1/* 2 * GIT - The information manager from hell 3 * 4 * Copyright (C) Linus Torvalds, 2005 5 */ 6#define NO_THE_INDEX_COMPATIBILITY_MACROS 7#include"cache.h" 8#include"cache-tree.h" 9#include"refs.h" 10#include"dir.h" 11#include"tree.h" 12#include"commit.h" 13#include"blob.h" 14#include"resolve-undo.h" 15#include"strbuf.h" 16#include"varint.h" 17 18static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,int really); 19 20/* Mask for the name length in ce_flags in the on-disk index */ 21 22#define CE_NAMEMASK (0x0fff) 23 24/* Index extensions. 25 * 26 * The first letter should be 'A'..'Z' for extensions that are not 27 * necessary for a correct operation (i.e. optimization data). 28 * When new extensions are added that _needs_ to be understood in 29 * order to correctly interpret the index file, pick character that 30 * is outside the range, to cause the reader to abort. 31 */ 32 33#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) ) 34#define CACHE_EXT_TREE 0x54524545/* "TREE" */ 35#define CACHE_EXT_RESOLVE_UNDO 0x52455543/* "REUC" */ 36 37struct index_state the_index; 38 39static voidset_index_entry(struct index_state *istate,int nr,struct cache_entry *ce) 40{ 41 istate->cache[nr] = ce; 42add_name_hash(istate, ce); 43} 44 45static voidreplace_index_entry(struct index_state *istate,int nr,struct cache_entry *ce) 46{ 47struct cache_entry *old = istate->cache[nr]; 48 49remove_name_hash(istate, old); 50set_index_entry(istate, nr, ce); 51 istate->cache_changed =1; 52} 53 54voidrename_index_entry_at(struct index_state *istate,int nr,const char*new_name) 55{ 56struct cache_entry *old = istate->cache[nr], *new; 57int namelen =strlen(new_name); 58 59new=xmalloc(cache_entry_size(namelen)); 60copy_cache_entry(new, old); 61new->ce_flags &= ~CE_STATE_MASK; 62new->ce_namelen = namelen; 63memcpy(new->name, new_name, namelen +1); 64 65cache_tree_invalidate_path(istate->cache_tree, old->name); 66remove_index_entry_at(istate, nr); 67add_index_entry(istate,new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE); 68} 69 70/* 71 * This only updates the "non-critical" parts of the directory 72 * cache, ie the parts that aren't tracked by GIT, and only used 73 * to validate the cache. 74 */ 75voidfill_stat_cache_info(struct cache_entry *ce,struct stat *st) 76{ 77 ce->ce_ctime.sec = (unsigned int)st->st_ctime; 78 ce->ce_mtime.sec = (unsigned int)st->st_mtime; 79 ce->ce_ctime.nsec =ST_CTIME_NSEC(*st); 80 ce->ce_mtime.nsec =ST_MTIME_NSEC(*st); 81 ce->ce_dev = st->st_dev; 82 ce->ce_ino = st->st_ino; 83 ce->ce_uid = st->st_uid; 84 ce->ce_gid = st->st_gid; 85 ce->ce_size = st->st_size; 86 87if(assume_unchanged) 88 ce->ce_flags |= CE_VALID; 89 90if(S_ISREG(st->st_mode)) 91ce_mark_uptodate(ce); 92} 93 94static intce_compare_data(struct cache_entry *ce,struct stat *st) 95{ 96int match = -1; 97int fd =open(ce->name, O_RDONLY); 98 99if(fd >=0) { 100unsigned char sha1[20]; 101if(!index_fd(sha1, fd, st, OBJ_BLOB, ce->name,0)) 102 match =hashcmp(sha1, ce->sha1); 103/* index_fd() closed the file descriptor already */ 104} 105return match; 106} 107 108static intce_compare_link(struct cache_entry *ce,size_t expected_size) 109{ 110int match = -1; 111void*buffer; 112unsigned long size; 113enum object_type type; 114struct strbuf sb = STRBUF_INIT; 115 116if(strbuf_readlink(&sb, ce->name, expected_size)) 117return-1; 118 119 buffer =read_sha1_file(ce->sha1, &type, &size); 120if(buffer) { 121if(size == sb.len) 122 match =memcmp(buffer, sb.buf, size); 123free(buffer); 124} 125strbuf_release(&sb); 126return match; 127} 128 129static intce_compare_gitlink(struct cache_entry *ce) 130{ 131unsigned char sha1[20]; 132 133/* 134 * We don't actually require that the .git directory 135 * under GITLINK directory be a valid git directory. It 136 * might even be missing (in case nobody populated that 137 * sub-project). 138 * 139 * If so, we consider it always to match. 140 */ 141if(resolve_gitlink_ref(ce->name,"HEAD", sha1) <0) 142return0; 143returnhashcmp(sha1, ce->sha1); 144} 145 146static intce_modified_check_fs(struct cache_entry *ce,struct stat *st) 147{ 148switch(st->st_mode & S_IFMT) { 149case S_IFREG: 150if(ce_compare_data(ce, st)) 151return DATA_CHANGED; 152break; 153case S_IFLNK: 154if(ce_compare_link(ce,xsize_t(st->st_size))) 155return DATA_CHANGED; 156break; 157case S_IFDIR: 158if(S_ISGITLINK(ce->ce_mode)) 159returnce_compare_gitlink(ce) ? DATA_CHANGED :0; 160default: 161return TYPE_CHANGED; 162} 163return0; 164} 165 166static intce_match_stat_basic(struct cache_entry *ce,struct stat *st) 167{ 168unsigned int changed =0; 169 170if(ce->ce_flags & CE_REMOVE) 171return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; 