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#include "strbuf.h" 16#include "varint.h" 17 18static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really); 19 20/* Mask for the name length in ce_flags in the on-disk index */ 21 22#define CE_NAMEMASK (0x0fff) 23 24/* Index extensions. 25 * 26 * The first letter should be 'A'..'Z' for extensions that are not 27 * necessary for a correct operation (i.e. optimization data). 28 * When new extensions are added that _needs_ to be understood in 29 * order to correctly interpret the index file, pick character that 30 * is outside the range, to cause the reader to abort. 31 */ 32 33#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) ) 34#define CACHE_EXT_TREE 0x54524545 /* "TREE" */ 35#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */ 36 37struct index_state the_index; 38 39static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 40{ 41 istate->cache[nr] = ce; 42 add_name_hash(istate, ce); 43} 44 45static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 46{ 47 struct cache_entry *old = istate->cache[nr]; 48 49 remove_name_hash(istate, old); 50 set_index_entry(istate, nr, ce); 51 istate->cache_changed = 1; 52} 53 54void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name) 55{ 56 struct cache_entry *old = istate->cache[nr], *new; 57 int namelen = strlen(new_name); 58 59 new = xmalloc(cache_entry_size(namelen)); 60 copy_cache_entry(new, old); 61 new->ce_flags &= ~CE_STATE_MASK; 62 new->ce_namelen = namelen; 63 memcpy(new->name, new_name, namelen + 1); 64 65 cache_tree_invalidate_path(istate->cache_tree, old->name); 66 remove_index_entry_at(istate, nr); 67 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE); 68} 69 70/* 71 * This only updates the "non-critical" parts of the directory 72 * cache, ie the parts that aren't tracked by GIT, and only used 73 * to validate the cache. 74 */ 75void fill_stat_cache_info(struct cache_entry *ce, struct stat *st) 76{ 77 ce->ce_ctime.sec = (unsigned int)st->st_ctime; 78 ce->ce_mtime.sec = (unsigned int)st->st_mtime; 79 ce->ce_ctime.nsec = ST_CTIME_NSEC(*st); 80 ce->ce_mtime.nsec = ST_MTIME_NSEC(*st); 81 ce->ce_dev = st->st_dev; 82 ce->ce_ino = st->st_ino; 83 ce->ce_uid = st->st_uid; 84 ce->ce_gid = st->st_gid; 85 ce->ce_size = st->st_size; 86 87 if (assume_unchanged) 88 ce->ce_flags |= CE_VALID; 89 90 if (S_ISREG(st->st_mode)) 91 ce_mark_uptodate(ce); 92} 93 94static int ce_compare_data(const struct cache_entry *ce, struct stat *st) 95{ 96 int match = -1; 97 int fd = open(ce->name, O_RDONLY); 98 99 if (fd >= 0) { 100 unsigned char sha1[20]; 101 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0)) 102 match = hashcmp(sha1, ce->sha1); 103 /* index_fd() closed the file descriptor already */ 104 } 105 return match; 106} 107 108static int ce_compare_link(const struct cache_entry *ce, size_t expected_size) 109{ 110 int match = -1; 111 void *buffer; 112 unsigned long size; 113 enum object_type type; 114 struct strbuf sb = STRBUF_INIT; 115 116 if (strbuf_readlink(&sb, ce->name, expected_size)) 117 return -1; 118 119 buffer = read_sha1_file(ce->sha1, &type, &size); 120 if (buffer) { 121 if (size == sb.len) 122 match = memcmp(buffer, sb.buf, size); 123 free(buffer); 124 } 125 strbuf_release(&sb); 126 return match; 127} 128 129static int ce_compare_gitlink(const struct cache_entry *ce) 130{ 131 unsigned char sha1[20]; 132 133 /* 134 * We don't actually require that the .git directory 135 * under GITLINK directory be a valid git directory. It 136 * might even be missing (in case nobody populated that 137 * sub-project). 138 * 139 * If so, we consider it always to match. 140 */ 141 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0) 142 return 0; 143 return hashcmp(sha1, ce->sha1); 144} 145 146static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st) 147{ 148 switch (st->st_mode & S_IFMT) { 149 case S_IFREG: 150 if (ce_compare_data(ce, st)) 151 return DATA_CHANGED; 152 break; 153 case S_IFLNK: 154 if (ce_compare_link(ce, xsize_t(st->st_size))) 155 return DATA_CHANGED; 156 break; 157 case S_IFDIR: 158 if (S_ISGITLINK(ce->ce_mode)) 159 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0; 160 default: 161 return TYPE_CHANGED; 162 } 163 return 0; 164} 165 166static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st) 167{ 168 unsigned int changed = 0; 169 170 if (ce->ce_flags & CE_REMOVE) 171 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; 172 173 switch (ce->ce_mode & S_IFMT) { 174 case S_IFREG: 175 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0; 176 /* We consider only the owner x bit to be relevant for 177 * "mode changes" 178 */ 179 if (trust_executable_bit && 180 (0100 & (ce->ce_mode ^ st->st_mode))) 181 changed |= MODE_CHANGED; 182 break; 183 case S_IFLNK: 184 if (!S_ISLNK(st->st_mode) && 185 (has_symlinks || !S_ISREG(st->st_mode))) 186 changed |= TYPE_CHANGED; 187 break; 188 case S_IFGITLINK: 189 /* We ignore most of the st_xxx fields for gitlinks */ 190 if (!S_ISDIR(st->st_mode)) 191 changed |= TYPE_CHANGED; 192 else if (ce_compare_gitlink(ce)) 193 changed |= DATA_CHANGED; 194 return changed; 195 default: 196 die("internal error: ce_mode is %o", ce->ce_mode); 197 } 198 if (ce->ce_mtime.sec != (unsigned int)st->st_mtime) 199 changed |= MTIME_CHANGED; 200 if (trust_ctime && check_stat && 201 ce->ce_ctime.sec != (unsigned int)st->st_ctime) 202 changed |= CTIME_CHANGED; 203 204#ifdef USE_NSEC 205 if (check_stat && ce->ce_mtime.nsec != ST_MTIME_NSEC(*st)) 206 changed |= MTIME_CHANGED; 207 if (trust_ctime && check_stat && 208 ce->ce_ctime.nsec != ST_CTIME_NSEC(*st)) 209 changed |= CTIME_CHANGED; 210#endif 211 212 if (check_stat) { 213 if (ce->ce_uid != (unsigned int) st->st_uid || 214 ce->ce_gid != (unsigned int) st->st_gid) 215 changed |= OWNER_CHANGED; 216 if (ce->ce_ino != (unsigned int) st->st_ino) 217 changed |= INODE_CHANGED; 218 } 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 (check_stat && 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, 243 const struct cache_entry *ce) 244{ 245 return (!S_ISGITLINK(ce->ce_mode) && 246 istate->timestamp.sec && 247#ifdef USE_NSEC 248 /* nanosecond timestamped files can also be racy! */ 249 (istate->timestamp.sec < ce->ce_mtime.sec || 250 (istate->timestamp.sec == ce->ce_mtime.sec && 251 istate->timestamp.nsec <= ce->ce_mtime.nsec)) 252#else 253 istate->timestamp.sec <= ce->ce_mtime.sec 254#endif 255 ); 256} 257 258int ie_match_stat(const struct index_state *istate, 259 const struct cache_entry *ce, struct stat *st, 260 unsigned int options) 261{ 262 unsigned int changed; 263 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 264 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE; 265 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 266 267 /* 268 * If it's marked as always valid in the index, it's 269 * valid whatever the checked-out copy says. 270 * 271 * skip-worktree has the same effect with higher precedence 272 */ 273 if (!ignore_skip_worktree && ce_skip_worktree(ce)) 274 return 0; 275 if (!ignore_valid && (ce->ce_flags & CE_VALID)) 276 return 0; 277 278 /* 279 * Intent-to-add entries have not been added, so the index entry 280 * by definition never matches what is in the work tree until it 281 * actually gets added. 