_stat_cache_info(ce, st); 571 else 572 ce->ce_flags |= CE_INTENT_TO_ADD; 573 574 if (trust_executable_bit && has_symlinks) 575 ce->ce_mode = create_ce_mode(st_mode); 576 else { 577 /* If there is an existing entry, pick the mode bits and type 578 * from it, otherwise assume unexecutable regular file. 579 */ 580 struct cache_entry *ent; 581 int pos = index_name_pos_also_unmerged(istate, path, namelen); 582 583 ent = (0 <= pos) ? istate->cache[pos] : NULL; 584 ce->ce_mode = ce_mode_from_stat(ent, st_mode); 585 } 586 587 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case); 588 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) { 589 /* Nothing changed, really */ 590 free(ce); 591 ce_mark_uptodate(alias); 592 alias->ce_flags |= CE_ADDED; 593 return 0; 594 } 595 if (!intent_only) { 596 if (index_path(ce->sha1, path, st, 1)) 597 return error("unable to index file %s", path); 598 } else 599 record_intent_to_add(ce); 600 601 if (ignore_case && alias && different_name(ce, alias)) 602 ce = create_alias_ce(ce, alias); 603 ce->ce_flags |= CE_ADDED; 604 605 /* It was suspected to be racily clean, but it turns out to be Ok */ 606 was_same = (alias && 607 !ce_stage(alias) && 608 !hashcmp(alias->sha1, ce->sha1) && 609 ce->ce_mode == alias->ce_mode); 610 611 if (pretend) 612 ; 613 else if (add_index_entry(istate, ce, add_option)) 614 return error("unable to add %s to index",path); 615 if (verbose && !was_same) 616 printf("add '%s'\n", path); 617 return 0; 618} 619 620int add_file_to_index(struct index_state *istate, const char *path, int flags) 621{ 622 struct stat st; 623 if (lstat(path, &st)) 624 die("%s: unable to stat (%s)", path, strerror(errno)); 625 return add_to_index(istate, path, &st, flags); 626} 627 628struct cache_entry *make_cache_entry(unsigned int mode, 629 const unsigned char *sha1, const char *path, int stage, 630 int refresh) 631{ 632 int size, len; 633 struct cache_entry *ce; 634 635 if (!verify_path(path)) { 636 error("Invalid path '%s'", path); 637 return NULL; 638 } 639 640 len = strlen(path); 641 size = cache_entry_size(len); 642 ce = xcalloc(1, size); 643 644 hashcpy(ce->sha1, sha1); 645 memcpy(ce->name, path, len); 646 ce->ce_flags = create_ce_flags(len, stage); 647 ce->ce_mode = create_ce_mode(mode); 648 649 if (refresh) 650 return refresh_cache_entry(ce, 0); 651 652 return ce; 653} 654 655int ce_same_name(struct cache_entry *a, struct cache_entry *b) 656{ 657 int len = ce_namelen(a); 658 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 659} 660 661int ce_path_match(const struct cache_entry *ce, const char **pathspec) 662{ 663 const char *match, *name; 664 int len; 665 666 if (!pathspec) 667 return 1; 668 669 len = ce_namelen(ce); 670 name = ce->name; 671 while ((match = *pathspec++) != NULL) { 672 int matchlen = strlen(match); 673 if (matchlen > len) 674 continue; 675 if (memcmp(name, match, matchlen)) 676 continue; 677 if (matchlen && name[matchlen-1] == '/') 678 return 1; 679 if (name[matchlen] == '/' || !name[matchlen]) 680 return 1; 681 if (!matchlen) 682 return 1; 683 } 684 return 0; 685} 686 687/* 688 * We fundamentally don't like some paths: we don't want 689 * dot or dot-dot anywhere, and for obvious reasons don't 690 * want to recurse into ".git" either. 691 * 692 * Also, we don't want double slashes or slashes at the 693 * end that can make pathnames ambiguous. 