172 173switch(ce->ce_mode & S_IFMT) { 174case S_IFREG: 175 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED :0; 176/* We consider only the owner x bit to be relevant for 177 * "mode changes" 178 */ 179if(trust_executable_bit && 180(0100& (ce->ce_mode ^ st->st_mode))) 181 changed |= MODE_CHANGED; 182break; 183case S_IFLNK: 184if(!S_ISLNK(st->st_mode) && 185(has_symlinks || !S_ISREG(st->st_mode))) 186 changed |= TYPE_CHANGED; 187break; 188case S_IFGITLINK: 189/* We ignore most of the st_xxx fields for gitlinks */ 190if(!S_ISDIR(st->st_mode)) 191 changed |= TYPE_CHANGED; 192else if(ce_compare_gitlink(ce)) 193 changed |= DATA_CHANGED; 194return changed; 195default: 196die("internal error: ce_mode is%o", ce->ce_mode); 197} 198if(ce->ce_mtime.sec != (unsigned int)st->st_mtime) 199 changed |= MTIME_CHANGED; 200if(trust_ctime && check_stat && 201 ce->ce_ctime.sec != (unsigned int)st->st_ctime) 202 changed |= CTIME_CHANGED; 203 204#ifdef USE_NSEC 205if(check_stat && ce->ce_mtime.nsec !=ST_MTIME_NSEC(*st)) 206 changed |= MTIME_CHANGED; 207if(trust_ctime && check_stat && 208 ce->ce_ctime.nsec !=ST_CTIME_NSEC(*st)) 209 changed |= CTIME_CHANGED; 210#endif 211 212if(check_stat) { 213if(ce->ce_uid != (unsigned int) st->st_uid || 214 ce->ce_gid != (unsigned int) st->st_gid) 215 changed |= OWNER_CHANGED; 216if(ce->ce_ino != (unsigned int) st->st_ino) 217 changed |= INODE_CHANGED; 218} 219 220#ifdef USE_STDEV 221/* 222 * st_dev breaks on network filesystems where different 223 * clients will have different views of what "device" 224 * the filesystem is on 225 */ 226if(check_stat && ce->ce_dev != (unsigned int) st->st_dev) 227 changed |= INODE_CHANGED; 228#endif 229 230if(ce->ce_size != (unsigned int) st->st_size) 231 changed |= DATA_CHANGED; 232 233/* Racily smudged entry? */ 234if(!ce->ce_size) { 235if(!is_empty_blob_sha1(ce->sha1)) 236 changed |= DATA_CHANGED; 237} 238 239return changed; 240} 241 242static intis_racy_timestamp(const struct index_state *istate,struct cache_entry *ce) 243{ 244return(!S_ISGITLINK(ce->ce_mode) && 245 istate->timestamp.sec && 246#ifdef USE_NSEC 247/* nanosecond timestamped files can also be racy! */ 248(istate->timestamp.sec < ce->ce_mtime.sec || 249(istate->timestamp.sec == ce->ce_mtime.sec && 250 istate->timestamp.nsec <= ce->ce_mtime.nsec)) 251#else 252 istate->timestamp.sec <= ce->ce_mtime.sec 253#endif 254); 255} 256 257intie_match_stat(const struct index_state *istate, 258struct cache_entry *ce,struct stat *st, 259unsigned int options) 260{ 261unsigned int changed; 262int ignore_valid = options & CE_MATCH_IGNORE_VALID; 263int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE; 264int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 265 266/* 267 * If it's marked as always valid in the index, it's 268 * valid whatever the checked-out copy says. 269 * 270 * skip-worktree has the same effect with higher precedence 271 */ 272if(!ignore_skip_worktree &&ce_skip_worktree(ce)) 273return0; 274if(!ignore_valid && (ce->ce_flags & CE_VALID)) 275return0; 276 277/* 278 * Intent-to-add entries have not been added, so the index entry 279 * by definition never matches what is in the work tree until it 280 * actually gets added. 281 */ 282if(ce->ce_flags & CE_INTENT_TO_ADD) 283return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED; 284 285 changed =ce_match_stat_basic(ce, st); 286 287/* 288 * Within 1 second of this sequence: 289 * echo xyzzy >file && git-update-index --add file 290 * running this command: 291 * echo frotz >file 292 * would give a falsely clean cache entry. The mtime and 293 * length match the cache, and other stat fields do not change. 294 * 295 * We could detect this at update-index time (the cache entry 296 * being registered/updated records the same time as "now") 297 * and delay the return from git-update-index, but that would 298 * effectively mean we can make at most one commit per second, 299 * which is not acceptable. Instead, we check cache entries 300 * whose mtime are the same as the index file timestamp more 301 * carefully than others. 302 */ 303if(!changed &&is_racy_timestamp(istate, ce)) { 304if(assume_racy_is_modified) 305 changed |= DATA_CHANGED; 306else 307 changed |=ce_modified_check_fs(ce, st); 308} 309 310return changed; 311} 312 313intie_modified(const struct index_state *istate, 314struct cache_entry *ce,struct stat *st,unsigned int options) 315{ 316int changed, changed_fs; 317 318 changed =ie_match_stat(istate, ce, st, options); 319if(!changed) 320return0; 321/* 322 * If the mode or type has changed, there's no point in trying 323 * to refresh the entry - it's not going to match 324 */ 325if(changed & (MODE_CHANGED | TYPE_CHANGED)) 326return changed; 327 328/* 329 * Immediately after read-tree or update-index --cacheinfo, 330 * the length field is zero, as we have never even read the 331 * lstat(2) information once, and we cannot trust DATA_CHANGED 332 * returned by ie_match_stat() which in turn was returned by 333 * ce_match_stat_basic() to signal that the filesize of the 334 * blob changed. We have to actually go to the filesystem to 335 * see if the contents match, and if so, should answer "unchanged". 