1/* 2 * GIT - The information manager from hell 3 * 4 * Copyright (C) Linus Torvalds, 2005 5 */ 6#define NO_THE_INDEX_COMPATIBILITY_MACROS 7#include "cache.h" 8#include "cache-tree.h" 9#include "refs.h" 10#include "dir.h" 11#include "tree.h" 12#include "commit.h" 13#include "diff.h" 14#include "diffcore.h" 15#include "revision.h" 16#include "blob.h" 17#include "resolve-undo.h" 18 19static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really); 20 21/* Index extensions. 22 * 23 * The first letter should be 'A'..'Z' for extensions that are not 24 * necessary for a correct operation (i.e. optimization data). 25 * When new extensions are added that _needs_ to be understood in 26 * order to correctly interpret the index file, pick character that 27 * is outside the range, to cause the reader to abort. 28 */ 29 30#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) ) 31#define CACHE_EXT_TREE 0x54524545 /* "TREE" */ 32#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUN" */ 33 34struct index_state the_index; 35 36static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 37{ 38 istate->cache[nr] = ce; 39 add_name_hash(istate, ce); 40} 41 42static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 43{ 44 struct cache_entry *old = istate->cache[nr]; 45 46 remove_name_hash(old); 47 set_index_entry(istate, nr, ce); 48 istate->cache_changed = 1; 49} 50 51void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name) 52{ 53 struct cache_entry *old = istate->cache[nr], *new; 54 int namelen = strlen(new_name); 55 56 new = xmalloc(cache_entry_size(namelen)); 57 copy_cache_entry(new, old); 58 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK); 59 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen); 60 memcpy(new->name, new_name, namelen + 1); 61 62 cache_tree_invalidate_path(istate->cache_tree, old->name); 63 remove_index_entry_at(istate, nr); 64 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE); 65} 66 67/* 68 * This only updates the "non-critical" parts of the directory 69 * cache, ie the parts that aren't tracked by GIT, and only used 70 * to validate the cache. 71 */ 72void fill_stat_cache_info(struct cache_entry *ce, struct stat *st) 73{ 74 ce->ce_ctime.sec = (unsigned int)st->st_ctime; 75 ce->ce_mtime.sec = (unsigned int)st->st_mtime; 76 ce->ce_ctime.nsec = ST_CTIME_NSEC(*st); 77 ce->ce_mtime.nsec = ST_MTIME_NSEC(*st); 78 ce->ce_dev = st->st_dev; 79 ce->ce_ino = st->st_ino; 80 ce->ce_uid = st->st_uid; 81 ce->ce_gid = st->st_gid; 82 ce->ce_size = st->st_size; 83 84 if (assume_unchanged) 85 ce->ce_flags |= CE_VALID; 86 87 if (S_ISREG(st->st_mode)) 88 ce_mark_uptodate(ce); 89} 90 91static int ce_compare_data(struct cache_entry *ce, struct stat *st) 92{ 93 int match = -1; 94 int fd = open(ce->name, O_RDONLY); 95 96 if (fd >= 0) { 97 unsigned char sha1[20]; 98 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name)) 99 match = hashcmp(sha1, ce->sha1); 100 /* index_fd() closed the file descriptor already */ 101 } 102 return match; 103} 104 105static int ce_compare_link(struct cache_entry *ce, size_t expected_size) 106{ 107 int match = -1; 108 void *buffer; 109 unsigned long size; 110 enum object_type type; 111 struct strbuf sb = STRBUF_INIT; 112 113 if (strbuf_readlink(&sb, ce->name, expected_size)) 114 return -1; 115 116 buffer = read_sha1_file(ce->sha1, &type, &size); 117 if (buffer) { 118 if (size == sb.len) 119 match = memcmp(buffer, sb.buf, size); 120 free(buffer); 121 } 122 strbuf_release(&sb); 123 return match; 124} 125 126static int ce_compare_gitlink(struct cache_entry *ce) 127{ 128 unsigned char sha1[20]; 129 130 /* 131 * We don't actually require that the .git directory 132 * under GITLINK directory be a valid git directory. It 133 * might even be missing (in case nobody populated that 134 * sub-project). 135 * 136 * If so, we consider it always to match. 137 */ 138 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0) 139 return 0; 140 return hashcmp(sha1, ce->sha1); 141} 142 143static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st) 144{ 145 switch (st->st_mode & S_IFMT) { 146 case S_IFREG: 147 if (ce_compare_data(ce, st)) 148 return DATA_CHANGED; 149 break; 150 case S_IFLNK: 151 if (ce_compare_link(ce, xsize_t(st->st_size))) 152 return DATA_CHANGED; 153 break; 154 case S_IFDIR: 155 if (S_ISGITLINK(ce->ce_mode)) 156 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0; 157 default: 158 return TYPE_CHANGED; 159 } 160 return 0; 161} 162 163static int is_empty_blob_sha1(const unsigned char *sha1) 164{ 165 static const unsigned char empty_blob_sha1[20] = { 166 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b, 167 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91 168 }; 169 170 return !hashcmp(sha1, empty_blob_sha1); 171} 172 173static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st) 174{ 175 unsigned int changed = 0; 176 177 if (ce->ce_flags & CE_REMOVE) 178 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; 179 180 switch (ce->ce_mode & S_IFMT) { 181 case S_IFREG: 182 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0; 183 /* We consider only the owner x bit to be relevant for 184 * "mode changes" 185 */ 186 if (trust_executable_bit && 187 (0100 & (ce->ce_mode ^ st->st_mode))) 188 changed |= MODE_CHANGED; 189 break; 190 case S_IFLNK: 191 if (!S_ISLNK(st->st_mode) && 192 (has_symlinks || !S_ISREG(st->st_mode))) 193 changed |= TYPE_CHANGED; 194 break; 195 case S_IFGITLINK: 196 /* We ignore most of the st_xxx fields for gitlinks */ 197 if (!S_ISDIR(st->st_mode)) 198 changed |= TYPE_CHANGED; 199 else if (ce_compare_gitlink(ce)) 200 changed |= DATA_CHANGED; 201 return changed; 202 default: 203 die("internal error: ce_mode is %o", ce->ce_mode); 204 } 205 if (ce->ce_mtime.sec != (unsigned int)st->st_mtime) 206 changed |= MTIME_CHANGED; 207 if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime) 208 changed |= CTIME_CHANGED; 209 210#ifdef USE_NSEC 211 if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st)) 212 changed |= MTIME_CHANGED; 213 if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st)) 214 changed |= CTIME_CHANGED; 215#endif 216 217 if (ce->ce_uid != (unsigned int) st->st_uid || 218 ce->ce_gid != (unsigned int) st->st_gid) 219 changed |= OWNER_CHANGED; 220 if (ce->ce_ino != (unsigned int) st->st_ino) 221 changed |= INODE_CHANGED; 222 223#ifdef USE_STDEV 224 /* 225 * st_dev breaks on network filesystems where different 226 * clients will have different views of what "device" 227 * the filesystem is on 228 */ 229 if (ce->ce_dev != (unsigned int) st->st_dev) 230 changed |= INODE_CHANGED; 231#endif 232 233 if (ce->ce_size != (unsigned int) st->st_size) 234 changed |= DATA_CHANGED; 235 236 /* Racily smudged entry? */ 237 if (!ce->ce_size) { 238 if (!is_empty_blob_sha1(ce->sha1)) 239 changed |= DATA_CHANGED; 240 } 241 242 return changed; 243} 244 245static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce) 246{ 247 return (!S_ISGITLINK(ce->ce_mode) && 248 istate->timestamp.sec && 249#ifdef USE_NSEC 250 /* nanosecond timestamped files can also be racy! */ 251 (istate->timestamp.sec < ce->ce_mtime.sec || 252 (istate->timestamp.sec == ce->ce_mtime.sec && 253 istate->timestamp.nsec <= ce->ce_mtime.nsec)) 254#else 255 istate->timestamp.sec <= ce->ce_mtime.sec 256#endif 257 ); 258} 259 260int ie_match_stat(const struct index_state *istate, 261 struct cache_entry *ce, struct stat *st, 262 unsigned int options) 263{ 264 unsigned int changed; 265 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 266 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE; 267 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 268 269 /* 270 * If it's marked as always valid in the index, it's 271 * valid whatever the checked-out copy says. 272 * 273 * skip-worktree has the same effect with higher precedence 274 */ 275 if (!ignore_skip_worktree && ce_skip_worktree(ce)) 276 return 0; 277 if (!ignore_valid && (ce->ce_flags & CE_VALID)) 278 return 0; 279 280 /* 281 * Intent-to-add entries have not been added, so the index entry 282 * by definition never matches what is in the work tree until it 283 * actually gets added. 284 */ 285 if (ce->ce_flags & CE_INTENT_TO_ADD) 286 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED; 287 288 changed = ce_match_stat_basic(ce, st); 289 290 /* 291 * Within 1 second of this sequence: 292 * echo xyzzy >file && git-update-index --add file 293 * running this command: 294 * echo frotz >file 295 * would give a falsely clean cache entry. The mtime and 296 * length match the cache, and other stat fields do not change. 297 * 298 * We could detect this at update-index time (the cache entry 299 * being registered/updated records the same time as "now") 300 * and delay the return from git-update-index, but that would 301 * effectively mean we can make at most one commit per second, 302 * which is not acceptable. Instead, we check cache entries 303 * whose mtime are the same as the index file timestamp more 304 * carefully than others. 305 */ 306 if (!changed && is_racy_timestamp(istate, ce)) { 307 if (assume_racy_is_modified) 308 changed |= DATA_CHANGED; 309 else 310 changed |= ce_modified_check_fs(ce, st); 311 } 312 313 return changed; 314} 315 316int ie_modified(const struct index_state *istate, 317 struct cache_entry *ce, struct stat *st, unsigned int options) 318{ 319 int changed, changed_fs; 320 321 changed = ie_match_stat(istate, ce, st, options); 322 if (!changed) 323 return 0; 324 /* 325 * If the mode or type has changed, there's no point in trying 326 * to refresh the entry - it's not going to match 327 */ 328 if (changed & (MODE_CHANGED | TYPE_CHANGED)) 329 return changed; 330 331 /* 332 * Immediately after read-tree or update-index --cacheinfo, 333 * the length field is zero, as we have never even read the 334 * lstat(2) information once, and we cannot trust DATA_CHANGED 335 * returned by ie_match_stat() which in turn was returned by 336 * ce_match_stat_basic() to signal that the filesize of the 337 * blob changed. We have to actually go to the filesystem to 338 * see if the contents match, and if so, should answer "unchanged". 