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 "tempfile.h" 9#include "lockfile.h" 10#include "cache-tree.h" 11#include "refs.h" 12#include "dir.h" 13#include "tree.h" 14#include "commit.h" 15#include "blob.h" 16#include "resolve-undo.h" 17#include "strbuf.h" 18#include "varint.h" 19#include "split-index.h" 20#include "utf8.h" 21 22/* Mask for the name length in ce_flags in the on-disk index */ 23 24#define CE_NAMEMASK (0x0fff) 25 26/* Index extensions. 27 * 28 * The first letter should be 'A'..'Z' for extensions that are not 29 * necessary for a correct operation (i.e. optimization data). 30 * When new extensions are added that _needs_ to be understood in 31 * order to correctly interpret the index file, pick character that 32 * is outside the range, to cause the reader to abort. 33 */ 34 35#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) ) 36#define CACHE_EXT_TREE 0x54524545 /* "TREE" */ 37#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */ 38#define CACHE_EXT_LINK 0x6c696e6b /* "link" */ 39#define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */ 40 41/* changes that can be kept in $GIT_DIR/index (basically all extensions) */ 42#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \ 43 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \ 44 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED) 45 46struct index_state the_index; 47static const char *alternate_index_output; 48 49static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 50{ 51 istate->cache[nr] = ce; 52 add_name_hash(istate, ce); 53} 54 55static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 56{ 57 struct cache_entry *old = istate->cache[nr]; 58 59 replace_index_entry_in_base(istate, old, ce); 60 remove_name_hash(istate, old); 61 free(old); 62 set_index_entry(istate, nr, ce); 63 ce->ce_flags |= CE_UPDATE_IN_BASE; 64 istate->cache_changed |= CE_ENTRY_CHANGED; 65} 66 67void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name) 68{ 69 struct cache_entry *old = istate->cache[nr], *new; 70 int namelen = strlen(new_name); 71 72 new = xmalloc(cache_entry_size(namelen)); 73 copy_cache_entry(new, old); 74 new->ce_flags &= ~CE_HASHED; 75 new->ce_namelen = namelen; 76 new->index = 0; 77 memcpy(new->name, new_name, namelen + 1); 78 79 cache_tree_invalidate_path(istate, old->name); 80 untracked_cache_remove_from_index(istate, old->name); 81 remove_index_entry_at(istate, nr); 82 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE); 83} 84 85void fill_stat_data(struct stat_data *sd, struct stat *st) 86{ 87 sd->sd_ctime.sec = (unsigned int)st->st_ctime; 88 sd->sd_mtime.sec = (unsigned int)st->st_mtime; 89 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st); 90 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st); 91 sd->sd_dev = st->st_dev; 92 sd->sd_ino = st->st_ino; 93 sd->sd_uid = st->st_uid; 94 sd->sd_gid = st->st_gid; 95 sd->sd_size = st->st_size; 96} 97 98int match_stat_data(const struct stat_data *sd, struct stat *st) 99{ 100 int changed = 0; 101 102 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime) 103 changed |= MTIME_CHANGED; 104 if (trust_ctime && check_stat && 105 sd->sd_ctime.sec != (unsigned int)st->st_ctime) 106 changed |= CTIME_CHANGED; 107 108#ifdef USE_NSEC 109 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st)) 110 changed |= MTIME_CHANGED; 111 if (trust_ctime && check_stat && 112 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st)) 113 changed |= CTIME_CHANGED; 114#endif 115 116 if (check_stat) { 117 if (sd->sd_uid != (unsigned int) st->st_uid || 118 sd->sd_gid != (unsigned int) st->st_gid) 119 changed |= OWNER_CHANGED; 120 if (sd->sd_ino != (unsigned int) st->st_ino) 121 changed |= INODE_CHANGED; 122 } 123 124#ifdef USE_STDEV 125 /* 126 * st_dev breaks on network filesystems where different 127 * clients will have different views of what "device" 128 * the filesystem is on 129 */ 130 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev) 131 changed |= INODE_CHANGED; 132#endif 133 134 if (sd->sd_size != (unsigned int) st->st_size) 135 changed |= DATA_CHANGED; 136 137 return changed; 138} 139 140/* 141 * This only updates the "non-critical" parts of the directory 142 * cache, ie the parts that aren't tracked by GIT, and only used 143 * to validate the cache. 144 */ 145void fill_stat_cache_info(struct cache_entry *ce, struct stat *st) 146{ 147 fill_stat_data(&ce->ce_stat_data, st); 148 149 if (assume_unchanged) 150 ce->ce_flags |= CE_VALID; 151 152 if (S_ISREG(st->st_mode)) 153 ce_mark_uptodate(ce); 154} 155 156static int ce_compare_data(const struct cache_entry *ce, struct stat *st) 157{ 158 int match = -1; 159 int fd = git_open_cloexec(ce->name, O_RDONLY); 160 161 if (fd >= 0) { 162 unsigned char sha1[20]; 163 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0)) 164 match = hashcmp(sha1, ce->oid.hash); 165 /* index_fd() closed the file descriptor already */ 166 } 167 return match; 168} 169 170static int ce_compare_link(const struct cache_entry *ce, size_t expected_size) 171{ 172 int match = -1; 173 void *buffer; 174 unsigned long size; 175 enum object_type type; 176 struct strbuf sb = STRBUF_INIT; 177 178 if (strbuf_readlink(&sb, ce->name, expected_size)) 179 return -1; 180 181 buffer = read_sha1_file(ce->oid.hash, &type, &size); 182 if (buffer) { 183 if (size == sb.len) 184 match = memcmp(buffer, sb.buf, size); 185 free(buffer); 186 } 187 strbuf_release(&sb); 188 return match; 189} 190 191static int ce_compare_gitlink(const struct cache_entry *ce) 192{ 193 unsigned char sha1[20]; 194 195 /* 196 * We don't actually require that the .git directory 197 * under GITLINK directory be a valid git directory. It 198 * might even be missing (in case nobody populated that 199 * sub-project). 200 * 201 * If so, we consider it always to match. 202 */ 203 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0) 204 return 0; 205 return hashcmp(sha1, ce->oid.hash); 206} 207 208static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st) 209{ 210 switch (st->st_mode & S_IFMT) { 211 case S_IFREG: 212 if (ce_compare_data(ce, st)) 213 return DATA_CHANGED; 214 break; 215 case S_IFLNK: 216 if (ce_compare_link(ce, xsize_t(st->st_size))) 217 return DATA_CHANGED; 218 break; 219 case S_IFDIR: 220 if (S_ISGITLINK(ce->ce_mode)) 221 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0; 222 default: 223 return TYPE_CHANGED; 224 } 225 return 0; 226} 227 228static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st) 229{ 230 unsigned int changed = 0; 231 232 if (ce->ce_flags & CE_REMOVE) 233 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; 234 235 switch (ce->ce_mode & S_IFMT) { 236 case S_IFREG: 237 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0; 238 /* We consider only the owner x bit to be relevant for 239 * "mode changes" 240 */ 241 if (trust_executable_bit && 242 (0100 & (ce->ce_mode ^ st->st_mode))) 243 changed |= MODE_CHANGED; 244 break; 245 case S_IFLNK: 246 if (!S_ISLNK(st->st_mode) && 247 (has_symlinks || !S_ISREG(st->st_mode))) 248 changed |= TYPE_CHANGED; 249 break; 250 case S_IFGITLINK: 251 /* We ignore most of the st_xxx fields for gitlinks */ 252 if (!S_ISDIR(st->st_mode)) 253 changed |= TYPE_CHANGED; 254 else if (ce_compare_gitlink(ce)) 255 changed |= DATA_CHANGED; 256 return changed; 257 default: 258 die("internal error: ce_mode is %o", ce->ce_mode); 259 } 260 261 changed |= match_stat_data(&ce->ce_stat_data, st); 262 263 /* Racily smudged entry? */ 264 if (!ce->ce_stat_data.sd_size) { 265 if (!is_empty_blob_sha1(ce->oid.hash)) 266 changed |= DATA_CHANGED; 267 } 268 269 return changed; 270} 271 272static int is_racy_stat(const struct index_state *istate, 273 const struct stat_data *sd) 274{ 275 return (istate->timestamp.sec && 276#ifdef USE_NSEC 277 /* nanosecond timestamped files can also be racy! */ 278 (istate->timestamp.sec < sd->sd_mtime.sec || 279 (istate->timestamp.sec == sd->sd_mtime.sec && 280 istate->timestamp.nsec <= sd->sd_mtime.nsec)) 281#else 282 istate->timestamp.sec <= sd->sd_mtime.sec 283#endif 284 ); 285} 286 287static int is_racy_timestamp(const struct index_state *istate, 288 const struct cache_entry *ce) 289{ 290 return (!S_ISGITLINK(ce->ce_mode) && 291 is_racy_stat(istate, &ce->ce_stat_data)); 292} 293 294int match_stat_data_racy(const struct index_state *istate, 295 const struct stat_data *sd, struct stat *st) 296{ 297 if (is_racy_stat(istate, sd)) 298 return MTIME_CHANGED; 299 return match_stat_data(sd, st); 300} 301 302int ie_match_stat(const struct index_state *istate, 303 const struct cache_entry *ce, struct stat *st, 304 unsigned int options) 305{ 306 unsigned int changed; 307 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 308 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE; 309 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 310 311 /* 312 * If it's marked as always valid in the index, it's 313 * valid whatever the checked-out copy says. 314 * 315 * skip-worktree has the same effect with higher precedence 316 */ 317 if (!ignore_skip_worktree && ce_skip_worktree(ce)) 318 return 0; 319 if (!ignore_valid && (ce->ce_flags & CE_VALID)) 320 return 0; 321 322 /* 323 * Intent-to-add entries have not been added, so the index entry 324 * by definition never matches what is in the work tree until it 325 * actually gets added. 326 */ 327 if (ce_intent_to_add(ce)) 328 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED; 329 330 changed = ce_match_stat_basic(ce, st); 331 332 /* 333 * Within 1 second of this sequence: 334 * echo xyzzy >file && git-update-index --add file 335 * running this command: 336 * echo frotz >file 337 * would give a falsely clean cache entry. The mtime and 338 * length match the cache, and other stat fields do not change. 