336 * 337 * The logic does not apply to gitlinks, as ce_match_stat_basic() 338 * already has checked the actual HEAD from the filesystem in the 339 * subproject. If ie_match_stat() already said it is different, 340 * then we know it is. 341 */ 342if((changed & DATA_CHANGED) && 343(S_ISGITLINK(ce->ce_mode) || ce->ce_size !=0)) 344return changed; 345 346 changed_fs =ce_modified_check_fs(ce, st); 347if(changed_fs) 348return changed | changed_fs; 349return0; 350} 351 352intbase_name_compare(const char*name1,int len1,int mode1, 353const char*name2,int len2,int mode2) 354{ 355unsigned char c1, c2; 356int len = len1 < len2 ? len1 : len2; 357int cmp; 358 359 cmp =memcmp(name1, name2, len); 360if(cmp) 361return cmp; 362 c1 = name1[len]; 363 c2 = name2[len]; 364if(!c1 &&S_ISDIR(mode1)) 365 c1 ='/'; 366if(!c2 &&S_ISDIR(mode2)) 367 c2 ='/'; 368return(c1 < c2) ? -1: (c1 > c2) ?1:0; 369} 370 371/* 372 * df_name_compare() is identical to base_name_compare(), except it 373 * compares conflicting directory/file entries as equal. Note that 374 * while a directory name compares as equal to a regular file, they 375 * then individually compare _differently_ to a filename that has 376 * a dot after the basename (because '\0' < '.' < '/'). 377 * 378 * This is used by routines that want to traverse the git namespace 379 * but then handle conflicting entries together when possible. 380 */ 381intdf_name_compare(const char*name1,int len1,int mode1, 382const char*name2,int len2,int mode2) 383{ 384int len = len1 < len2 ? len1 : len2, cmp; 385unsigned char c1, c2; 386 387 cmp =memcmp(name1, name2, len); 388if(cmp) 389return cmp; 390/* Directories and files compare equal (same length, same name) */ 391if(len1 == len2) 392return0; 393 c1 = name1[len]; 394if(!c1 &&S_ISDIR(mode1)) 395 c1 ='/'; 396 c2 = name2[len]; 397if(!c2 &&S_ISDIR(mode2)) 398 c2 ='/'; 399if(c1 =='/'&& !c2) 400return0; 401if(c2 =='/'&& !c1) 402return0; 403return c1 - c2; 404} 405 406intcache_name_stage_compare(const char*name1,int len1,int stage1,const char*name2,int len2,int stage2) 407{ 408int len = len1 < len2 ? len1 : len2; 409int cmp; 410 411 cmp =memcmp(name1, name2, len); 412if(cmp) 413return cmp; 414if(len1 < len2) 415return-1; 416if(len1 > len2) 417return1; 418 419if(stage1 < stage2) 420return-1; 421if(stage1 > stage2) 422return1; 423return0; 424} 425 426intcache_name_compare(const char*name1,int len1,const char*name2,int len2) 427{ 428returncache_name_stage_compare(name1, len1,0, name2, len2,0); 429} 430 431static intindex_name_stage_pos(const struct index_state *istate,const char*name,int namelen,int stage) 432{ 433int first, last; 434 435 first =0; 436 last = istate->cache_nr; 437while(last > first) { 438int next = (last + first) >>1; 439struct cache_entry *ce = istate->cache[next]; 440int cmp =cache_name_stage_compare(name, namelen, stage, ce->name,ce_namelen(ce),ce_stage(ce)); 441if(!cmp) 442return next; 443if(cmp <0) { 444 last = next; 445continue; 446} 447 first = next+1; 448} 449return-first-1; 450} 451 452intindex_name_pos(const struct index_state *istate,const char*name,int namelen) 453{ 454returnindex_name_stage_pos(istate, name, namelen,0); 455} 456 457/* Remove entry, return true if there are more entries to go.. */ 458intremove_index_entry_at(struct index_state *istate,int pos) 459{ 460struct cache_entry *ce = istate->cache[pos]; 461 462record_resolve_undo(istate, ce); 463remove_name_hash(istate, ce); 464 istate->cache_changed =1; 465 istate->cache_nr--; 466if(pos >= istate->cache_nr) 467return0; 468memmove(istate->cache + pos, 469 istate->cache + pos +1, 470(istate->cache_nr - pos) *sizeof(struct cache_entry *)); 471return1; 472} 473 474/* 475 * Remove all cache ententries marked for removal, that is where 476 * CE_REMOVE is set in ce_flags. This is much more effective than 477 * calling remove_index_entry_at() for each entry to be removed. 