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 "blob.h" 14#include "resolve-undo.h" 15 16static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really); 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#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */ 30 31struct index_state the_index; 32 33static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 34{ 35 istate->cache[nr] = ce; 36 add_name_hash(istate, ce); 37} 38 39static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 40{ 41 struct cache_entry *old = istate->cache[nr]; 42 43 remove_name_hash(old); 44 set_index_entry(istate, nr, ce); 45 istate->cache_changed = 1; 46} 47 48void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name) 49{ 50 struct cache_entry *old = istate->cache[nr], *new; 51 int namelen = strlen(new_name); 52 53 new = xmalloc(cache_entry_size(namelen)); 54 copy_cache_entry(new, old); 55 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK); 56 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen); 57 memcpy(new->name, new_name, namelen + 1); 58 59 cache_tree_invalidate_path(istate->cache_tree, old->name); 60 remove_index_entry_at(istate, nr); 61 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE); 62} 63 64/* 65 * This only updates the "non-critical" parts of the directory 66 * cache, ie the parts that aren't tracked by GIT, and only used 67 * to validate the cache. 68 */ 69void fill_stat_cache_info(struct cache_entry *ce, struct stat *st) 70{ 71 ce->ce_ctime.sec = (unsigned int)st->st_ctime; 72 ce->ce_mtime.sec = (unsigned int)st->st_mtime; 73 ce->ce_ctime.nsec = ST_CTIME_NSEC(*st); 74 ce->ce_mtime.nsec = ST_MTIME_NSEC(*st); 75 ce->ce_dev = st->st_dev; 76 ce->ce_ino = st->st_ino; 77 ce->ce_uid = st->st_uid; 78 ce->ce_gid = st->st_gid; 79 ce->ce_size = st->st_size; 80 81 if (assume_unchanged) 82 ce->ce_flags |= CE_VALID; 83 84 if (S_ISREG(st->st_mode)) 85 ce_mark_uptodate(ce); 86} 87 88static int ce_compare_data(struct cache_entry *ce, struct stat *st) 89{ 90 int match = -1; 91 int fd = open(ce->name, O_RDONLY); 92 93 if (fd >= 0) { 94 unsigned char sha1[20]; 95 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0)) 96 match = hashcmp(sha1, ce->sha1); 97 /* index_fd() closed the file descriptor already */ 98 } 99 return match; 100} 101 102static int ce_compare_link(struct cache_entry *ce, size_t expected_size) 103{ 104 int match = -1; 105 void *buffer; 106 unsigned long size; 107 enum object_type type; 108 struct strbuf sb = STRBUF_INIT; 109 110 if (strbuf_readlink(&sb, ce->name, expected_size)) 111 return -1; 112 113 buffer = read_sha1_file(ce->sha1, &type, &size); 114 if (buffer) { 115 if (size == sb.len) 116 match = memcmp(buffer, sb.buf, size); 117 free(buffer); 118 } 119 strbuf_release(&sb); 120 return match; 121} 122 123static int ce_compare_gitlink(struct cache_entry *ce) 124{ 125 unsigned char sha1[20]; 126 127 /* 128 * We don't actually require that the .git directory 129 * under GITLINK directory be a valid git directory. It 130 * might even be missing (in case nobody populated that 131 * sub-project). 132 * 133 * If so, we consider it always to match. 134 */ 135 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0) 136 return 0; 137 return hashcmp(sha1, ce->sha1); 138} 139 140static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st) 141{ 142 switch (st->st_mode & S_IFMT) { 143 case S_IFREG: 144 if (ce_compare_data(ce, st)) 145 return DATA_CHANGED; 146 break; 147 case S_IFLNK: 148 if (ce_compare_link(ce, xsize_t(st->st_size))) 149 return DATA_CHANGED; 150 break; 151 case S_IFDIR: 152 if (S_ISGITLINK(ce->ce_mode)) 153 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0; 154 default: 155 return TYPE_CHANGED; 156 } 157 return 0; 158} 159 160static int is_empty_blob_sha1(const unsigned char *sha1) 161{ 162 static const unsigned char empty_blob_sha1[20] = { 163 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b, 164 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91 165 }; 166 167 return !hashcmp(sha1, empty_blob_sha1); 168} 169 170static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st) 171{ 172 unsigned int changed = 0; 173 174 if (ce->ce_flags & CE_REMOVE) 175 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; 176 177 switch (ce->ce_mode & S_IFMT) { 178 case S_IFREG: 179 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0; 180 /* We consider only the owner x bit to be relevant for 181 * "mode changes" 182 */ 183 if (trust_executable_bit && 184 (0100 & (ce->ce_mode ^ st->st_mode))) 185 changed |= MODE_CHANGED; 186 break; 187 case S_IFLNK: 188 if (!S_ISLNK(st->st_mode) && 189 (has_symlinks || !S_ISREG(st->st_mode))) 190 changed |= TYPE_CHANGED; 191 break; 192 case S_IFGITLINK: 193 /* We ignore most of the st_xxx fields for gitlinks */ 194 if (!S_ISDIR(st->st_mode)) 195 changed |= TYPE_CHANGED; 196 else if (ce_compare_gitlink(ce)) 197 changed |= DATA_CHANGED; 198 return changed; 199 default: 200 die("internal error: ce_mode is %o", ce->ce_mode); 201 } 202 if (ce->ce_mtime.sec != (unsigned int)st->st_mtime) 203 changed |= MTIME_CHANGED; 204 if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime) 205 changed |= CTIME_CHANGED; 206 207#ifdef USE_NSEC 208 if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st)) 209 changed |= MTIME_CHANGED; 210 if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st)) 211 changed |= CTIME_CHANGED; 212#endif 213 214 if (ce->ce_uid != (unsigned int) st->st_uid || 215 ce->ce_gid != (unsigned int) st->st_gid) 216 changed |= OWNER_CHANGED; 217 if (ce->ce_ino != (unsigned int) st->st_ino) 218 changed |= INODE_CHANGED; 219 220#ifdef USE_STDEV 221 /* 222 * st_dev breaks on network filesystems where different 223 * clients will have different views of what "device" 224 * the filesystem is on 225 */ 226 if (ce->ce_dev != (unsigned int) st->st_dev) 227 changed |= INODE_CHANGED; 228#endif 229 230 if (ce->ce_size != (unsigned int) st->st_size) 231 changed |= DATA_CHANGED; 232 233 /* Racily smudged entry? */ 234 if (!ce->ce_size) { 235 if (!is_empty_blob_sha1(ce->sha1)) 236 changed |= DATA_CHANGED; 237 } 238 239 return changed; 