339 * 340 * We could detect this at update-index time (the cache entry 341 * being registered/updated records the same time as "now") 342 * and delay the return from git-update-index, but that would 343 * effectively mean we can make at most one commit per second, 344 * which is not acceptable. Instead, we check cache entries 345 * whose mtime are the same as the index file timestamp more 346 * carefully than others. 347 */ 348 if (!changed && is_racy_timestamp(istate, ce)) { 349 if (assume_racy_is_modified) 350 changed |= DATA_CHANGED; 351 else 352 changed |= ce_modified_check_fs(ce, st); 353 } 354 355 return changed; 356} 357 358int ie_modified(const struct index_state *istate, 359 const struct cache_entry *ce, 360 struct stat *st, unsigned int options) 361{ 362 int changed, changed_fs; 363 364 changed = ie_match_stat(istate, ce, st, options); 365 if (!changed) 366 return 0; 367 /* 368 * If the mode or type has changed, there's no point in trying 369 * to refresh the entry - it's not going to match 370 */ 371 if (changed & (MODE_CHANGED | TYPE_CHANGED)) 372 return changed; 373 374 /* 375 * Immediately after read-tree or update-index --cacheinfo, 376 * the length field is zero, as we have never even read the 377 * lstat(2) information once, and we cannot trust DATA_CHANGED 378 * returned by ie_match_stat() which in turn was returned by 379 * ce_match_stat_basic() to signal that the filesize of the 380 * blob changed. We have to actually go to the filesystem to 381 * see if the contents match, and if so, should answer "unchanged". 382 * 383 * The logic does not apply to gitlinks, as ce_match_stat_basic() 384 * already has checked the actual HEAD from the filesystem in the 385 * subproject. If ie_match_stat() already said it is different, 386 * then we know it is. 387 */ 388 if ((changed & DATA_CHANGED) && 389 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0)) 390 return changed; 391 392 changed_fs = ce_modified_check_fs(ce, st); 393 if (changed_fs) 394 return changed | changed_fs; 395 return 0; 396} 397 398int base_name_compare(const char *name1, int len1, int mode1, 399 const char *name2, int len2, int mode2) 400{ 401 unsigned char c1, c2; 402 int len = len1 < len2 ? len1 : len2; 403 int cmp; 404 405 cmp = memcmp(name1, name2, len); 406 if (cmp) 407 return cmp; 408 c1 = name1[len]; 409 c2 = name2[len]; 410 if (!c1 && S_ISDIR(mode1)) 411 c1 = '/'; 412 if (!c2 && S_ISDIR(mode2)) 413 c2 = '/'; 414 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0; 415} 416 417/* 418 * df_name_compare() is identical to base_name_compare(), except it 419 * compares conflicting directory/file entries as equal. Note that 420 * while a directory name compares as equal to a regular file, they 421 * then individually compare _differently_ to a filename that has 422 * a dot after the basename (because '\0' < '.' < '/'). 423 * 424 * This is used by routines that want to traverse the git namespace 425 * but then handle conflicting entries together when possible. 426 */ 427int df_name_compare(const char *name1, int len1, int mode1, 428 const char *name2, int len2, int mode2) 429{ 430 int len = len1 < len2 ? len1 : len2, cmp; 431 unsigned char c1, c2; 432 433 cmp = memcmp(name1, name2, len); 434 if (cmp) 435 return cmp; 436 /* Directories and files compare equal (same length, same name) */ 437 if (len1 == len2) 438 return 0; 439 c1 = name1[len]; 440 if (!c1 && S_ISDIR(mode1)) 441 c1 = '/'; 442 c2 = name2[len]; 443 if (!c2 && S_ISDIR(mode2)) 444 c2 = '/'; 445 if (c1 == '/' && !c2) 446 return 0; 447 if (c2 == '/' && !c1) 448 return 0; 449 return c1 - c2; 450} 451 452int name_compare(const char *name1, size_t len1, const char *name2, size_t len2) 453{ 454 size_t min_len = (len1 < len2) ? len1 : len2; 455 int cmp = memcmp(name1, name2, min_len); 456 if (cmp) 457 return cmp; 458 if (len1 < len2) 459 return -1; 460 if (len1 > len2) 461 return 1; 462 return 0; 463} 464 465int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2) 466{ 467 int cmp; 468 469 cmp = name_compare(name1, len1, name2, len2); 470 if (cmp) 471 return cmp; 472 473 if (stage1 < stage2) 474 return -1; 475 if (stage1 > stage2) 476 return 1; 477 return 0; 478} 479 480static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage) 481{ 482 int first, last; 483 484 first = 0; 485 last = istate->cache_nr; 486 while (last > first) { 487 int next = (last + first) >> 1; 488 struct cache_entry *ce = istate->cache[next]; 489 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce)); 490 if (!cmp) 491 return next; 492 if (cmp < 0) { 493 last = next; 494 continue; 495 } 496 first = next+1; 497 } 498 return -first-1; 499} 500 501int index_name_pos(const struct index_state *istate, const char *name, int namelen) 502{ 503 return index_name_stage_pos(istate, name, namelen, 0); 504} 505 506int remove_index_entry_at(struct index_state *istate, int pos) 507{ 508 struct cache_entry *ce = istate->cache[pos]; 509 510 record_resolve_undo(istate, ce); 511 remove_name_hash(istate, ce); 512 save_or_free_index_entry(istate, ce); 513 istate->cache_changed |= CE_ENTRY_REMOVED; 514 istate->cache_nr--; 515 if (pos >= istate->cache_nr) 516 return 0; 517 memmove(istate->cache + pos, 518 istate->cache + pos + 1, 519 (istate->cache_nr - pos) * sizeof(struct cache_entry *)); 520 return 1; 521} 522 523/* 524 * Remove all cache entries marked for removal, that is where 525 * CE_REMOVE is set in ce_flags. This is much more effective than 526 * calling remove_index_entry_at() for each entry to be removed. 527 */ 528void remove_marked_cache_entries(struct index_state *istate) 529{ 530 struct cache_entry **ce_array = istate->cache; 531 unsigned int i, j; 532 533 for (i = j = 0; i < istate->cache_nr; i++) { 534 if (ce_array[i]->ce_flags & CE_REMOVE) { 535 remove_name_hash(istate, ce_array[i]); 536 save_or_free_index_entry(istate, ce_array[i]); 537 } 538 else 539 ce_array[j++] = ce_array[i]; 540 } 541 if (j == istate->cache_nr) 542 return; 543 istate->cache_changed |= CE_ENTRY_REMOVED; 544 istate->cache_nr = j; 545} 546 547int remove_file_from_index(struct index_state *istate, const char *path) 548{ 549 int pos = index_name_pos(istate, path, strlen(path)); 550 if (pos < 0) 551 pos = -pos-1; 552 cache_tree_invalidate_path(istate, path); 553 untracked_cache_remove_from_index(istate, path); 554 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 555 remove_index_entry_at(istate, pos); 556 return 0; 557} 558 559static int compare_name(struct cache_entry *ce, const char *path, int namelen) 560{ 561 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); 562} 563 564static int index_name_pos_also_unmerged(struct index_state *istate, 565 const char *path, int namelen) 566{ 567 int pos = index_name_pos(istate, path, namelen); 568 struct cache_entry *ce; 569 570 if (pos >= 0) 571 return pos; 572 573 /* maybe unmerged? */ 574 pos = -1 - pos; 575 if (pos >= istate->cache_nr || 576 compare_name((ce = istate->cache[pos]), path, namelen)) 577 return -1; 578 579 /* order of preference: stage 2, 1, 3 */ 580 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && 581 ce_stage((ce = istate->cache[pos + 1])) == 2 && 582 !compare_name(ce, path, namelen)) 583 pos++; 584 return pos; 585} 586 587static int different_name(struct cache_entry *ce, struct cache_entry *alias) 588{ 589 int len = ce_namelen(ce); 590 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len); 591} 592 593/* 594 * If we add a filename that aliases in the cache, we will use the 595 * name that we already have - but we don't want to update the same 596 * alias twice, because that implies that there were actually two 597 * different files with aliasing names! 598 * 599 * So we use the CE_ADDED flag to verify that the alias was an old 600 * one before we accept it as 601 */ 602static struct cache_entry *create_alias_ce(struct index_state *istate, 603 struct cache_entry *ce, 604 struct cache_entry *alias) 605{ 606 int len; 607 struct cache_entry *new; 608 609 if (alias->ce_flags & CE_ADDED) 610 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); 611 612 /* Ok, create the new entry using the name of the existing alias */ 613 len = ce_namelen(alias); 614 new = xcalloc(1, cache_entry_size(len)); 615 memcpy(new->name, alias->name, len); 616 copy_cache_entry(new, ce); 617 save_or_free_index_entry(istate, ce); 618 return new; 619} 620 621void set_object_name_for_intent_to_add_entry(struct cache_entry *ce) 622{ 623 unsigned char sha1[20]; 624 if (write_sha1_file("", 0, blob_type, sha1)) 625 die("cannot create an empty blob in the object database"); 626 hashcpy(ce->oid.hash, sha1); 627} 628 629int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags) 630{ 631 int size, namelen, was_same; 632 mode_t st_mode = st->st_mode; 633 struct cache_entry *ce, *alias; 634 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY; 635 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); 636 int pretend = flags & ADD_CACHE_PRETEND; 637 int intent_only = flags & ADD_CACHE_INTENT; 638 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| 639 (intent_only ? ADD_CACHE_NEW_ONLY : 0)); 640 641 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 642 return error("%s: can only add regular files, symbolic links or git-directories", path); 643 644 namelen = strlen(path); 645 if (S_ISDIR(st_mode)) { 646 while (namelen && path[namelen-1] == '/') 647 namelen--; 648 } 649 size = cache_entry_size(namelen); 650 ce = xcalloc(1, size); 651 memcpy(ce->name, path, namelen); 652 ce->ce_namelen = namelen; 653 if (!intent_only) 654 fill_stat_cache_info(ce, st); 655 else 656 ce->ce_flags |= CE_INTENT_TO_ADD; 657 658 659 if (trust_executable_bit && has_symlinks) { 660 ce->ce_mode = create_ce_mode(st_mode); 661 } else { 662 /* If there is an existing entry, pick the mode bits and type 663 * from it, otherwise assume unexecutable regular file. 664 */ 665 struct cache_entry *ent; 666 int pos = index_name_pos_also_unmerged(istate, path, namelen); 667 668 ent = (0 <= pos) ? istate->cache[pos] : NULL; 669 ce->ce_mode = ce_mode_from_stat(ent, st_mode); 670 } 671 672 /* When core.ignorecase=true, determine if a directory of the same name but differing 673 * case already exists within the Git repository. If it does, ensure the directory 674 * case of the file being added to the repository matches (is folded into) the existing 675 * entry's directory case. 676 */ 677 if (ignore_case) { 678 adjust_dirname_case(istate, ce->name); 679 } 680 681 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case); 682 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) { 683 /* Nothing changed, really */ 684 if (!S_ISGITLINK(alias->ce_mode)) 685 ce_mark_uptodate(alias); 686 alias->ce_flags |= CE_ADDED; 687 688 free(ce); 689 return 0; 690 } 691 if (!intent_only) { 692 if (index_path(ce->oid.hash, path, st, HASH_WRITE_OBJECT)) { 693 free(ce); 694 return error("unable to index file %s", path); 695 } 696 } else 697 set_object_name_for_intent_to_add_entry(ce); 698 699 if (ignore_case && alias && different_name(ce, alias)) 700 ce = create_alias_ce(istate, ce, alias); 701 ce->ce_flags |= CE_ADDED; 702 703 /* It was suspected to be racily clean, but it turns out to be Ok */ 704 was_same = (alias && 705 !ce_stage(alias) && 706 !oidcmp(&alias->oid, &ce->oid) && 707 ce->ce_mode == alias->ce_mode); 708 709 if (pretend) 710 free(ce); 711 else if (add_index_entry(istate, ce, add_option)) { 712 free(ce); 713 return error("unable to add %s to index", path); 714 } 715 if (verbose && !was_same) 716 printf("add '%s'\n", path); 717 return 0; 718} 719 720int add_file_to_index(struct index_state *istate, const char *path, int flags) 721{ 722 struct stat st; 723 if (lstat(path, &st)) 724 die_errno("unable to stat '%s'", path); 725 return add_to_index(istate, path, &st, flags); 726} 727 728struct cache_entry *make_cache_entry(unsigned int mode, 729 const unsigned char *sha1, const char *path, int stage, 730 unsigned int refresh_options) 731{ 732 int size, len; 733 struct cache_entry *ce, *ret; 734 735 if (!verify_path(path)) { 736 error("Invalid path '%s'", path); 737 return NULL; 738 } 739 740 len = strlen(path); 741 size = cache_entry_size(len); 742 ce = xcalloc(1, size); 743 744 hashcpy(ce->oid.hash, sha1); 745 memcpy(ce->name, path, len); 746 ce->ce_flags = create_ce_flags(stage); 747 ce->ce_namelen = len; 748 ce->ce_mode = create_ce_mode(mode); 749 750 ret = refresh_cache_entry(ce, refresh_options); 751 if (ret != ce) 752 free(ce); 753 return ret; 754} 755 756/* 757 * Chmod an index entry with either +x or -x. 758 * 759 * Returns -1 if the chmod for the particular cache entry failed (if it's 760 * not a regular file), -2 if an invalid flip argument is passed in, 0 761 * otherwise. 762 */ 763int chmod_index_entry(struct index_state *istate, struct cache_entry *ce, 764 char flip) 765{ 766 if (!S_ISREG(ce->ce_mode)) 767 return -1; 768 switch (flip) { 769 case '+': 770 ce->ce_mode |= 0111; 771 break; 772 case '-': 773 ce->ce_mode &= ~0111; 774 break; 775 default: 776 return -2; 777 } 778 cache_tree_invalidate_path(istate, ce->name); 779 ce->ce_flags |= CE_UPDATE_IN_BASE; 780 istate->cache_changed |= CE_ENTRY_CHANGED; 781 782 return 0; 783} 784 785int ce_same_name(const struct cache_entry *a, const struct cache_entry *b) 786{ 787 int len = ce_namelen(a); 788 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 789} 790 791/* 792 * We fundamentally don't like some paths: we don't want 793 * dot or dot-dot anywhere, and for obvious reasons don't 794 * want to recurse into ".git" either. 795 * 796 * Also, we don't want double slashes or slashes at the 797 * end that can make pathnames ambiguous. 798 */ 799static int verify_dotfile(const char *rest) 800{ 801 /* 802 * The first character was '.', but that 803 * has already been discarded, we now test 804 * the rest. 805 */ 806 807 /* "." is not allowed */ 808 if (*rest == '\0' || is_dir_sep(*rest)) 809 return 0; 810 811 switch (*rest) { 812 /* 813 * ".git" followed by NUL or slash is bad. Note that we match 814 * case-insensitively here, even if ignore_case is not set. 815 * This outlaws ".GIT" everywhere out of an abundance of caution, 816 * since there's really no good reason to allow it. 817 */ 818 case 'g': 819 case 'G': 820 if (rest[1] != 'i' && rest[1] != 'I') 821 break; 822 if (rest[2] != 't' && rest[2] != 'T') 823 break; 824 if (rest[3] == '\0' || is_dir_sep(rest[3])) 825 return 0; 826 break; 827 case '.': 828 if (rest[1] == '\0' || is_dir_sep(rest[1])) 829 return 0; 830 } 831 return 1; 832} 833 834int verify_path(const char *path) 835{ 836 char c; 837 838 if (has_dos_drive_prefix(path)) 839 return 0; 840 841 goto inside; 842 for (;;) { 843 if (!c) 844 return 1; 845 if (is_dir_sep(c)) { 846inside: 847 if (protect_hfs && is_hfs_dotgit(path)) 848 return 0; 849 if (protect_ntfs && is_ntfs_dotgit(path)) 850 return 0; 851 c = *path++; 852 if ((c == '.' && !verify_dotfile(path)) || 853 is_dir_sep(c) || c == '\0') 854 return 0; 855 } 856 c = *path++; 857 } 858} 859 860/* 861 * Do we have another file that has the beginning components being a 862 * proper superset of the name we're trying to add? 863 */ 864static int has_file_name(struct index_state *istate, 865 const struct cache_entry *ce, int pos, int ok_to_replace) 866{ 867 int retval = 0; 868 int len = ce_namelen(ce); 869 int stage = ce_stage(ce); 870 const char *name = ce->name; 871 872 while (pos < istate->cache_nr) { 873 struct cache_entry *p = istate->cache[pos++]; 874 875 if (len >= ce_namelen(p)) 876 break; 877 if (memcmp(name, p->name, len)) 878 break; 879 if (ce_stage(p) != stage) 880 continue; 881 if (p->name[len] != '/') 882 continue; 883 if (p->ce_flags & CE_REMOVE) 884 continue; 885 retval = -1; 886 if (!ok_to_replace) 887 break; 888 remove_index_entry_at(istate, --pos); 889 } 890 return retval; 891} 892 893 894/* 895 * Like strcmp(), but also return the offset of the first change. 896 * If strings are equal, return the length. 897 */ 898int strcmp_offset(const char *s1, const char *s2, size_t *first_change) 899{ 900 size_t k; 901 902 if (!first_change) 903 return strcmp(s1, s2); 904 905 for (k = 0; s1[k] == s2[k]; k++) 906 if (s1[k] == '\0') 907 break; 908 909 *first_change = k; 910 return (unsigned char)s1[k] - (unsigned char)s2[k]; 911} 912 913/* 914 * Do we have another file with a pathname that is a proper 915 * subset of the name we're trying to add? 916 * 917 * That is, is there another file in the index with a path 918 * that matches a sub-directory in the given entry? 919 */ 920static int has_dir_name(struct index_state *istate, 921 const struct cache_entry *ce, int pos, int ok_to_replace) 922{ 923 int retval = 0; 924 int stage = ce_stage(ce); 925 const char *name = ce->name; 926 const char *slash = name + ce_namelen(ce); 927 size_t len_eq_last; 928 int cmp_last = 0; 929 930 /* 931 * We are frequently called during an iteration on a sorted 932 * list of pathnames and while building a new index. Therefore, 933 * there is a high probability that this entry will eventually 934 * be appended to the index, rather than inserted in the middle. 935 * If we can confirm that, we can avoid binary searches on the 936 * components of the pathname. 937 * 938 * Compare the entry's full path with the last path in the index. 939 */ 940 if (istate->cache_nr > 0) { 941 cmp_last = strcmp_offset(name, 942 istate->cache[istate->cache_nr - 1]->name, 943 &len_eq_last); 944 if (cmp_last > 0) { 945 if (len_eq_last == 0) { 946 /* 947 * The entry sorts AFTER the last one in the 948 * index and their paths have no common prefix, 949 * so there cannot be a F/D conflict. 950 */ 951 return retval; 952 } else { 953 /* 954 * The entry sorts AFTER the last one in the 955 * index, but has a common prefix. Fall through 956 * to the loop below to disect the entry's path 957 * and see where the difference is. 958 */ 959 } 960 } else if (cmp_last == 0) { 961 /* 962 * The entry exactly matches the last one in the 963 * index, but because of multiple stage and CE_REMOVE 964 * items, we fall through and let the regular search 965 * code handle it. 966 */ 967 } 968 } 969 970 for (;;) { 971 size_t len; 972 973 for (;;) { 974 if (*--slash == '/') 975 break; 976 if (slash <= ce->name) 977 return retval; 978 } 979 len = slash - name; 980 981 if (cmp_last > 0) { 982 /* 983 * (len + 1) is a directory boundary (including 984 * the trailing slash). And since the loop is 985 * decrementing "slash", the first iteration is 986 * the longest directory prefix; subsequent 987 * iterations consider parent directories. 988 */ 989 990 if (len + 1 <= len_eq_last) { 991 /* 992 * The directory prefix (including the trailing 993 * slash) also appears as a prefix in the last 994 * entry, so the remainder cannot collide (because 995 * strcmp said the whole path was greater). 