339 * 340 * The logic does not apply to gitlinks, as ce_match_stat_basic() 341 * already has checked the actual HEAD from the filesystem in the 342 * subproject. If ie_match_stat() already said it is different, 343 * then we know it is. 344 */ 345 if ((changed & DATA_CHANGED) && 346 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0)) 347 return changed; 348 349 changed_fs = ce_modified_check_fs(ce, st); 350 if (changed_fs) 351 return changed | changed_fs; 352 return 0; 353} 354 355int base_name_compare(const char *name1, int len1, int mode1, 356 const char *name2, int len2, int mode2) 357{ 358 unsigned char c1, c2; 359 int len = len1 < len2 ? len1 : len2; 360 int cmp; 361 362 cmp = memcmp(name1, name2, len); 363 if (cmp) 364 return cmp; 365 c1 = name1[len]; 366 c2 = name2[len]; 367 if (!c1 && S_ISDIR(mode1)) 368 c1 = '/'; 369 if (!c2 && S_ISDIR(mode2)) 370 c2 = '/'; 371 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0; 372} 373 374/* 375 * df_name_compare() is identical to base_name_compare(), except it 376 * compares conflicting directory/file entries as equal. Note that 377 * while a directory name compares as equal to a regular file, they 378 * then individually compare _differently_ to a filename that has 379 * a dot after the basename (because '\0' < '.' < '/'). 380 * 381 * This is used by routines that want to traverse the git namespace 382 * but then handle conflicting entries together when possible. 383 */ 384int df_name_compare(const char *name1, int len1, int mode1, 385 const char *name2, int len2, int mode2) 386{ 387 int len = len1 < len2 ? len1 : len2, cmp; 388 unsigned char c1, c2; 389 390 cmp = memcmp(name1, name2, len); 391 if (cmp) 392 return cmp; 393 /* Directories and files compare equal (same length, same name) */ 394 if (len1 == len2) 395 return 0; 396 c1 = name1[len]; 397 if (!c1 && S_ISDIR(mode1)) 398 c1 = '/'; 399 c2 = name2[len]; 400 if (!c2 && S_ISDIR(mode2)) 401 c2 = '/'; 402 if (c1 == '/' && !c2) 403 return 0; 404 if (c2 == '/' && !c1) 405 return 0; 406 return c1 - c2; 407} 408 409int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2) 410{ 411 int len1 = flags1 & CE_NAMEMASK; 412 int len2 = flags2 & CE_NAMEMASK; 413 int len = len1 < len2 ? len1 : len2; 414 int cmp; 415 416 cmp = memcmp(name1, name2, len); 417 if (cmp) 418 return cmp; 419 if (len1 < len2) 420 return -1; 421 if (len1 > len2) 422 return 1; 423 424 /* Compare stages */ 425 flags1 &= CE_STAGEMASK; 426 flags2 &= CE_STAGEMASK; 427 428 if (flags1 < flags2) 429 return -1; 430 if (flags1 > flags2) 431 return 1; 432 return 0; 433} 434 435int index_name_pos(const struct index_state *istate, const char *name, int namelen) 436{ 437 int first, last; 438 439 first = 0; 440 last = istate->cache_nr; 441 while (last > first) { 442 int next = (last + first) >> 1; 443 struct cache_entry *ce = istate->cache[next]; 444 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags); 445 if (!cmp) 446 return next; 447 if (cmp < 0) { 448 last = next; 449 continue; 450 } 451 first = next+1; 452 } 453 return -first-1; 454} 455 456/* Remove entry, return true if there are more entries to go.. */ 457int remove_index_entry_at(struct index_state *istate, int pos) 458{ 459 struct cache_entry *ce = istate->cache[pos]; 460 461 record_resolve_undo(istate, ce); 462 remove_name_hash(ce); 463 istate->cache_changed = 1; 464 istate->cache_nr--; 465 if (pos >= istate->cache_nr) 466 return 0; 467 memmove(istate->cache + pos, 468 istate->cache + pos + 1, 469 (istate->cache_nr - pos) * sizeof(struct cache_entry *)); 470 return 1; 471} 472 473/* 474 * Remove all cache ententries marked for removal, that is where 475 * CE_REMOVE is set in ce_flags. This is much more effective than 476 * calling remove_index_entry_at() for each entry to be removed. 477 */ 478void remove_marked_cache_entries(struct index_state *istate) 479{ 480 struct cache_entry **ce_array = istate->cache; 481 unsigned int i, j; 482 483 for (i = j = 0; i < istate->cache_nr; i++) { 484 if (ce_array[i]->ce_flags & CE_REMOVE) 485 remove_name_hash(ce_array[i]); 486 else 487 ce_array[j++] = ce_array[i]; 488 } 489 istate->cache_changed = 1; 490 istate->cache_nr = j; 491} 492 493int remove_file_from_index(struct index_state *istate, const char *path) 494{ 495 int pos = index_name_pos(istate, path, strlen(path)); 496 if (pos < 0) 497 pos = -pos-1; 498 cache_tree_invalidate_path(istate->cache_tree, path); 499 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 500 remove_index_entry_at(istate, pos); 501 return 0; 502} 503 504static int compare_name(struct cache_entry *ce, const char *path, int namelen) 505{ 506 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); 507} 508 509static int index_name_pos_also_unmerged(struct index_state *istate, 510 const char *path, int namelen) 511{ 512 int pos = index_name_pos(istate, path, namelen); 513 struct cache_entry *ce; 514 515 if (pos >= 0) 516 return pos; 517 518 /* maybe unmerged? */ 519 pos = -1 - pos; 520 if (pos >= istate->cache_nr || 521 compare_name((ce = istate->cache[pos]), path, namelen)) 522 return -1; 523 524 /* order of preference: stage 2, 1, 3 */ 525 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && 526 ce_stage((ce = istate->cache[pos + 1])) == 2 && 527 !compare_name(ce, path, namelen)) 528 pos++; 529 return pos; 530} 531 532static int different_name(struct cache_entry *ce, struct cache_entry *alias) 533{ 534 int len = ce_namelen(ce); 535 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len); 536} 537 538/* 539 * If we add a filename that aliases in the cache, we will use the 540 * name that we already have - but we don't want to update the same 541 * alias twice, because that implies that there were actually two 542 * different files with aliasing names! 