478 */ 479voidremove_marked_cache_entries(struct index_state *istate) 480{ 481struct cache_entry **ce_array = istate->cache; 482unsigned int i, j; 483 484for(i = j =0; i < istate->cache_nr; i++) { 485if(ce_array[i]->ce_flags & CE_REMOVE) 486remove_name_hash(istate, ce_array[i]); 487else 488 ce_array[j++] = ce_array[i]; 489} 490 istate->cache_changed =1; 491 istate->cache_nr = j; 492} 493 494intremove_file_from_index(struct index_state *istate,const char*path) 495{ 496int pos =index_name_pos(istate, path,strlen(path)); 497if(pos <0) 498 pos = -pos-1; 499cache_tree_invalidate_path(istate->cache_tree, path); 500while(pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 501remove_index_entry_at(istate, pos); 502return0; 503} 504 505static intcompare_name(struct cache_entry *ce,const char*path,int namelen) 506{ 507return namelen !=ce_namelen(ce) ||memcmp(path, ce->name, namelen); 508} 509 510static intindex_name_pos_also_unmerged(struct index_state *istate, 511const char*path,int namelen) 512{ 513int pos =index_name_pos(istate, path, namelen); 514struct cache_entry *ce; 515 516if(pos >=0) 517return pos; 518 519/* maybe unmerged? */ 520 pos = -1- pos; 521if(pos >= istate->cache_nr || 522compare_name((ce = istate->cache[pos]), path, namelen)) 523return-1; 524 525/* order of preference: stage 2, 1, 3 */ 526if(ce_stage(ce) ==1&& pos +1< istate->cache_nr && 527ce_stage((ce = istate->cache[pos +1])) ==2&& 528!compare_name(ce, path, namelen)) 529 pos++; 530return pos; 531} 532 533static intdifferent_name(struct cache_entry *ce,struct cache_entry *alias) 534{ 535int len =ce_namelen(ce); 536returnce_namelen(alias) != len ||memcmp(ce->name, alias->name, len); 537} 538 539/* 540 * If we add a filename that aliases in the cache, we will use the 541 * name that we already have - but we don't want to update the same 542 * alias twice, because that implies that there were actually two 543 * different files with aliasing names! 544 * 545 * So we use the CE_ADDED flag to verify that the alias was an old 546 * one before we accept it as 547 */ 548static struct cache_entry *create_alias_ce(struct cache_entry *ce,struct cache_entry *alias) 549{ 550int len; 551struct cache_entry *new; 552 553if(alias->ce_flags & CE_ADDED) 554die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); 555 556/* Ok, create the new entry using the name of the existing alias */ 557 len =ce_namelen(alias); 558new=xcalloc(1,cache_entry_size(len)); 559memcpy(new->name, alias->name, len); 560copy_cache_entry(new, ce); 561free(ce); 562return new; 563} 564 565static voidrecord_intent_to_add(struct cache_entry *ce) 566{ 567unsigned char sha1[20]; 568if(write_sha1_file("",0, blob_type, sha1)) 569die("cannot create an empty blob in the object database"); 570hashcpy(ce->sha1, sha1); 571} 572 573intadd_to_index(struct index_state *istate,const char*path,struct stat *st,int flags) 574{ 575int size, namelen, was_same; 576 mode_t st_mode = st->st_mode; 577struct cache_entry *ce, *alias; 578unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY; 579int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); 580int pretend = flags & ADD_CACHE_PRETEND; 581int intent_only = flags & ADD_CACHE_INTENT; 582int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| 583(intent_only ? ADD_CACHE_NEW_ONLY :0)); 584 585if(!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 586returnerror("%s: can only add regular files, symbolic links or git-directories", path); 587 588 namelen =strlen(path); 589if(S_ISDIR(st_mode)) { 590while(namelen && path[namelen-1] =='/') 591 namelen--; 592} 593 size =cache_entry_size(namelen); 594 ce =xcalloc(1, size); 595memcpy(ce->name, path, namelen); 596 ce->ce_namelen = namelen; 597if(!intent_only) 598fill_stat_cache_info(ce, st); 599else 600 ce->ce_flags |= CE_INTENT_TO_ADD; 601 602if(trust_executable_bit && has_symlinks) 603 ce->ce_mode =create_ce_mode(st_mode); 604else{ 605/* If there is an existing entry, pick the mode bits and type 606 * from it, otherwise assume unexecutable regular file. 607 */ 608struct cache_entry *ent; 609int pos =index_name_pos_also_unmerged(istate, path, namelen); 610 611 ent = (0<= pos) ? istate->cache[pos] : NULL; 612 ce->ce_mode =ce_mode_from_stat(ent, st_mode); 613} 614 615/* When core.ignorecase=true, determine if a directory of the same name but differing 616 * case already exists within the Git repository. If it does, ensure the directory 617 * case of the file being added to the repository matches (is folded into) the existing 618 * entry's directory case. 619 */ 620if(ignore_case) { 621const char*startPtr = ce->name; 622const char*ptr = startPtr; 623while(*ptr) { 624while(*ptr && *ptr !