694 */ 695static int verify_dotfile(const char *rest) 696{ 697 /* 698 * The first character was '.', but that 699 * has already been discarded, we now test 700 * the rest. 701 */ 702 switch (*rest) { 703 /* "." is not allowed */ 704 case '\0': case '/': 705 return 0; 706 707 /* 708 * ".git" followed by NUL or slash is bad. This 709 * shares the path end test with the ".." case. 710 */ 711 case 'g': 712 if (rest[1] != 'i') 713 break; 714 if (rest[2] != 't') 715 break; 716 rest += 2; 717 /* fallthrough */ 718 case '.': 719 if (rest[1] == '\0' || rest[1] == '/') 720 return 0; 721 } 722 return 1; 723} 724 725int verify_path(const char *path) 726{ 727 char c; 728 729 goto inside; 730 for (;;) { 731 if (!c) 732 return 1; 733 if (c == '/') { 734inside: 735 c = *path++; 736 switch (c) { 737 default: 738 continue; 739 case '/': case '\0': 740 break; 741 case '.': 742 if (verify_dotfile(path)) 743 continue; 744 } 745 return 0; 746 } 747 c = *path++; 748 } 749} 750 751/* 752 * Do we have another file that has the beginning components being a 753 * proper superset of the name we're trying to add? 754 */ 755static int has_file_name(struct index_state *istate, 756 const struct cache_entry *ce, int pos, int ok_to_replace) 757{ 758 int retval = 0; 759 int len = ce_namelen(ce); 760 int stage = ce_stage(ce); 761 const char *name = ce->name; 762 763 while (pos < istate->cache_nr) { 764 struct cache_entry *p = istate->cache[pos++]; 765 766 if (len >= ce_namelen(p)) 767 break; 768 if (memcmp(name, p->name, len)) 769 break; 770 if (ce_stage(p) != stage) 771 continue; 772 if (p->name[len] != '/') 773 continue; 774 if (p->ce_flags & CE_REMOVE) 775 continue; 776 retval = -1; 777 if (!ok_to_replace) 778 break; 779 remove_index_entry_at(istate, --pos); 780 } 781 return retval; 782} 783 784/* 785 * Do we have another file with a pathname that is a proper 786 * subset of the name we're trying to add? 787 */ 788static int has_dir_name(struct index_state *istate, 789 const struct cache_entry *ce, int pos, int ok_to_replace) 790{ 791 int retval = 0; 792 int stage = ce_stage(ce); 793 const char *name = ce->name; 794 const char *slash = name + ce_namelen(ce); 795 796 for (;;) { 797 int len; 798 799 for (;;) { 800 if (*--slash == '/') 801 break; 802 if (slash <= ce->name) 803 return retval; 804 } 805 len = slash - name; 806 807 pos = index_name_pos(istate, name, create_ce_flags(len, stage)); 808 if (pos >= 0) { 809 /* 810 * Found one, but not so fast. This could 811 * be a marker that says "I was here, but 812 * I am being removed". Such an entry is 813 * not a part of the resulting tree, and 814 * it is Ok to have a directory at the same 815 * path. 816 */ 817 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) { 818 retval = -1; 819 if (!ok_to_replace) 820 break; 821 remove_index_entry_at(istate, pos); 822 continue; 823 } 824 } 825 else 826 pos = -pos-1; 827 828 /* 829 * Trivial optimization: if we find an entry that 830 * already matches the sub-directory, then we know 831 * we're ok, and we can exit. 832 */ 833 while (pos < istate->cache_nr) { 834 struct cache_entry *p = istate->cache[pos]; 835 if ((ce_namelen(p) <= len) || 836 (p->name[len] != '/') || 837 memcmp(p->name, name, len)) 838 break; /* not our subdirectory */ 839 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE)) 840 /* 841 * p is at the same stage as our entry, and 842 * is a subdirectory of what we are looking 843 * at, so we cannot have conflicts at our 844 * level or anything shorter. 845 */ 846 return retval; 847 pos++; 848 } 849 } 850 return retval; 851} 852 853/* We may be in a situation where we already have path/file and path 854 * is being added, or we already have path and path/file is being 855 * added. Either one would result in a nonsense tree that has path 856 * twice when git-write-tree tries to write it out. Prevent it. 857 * 858 * If ok-to-replace is specified, we remove the conflicting entries 859 * from the cache so the caller should recompute the insert position. 860 * When this happens, we return non-zero. 861 */ 862static int check_file_directory_conflict(struct index_state *istate, 863 const struct cache_entry *ce, 864 int pos, int ok_to_replace) 865{ 866 int retval; 867 868 /* 869 * When ce is an "I am going away" entry, we allow it to be added 870 */ 871 if (ce->ce_flags & CE_REMOVE) 872 return 0; 873 874 /* 875 * We check if the path is a sub-path of a subsequent pathname 876 * first, since removing those will not change the position 877 * in the array. 