240} 241 242static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce) 243{ 244 return (!S_ISGITLINK(ce->ce_mode) && 245 istate->timestamp.sec && 246#ifdef USE_NSEC 247 /* nanosecond timestamped files can also be racy! */ 248 (istate->timestamp.sec < ce->ce_mtime.sec || 249 (istate->timestamp.sec == ce->ce_mtime.sec && 250 istate->timestamp.nsec <= ce->ce_mtime.nsec)) 251#else 252 istate->timestamp.sec <= ce->ce_mtime.sec 253#endif 254 ); 255} 256 257int ie_match_stat(const struct index_state *istate, 258 struct cache_entry *ce, struct stat *st, 259 unsigned int options) 260{ 261 unsigned int changed; 262 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 263 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE; 264 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 265 266 /* 267 * If it's marked as always valid in the index, it's 268 * valid whatever the checked-out copy says. 269 * 270 * skip-worktree has the same effect with higher precedence 271 */ 272 if (!ignore_skip_worktree && ce_skip_worktree(ce)) 273 return 0; 274 if (!ignore_valid && (ce->ce_flags & CE_VALID)) 275 return 0; 276 277 /* 278 * Intent-to-add entries have not been added, so the index entry 279 * by definition never matches what is in the work tree until it 280 * actually gets added. 281 */ 282 if (ce->ce_flags & CE_INTENT_TO_ADD) 283 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED; 284 285 changed = ce_match_stat_basic(ce, st); 286 287 /* 288 * Within 1 second of this sequence: 289 * echo xyzzy >file && git-update-index --add file 290 * running this command: 291 * echo frotz >file 292 * would give a falsely clean cache entry. The mtime and 293 * length match the cache, and other stat fields do not change. 294 * 295 * We could detect this at update-index time (the cache entry 296 * being registered/updated records the same time as "now") 297 * and delay the return from git-update-index, but that would 298 * effectively mean we can make at most one commit per second, 299 * which is not acceptable. Instead, we check cache entries 300 * whose mtime are the same as the index file timestamp more 301 * carefully than others. 302 */ 303 if (!changed && is_racy_timestamp(istate, ce)) { 304 if (assume_racy_is_modified) 305 changed |= DATA_CHANGED; 306 else 307 changed |= ce_modified_check_fs(ce, st); 308 } 309 310 return changed; 311} 312 313int ie_modified(const struct index_state *istate, 314 struct cache_entry *ce, struct stat *st, unsigned int options) 315{ 316 int changed, changed_fs; 317 318 changed = ie_match_stat(istate, ce, st, options); 319 if (!changed) 320 return 0; 321 /* 322 * If the mode or type has changed, there's no point in trying 323 * to refresh the entry - it's not going to match 324 */ 325 if (changed & (MODE_CHANGED | TYPE_CHANGED)) 326 return changed; 327 328 /* 329 * Immediately after read-tree or update-index --cacheinfo, 330 * the length field is zero, as we have never even read the 331 * lstat(2) information once, and we cannot trust DATA_CHANGED 332 * returned by ie_match_stat() which in turn was returned by 333 * ce_match_stat_basic() to signal that the filesize of the 334 * blob changed. We have to actually go to the filesystem to 335 * see if the contents match, and if so, should answer "unchanged". 336 * 337 * The logic does not apply to gitlinks, as ce_match_stat_basic() 338 * already has checked the actual HEAD from the filesystem in the 339 * subproject. If ie_match_stat() already said it is different, 340 * then we know it is. 341 */ 342 if ((changed & DATA_CHANGED) && 343 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0)) 344 return changed; 345 346 changed_fs = ce_modified_check_fs(ce, st); 347 if (changed_fs) 348 return changed | changed_fs; 349 return 0; 350} 351 352int base_name_compare(const char *name1, int len1, int mode1, 353 const char *name2, int len2, int mode2) 354{ 355 unsigned char c1, c2; 356 int len = len1 < len2 ? len1 : len2; 357 int cmp; 358 359 cmp = memcmp(name1, name2, len); 360 if (cmp) 361 return cmp; 362 c1 = name1[len]; 363 c2 = name2[len]; 364 if (!c1 && S_ISDIR(mode1)) 365 c1 = '/'; 366 if (!c2 && S_ISDIR(mode2)) 367 c2 = '/'; 368 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0; 369} 370 371/* 372 * df_name_compare() is identical to base_name_compare(), except it 373 * compares conflicting directory/file entries as equal. Note that 374 * while a directory name compares as equal to a regular file, they 375 * then individually compare _differently_ to a filename that has 376 * a dot after the basename (because '\0' < '.' < '/'). 377 * 378 * This is used by routines that want to traverse the git namespace 379 * but then handle conflicting entries together when possible. 380 */ 381int df_name_compare(const char *name1, int len1, int mode1, 382 const char *name2, int len2, int mode2) 383{ 384 int len = len1 < len2 ? len1 : len2, cmp; 385 unsigned char c1, c2; 386 387 cmp = memcmp(name1, name2, len); 388 if (cmp) 389 return cmp; 390 /* Directories and files compare equal (same length, same name) */ 391 if (len1 == len2) 392 return 0; 393 c1 = name1[len]; 394 if (!c1 && S_ISDIR(mode1)) 395 c1 = '/'; 396 c2 = name2[len]; 397 if (!c2 && S_ISDIR(mode2)) 398 c2 = '/'; 399 if (c1 == '/' && !c2) 400 return 0; 401 if (c2 == '/' && !c1) 402 return 0; 403 return c1 - c2; 404} 405 406int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2) 407{ 408 int len1 = flags1 & CE_NAMEMASK; 409 int len2 = flags2 & CE_NAMEMASK; 410 int len = len1 < len2 ? len1 : len2; 411 int cmp; 412 413 cmp = memcmp(name1, name2, len); 414 if (cmp) 415 return cmp; 416 if (len1 < len2) 417 return -1; 418 if (len1 > len2) 419 return 1; 420 421 /* Compare stages */ 422 flags1 &= CE_STAGEMASK; 423 flags2 &= CE_STAGEMASK; 424 425 if (flags1 < flags2) 426 return -1; 427 if (flags1 > flags2) 428 return 1; 429 return 0; 430} 431 432int index_name_pos(const struct index_state *istate, const char *name, int namelen) 433{ 434 int first, last; 435 436 first = 0; 437 last = istate->cache_nr; 438 while (last > first) { 439 int next = (last + first) >> 1; 440 struct cache_entry *ce = istate->cache[next]; 441 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags); 442 if (!cmp) 443 return next; 444 if (cmp < 0) { 445 last = next; 446 continue; 447 } 448 first = next+1; 449 } 450 return -first-1; 451} 452 453/* Remove entry, return true if there are more entries to go.. */ 454int remove_index_entry_at(struct index_state *istate, int pos) 455{ 456 struct cache_entry *ce = istate->cache[pos]; 457 458 record_resolve_undo(istate, ce); 459 remove_name_hash(ce); 460 istate->cache_changed = 1; 461 istate->cache_nr--; 462 if (pos >= istate->cache_nr) 463 return 0; 464 memmove(istate->cache + pos, 465 istate->cache + pos + 1, 466 (istate->cache_nr - pos) * sizeof(struct cache_entry *)); 467 return 1; 468} 469 470/* 471 * Remove all cache ententries marked for removal, that is where 472 * CE_REMOVE is set in ce_flags. This is much more effective than 473 * calling remove_index_entry_at() for each entry to be removed. 474 */ 475void remove_marked_cache_entries(struct index_state *istate) 476{ 477 struct cache_entry **ce_array = istate->cache; 478 unsigned int i, j; 479 480 for (i = j = 0; i < istate->cache_nr; i++) { 481 if (ce_array[i]->ce_flags & CE_REMOVE) 482 remove_name_hash(ce_array[i]); 483 else 484 ce_array[j++] = ce_array[i]; 485 } 486 istate->cache_changed = 1; 487 istate->cache_nr = j; 488} 489 490int remove_file_from_index(struct index_state *istate, const char *path) 491{ 492 int pos = index_name_pos(istate, path, strlen(path)); 493 if (pos < 0) 494 pos = -pos-1; 495 cache_tree_invalidate_path(istate->cache_tree, path); 496 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 497 remove_index_entry_at(istate, pos); 498 return 0; 499} 500 501static int compare_name(struct cache_entry *ce, const char *path, int namelen) 502{ 503 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); 504} 505 506static int index_name_pos_also_unmerged(struct index_state *istate, 507 const char *path, int namelen) 508{ 509 int pos = index_name_pos(istate, path, namelen); 510 struct cache_entry *ce; 511 512 if (pos >= 0) 513 return pos; 514 515 /* maybe unmerged? */ 516 pos = -1 - pos; 517 if (pos >= istate->cache_nr || 518 compare_name((ce = istate->cache[pos]), path, namelen)) 519 return -1; 520 521 /* order of preference: stage 2, 1, 3 */ 522 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && 523 ce_stage((ce = istate->cache[pos + 1])) == 2 && 524 !compare_name(ce, path, namelen)) 525 pos++; 526 return pos; 527} 528 529static int different_name(struct cache_entry *ce, struct cache_entry *alias) 530{ 531 int len = ce_namelen(ce); 532 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len); 533} 534 535/* 536 * If we add a filename that aliases in the cache, we will use the 537 * name that we already have - but we don't want to update the same 538 * alias twice, because that implies that there were actually two 539 * different files with aliasing names! 540 * 541 * So we use the CE_ADDED flag to verify that the alias was an old 542 * one before we accept it as 543 */ 544static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias) 545{ 546 int len; 547 struct cache_entry *new; 548 549 if (alias->ce_flags & CE_ADDED) 550 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); 551 552 /* Ok, create the new entry using the name of the existing alias */ 553 len = ce_namelen(alias); 554 new = xcalloc(1, cache_entry_size(len)); 555 memcpy(new->name, alias->name, len); 556 copy_cache_entry(new, ce); 557 free(ce); 558 return new; 559} 560 561static void record_intent_to_add(struct cache_entry *ce) 562{ 563 unsigned char sha1[20]; 564 if (write_sha1_file("", 0, blob_type, sha1)) 565 die("cannot create an empty blob in the object database"); 566 hashcpy(ce->sha1, sha1); 567} 568 569int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags) 570{ 571 int size, namelen, was_same; 572 mode_t st_mode = st->st_mode; 573 struct cache_entry *ce, *alias; 574 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY; 575 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); 576 int pretend = flags & ADD_CACHE_PRETEND; 577 int intent_only = flags & ADD_CACHE_INTENT; 578 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| 579 (intent_only ? ADD_CACHE_NEW_ONLY : 0)); 580 581 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 582 return error("%s: can only add regular files, symbolic links or git-directories", path); 583 584 namelen = strlen(path); 585 if (S_ISDIR(st_mode)) { 586 while (namelen && path[namelen-1] == '/') 587 namelen--; 588 } 589 size = cache_entry_size(namelen); 590 ce = xcalloc(1, size); 591 memcpy(ce->name, path, namelen); 592 ce->ce_flags = namelen; 593 if (!intent_only) 594 fill_stat_cache_info(ce, st); 595 else 596 ce->ce_flags |= CE_INTENT_TO_ADD; 597 598 if (trust_executable_bit && has_symlinks) 599 ce->ce_mode = create_ce_mode(st_mode); 600 else { 601 /* If there is an existing entry, pick the mode bits and type 602 * from it, otherwise assume unexecutable regular file. 603 */ 604 struct cache_entry *ent; 605 int pos = index_name_pos_also_unmerged(istate, path, namelen); 606 607 ent = (0 <= pos) ? istate->cache[pos] : NULL; 608 ce->ce_mode = ce_mode_from_stat(ent, st_mode); 609 } 610 611 /* When core.ignorecase=true, determine if a directory of the same name but differing 612 * case already exists within the Git repository. If it does, ensure the directory 613 * case of the file being added to the repository matches (is folded into) the existing 614 * entry's directory case. 615 */ 616 if (ignore_case) { 617 const char *startPtr = ce->name; 618 const char *ptr = startPtr; 619 while (*ptr) { 620 while (*ptr && *ptr != '/') 621 ++ptr; 622 if (*ptr == '/') { 623 struct cache_entry *foundce; 624 ++ptr; 625 foundce = index_name_exists(&the_index, ce->name, ptr - ce->name, ignore_case); 626 if (foundce) { 627 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr); 628 startPtr = ptr; 629 } 630 } 631 } 632 } 633 634 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case); 635 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) { 636 /* Nothing changed, really */ 637 free(ce); 638 