996 * 997 * EQ: last: xxx/A 998 * this: xxx/B 999 *1000 * LT: last: xxx/file_A1001 * this: xxx/file_B1002 */1003 return retval;1004 }10051006 if (len > len_eq_last) {1007 /*1008 * This part of the directory prefix (excluding1009 * the trailing slash) is longer than the known1010 * equal portions, so this sub-directory cannot1011 * collide with a file.1012 *1013 * GT: last: xxxA1014 * this: xxxB/file1015 */1016 return retval;1017 }10181019 if (istate->cache_nr > 0 &&1020 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {1021 /*1022 * The directory prefix lines up with part of1023 * a longer file or directory name, but sorts1024 * after it, so this sub-directory cannot1025 * collide with a file.1026 *1027 * last: xxx/yy-file (because '-' sorts before '/')1028 * this: xxx/yy/abc1029 */1030 return retval;1031 }10321033 /*1034 * This is a possible collision. Fall through and1035 * let the regular search code handle it.1036 *1037 * last: xxx1038 * this: xxx/file1039 */1040 }10411042 pos = index_name_stage_pos(istate, name, len, stage);1043 if (pos >= 0) {1044 /*1045 * Found one, but not so fast. This could1046 * be a marker that says "I was here, but1047 * I am being removed". Such an entry is1048 * not a part of the resulting tree, and1049 * it is Ok to have a directory at the same1050 * path.1051 */1052 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {1053 retval = -1;1054 if (!ok_to_replace)1055 break;1056 remove_index_entry_at(istate, pos);1057 continue;1058 }1059 }1060 else1061 pos = -pos-1;10621063 /*1064 * Trivial optimization: if we find an entry that1065 * already matches the sub-directory, then we know1066 * we're ok, and we can exit.1067 */1068 while (pos < istate->cache_nr) {1069 struct cache_entry *p = istate->cache[pos];1070 if ((ce_namelen(p) <= len) ||1071 (p->name[len] != '/') ||1072 memcmp(p->name, name, len))1073 break; /* not our subdirectory */1074 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))1075 /*1076 * p is at the same stage as our entry, and1077 * is a subdirectory of what we are looking1078 * at, so we cannot have conflicts at our1079 * level or anything shorter.1080 */1081 return retval;1082 pos++;1083 }1084 }1085 return retval;1086}10871088/* We may be in a situation where we already have path/file and path1089 * is being added, or we already have path and path/file is being1090 * added. Either one would result in a nonsense tree that has path1091 * twice when git-write-tree tries to write it out. Prevent it.1092 *1093 * If ok-to-replace is specified, we remove the conflicting entries1094 * from the cache so the caller should recompute the insert position.1095 * When this happens, we return non-zero.1096 */1097static int check_file_directory_conflict(struct index_state *istate,1098 const struct cache_entry *ce,1099 int pos, int ok_to_replace)1100{1101 int retval;11021103 /*1104 * When ce is an "I am going away" entry, we allow it to be added1105 */1106 if (ce->ce_flags & CE_REMOVE)1107 return 0;11081109 /*1110 * We check if the path is a sub-path of a subsequent pathname1111 * first, since removing those will not change the position1112 * in the array.1113 */1114 retval = has_file_name(istate, ce, pos, ok_to_replace);11151116 /*1117 * Then check if the path might have a clashing sub-directory1118 * before it.1119 */1120 return retval + has_dir_name(istate, ce, pos, ok_to_replace);1121}11221123static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)1124{1125 int pos;1126 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;1127 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;1128 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;1129 int new_only = option & ADD_CACHE_NEW_ONLY;11301131 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))1132 cache_tree_invalidate_path(istate, ce->name);11331134 /*1135 * If this entry's path sorts after the last entry in the index,1136 * we can avoid searching for it.1137 */1138 if (istate->cache_nr > 0 &&1139 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)1140 pos = -istate->cache_nr - 1;1141 else1142 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));11431144 /* existing match? Just replace it. */1145 if (pos >= 0) {1146 if (!new_only)1147 replace_index_entry(istate, pos, ce);1148 return 0;1149 }1150 pos = -pos-1;11511152 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))1153 untracked_cache_add_to_index(istate, ce->name);11541155 /*1156 * Inserting a merged entry ("stage 0") into the index1157 * will always replace all non-merged entries..1158 */1159 if (pos < istate->cache_nr && ce_stage(ce) == 0) {1160 while (ce_same_name(istate->cache[pos], ce)) {1161 ok_to_add = 1;1162 if (!remove_index_entry_at(istate, pos))1163 break;1164 }1165 }11661167 if (!ok_to_add)1168 return -1;1169 if (!verify_path(ce->name))1170 return error("Invalid path '%s'", ce->name);11711172 if (!skip_df_check &&1173 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {1174 if (!ok_to_replace)1175 return error("'%s' appears as both a file and as a directory",1176 ce->name);1177 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));1178 pos = -pos-1;1179 }1180 return pos + 1;1181}11821183int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)1184{1185 int pos;11861187 if (option & ADD_CACHE_JUST_APPEND)1188 pos = istate->cache_nr;1189 else {1190 int ret;1191 ret = add_index_entry_with_check(istate, ce, option);1192 if (ret <= 0)1193 return ret;1194 pos = ret - 1;1195 }11961197 /* Make sure the array is big enough .. */1198 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);11991200 /* Add it in.. */1201 istate->cache_nr++;1202 if (istate->cache_nr > pos + 1)1203 memmove(istate->cache + pos + 1,1204 istate->cache + pos,1205 (istate->cache_nr - pos - 1) * sizeof(ce));1206 set_index_entry(istate, pos, ce);1207 istate->cache_changed |= CE_ENTRY_ADDED;1208 return 0;1209}12101211/*1212 * "refresh" does not calculate a new sha1 file or bring the1213 * cache up-to-date for mode/content changes. But what it1214 * _does_ do is to "re-match" the stat information of a file1215 * with the cache, so that you can refresh the cache for a1216 * file that hasn't been changed but where the stat entry is1217 * out of date.1218 *1219 * For example, you'd want to do this after doing a "git-read-tree",1220 * to link up the stat cache details with the proper files.1221 */1222static struct cache_entry *refresh_cache_ent(struct index_state *istate,1223 struct cache_entry *ce,1224 unsigned int options, int *err,1225 int *changed_ret)1226{1227 struct stat st;1228 struct cache_entry *updated;1229 int changed, size;1230 int refresh = options & CE_MATCH_REFRESH;1231 int ignore_valid = options & CE_MATCH_IGNORE_VALID;1232 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;1233 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;12341235 if (!refresh || ce_uptodate(ce))1236 return ce;12371238 /*1239 * CE_VALID or CE_SKIP_WORKTREE means the user promised us1240 * that the change to the work tree does not matter and told1241 * us not to worry.1242 */1243 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {1244 ce_mark_uptodate(ce);1245 return ce;1246 }1247 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {1248 ce_mark_uptodate(ce);1249 return ce;1250 }12511252 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {1253 if (ignore_missing)1254 return ce;1255 if (err)1256 *err = ENOENT;1257 return NULL;1258 }12591260 if (lstat(ce->name, &st) < 0) {1261 if (ignore_missing && errno == ENOENT)1262 return ce;1263 if (err)1264 *err = errno;1265 return NULL;1266 }12671268 changed = ie_match_stat(istate, ce, &st, options);1269 if (changed_ret)1270 *changed_ret = changed;1271 if (!changed) {1272 /*1273 * The path is unchanged. If we were told to ignore1274 * valid bit, then we did the actual stat check and1275 * found that the entry is unmodified. If the entry1276 * is not marked VALID, this is the place to mark it1277 * valid again, under "assume unchanged" mode.1278 */1279 if (ignore_valid && assume_unchanged &&1280 !(ce->ce_flags & CE_VALID))1281 ; /* mark this one VALID again */1282 else {1283 /*1284 * We do not mark the index itself "modified"1285 * because CE_UPTODATE flag is in-core only;1286 * we are not going to write this change out.1287 */1288 if (!S_ISGITLINK(ce->ce_mode))1289 ce_mark_uptodate(ce);1290 return ce;1291 }1292 }12931294 if (ie_modified(istate, ce, &st, options)) {1295 if (err)1296 *err = EINVAL;1297 return NULL;1298 }12991300 size = ce_size(ce);1301 updated = xmalloc(size);1302 memcpy(updated, ce, size);1303 fill_stat_cache_info(updated, &st);1304 /*1305 * If ignore_valid is not set, we should leave CE_VALID bit1306 * alone. Otherwise, paths marked with --no-assume-unchanged1307 * (i.e. things to be edited) will reacquire CE_VALID bit1308 * automatically, which is not really what we want.1309 */1310 if (!ignore_valid && assume_unchanged &&1311 !