878 */ 879 retval = has_file_name(istate, ce, pos, ok_to_replace); 880 881 /* 882 * Then check if the path might have a clashing sub-directory 883 * before it. 884 */ 885 return retval + has_dir_name(istate, ce, pos, ok_to_replace); 886} 887 888static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option) 889{ 890 int pos; 891 int ok_to_add = option & ADD_CACHE_OK_TO_ADD; 892 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE; 893 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK; 894 int new_only = option & ADD_CACHE_NEW_ONLY; 895 896 cache_tree_invalidate_path(istate->cache_tree, ce->name); 897 pos = index_name_pos(istate, ce->name, ce->ce_flags); 898 899 /* existing match? Just replace it. */ 900 if (pos >= 0) { 901 if (!new_only) 902 replace_index_entry(istate, pos, ce); 903 return 0; 904 } 905 pos = -pos-1; 906 907 /* 908 * Inserting a merged entry ("stage 0") into the index 909 * will always replace all non-merged entries.. 910 */ 911 if (pos < istate->cache_nr && ce_stage(ce) == 0) { 912 while (ce_same_name(istate->cache[pos], ce)) { 913 ok_to_add = 1; 914 if (!remove_index_entry_at(istate, pos)) 915 break; 916 } 917 } 918 919 if (!ok_to_add) 920 return -1; 921 if (!verify_path(ce->name)) 922 return error("Invalid path '%s'", ce->name); 923 924 if (!skip_df_check && 925 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) { 926 if (!ok_to_replace) 927 return error("'%s' appears as both a file and as a directory", 928 ce->name); 929 pos = index_name_pos(istate, ce->name, ce->ce_flags); 930 pos = -pos-1; 931 } 932 return pos + 1; 933} 934 935int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option) 936{ 937 int pos; 938 939 if (option & ADD_CACHE_JUST_APPEND) 940 pos = istate->cache_nr; 941 else { 942 int ret; 943 ret = add_index_entry_with_check(istate, ce, option); 944 if (ret <= 0) 945 return ret; 946 pos = ret - 1; 947 } 948 949 /* Make sure the array is big enough .. */ 950 if (istate->cache_nr == istate->cache_alloc) { 951 istate->cache_alloc = alloc_nr(istate->cache_alloc); 952 istate->cache = xrealloc(istate->cache, 953 istate->cache_alloc * sizeof(struct cache_entry *)); 954 } 955 956 /* Add it in.. */ 957 istate->cache_nr++; 958 if (istate->cache_nr > pos + 1) 959 memmove(istate->cache + pos + 1, 960 istate->cache + pos, 961 (istate->cache_nr - pos - 1) * sizeof(ce)); 962 set_index_entry(istate, pos, ce); 963 istate->cache_changed = 1; 964 return 0; 965} 966 967/* 968 * "refresh" does not calculate a new sha1 file or bring the 969 * cache up-to-date for mode/content changes. But what it 970 * _does_ do is to "re-match" the stat information of a file 971 * with the cache, so that you can refresh the cache for a 972 * file that hasn't been changed but where the stat entry is 973 * out of date. 974 * 975 * For example, you'd want to do this after doing a "git-read-tree", 976 * to link up the stat cache details with the proper files. 977 */ 978static struct cache_entry *refresh_cache_ent(struct index_state *istate, 979 struct cache_entry *ce, 980 unsigned int options, int *err) 981{ 982 struct stat st; 983 struct cache_entry *updated; 984 int changed, size; 985 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 986 987 if (ce_uptodate(ce)) 988 return ce; 989 990 /* 991 * CE_VALID means the user promised us that the change to 992 * the work tree does not matter and told us not to worry. 993 */ 994 if (!ignore_valid && (ce->ce_flags & CE_VALID)) { 995 ce_mark_uptodate(ce); 996 return ce; 997 } 998 999 if (lstat(ce->name, &st) < 0) {1000 if (err)1001 *err = errno;1002 return NULL;1003 }10041005 changed = ie_match_stat(istate, ce, &st, options);1006 if (!changed) {1007 /*1008 * The path is unchanged. If we were told to ignore1009 * valid bit, then we did the actual stat check and1010 * found that the entry is unmodified. If the entry1011 * is not marked VALID, this is the place to mark it1012 * valid again, under "assume unchanged" mode.1013 */1014 if (ignore_valid && assume_unchanged &&1015 !