543 * 544 * So we use the CE_ADDED flag to verify that the alias was an old 545 * one before we accept it as 546 */ 547static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias) 548{ 549 int len; 550 struct cache_entry *new; 551 552 if (alias->ce_flags & CE_ADDED) 553 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); 554 555 /* Ok, create the new entry using the name of the existing alias */ 556 len = ce_namelen(alias); 557 new = xcalloc(1, cache_entry_size(len)); 558 memcpy(new->name, alias->name, len); 559 copy_cache_entry(new, ce); 560 free(ce); 561 return new; 562} 563 564static void record_intent_to_add(struct cache_entry *ce) 565{ 566 unsigned char sha1[20]; 567 if (write_sha1_file("", 0, blob_type, sha1)) 568 die("cannot create an empty blob in the object database"); 569 hashcpy(ce->sha1, sha1); 570} 571 572int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags) 573{ 574 int size, namelen, was_same; 575 mode_t st_mode = st->st_mode; 576 struct cache_entry *ce, *alias; 577 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY; 578 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); 579 int pretend = flags & ADD_CACHE_PRETEND; 580 int intent_only = flags & ADD_CACHE_INTENT; 581 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| 582 (intent_only ? ADD_CACHE_NEW_ONLY : 0)); 583 584 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 585 return error("%s: can only add regular files, symbolic links or git-directories", path); 586 587 namelen = strlen(path); 588 if (S_ISDIR(st_mode)) { 589 while (namelen && path[namelen-1] == '/') 590 namelen--; 591 } 592 size = cache_entry_size(namelen); 593 ce = xcalloc(1, size); 594 memcpy(ce->name, path, namelen); 595 ce->ce_flags = namelen; 596 if (!intent_only) 597 fill_stat_cache_info(ce, st); 598 else 599 ce->ce_flags |= CE_INTENT_TO_ADD; 600 601 if (trust_executable_bit && has_symlinks) 602 ce->ce_mode = create_ce_mode(st_mode); 603 else { 604 /* If there is an existing entry, pick the mode bits and type 605 * from it, otherwise assume unexecutable regular file. 606 */ 607 struct cache_entry *ent; 608 int pos = index_name_pos_also_unmerged(istate, path, namelen); 609 610 ent = (0 <= pos) ? istate->cache[pos] : NULL; 611 ce->ce_mode = ce_mode_from_stat(ent, st_mode); 612 } 613 614 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case); 615 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) { 616 /* Nothing changed, really */ 617 free(ce); 618 ce_mark_uptodate(alias); 619 alias->ce_flags |= CE_ADDED; 620 return 0; 621 } 622 if (!intent_only) { 623 if (index_path(ce->sha1, path, st, 1)) 624 return error("unable to index file %s", path); 625 } else 626 record_intent_to_add(ce); 627 628 if (ignore_case && alias && different_name(ce, alias)) 629 ce = create_alias_ce(ce, alias); 630 ce->ce_flags |= CE_ADDED; 631 632 /* It was suspected to be racily clean, but it turns out to be Ok */ 633 was_same = (alias && 634 !ce_stage(alias) && 635 !hashcmp(alias->sha1, ce->sha1) && 636 ce->ce_mode == alias->ce_mode); 637 638 if (pretend) 639 ; 640 else if (add_index_entry(istate, ce, add_option)) 641 return error("unable to add %s to index",path); 642 if (verbose && !was_same) 643 printf("add '%s'\n", path); 644 return 0; 645} 646 647int add_file_to_index(struct index_state *istate, const char *path, int flags) 648{ 649 struct stat st; 650 if (lstat(path, &st)) 651 die_errno("unable to stat '%s'", path); 652 return add_to_index(istate, path, &st, flags); 653} 654 655struct cache_entry *make_cache_entry(unsigned int mode, 656 const unsigned char *sha1, const char *path, int stage, 657 int refresh) 658{ 659 int size, len; 660 struct cache_entry *ce; 661 662 if (!verify_path(path)) { 663 error("Invalid path '%s'", path); 664 return NULL; 665 } 666 667 len = strlen(path); 668 size = cache_entry_size(len); 669 ce = xcalloc(1, size); 670 671 hashcpy(ce->sha1, sha1); 672 memcpy(ce->name, path, len); 673 ce->ce_flags = create_ce_flags(len, stage); 674 ce->ce_mode = create_ce_mode(mode); 675 676 if (refresh) 677 return refresh_cache_entry(ce, 0); 678 679 return ce; 680} 681 682int ce_same_name(struct cache_entry *a, struct cache_entry *b) 683{ 684 int len = ce_namelen(a); 685 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 686} 687 688int ce_path_match(const struct cache_entry *ce, const char **pathspec) 689{ 690 const char *match, *name; 691 int len; 692 693 if (!pathspec) 694 return 1; 695 696 len = ce_namelen(ce); 697 name = ce->name; 698 while ((match = *pathspec++) != NULL) { 699 int matchlen = strlen(match); 700 if (matchlen > len) 701 continue; 702 if (memcmp(name, match, matchlen)) 703 continue; 704 if (matchlen && name[matchlen-1] == '/') 705 return 1; 706 if (name[matchlen] == '/' || !name[matchlen]) 707 return 1; 708 if (!matchlen) 709 return 1; 710 } 711 return 0; 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 ce_mark_uptodate(ce);1057 return ce;1058 }1059 }10601061 if (ie_modified(istate, ce, &st, options)) {1062 if (err)1063 *err = EINVAL;1064 return NULL;1065 }10661067 size = ce_size(ce);1068 updated = xmalloc(size);1069 memcpy(updated, ce, size);1070 fill_stat_cache_info(updated, &st);1071 /*1072 * If ignore_valid is not set, we should leave CE_VALID bit1073 * alone. Otherwise, paths marked with --no-assume-unchanged1074 * (i.e. things to be edited) will reacquire CE_VALID bit1075 * automatically, which is not really what we want.1076 */1077 if (!ignore_valid && assume_unchanged &&1078 !