282 */ 283 if (ce->ce_flags & CE_INTENT_TO_ADD) 284 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED; 285 286 changed = ce_match_stat_basic(ce, st); 287 288 /* 289 * Within 1 second of this sequence: 290 * echo xyzzy >file && git-update-index --add file 291 * running this command: 292 * echo frotz >file 293 * would give a falsely clean cache entry. The mtime and 294 * length match the cache, and other stat fields do not change. 295 * 296 * We could detect this at update-index time (the cache entry 297 * being registered/updated records the same time as "now") 298 * and delay the return from git-update-index, but that would 299 * effectively mean we can make at most one commit per second, 300 * which is not acceptable. Instead, we check cache entries 301 * whose mtime are the same as the index file timestamp more 302 * carefully than others. 303 */ 304 if (!changed && is_racy_timestamp(istate, ce)) { 305 if (assume_racy_is_modified) 306 changed |= DATA_CHANGED; 307 else 308 changed |= ce_modified_check_fs(ce, st); 309 } 310 311 return changed; 312} 313 314int ie_modified(const struct index_state *istate, 315 const struct cache_entry *ce, 316 struct stat *st, unsigned int options) 317{ 318 int changed, changed_fs; 319 320 changed = ie_match_stat(istate, ce, st, options); 321 if (!changed) 322 return 0; 323 /* 324 * If the mode or type has changed, there's no point in trying 325 * to refresh the entry - it's not going to match 326 */ 327 if (changed & (MODE_CHANGED | TYPE_CHANGED)) 328 return changed; 329 330 /* 331 * Immediately after read-tree or update-index --cacheinfo, 332 * the length field is zero, as we have never even read the 333 * lstat(2) information once, and we cannot trust DATA_CHANGED 334 * returned by ie_match_stat() which in turn was returned by 335 * ce_match_stat_basic() to signal that the filesize of the 336 * blob changed. We have to actually go to the filesystem to 337 * see if the contents match, and if so, should answer "unchanged". 338 * 339 * The logic does not apply to gitlinks, as ce_match_stat_basic() 340 * already has checked the actual HEAD from the filesystem in the 341 * subproject. If ie_match_stat() already said it is different, 342 * then we know it is. 343 */ 344 if ((changed & DATA_CHANGED) && 345 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0)) 346 return changed; 347 348 changed_fs = ce_modified_check_fs(ce, st); 349 if (changed_fs) 350 return changed | changed_fs; 351 return 0; 352} 353 354int base_name_compare(const char *name1, int len1, int mode1, 355 const char *name2, int len2, int mode2) 356{ 357 unsigned char c1, c2; 358 int len = len1 < len2 ? len1 : len2; 359 int cmp; 360 361 cmp = memcmp(name1, name2, len); 362 if (cmp) 363 return cmp; 364 c1 = name1[len]; 365 c2 = name2[len]; 366 if (!c1 && S_ISDIR(mode1)) 367 c1 = '/'; 368 if (!c2 && S_ISDIR(mode2)) 369 c2 = '/'; 370 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0; 371} 372 373/* 374 * df_name_compare() is identical to base_name_compare(), except it 375 * compares conflicting directory/file entries as equal. Note that 376 * while a directory name compares as equal to a regular file, they 377 * then individually compare _differently_ to a filename that has 378 * a dot after the basename (because '\0' < '.' < '/'). 379 * 380 * This is used by routines that want to traverse the git namespace 381 * but then handle conflicting entries together when possible. 382 */ 383int df_name_compare(const char *name1, int len1, int mode1, 384 const char *name2, int len2, int mode2) 385{ 386 int len = len1 < len2 ? len1 : len2, cmp; 387 unsigned char c1, c2; 388 389 cmp = memcmp(name1, name2, len); 390 if (cmp) 391 return cmp; 392 /* Directories and files compare equal (same length, same name) */ 393 if (len1 == len2) 394 return 0; 395 c1 = name1[len]; 396 if (!c1 && S_ISDIR(mode1)) 397 c1 = '/'; 398 c2 = name2[len]; 399 if (!c2 && S_ISDIR(mode2)) 400 c2 = '/'; 401 if (c1 == '/' && !c2) 402 return 0; 403 if (c2 == '/' && !c1) 404 return 0; 405 return c1 - c2; 406} 407 408int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2) 409{ 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 if (stage1 < stage2) 422 return -1; 423 if (stage1 > stage2) 424 return 1; 425 return 0; 426} 427 428int cache_name_compare(const char *name1, int len1, const char *name2, int len2) 429{ 430 return cache_name_stage_compare(name1, len1, 0, name2, len2, 0); 431} 432 433static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage) 434{ 435 int first, last; 436 437 first = 0; 438 last = istate->cache_nr; 439 while (last > first) { 440 int next = (last + first) >> 1; 441 struct cache_entry *ce = istate->cache[next]; 442 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce)); 443 if (!cmp) 444 return next; 445 if (cmp < 0) { 446 last = next; 447 continue; 448 } 449 first = next+1; 450 } 451 return -first-1; 452} 453 454int index_name_pos(const struct index_state *istate, const char *name, int namelen) 455{ 456 return index_name_stage_pos(istate, name, namelen, 0); 457} 458 459/* Remove entry, return true if there are more entries to go.. */ 460int remove_index_entry_at(struct index_state *istate, int pos) 461{ 462 struct cache_entry *ce = istate->cache[pos]; 463 464 record_resolve_undo(istate, ce); 465 remove_name_hash(istate, ce); 466 istate->cache_changed = 1; 467 istate->cache_nr--; 468 if (pos >= istate->cache_nr) 469 return 0; 470 memmove(istate->cache + pos, 471 istate->cache + pos + 1, 472 (istate->cache_nr - pos) * sizeof(struct cache_entry *)); 473 return 1; 474} 475 476/* 477 * Remove all cache ententries marked for removal, that is where 478 * CE_REMOVE is set in ce_flags. This is much more effective than 479 * calling remove_index_entry_at() for each entry to be removed. 