if (!S_ISGITLINK(alias->ce_mode)) 639 ce_mark_uptodate(alias); 640 alias->ce_flags |= CE_ADDED; 641 return 0; 642 } 643 if (!intent_only) { 644 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT)) 645 return error("unable to index file %s", path); 646 } else 647 record_intent_to_add(ce); 648 649 if (ignore_case && alias && different_name(ce, alias)) 650 ce = create_alias_ce(ce, alias); 651 ce->ce_flags |= CE_ADDED; 652 653 /* It was suspected to be racily clean, but it turns out to be Ok */ 654 was_same = (alias && 655 !ce_stage(alias) && 656 !hashcmp(alias->sha1, ce->sha1) && 657 ce->ce_mode == alias->ce_mode); 658 659 if (pretend) 660 ; 661 else if (add_index_entry(istate, ce, add_option)) 662 return error("unable to add %s to index",path); 663 if (verbose && !was_same) 664 printf("add '%s'\n", path); 665 return 0; 666} 667 668int add_file_to_index(struct index_state *istate, const char *path, int flags) 669{ 670 struct stat st; 671 if (lstat(path, &st)) 672 die_errno("unable to stat '%s'", path); 673 return add_to_index(istate, path, &st, flags); 674} 675 676struct cache_entry *make_cache_entry(unsigned int mode, 677 const unsigned char *sha1, const char *path, int stage, 678 int refresh) 679{ 680 int size, len; 681 struct cache_entry *ce; 682 683 if (!verify_path(path)) { 684 error("Invalid path '%s'", path); 685 return NULL; 686 } 687 688 len = strlen(path); 689 size = cache_entry_size(len); 690 ce = xcalloc(1, size); 691 692 hashcpy(ce->sha1, sha1); 693 memcpy(ce->name, path, len); 694 ce->ce_flags = create_ce_flags(len, stage); 695 ce->ce_mode = create_ce_mode(mode); 696 697 if (refresh) 698 return refresh_cache_entry(ce, 0); 699 700 return ce; 701} 702 703int ce_same_name(struct cache_entry *a, struct cache_entry *b) 704{ 705 int len = ce_namelen(a); 706 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 707} 708 709int ce_path_match(const struct cache_entry *ce, const struct pathspec *pathspec) 710{ 711 return match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, NULL); 712} 713 714/* 715 * We fundamentally don't like some paths: we don't want 716 * dot or dot-dot anywhere, and for obvious reasons don't 717 * want to recurse into ".git" either. 718 * 719 * Also, we don't want double slashes or slashes at the 720 * end that can make pathnames ambiguous. 721 */ 722static int verify_dotfile(const char *rest) 723{ 724 /* 725 * The first character was '.', but that 726 * has already been discarded, we now test 727 * the rest. 728 */ 729 switch (*rest) { 730 /* "." is not allowed */ 731 case '\0': case '/': 732 return 0; 733 734 /* 735 * ".git" followed by NUL or slash is bad. This 736 * shares the path end test with the ".." case. 737 */ 738 case 'g': 739 if (rest[1] != 'i') 740 break; 741 if (rest[2] != 't') 742 break; 743 rest += 2; 744 /* fallthrough */ 745 case '.': 746 if (rest[1] == '\0' || rest[1] == '/') 747 return 0; 748 } 749 return 1; 750} 751 752int verify_path(const char *path) 753{ 754 char c; 755 756 goto inside; 757 for (;;) { 758 if (!c) 759 return 1; 760 if (c == '/') { 761inside: 762 c = *path++; 763 switch (c) { 764 default: 765 continue; 766 case '/': case '\0': 767 break; 768 case '.': 769 if (verify_dotfile(path)) 770 continue; 771 } 772 return 0; 773 } 774 c = *path++; 775 } 776} 777 778/* 779 * Do we have another file that has the beginning components being a 780 * proper superset of the name we're trying to add? 781 */ 782static int has_file_name(struct index_state *istate, 783 const struct cache_entry *ce, int pos, int ok_to_replace) 784{ 785 int retval = 0; 786 int len = ce_namelen(ce); 787 int stage = ce_stage(ce); 788 const char *name = ce->name; 789 790 while (pos < istate->cache_nr) { 791 struct cache_entry *p = istate->cache[pos++]; 792 793 if (len >= ce_namelen(p)) 794 break; 795 if (memcmp(name, p->name, len)) 796 break; 797 if (ce_stage(p) != stage) 798 continue; 799 if (p->name[len] != '/') 800 continue; 801 if (p->ce_flags & CE_REMOVE) 802 continue; 803 retval = -1; 804 if (!ok_to_replace) 805 break; 806 remove_index_entry_at(istate, --pos); 807 } 808 return retval; 809} 810 811/* 812 * Do we have another file with a pathname that is a proper 813 * subset of the name we're trying to add? 814 */ 815static int has_dir_name(struct index_state *istate, 816 const struct cache_entry *ce, int pos, int ok_to_replace) 817{ 818 int retval = 0; 819 int stage = ce_stage(ce); 820 const char *name = ce->name; 821 const char *slash = name + ce_namelen(ce); 822 823 for (;;) { 824 int len; 825 826 for (;;) { 827 if (*--slash == '/') 828 break; 829 if (slash <= ce->name) 830 return retval; 831 } 832 len = slash - name; 833 834 pos = index_name_pos(istate, name, create_ce_flags(len, stage)); 835 if (pos >= 0) { 836 /* 837 * Found one, but not so fast. This could 838 * be a marker that says "I was here, but 839 * I am being removed". Such an entry is 840 * not a part of the resulting tree, and 841 * it is Ok to have a directory at the same 842 * path. 843 */ 844 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) { 845 retval = -1; 846 if (!ok_to_replace) 847 break; 848 remove_index_entry_at(istate, pos); 849 continue; 850 } 851 } 852 else 853 pos = -pos-1; 854 855 /* 856 * Trivial optimization: if we find an entry that 857 * already matches the sub-directory, then we know 858 * we're ok, and we can exit. 859 */ 860 while (pos < istate->cache_nr) { 861 struct cache_entry *p = istate->cache[pos]; 862 if ((ce_namelen(p) <= len) || 863 (p->name[len] != '/') || 864 memcmp(p->name, name, len)) 865 break; /* not our subdirectory */ 866 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE)) 867 /* 868 * p is at the same stage as our entry, and 869 * is a subdirectory of what we are looking 870 * at, so we cannot have conflicts at our 871 * level or anything shorter. 872 */ 873 return retval; 874 pos++; 875 } 876 } 877 return retval; 878} 879 880/* We may be in a situation where we already have path/file and path 881 * is being added, or we already have path and path/file is being 882 * added. Either one would result in a nonsense tree that has path 883 * twice when git-write-tree tries to write it out. Prevent it. 884 * 885 * If ok-to-replace is specified, we remove the conflicting entries 886 * from the cache so the caller should recompute the insert position. 