(ce->ce_flags & CE_VALID))1079 updated->ce_flags &= ~CE_VALID;10801081 return updated;1082}10831084static void show_file(const char * fmt, const char * name, int in_porcelain,1085 int * first, char *header_msg)1086{1087 if (in_porcelain && *first && header_msg) {1088 printf("%s\n", header_msg);1089 *first=0;1090 }1091 printf(fmt, name);1092}10931094int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,1095 char *seen, char *header_msg)1096{1097 int i;1098 int has_errors = 0;1099 int really = (flags & REFRESH_REALLY) != 0;1100 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;1101 int quiet = (flags & REFRESH_QUIET) != 0;1102 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;1103 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;1104 int first = 1;1105 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);1106 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;1107 const char *needs_update_fmt;1108 const char *needs_merge_fmt;11091110 needs_update_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");1111 needs_merge_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");1112 for (i = 0; i < istate->cache_nr; i++) {1113 struct cache_entry *ce, *new;1114 int cache_errno = 0;11151116 ce = istate->cache[i];1117 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))1118 continue;11191120 if (ce_stage(ce)) {1121 while ((i < istate->cache_nr) &&1122 ! strcmp(istate->cache[i]->name, ce->name))1123 i++;1124 i--;1125 if (allow_unmerged)1126 continue;1127 show_file(needs_merge_fmt, ce->name, in_porcelain, &first, header_msg);1128 has_errors = 1;1129 continue;1130 }11311132 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))1133 continue;11341135 new = refresh_cache_ent(istate, ce, options, &cache_errno);1136 if (new == ce)1137 continue;1138 if (!new) {1139 if (not_new && cache_errno == ENOENT)1140 continue;1141 if (really && cache_errno == EINVAL) {1142 /* If we are doing --really-refresh that1143 * means the index is not valid anymore.1144 */1145 ce->ce_flags &= ~CE_VALID;1146 istate->cache_changed = 1;1147 }1148 if (quiet)1149 continue;1150 show_file(needs_update_fmt, ce->name, in_porcelain, &first, header_msg);1151 has_errors = 1;1152 continue;1153 }11541155 replace_index_entry(istate, i, new);1156 }1157 return has_errors;1158}11591160static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)1161{1162 return refresh_cache_ent(&the_index, ce, really, NULL);1163}11641165static int verify_hdr(struct cache_header *hdr, unsigned long size)1166{1167 git_SHA_CTX c;1168 unsigned char sha1[20];11691170 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))1171 return error("bad signature");1172 if (hdr->hdr_version != htonl(2) && hdr->hdr_version != htonl(3))1173 return error("bad index version");1174 git_SHA1_Init(&c);1175 git_SHA1_Update(&c, hdr, size - 20);1176 git_SHA1_Final(sha1, &c);1177 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))1178 return error("bad index file sha1 signature");1179 return 0;1180}11811182static int read_index_extension(struct index_state *istate,1183 const char *ext, void *data, unsigned long sz)1184{1185 switch (CACHE_EXT(ext)) {1186 case CACHE_EXT_TREE:1187 istate->cache_tree = cache_tree_read(data, sz);1188 break;1189 case CACHE_EXT_RESOLVE_UNDO:1190 istate->resolve_undo = resolve_undo_read(data, sz);1191 break;1192 default:1193 if (*ext < 'A' || 'Z' < *ext)1194 return error("index uses %.4s extension, which we do not understand",1195 ext);1196 fprintf(stderr, "ignoring %.4s extension\n", ext);1197 break;1198 }1199 return 0;1200}12011202int read_index(struct index_state *istate)1203{1204 return read_index_from(istate, get_index_file());1205}12061207static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)1208{1209 size_t len;1210 const char *name;12111212 ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);1213 ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);1214 ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);1215 ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);1216 ce->ce_dev = ntohl(ondisk->dev);1217 ce->ce_ino = ntohl(ondisk->ino);1218 ce->ce_mode = ntohl(ondisk->mode);1219 ce->ce_uid = ntohl(ondisk->uid);1220 ce->ce_gid = ntohl(ondisk->gid);1221 ce->ce_size = ntohl(ondisk->size);1222 /* On-disk flags are just 16 bits */1223 ce->ce_flags = ntohs(ondisk->flags);12241225 hashcpy(ce->sha1, ondisk->sha1);12261227 len = ce->ce_flags & CE_NAMEMASK;12281229 if (ce->ce_flags & CE_EXTENDED) {1230 struct ondisk_cache_entry_extended *ondisk2;1231 int extended_flags;1232 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1233 extended_flags = ntohs(ondisk2->flags2) << 16;1234 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1235 