480 */ 481void remove_marked_cache_entries(struct index_state *istate) 482{ 483 struct cache_entry **ce_array = istate->cache; 484 unsigned int i, j; 485 486 for (i = j = 0; i < istate->cache_nr; i++) { 487 if (ce_array[i]->ce_flags & CE_REMOVE) 488 remove_name_hash(istate, ce_array[i]); 489 else 490 ce_array[j++] = ce_array[i]; 491 } 492 istate->cache_changed = 1; 493 istate->cache_nr = j; 494} 495 496int remove_file_from_index(struct index_state *istate, const char *path) 497{ 498 int pos = index_name_pos(istate, path, strlen(path)); 499 if (pos < 0) 500 pos = -pos-1; 501 cache_tree_invalidate_path(istate->cache_tree, path); 502 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 503 remove_index_entry_at(istate, pos); 504 return 0; 505} 506 507static int compare_name(struct cache_entry *ce, const char *path, int namelen) 508{ 509 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); 510} 511 512static int index_name_pos_also_unmerged(struct index_state *istate, 513 const char *path, int namelen) 514{ 515 int pos = index_name_pos(istate, path, namelen); 516 struct cache_entry *ce; 517 518 if (pos >= 0) 519 return pos; 520 521 /* maybe unmerged? */ 522 pos = -1 - pos; 523 if (pos >= istate->cache_nr || 524 compare_name((ce = istate->cache[pos]), path, namelen)) 525 return -1; 526 527 /* order of preference: stage 2, 1, 3 */ 528 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && 529 ce_stage((ce = istate->cache[pos + 1])) == 2 && 530 !compare_name(ce, path, namelen)) 531 pos++; 532 return pos; 533} 534 535static int different_name(struct cache_entry *ce, struct cache_entry *alias) 536{ 537 int len = ce_namelen(ce); 538 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len); 539} 540 541/* 542 * If we add a filename that aliases in the cache, we will use the 543 * name that we already have - but we don't want to update the same 544 * alias twice, because that implies that there were actually two 545 * different files with aliasing names! 546 * 547 * So we use the CE_ADDED flag to verify that the alias was an old 548 * one before we accept it as 549 */ 550static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias) 551{ 552 int len; 553 struct cache_entry *new; 554 555 if (alias->ce_flags & CE_ADDED) 556 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); 557 558 /* Ok, create the new entry using the name of the existing alias */ 559 len = ce_namelen(alias); 560 new = xcalloc(1, cache_entry_size(len)); 561 memcpy(new->name, alias->name, len); 562 copy_cache_entry(new, ce); 563 free(ce); 564 return new; 565} 566 567static void record_intent_to_add(struct cache_entry *ce) 568{ 569 unsigned char sha1[20]; 570 if (write_sha1_file("", 0, blob_type, sha1)) 571 die("cannot create an empty blob in the object database"); 572 hashcpy(ce->sha1, sha1); 573} 574 575int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags) 576{ 577 int size, namelen, was_same; 578 mode_t st_mode = st->st_mode; 579 struct cache_entry *ce, *alias; 580 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY; 581 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); 582 int pretend = flags & ADD_CACHE_PRETEND; 583 int intent_only = flags & ADD_CACHE_INTENT; 584 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| 585 (intent_only ? ADD_CACHE_NEW_ONLY : 0)); 586 587 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 588 return error("%s: can only add regular files, symbolic links or git-directories", path); 589 590 namelen = strlen(path); 591 if (S_ISDIR(st_mode)) { 592 while (namelen && path[namelen-1] == '/') 593 namelen--; 594 } 595 size = cache_entry_size(namelen); 596 ce = xcalloc(1, size); 597 memcpy(ce->name, path, namelen); 598 ce->ce_namelen = namelen; 599 if (!intent_only) 600 fill_stat_cache_info(ce, st); 601 else 602 ce->ce_flags |= CE_INTENT_TO_ADD; 603 604 if (trust_executable_bit && has_symlinks) 605 ce->ce_mode = create_ce_mode(st_mode); 606 else { 607 /* If there is an existing entry, pick the mode bits and type 608 * from it, otherwise assume unexecutable regular file. 609 */ 610 struct cache_entry *ent; 611 int pos = index_name_pos_also_unmerged(istate, path, namelen); 612 613 ent = (0 <= pos) ? istate->cache[pos] : NULL; 614 ce->ce_mode = ce_mode_from_stat(ent, st_mode); 615 } 616 617 /* When core.ignorecase=true, determine if a directory of the same name but differing 618 * case already exists within the Git repository. If it does, ensure the directory 619 * case of the file being added to the repository matches (is folded into) the existing 620 * entry's directory case. 621 */ 622 if (ignore_case) { 623 const char *startPtr = ce->name; 624 const char *ptr = startPtr; 625 while (*ptr) { 626 while (*ptr && *ptr != '/') 627 ++ptr; 628 if (*ptr == '/') { 629 struct cache_entry *foundce; 630 ++ptr; 631 foundce = index_name_exists(istate, ce->name, ptr - ce->name, ignore_case); 632 if (foundce) { 633 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr); 634 startPtr = ptr; 635 } 636 } 637 } 638 } 639 640 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case); 641 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) { 642 /* Nothing changed, really */ 643 free(ce); 644 if (!S_ISGITLINK(alias->ce_mode)) 645 ce_mark_uptodate(alias); 646 alias->ce_flags |= CE_ADDED; 647 return 0; 648 } 649 if (!intent_only) { 650 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT)) 651 return error("unable to index file %s", path); 652 } else 653 record_intent_to_add(ce); 654 655 if (ignore_case && alias && different_name(ce, alias)) 656 ce = create_alias_ce(ce, alias); 657 ce->ce_flags |= CE_ADDED; 658 659 /* It was suspected to be racily clean, but it turns out to be Ok */ 660 was_same = (alias && 661 !ce_stage(alias) && 662 !hashcmp(alias->sha1, ce->sha1) && 663 ce->ce_mode == alias->ce_mode); 664 665 if (pretend) 666 ; 667 else if (add_index_entry(istate, ce, add_option)) 668 return error("unable to add %s to index",path); 669 if (verbose && !was_same) 670 printf("add '%s'\n", path); 671 return 0; 672} 673 674int add_file_to_index(struct index_state *istate, const char *path, int flags) 675{ 676 struct stat st; 677 if (lstat(path, &st)) 678 die_errno("unable to stat '%s'", path); 679 return add_to_index(istate, path, &st, flags); 680} 681 682struct cache_entry *make_cache_entry(unsigned int mode, 683 const unsigned char *sha1, const char *path, int stage, 684 int refresh) 685{ 686 int size, len; 687 struct cache_entry *ce; 688 689 if (!verify_path(path)) { 690 error("Invalid path '%s'", path); 691 return NULL; 692 } 693 694 len = strlen(path); 695 size = cache_entry_size(len); 696 ce = xcalloc(1, size); 697 698 hashcpy(ce->sha1, sha1); 699 memcpy(ce->name, path, len); 700 ce->ce_flags = create_ce_flags(stage); 701 ce->ce_namelen = len; 702 ce->ce_mode = create_ce_mode(mode); 703 704 if (refresh) 705 return refresh_cache_entry(ce, 0); 706 707 return ce; 708} 709 710int ce_same_name(struct cache_entry *a, struct cache_entry *b) 711{ 712 int len = ce_namelen(a); 713 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 714} 715 716int ce_path_match(const struct cache_entry *ce, const struct pathspec *pathspec) 717{ 718 return match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, NULL); 719} 720 721/* 722 * We fundamentally don't like some paths: we don't want 723 * dot or dot-dot anywhere, and for obvious reasons don't 724 * want to recurse into ".git" either. 725 * 726 * Also, we don't want double slashes or slashes at the 727 * end that can make pathnames ambiguous. 