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.sec = (unsigned int)st->st_ctime; 71 ce->ce_mtime.sec = (unsigned int)st->st_mtime; 72#ifdef USE_NSEC 73 ce->ce_ctime.nsec = (unsigned int)st->st_ctim.tv_nsec; 74 ce->ce_mtime.nsec = (unsigned int)st->st_mtim.tv_nsec; 75#else 76 ce->ce_ctime.nsec = 0; 77 ce->ce_mtime.nsec = 0; 78#endif 79 ce->ce_dev = st->st_dev; 80 ce->ce_ino = st->st_ino; 81 ce->ce_uid = st->st_uid; 82 ce->ce_gid = st->st_gid; 83 ce->ce_size = st->st_size; 84 85 if (assume_unchanged) 86 ce->ce_flags |= CE_VALID; 87 88 if (S_ISREG(st->st_mode)) 89 ce_mark_uptodate(ce); 90} 91 92static int ce_compare_data(struct cache_entry *ce, struct stat *st) 93{ 94 int match = -1; 95 int fd = open(ce->name, O_RDONLY); 96 97 if (fd >= 0) { 98 unsigned char sha1[20]; 99 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name)) 100 match = hashcmp(sha1, ce->sha1); 101 /* index_fd() closed the file descriptor already */ 102 } 103 return match; 104} 105 106static int ce_compare_link(struct cache_entry *ce, size_t expected_size) 107{ 108 int match = -1; 109 void *buffer; 110 unsigned long size; 111 enum object_type type; 112 struct strbuf sb = STRBUF_INIT; 113 114 if (strbuf_readlink(&sb, ce->name, expected_size)) 115 return -1; 116 117 buffer = read_sha1_file(ce->sha1, &type, &size); 118 if (buffer) { 119 if (size == sb.len) 120 match = memcmp(buffer, sb.buf, size); 121 free(buffer); 122 } 123 strbuf_release(&sb); 124 return match; 125} 126 127static int ce_compare_gitlink(struct cache_entry *ce) 128{ 129 unsigned char sha1[20]; 130 131 /* 132 * We don't actually require that the .git directory 133 * under GITLINK directory be a valid git directory. It 134 * might even be missing (in case nobody populated that 135 * sub-project). 136 * 137 * If so, we consider it always to match. 138 */ 139 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0) 140 return 0; 141 return hashcmp(sha1, ce->sha1); 142} 143 144static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st) 145{ 146 switch (st->st_mode & S_IFMT) { 147 case S_IFREG: 148 if (ce_compare_data(ce, st)) 149 return DATA_CHANGED; 150 break; 151 case S_IFLNK: 152 if (ce_compare_link(ce, xsize_t(st->st_size))) 153 return DATA_CHANGED; 154 break; 155 case S_IFDIR: 156 if (S_ISGITLINK(ce->ce_mode)) 157 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0; 158 default: 159 return TYPE_CHANGED; 160 } 161 return 0; 162} 163 164int is_empty_blob_sha1(const unsigned char *sha1) 165{ 166 static const unsigned char empty_blob_sha1[20] = { 167 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b, 168 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91 169 }; 170 171 return !hashcmp(sha1, empty_blob_sha1); 172} 173 174static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st) 175{ 176 unsigned int changed = 0; 177 178 if (ce->ce_flags & CE_REMOVE) 179 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; 180 181 switch (ce->ce_mode & S_IFMT) { 182 case S_IFREG: 183 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0; 184 /* We consider only the owner x bit to be relevant for 185 * "mode changes" 186 */ 187 if (trust_executable_bit && 188 (0100 & (ce->ce_mode ^ st->st_mode))) 189 changed |= MODE_CHANGED; 190 break; 191 case S_IFLNK: 192 if (!S_ISLNK(st->st_mode) && 193 (has_symlinks || !S_ISREG(st->st_mode))) 194 changed |= TYPE_CHANGED; 195 break; 196 case S_IFGITLINK: 197 /* We ignore most of the st_xxx fields for gitlinks */ 198 if (!S_ISDIR(st->st_mode)) 199 changed |= TYPE_CHANGED; 200 else if (ce_compare_gitlink(ce)) 201 changed |= DATA_CHANGED; 202 return changed; 203 default: 204 die("internal error: ce_mode is %o", ce->ce_mode); 205 } 206 if (ce->ce_mtime.sec != (unsigned int)st->st_mtime) 207 changed |= MTIME_CHANGED; 208 if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime) 209 changed |= CTIME_CHANGED; 210 211#ifdef USE_NSEC 212 if (ce->ce_mtime.nsec != (unsigned int)st->st_mtim.tv_nsec) 213 changed |= MTIME_CHANGED; 214 if (trust_ctime && ce->ce_ctime.nsec != (unsigned int)st->st_ctim.tv_nsec) 215 changed |= CTIME_CHANGED; 216#endif 217 218 if (ce->ce_uid != (unsigned int) st->st_uid || 219 ce->ce_gid != (unsigned int) st->st_gid) 220 changed |= OWNER_CHANGED; 221 if (ce->ce_ino != (unsigned int) st->st_ino) 222 changed |= INODE_CHANGED; 223 224#ifdef USE_STDEV 225 /* 226 * st_dev breaks on network filesystems where different 227 * clients will have different views of what "device" 228 * the filesystem is on 229 */ 230 if (ce->ce_dev != (unsigned int) st->st_dev) 231 changed |= INODE_CHANGED; 232#endif 233 234 if (ce->ce_size != (unsigned int) st->st_size) 235 changed |= DATA_CHANGED; 236 237 /* Racily smudged entry? */ 238 if (!ce->ce_size) { 239 if (!is_empty_blob_sha1(ce->sha1)) 240 changed |= DATA_CHANGED; 241 } 242 243 return changed; 244} 245 246static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce) 247{ 248 return (!S_ISGITLINK(ce->ce_mode) && 249 istate->timestamp.sec && 250#ifdef USE_NSEC 251 /* nanosecond timestamped files can also be racy! */ 252 (istate->timestamp.sec < ce->ce_mtime.sec || 253 (istate->timestamp.sec == ce->ce_mtime.sec && 254 istate->timestamp.nsec <= ce->ce_mtime.nsec)) 255#else 256 istate->timestamp.sec <= ce->ce_mtime.sec 257#endif 258 ); 259} 260 261int ie_match_stat(const struct index_state *istate, 262 struct cache_entry *ce, struct stat *st, 263 unsigned int options) 264{ 265 unsigned int changed; 266 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 267 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 268 269 /* 270 * If it's marked as always valid in the index, it's 271 * valid whatever the checked-out copy says. 272 */ 273 if (!ignore_valid && (ce->ce_flags & CE_VALID)) 274 return 0; 275 276 /* 277 * Intent-to-add entries have not been added, so the index entry 278 * by definition never matches what is in the work tree until it 279 * actually gets added. 