(ce->ce_flags & CE_VALID))1312 updated->ce_flags &= ~CE_VALID;13131314 /* istate->cache_changed is updated in the caller */1315 return updated;1316}13171318static void show_file(const char * fmt, const char * name, int in_porcelain,1319 int * first, const char *header_msg)1320{1321 if (in_porcelain && *first && header_msg) {1322 printf("%s\n", header_msg);1323 *first = 0;1324 }1325 printf(fmt, name);1326}13271328int refresh_index(struct index_state *istate, unsigned int flags,1329 const struct pathspec *pathspec,1330 char *seen, const char *header_msg)1331{1332 int i;1333 int has_errors = 0;1334 int really = (flags & REFRESH_REALLY) != 0;1335 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;1336 int quiet = (flags & REFRESH_QUIET) != 0;1337 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;1338 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;1339 int first = 1;1340 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);1341 unsigned int options = (CE_MATCH_REFRESH |1342 (really ? CE_MATCH_IGNORE_VALID : 0) |1343 (not_new ? CE_MATCH_IGNORE_MISSING : 0));1344 const char *modified_fmt;1345 const char *deleted_fmt;1346 const char *typechange_fmt;1347 const char *added_fmt;1348 const char *unmerged_fmt;13491350 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");1351 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");1352 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");1353 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");1354 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");1355 for (i = 0; i < istate->cache_nr; i++) {1356 struct cache_entry *ce, *new;1357 int cache_errno = 0;1358 int changed = 0;1359 int filtered = 0;13601361 ce = istate->cache[i];1362 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))1363 continue;13641365 if (pathspec && !ce_path_match(ce, pathspec, seen))1366 filtered = 1;13671368 if (ce_stage(ce)) {1369 while ((i < istate->cache_nr) &&1370 ! strcmp(istate->cache[i]->name, ce->name))1371 i++;1372 i--;1373 if (allow_unmerged)1374 continue;1375 if (!filtered)1376 show_file(unmerged_fmt, ce->name, in_porcelain,1377 &first, header_msg);1378 has_errors = 1;1379 continue;1380 }13811382 if (filtered)1383 continue;13841385 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);1386 if (new == ce)1387 continue;1388 if (!new) {1389 const char *fmt;13901391 if (really && cache_errno == EINVAL) {1392 /* If we are doing --really-refresh that1393 * means the index is not valid anymore.1394 */1395 ce->ce_flags &= ~CE_VALID;1396 ce->ce_flags |= CE_UPDATE_IN_BASE;1397 istate->cache_changed |= CE_ENTRY_CHANGED;1398 }1399 if (quiet)1400 continue;14011402 if (cache_errno == ENOENT)1403 fmt = deleted_fmt;1404 else if (ce_intent_to_add(ce))1405 fmt = added_fmt; /* must be before other checks */1406 else if (changed & TYPE_CHANGED)1407 fmt = typechange_fmt;1408 else1409 fmt = modified_fmt;1410 show_file(fmt,1411 ce->name, in_porcelain, &first, header_msg);1412 has_errors = 1;1413 continue;1414 }14151416 replace_index_entry(istate, i, new);1417 }1418 return has_errors;1419}14201421struct cache_entry *refresh_cache_entry(struct cache_entry *ce,1422 unsigned int options)1423{1424 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);1425}142614271428/*****************************************************************1429 * Index File I/O1430 *****************************************************************/14311432#define INDEX_FORMAT_DEFAULT 314331434static unsigned int get_index_format_default(void)1435{1436 char *envversion = getenv("GIT_INDEX_VERSION");1437 char *endp;1438 int value;1439 unsigned int version = INDEX_FORMAT_DEFAULT;14401441 if (!envversion) {1442 if (!git_config_get_int("index.version", &value))1443 version = value;1444 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {1445 warning(_("index.version set, but the value is invalid.\n"1446 "Using version %i"), INDEX_FORMAT_DEFAULT);1447 return INDEX_FORMAT_DEFAULT;1448 }1449 return version;1450 }14511452 version = strtoul(envversion, &endp, 10);1453 if (*endp ||1454 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {1455 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"1456 "Using version %i"), INDEX_FORMAT_DEFAULT);1457 version = INDEX_FORMAT_DEFAULT;1458 }1459 return version;1460}14611462/*1463 * dev/ino/uid/gid/size are also just tracked to the low 32 bits1464 * Again - this is just a (very strong in practice) heuristic that1465 * the inode hasn't changed.1466 *1467 * We save the fields in big-endian order to allow using the1468 * index file over NFS transparently.1469 */1470struct ondisk_cache_entry {1471 struct cache_time ctime;1472 struct cache_time mtime;1473 uint32_t dev;1474 uint32_t ino;1475 uint32_t mode;1476 uint32_t uid;1477 uint32_t gid;1478 uint32_t size;1479 unsigned char sha1[20];1480 uint16_t flags;1481 char name[FLEX_ARRAY]; /* more */1482};14831484/*1485 * This struct is used when CE_EXTENDED bit is 11486 * The struct must match ondisk_cache_entry exactly from1487 * ctime till flags1488 */1489struct ondisk_cache_entry_extended {1490 struct cache_time ctime;1491 struct cache_time mtime;1492 uint32_t dev;1493 uint32_t ino;1494 uint32_t mode;1495 uint32_t uid;1496 uint32_t gid;1497 uint32_t size;1498 unsigned char sha1[20];1499 uint16_t flags;1500 uint16_t flags2;1501 char name[FLEX_ARRAY]; /* more */1502};15031504/* These are only used for v3 or lower */1505#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)1506#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)1507#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)1508#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \1509 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \1510 ondisk_cache_entry_size(ce_namelen(ce)))15111512/* Allow fsck to force verification of the index checksum. */1513int verify_index_checksum;15141515static int verify_hdr(struct cache_header *hdr, unsigned long size)1516{1517 git_SHA_CTX c;1518 unsigned char sha1[20];1519 int hdr_version;15201521 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))1522 return error("bad signature");1523 hdr_version = ntohl(hdr->hdr_version);1524 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)1525 return error("bad index version %d", hdr_version);15261527 if (!verify_index_checksum)1528 return 0;15291530 git_SHA1_Init(&c);1531 git_SHA1_Update(&c, hdr, size - 20);1532 git_SHA1_Final(sha1, &c);1533 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))1534 return error("bad index file sha1 signature");1535 return 0;1536}15371538static int read_index_extension(struct index_state *istate,1539 const char *ext, void *data, unsigned long sz)1540{1541 switch (CACHE_EXT(ext)) {1542 case CACHE_EXT_TREE:1543 istate->cache_tree = cache_tree_read(data, sz);1544 break;1545 case CACHE_EXT_RESOLVE_UNDO:1546 istate->resolve_undo = resolve_undo_read(data, sz);1547 break;1548 case CACHE_EXT_LINK:1549 if (read_link_extension(istate, data, sz))1550 return -1;1551 break;1552 case CACHE_EXT_UNTRACKED:1553 istate->untracked = read_untracked_extension(data, sz);1554 break;1555 default:1556 if (*ext < 'A' || 'Z' < *ext)1557 return error("index uses %.4s extension, which we do not understand",1558 ext);1559 fprintf(stderr, "ignoring %.4s extension\n", ext);1560 break;1561 }1562 return 0;1563}15641565int hold_locked_index(struct lock_file *lk, int lock_flags)1566{1567 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);1568}15691570int read_index(struct index_state *istate)1571{1572 return read_index_from(istate, get_index_file());1573}15741575static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,1576 unsigned int flags,1577 const char *name,1578 size_t len)1579{1580 struct cache_entry *ce = xmalloc(cache_entry_size(len));15811582 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);1583 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);1584 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);1585 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);1586 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);1587 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);1588 ce->ce_mode = get_be32(&ondisk->mode);1589 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);1590 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);1591 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);1592 ce->ce_flags = flags & ~CE_NAMEMASK;1593 ce->ce_namelen = len;1594 ce->index = 0;1595 hashcpy(ce->oid.hash, ondisk->sha1);1596 memcpy(ce->name, name, len);1597 ce->name[len] = '\0';1598 return ce;1599}16001601/*1602 * Adjacent cache entries tend to share the leading paths, so it makes1603 * sense to only store the differences in later entries. In the v41604 * on-disk format of the index, each on-disk cache entry stores the1605 * number of bytes to be stripped from the end of the previous name,1606 * and the bytes to append to the result, to come up with its name.1607 */1608static unsigned long expand_name_field(struct strbuf *name, const char *cp_)1609{1610 const unsigned char *ep, *cp = (const unsigned char *)cp_;1611 size_t len = decode_varint(&cp);16121613 if (name->len < len)1614 die("malformed name field in the index");1615 strbuf_remove(name, name->len - len, len);1616 for (ep = cp; *ep; ep++)1617 ; /* find the end */1618 strbuf_add(name, cp, ep - cp);1619 return (const char *)ep + 1 - cp_;1620}16211622static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,1623 unsigned long *ent_size,1624 struct strbuf *previous_name)1625{1626 struct cache_entry *ce;1627 size_t len;1628 const char *name;1629 unsigned int flags;16301631 /* On-disk flags are just 16 bits */1632 flags = get_be16(&ondisk->flags);1633 len = flags & CE_NAMEMASK;16341635 if (flags & CE_EXTENDED) {1636 struct ondisk_cache_entry_extended *ondisk2;1637 int extended_flags;1638 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1639 extended_flags = get_be16(&ondisk2->flags2) << 16;1640 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1641 if (extended_flags & ~CE_EXTENDED_FLAGS)1642 die("Unknown index entry format %08x", extended_flags);1643 flags |= extended_flags;1644 name = ondisk2->name;1645 }1646 else1647 name = ondisk->name;16481649 if (!previous_name) {1650 /* v3 and earlier */1651 if (len == CE_NAMEMASK)1652 