728 */ 729static int verify_dotfile(const char *rest) 730{ 731 /* 732 * The first character was '.', but that 733 * has already been discarded, we now test 734 * the rest. 735 */ 736 737 /* "." is not allowed */ 738 if (*rest == '\0' || is_dir_sep(*rest)) 739 return 0; 740 741 switch (*rest) { 742 /* 743 * ".git" followed by NUL or slash is bad. This 744 * shares the path end test with the ".." case. 745 */ 746 case 'g': 747 if (rest[1] != 'i') 748 break; 749 if (rest[2] != 't') 750 break; 751 rest += 2; 752 /* fallthrough */ 753 case '.': 754 if (rest[1] == '\0' || is_dir_sep(rest[1])) 755 return 0; 756 } 757 return 1; 758} 759 760int verify_path(const char *path) 761{ 762 char c; 763 764 if (has_dos_drive_prefix(path)) 765 return 0; 766 767 goto inside; 768 for (;;) { 769 if (!c) 770 return 1; 771 if (is_dir_sep(c)) { 772inside: 773 c = *path++; 774 if ((c == '.' && !verify_dotfile(path)) || 775 is_dir_sep(c) || c == '\0') 776 return 0; 777 } 778 c = *path++; 779 } 780} 781 782/* 783 * Do we have another file that has the beginning components being a 784 * proper superset of the name we're trying to add? 785 */ 786static int has_file_name(struct index_state *istate, 787 const struct cache_entry *ce, int pos, int ok_to_replace) 788{ 789 int retval = 0; 790 int len = ce_namelen(ce); 791 int stage = ce_stage(ce); 792 const char *name = ce->name; 793 794 while (pos < istate->cache_nr) { 795 struct cache_entry *p = istate->cache[pos++]; 796 797 if (len >= ce_namelen(p)) 798 break; 799 if (memcmp(name, p->name, len)) 800 break; 801 if (ce_stage(p) != stage) 802 continue; 803 if (p->name[len] != '/') 804 continue; 805 if (p->ce_flags & CE_REMOVE) 806 continue; 807 retval = -1; 808 if (!ok_to_replace) 809 break; 810 remove_index_entry_at(istate, --pos); 811 } 812 return retval; 813} 814 815/* 816 * Do we have another file with a pathname that is a proper 817 * subset of the name we're trying to add? 818 */ 819static int has_dir_name(struct index_state *istate, 820 const struct cache_entry *ce, int pos, int ok_to_replace) 821{ 822 int retval = 0; 823 int stage = ce_stage(ce); 824 const char *name = ce->name; 825 const char *slash = name + ce_namelen(ce); 826 827 for (;;) { 828 int len; 829 830 for (;;) { 831 if (*--slash == '/') 832 break; 833 if (slash <= ce->name) 834 return retval; 835 } 836 len = slash - name; 837 838 pos = index_name_stage_pos(istate, name, len, stage); 839 if (pos >= 0) { 840 /* 841 * Found one, but not so fast. This could 842 * be a marker that says "I was here, but 843 * I am being removed". Such an entry is 844 * not a part of the resulting tree, and 845 * it is Ok to have a directory at the same 846 * path. 847 */ 848 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) { 849 retval = -1; 850 if (!ok_to_replace) 851 break; 852 remove_index_entry_at(istate, pos); 853 continue; 854 } 855 } 856 else 857 pos = -pos-1; 858 859 /* 860 * Trivial optimization: if we find an entry that 861 * already matches the sub-directory, then we know 862 * we're ok, and we can exit. 863 */ 864 while (pos < istate->cache_nr) { 865 struct cache_entry *p = istate->cache[pos]; 866 if ((ce_namelen(p) <= len) || 867 (p->name[len] != '/') || 868 memcmp(p->name, name, len)) 869 break; /* not our subdirectory */ 870 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE)) 871 /* 872 * p is at the same stage as our entry, and 873 * is a subdirectory of what we are looking 874 * at, so we cannot have conflicts at our 875 * level or anything shorter. 876 */ 877 return retval; 878 pos++; 879 } 880 } 881 return retval; 882} 883 884/* We may be in a situation where we already have path/file and path 885 * is being added, or we already have path and path/file is being 886 * added. Either one would result in a nonsense tree that has path 887 * twice when git-write-tree tries to write it out. Prevent it. 888 * 889 * If ok-to-replace is specified, we remove the conflicting entries 890 * from the cache so the caller should recompute the insert position. 891 * When this happens, we return non-zero. 892 */ 893static int check_file_directory_conflict(struct index_state *istate, 894 const struct cache_entry *ce, 895 int pos, int ok_to_replace) 896{ 897 int retval; 898 899 /* 900 * When ce is an "I am going away" entry, we allow it to be added 901 */ 902 if (ce->ce_flags & CE_REMOVE) 903 return 0; 904 905 /* 906 * We check if the path is a sub-path of a subsequent pathname 907 * first, since removing those will not change the position 908 * in the array. 909 */ 910 retval = has_file_name(istate, ce, pos, ok_to_replace); 911 912 /* 913 * Then check if the path might have a clashing sub-directory 914 * before it. 915 */ 916 return retval + has_dir_name(istate, ce, pos, ok_to_replace); 917} 918 919static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option) 920{ 921 int pos; 922 int ok_to_add = option & ADD_CACHE_OK_TO_ADD; 923 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE; 924 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK; 925 int new_only = option & ADD_CACHE_NEW_ONLY; 926 927 cache_tree_invalidate_path(istate->cache_tree, ce->name); 928 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce)); 929 930 /* existing match? Just replace it. */ 931 if (pos >= 0) { 932 if (!new_only) 933 replace_index_entry(istate, pos, ce); 934 return 0; 935 } 936 pos = -pos-1; 937 938 /* 939 * Inserting a merged entry ("stage 0") into the index 940 * will always replace all non-merged entries.. 941 */ 942 if (pos < istate->cache_nr && ce_stage(ce) == 0) { 943 while (ce_same_name(istate->cache[pos], ce)) { 944 ok_to_add = 1; 945 if (!remove_index_entry_at(istate, pos)) 946 break; 947 } 948 } 949 950 if (!ok_to_add) 951 return -1; 952 if (!verify_path(ce->name)) 953 return error("Invalid path '%s'", ce->name); 954 955 if (!skip_df_check && 956 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) { 957 if (!ok_to_replace) 958 return error("'%s' appears as both a file and as a directory", 959 ce->name); 960 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce)); 961 pos = -pos-1; 962 } 963 return pos + 1; 964} 965 966int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option) 967{ 968 int pos; 969 970 if (option & ADD_CACHE_JUST_APPEND) 971 pos = istate->cache_nr; 972 else { 973 int ret; 974 ret = add_index_entry_with_check(istate, ce, option); 975 if (ret <= 0) 976 return ret; 977 pos = ret - 1; 978 } 979 980 /* Make sure the array is big enough .. */ 981 if (istate->cache_nr == istate->cache_alloc) { 982 istate->cache_alloc = alloc_nr(istate->cache_alloc); 983 istate->cache = xrealloc(istate->cache, 984 istate->cache_alloc * sizeof(*istate->cache)); 985 } 986 987 /* Add it in.. */ 988 istate->cache_nr++; 989 if (istate->cache_nr > pos + 1) 990 memmove(istate->cache + pos + 1, 991 istate->cache + pos, 992 (istate->cache_nr - pos - 1) * sizeof(ce)); 993 set_index_entry(istate, pos, ce); 994 istate->cache_changed = 1; 995 return 0; 996} 997 998/* 999 * "refresh" does not calculate a new sha1 file or bring the1000 * cache up-to-date for mode/content changes. But