280 */ 281 if (ce->ce_flags & CE_INTENT_TO_ADD) 282 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED; 283 284 changed = ce_match_stat_basic(ce, st); 285 286 /* 287 * Within 1 second of this sequence: 288 * echo xyzzy >file && git-update-index --add file 289 * running this command: 290 * echo frotz >file 291 * would give a falsely clean cache entry. The mtime and 292 * length match the cache, and other stat fields do not change. 293 * 294 * We could detect this at update-index time (the cache entry 295 * being registered/updated records the same time as "now") 296 * and delay the return from git-update-index, but that would 297 * effectively mean we can make at most one commit per second, 298 * which is not acceptable. Instead, we check cache entries 299 * whose mtime are the same as the index file timestamp more 300 * carefully than others. 301 */ 302 if (!changed && is_racy_timestamp(istate, ce)) { 303 if (assume_racy_is_modified) 304 changed |= DATA_CHANGED; 305 else 306 changed |= ce_modified_check_fs(ce, st); 307 } 308 309 return changed; 310} 311 312int ie_modified(const struct index_state *istate, 313 struct cache_entry *ce, struct stat *st, unsigned int options) 314{ 315 int changed, changed_fs; 316 317 changed = ie_match_stat(istate, ce, st, options); 318 if (!changed) 319 return 0; 320 /* 321 * If the mode or type has changed, there's no point in trying 322 * to refresh the entry - it's not going to match 323 */ 324 if (changed & (MODE_CHANGED | TYPE_CHANGED)) 325 return changed; 326 327 /* 328 * Immediately after read-tree or update-index --cacheinfo, 329 * the length field is zero, as we have never even read the 330 * lstat(2) information once, and we cannot trust DATA_CHANGED 331 * returned by ie_match_stat() which in turn was returned by 332 * ce_match_stat_basic() to signal that the filesize of the 333 * blob changed. We have to actually go to the filesystem to 334 * see if the contents match, and if so, should answer "unchanged". 335 * 336 * The logic does not apply to gitlinks, as ce_match_stat_basic() 337 * already has checked the actual HEAD from the filesystem in the 338 * subproject. If ie_match_stat() already said it is different, 339 * then we know it is. 340 */ 341 if ((changed & DATA_CHANGED) && 342 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0)) 343 return changed; 344 345 changed_fs = ce_modified_check_fs(ce, st); 346 if (changed_fs) 347 return changed | changed_fs; 348 return 0; 349} 350 351int base_name_compare(const char *name1, int len1, int mode1, 352 const char *name2, int len2, int mode2) 353{ 354 unsigned char c1, c2; 355 int len = len1 < len2 ? len1 : len2; 356 int cmp; 357 358 cmp = memcmp(name1, name2, len); 359 if (cmp) 360 return cmp; 361 c1 = name1[len]; 362 c2 = name2[len]; 363 if (!c1 && S_ISDIR(mode1)) 364 c1 = '/'; 365 if (!c2 && S_ISDIR(mode2)) 366 c2 = '/'; 367 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0; 368} 369 370/* 371 * df_name_compare() is identical to base_name_compare(), except it 372 * compares conflicting directory/file entries as equal. Note that 373 * while a directory name compares as equal to a regular file, they 374 * then individually compare _differently_ to a filename that has 375 * a dot after the basename (because '\0' < '.' < '/'). 376 * 377 * This is used by routines that want to traverse the git namespace 378 * but then handle conflicting entries together when possible. 379 */ 380int df_name_compare(const char *name1, int len1, int mode1, 381 const char *name2, int len2, int mode2) 382{ 383 int len = len1 < len2 ? len1 : len2, cmp; 384 unsigned char c1, c2; 385 386 cmp = memcmp(name1, name2, len); 387 if (cmp) 388 return cmp; 389 /* Directories and files compare equal (same length, same name) */ 390 if (len1 == len2) 391 return 0; 392 c1 = name1[len]; 393 if (!c1 && S_ISDIR(mode1)) 394 c1 = '/'; 395 c2 = name2[len]; 396 if (!c2 && S_ISDIR(mode2)) 397 c2 = '/'; 398 if (c1 == '/' && !c2) 399 return 0; 400 if (c2 == '/' && !c1) 401 return 0; 402 return c1 - c2; 403} 404 405int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2) 406{ 407 int len1 = flags1 & CE_NAMEMASK; 408 int len2 = flags2 & CE_NAMEMASK; 409 int len = len1 < len2 ? len1 : len2; 410 int cmp; 411 412 cmp = memcmp(name1, name2, len); 413 if (cmp) 414 return cmp; 415 if (len1 < len2) 416 return -1; 417 if (len1 > len2) 418 return 1; 419 420 /* Compare stages */ 421 flags1 &= CE_STAGEMASK; 422 flags2 &= CE_STAGEMASK; 423 424 if (flags1 < flags2) 425 return -1; 426 if (flags1 > flags2) 427 return 1; 428 return 0; 429} 430 431int index_name_pos(const struct index_state *istate, const char *name, int namelen) 432{ 433 int first, last; 434 435 first = 0; 436 last = istate->cache_nr; 437 while (last > first) { 438 int next = (last + first) >> 1; 439 struct cache_entry *ce = istate->cache[next]; 440 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags); 441 if (!cmp) 442 return next; 443 if (cmp < 0) { 444 last = next; 445 continue; 446 } 447 first = next+1; 448 } 449 return -first-1; 450} 451 452/* Remove entry, return true if there are more entries to go.. */ 453int remove_index_entry_at(struct index_state *istate, int pos) 454{ 455 struct cache_entry *ce = istate->cache[pos]; 456 457 remove_name_hash(ce); 458 istate->cache_changed = 1; 459 istate->cache_nr--; 460 if (pos >= istate->cache_nr) 461 return 0; 462 memmove(istate->cache + pos, 463 istate->cache + pos + 1, 464 (istate->cache_nr - pos) * sizeof(struct cache_entry *)); 465 return 1; 466} 467 468/* 469 * Remove all cache ententries marked for removal, that is where 470 * CE_REMOVE is set in ce_flags. This is much more effective than 471 * calling remove_index_entry_at() for each entry to be removed. 