887 * When this happens, we return non-zero. 888 */ 889static int check_file_directory_conflict(struct index_state *istate, 890 const struct cache_entry *ce, 891 int pos, int ok_to_replace) 892{ 893 int retval; 894 895 /* 896 * When ce is an "I am going away" entry, we allow it to be added 897 */ 898 if (ce->ce_flags & CE_REMOVE) 899 return 0; 900 901 /* 902 * We check if the path is a sub-path of a subsequent pathname 903 * first, since removing those will not change the position 904 * in the array. 905 */ 906 retval = has_file_name(istate, ce, pos, ok_to_replace); 907 908 /* 909 * Then check if the path might have a clashing sub-directory 910 * before it. 911 */ 912 return retval + has_dir_name(istate, ce, pos, ok_to_replace); 913} 914 915static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option) 916{ 917 int pos; 918 int ok_to_add = option & ADD_CACHE_OK_TO_ADD; 919 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE; 920 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK; 921 int new_only = option & ADD_CACHE_NEW_ONLY; 922 923 cache_tree_invalidate_path(istate->cache_tree, ce->name); 924 pos = index_name_pos(istate, ce->name, ce->ce_flags); 925 926 /* existing match? Just replace it. */ 927 if (pos >= 0) { 928 if (!new_only) 929 replace_index_entry(istate, pos, ce); 930 return 0; 931 } 932 pos = -pos-1; 933 934 /* 935 * Inserting a merged entry ("stage 0") into the index 936 * will always replace all non-merged entries.. 937 */ 938 if (pos < istate->cache_nr && ce_stage(ce) == 0) { 939 while (ce_same_name(istate->cache[pos], ce)) { 940 ok_to_add = 1; 941 if (!remove_index_entry_at(istate, pos)) 942 break; 943 } 944 } 945 946 if (!ok_to_add) 947 return -1; 948 if (!verify_path(ce->name)) 949 return error("Invalid path '%s'", ce->name); 950 951 if (!skip_df_check && 952 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) { 953 if (!ok_to_replace) 954 return error("'%s' appears as both a file and as a directory", 955 ce->name); 956 pos = index_name_pos(istate, ce->name, ce->ce_flags); 957 pos = -pos-1; 958 } 959 return pos + 1; 960} 961 962int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option) 963{ 964 int pos; 965 966 if (option & ADD_CACHE_JUST_APPEND) 967 pos = istate->cache_nr; 968 else { 969 int ret; 970 ret = add_index_entry_with_check(istate, ce, option); 971 if (ret <= 0) 972 return ret; 973 pos = ret - 1; 974 } 975 976 /* Make sure the array is big enough .. */ 977 if (istate->cache_nr == istate->cache_alloc) { 978 istate->cache_alloc = alloc_nr(istate->cache_alloc); 979 istate->cache = xrealloc(istate->cache, 980 istate->cache_alloc * sizeof(struct cache_entry *)); 981 } 982 983 /* Add it in.. */ 984 istate->cache_nr++; 985 if (istate->cache_nr > pos + 1) 986 memmove(istate->cache + pos + 1, 987 istate->cache + pos, 988 (istate->cache_nr - pos - 1) * sizeof(ce)); 989 set_index_entry(istate, pos, ce); 990 istate->cache_changed = 1; 991 return 0; 992} 993 994/* 995 * "refresh" does not calculate a new sha1 file or bring the 996 * cache up-to-date for mode/content changes. But what it 997 * _does_ do is to "re-match" the stat information of a file 998 * with the cache, so that you can refresh the cache for a 999 * file that hasn't been changed but where the stat entry is1000 * out of date.1001 *1002 * For example, you'd want to do this after doing a "git-read-tree",1003 * to link up the stat cache details with the proper files.1004 */1005static struct cache_entry *refresh_cache_ent(struct index_state *istate,1006 struct cache_entry *ce,1007 unsigned int options, int *err)1008{1009 struct stat st;1010 struct cache_entry *updated;1011 int changed, size;1012 int ignore_valid = options & CE_MATCH_IGNORE_VALID;1013 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;10141015 if (ce_uptodate(ce))1016 return ce;10171018 /*1019 * CE_VALID or CE_SKIP_WORKTREE means the user promised us1020 * that the change to the work tree does not matter and told1021 * us not to worry.1022 */1023 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {1024 ce_mark_uptodate(ce);1025 return ce;1026 }1027 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {1028 ce_mark_uptodate(ce);1029 return ce;1030 }10311032 if (lstat(ce->name, &st) < 0) {1033 if (err)1034 *err = errno;1035 return NULL;1036 }10371038 changed = ie_match_stat(istate, ce, &st, options);1039 if (!changed) {1040 /*1041 * The path is unchanged. If we were told to ignore1042 * valid bit, then we did the actual stat check and1043 * found that the entry is unmodified. If the entry1044 * is not marked VALID, this is the place to mark it1045 * valid again, under "assume unchanged" mode.1046 */1047 if (ignore_valid && assume_unchanged &&1048 !(ce->ce_flags & CE_VALID))1049 ; /* mark this one VALID again */1050 else {1051 /*1052 * We do not mark the index itself "modified"1053 * because CE_UPTODATE flag is in-core only;1054 * we are not going to write this change out.1055 */1056 if (!S_ISGITLINK(ce->ce_mode))1057 ce_mark_uptodate(ce);1058 return ce;1059 }1060 }10611062 if (ie_modified(istate, ce, &st, options)) {1063 if (err)1064 *err = EINVAL;1065 return NULL;1066 }10671068 size = ce_size(ce);1069 updated = xmalloc(size);1070 memcpy(updated, ce, size);1071 fill_stat_cache_info(updated, &st);1072 /*1073 * If ignore_valid is not set, we should leave CE_VALID bit1074 * alone. Otherwise, paths marked with --no-assume-unchanged1075 * (i.e. things to be edited) will reacquire CE_VALID bit1076 * automatically, which is not really what we want.1077 */1078 if (!ignore_valid && assume_unchanged &&1079 !