='/') 625++ptr; 626if(*ptr =='/') { 627struct cache_entry *foundce; 628++ptr; 629 foundce =index_name_exists(istate, ce->name, ptr - ce->name, ignore_case); 630if(foundce) { 631memcpy((void*)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr); 632 startPtr = ptr; 633} 634} 635} 636} 637 638 alias =index_name_exists(istate, ce->name,ce_namelen(ce), ignore_case); 639if(alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) { 640/* Nothing changed, really */ 641free(ce); 642if(!S_ISGITLINK(alias->ce_mode)) 643ce_mark_uptodate(alias); 644 alias->ce_flags |= CE_ADDED; 645return0; 646} 647if(!intent_only) { 648if(index_path(ce->sha1, path, st, HASH_WRITE_OBJECT)) 649returnerror("unable to index file%s", path); 650}else 651record_intent_to_add(ce); 652 653if(ignore_case && alias &&different_name(ce, alias)) 654 ce =create_alias_ce(ce, alias); 655 ce->ce_flags |= CE_ADDED; 656 657/* It was suspected to be racily clean, but it turns out to be Ok */ 658 was_same = (alias && 659!ce_stage(alias) && 660!hashcmp(alias->sha1, ce->sha1) && 661 ce->ce_mode == alias->ce_mode); 662 663if(pretend) 664; 665else if(add_index_entry(istate, ce, add_option)) 666returnerror("unable to add%sto index",path); 667if(verbose && !was_same) 668printf("add '%s'\n", path); 669return0; 670} 671 672intadd_file_to_index(struct index_state *istate,const char*path,int flags) 673{ 674struct stat st; 675if(lstat(path, &st)) 676die_errno("unable to stat '%s'", path); 677returnadd_to_index(istate, path, &st, flags); 678} 679 680struct cache_entry *make_cache_entry(unsigned int mode, 681const unsigned char*sha1,const char*path,int stage, 682int refresh) 683{ 684int size, len; 685struct cache_entry *ce; 686 687if(!verify_path(path)) { 688error("Invalid path '%s'", path); 689return NULL; 690} 691 692 len =strlen(path); 693 size =cache_entry_size(len); 694 ce =xcalloc(1, size); 695 696hashcpy(ce->sha1, sha1); 697memcpy(ce->name, path, len); 698 ce->ce_flags =create_ce_flags(stage); 699 ce->ce_namelen = len; 700 ce->ce_mode =create_ce_mode(mode); 701 702if(refresh) 703returnrefresh_cache_entry(ce,0); 704 705return ce; 706} 707 708intce_same_name(struct cache_entry *a,struct cache_entry *b) 709{ 710int len =ce_namelen(a); 711returnce_namelen(b) == len && !memcmp(a->name, b->name, len); 712} 713 714intce_path_match(const struct cache_entry *ce,const struct pathspec *pathspec) 715{ 716returnmatch_pathspec_depth(pathspec, ce->name,ce_namelen(ce),0, NULL); 717} 718 719/* 720 * We fundamentally don't like some paths: we don't want 721 * dot or dot-dot anywhere, and for obvious reasons don't 722 * want to recurse into ".git" either. 723 * 724 * Also, we don't want double slashes or slashes at the 725 * end that can make pathnames ambiguous. 726 */ 727static intverify_dotfile(const char*rest) 728{ 729/* 730 * The first character was '.', but that 731 * has already been discarded, we now test 732 * the rest. 733 */ 734 735/* "." is not allowed */ 736if(*rest =='\0'||is_dir_sep(*rest)) 737return0; 738 739switch(*rest) { 740/* 741 * ".git" followed by NUL or slash is bad. This 742 * shares the path end test with the ".." case. 743 */ 744case'g': 745if(rest[1] !='i') 746break; 747if(rest[2] !='t') 748break; 749 rest +=2; 750/* fallthrough */ 751case'.': 752if(rest[1] =='\0'||is_dir_sep(rest[1])) 753return0; 754} 755return1; 756} 757 758intverify_path(const char*path) 759{ 760char c; 761 762if(has_dos_drive_prefix(path)) 763return0; 764 765goto inside; 766for(;;) { 767if(!c) 768return1; 769if(is_dir_sep(c)) { 770inside: 771 c = *path++; 772if((c =='.'&& !verify_dotfile(path)) || 773is_dir_sep(c) || c =='\0') 774return0; 775} 776 c = *path++; 777} 778} 779 780/* 781 * Do we have another file that has the beginning components being a 782 * proper superset of the name we're trying to add? 783 */ 784static inthas_file_name(struct index_state *istate, 785const struct cache_entry *ce,int pos,int ok_to_replace) 786{ 787int retval =0; 788int len =ce_namelen(ce); 789int stage =ce_stage(ce); 790const char*name = ce->name; 791 792while(pos < istate->cache_nr) { 793struct cache_entry *p = istate->cache[pos++]; 794 795if(len >=ce_namelen(p)) 796break; 797if(memcmp(name, p->name, len)) 798break; 799if(ce_stage(p) != stage) 800continue; 801if(p->name[len] !='/') 802continue; 803if(p->ce_flags & CE_REMOVE) 804continue; 805 retval = -1; 806if(!ok_to_replace) 807break; 808remove_index_entry_at(istate, --pos); 809} 810return retval; 811} 812 813/* 814 * Do we have another file with a pathname that is a proper 815 * subset of the name we're trying to add? 816 */ 817static inthas_dir_name(struct index_state *istate, 818const struct cache_entry *ce,int pos,int ok_to_replace) 819{ 820int retval =0; 821int stage =ce_stage(ce); 822const char*name = ce->name; 823const char*slash = name +ce_namelen(ce); 824 825for(;;) { 826int len; 827 828for(;;) { 829if(*--slash =='/') 830break; 831if(slash <= ce->name) 832return retval; 833} 834 len = slash - name; 835 836 pos =index_name_stage_pos(istate, name, len, stage); 837if(pos >=0) { 838/* 839 * Found one, but not so fast. This could 840 * be a marker that says "I was here, but 841 * I am being removed". Such an entry is 842 * not a part of the resulting tree, and 843 * it is Ok to have a directory at the same 844 * path. 845 */ 846if(!