472 */ 473void remove_marked_cache_entries(struct index_state *istate) 474{ 475 struct cache_entry **ce_array = istate->cache; 476 unsigned int i, j; 477 478 for (i = j = 0; i < istate->cache_nr; i++) { 479 if (ce_array[i]->ce_flags & CE_REMOVE) 480 remove_name_hash(ce_array[i]); 481 else 482 ce_array[j++] = ce_array[i]; 483 } 484 istate->cache_changed = 1; 485 istate->cache_nr = j; 486} 487 488int remove_file_from_index(struct index_state *istate, const char *path) 489{ 490 int pos = index_name_pos(istate, path, strlen(path)); 491 if (pos < 0) 492 pos = -pos-1; 493 cache_tree_invalidate_path(istate->cache_tree, path); 494 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 495 remove_index_entry_at(istate, pos); 496 return 0; 497} 498 499static int compare_name(struct cache_entry *ce, const char *path, int namelen) 500{ 501 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); 502} 503 504static int index_name_pos_also_unmerged(struct index_state *istate, 505 const char *path, int namelen) 506{ 507 int pos = index_name_pos(istate, path, namelen); 508 struct cache_entry *ce; 509 510 if (pos >= 0) 511 return pos; 512 513 /* maybe unmerged? */ 514 pos = -1 - pos; 515 if (pos >= istate->cache_nr || 516 compare_name((ce = istate->cache[pos]), path, namelen)) 517 return -1; 518 519 /* order of preference: stage 2, 1, 3 */ 520 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && 521 ce_stage((ce = istate->cache[pos + 1])) == 2 && 522 !compare_name(ce, path, namelen)) 523 pos++; 524 return pos; 525} 526 527static int different_name(struct cache_entry *ce, struct cache_entry *alias) 528{ 529 int len = ce_namelen(ce); 530 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len); 531} 532 533/* 534 * If we add a filename that aliases in the cache, we will use the 535 * name that we already have - but we don't want to update the same 536 * alias twice, because that implies that there were actually two 537 * different files with aliasing names! 538 * 539 * So we use the CE_ADDED flag to verify that the alias was an old 540 * one before we accept it as 541 */ 542static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias) 543{ 544 int len; 545 struct cache_entry *new; 546 547 if (alias->ce_flags & CE_ADDED) 548 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); 549 550 /* Ok, create the new entry using the name of the existing alias */ 551 len = ce_namelen(alias); 552 new = xcalloc(1, cache_entry_size(len)); 553 memcpy(new->name, alias->name, len); 554 copy_cache_entry(new, ce); 555 free(ce); 556 return new; 557} 558 559static void record_intent_to_add(struct cache_entry *ce) 560{ 561 unsigned char sha1[20]; 562 if (write_sha1_file("", 0, blob_type, sha1)) 563 die("cannot create an empty blob in the object database"); 564 hashcpy(ce->sha1, sha1); 565} 566 567int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags) 568{ 569 int size, namelen, was_same; 570 mode_t st_mode = st->st_mode; 571 struct cache_entry *ce, *alias; 572 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY; 573 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); 574 int pretend = flags & ADD_CACHE_PRETEND; 575 int intent_only = flags & ADD_CACHE_INTENT; 576 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| 577 (intent_only ? ADD_CACHE_NEW_ONLY : 0)); 578 579 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 580 return error("%s: can only add regular files, symbolic links or git-directories", path); 581 582 namelen = strlen(path); 583 if (S_ISDIR(st_mode)) { 584 while (namelen && path[namelen-1] == '/') 585 namelen--; 586 } 587 size = cache_entry_size(namelen); 588 ce = xcalloc(1, size); 589 memcpy(ce->name, path, namelen); 590 ce->ce_flags = namelen; 591 if (!intent_only) 592 fill_stat_cache_info(ce, st); 593 else 594 ce->ce_flags |= CE_INTENT_TO_ADD; 595 596 if (trust_executable_bit && has_symlinks) 597 ce->ce_mode = create_ce_mode(st_mode); 598 else { 599 /* If there is an existing entry, pick the mode bits and type 600 * from it, otherwise assume unexecutable regular file. 601 */ 602 struct cache_entry *ent; 603 int pos = index_name_pos_also_unmerged(istate, path, namelen); 604 605 ent = (0 <= pos) ? istate->cache[pos] : NULL; 606 ce->ce_mode = ce_mode_from_stat(ent, st_mode); 607 } 608 609 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case); 610 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) { 611 /* Nothing changed, really */ 612 free(ce); 613 ce_mark_uptodate(alias); 614 alias->ce_flags |= CE_ADDED; 615 return 0; 616 } 617 if (!intent_only) { 618 if (index_path(ce->sha1, path, st, 1)) 619 return error("unable to index file %s", path); 620 } else 621 record_intent_to_add(ce); 622 623 if (ignore_case && alias && different_name(ce, alias)) 624 ce = create_alias_ce(ce, alias); 625 ce->ce_flags |= CE_ADDED; 626 627 /* It was suspected to be racily clean, but it turns out to be Ok */ 628 was_same = (alias && 629 !ce_stage(alias) && 630 !hashcmp(alias->sha1, ce->sha1) && 631 ce->ce_mode == alias->ce_mode); 632 633 if (pretend) 634 ; 635 else if (add_index_entry(istate, ce, add_option)) 636 return error("unable to add %s to index",path); 637 if (verbose && !was_same) 638 printf("add '%s'\n", path); 639 return 0; 640} 641 642int add_file_to_index(struct index_state *istate, const char *path, int flags) 643{ 644 struct stat st; 645 if (lstat(path, &st)) 646 die("%s: unable to stat (%s)", path, strerror(errno)); 647 return add_to_index(istate, path, &st, flags); 648} 649 650struct cache_entry *make_cache_entry(unsigned int mode, 651 const unsigned char *sha1, const char *path, int stage, 652 int refresh) 653{ 654 int size, len; 655 struct cache_entry *ce; 656 657 if (!verify_path(path)) { 658 error("Invalid path '%s'", path); 659 return NULL; 660 } 661 662 len = strlen(path); 663 size = cache_entry_size(len); 664 ce = xcalloc(1, size); 665 666 hashcpy(ce->sha1, sha1); 667 memcpy(ce->name, path, len); 668 ce->ce_flags = create_ce_flags(len, stage); 669 ce->ce_mode = create_ce_mode(mode); 670 671 if (refresh) 672 return refresh_cache_entry(ce, 0); 673 674 return ce; 675} 676 677int ce_same_name(struct cache_entry *a, struct cache_entry *b) 678{ 679 int len = ce_namelen(a); 680 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 681} 682 683int ce_path_match(const struct cache_entry *ce, const char **pathspec) 684{ 685 const char *match, *name; 686 int len; 687 688 if (!pathspec) 689 return 1; 690 691 len = ce_namelen(ce); 692 name = ce->name; 693 while ((match = *pathspec++) != NULL) { 694 int matchlen = strlen(match); 695 if (matchlen > len) 696 continue; 697 if (memcmp(name, match, matchlen)) 698 continue; 699 if (matchlen && name[matchlen-1] == '/') 700 return 1; 701 if (name[matchlen] == '/' || !name[matchlen]) 702 return 1; 703 if (!matchlen) 704 return 1; 705 } 706 return 0; 707} 708 709/* 710 * We fundamentally don't like some paths: we don't want 711 * dot or dot-dot anywhere, and for obvious reasons don't 712 * want to recurse into ".git" either. 