if (extended_flags & ~CE_EXTENDED_FLAGS)1236 die("Unknown index entry format %08x", extended_flags);1237 ce->ce_flags |= extended_flags;1238 name = ondisk2->name;1239 }1240 else1241 name = ondisk->name;12421243 if (len == CE_NAMEMASK)1244 len = strlen(name);1245 /*1246 * NEEDSWORK: If the original index is crafted, this copy could1247 * go unchecked.1248 */1249 memcpy(ce->name, name, len + 1);1250}12511252static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)1253{1254 long per_entry;12551256 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);12571258 /*1259 * Alignment can cause differences. This should be "alignof", but1260 * since that's a gcc'ism, just use the size of a pointer.1261 */1262 per_entry += sizeof(void *);1263 return ondisk_size + entries*per_entry;1264}12651266/* remember to discard_cache() before reading a different cache! */1267int read_index_from(struct index_state *istate, const char *path)1268{1269 int fd, i;1270 struct stat st;1271 unsigned long src_offset, dst_offset;1272 struct cache_header *hdr;1273 void *mmap;1274 size_t mmap_size;12751276 errno = EBUSY;1277 if (istate->initialized)1278 return istate->cache_nr;12791280 errno = ENOENT;1281 istate->timestamp.sec = 0;1282 istate->timestamp.nsec = 0;1283 fd = open(path, O_RDONLY);1284 if (fd < 0) {1285 if (errno == ENOENT)1286 return 0;1287 die_errno("index file open failed");1288 }12891290 if (fstat(fd, &st))1291 die_errno("cannot stat the open index");12921293 errno = EINVAL;1294 mmap_size = xsize_t(st.st_size);1295 if (mmap_size < sizeof(struct cache_header) + 20)1296 die("index file smaller than expected");12971298 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);1299 close(fd);1300 if (mmap == MAP_FAILED)1301 die_errno("unable to map index file");13021303 hdr = mmap;1304 if (verify_hdr(hdr, mmap_size) < 0)1305 goto unmap;13061307 istate->cache_nr = ntohl(hdr->hdr_entries);1308 istate->cache_alloc = alloc_nr(istate->cache_nr);1309 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));13101311 /*1312 * The disk format is actually larger than the in-memory format,1313 * due to space for nsec etc, so even though the in-memory one1314 * has room for a few more flags, we can allocate using the same1315 * index size1316 */1317 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));1318 istate->initialized = 1;13191320 src_offset = sizeof(*hdr);1321 dst_offset = 0;1322 for (i = 0; i < istate->cache_nr; i++) {1323 struct ondisk_cache_entry *disk_ce;1324 struct cache_entry *ce;13251326 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);1327 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);1328 convert_from_disk(disk_ce, ce);1329 set_index_entry(istate, i, ce);13301331 src_offset += ondisk_ce_size(ce);1332 dst_offset += ce_size(ce);1333 }1334 istate->timestamp.sec = st.st_mtime;1335 istate->timestamp.nsec = ST_MTIME_NSEC(st);13361337 while (src_offset <= mmap_size - 20 - 8) {1338 /* After an array of active_nr index entries,1339 * there can be arbitrary number of extended1340 * sections, each of which is prefixed with1341 * extension name (4-byte) and section length1342 * in 4-byte network byte order.1343 */1344 uint32_t extsize;1345 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);1346 extsize = ntohl(extsize);1347 if (read_index_extension(istate,1348 (const char *) mmap + src_offset,1349 (char *) mmap + src_offset + 8,1350 extsize) < 0)1351 goto unmap;1352 src_offset += 8;1353 src_offset += extsize;1354 }1355 munmap(mmap, mmap_size);1356 return istate->cache_nr;13571358unmap:1359 munmap(mmap, mmap_size);1360 errno = EINVAL;1361 die("index file corrupt");1362}13631364int is_index_unborn(struct index_state *istate)1365{1366 return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);1367}13681369int discard_index(struct index_state *istate)1370{1371 resolve_undo_clear_index(istate);1372 istate->cache_nr = 0;1373 istate->cache_changed = 0;1374 istate->timestamp.sec = 0;1375 istate->timestamp.nsec = 0;1376 istate->name_hash_initialized = 0;1377 free_hash(&istate->name_hash);1378 cache_tree_free(&(istate->cache_tree));1379 free(istate->alloc);1380 istate->alloc = NULL;1381 istate->initialized = 0;13821383 /* no need to throw away allocated active_cache */1384 return 0;1385}13861387int unmerged_index(const struct index_state *istate)1388{1389 int i;1390 for (i = 0; i < istate->cache_nr; i++) {1391 if (ce_stage(istate->cache[i]))1392 return 1;1393 }1394 return 0;1395}13961397#define WRITE_BUFFER_SIZE 81921398static unsigned char write_buffer[WRITE_BUFFER_SIZE];1399static unsigned long write_buffer_len;14001401static int ce_write_flush(git_SHA_CTX *context, int fd)1402{1403 unsigned int buffered = write_buffer_len;1404 if (buffered) {1405 git_SHA1_Update(context, write_buffer, buffered);1406 if (write_in_full(fd, write_buffer, buffered) != buffered)1407 return -1;1408 write_buffer_len = 0;1409 }1410 return 0;1411}14121413static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)1414{1415 while (len) {1416 unsigned int buffered = write_buffer_len;1417 unsigned int partial = WRITE_BUFFER_SIZE - buffered;1418 if (partial > len)1419 partial = len;1420 memcpy(write_buffer + buffered, data, partial);1421 