(ce->ce_flags & CE_VALID))1080 updated->ce_flags &= ~CE_VALID;10811082 return updated;1083}10841085static void show_file(const char * fmt, const char * name, int in_porcelain,1086 int * first, const char *header_msg)1087{1088 if (in_porcelain && *first && header_msg) {1089 printf("%s\n", header_msg);1090 *first=0;1091 }1092 printf(fmt, name);1093}10941095int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,1096 char *seen, const char *header_msg)1097{1098 int i;1099 int has_errors = 0;1100 int really = (flags & REFRESH_REALLY) != 0;1101 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;1102 int quiet = (flags & REFRESH_QUIET) != 0;1103 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;1104 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;1105 int first = 1;1106 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);1107 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;1108 const char *needs_update_fmt;1109 const char *needs_merge_fmt;11101111 needs_update_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");1112 needs_merge_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");1113 for (i = 0; i < istate->cache_nr; i++) {1114 struct cache_entry *ce, *new;1115 int cache_errno = 0;11161117 ce = istate->cache[i];1118 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))1119 continue;11201121 if (ce_stage(ce)) {1122 while ((i < istate->cache_nr) &&1123 ! strcmp(istate->cache[i]->name, ce->name))1124 i++;1125 i--;1126 if (allow_unmerged)1127 continue;1128 show_file(needs_merge_fmt, ce->name, in_porcelain, &first, header_msg);1129 has_errors = 1;1130 continue;1131 }11321133 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))1134 continue;11351136 new = refresh_cache_ent(istate, ce, options, &cache_errno);1137 if (new == ce)1138 continue;1139 if (!new) {1140 if (not_new && cache_errno == ENOENT)1141 continue;1142 if (really && cache_errno == EINVAL) {1143 /* If we are doing --really-refresh that1144 * means the index is not valid anymore.1145 */1146 ce->ce_flags &= ~CE_VALID;1147 istate->cache_changed = 1;1148 }1149 if (quiet)1150 continue;1151 show_file(needs_update_fmt, ce->name, in_porcelain, &first, header_msg);1152 has_errors = 1;1153 continue;1154 }11551156 replace_index_entry(istate, i, new);1157 }1158 return has_errors;1159}11601161static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)1162{1163 return refresh_cache_ent(&the_index, ce, really, NULL);1164}11651166static int verify_hdr(struct cache_header *hdr, unsigned long size)1167{1168 git_SHA_CTX c;1169 unsigned char sha1[20];11701171 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))1172 return error("bad signature");1173 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))1174 return error("bad index version");1175 git_SHA1_Init(&c);1176 git_SHA1_Update(&c, hdr, size - 20);1177 git_SHA1_Final(sha1, &c);1178 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))1179 return error("bad index file sha1 signature");1180 return 0;1181}11821183static int read_index_extension(struct index_state *istate,1184 const char *ext, void *data, unsigned long sz)1185{1186 switch (CACHE_EXT(ext)) {1187 case CACHE_EXT_TREE:1188 istate->cache_tree = cache_tree_read(data, sz);1189 break;1190 case CACHE_EXT_RESOLVE_UNDO:1191 istate->resolve_undo = resolve_undo_read(data, sz);1192 break;1193 default:1194 if (*ext < 'A' || 'Z' < *ext)1195 return error("index uses %.4s extension, which we do not understand",1196 ext);1197 fprintf(stderr, "ignoring %.4s extension\n", ext);1198 break;1199 }1200 return 0;1201}12021203int read_index(struct index_state *istate)1204{1205 return read_index_from(istate, get_index_file());1206}12071208static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)1209{1210 size_t len;1211 const char *name;12121213 ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);1214 ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);1215 ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);1216 ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);1217 ce->ce_dev = ntohl(ondisk->dev);1218 ce->ce_ino = ntohl(ondisk->ino);1219 ce->ce_mode = ntohl(ondisk->mode);1220 ce->ce_uid = ntohl(ondisk->uid);1221 ce->ce_gid = ntohl(ondisk->gid);1222 ce->ce_size = ntohl(ondisk->size);1223 /* On-disk flags are just 16 bits */1224 ce->ce_flags = ntohs(ondisk->flags);12251226 hashcpy(ce->sha1, ondisk->sha1);12271228 len = ce->ce_flags & CE_NAMEMASK;12291230 if (ce->ce_flags & CE_EXTENDED) {1231 struct ondisk_cache_entry_extended *ondisk2;1232 int extended_flags;1233 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1234 extended_flags = ntohs(ondisk2->flags2) << 16;1235 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1236 if (extended_flags & ~CE_EXTENDED_FLAGS)1237 die("Unknown index entry format %08x", extended_flags);1238 ce->ce_flags |= extended_flags;1239 name = ondisk2->name;1240 }1241 else1242 name = ondisk->name;12431244 if (len == CE_NAMEMASK)1245 len = strlen(name);1246 /*1247 * NEEDSWORK: If the original index is crafted, this copy could1248 * go unchecked.1249 */1250 memcpy(ce->name, name, len + 1);1251}12521253static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)1254{1255 long per_entry;12561257 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);12581259 /*1260 * Alignment can cause differences. This should be "alignof", but1261 * since that's a gcc'ism, just use the size of a pointer.1262 */1263 per_entry += sizeof(void *);1264 return ondisk_size + entries*per_entry;1265}12661267/* remember to discard_cache() before reading a different cache! */1268int read_index_from(struct index_state *istate, const char *path)1269{1270 int fd, i;1271 struct stat st;1272 unsigned long src_offset, dst_offset;1273 struct cache_header *hdr;1274 void *mmap;1275 size_t mmap_size;12761277 errno = EBUSY;1278 if (istate->initialized)1279 return istate->cache_nr;12801281 errno = ENOENT;1282 istate->timestamp.sec = 0;1283 istate->timestamp.nsec = 0;1284 fd = open(path, O_RDONLY);1285 if (fd < 0) {1286 if (errno == ENOENT)1287 return 0;1288 die_errno("index file open failed");1289 }12901291 if (fstat(fd, &st))1292 die_errno("cannot stat the open index");12931294 errno = EINVAL;1295 mmap_size = xsize_t(st.st_size);1296 if (mmap_size < sizeof(struct cache_header) + 20)1297 die("index file smaller than expected");12981299 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);1300 close(fd);1301 if (mmap == MAP_FAILED)1302 die_errno("unable to map index