(istate->cache[pos]->ce_flags & CE_REMOVE)) { 847 retval = -1; 848if(!ok_to_replace) 849break; 850remove_index_entry_at(istate, pos); 851continue; 852} 853} 854else 855 pos = -pos-1; 856 857/* 858 * Trivial optimization: if we find an entry that 859 * already matches the sub-directory, then we know 860 * we're ok, and we can exit. 861 */ 862while(pos < istate->cache_nr) { 863struct cache_entry *p = istate->cache[pos]; 864if((ce_namelen(p) <= len) || 865(p->name[len] !='/') || 866memcmp(p->name, name, len)) 867break;/* not our subdirectory */ 868if(ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE)) 869/* 870 * p is at the same stage as our entry, and 871 * is a subdirectory of what we are looking 872 * at, so we cannot have conflicts at our 873 * level or anything shorter. 874 */ 875return retval; 876 pos++; 877} 878} 879return retval; 880} 881 882/* We may be in a situation where we already have path/file and path 883 * is being added, or we already have path and path/file is being 884 * added. Either one would result in a nonsense tree that has path 885 * twice when git-write-tree tries to write it out. Prevent it. 886 * 887 * If ok-to-replace is specified, we remove the conflicting entries 888 * from the cache so the caller should recompute the insert position. 889 * When this happens, we return non-zero. 890 */ 891static intcheck_file_directory_conflict(struct index_state *istate, 892const struct cache_entry *ce, 893int pos,int ok_to_replace) 894{ 895int retval; 896 897/* 898 * When ce is an "I am going away" entry, we allow it to be added 899 */ 900if(ce->ce_flags & CE_REMOVE) 901return0; 902 903/* 904 * We check if the path is a sub-path of a subsequent pathname 905 * first, since removing those will not change the position 906 * in the array. 907 */ 908 retval =has_file_name(istate, ce, pos, ok_to_replace); 909 910/* 911 * Then check if the path might have a clashing sub-directory 912 * before it. 913 */ 914return retval +has_dir_name(istate, ce, pos, ok_to_replace); 915} 916 917static intadd_index_entry_with_check(struct index_state *istate,struct cache_entry *ce,int option) 918{ 919int pos; 920int ok_to_add = option & ADD_CACHE_OK_TO_ADD; 921int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE; 922int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK; 923int new_only = option & ADD_CACHE_NEW_ONLY; 924 925cache_tree_invalidate_path(istate->cache_tree, ce->name); 926 pos =index_name_stage_pos(istate, ce->name,ce_namelen(ce),ce_stage(ce)); 927 928/* existing match? Just replace it. */ 929if(pos >=0) { 930if(!new_only) 931replace_index_entry(istate, pos, ce); 932return0; 933} 934 pos = -pos-1; 935 936/* 937 * Inserting a merged entry ("stage 0") into the index 938 * will always replace all non-merged entries.. 939 */ 940if(pos < istate->cache_nr &&ce_stage(ce) ==0) { 941while(ce_same_name(istate->cache[pos], ce)) { 942 ok_to_add =1; 943if(!remove_index_entry_at(istate, pos)) 944break; 945} 946} 947 948if(!ok_to_add) 949return-1; 950if(!verify_path(ce->name)) 951returnerror("Invalid path '%s'", ce->name); 952 953if(!skip_df_check && 954check_file_directory_conflict(istate, ce, pos, ok_to_replace)) { 955if(!ok_to_replace) 956returnerror("'%s' appears as both a file and as a directory", 957 ce->name); 958 pos =index_name_stage_pos(istate, ce->name,ce_namelen(ce),ce_stage(ce)); 959 pos = -pos-1; 960} 961return pos +1; 962} 963 964intadd_index_entry(struct index_state *istate,struct cache_entry *ce,int option) 965{ 966int pos; 967 968if(option & ADD_CACHE_JUST_APPEND) 969 pos = istate->cache_nr; 970else{ 971int ret; 972 ret =add_index_entry_with_check(istate, ce, option); 973if(ret <=0) 974return ret; 975 pos = ret -1; 976} 977 978/* Make sure the array is big enough .. */ 979if(istate->cache_nr == istate->cache_alloc) { 980 istate->cache_alloc =alloc_nr(istate->cache_alloc); 981 istate->cache =xrealloc(istate->cache, 982 istate->cache_alloc *sizeof(*istate->cache)); 983} 984 985/* Add it in.. */ 986 istate->cache_nr++; 987if(istate->cache_nr > pos +1) 988memmove(istate->cache + pos +1, 989 istate->cache + pos, 990(istate->cache_nr - pos -1) *sizeof(ce)); 991set_index_entry(istate, pos, ce); 992 istate->cache_changed =1; 993return0; 994} 995 996/* 997 * "refresh" does not calculate a new sha1 file or bring the 998 * cache up-to-date for mode/content