what it1001 * _does_ do is to "re-match" the stat information of a file1002 * with the cache, so that you can refresh the cache for a1003 * file that hasn't been changed but where the stat entry is1004 * out of date.1005 *1006 * For example, you'd want to do this after doing a "git-read-tree",1007 * to link up the stat cache details with the proper files.1008 */1009static struct cache_entry *refresh_cache_ent(struct index_state *istate,1010 struct cache_entry *ce,1011 unsigned int options, int *err,1012 int *changed_ret)1013{1014 struct stat st;1015 struct cache_entry *updated;1016 int changed, size;1017 int ignore_valid = options & CE_MATCH_IGNORE_VALID;1018 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;10191020 if (ce_uptodate(ce))1021 return ce;10221023 /*1024 * CE_VALID or CE_SKIP_WORKTREE means the user promised us1025 * that the change to the work tree does not matter and told1026 * us not to worry.1027 */1028 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {1029 ce_mark_uptodate(ce);1030 return ce;1031 }1032 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {1033 ce_mark_uptodate(ce);1034 return ce;1035 }10361037 if (lstat(ce->name, &st) < 0) {1038 if (err)1039 *err = errno;1040 return NULL;1041 }10421043 changed = ie_match_stat(istate, ce, &st, options);1044 if (changed_ret)1045 *changed_ret = changed;1046 if (!changed) {1047 /*1048 * The path is unchanged. If we were told to ignore1049 * valid bit, then we did the actual stat check and1050 * found that the entry is unmodified. If the entry1051 * is not marked VALID, this is the place to mark it1052 * valid again, under "assume unchanged" mode.1053 */1054 if (ignore_valid && assume_unchanged &&1055 !(ce->ce_flags & CE_VALID))1056 ; /* mark this one VALID again */1057 else {1058 /*1059 * We do not mark the index itself "modified"1060 * because CE_UPTODATE flag is in-core only;1061 * we are not going to write this change out.1062 */1063 if (!S_ISGITLINK(ce->ce_mode))1064 ce_mark_uptodate(ce);1065 return ce;1066 }1067 }10681069 if (ie_modified(istate, ce, &st, options)) {1070 if (err)1071 *err = EINVAL;1072 return NULL;1073 }10741075 size = ce_size(ce);1076 updated = xmalloc(size);1077 memcpy(updated, ce, size);1078 fill_stat_cache_info(updated, &st);1079 /*1080 * If ignore_valid is not set, we should leave CE_VALID bit1081 * alone. Otherwise, paths marked with --no-assume-unchanged1082 * (i.e. things to be edited) will reacquire CE_VALID bit1083 * automatically, which is not really what we want.1084 */1085 if (!ignore_valid && assume_unchanged &&1086 !(ce->ce_flags & CE_VALID))1087 updated->ce_flags &= ~CE_VALID;10881089 return updated;1090}10911092static void show_file(const char * fmt, const char * name, int in_porcelain,1093 int * first, const char *header_msg)1094{1095 if (in_porcelain && *first && header_msg) {1096 printf("%s\n", header_msg);1097 *first = 0;1098 }1099 printf(fmt, name);1100}11011102int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,1103 char *seen, const char *header_msg)1104{1105 int i;1106 int has_errors = 0;1107 int really = (flags & REFRESH_REALLY) != 0;1108 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;1109 int quiet = (flags & REFRESH_QUIET) != 0;1110 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;1111 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;1112 int first = 1;1113 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);1114 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;1115 const char *modified_fmt;1116 const char *deleted_fmt;1117 const char *typechange_fmt;1118 const char *added_fmt;1119 const char *unmerged_fmt;11201121 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");1122 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");1123 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");1124 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");1125 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");1126 for (i = 0; i < istate->cache_nr; i++) {1127 struct cache_entry *ce, *new;1128 int cache_errno = 0;1129 int changed = 0;1130 int filtered = 0;11311132 ce = istate->cache[i];1133 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))1134 continue;11351136 if (pathspec &&1137 !match_pathspec(pathspec, ce->name, ce_namelen(ce), 0, seen))1138 filtered = 1;11391140 if (ce_stage(ce)) {1141 while ((i < istate->cache_nr) &&1142 ! strcmp(istate->cache[i]->name, ce->name))1143 i++;1144 i--;1145 if (allow_unmerged)1146 continue;1147 if (!filtered)1148 show_file(unmerged_fmt, ce->name, in_porcelain,1149 &first, header_msg);1150 has_errors = 1;1151 continue;1152 }11531154 if (filtered)1155 continue;11561157 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);1158 if (new == ce)1159 continue;1160 if (!new) {1161 const char *fmt;11621163 if (not_new && cache_errno == ENOENT)1164 continue;1165 if (really && cache_errno == EINVAL) {1166 /* If we are doing --really-refresh that1167 * means the index is not valid anymore.1168 */1169 ce->ce_flags &= ~CE_VALID;1170 istate->cache_changed = 1;1171 }1172 if (quiet)1173 continue;11741175 if (cache_errno == ENOENT)1176 fmt = deleted_fmt;1177 else if (ce->ce_flags & CE_INTENT_TO_ADD)1178 fmt = added_fmt; /* must be before other checks */1179 else if (changed & TYPE_CHANGED)1180 fmt = typechange_fmt;1181 else1182 fmt = modified_fmt;1183 show_file(fmt,1184 ce->name, in_porcelain, &first, header_msg);1185 has_errors = 1;1186 continue;1187 }11881189 replace_index_entry(istate, i, new);1190 }1191 return has_errors;1192}11931194static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)1195{1196 return refresh_cache_ent(&the_index, ce, really, NULL, NULL);1197}119811991200/*****************************************************************1201 * Index File I/O1202 *****************************************************************/12031204#define INDEX_FORMAT_DEFAULT 312051206/*1207 * dev/ino/uid/gid/size are also just tracked to the low 32 bits1208 * Again - this is just a (very strong in practice) heuristic that1209 * the inode hasn't changed.1210 *1211 * We save the fields in big-endian order to allow using the1212 * index file over NFS transparently.1213 */1214struct ondisk_cache_entry {1215 struct cache_time ctime;1216 struct cache_time mtime;1217 unsigned int dev;1218 unsigned int ino;1219 unsigned int mode;1220 unsigned int uid;1221 unsigned int gid;1222 unsigned int size;1223 unsigned char sha1[20];1224 unsigned short flags;1225 char name[FLEX_ARRAY]; /* more */1226};12271228/*1229 * This struct is used when CE_EXTENDED bit is 11230 * The struct must match ondisk_cache_entry exactly from1231 * ctime till flags1232 */1233struct ondisk_cache_entry_extended {1234 struct cache_time ctime;1235 struct cache_time mtime;1236 unsigned int dev;1237 unsigned int ino;1238 unsigned int mode;1239 unsigned int uid;1240 unsigned int gid;1241 unsigned int size;1242 unsigned char sha1[20];1243 unsigned