713 * 714 * Also, we don't want double slashes or slashes at the 715 * end that can make pathnames ambiguous. 716 */ 717static int verify_dotfile(const char *rest) 718{ 719 /* 720 * The first character was '.', but that 721 * has already been discarded, we now test 722 * the rest. 723 */ 724 switch (*rest) { 725 /* "." is not allowed */ 726 case '\0': case '/': 727 return 0; 728 729 /* 730 * ".git" followed by NUL or slash is bad. This 731 * shares the path end test with the ".." case. 732 */ 733 case 'g': 734 if (rest[1] != 'i') 735 break; 736 if (rest[2] != 't') 737 break; 738 rest += 2; 739 /* fallthrough */ 740 case '.': 741 if (rest[1] == '\0' || rest[1] == '/') 742 return 0; 743 } 744 return 1; 745} 746 747int verify_path(const char *path) 748{ 749 char c; 750 751 goto inside; 752 for (;;) { 753 if (!c) 754 return 1; 755 if (c == '/') { 756inside: 757 c = *path++; 758 switch (c) { 759 default: 760 continue; 761 case '/': case '\0': 762 break; 763 case '.': 764 if (verify_dotfile(path)) 765 continue; 766 } 767 return 0; 768 } 769 c = *path++; 770 } 771} 772 773/* 774 * Do we have another file that has the beginning components being a 775 * proper superset of the name we're trying to add? 776 */ 777static int has_file_name(struct index_state *istate, 778 const struct cache_entry *ce, int pos, int ok_to_replace) 779{ 780 int retval = 0; 781 int len = ce_namelen(ce); 782 int stage = ce_stage(ce); 783 const char *name = ce->name; 784 785 while (pos < istate->cache_nr) { 786 struct cache_entry *p = istate->cache[pos++]; 787 788 if (len >= ce_namelen(p)) 789 break; 790 if (memcmp(name, p->name, len)) 791 break; 792 if (ce_stage(p) != stage) 793 continue; 794 if (p->name[len] != '/') 795 continue; 796 if (p->ce_flags & CE_REMOVE) 797 continue; 798 retval = -1; 799 if (!ok_to_replace) 800 break; 801 remove_index_entry_at(istate, --pos); 802 } 803 return retval; 804} 805 806/* 807 * Do we have another file with a pathname that is a proper 808 * subset of the name we're trying to add? 809 */ 810static int has_dir_name(struct index_state *istate, 811 const struct cache_entry *ce, int pos, int ok_to_replace) 812{ 813 int retval = 0; 814 int stage = ce_stage(ce); 815 const char *name = ce->name; 816 const char *slash = name + ce_namelen(ce); 817 818 for (;;) { 819 int len; 820 821 for (;;) { 822 if (*--slash == '/') 823 break; 824 if (slash <= ce->name) 825 return retval; 826 } 827 len = slash - name; 828 829 pos = index_name_pos(istate, name, create_ce_flags(len, stage)); 830 if (pos >= 0) { 831 /* 832 * Found one, but not so fast. This could 833 * be a marker that says "I was here, but 834 * I am being removed". Such an entry is 835 * not a part of the resulting tree, and 836 * it is Ok to have a directory at the same 837 * path. 838 */ 839 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) { 840 retval = -1; 841 if (!ok_to_replace) 842 break; 843 remove_index_entry_at(istate, pos); 844 continue; 845 } 846 } 847 else 848 pos = -pos-1; 849 850 /* 851 * Trivial optimization: if we find an entry that 852 * already matches the sub-directory, then we know 853 * we're ok, and we can exit. 854 */ 855 while (pos < istate->cache_nr) { 856 struct cache_entry *p = istate->cache[pos]; 857 if ((ce_namelen(p) <= len) || 858 (p->name[len] != '/') || 859 memcmp(p->name, name, len)) 860 break; /* not our subdirectory */ 861 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE)) 862 /* 863 * p is at the same stage as our entry, and 864 * is a subdirectory of what we are looking 865 * at, so we cannot have conflicts at our 866 * level or anything shorter. 867 */ 868 return retval; 869 pos++; 870 } 871 } 872 return retval; 873} 874 875/* We may be in a situation where we already have path/file and path 876 * is being added, or we already have path and path/file is being 877 * added. Either one would result in a nonsense tree that has path 878 * twice when git-write-tree tries to write it out. Prevent it. 879 * 880 * If ok-to-replace is specified, we remove the conflicting entries 881 * from the cache so the caller should recompute the insert position. 882 * When this happens, we return non-zero. 883 */ 884static int check_file_directory_conflict(struct index_state *istate, 885 const struct cache_entry *ce, 886 int pos, int ok_to_replace) 887{ 888 int retval; 889 890 /* 891 * When ce is an "I am going away" entry, we allow it to be added 892 */ 893 if (ce->ce_flags & CE_REMOVE) 894 return 0; 895 896 /* 897 * We check if the path is a sub-path of a subsequent pathname 898 * first, since removing those will not change the position 899 * in the array. 900 */ 901 retval = has_file_name(istate, ce, pos, ok_to_replace); 902 903 /* 904 * Then check if the path might have a clashing sub-directory 905 * before it. 906 */ 907 return retval + has_dir_name(istate, ce, pos, ok_to_replace); 908} 909 910static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option) 911{ 912 int pos; 913 int ok_to_add = option & ADD_CACHE_OK_TO_ADD; 914 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE; 915 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK; 916 int new_only = option & ADD_CACHE_NEW_ONLY; 917 918 cache_tree_invalidate_path(istate->cache_tree, ce->name); 919 pos = index_name_pos(istate, ce->name, ce->ce_flags); 920 921 /* existing match? Just replace it. */ 922 if (pos >= 0) { 923 if (!new_only) 924 replace_index_entry(istate, pos, ce); 925 return 0; 926 } 927 pos = -pos-1; 928 929 /* 930 * Inserting a merged entry ("stage 0") into the index 931 * will always replace all non-merged entries.. 