changes. But what it 999 * _does_ do is to "re-match" the stat information of a file1000 * with the cache, so that you can refresh the cache for a1001 * file that hasn't been changed but where the stat entry is1002 * out of date.1003 *1004 * For example, you'd want to do this after doing a "git-read-tree",1005 * to link up the stat cache details with the proper files.1006 */1007static struct cache_entry *refresh_cache_ent(struct index_state *istate,1008struct cache_entry *ce,1009unsigned int options,int*err,1010int*changed_ret)1011{1012struct stat st;1013struct cache_entry *updated;1014int changed, size;1015int ignore_valid = options & CE_MATCH_IGNORE_VALID;1016int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;10171018if(ce_uptodate(ce))1019return ce;10201021/*1022 * CE_VALID or CE_SKIP_WORKTREE means the user promised us1023 * that the change to the work tree does not matter and told1024 * us not to worry.1025 */1026if(!ignore_skip_worktree &&ce_skip_worktree(ce)) {1027ce_mark_uptodate(ce);1028return ce;1029}1030if(!ignore_valid && (ce->ce_flags & CE_VALID)) {1031ce_mark_uptodate(ce);1032return ce;1033}10341035if(lstat(ce->name, &st) <0) {1036if(err)1037*err = errno;1038return NULL;1039}10401041 changed =ie_match_stat(istate, ce, &st, options);1042if(changed_ret)1043*changed_ret = changed;1044if(!changed) {1045/*1046 * The path is unchanged. If we were told to ignore1047 * valid bit, then we did the actual stat check and1048 * found that the entry is unmodified. If the entry1049 * is not marked VALID, this is the place to mark it1050 * valid again, under "assume unchanged" mode.1051 */1052if(ignore_valid && assume_unchanged &&1053!(ce->ce_flags & CE_VALID))1054;/* mark this one VALID again */1055else{1056/*1057 * We do not mark the index itself "modified"1058 * because CE_UPTODATE flag is in-core only;1059 * we are not going to write this change out.1060 */1061if(!S_ISGITLINK(ce->ce_mode))1062ce_mark_uptodate(ce);1063return ce;1064}1065}10661067if(ie_modified(istate, ce, &st, options)) {1068if(err)1069*err = EINVAL;1070return NULL;1071}10721073 size =ce_size(ce);1074 updated =xmalloc(size);1075memcpy(updated, ce, size);1076fill_stat_cache_info(updated, &st);1077/*1078 * If ignore_valid is not set, we should leave CE_VALID bit1079 * alone. Otherwise, paths marked with --no-assume-unchanged1080 * (i.e. things to be edited) will reacquire CE_VALID bit1081 * automatically, which is not really what we want.1082 */1083if(!ignore_valid && assume_unchanged &&1084!(ce->ce_flags & CE_VALID))1085 updated->ce_flags &= ~CE_VALID;10861087return updated;1088}10891090static voidshow_file(const char* fmt,const char* name,int in_porcelain,1091int* first,const char*header_msg)1092{1093if(in_porcelain && *first && header_msg) {1094printf("%s\n", header_msg);1095*first =0;1096}1097printf(fmt, name);1098}10991100intrefresh_index(struct index_state *istate,unsigned int flags,const char**pathspec,1101char*seen,const char*header_msg)1102{1103int i;1104int has_errors =0;1105int really = (flags & REFRESH_REALLY) !=0;1106int allow_unmerged = (flags & REFRESH_UNMERGED) !=0;1107int quiet = (flags & REFRESH_QUIET) !=0;1108int not_new = (flags & REFRESH_IGNORE_MISSING) !=0;1109int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) !=0;1110int first =1;1111int in_porcelain = (flags & REFRESH_IN_PORCELAIN);1112unsigned int options = really ? CE_MATCH_IGNORE_VALID :0;1113const char*modified_fmt;1114const char*deleted_fmt;1115const char*typechange_fmt;1116const char*added_fmt;1117const char*unmerged_fmt;11181119 modified_fmt = (in_porcelain ?"M\t%s\n":"%s: needs update\n");1120 deleted_fmt = (in_porcelain ?"D\t%s\n":"%s: needs update\n");1121 typechange_fmt = (in_porcelain ?"T\t%s\n":"%sneeds update\n");1122 added_fmt = (in_porcelain ?"A\t%s\n":"%sneeds update\n");1123 unmerged_fmt = (in_porcelain ?"U\t%s\n":"%s: needs merge\n");1124for(i =0; i < istate->cache_nr; i++) {1125struct cache_entry *ce, *new;1126int cache_errno =0;1127int changed =0;1128int filtered =0;11291130 ce = istate->cache[i];1131if(ignore_submodules &&S_ISGITLINK(ce->ce_mode))1132continue;11331134if(pathspec &&1135!match_pathspec(pathspec, ce->name,ce_namelen(ce),0, seen))1136 filtered =1;11371138if(ce_stage(ce)) {1139while((i < istate->cache_nr) &&1140!strcmp(istate->cache[i]->name, ce->name))1141 i++;1142 i--;1143if(allow_unmerged)1144continue;1145if(!filtered)1146show_file(unmerged_fmt, ce->name, in_porcelain,1147&first, header_msg);1148 has_errors =1;1149continue;1150}11511152if(filtered)1153continue;11541155new=refresh_cache_ent(istate, ce, options, &cache_errno, &changed);1156if(new== ce)1157continue;1158if(!new) {1159const