file");13031304 hdr = mmap;1305 if (verify_hdr(hdr, mmap_size) < 0)1306 goto unmap;13071308 istate->cache_nr = ntohl(hdr->hdr_entries);1309 istate->cache_alloc = alloc_nr(istate->cache_nr);1310 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));13111312 /*1313 * The disk format is actually larger than the in-memory format,1314 * due to space for nsec etc, so even though the in-memory one1315 * has room for a few more flags, we can allocate using the same1316 * index size1317 */1318 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));1319 istate->initialized = 1;13201321 src_offset = sizeof(*hdr);1322 dst_offset = 0;1323 for (i = 0; i < istate->cache_nr; i++) {1324 struct ondisk_cache_entry *disk_ce;1325 struct cache_entry *ce;13261327 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);1328 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);1329 convert_from_disk(disk_ce, ce);1330 set_index_entry(istate, i, ce);13311332 src_offset += ondisk_ce_size(ce);1333 dst_offset += ce_size(ce);1334 }1335 istate->timestamp.sec = st.st_mtime;1336 istate->timestamp.nsec = ST_MTIME_NSEC(st);13371338 while (src_offset <= mmap_size - 20 - 8) {1339 /* After an array of active_nr index entries,1340 * there can be arbitrary number of extended1341 * sections, each of which is prefixed with1342 * extension name (4-byte) and section length1343 * in 4-byte network byte order.1344 */1345 uint32_t extsize;1346 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);1347 extsize = ntohl(extsize);1348 if (read_index_extension(istate,1349 (const char *) mmap + src_offset,1350 (char *) mmap + src_offset + 8,1351 extsize) < 0)1352 goto unmap;1353 src_offset += 8;1354 src_offset += extsize;1355 }1356 munmap(mmap, mmap_size);1357 return istate->cache_nr;13581359unmap:1360 munmap(mmap, mmap_size);1361 errno = EINVAL;1362 die("index file corrupt");1363}13641365int is_index_unborn(struct index_state *istate)1366{1367 return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);1368}13691370int discard_index(struct index_state *istate)1371{1372 resolve_undo_clear_index(istate);1373 istate->cache_nr = 0;1374 istate->cache_changed = 0;1375 istate->timestamp.sec = 0;1376 istate->timestamp.nsec = 0;1377 istate->name_hash_initialized = 0;1378 free_hash(&istate->name_hash);1379 cache_tree_free(&(istate->cache_tree));1380 free(istate->alloc);1381 istate->alloc = NULL;1382 istate->initialized = 0;13831384 /* no need to throw away allocated active_cache */1385 return 0;1386}13871388int unmerged_index(const struct index_state *istate)1389{1390 int i;1391 for (i = 0; i < istate->cache_nr; i++) {1392 if (ce_stage(istate->cache[i]))1393 return 1;1394 }1395 return 0;1396}13971398#define WRITE_BUFFER_SIZE 81921399static unsigned char write_buffer[WRITE_BUFFER_SIZE];1400static unsigned long write_buffer_len;14011402static int ce_write_flush(git_SHA_CTX *context, int fd)1403{1404 unsigned int buffered = write_buffer_len;1405 if (buffered) {1406 git_SHA1_Update(context, write_buffer, buffered);1407 if (write_in_full(fd, write_buffer, buffered) != buffered)1408 return -1;1409 write_buffer_len = 0;1410 }1411 return 0;1412}14131414static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)1415{1416 while (len) {1417 unsigned int buffered = write_buffer_len;1418 unsigned int partial = WRITE_BUFFER_SIZE - buffered;1419 if (partial > len)1420 partial = len;1421 memcpy(write_buffer + buffered, data, partial);1422 buffered += partial;1423 if (buffered == WRITE_BUFFER_SIZE) {1424 write_buffer_len = buffered;1425 if (ce_write_flush(context, fd))1426 return -1;1427 buffered = 0;1428 }1429 write_buffer_len = buffered;1430 len -= partial;1431 data = (char *) data + partial;1432 }1433 return 0;1434}14351436static int write_index_ext_header(git_SHA_CTX *context, int fd,1437 unsigned int ext, unsigned int sz)1438{1439 ext = htonl(ext);1440 sz = htonl(sz);1441 return ((ce_write(context, fd, &ext, 4) < 0) ||1442 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;1443}14441445static int ce_flush(git_SHA_CTX *context, int fd)1446{1447 unsigned int left = write_buffer_len;14481449 if (left) {1450 write_buffer_len = 0;1451 git_SHA1_Update(context, write_buffer, left);1452 }14531454 /* Flush first if not enough space for SHA1 signature */1455 if (left + 20 > WRITE_BUFFER_SIZE) {1456 if (write_in_full(fd, write_buffer, left) != left)1457 return -1;1458 left = 0;1459 }14601461 /* Append the SHA1 signature at the end */1462 git_SHA1_Final(write_buffer + left, context);1463 left += 20;1464 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;1465}14661467static void ce_smudge_racily_clean_entry(struct cache_entry *ce)1468{1469 /*1470 * The only thing we care about in this function is to smudge the1471 * falsely clean entry due to touch-update-touch race, so we leave1472 * everything else as they are. We are called for entries whose1473 * ce_mtime match the index file mtime.1474 *1475 * Note that this actually does not do much for gitlinks, for1476 * which ce_match_stat_basic() always goes to the actual1477 * contents. The caller checks with is_racy_timestamp() which1478 * always says "no" for gitlinks, so we are not called for them ;-)1479 */1480 struct stat st;14811482 if (lstat(ce->name, &st) < 0)1483 return;1484 if (ce_match_stat_basic(ce, &st))1485 return;1486 if (ce_modified_check_fs(ce, &st)) {1487 /* This is "racily clean"; smudge it. Note that this1488 * is a tricky code. At first glance, it may appear1489 * that it can break with this sequence:1490 *1491 * $ echo xyzzy >frotz1492 * $ git-update-index --add frotz1493 * $ : >frotz1494 * $ sleep 31495 * $ echo filfre >nitfol1496 * $ git-update-index --add nitfol1497 *1498 * but it does not. When the second update-index runs,1499 * it notices that the entry "frotz" has the same timestamp1500 * as index, and if we were to smudge it by resetting its1501 * size to zero here, then the object name recorded1502 * in index is the 6-byte file but the cached stat information1503 * becomes zero --- which would then match what we would1504 * obtain from the filesystem next time we stat("frotz").1505 *1506 * However, the second update-index, before calling1507 * this function, notices that the cached size is 61508 * bytes and what is on the filesystem is an empty1509 * file, and never calls us, so the cached size information1510 * for "frotz" stays 6 which does not match the filesystem.1511 */1512 ce->ce_size = 0;1513 }1514}15151516static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)1517{1518 int size = ondisk_ce_size(ce);1519 struct ondisk_cache_entry *ondisk = xcalloc(1, size);1520 char *name;1521 int result;15221523 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);1524 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);1525 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);1526 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);1527 ondisk->dev = htonl(ce->ce_dev);1528 ondisk->ino = htonl(ce->ce_ino);1529 ondisk->mode = htonl(ce->ce_mode);1530 ondisk->uid = htonl(ce->ce_uid);1531 ondisk->gid = htonl(ce->ce_gid);1532 ondisk->size = htonl(ce->ce_size);1533 hashcpy(ondisk->sha1, ce->sha1);1534 ondisk->flags = htons(ce->ce_flags);1535 if (ce->ce_flags & CE_EXTENDED) {1536 struct ondisk_cache_entry_extended *ondisk2;1537 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1538 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);1539 name = ondisk2->name;1540 }1541 else1542 name = ondisk->name;1543 memcpy(name, ce->name, ce_namelen(ce));15441545 result = ce_write(c, fd, ondisk, size);1546 free(ondisk);1547 return result;1548}15491550static int has_racy_timestamp(struct index_state *istate)1551{1552 int entries = istate->cache_nr;1553 int i;15541555 for (i = 0; i < entries; i++) {1556 struct cache_entry *ce = istate->cache[i];1557 if (is_racy_timestamp(istate, ce))1558 return 1;1559 }1560 return 0;1561}15621563/*1564 * Opportunisticly update the index but do not complain if we can't1565 */1566void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)1567{1568 if ((istate->cache_changed || has_racy_timestamp(istate)) &&1569 !write_index(istate, lockfile->fd))1570 commit_locked_index(lockfile);1571 else1572 rollback_lock_file(lockfile);1573}15741575int write_index(struct index_state *istate, int newfd)1576{1577 git_SHA_CTX c;1578 struct cache_header hdr;1579 int i, err, removed, extended;1580 struct cache_entry **cache = istate->cache;1581 int entries = istate->cache_nr;1582 struct stat st;15831584 for (i = removed = extended = 0; i < entries; i++) {1585 if (cache[i]->ce_flags & CE_REMOVE)1586 removed++;15871588 /* reduce extended entries if possible */1589 cache[i]->ce_flags &= ~CE_EXTENDED;1590 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {1591 extended++;1592 cache[i]->ce_flags |= CE_EXTENDED;1593 }1594 }15951596 hdr.hdr_signature = htonl(CACHE_SIGNATURE);1597 /* for extended format, increase version so older git won't try to read it */1598 hdr.hdr_version = htonl(extended ? 3 : 2);1599 hdr.hdr_entries = htonl(entries - removed);16001601 git_SHA1_Init(&c);1602 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)1603 return -1;16041605 for (i = 0; i < entries; i++) {1606 struct cache_entry *ce = cache[i];1607 if (ce->ce_flags & CE_REMOVE)1608 continue;1609 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))1610 ce_smudge_racily_clean_entry(ce);1611 if (ce_write_entry(&c, newfd, ce) < 0)1612 return -1;1613 }16141615 /* Write extension data here */1616 if (istate->cache_tree) {1617 struct strbuf sb = STRBUF_INIT;16181619 cache_tree_write(&sb, istate->cache_tree);1620 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 01621 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1622 strbuf_release(&sb);1623 if (err)1624 return -1;1625 }1626 if (istate->resolve_undo) {1627 struct strbuf sb = STRBUF_INIT;16281629 resolve_undo_write(&sb, istate->resolve_undo);1630 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,1631 sb.len) < 01632 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1633 strbuf_release(&sb);1634 if (err)1635 return -1;1636 }16371638 if (ce_flush(&c, newfd) || fstat(newfd, &st))1639 return -1;1640 istate->timestamp.sec = (unsigned int)st.st_mtime;1641 istate->timestamp.nsec = ST_MTIME_NSEC(st);1642 return 0;1643}16441645/*1646 * Read the index file that is potentially unmerged into given1647 * index_state, dropping any unmerged entries. Returns true if1648 * the index is unmerged. Callers who want to refuse to work1649 * from an unmerged state can call this and check its return value,1650 * instead of calling read_cache().1651 */1652int read_index_unmerged(struct index_state *istate)1653{1654 int i;1655 int unmerged = 0;16561657 read_index(istate);1658 for (i = 0; i < istate->cache_nr; i++) {1659 struct cache_entry *ce = istate->cache[i];1660 struct cache_entry *new_ce;1661 int size, len;16621663 if (!ce_stage(ce))1664 continue;1665 unmerged = 1;1666 len = strlen(ce->name);1667 size = cache_entry_size(len);1668 new_ce = xcalloc(1, size);1669 memcpy(new_ce->name, ce->name, len);1670 new_ce->ce_flags = create_ce_flags(len, 0) | CE_CONFLICTED;1671 new_ce->ce_mode = ce->ce_mode;1672 if (add_index_entry(istate, new_ce, 0))1673 return error("%s: cannot drop to stage #0",1674 ce->name);1675 i = index_name_pos(istate, new_ce->name, len);1676 }1677 return unmerged;1678}16791680/*1681 * Returns 1 if the path is an "other" path with respect to1682 * the index; that is, the path is not mentioned in the index at all,1683 * either as a file, a directory with some files in the index,1684 * or as an unmerged entry.1685 *1686 * We helpfully remove a trailing "/" from directories so that1687 * the output of read_directory can be used as-is.1688 */1689int index_name_is_other(const struct index_state *istate, const char *name,1690 int namelen)1691{1692 int pos;1693 if (namelen && name[namelen - 1] == '/')1694 namelen--;1695 pos = index_name_pos(istate, name, namelen);1696 if (0 <= pos)1697 return 0; /* exact match */1698 pos = -pos - 1;1699 if (pos < istate->cache_nr) {1700 struct cache_entry *ce = istate->cache[pos];1701 if (ce_namelen(ce) == namelen &&1702 !memcmp(ce->name, name, namelen))1703 return 0; /* Yup, this one exists unmerged */1704 }1705 return 1;1706}