len = strlen(name);1653 ce = cache_entry_from_ondisk(ondisk, flags, name, len);16541655 *ent_size = ondisk_ce_size(ce);1656 } else {1657 unsigned long consumed;1658 consumed = expand_name_field(previous_name, name);1659 ce = cache_entry_from_ondisk(ondisk, flags,1660 previous_name->buf,1661 previous_name->len);16621663 *ent_size = (name - ((char *)ondisk)) + consumed;1664 }1665 return ce;1666}16671668static void check_ce_order(struct index_state *istate)1669{1670 unsigned int i;16711672 for (i = 1; i < istate->cache_nr; i++) {1673 struct cache_entry *ce = istate->cache[i - 1];1674 struct cache_entry *next_ce = istate->cache[i];1675 int name_compare = strcmp(ce->name, next_ce->name);16761677 if (0 < name_compare)1678 die("unordered stage entries in index");1679 if (!name_compare) {1680 if (!ce_stage(ce))1681 die("multiple stage entries for merged file '%s'",1682 ce->name);1683 if (ce_stage(ce) > ce_stage(next_ce))1684 die("unordered stage entries for '%s'",1685 ce->name);1686 }1687 }1688}16891690static void tweak_untracked_cache(struct index_state *istate)1691{1692 switch (git_config_get_untracked_cache()) {1693 case -1: /* keep: do nothing */1694 break;1695 case 0: /* false */1696 remove_untracked_cache(istate);1697 break;1698 case 1: /* true */1699 add_untracked_cache(istate);1700 break;1701 default: /* unknown value: do nothing */1702 break;1703 }1704}17051706static void tweak_split_index(struct index_state *istate)1707{1708 switch (git_config_get_split_index()) {1709 case -1: /* unset: do nothing */1710 break;1711 case 0: /* false */1712 remove_split_index(istate);1713 break;1714 case 1: /* true */1715 add_split_index(istate);1716 break;1717 default: /* unknown value: do nothing */1718 break;1719 }1720}17211722static void post_read_index_from(struct index_state *istate)1723{1724 check_ce_order(istate);1725 tweak_untracked_cache(istate);1726 tweak_split_index(istate);1727}17281729/* remember to discard_cache() before reading a different cache! */1730int do_read_index(struct index_state *istate, const char *path, int must_exist)1731{1732 int fd, i;1733 struct stat st;1734 unsigned long src_offset;1735 struct cache_header *hdr;1736 void *mmap;1737 size_t mmap_size;1738 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;17391740 if (istate->initialized)1741 return istate->cache_nr;17421743 istate->timestamp.sec = 0;1744 istate->timestamp.nsec = 0;1745 fd = open(path, O_RDONLY);1746 if (fd < 0) {1747 if (!must_exist && errno == ENOENT)1748 return 0;1749 die_errno("%s: index file open failed", path);1750 }17511752 if (fstat(fd, &st))1753 die_errno("cannot stat the open index");17541755 mmap_size = xsize_t(st.st_size);1756 if (mmap_size < sizeof(struct cache_header) + 20)1757 die("index file smaller than expected");17581759 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);1760 if (mmap == MAP_FAILED)1761 die_errno("unable to map index file");1762 close(fd);17631764 hdr = mmap;1765 if (verify_hdr(hdr, mmap_size) < 0)1766 goto unmap;17671768 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);1769 istate->version = ntohl(hdr->hdr_version);1770 istate->cache_nr = ntohl(hdr->hdr_entries);1771 istate->cache_alloc = alloc_nr(istate->cache_nr);1772 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));1773 istate->initialized = 1;17741775 if (istate->version == 4)1776 previous_name = &previous_name_buf;1777 else1778 previous_name = NULL;17791780 src_offset = sizeof(*hdr);1781 for (i = 0; i < istate->cache_nr; i++) {1782 struct ondisk_cache_entry *disk_ce;1783 struct cache_entry *ce;1784 unsigned long consumed;17851786 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);1787 ce = create_from_disk(disk_ce, &consumed, previous_name);1788 set_index_entry(istate, i, ce);17891790 src_offset += consumed;1791 }1792 strbuf_release(&previous_name_buf);1793 istate->timestamp.sec = st.st_mtime;1794 istate->timestamp.nsec = ST_MTIME_NSEC(st);17951796 while (src_offset <= mmap_size - 20 - 8) {1797 /* After an array of active_nr index entries,1798 * there can be arbitrary number of extended1799 * sections, each of which is prefixed with1800 * extension name (4-byte) and section length1801 * in 4-byte network byte order.1802 */1803 uint32_t extsize;1804 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);1805 extsize = ntohl(extsize);1806 if (read_index_extension(istate,1807 (const char *) mmap + src_offset,1808 (char *) mmap + src_offset + 8,1809 extsize) < 0)1810 goto unmap;1811 src_offset += 8;1812 src_offset += extsize;1813 }1814 munmap(mmap, mmap_size);1815 return istate->cache_nr;18161817unmap:1818 munmap(mmap, mmap_size);1819 die("index file corrupt");1820}18211822/*1823 * Signal that the shared index is used by updating its mtime.1824 *1825 * This way, shared index can be removed if they have not been used1826 * for some time.1827 */1828static void freshen_shared_index(char *base_sha1_hex, int warn)1829{1830 char *shared_index = git_pathdup("sharedindex.%s", base_sha1_hex);1831 if (!check_and_freshen_file(shared_index, 1) && warn)1832 warning("could not freshen shared index '%s'", shared_index);1833 free(shared_index);1834}18351836int read_index_from(struct index_state *istate, const char *path)1837{1838 struct split_index *split_index;1839 int ret;1840 char *base_sha1_hex;1841 const char *base_path;18421843 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */1844 if (istate->initialized)1845 return istate->cache_nr;18461847 ret = do_read_index(istate, path, 0);18481849 split_index = istate->split_index;1850 if (!split_index || is_null_sha1(split_index->base_sha1)) {1851 post_read_index_from(istate);1852 return ret;1853 }18541855 if (split_index->base)1856 discard_index(split_index->base);1857 else1858 split_index->base = xcalloc(1, sizeof(*split_index->base));18591860 base_sha1_hex = sha1_to_hex(split_index->base_sha1);1861 base_path = git_path("sharedindex.%s", base_sha1_hex);1862 ret = do_read_index(split_index->base, base_path, 1);1863 if (hashcmp(split_index->base_sha1, split_index->base->sha1))1864 die("broken index, expect %s in %s, got %s",1865 base_sha1_hex, base_path,1866 sha1_to_hex(split_index->base->sha1));18671868 freshen_shared_index(base_sha1_hex, 0);1869 merge_base_index(istate);1870 post_read_index_from(istate);1871 return ret;1872}18731874int is_index_unborn(struct index_state *istate)1875{1876 return (!istate->cache_nr && !istate->timestamp.sec);1877}18781879int discard_index(struct index_state *istate)1880{1881 int i;18821883 for (i = 0; i < istate->cache_nr; i++) {1884 if (istate->cache[i]->index &&1885 istate->split_index &&1886 istate->split_index->base &&1887 istate->cache[i]->index <= istate->split_index->base->cache_nr &&1888 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])1889 continue;1890 free(istate->cache[i]);1891 }1892 resolve_undo_clear_index(istate);1893 istate->cache_nr = 0;1894 istate->cache_changed = 0;1895 istate->timestamp.sec = 0;1896 istate->timestamp.nsec = 0;1897 free_name_hash(istate);1898 cache_tree_free(&(istate->cache_tree));1899 istate->initialized = 0;1900 free(istate->cache);1901 istate->cache = NULL;1902 istate->cache_alloc = 0;1903 discard_split_index(istate);1904 free_untracked_cache(istate->untracked);1905 istate->untracked = NULL;1906 return 0;1907}19081909int unmerged_index(const struct index_state *istate)1910{1911 int i;1912 for (i = 0; i < istate->cache_nr; i++) {1913 if (ce_stage(istate->cache[i]))1914 return 1;1915 }1916 return 0;1917}19181919#define WRITE_BUFFER_SIZE 81921920static unsigned char write_buffer[WRITE_BUFFER_SIZE];1921static unsigned long write_buffer_len;19221923static int ce_write_flush(git_SHA_CTX *context, int fd)1924{1925 unsigned int buffered = write_buffer_len;1926 if (buffered) {1927 git_SHA1_Update(context, write_buffer, buffered);1928 if (write_in_full(fd, write_buffer, buffered) != buffered)1929 return -1;1930 write_buffer_len = 0;1931 }1932 return 0;1933}19341935static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)1936{1937 while (len) {1938 unsigned int buffered = write_buffer_len;1939 unsigned int partial = WRITE_BUFFER_SIZE - buffered;1940 if (partial > len)1941 partial = len;1942 memcpy(write_buffer + buffered, data, partial);1943 buffered += partial;1944 if (buffered == WRITE_BUFFER_SIZE) {1945 write_buffer_len = buffered;1946 if (ce_write_flush(context, fd))1947 return -1;1948 buffered = 0;1949 }1950 write_buffer_len = buffered;1951 len -= partial;1952 data = (char *) data + partial;1953 }1954 return 0;1955}19561957static int write_index_ext_header(git_SHA_CTX *context, int fd,1958 unsigned int ext, unsigned int sz)1959{1960 ext = htonl(ext);1961 sz = htonl(sz);1962 return ((ce_write(context, fd, &ext, 4) < 0) ||1963 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;1964}19651966static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)1967{1968 unsigned int left = write_buffer_len;19691970 if (left) {1971 write_buffer_len = 0;1972 git_SHA1_Update(context, write_buffer, left);1973 }19741975 /* Flush first if not enough space for SHA1 signature */1976 if (left + 20 > WRITE_BUFFER_SIZE) {1977 if (write_in_full(fd, write_buffer, left) != left)1978 return -1;1979 left = 0;1980 }19811982 /* Append the SHA1 signature at the end */1983 git_SHA1_Final(write_buffer + left, context);1984 hashcpy(sha1, write_buffer + left);1985 left += 20;1986 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;1987}19881989static void ce_smudge_racily_clean_entry(struct cache_entry *ce)1990{1991 /*1992 * The only thing we care about in this