short flags;1244 unsigned short flags2;1245 char name[FLEX_ARRAY]; /* more */1246};12471248/* These are only used for v3 or lower */1249#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)1250#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)1251#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)1252#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \1253 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \1254 ondisk_cache_entry_size(ce_namelen(ce)))12551256static int verify_hdr(struct cache_header *hdr, unsigned long size)1257{1258 git_SHA_CTX c;1259 unsigned char sha1[20];1260 int hdr_version;12611262 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))1263 return error("bad signature");1264 hdr_version = ntohl(hdr->hdr_version);1265 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)1266 return error("bad index version %d", hdr_version);1267 git_SHA1_Init(&c);1268 git_SHA1_Update(&c, hdr, size - 20);1269 git_SHA1_Final(sha1, &c);1270 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))1271 return error("bad index file sha1 signature");1272 return 0;1273}12741275static int read_index_extension(struct index_state *istate,1276 const char *ext, void *data, unsigned long sz)1277{1278 switch (CACHE_EXT(ext)) {1279 case CACHE_EXT_TREE:1280 istate->cache_tree = cache_tree_read(data, sz);1281 break;1282 case CACHE_EXT_RESOLVE_UNDO:1283 istate->resolve_undo = resolve_undo_read(data, sz);1284 break;1285 default:1286 if (*ext < 'A' || 'Z' < *ext)1287 return error("index uses %.4s extension, which we do not understand",1288 ext);1289 fprintf(stderr, "ignoring %.4s extension\n", ext);1290 break;1291 }1292 return 0;1293}12941295int read_index(struct index_state *istate)1296{1297 return read_index_from(istate, get_index_file());1298}12991300#ifndef NEEDS_ALIGNED_ACCESS1301#define ntoh_s(var) ntohs(var)1302#define ntoh_l(var) ntohl(var)1303#else1304static inline uint16_t ntoh_s_force_align(void *p)1305{1306 uint16_t x;1307 memcpy(&x, p, sizeof(x));1308 return ntohs(x);1309}1310static inline uint32_t ntoh_l_force_align(void *p)1311{1312 uint32_t x;1313 memcpy(&x, p, sizeof(x));1314 return ntohl(x);1315}1316#define ntoh_s(var) ntoh_s_force_align(&(var))1317#define ntoh_l(var) ntoh_l_force_align(&(var))1318#endif13191320static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,1321 unsigned int flags,1322 const char *name,1323 size_t len)1324{1325 struct cache_entry *ce = xmalloc(cache_entry_size(len));13261327 ce->ce_ctime.sec = ntoh_l(ondisk->ctime.sec);1328 ce->ce_mtime.sec = ntoh_l(ondisk->mtime.sec);1329 ce->ce_ctime.nsec = ntoh_l(ondisk->ctime.nsec);1330 ce->ce_mtime.nsec = ntoh_l(ondisk->mtime.nsec);1331 ce->ce_dev = ntoh_l(ondisk->dev);1332 ce->ce_ino = ntoh_l(ondisk->ino);1333 ce->ce_mode = ntoh_l(ondisk->mode);1334 ce->ce_uid = ntoh_l(ondisk->uid);1335 ce->ce_gid = ntoh_l(ondisk->gid);1336 ce->ce_size = ntoh_l(ondisk->size);1337 ce->ce_flags = flags & ~CE_NAMEMASK;1338 ce->ce_namelen = len;1339 hashcpy(ce->sha1, ondisk->sha1);1340 memcpy(ce->name, name, len);1341 ce->name[len] = '\0';1342 return ce;1343}13441345/*1346 * Adjacent cache entries tend to share the leading paths, so it makes1347 * sense to only store the differences in later entries. In the v41348 * on-disk format of the index, each on-disk cache entry stores the1349 * number of bytes to be stripped from the end of the previous name,1350 * and the bytes to append to the result, to come up with its name.1351 */1352static unsigned long expand_name_field(struct strbuf *name, const char *cp_)1353{1354 const unsigned char *ep, *cp = (const unsigned char *)cp_;1355 size_t len = decode_varint(&cp);13561357 if (name->len < len)1358 die("malformed name field in the index");1359 strbuf_remove(name, name->len - len, len);1360 for (ep = cp; *ep; ep++)1361 ; /* find the end */1362 strbuf_add(name, cp, ep - cp);1363 return (const char *)ep + 1 - cp_;1364}13651366static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,1367 unsigned long *ent_size,1368 struct strbuf *previous_name)1369{1370 struct cache_entry *ce;1371 size_t len;1372 const char *name;1373 unsigned int flags;13741375 /* On-disk flags are just 16 bits */1376 flags = ntoh_s(ondisk->flags);1377 len = flags & CE_NAMEMASK;13781379 if (flags & CE_EXTENDED) {1380 struct ondisk_cache_entry_extended *ondisk2;1381 int extended_flags;1382 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1383 extended_flags = ntoh_s(ondisk2->flags2) << 16;1384 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1385 if (extended_flags & ~CE_EXTENDED_FLAGS)1386 die("Unknown index entry format %08x", extended_flags);1387 flags |= extended_flags;1388 name = ondisk2->name;1389 }1390 else1391 name = ondisk->name;13921393 if (!previous_name) {1394 /* v3 and earlier */1395 if (len == CE_NAMEMASK)1396 len = strlen(name);1397 ce = cache_entry_from_ondisk(ondisk, flags, name, len);13981399 *ent_size = ondisk_ce_size(ce);1400 } else {1401 unsigned long consumed;1402 consumed = expand_name_field(previous_name, name);1403 ce = cache_entry_from_ondisk(ondisk, flags,1404 previous_name->buf,1405 previous_name->len);14061407 *ent_size = (name - ((char *)ondisk)) + consumed;1408 }1409 return ce;1410}14111412/* remember to discard_cache() before reading a different cache! */1413int read_index_from(struct index_state *istate, const char *path)1414{1415 int fd, i;1416 struct stat st;1417 unsigned long src_offset;1418 struct cache_header *hdr;1419 void *mmap;1420 size_t mmap_size;1421 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;14221423 if (istate->initialized)1424 return istate->cache_nr;14251426 istate->timestamp.sec = 0;1427 istate->timestamp.nsec = 0;1428 fd = open(path, O_RDONLY);1429 if (fd < 0) {1430 if (errno == ENOENT)1431 return 0;1432 die_errno("index file open failed");1433 }14341435 if (fstat(fd, &st))1436 die_errno("cannot stat the open index");14371438 mmap_size = xsize_t(st.st_size);1439 if (mmap_size < sizeof(struct cache_header) + 20)1440 die("index file smaller than expected");14411442 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);1443 if (mmap == MAP_FAILED)1444 die_errno("unable to map index file");1445 close(fd);14461447 hdr = mmap;1448 if (verify_hdr(hdr, mmap_size) < 0)1449 goto unmap;14501451 istate->version = ntohl(hdr->hdr_version);1452 istate->cache_nr = ntohl(hdr->hdr_entries);1453 istate->cache_alloc = alloc_nr(istate->cache_nr);1454 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));1455 istate->initialized = 1;14561457 if (istate->version == 4)1458 previous_name = &previous_name_buf;1459 else1460 previous_name = NULL;14611462 src_offset = sizeof(*hdr);1463 for (i = 0; i < istate->cache_nr; i++) {1464 struct ondisk_cache_entry *disk_ce;1465 struct cache_entry *ce;1466 unsigned long consumed;14671468 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);1469 ce = create_from_disk(disk_ce, &consumed, previous_name);1470 