(ce->ce_flags & CE_VALID))1016 ; /* mark this one VALID again */1017 else {1018 /*1019 * We do not mark the index itself "modified"1020 * because CE_UPTODATE flag is in-core only;1021 * we are not going to write this change out.1022 */1023 ce_mark_uptodate(ce);1024 return ce;1025 }1026 }10271028 if (ie_modified(istate, ce, &st, options)) {1029 if (err)1030 *err = EINVAL;1031 return NULL;1032 }10331034 size = ce_size(ce);1035 updated = xmalloc(size);1036 memcpy(updated, ce, size);1037 fill_stat_cache_info(updated, &st);1038 /*1039 * If ignore_valid is not set, we should leave CE_VALID bit1040 * alone. Otherwise, paths marked with --no-assume-unchanged1041 * (i.e. things to be edited) will reacquire CE_VALID bit1042 * automatically, which is not really what we want.1043 */1044 if (!ignore_valid && assume_unchanged &&1045 !(ce->ce_flags & CE_VALID))1046 updated->ce_flags &= ~CE_VALID;10471048 return updated;1049}10501051int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)1052{1053 int i;1054 int has_errors = 0;1055 int really = (flags & REFRESH_REALLY) != 0;1056 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;1057 int quiet = (flags & REFRESH_QUIET) != 0;1058 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;1059 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;1060 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;1061 const char *needs_update_message;10621063 needs_update_message = ((flags & REFRESH_SAY_CHANGED)1064 ? "locally modified" : "needs update");1065 for (i = 0; i < istate->cache_nr; i++) {1066 struct cache_entry *ce, *new;1067 int cache_errno = 0;10681069 ce = istate->cache[i];1070 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))1071 continue;10721073 if (ce_stage(ce)) {1074 while ((i < istate->cache_nr) &&1075 ! strcmp(istate->cache[i]->name, ce->name))1076 i++;1077 i--;1078 if (allow_unmerged)1079 continue;1080 printf("%s: needs merge\n", ce->name);1081 has_errors = 1;1082 continue;1083 }10841085 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))1086 continue;10871088 new = refresh_cache_ent(istate, ce, options, &cache_errno);1089 if (new == ce)1090 continue;1091 if (!new) {1092 if (not_new && cache_errno == ENOENT)1093 continue;1094 if (really && cache_errno == EINVAL) {1095 /* If we are doing --really-refresh that1096 * means the index is not valid anymore.1097 */1098 ce->ce_flags &= ~CE_VALID;1099 istate->cache_changed = 1;1100 }1101 if (quiet)1102 continue;1103 printf("%s: %s\n", ce->name, needs_update_message);1104 has_errors = 1;1105 continue;1106 }11071108 replace_index_entry(istate, i, new);1109 }1110 return has_errors;1111}11121113struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)1114{1115 return refresh_cache_ent(&the_index, ce, really, NULL);1116}11171118static int verify_hdr(struct cache_header *hdr, unsigned long size)1119{1120 git_SHA_CTX c;1121 unsigned char sha1[20];11221123 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))1124 return error("bad signature");1125 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))1126 return error("bad index version");1127 git_SHA1_Init(&c);1128 git_SHA1_Update(&c, hdr, size - 20);1129 git_SHA1_Final(sha1, &c);1130 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))1131 return error("bad index file sha1 signature");1132 return 0;1133}11341135static int read_index_extension(struct index_state *istate,1136 const char *ext, void *data, unsigned long sz)1137{1138 switch (CACHE_EXT(ext)) {1139 case CACHE_EXT_TREE:1140 istate->cache_tree = cache_tree_read(data, sz);1141 break;1142 default:1143 if (*ext < 'A' || 'Z' < *ext)1144 return error("index uses %.4s extension, which we do not understand",1145 ext);1146 fprintf(stderr, "ignoring %.4s extension\n", ext);1147 break;1148 }1149 return 0;1150}11511152int read_index(struct index_state *istate)1153{1154 return read_index_from(istate, get_index_file());1155}11561157static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)1158{1159 