buffered += partial;1422 if (buffered == WRITE_BUFFER_SIZE) {1423 write_buffer_len = buffered;1424 if (ce_write_flush(context, fd))1425 return -1;1426 buffered = 0;1427 }1428 write_buffer_len = buffered;1429 len -= partial;1430 data = (char *) data + partial;1431 }1432 return 0;1433}14341435static int write_index_ext_header(git_SHA_CTX *context, int fd,1436 unsigned int ext, unsigned int sz)1437{1438 ext = htonl(ext);1439 sz = htonl(sz);1440 return ((ce_write(context, fd, &ext, 4) < 0) ||1441 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;1442}14431444static int ce_flush(git_SHA_CTX *context, int fd)1445{1446 unsigned int left = write_buffer_len;14471448 if (left) {1449 write_buffer_len = 0;1450 git_SHA1_Update(context, write_buffer, left);1451 }14521453 /* Flush first if not enough space for SHA1 signature */1454 if (left + 20 > WRITE_BUFFER_SIZE) {1455 if (write_in_full(fd, write_buffer, left) != left)1456 return -1;1457 left = 0;1458 }14591460 /* Append the SHA1 signature at the end */1461 git_SHA1_Final(write_buffer + left, context);1462 left += 20;1463 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;1464}14651466static void ce_smudge_racily_clean_entry(struct cache_entry *ce)1467{1468 /*1469 * The only thing we care about in this function is to smudge the1470 * falsely clean entry due to touch-update-touch race, so we leave1471 * everything else as they are. We are called for entries whose1472 * ce_mtime match the index file mtime.1473 *1474 * Note that this actually does not do much for gitlinks, for1475 * which ce_match_stat_basic() always goes to the actual1476 * contents. The caller checks with is_racy_timestamp() which1477 * always says "no" for gitlinks, so we are not called for them ;-)1478 */1479 struct stat st;14801481 if (lstat(ce->name, &st) < 0)1482 return;1483 if (ce_match_stat_basic(ce, &st))1484 return;1485 if (ce_modified_check_fs(ce, &st)) {1486 /* This is "racily clean"; smudge it. Note that this1487 * is a tricky code. At first glance, it may appear1488 * that it can break with this sequence:1489 *1490 * $ echo xyzzy >frotz1491 * $ git-update-index --add frotz1492 * $ : >frotz1493 * $ sleep 31494 * $ echo filfre >nitfol1495 * $ git-update-index --add nitfol1496 *1497 * but it does not. When the second update-index runs,1498 * it notices that the entry "frotz" has the same timestamp1499 * as index, and if we were to smudge it by resetting its1500 * size to zero here, then the object name recorded1501 * in index is the 6-byte file but the cached stat information1502 * becomes zero --- which would then match what we would1503 * obtain from the filesystem next time we stat("frotz").1504 *1505 * However, the second update-index, before calling1506 * this function, notices that the cached size is 61507 * bytes and what is on the filesystem is an empty1508 * file, and never calls us, so the cached size information1509 * for "frotz" stays 6 which does not match the filesystem.1510 */1511 ce->ce_size = 0;1512 }1513}15141515static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)1516{1517 int size = ondisk_ce_size(ce);1518 struct ondisk_cache_entry *ondisk = xcalloc(1, size);1519 char *name;15201521 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);1522 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);1523 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);1524 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);1525 ondisk->dev = htonl(ce->ce_dev);1526 ondisk->ino = htonl(ce->ce_ino);1527 ondisk->mode = htonl(ce->ce_mode);1528 ondisk->uid = htonl(ce->ce_uid);1529 ondisk->gid = htonl(ce->ce_gid);1530 ondisk->size = htonl(ce->ce_size);1531 hashcpy(ondisk->sha1, ce->sha1);1532 ondisk->flags = htons(ce->ce_flags);1533 if (ce->ce_flags & CE_EXTENDED) {1534 struct ondisk_cache_entry_extended *ondisk2;1535 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1536 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);1537 name = ondisk2->name;1538 }1539 else1540 name = ondisk->name;1541 memcpy(name, ce->name, ce_namelen(ce));15421543 return ce_write(c, fd, ondisk, size);1544}15451546int write_index(struct index_state *istate, int newfd)1547{1548 git_SHA_CTX c;1549 struct cache_header hdr;1550 int i, err, removed, extended;1551 struct cache_entry **cache = istate->cache;1552 int entries = istate->cache_nr;1553 struct stat st;15541555 for (i = removed = extended = 0; i < entries; i++) {1556 if (cache[i]->ce_flags & CE_REMOVE)1557 removed++;15581559 /* reduce extended entries if possible */1560 cache[i]->ce_flags &= ~CE_EXTENDED;1561 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {1562 extended++;1563 cache[i]->ce_flags |= CE_EXTENDED;1564 }1565 }15661567 hdr.hdr_signature = htonl(CACHE_SIGNATURE);1568 /* for extended format, increase version so older git won't try to read it */1569 hdr.hdr_version = htonl(extended ? 