set_index_entry(istate, i, ce);14711472 src_offset += consumed;1473 }1474 strbuf_release(&previous_name_buf);1475 istate->timestamp.sec = st.st_mtime;1476 istate->timestamp.nsec = ST_MTIME_NSEC(st);14771478 while (src_offset <= mmap_size - 20 - 8) {1479 /* After an array of active_nr index entries,1480 * there can be arbitrary number of extended1481 * sections, each of which is prefixed with1482 * extension name (4-byte) and section length1483 * in 4-byte network byte order.1484 */1485 uint32_t extsize;1486 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);1487 extsize = ntohl(extsize);1488 if (read_index_extension(istate,1489 (const char *) mmap + src_offset,1490 (char *) mmap + src_offset + 8,1491 extsize) < 0)1492 goto unmap;1493 src_offset += 8;1494 src_offset += extsize;1495 }1496 munmap(mmap, mmap_size);1497 return istate->cache_nr;14981499unmap:1500 munmap(mmap, mmap_size);1501 die("index file corrupt");1502}15031504int is_index_unborn(struct index_state *istate)1505{1506 return (!istate->cache_nr && !istate->timestamp.sec);1507}15081509int discard_index(struct index_state *istate)1510{1511 int i;15121513 for (i = 0; i < istate->cache_nr; i++)1514 free(istate->cache[i]);1515 resolve_undo_clear_index(istate);1516 istate->cache_nr = 0;1517 istate->cache_changed = 0;1518 istate->timestamp.sec = 0;1519 istate->timestamp.nsec = 0;1520 free_name_hash(istate);1521 cache_tree_free(&(istate->cache_tree));1522 istate->initialized = 0;15231524 /* no need to throw away allocated active_cache */1525 return 0;1526}15271528int unmerged_index(const struct index_state *istate)1529{1530 int i;1531 for (i = 0; i < istate->cache_nr; i++) {1532 if (ce_stage(istate->cache[i]))1533 return 1;1534 }1535 return 0;1536}15371538#define WRITE_BUFFER_SIZE 81921539static unsigned char write_buffer[WRITE_BUFFER_SIZE];1540static unsigned long write_buffer_len;15411542static int ce_write_flush(git_SHA_CTX *context, int fd)1543{1544 unsigned int buffered = write_buffer_len;1545 if (buffered) {1546 git_SHA1_Update(context, write_buffer, buffered);1547 if (write_in_full(fd, write_buffer, buffered) != buffered)1548 return -1;1549 write_buffer_len = 0;1550 }1551 return 0;1552}15531554static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)1555{1556 while (len) {1557 unsigned int buffered = write_buffer_len;1558 unsigned int partial = WRITE_BUFFER_SIZE - buffered;1559 if (partial > len)1560 partial = len;1561 memcpy(write_buffer + buffered, data, partial);1562 buffered += partial;1563 if (buffered == WRITE_BUFFER_SIZE) {1564 write_buffer_len = buffered;1565 if (ce_write_flush(context, fd))1566 return -1;1567 buffered = 0;1568 }1569 write_buffer_len = buffered;1570 len -= partial;1571 data = (char *) data + partial;1572 }1573 return 0;1574}15751576static int write_index_ext_header(git_SHA_CTX *context, int fd,1577 unsigned int ext, unsigned int sz)1578{1579 ext = htonl(ext);1580 sz = htonl(sz);1581 return ((ce_write(context, fd, &ext, 4) < 0) ||1582 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;1583}15841585static int ce_flush(git_SHA_CTX *context, int fd)1586{1587 unsigned int left = write_buffer_len;15881589 if (left) {1590 write_buffer_len = 0;1591 git_SHA1_Update(context, write_buffer, left);1592 }15931594 /* Flush first if not enough space for SHA1 signature */1595 if (left + 20 > WRITE_BUFFER_SIZE) {1596 if (write_in_full(fd, write_buffer, left) != left)1597 return -1;1598 left = 0;1599 }16001601 /* Append the SHA1 signature at the end */1602 git_SHA1_Final(write_buffer + left, context);1603 left += 20;1604 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;1605}16061607static void ce_smudge_racily_clean_entry(struct cache_entry *ce)1608{1609 /*1610 * The only thing we care about in this function is to smudge the1611 * falsely clean entry due to touch-update-touch race, so we leave1612 * everything else as they are. We are called for entries whose1613 * ce_mtime match the index file mtime.1614 *1615 * Note that this actually does not do much for gitlinks, for1616 * which ce_match_stat_basic() always goes to the actual1617 * contents. The caller checks with is_racy_timestamp() which1618 * always says "no" for gitlinks, so we are not called for them ;-)1619 */1620 struct stat st;16211622 if (lstat(ce->name, &st) < 0)1623 return;1624 if (ce_match_stat_basic(ce, &st))1625 return;1626 if (ce_modified_check_fs(ce, &st)) {1627 /* This is "racily clean"; smudge it. Note that this1628 * is a tricky code. At first glance, it may appear1629 * that it can break with this sequence:1630 *1631 * $ echo xyzzy >frotz1632 * $ git-update-index --add frotz1633 * $ : >frotz1634 * $ sleep 31635 * $ echo filfre >nitfol1636 * $ git-update-index --add nitfol1637 *1638 * but it does not. When the second update-index runs,1639 * it notices that the entry "frotz" has the same timestamp1640 * as index, and if we were to smudge it by resetting its1641 * size to zero here, then the object name recorded1642 * in index is the 6-byte file but the cached stat information1643 * becomes zero --- which would then match what we would1644 * obtain from the filesystem next time we stat("frotz").1645 *1646 * However, the second update-index, before calling1647 * this function, notices that the cached size is 61648 * bytes and what is on the filesystem is an empty1649 * file, and never calls us, so the cached size information1650 * for "frotz" stays 6 which does not match the filesystem.1651 */1652 ce->ce_size = 0;1653 }1654}16551656/* Copy miscellaneous fields but not the name */1657static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,1658 struct cache_entry *ce)1659{1660 short flags;16611662 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);1663 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);1664 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);1665 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);1666 ondisk->dev = htonl(ce->ce_dev);1667 ondisk->ino = htonl(ce->ce_ino);1668 ondisk->mode = htonl(ce->ce_mode);1669 ondisk->uid = htonl(ce->ce_uid);1670 ondisk->gid = htonl(ce->ce_gid);1671 ondisk->size = htonl(ce->ce_size);1672 hashcpy(ondisk->sha1, ce->sha1);16731674 flags = ce->ce_flags;1675 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));1676 ondisk->flags = htons(flags);1677 if (ce->ce_flags & CE_EXTENDED) {1678 struct ondisk_cache_entry_extended *ondisk2;1679 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1680 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);1681 return ondisk2->name;1682 }1683 else {1684 return ondisk->name;1685 }1686}16871688static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,1689 struct strbuf *previous_name)1690{1691 int size;1692 struct ondisk_cache_entry *ondisk;1693 char *name;1694 int result;16951696 if (!previous_name) {1697 size = ondisk_ce_size(ce);1698 ondisk = xcalloc(1, size);1699 name = copy_cache_entry_to_ondisk(ondisk, ce);1700 