932 */ 933 if (pos < istate->cache_nr && ce_stage(ce) == 0) { 934 while (ce_same_name(istate->cache[pos], ce)) { 935 ok_to_add = 1; 936 if (!remove_index_entry_at(istate, pos)) 937 break; 938 } 939 } 940 941 if (!ok_to_add) 942 return -1; 943 if (!verify_path(ce->name)) 944 return error("Invalid path '%s'", ce->name); 945 946 if (!skip_df_check && 947 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) { 948 if (!ok_to_replace) 949 return error("'%s' appears as both a file and as a directory", 950 ce->name); 951 pos = index_name_pos(istate, ce->name, ce->ce_flags); 952 pos = -pos-1; 953 } 954 return pos + 1; 955} 956 957int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option) 958{ 959 int pos; 960 961 if (option & ADD_CACHE_JUST_APPEND) 962 pos = istate->cache_nr; 963 else { 964 int ret; 965 ret = add_index_entry_with_check(istate, ce, option); 966 if (ret <= 0) 967 return ret; 968 pos = ret - 1; 969 } 970 971 /* Make sure the array is big enough .. */ 972 if (istate->cache_nr == istate->cache_alloc) { 973 istate->cache_alloc = alloc_nr(istate->cache_alloc); 974 istate->cache = xrealloc(istate->cache, 975 istate->cache_alloc * sizeof(struct cache_entry *)); 976 } 977 978 /* Add it in.. */ 979 istate->cache_nr++; 980 if (istate->cache_nr > pos + 1) 981 memmove(istate->cache + pos + 1, 982 istate->cache + pos, 983 (istate->cache_nr - pos - 1) * sizeof(ce)); 984 set_index_entry(istate, pos, ce); 985 istate->cache_changed = 1; 986 return 0; 987} 988 989/* 990 * "refresh" does not calculate a new sha1 file or bring the 991 * cache up-to-date for mode/content changes. But what it 992 * _does_ do is to "re-match" the stat information of a file 993 * with the cache, so that you can refresh the cache for a 994 * file that hasn't been changed but where the stat entry is 995 * out of date. 996 * 997 * For example, you'd want to do this after doing a "git-read-tree", 998 * to link up the stat cache details with the proper files. 999 */1000static struct cache_entry *refresh_cache_ent(struct index_state *istate,1001 struct cache_entry *ce,1002 unsigned int options, int *err)1003{1004 struct stat st;1005 struct cache_entry *updated;1006 int changed, size;1007 int ignore_valid = options & CE_MATCH_IGNORE_VALID;10081009 if (ce_uptodate(ce))1010 return ce;10111012 /*1013 * CE_VALID means the user promised us that the change to1014 * the work tree does not matter and told us not to worry.1015 */1016 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {1017 ce_mark_uptodate(ce);1018 return ce;1019 }10201021 if (lstat(ce->name, &st) < 0) {1022 if (err)1023 *err = errno;1024 return NULL;1025 }10261027 changed = ie_match_stat(istate, ce, &st, options);1028 if (!changed) {1029 /*1030 * The path is unchanged. If we were told to ignore1031 * valid bit, then we did the actual stat check and1032 * found that the entry is unmodified. If the entry1033 * is not marked VALID, this is the place to mark it1034 * valid again, under "assume unchanged" mode.1035 */1036 if (ignore_valid && assume_unchanged &&1037 !(ce->ce_flags & CE_VALID))1038 ; /* mark this one VALID again */1039 else {1040 /*1041 * We do not mark the index itself "modified"1042 * because CE_UPTODATE flag is in-core only;1043 * we are not going to write this change out.1044 */1045 ce_mark_uptodate(ce);1046 return ce;1047 }1048 }10491050 if (ie_modified(istate, ce, &st, options)) {1051 if (err)1052 *err = EINVAL;1053 return NULL;1054 }10551056 size = ce_size(ce);1057 updated = xmalloc(size);1058 memcpy(updated, ce, size);1059 fill_stat_cache_info(updated, &st);1060 /*1061 * If ignore_valid is not set, we should leave CE_VALID bit1062 * alone. Otherwise, paths marked with --no-assume-unchanged1063 * (i.e. things to be edited) will reacquire CE_VALID bit1064 * automatically, which is not really what we want.1065 */1066 if (!ignore_valid && assume_unchanged &&1067 !(ce->ce_flags & CE_VALID))1068 updated->ce_flags &= ~CE_VALID;10691070 return updated;1071}10721073int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)1074{1075 int i;1076 int has_errors = 0;1077 int really = (flags & REFRESH_REALLY) != 0;1078 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;1079 int quiet = (flags & REFRESH_QUIET) != 0;1080 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;1081 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;1082 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;1083 const char *needs_update_message;10841085 needs_update_message = ((flags & REFRESH_SAY_CHANGED)1086 ? "locally modified" : "needs update");1087 for (i = 0; i < istate->cache_nr; i++) {1088 struct cache_entry *ce, *new;1089 int cache_errno = 0;10901091 ce = istate->cache[i];1092 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))1093 continue;10941095 if (ce_stage(ce)) {1096 while ((i < istate->cache_nr) &&1097 ! strcmp(istate->cache[i]->name, ce->name))1098 i++;1099 i--;1100 if (allow_unmerged)1101 continue;1102 printf("%s: needs merge\n", ce->name);1103 has_errors = 1;1104 continue;1105 }11061107 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))1108 continue;11091110 new = refresh_cache_ent(istate, ce, options, &cache_errno);1111 if (new == ce)1112 continue;1113 if (!new) {1114 if (not_new && cache_errno == ENOENT)1115 continue;1116 if (really && cache_errno == EINVAL) {1117 /* If we are doing --really-refresh that1118 * means the index is not valid anymore.1119 */1120 ce->ce_flags &= ~CE_VALID;1121 istate->cache_changed = 1;1122 }1123 if (quiet)1124 continue;1125 printf("%s: %s\n", ce->name, needs_update_message);1126 