size_t len;1160 const char *name;11611162 ce->ce_ctime = ntohl(ondisk->ctime.sec);1163 ce->ce_mtime = ntohl(ondisk->mtime.sec);1164 ce->ce_dev = ntohl(ondisk->dev);1165 ce->ce_ino = ntohl(ondisk->ino);1166 ce->ce_mode = ntohl(ondisk->mode);1167 ce->ce_uid = ntohl(ondisk->uid);1168 ce->ce_gid = ntohl(ondisk->gid);1169 ce->ce_size = ntohl(ondisk->size);1170 /* On-disk flags are just 16 bits */1171 ce->ce_flags = ntohs(ondisk->flags);11721173 hashcpy(ce->sha1, ondisk->sha1);11741175 len = ce->ce_flags & CE_NAMEMASK;11761177 if (ce->ce_flags & CE_EXTENDED) {1178 struct ondisk_cache_entry_extended *ondisk2;1179 int extended_flags;1180 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1181 extended_flags = ntohs(ondisk2->flags2) << 16;1182 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1183 if (extended_flags & ~CE_EXTENDED_FLAGS)1184 die("Unknown index entry format %08x", extended_flags);1185 ce->ce_flags |= extended_flags;1186 name = ondisk2->name;1187 }1188 else1189 name = ondisk->name;11901191 if (len == CE_NAMEMASK)1192 len = strlen(name);1193 /*1194 * NEEDSWORK: If the original index is crafted, this copy could1195 * go unchecked.1196 */1197 memcpy(ce->name, name, len + 1);1198}11991200static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)1201{1202 long per_entry;12031204 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);12051206 /*1207 * Alignment can cause differences. This should be "alignof", but1208 * since that's a gcc'ism, just use the size of a pointer.1209 */1210 per_entry += sizeof(void *);1211 return ondisk_size + entries*per_entry;1212}12131214/* remember to discard_cache() before reading a different cache! */1215int read_index_from(struct index_state *istate, const char *path)1216{1217 int fd, i;1218 struct stat st;1219 unsigned long src_offset, dst_offset;1220 struct cache_header *hdr;1221 void *mmap;1222 size_t mmap_size;12231224 errno = EBUSY;1225 if (istate->initialized)1226 return istate->cache_nr;12271228 errno = ENOENT;1229 istate->timestamp = 0;1230 fd = open(path, O_RDONLY);1231 if (fd < 0) {1232 if (errno == ENOENT)1233 return 0;1234 die("index file open failed (%s)", strerror(errno));1235 }12361237 if (fstat(fd, &st))1238 die("cannot stat the open index (%s)", strerror(errno));12391240 errno = EINVAL;1241 mmap_size = xsize_t(st.st_size);1242 if (mmap_size < sizeof(struct cache_header) + 20)1243 die("index file smaller than expected");12441245 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);1246 close(fd);1247 if (mmap == MAP_FAILED)1248 die("unable to map index file");12491250 hdr = mmap;1251 if (verify_hdr(hdr, mmap_size) < 0)1252 goto unmap;12531254 istate->cache_nr = ntohl(hdr->hdr_entries);1255 istate->cache_alloc = alloc_nr(istate->cache_nr);1256 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));12571258 /*1259 * The disk format is actually larger than the in-memory format,1260 * due to space for nsec etc, so even though the in-memory one1261 * has room for a few more flags, we can allocate using the same1262 * index size1263 */1264 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));1265 istate->initialized = 1;12661267 src_offset = sizeof(*hdr);1268 dst_offset = 0;1269 for (i = 0; i < istate->cache_nr; i++) {1270 struct ondisk_cache_entry *disk_ce;1271 struct cache_entry *ce;12721273 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);1274 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);1275 convert_from_disk(disk_ce, ce);1276 set_index_entry(istate, i, ce);12771278 src_offset += ondisk_ce_size(ce);1279 dst_offset += ce_size(ce);1280 }1281 istate->timestamp = st.st_mtime;1282 while (src_offset <= mmap_size - 20 - 8) {1283 /* After an array of active_nr index entries,1284 * there can be arbitrary number of extended1285 * sections, each of which is prefixed with1286 * extension name (4-byte) and section length1287 * in 4-byte network byte order.1288 */1289 unsigned long extsize;1290 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);1291 extsize = ntohl(extsize);1292 if (read_index_extension(istate,1293 (const char *) mmap + src_offset,1294 (char *) mmap + src_offset + 8,1295 extsize) < 0)1296 goto unmap;1297 src_offset += 8;1298 src_offset += extsize;1299 }1300 munmap(mmap, mmap_size);1301 return istate->cache_nr;13021303unmap:1304 munmap(mmap, mmap_size);1305 errno = EINVAL;1306 die("index file corrupt");1307}13081309int is_index_unborn(struct index_state *istate)1310{1311 return (!istate->cache_nr && !istate->alloc && !istate->timestamp);1312}13131314int discard_index(struct index_state *istate)1315{1316 istate->cache_nr = 0;1317 istate->cache_changed = 0;1318 istate->timestamp = 0;1319 istate->name_hash_initialized = 0;1320 free_hash(&istate->name_hash);1321 cache_tree_free(&(istate->cache_tree));1322 free(istate->alloc);1323 istate->alloc = NULL;1324 istate->initialized = 0;13251326 /* no need to throw away allocated active_cache */1327 return 0;1328}13291330int unmerged_index(const struct index_state *istate)1331{1332 int i;1333 for (i = 0; i < istate->cache_nr; i++) {1334 if (ce_stage(istate->cache[i]))1335 return 1;1336 }1337 return 0;1338}13391340#define WRITE_BUFFER_SIZE 81921341static unsigned char write_buffer[WRITE_BUFFER_SIZE];1342static unsigned long write_buffer_len;13431344static int ce_write_flush(git_SHA_CTX *context, int fd)1345{1346 unsigned int buffered = write_buffer_len;1347 if (buffered) {1348 git_SHA1_Update(context, write_buffer, buffered);1349 if (write_in_full(fd, write_buffer, buffered) != buffered)1350 return -1;1351 write_buffer_len = 0;1352 }1353 return 0;1354}13551356static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)1357{1358 while (len) {1359 unsigned int buffered = write_buffer_len;1360 unsigned int partial = WRITE_BUFFER_SIZE - buffered;1361 if (partial > len)1362 partial = len;1363 memcpy(write_buffer + buffered, data, partial);1364 buffered += partial;1365 if (buffered == WRITE_BUFFER_SIZE) {1366 write_buffer_len = buffered;1367 if (ce_write_flush(context, fd))1368 return -1;1369 buffered = 0;1370 }1371 write_buffer_len = buffered;1372 len -= partial;1373 data = (char *) data + partial;1374 }1375 return 0;1376}13771378static int write_index_ext_header(git_SHA_CTX *context, int fd,1379 unsigned int ext, unsigned int sz)1380{1381 ext = htonl(ext);1382 sz = htonl(sz);1383 return ((ce_write(context, fd, &ext, 4) < 0) ||1384 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;1385}13861387static int ce_flush(git_SHA_CTX *context, int fd)1388{1389 unsigned int left = write_buffer_len;13901391 if (left) {1392 write_buffer_len = 0;1393 git_SHA1_Update(context, write_buffer, left);1394 }13951396 /* Flush first if not enough space for SHA1 signature */1397 if (left + 20 > WRITE_BUFFER_SIZE) {1398 if (write_in_full(fd, write_buffer, left) != left)1399 return -1;1400 left = 0;1401 }14021403 /* Append the SHA1 signature at the end */1404 git_SHA1_Final(write_buffer + left, context);1405 left += 20;1406 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;1407}14081409static void ce_smudge_racily_clean_entry(struct cache_entry *ce)1410{1411 /*1412 * The only thing we care about in this function is to smudge the1413 * falsely clean entry due to touch-update-touch race, so we leave1414 * everything else as they are. We are called for entries whose1415 * ce_mtime match the index file mtime.1416 *1417 * Note that this actually does not do much for gitlinks, for1418 * which ce_match_stat_basic() always goes to the actual1419 * contents. The caller checks with is_racy_timestamp() which1420 * always says "no" for gitlinks, so we are not called for them ;-)1421 */1422 struct stat st;14231424 if (lstat(ce->name, &st) < 0)1425 return;1426 if (ce_match_stat_basic(ce, &st))1427 return;1428 if (ce_modified_check_fs(ce, &st)) {1429 /* This is "racily clean"; smudge it. Note that this1430 * is