function is to smudge the1993 * falsely clean entry due to touch-update-touch race, so we leave1994 * everything else as they are. We are called for entries whose1995 * ce_stat_data.sd_mtime match the index file mtime.1996 *1997 * Note that this actually does not do much for gitlinks, for1998 * which ce_match_stat_basic() always goes to the actual1999 * contents. The caller checks with is_racy_timestamp() which2000 * always says "no" for gitlinks, so we are not called for them ;-)2001 */2002 struct stat st;20032004 if (lstat(ce->name, &st) < 0)2005 return;2006 if (ce_match_stat_basic(ce, &st))2007 return;2008 if (ce_modified_check_fs(ce, &st)) {2009 /* This is "racily clean"; smudge it. Note that this2010 * is a tricky code. At first glance, it may appear2011 * that it can break with this sequence:2012 *2013 * $ echo xyzzy >frotz2014 * $ git-update-index --add frotz2015 * $ : >frotz2016 * $ sleep 32017 * $ echo filfre >nitfol2018 * $ git-update-index --add nitfol2019 *2020 * but it does not. When the second update-index runs,2021 * it notices that the entry "frotz" has the same timestamp2022 * as index, and if we were to smudge it by resetting its2023 * size to zero here, then the object name recorded2024 * in index is the 6-byte file but the cached stat information2025 * becomes zero --- which would then match what we would2026 * obtain from the filesystem next time we stat("frotz").2027 *2028 * However, the second update-index, before calling2029 * this function, notices that the cached size is 62030 * bytes and what is on the filesystem is an empty2031 * file, and never calls us, so the cached size information2032 * for "frotz" stays 6 which does not match the filesystem.2033 */2034 ce->ce_stat_data.sd_size = 0;2035 }2036}20372038/* Copy miscellaneous fields but not the name */2039static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,2040 struct cache_entry *ce)2041{2042 short flags;20432044 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);2045 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);2046 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);2047 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);2048 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);2049 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);2050 ondisk->mode = htonl(ce->ce_mode);2051 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);2052 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);2053 ondisk->size = htonl(ce->ce_stat_data.sd_size);2054 hashcpy(ondisk->sha1, ce->oid.hash);20552056 flags = ce->ce_flags & ~CE_NAMEMASK;2057 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));2058 ondisk->flags = htons(flags);2059 if (ce->ce_flags & CE_EXTENDED) {2060 struct ondisk_cache_entry_extended *ondisk2;2061 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;2062 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);2063 return ondisk2->name;2064 }2065 else {2066 return ondisk->name;2067 }2068}20692070static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,2071 struct strbuf *previous_name)2072{2073 int size;2074 struct ondisk_cache_entry *ondisk;2075 int saved_namelen = saved_namelen; /* compiler workaround */2076 char *name;2077 int result;20782079 if (ce->ce_flags & CE_STRIP_NAME) {2080 saved_namelen = ce_namelen(ce);2081 ce->ce_namelen = 0;2082 }20832084 if (!previous_name) {2085 size = ondisk_ce_size(ce);2086 ondisk = xcalloc(1, size);2087 name = copy_cache_entry_to_ondisk(ondisk, ce);2088 memcpy(name, ce->name, ce_namelen(ce));2089 } else {2090 int common, to_remove, prefix_size;2091 unsigned char to_remove_vi[16];2092 for (common = 0;2093 (ce->name[common] &&2094 common < previous_name->len &&2095 ce->name[common] == previous_name->buf[common]);2096 common++)2097 ; /* still matching */2098 to_remove = previous_name->len - common;2099 prefix_size = encode_varint(to_remove, to_remove_vi);21002101 if (ce->ce_flags & CE_EXTENDED)2102 size = offsetof(struct ondisk_cache_entry_extended, name);2103 else2104 size = offsetof(struct ondisk_cache_entry, name);2105 size += prefix_size + (ce_namelen(ce) - common + 1);21062107 ondisk = xcalloc(1, size);2108 name = copy_cache_entry_to_ondisk(ondisk, ce);2109 memcpy(name, to_remove_vi, prefix_size);2110 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);21112112 strbuf_splice(previous_name, common, to_remove,2113 ce->name + common, ce_namelen(ce) - common);2114 }2115 if (ce->ce_flags & CE_STRIP_NAME) {2116 ce->ce_namelen = saved_namelen;2117 ce->ce_flags &= ~CE_STRIP_NAME;2118 }21192120 result = ce_write(c, fd, ondisk, size);2121 free(ondisk);2122 return result;2123}21242125/*2126 * This function verifies if index_state has the correct sha1 of the2127 * index file. Don't die if we have any other failure, just return 0.2128 */2129static int verify_index_from(const struct index_state *istate, const char *path)2130{2131 int fd;2132 ssize_t n;2133 struct stat st;2134 unsigned char sha1[20];21352136 if (!istate->initialized)2137 return 0;21382139 fd = open(path, O_RDONLY);2140 if (fd < 0)2141 return 0;21422143 if (fstat(fd, &st))2144 goto out;21452146 if (st.st_size < sizeof(struct cache_header) + 20)2147 goto out;21482149 n = pread_in_full(fd, sha1, 20, st.st_size - 20);2150 if (n != 20)2151 goto out;21522153 if (hashcmp(istate->sha1, sha1))2154 goto out;21552156 close(fd);2157 return 1;21582159out:2160 close(fd);2161 return 0;2162}21632164static int verify_index(const struct index_state *istate)2165{2166 return verify_index_from(istate, get_index_file());2167}21682169static int has_racy_timestamp(struct index_state *istate)2170{2171 int entries = istate->cache_nr;2172 int i;21732174 for (i = 0; i < entries; i++) {2175 struct cache_entry *ce = istate->cache[i];2176 if (is_racy_timestamp(istate, ce))2177 return 1;2178 }2179 return 0;2180}21812182/*2183 * Opportunistically update the index but do not complain if we can't2184 */2185void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)2186{2187 if ((istate->cache_changed || has_racy_timestamp(istate)) &&2188 verify_index(istate) &&2189 write_locked_index(istate, lockfile, COMMIT_LOCK))2190 rollback_lock_file(lockfile);2191}21922193static int do_write_index(struct index_state *istate, struct tempfile *tempfile,2194 int strip_extensions)2195{2196 int newfd = tempfile->fd;2197 git_SHA_CTX c;2198 struct cache_header hdr;2199 int i, err, removed, extended, hdr_version;2200 struct cache_entry **cache = istate->cache;2201 int entries = istate->cache_nr;2202 struct stat st;2203 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;22042205 for (i = removed = extended = 0; i < entries; i++) {2206 if (cache[i]->ce_flags & CE_REMOVE)2207 removed++;22082209 /* reduce extended entries if possible */2210 cache[i]->ce_flags &= ~CE_EXTENDED;2211 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {2212 extended++;2213 cache[i]->ce_flags |= CE_EXTENDED;2214 }2215 }22162217 if (!istate->version) {2218 istate->version = get_index_format_default();2219 if (getenv("GIT_TEST_SPLIT_INDEX"))2220 init_split_index(istate);2221 }22222223 /* demote version 3 to version 2 when the latter suffices */2224 if (istate->version == 3 || istate->version == 2)2225 istate->version = extended ? 3 : 2;22262227 hdr_version = istate->version;22282229 hdr.hdr_signature = htonl(CACHE_SIGNATURE);2230 hdr.hdr_version = htonl(hdr_version);2231 hdr.hdr_entries = htonl(entries - removed);22322233 git_SHA1_Init(&c);2234 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)2235 return -1;22362237 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;2238 for (i = 0; i < entries; i++) {2239 struct cache_entry *ce = cache[i];2240 if (ce->ce_flags & CE_REMOVE)2241 continue;2242 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))2243 ce_smudge_racily_clean_entry(ce);2244 if (is_null_oid(&ce->oid)) {2245 static const char msg[] = "cache entry has null sha1: %s";2246 static int allow = -1;22472248 if (allow < 0)2249 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);2250 if (allow)2251 warning(msg, ce->name);2252 else2253 return error(msg, ce->name);2254 }2255 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)2256 return -1;2257 }2258 strbuf_release(&previous_name_buf);22592260 /* Write extension data here */2261 if (!strip_extensions && istate->split_index) {2262 struct strbuf sb = STRBUF_INIT;22632264 err = write_link_extension(&sb, istate) < 0 ||2265 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,2266 sb.len) < 0 ||2267 ce_write(&c, newfd, sb.buf, sb.len) < 0;2268 strbuf_release(&sb);2269 if (err)2270 return -1;2271 }2272 if (!strip_extensions && istate->cache_tree) {2273 struct strbuf sb = STRBUF_INIT;22742275 cache_tree_write(&sb, istate->cache_tree);2276 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 02277 || ce_write(&c, newfd, sb.buf, sb.len) < 0;2278 strbuf_release(&sb);2279 if (err)2280 return -1;2281 }2282 if (!strip_extensions && istate->resolve_undo) {2283 struct strbuf sb = STRBUF_INIT;22842285 resolve_undo_write(&sb, istate->resolve_undo);2286 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,2287 sb.len) < 02288 || ce_write(&c, newfd, sb.buf, sb.len) < 0;2289 strbuf_release(&sb);2290 if (err)2291 return -1;2292 }2293 if (!strip_extensions && istate->untracked) {2294 struct strbuf sb = STRBUF_INIT;22952296 write_untracked_extension(&sb, istate->untracked);2297 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,2298 sb.len) < 0 ||2299 ce_write(&c, newfd, sb.buf, sb.len) < 0;2300 strbuf_release(&sb);2301 if (err)2302 return -1;2303 }23042305 if (ce_flush(&c, newfd, istate->sha1))2306 return -1;2307 if (close_tempfile(tempfile))2308 return error(_("could not close '%s'"), tempfile->filename.buf);2309 if (stat(tempfile->filename.buf, &st))2310 return -1;2311 istate->timestamp.sec = (unsigned int)st.st_mtime;2312 istate->timestamp.nsec = ST_MTIME_NSEC(st);2313 return 0;2314}23152316void set_alternate_index_output(const char *name)2317{2318 alternate_index_output = name;2319}23202321static int commit_locked_index(struct lock_file *lk)2322{2323 if (alternate_index_output)2324 return commit_lock_file_to(lk, alternate_index_output);2325 else2326 return commit_lock_file(lk);2327}23282329static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,2330 unsigned flags)2331{2332 int ret = do_write_index(istate, &lock->tempfile, 0);2333 if (ret)2334 return ret;2335 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=2336 (COMMIT_LOCK | CLOSE_LOCK));2337 if (flags & COMMIT_LOCK)2338 return commit_locked_index(lock);2339 else if (flags & CLOSE_LOCK)2340 return close_lock_file(lock);2341 else2342 return ret;2343}23442345static int write_split_index(struct index_state *istate,2346 struct lock_file *lock,2347 unsigned flags)2348{2349 int ret;2350 prepare_to_write_split_index(istate);2351 ret = do_write_locked_index(istate, lock, flags);2352 finish_writing_split_index(istate);2353 return ret;2354}23552356static const char *shared_index_expire = "2.weeks.ago";23572358static unsigned long get_shared_index_expire_date(void)2359{2360 static unsigned long shared_index_expire_date;2361 static int shared_index_expire_date_prepared;23622363 if (!shared_index_expire_date_prepared) {2364 git_config_get_expiry("splitindex.sharedindexexpire",2365 &shared_index_expire);2366 shared_index_expire_date = approxidate(shared_index_expire);2367 shared_index_expire_date_prepared = 1;2368 }23692370 return shared_index_expire_date;2371}23722373static int should_delete_shared_index(const char *shared_index_path)2374{2375 struct stat st;2376 unsigned long expiration;23772378 /* Check timestamp */2379 expiration = get_shared_index_expire_date();2380 if (!expiration)2381 return 0;2382 if (stat(shared_index_path, &st))2383 return error_errno(_("could not stat '%s'"), shared_index_path);2384 if (st.st_mtime > expiration)2385 return 0;23862387 return 1;2388}23892390static int clean_shared_index_files(const char *current_hex)2391{2392 struct dirent *de;2393 DIR *dir = opendir(get_git_dir());23942395 if (!dir)2396 return error_errno(_("unable to open git dir: %s"), get_git_dir());23972398 while ((de = readdir(dir)) != NULL) {2399 const char *sha1_hex;2400 const char *shared_index_path;2401 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))2402 continue;2403 if (!strcmp(sha1_hex, current_hex))2404 continue;2405 shared_index_path = git_path("%s", de->d_name);2406 if (should_delete_shared_index(shared_index_path) > 0 &&2407 unlink(shared_index_path))2408 warning_errno(_("unable to unlink: %s"), shared_index_path);2409 }2410 closedir(dir);24112412 return 0;2413}24142415static struct tempfile temporary_sharedindex;24162417static int write_shared_index(struct index_state *istate,2418 struct lock_file *lock, unsigned flags)2419{2420 struct split_index *si = istate->split_index;2421 int fd, ret;24222423 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));2424 if (fd < 0) {2425 hashclr(si->base_sha1);2426 return do_write_locked_index(istate, lock, flags);2427 }2428 move_cache_to_base_index(istate);2429 ret = do_write_index(si->base, &temporary_sharedindex, 1);2430 if (ret) {2431 delete_tempfile(&temporary_sharedindex);2432 return ret;2433 }2434 ret = adjust_shared_perm(get_tempfile_path(&temporary_sharedindex));2435 if (ret) {2436 int save_errno = errno;2437 error("cannot fix permission bits on %s", get_tempfile_path(&temporary_sharedindex));2438 delete_tempfile(&temporary_sharedindex);2439 errno = save_errno;2440 return ret;2441 }2442 ret = rename_tempfile(&temporary_sharedindex,2443 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));2444 if (!ret) {2445 hashcpy(si->base_sha1, si->base->sha1);2446 clean_shared_index_files(sha1_to_hex(si->base->sha1));2447 }24482449 return ret;2450}24512452static const int default_max_percent_split_change = 20;24532454static int too_many_not_shared_entries(struct index_state *istate)2455{2456 int i, not_shared = 0;2457 int max_split = git_config_get_max_percent_split_change();24582459 switch (max_split) {2460 case -1:2461 /* not or badly configured: use the default value */2462 max_split = default_max_percent_split_change;2463 break;2464 case 0:2465 return 1; /* 0% means always write a new shared index */2466 case 100:2467 return 0; /* 100% means never write a new shared index */2468 default:2469 break; /* just use the configured value */2470 }24712472 /* Count not shared entries */2473 for (i = 0; i < istate->cache_nr; i++) {2474 struct cache_entry *ce = istate->cache[i];2475 if (!ce->index)2476 not_shared++;2477 }24782479 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;2480}24812482int write_locked_index(struct index_state *istate, struct lock_file *lock,2483 unsigned flags)2484{2485 int new_shared_index, ret;2486 struct split_index *si = istate->split_index;24872488 if (!si || alternate_index_output ||2489 (istate->cache_changed & ~EXTMASK)) {2490 if (si)2491 hashclr(si->base_sha1);2492 return do_write_locked_index(istate, lock, flags);2493 }24942495 if (getenv("GIT_TEST_SPLIT_INDEX")) {2496 int v = si->base_sha1[0];2497 if ((v & 15) < 6)2498 istate->cache_changed |= SPLIT_INDEX_ORDERED;2499 }2500 if (too_many_not_shared_entries(istate))2501 istate->cache_changed |= SPLIT_INDEX_ORDERED;25022503 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;25042505 if (new_shared_index) {2506 ret = write_shared_index(istate, lock, flags);2507 if (ret)2508 return ret;2509 }25102511 ret = write_split_index(istate, lock, flags);25122513 /* Freshen the shared index only if the split-index was written */2514 if (!ret && !new_shared_index)2515 freshen_shared_index(sha1_to_hex(si->base_sha1), 1);25162517 return ret;2518}25192520/*2521 * Read the index file that is potentially unmerged into given2522 * index_state, dropping any unmerged entries. Returns true if2523 * the index is unmerged. Callers who want to refuse to work2524 * from an unmerged state can call this and check its return value,2525 * instead of calling read_cache().2526 */2527int read_index_unmerged(struct index_state *istate)2528{2529 int i;2530 int unmerged = 0;25312532 read_index(istate);2533 for (i = 0; i < istate->cache_nr; i++) {2534 struct cache_entry *ce = istate->cache[i];2535 struct cache_entry *new_ce;2536 int size, len;25372538 if (!ce_stage(ce))2539 continue;2540 unmerged = 1;2541 len = ce_namelen(ce);2542 size = cache_entry_size(len);2543 new_ce = xcalloc(1, size);2544 memcpy(new_ce->name, ce->name, len);2545 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;2546 new_ce->ce_namelen = len;2547 new_ce->ce_mode = ce->ce_mode;2548 if (add_index_entry(istate, new_ce, 0))2549 return error("%s: cannot drop to stage #0",2550 new_ce->name);2551 }2552 return unmerged;2553}25542555/*2556 * Returns 1 if the path is an "other" path with respect to2557 * the index; that is, the path is not mentioned in the index at all,2558 * either as a file, a directory with some files in the index,2559 * or as an unmerged entry.2560 *2561 * We helpfully remove a trailing "/" from directories so that2562 * the output of read_directory can be used as-is.2563 */2564int index_name_is_other(const struct index_state *istate, const char *name,2565 int namelen)2566{2567 int pos;2568 if (namelen && name[namelen - 1] == '/')2569 namelen--;2570 pos = index_name_pos(istate, name, namelen);2571 if (0 <= pos)2572 return 0; /* exact match */2573 pos = -pos - 1;2574 if (pos < istate->cache_nr) {2575 struct cache_entry *ce = istate->cache[pos];2576 if (ce_namelen(ce) == namelen &&2577 !memcmp(ce->name, name, namelen))2578 return 0; /* Yup, this one exists unmerged */2579 }2580 return 1;2581}25822583void *read_blob_data_from_index(const struct index_state *istate,2584 const char *path, unsigned long *size)2585{2586 int pos, len;2587 unsigned long sz;2588 enum object_type type;2589 void *data;25902591 len = strlen(path);2592 pos = index_name_pos(istate, path, len);2593 if (pos < 0) {2594 /*2595 * We might be in the middle of a merge, in which2596 * case we would read stage #2 (ours).2597 */2598 int i;2599 for (i = -pos - 1;2600 (pos < 0 && i < istate->cache_nr &&2601 !strcmp(istate->cache[i]->name, path));2602 i++)2603 if (ce_stage(istate->cache[i]) == 2)2604 pos = i;2605 }2606 if (pos < 0)2607 return NULL;2608 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);2609 if (!data || type != OBJ_BLOB) {2610 free(data);2611 return NULL;2612 }2613 if (size)2614 *size = sz;2615 return data;2616}26172618void stat_validity_clear(struct stat_validity *sv)2619{2620 free(sv->sd);2621 sv->sd = NULL;2622}26232624int stat_validity_check(struct stat_validity *sv, const char *path)2625{2626 struct stat st;26272628 if (stat(path, &st) < 0)2629 return sv->sd == NULL;2630 if (!sv->sd)2631 return 0;2632 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);2633}26342635void stat_validity_update(struct stat_validity *sv, int fd)2636{2637 struct stat st;26382639 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))2640 stat_validity_clear(sv);2641 else {2642 if (!sv->sd)2643 sv->sd = xcalloc(1, sizeof(struct stat_data));2644 fill_stat_data(sv->sd, &st);2645 }2646}