has_errors = 1;1127 continue;1128 }11291130 replace_index_entry(istate, i, new);1131 }1132 return has_errors;1133}11341135struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)1136{1137 return refresh_cache_ent(&the_index, ce, really, NULL);1138}11391140static int verify_hdr(struct cache_header *hdr, unsigned long size)1141{1142 git_SHA_CTX c;1143 unsigned char sha1[20];11441145 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))1146 return error("bad signature");1147 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))1148 return error("bad index version");1149 git_SHA1_Init(&c);1150 git_SHA1_Update(&c, hdr, size - 20);1151 git_SHA1_Final(sha1, &c);1152 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))1153 return error("bad index file sha1 signature");1154 return 0;1155}11561157static int read_index_extension(struct index_state *istate,1158 const char *ext, void *data, unsigned long sz)1159{1160 switch (CACHE_EXT(ext)) {1161 case CACHE_EXT_TREE:1162 istate->cache_tree = cache_tree_read(data, sz);1163 break;1164 default:1165 if (*ext < 'A' || 'Z' < *ext)1166 return error("index uses %.4s extension, which we do not understand",1167 ext);1168 fprintf(stderr, "ignoring %.4s extension\n", ext);1169 break;1170 }1171 return 0;1172}11731174int read_index(struct index_state *istate)1175{1176 return read_index_from(istate, get_index_file());1177}11781179static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)1180{1181 size_t len;1182 const char *name;11831184 ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);1185 ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);1186#ifdef USE_NSEC1187 ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);1188 ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);1189#else1190 ce->ce_ctime.nsec = 0;1191 ce->ce_mtime.nsec = 0;1192#endif1193 ce->ce_dev = ntohl(ondisk->dev);1194 ce->ce_ino = ntohl(ondisk->ino);1195 ce->ce_mode = ntohl(ondisk->mode);1196 ce->ce_uid = ntohl(ondisk->uid);1197 ce->ce_gid = ntohl(ondisk->gid);1198 ce->ce_size = ntohl(ondisk->size);1199 /* On-disk flags are just 16 bits */1200 ce->ce_flags = ntohs(ondisk->flags);12011202 hashcpy(ce->sha1, ondisk->sha1);12031204 len = ce->ce_flags & CE_NAMEMASK;12051206 if (ce->ce_flags & CE_EXTENDED) {1207 struct ondisk_cache_entry_extended *ondisk2;1208 int extended_flags;1209 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1210 extended_flags = ntohs(ondisk2->flags2) << 16;1211 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1212 if (extended_flags & ~CE_EXTENDED_FLAGS)1213 die("Unknown index entry format %08x", extended_flags);1214 ce->ce_flags |= extended_flags;1215 name = ondisk2->name;1216 }1217 else1218 name = ondisk->name;12191220 if (len == CE_NAMEMASK)1221 len = strlen(name);1222 /*1223 * NEEDSWORK: If the original index is crafted, this copy could1224 * go unchecked.1225 */1226 memcpy(ce->name, name, len + 1);1227}12281229static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)1230{1231 long per_entry;12321233 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);12341235 /*1236 * Alignment can cause differences. This should be "alignof", but1237 * since that's a gcc'ism, just use the size of a pointer.1238 */1239 per_entry += sizeof(void *);1240 return ondisk_size + entries*per_entry;1241}12421243/* remember to discard_cache() before reading a different cache! */1244int read_index_from(struct index_state *istate, const char *path)1245{1246 int fd, i;1247 struct stat st;1248 unsigned long src_offset, dst_offset;1249 struct cache_header *hdr;1250 void *mmap;1251 size_t mmap_size;12521253 errno = EBUSY;1254 if (istate->initialized)1255 return istate->cache_nr;12561257 errno = ENOENT;1258 istate->timestamp.sec = 0;1259 istate->timestamp.nsec = 0;1260 fd = open(path, O_RDONLY);1261 if (fd < 0) {1262 if (errno == ENOENT)1263 return 0;1264 die("index file open failed (%s)", strerror(errno));1265 }12661267 if (fstat(fd, &st))1268 die("cannot stat the open index (%s)", strerror(errno));12691270 errno = EINVAL;1271 mmap_size = xsize_t(st.st_size);1272 if (mmap_size < sizeof(struct cache_header) + 20)1273 die("index file smaller than expected");12741275 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);1276 close(fd);1277 if (mmap == MAP_FAILED)1278 die("unable to map index file");12791280 hdr = mmap;1281 if (verify_hdr(hdr, mmap_size) < 0)1282 goto unmap;12831284 istate->cache_nr = ntohl(hdr->hdr_entries);1285 istate->cache_alloc = alloc_nr(istate->cache_nr);1286 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));12871288 /*1289 * The disk format is actually larger than the in-memory format,1290 * due to space for nsec etc, so even though the in-memory one1291 * has room for a few more flags, we can allocate using the same1292 * index size1293 */1294 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));1295 istate->initialized = 1;12961297 src_offset = sizeof(*hdr);1298 dst_offset = 0;1299 for (i = 0; i < istate->cache_nr; i++) {1300 struct ondisk_cache_entry *disk_ce;1301 struct cache_entry *ce;13021303 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);1304 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);1305 convert_from_disk(disk_ce, ce);1306 set_index_entry(istate, i, ce);13071308 src_offset += ondisk_ce_size(ce);1309 dst_offset += ce_size(ce);1310 }1311 istate->timestamp.sec = st.st_mtime;1312#ifdef USE_NSEC1313 istate->timestamp.nsec = (unsigned int)st.st_mtim.tv_nsec;1314#else1315 istate->timestamp.nsec = 0;1316#endif13171318 while (src_offset <= mmap_size - 20 - 8) {1319 /* After an array of active_nr index entries,1320 * there can be arbitrary number of extended1321 * sections, each of which is prefixed with1322 * extension name (4-byte) and section length1323 * in 4-byte network byte order.1324 */1325 unsigned long extsize;1326 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);1327 extsize = ntohl(extsize);1328 if (read_index_extension(istate,1329 (const