3 : 2);1570 hdr.hdr_entries = htonl(entries - removed);15711572 git_SHA1_Init(&c);1573 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)1574 return -1;15751576 for (i = 0; i < entries; i++) {1577 struct cache_entry *ce = cache[i];1578 if (ce->ce_flags & CE_REMOVE)1579 continue;1580 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))1581 ce_smudge_racily_clean_entry(ce);1582 if (ce_write_entry(&c, newfd, ce) < 0)1583 return -1;1584 }15851586 /* Write extension data here */1587 if (istate->cache_tree) {1588 struct strbuf sb = STRBUF_INIT;15891590 cache_tree_write(&sb, istate->cache_tree);1591 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 01592 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1593 strbuf_release(&sb);1594 if (err)1595 return -1;1596 }1597 if (istate->resolve_undo) {1598 struct strbuf sb = STRBUF_INIT;15991600 resolve_undo_write(&sb, istate->resolve_undo);1601 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,1602 sb.len) < 01603 || ce_write(&c, newfd, sb.buf, sb.len) < 0;1604 strbuf_release(&sb);1605 if (err)1606 return -1;1607 }16081609 if (ce_flush(&c, newfd) || fstat(newfd, &st))1610 return -1;1611 istate->timestamp.sec = (unsigned int)st.st_mtime;1612 istate->timestamp.nsec = ST_MTIME_NSEC(st);1613 return 0;1614}16151616/*1617 * Read the index file that is potentially unmerged into given1618 * index_state, dropping any unmerged entries. Returns true if1619 * the index is unmerged. Callers who want to refuse to work1620 * from an unmerged state can call this and check its return value,1621 * instead of calling read_cache().1622 */1623int read_index_unmerged(struct index_state *istate)1624{1625 int i;1626 int unmerged = 0;16271628 read_index(istate);1629 for (i = 0; i < istate->cache_nr; i++) {1630 struct cache_entry *ce = istate->cache[i];1631 struct cache_entry *new_ce;1632 int size, len;16331634 if (!ce_stage(ce))1635 continue;1636 unmerged = 1;1637 len = strlen(ce->name);1638 size = cache_entry_size(len);1639 new_ce = xcalloc(1, size);1640 memcpy(new_ce->name, ce->name, len);1641 new_ce->ce_flags = create_ce_flags(len, 0) | CE_CONFLICTED;1642 new_ce->ce_mode = ce->ce_mode;1643 if (add_index_entry(istate, new_ce, 0))1644 return error("%s: cannot drop to stage #0",1645 ce->name);1646 i = index_name_pos(istate, new_ce->name, len);1647 }1648 return unmerged;1649}16501651struct update_callback_data1652{1653 int flags;1654 int add_errors;1655};16561657static void update_callback(struct diff_queue_struct *q,1658 struct diff_options *opt, void *cbdata)1659{1660 int i;1661 struct update_callback_data *data = cbdata;16621663 for (i = 0; i < q->nr; i++) {1664 struct diff_filepair *p = q->queue[i];1665 const char *path = p->one->path;1666 switch (p->status) {1667 default:1668 die("unexpected diff status %c", p->status);1669 case DIFF_STATUS_UNMERGED:1670 /*1671 * ADD_CACHE_IGNORE_REMOVAL is unset if "git1672 * add -u" is calling us, In such a case, a1673 * missing work tree file needs to be removed1674 * if there is an unmerged entry at stage #2,1675 * but such a diff record is followed by1676 * another with DIFF_STATUS_DELETED (and if1677 * there is no stage #2, we won't see DELETED1678 * nor MODIFIED). We can simply continue1679 * either way.1680 */1681 if (!(data->flags & ADD_CACHE_IGNORE_REMOVAL))1682 continue;1683 /*1684 * Otherwise, it is "git add path" is asking1685 * to explicitly add it; we fall through. A1686 * missing work tree file is an error and is1687 * caught by add_file_to_index() in such a1688 * case.1689 */1690 case DIFF_STATUS_MODIFIED:1691 case DIFF_STATUS_TYPE_CHANGED:1692 if (add_file_to_index(&the_index, path, data->flags)) {1693 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))1694 die("updating files failed");1695 data->add_errors++;1696 }1697 break;1698 case DIFF_STATUS_DELETED:1699 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)1700 break;1701 if (!(data->flags & ADD_CACHE_PRETEND))1702 remove_file_from_index(&the_index, path);1703 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))1704 printf("remove '%s'\n", path);1705 break;1706 }1707 }1708}17091710int add_files_to_cache(const char *prefix, const char **pathspec, int flags)1711{1712 struct update_callback_data data;1713 struct rev_info rev;1714 init_revisions(&rev, prefix);1715 setup_revisions(0, NULL, &rev, NULL);1716 rev.prune_data = pathspec;1717 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;1718 rev.diffopt.format_callback = update_callback;1719 data.flags = flags;1720 data.add_errors = 0;1721 rev.diffopt.format_callback_data = &data;1722 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);1723 return !!data.add_errors;1724}17251726/*1727 * Returns 1 if the path is an "other" path with respect to1728 * the index; that is, the path is not mentioned in the index at all,1729 * either as a file, a directory with some files in the index,1730 * or as an unmerged entry.1731 *1732 * We helpfully remove a trailing "/" from directories so that1733 * the output of read_directory can be used as-is.1734 */1735int index_name_is_other(const struct index_state *istate, const char *name,1736 int namelen)1737{1738 int pos;1739 if (namelen && name[namelen - 1] == '/')1740 namelen--;1741 pos = index_name_pos(istate, name, namelen);1742 if (0 <= pos)1743 return 0; /* exact match */1744 pos = -pos - 1;1745 if (pos < istate->cache_nr) {1746 struct cache_entry *ce = istate->cache[pos];1747 if (ce_namelen(ce) == namelen &&1748 !memcmp(ce->name, name, namelen))1749 return 0; /* Yup, this one exists unmerged */1750 }1751 return 1;1752}