a tricky code. At first glance, it may appear1431 * that it can break with this sequence:1432 *1433 * $ echo xyzzy >frotz1434 * $ git-update-index --add frotz1435 * $ : >frotz1436 * $ sleep 31437 * $ echo filfre >nitfol1438 * $ git-update-index --add nitfol1439 *1440 * but it does not. When the second update-index runs,1441 * it notices that the entry "frotz" has the same timestamp1442 * as index, and if we were to smudge it by resetting its1443 * size to zero here, then the object name recorded1444 * in index is the 6-byte file but the cached stat information1445 * becomes zero --- which would then match what we would1446 * obtain from the filesystem next time we stat("frotz").1447 *1448 * However, the second update-index, before calling1449 * this function, notices that the cached size is 61450 * bytes and what is on the filesystem is an empty1451 * file, and never calls us, so the cached size information1452 * for "frotz" stays 6 which does not match the filesystem.1453 */1454 ce->ce_size = 0;1455 }1456}14571458static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)1459{1460 int size = ondisk_ce_size(ce);1461 struct ondisk_cache_entry *ondisk = xcalloc(1, size);1462 char *name;14631464 ondisk->ctime.sec = htonl(ce->ce_ctime);1465 ondisk->ctime.nsec = 0;1466 ondisk->mtime.sec = htonl(ce->ce_mtime);1467 ondisk->mtime.nsec = 0;1468 ondisk->dev = htonl(ce->ce_dev);1469 ondisk->ino = htonl(ce->ce_ino);1470 ondisk->mode = htonl(ce->ce_mode);1471 ondisk->uid = htonl(ce->ce_uid);1472 ondisk->gid = htonl(ce->ce_gid);1473 ondisk->size = htonl(ce->ce_size);1474 hashcpy(ondisk->sha1, ce->sha1);1475 ondisk->flags = htons(ce->ce_flags);1476 if (ce->ce_flags & CE_EXTENDED) {1477 struct ondisk_cache_entry_extended *ondisk2;1478 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1479 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);1480 name = ondisk2->name;1481 }1482 else1483 name = ondisk->name;1484 memcpy(name, ce->name, ce_namelen(ce));14851486 return ce_write(c, fd, ondisk, size);1487}14881489int write_index(const struct index_state *istate, int newfd)1490{1491 git_SHA_CTX c;1492 struct cache_header hdr;1493 int i, err, removed, extended;1494 struct cache_entry **cache = istate->cache;1495 int entries = istate->cache_nr;14961497 for (i = removed = extended = 0; i < entries; i++) {1498 if (cache[i]->ce_flags & CE_REMOVE)1499 removed++;15001501 /* reduce extended entries if possible */1502 cache[i]->ce_flags &= ~CE_EXTENDED;1503 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {1504 extended++;1505 cache[i]->ce_flags |= CE_EXTENDED;1506 }1507 }15081509 hdr.hdr_signature = htonl(CACHE_SIGNATURE);1510 /* for extended format, increase version so older git won't try to read it */1511 hdr.hdr_version = htonl(extended ? 3 : 2);1512 hdr.hdr_entries = htonl(entries - removed);15131514 git_SHA1_Init(&c);1515 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)1516 return -1;15171518 for (i = 0; i < entries; i++) {1519 struct cache_entry *ce = cache[i];1520 if (ce->ce_flags & CE_REMOVE)1521 continue;1522 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))1523 ce_smudge_racily_clean_entry(ce);1524 if (ce_write_entry(&c, newfd, ce) < 0)1525 return -1;1526 }15271528 /* Write extension data here */1529 if (istate->cache_tree) {1530 struct strbuf sb = STRBUF_INIT;15311532 cache_tree_write(&sb, istate->cache_tree);1533 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 01534 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1535 strbuf_release(&sb);1536 if (err)1537 return -1;1538 }1539 return ce_flush(&c, newfd);1540}15411542/*1543 * Read the index file that is potentially unmerged into given1544 * index_state, dropping any unmerged entries. Returns true if1545 * the index is unmerged. Callers who want to refuse to work1546 * from an unmerged state can call this and check its return value,1547 * instead of calling read_cache().1548 */1549int read_index_unmerged(struct