char *) mmap + src_offset,1330 (char *) mmap + src_offset + 8,1331 extsize) < 0)1332 goto unmap;1333 src_offset += 8;1334 src_offset += extsize;1335 }1336 munmap(mmap, mmap_size);1337 return istate->cache_nr;13381339unmap:1340 munmap(mmap, mmap_size);1341 errno = EINVAL;1342 die("index file corrupt");1343}13441345int is_index_unborn(struct index_state *istate)1346{1347 return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);1348}13491350int discard_index(struct index_state *istate)1351{1352 istate->cache_nr = 0;1353 istate->cache_changed = 0;1354 istate->timestamp.sec = 0;1355 istate->timestamp.nsec = 0;1356 istate->name_hash_initialized = 0;1357 free_hash(&istate->name_hash);1358 cache_tree_free(&(istate->cache_tree));1359 free(istate->alloc);1360 istate->alloc = NULL;1361 istate->initialized = 0;13621363 /* no need to throw away allocated active_cache */1364 return 0;1365}13661367int unmerged_index(const struct index_state *istate)1368{1369 int i;1370 for (i = 0; i < istate->cache_nr; i++) {1371 if (ce_stage(istate->cache[i]))1372 return 1;1373 }1374 return 0;1375}13761377#define WRITE_BUFFER_SIZE 81921378static unsigned char write_buffer[WRITE_BUFFER_SIZE];1379static unsigned long write_buffer_len;13801381static int ce_write_flush(git_SHA_CTX *context, int fd)1382{1383 unsigned int buffered = write_buffer_len;1384 if (buffered) {1385 git_SHA1_Update(context, write_buffer, buffered);1386 if (write_in_full(fd, write_buffer, buffered) != buffered)1387 return -1;1388 write_buffer_len = 0;1389 }1390 return 0;1391}13921393static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)1394{1395 while (len) {1396 unsigned int buffered = write_buffer_len;1397 unsigned int partial = WRITE_BUFFER_SIZE - buffered;1398 if (partial > len)1399 partial = len;1400 memcpy(write_buffer + buffered, data, partial);1401 buffered += partial;1402 if (buffered == WRITE_BUFFER_SIZE) {1403 write_buffer_len = buffered;1404 if (ce_write_flush(context, fd))1405 return -1;1406 buffered = 0;1407 }1408 write_buffer_len = buffered;1409 len -= partial;1410 data = (char *) data + partial;1411 }1412 return 0;1413}14141415static int write_index_ext_header(git_SHA_CTX *context, int fd,1416 unsigned int ext, unsigned int sz)1417{1418 ext = htonl(ext);1419 sz = htonl(sz);1420 return ((ce_write(context, fd, &ext, 4) < 0) ||1421 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;1422}14231424static int ce_flush(git_SHA_CTX *context, int fd)1425{1426 unsigned int left = write_buffer_len;14271428 if (left) {1429 write_buffer_len = 0;1430 git_SHA1_Update(context, write_buffer, left);1431 }14321433 /* Flush first if not enough space for SHA1 signature */1434 if (left + 20 > WRITE_BUFFER_SIZE) {1435 if (write_in_full(fd, write_buffer, left) != left)1436 return -1;1437 left = 0;1438 }14391440 /* Append the SHA1 signature at the end */1441 git_SHA1_Final(write_buffer + left, context);1442 left += 20;1443 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;1444}14451446static void ce_smudge_racily_clean_entry(struct cache_entry *ce)1447{1448 /*1449 * The only thing we care about in this function is to smudge the1450 * falsely clean entry due to touch-update-touch race, so we leave1451 * everything else as they are. We are called for entries whose1452 * ce_mtime match the index file mtime.1453 *1454 * Note that this actually does not do much for gitlinks, for1455 * which ce_match_stat_basic() always goes to the actual1456 * contents. The caller checks with is_racy_timestamp() which1457 * always says "no" for gitlinks, so we are not called for them ;-)1458 */1459 struct stat st;14601461 if (lstat(ce->name, &st) < 0)1462 return;1463 if (ce_match_stat_basic(ce, &st))1464 return;1465 if (ce_modified_check_fs(ce, &st)) {1466 /* This is "racily clean"; smudge it. Note that this1467 * is a tricky code. At first glance, it may appear1468 * that it can break with this sequence:1469 *1470 * $ echo xyzzy >frotz1471 * $ git-update-index --add frotz1472 * $ : >frotz1473 * $ sleep 31474 * $ echo filfre >nitfol1475 * $ git-update-index --add nitfol1476 *1477 * but it does not. When the second update-index runs,1478 * it notices that the entry "frotz" has the same timestamp1479 * as index, and if we were to smudge it by resetting its1480 * size to zero here, then the object name recorded1481 * in index is the 6-byte file but the cached stat information1482 * becomes zero --- which would then match what we would1483 * obtain from the filesystem next time we stat("frotz").1484 *1485 * However, the second update-index, before calling1486 * this function, notices that the cached size is 61487 * bytes and what is on the filesystem is an empty1488 * file, and never calls us, so the cached size information1489 * for "frotz" stays 6 which does not match the filesystem.1490 */1491 ce->ce_size = 0;1492 }1493}14941495static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)1496{1497 int size = ondisk_ce_size(ce);1498 struct ondisk_cache_entry *ondisk = xcalloc(1, size);1499 char *name;15001501 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);1502 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);1503#ifdef USE_NSEC1504 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);1505 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);1506#else1507 ondisk->ctime.nsec = 0;1508 ondisk->mtime.nsec = 0;1509#endif1510 ondisk->dev = htonl(ce->ce_dev);1511 ondisk->ino = htonl(ce->ce_ino);1512 ondisk->mode = htonl(ce->ce_mode);1513 ondisk->uid = htonl(ce->ce_uid);1514 ondisk->gid = htonl(ce->ce_gid);1515 ondisk->size = htonl(ce->ce_size);1516 hashcpy(ondisk->sha1, ce->sha1);1517 ondisk->flags = htons(ce->ce_flags);1518 if (ce->ce_flags & CE_EXTENDED) {1519 struct ondisk_cache_entry_extended *ondisk2;1520 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1521 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);1522 name = ondisk2->name;1523 }1524 else1525 name = ondisk->name;1526 memcpy(name, ce->name, ce_namelen(ce));15271528 return ce_write(c, fd, ondisk, size);1529}15301531int write_index(const struct index_state *istate, int newfd)1532{1533 git_SHA_CTX c;1534 struct cache_header hdr;1535 int i, err, removed, extended;1536 struct cache_entry **cache = istate->cache;1537 int entries = istate->cache_nr;15381539 for (i = removed = extended = 0; i < entries; i++) {1540 if (cache[i]->ce_flags & CE_REMOVE)1541 removed++;15421543 /* reduce extended entries if possible */1544 cache[i]->ce_flags &= ~CE_EXTENDED;1545 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {1546 extended++;1547 cache[i]->ce_flags |= CE_EXTENDED;1548 }1549 }15501551 hdr.hdr_signature = htonl(CACHE_SIGNATURE);1552 /* for extended format, increase version so older git won't try to read it */1553 hdr.hdr_version = htonl(extended ? 