memcpy(name, ce->name, ce_namelen(ce));1701 } else {1702 int common, to_remove, prefix_size;1703 unsigned char to_remove_vi[16];1704 for (common = 0;1705 (ce->name[common] &&1706 common < previous_name->len &&1707 ce->name[common] == previous_name->buf[common]);1708 common++)1709 ; /* still matching */1710 to_remove = previous_name->len - common;1711 prefix_size = encode_varint(to_remove, to_remove_vi);17121713 if (ce->ce_flags & CE_EXTENDED)1714 size = offsetof(struct ondisk_cache_entry_extended, name);1715 else1716 size = offsetof(struct ondisk_cache_entry, name);1717 size += prefix_size + (ce_namelen(ce) - common + 1);17181719 ondisk = xcalloc(1, size);1720 name = copy_cache_entry_to_ondisk(ondisk, ce);1721 memcpy(name, to_remove_vi, prefix_size);1722 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);17231724 strbuf_splice(previous_name, common, to_remove,1725 ce->name + common, ce_namelen(ce) - common);1726 }17271728 result = ce_write(c, fd, ondisk, size);1729 free(ondisk);1730 return result;1731}17321733static int has_racy_timestamp(struct index_state *istate)1734{1735 int entries = istate->cache_nr;1736 int i;17371738 for (i = 0; i < entries; i++) {1739 struct cache_entry *ce = istate->cache[i];1740 if (is_racy_timestamp(istate, ce))1741 return 1;1742 }1743 return 0;1744}17451746/*1747 * Opportunisticly update the index but do not complain if we can't1748 */1749void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)1750{1751 if ((istate->cache_changed || has_racy_timestamp(istate)) &&1752 !write_index(istate, lockfile->fd))1753 commit_locked_index(lockfile);1754 else1755 rollback_lock_file(lockfile);1756}17571758int write_index(struct index_state *istate, int newfd)1759{1760 git_SHA_CTX c;1761 struct cache_header hdr;1762 int i, err, removed, extended, hdr_version;1763 struct cache_entry **cache = istate->cache;1764 int entries = istate->cache_nr;1765 struct stat st;1766 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;17671768 for (i = removed = extended = 0; i < entries; i++) {1769 if (cache[i]->ce_flags & CE_REMOVE)1770 removed++;17711772 /* reduce extended entries if possible */1773 cache[i]->ce_flags &= ~CE_EXTENDED;1774 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {1775 extended++;1776 cache[i]->ce_flags |= CE_EXTENDED;1777 }1778 }17791780 if (!istate->version)1781 istate->version = INDEX_FORMAT_DEFAULT;17821783 /* demote version 3 to version 2 when the latter suffices */1784 if (istate->version == 3 || istate->version == 2)1785 istate->version = extended ? 3 : 2;17861787 hdr_version = istate->version;17881789 hdr.hdr_signature = htonl(CACHE_SIGNATURE);1790 hdr.hdr_version = htonl(hdr_version);1791 hdr.hdr_entries = htonl(entries - removed);17921793 git_SHA1_Init(&c);1794 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)1795 return -1;17961797 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;1798 for (i = 0; i < entries; i++) {1799 struct cache_entry *ce = cache[i];1800 if (ce->ce_flags & CE_REMOVE)1801 continue;1802 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))1803 ce_smudge_racily_clean_entry(ce);1804 if (is_null_sha1(ce->sha1))1805 return error("cache entry has null sha1: %s", ce->name);1806 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)1807 return -1;1808 }1809 strbuf_release(&previous_name_buf);18101811 /* Write extension data here */1812 if (istate->cache_tree) {1813 struct strbuf sb = STRBUF_INIT;18141815 cache_tree_write(&sb, istate->cache_tree);1816 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 01817 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1818 strbuf_release(&sb);1819 if (err)1820 return -1;1821 }1822 if (istate->resolve_undo) {1823 struct strbuf sb = STRBUF_INIT;18241825 resolve_undo_write(&sb, istate->resolve_undo);1826 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,1827 sb.len) < 01828 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1829 strbuf_release(&sb);1830 if (err)1831 return -1;1832 }18331834 if (ce_flush(&c, newfd) || fstat(newfd, &st))1835 return -1;1836 istate->timestamp.sec = (unsigned int)st.st_mtime;1837 istate->timestamp.nsec = ST_MTIME_NSEC(st);1838 return 0;1839}18401841/*1842 * Read the index file that is potentially unmerged into given1843 * index_state, dropping any unmerged entries. Returns true if1844 * the index is unmerged. Callers who want to refuse to work1845 * from an unmerged state can call this and check its return value,1846 * instead of calling read_cache().1847 */1848int read_index_unmerged(struct index_state *istate)1849{1850 int i;1851 int unmerged = 0;18521853 read_index(istate);1854 for (i = 0; i < istate->cache_nr; i++) {1855 struct cache_entry *ce = istate->cache[i];1856 struct cache_entry *new_ce;1857 int size, len;18581859 if (!ce_stage(ce))1860 continue;1861 unmerged = 1;1862 len = ce_namelen(ce);1863 size = cache_entry_size(len);1864 new_ce = xcalloc(1, size);1865 memcpy(new_ce->name, ce->name, len);1866 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;1867 new_ce->ce_namelen = len;1868 new_ce->ce_mode = ce->ce_mode;1869 if (add_index_entry(istate, new_ce, 0))1870 return error("%s: cannot drop to stage #0",1871 ce->name);1872 i = index_name_pos(istate, new_ce->name, len);1873 }1874 return unmerged;1875}18761877/*1878 * Returns 1 if the path is an "other" path with respect to1879 * the index; that is, the path is not mentioned in the index at all,1880 * either as a file, a directory with some files in the index,1881 * or as an unmerged entry.1882 *1883 * We helpfully remove a trailing "/" from directories so that1884 * the output of read_directory can be used as-is.1885 */1886int index_name_is_other(const struct index_state *istate, const char *name,1887 int namelen)1888{1889 int pos;1890 if (namelen && name[namelen - 1] == '/')1891 namelen--;1892 pos = index_name_pos(istate, name, namelen);1893 if (0 <= pos)1894 return 0; /* exact match */1895 pos = -pos - 1;1896 if (pos < istate->cache_nr) {1897 struct cache_entry *ce = istate->cache[pos];1898 if (ce_namelen(ce) == namelen &&1899 !memcmp(ce->name, name, namelen))1900 return 0; /* Yup, this one exists unmerged */1901 }1902 return 1;1903}19041905void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)1906{1907 int pos, len;1908 unsigned long sz;1909 enum object_type type;1910 void *data;19111912 len = strlen(path);1913 pos = index_name_pos(istate, path, len);1914 if (pos < 0) {1915 /*1916 * We might be in the middle of a merge, in which1917 * case we would read stage #2 (ours).1918 */1919 int i;1920 for (i = -pos - 1;1921 (pos < 0 && i < istate->cache_nr &&1922 !strcmp(istate->cache[i]->name, path));1923 i++)1924 if (ce_stage(istate->cache[i]) == 2)1925 pos = i;1926 }1927 if (pos < 0)1928 return NULL;1929 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);1930 if (!data || type != OBJ_BLOB) {1931 free(data);1932 return NULL;1933 }1934 if (size)1935 *size = sz;1936 return data;1937}