index_state *istate)1550{1551 int i;1552 int unmerged = 0;15531554 read_index(istate);1555 for (i = 0; i < istate->cache_nr; i++) {1556 struct cache_entry *ce = istate->cache[i];1557 struct cache_entry *new_ce;1558 int size, len;15591560 if (!ce_stage(ce))1561 continue;1562 unmerged = 1;1563 len = strlen(ce->name);1564 size = cache_entry_size(len);1565 new_ce = xcalloc(1, size);1566 hashcpy(new_ce->sha1, ce->sha1);1567 memcpy(new_ce->name, ce->name, len);1568 new_ce->ce_flags = create_ce_flags(len, 0);1569 new_ce->ce_mode = ce->ce_mode;1570 if (add_index_entry(istate, new_ce, 0))1571 return error("%s: cannot drop to stage #0",1572 ce->name);1573 i = index_name_pos(istate, new_ce->name, len);1574 }1575 return unmerged;1576}15771578struct update_callback_data1579{1580 int flags;1581 int add_errors;1582};15831584static void update_callback(struct diff_queue_struct *q,1585 struct diff_options *opt, void *cbdata)1586{1587 int i;1588 struct update_callback_data *data = cbdata;15891590 for (i = 0; i < q->nr; i++) {1591 struct diff_filepair *p = q->queue[i];1592 const char *path = p->one->path;1593 switch (p->status) {1594 default:1595 die("unexpected diff status %c", p->status);1596 case DIFF_STATUS_UNMERGED:1597 /*1598 * ADD_CACHE_IGNORE_REMOVAL is unset if "git1599 * add -u" is calling us, In such a case, a1600 * missing work tree file needs to be removed1601 * if there is an unmerged entry at stage #2,1602 * but such a diff record is followed by1603 * another with DIFF_STATUS_DELETED (and if1604 * there is no stage #2, we won't see DELETED1605 * nor MODIFIED). We can simply continue1606 * either way.1607 */1608 if (!(data->flags & ADD_CACHE_IGNORE_REMOVAL))1609 continue;1610 /*1611 * Otherwise, it is "git add path" is asking1612 * to explicitly add it; we fall through. A1613 * missing work tree file is an error and is1614 * caught by add_file_to_index() in such a1615 * case.1616 */1617 case DIFF_STATUS_MODIFIED:1618 case DIFF_STATUS_TYPE_CHANGED:1619 if (add_file_to_index(&the_index, path, data->flags)) {1620 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))1621 die("updating files failed");1622 data->add_errors++;1623 }1624 break;1625 case DIFF_STATUS_DELETED:1626 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)1627 break;1628 if (!(data->flags & ADD_CACHE_PRETEND))1629 remove_file_from_index(&the_index, path);1630 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))1631 printf("remove '%s'\n", path);1632 break;1633 }1634 }1635}16361637int add_files_to_cache(const char *prefix, const char **pathspec, int flags)1638{1639 struct update_callback_data data;1640 struct rev_info rev;1641 init_revisions(&rev, prefix);1642 setup_revisions(0, NULL, &rev, NULL);1643 rev.prune_data = pathspec;1644 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;1645 rev.diffopt.format_callback = update_callback;1646 data.flags = flags;1647 data.add_errors = 0;1648 rev.diffopt.format_callback_data = &data;1649 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);1650 return !!data.add_errors;1651}16521653/*1654 * Returns 1 if the path is an "other" path with respect to1655 * the index; that is, the path is not mentioned in the index at all,1656 * either as a file, a directory with some files in the index,1657 * or as an unmerged entry.1658 *1659 * We helpfully remove a trailing "/" from directories so that1660 * the output of read_directory can be used as-is.1661 */1662int index_name_is_other(const struct index_state *istate, const char *name,1663 int namelen)1664{1665 int pos;1666 if (namelen && name[namelen - 1] == '/')1667 namelen--;1668 pos = index_name_pos(istate, name, namelen);1669 if (0 <= pos)1670 return 0; /* exact match */1671 pos = -pos - 1;1672 if (pos < istate->cache_nr) {1673 struct cache_entry *ce = istate->cache[pos];1674 if (ce_namelen(ce) == namelen &&1675 !memcmp(ce->name, name, namelen))1676 return 0; /* Yup, this one exists unmerged */1677 }1678 return 1;1679}