3 : 2);1554 hdr.hdr_entries = htonl(entries - removed);15551556 git_SHA1_Init(&c);1557 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)1558 return -1;15591560 for (i = 0; i < entries; i++) {1561 struct cache_entry *ce = cache[i];1562 if (ce->ce_flags & CE_REMOVE)1563 continue;1564 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))1565 ce_smudge_racily_clean_entry(ce);1566 if (ce_write_entry(&c, newfd, ce) < 0)1567 return -1;1568 }15691570 /* Write extension data here */1571 if (istate->cache_tree) {1572 struct strbuf sb = STRBUF_INIT;15731574 cache_tree_write(&sb, istate->cache_tree);1575 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 01576 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1577 strbuf_release(&sb);1578 if (err)1579 return -1;1580 }1581 return ce_flush(&c, newfd);1582}15831584/*1585 * Read the index file that is potentially unmerged into given1586 * index_state, dropping any unmerged entries. Returns true if1587 * the index is unmerged. Callers who want to refuse to work1588 * from an unmerged state can call this and check its return value,1589 * instead of calling read_cache().1590 */1591int read_index_unmerged(struct index_state *istate)1592{1593 int i;1594 int unmerged = 0;15951596 read_index(istate);1597 for (i = 0; i < istate->cache_nr; i++) {1598 struct cache_entry *ce = istate->cache[i];1599 struct cache_entry *new_ce;1600 int size, len;16011602 if (!ce_stage(ce))1603 continue;1604 unmerged = 1;1605 len = strlen(ce->name);1606 size = cache_entry_size(len);1607 new_ce = xcalloc(1, size);1608 hashcpy(new_ce->sha1, ce->sha1);1609 memcpy(new_ce->name, ce->name, len);1610 new_ce->ce_flags = create_ce_flags(len, 0);1611 new_ce->ce_mode = ce->ce_mode;1612 if (add_index_entry(istate, new_ce, 0))1613 return error("%s: cannot drop to stage #0",1614 ce->name);1615 i = index_name_pos(istate, new_ce->name, len);1616 }1617 return unmerged;1618}16191620struct update_callback_data1621{1622 int flags;1623 int add_errors;1624};16251626static void update_callback(struct diff_queue_struct *q,1627 struct diff_options *opt, void *cbdata)1628{1629 int i;1630 struct update_callback_data *data = cbdata;16311632 for (i = 0; i < q->nr; i++) {1633 struct diff_filepair *p = q->queue[i];1634 const char *path = p->one->path;1635 switch (p->status) {1636 default:1637 die("unexpected diff status %c", p->status);1638 case DIFF_STATUS_UNMERGED:1639 /*1640 * ADD_CACHE_IGNORE_REMOVAL is unset if "git1641 * add -u" is calling us, In such a case, a1642 * missing work tree file needs to be removed1643 * if there is an unmerged entry at stage #2,1644 * but such a diff record is followed by1645 * another with DIFF_STATUS_DELETED (and if1646 * there is no stage #2, we won't see DELETED1647 * nor MODIFIED). We can simply continue1648 * either way.1649 */1650 if (!(data->flags & ADD_CACHE_IGNORE_REMOVAL))1651 continue;1652 /*1653 * Otherwise, it is "git add path" is asking1654 * to explicitly add it; we fall through. A1655 * missing work tree file is an error and is1656 * caught by add_file_to_index() in such a1657 * case.1658 */1659 case DIFF_STATUS_MODIFIED:1660 case DIFF_STATUS_TYPE_CHANGED:1661 if (add_file_to_index(&the_index, path, data->flags)) {1662 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))1663 die("updating files failed");1664 data->add_errors++;1665 }1666 break;1667 case DIFF_STATUS_DELETED:1668 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)1669 break;1670 if (!(data->flags & ADD_CACHE_PRETEND))1671 remove_file_from_index(&the_index, path);1672 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))1673 printf("remove '%s'\n", path);1674 break;1675 }1676 }1677}16781679int add_files_to_cache(const char *prefix, const char **pathspec, int flags)1680{1681 struct update_callback_data data;1682 struct rev_info rev;1683 init_revisions(&rev, prefix);1684 setup_revisions(0, NULL, &rev, NULL);1685 rev.prune_data = pathspec;1686 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;1687 rev.diffopt.format_callback = update_callback;1688 data.flags = flags;1689 data.add_errors = 0;1690 rev.diffopt.format_callback_data = &data;1691 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);1692 return !!data.add_errors;1693}16941695/*1696 * Returns 1 if the path is an "other" path with respect to1697 * the index; that is, the path is not mentioned in the index at all,1698 * either as a file, a directory with some files in the index,1699 * or as an unmerged entry.1700 *1701 * We helpfully remove a trailing "/" from directories so that1702 * the output of read_directory can be used as-is.1703 */1704int index_name_is_other(const struct index_state *istate, const char *name,1705 int namelen)1706{1707 int pos;1708 if (namelen && name[namelen - 1] == '/')1709 namelen--;1710 pos = index_name_pos(istate, name, namelen);1711 if (0 <= pos)1712 return 0; /* exact match */1713 pos = -pos - 1;1714 if (pos < istate->cache_nr) {1715 struct cache_entry *ce = istate->cache[pos];1716 if (ce_namelen(ce) == namelen &&1717 !memcmp(ce->name, name, namelen))1718 return 0; /* Yup, this one exists unmerged */1719 }1720 return 1;1721}