char*fmt;11601161if(not_new && cache_errno == ENOENT)1162continue;1163if(really && cache_errno == EINVAL) {1164/* If we are doing --really-refresh that1165 * means the index is not valid anymore.1166 */1167 ce->ce_flags &= ~CE_VALID;1168 istate->cache_changed =1;1169}1170if(quiet)1171continue;11721173if(cache_errno == ENOENT)1174 fmt = deleted_fmt;1175else if(ce->ce_flags & CE_INTENT_TO_ADD)1176 fmt = added_fmt;/* must be before other checks */1177else if(changed & TYPE_CHANGED)1178 fmt = typechange_fmt;1179else1180 fmt = modified_fmt;1181show_file(fmt,1182 ce->name, in_porcelain, &first, header_msg);1183 has_errors =1;1184continue;1185}11861187replace_index_entry(istate, i,new);1188}1189return has_errors;1190}11911192static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,int really)1193{1194returnrefresh_cache_ent(&the_index, ce, really, NULL, NULL);1195}119611971198/*****************************************************************1199 * Index File I/O1200 *****************************************************************/12011202#define INDEX_FORMAT_DEFAULT 312031204/*1205 * dev/ino/uid/gid/size are also just tracked to the low 32 bits1206 * Again - this is just a (very strong in practice) heuristic that1207 * the inode hasn't changed.1208 *1209 * We save the fields in big-endian order to allow using the1210 * index file over NFS transparently.1211 */1212struct ondisk_cache_entry {1213struct cache_time ctime;1214struct cache_time mtime;1215unsigned int dev;1216unsigned int ino;1217unsigned int mode;1218unsigned int uid;1219unsigned int gid;1220unsigned int size;1221unsigned char sha1[20];1222unsigned short flags;1223char name[FLEX_ARRAY];/* more */1224};12251226/*1227 * This struct is used when CE_EXTENDED bit is 11228 * The struct must match ondisk_cache_entry exactly from1229 * ctime till flags1230 */1231struct ondisk_cache_entry_extended {1232struct cache_time ctime;1233struct cache_time mtime;1234unsigned int dev;1235unsigned int ino;1236unsigned int mode;1237unsigned int uid;1238unsigned int gid;1239unsigned int size;1240unsigned char sha1[20];1241unsigned short flags;1242unsigned short flags2;1243char name[FLEX_ARRAY];/* more */1244};12451246/* These are only used for v3 or lower */1247#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)1248#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)1249#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)1250#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \1251 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \1252 ondisk_cache_entry_size(ce_namelen(ce)))12531254static intverify_hdr(struct cache_header *hdr,unsigned long size)1255{1256 git_SHA_CTX c;1257unsigned char sha1[20];1258int hdr_version;12591260if(hdr->hdr_signature !=htonl(CACHE_SIGNATURE))1261returnerror("bad signature");1262 hdr_version =ntohl(hdr->hdr_version);1263if(hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)1264returnerror("bad index version%d", hdr_version);1265git_SHA1_Init(&c);1266git_SHA1_Update(&c, hdr, size -20);1267git_SHA1_Final(sha1, &c);1268if(hashcmp(sha1, (unsigned char*)hdr + size -20))1269returnerror("bad index file sha1 signature");1270return0;1271}12721273static intread_index_extension(struct index_state *istate,1274const char*ext,void*data,unsigned long sz)1275{1276switch(CACHE_EXT(ext)) {1277case CACHE_EXT_TREE:1278 istate->cache_tree =cache_tree_read(data, sz);1279break;1280case CACHE_EXT_RESOLVE_UNDO:1281 istate->resolve_undo =resolve_undo_read(data, sz);1282break;1283default:1284if(*ext <'A'||'Z'< *ext)1285returnerror("index uses %.4s extension, which we do not understand",1286 ext);1287fprintf(stderr,"ignoring %.4s extension\n", ext);1288break;1289}1290return0;1291}12921293intread_index(struct index_state *istate)1294{1295returnread_index_from(istate,get_index_file());1296}12971298#ifndef NEEDS_ALIGNED_ACCESS1299#define ntoh_s(var) ntohs(var)1300#define ntoh_l(var) ntohl(var)1301#else1302staticinlineuint16_tntoh_s_force_align(void*p)1303{1304uint16_t x;1305memcpy(&x, p,sizeof(x));1306returnntohs(x);1307}1308staticinlineuint32_tntoh_l_force_align(void*p)1309{1310uint32_t x;1311memcpy(&x, p,sizeof(x));1312returnntohl(x);1313}1314#define ntoh_s(var) ntoh_s_force_align(&(var))1315#define ntoh_l(var) ntoh_l_force_align(&(var))1316#endif13171318static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,1319unsigned int flags,1320const char*name,1321size_t len)1322{1323struct cache_entry *ce =xmalloc(cache_entry_size(len));13241325 ce->ce_ctime.sec =ntoh_l(ondisk->ctime.sec);1326 ce->ce_mtime.sec =ntoh_l(ondisk->mtime.sec);1327 ce->ce_ctime.nsec =ntoh_l(ondisk->ctime.nsec);1328 ce->ce_mtime.nsec =ntoh_l(ondisk->mtime.nsec);1329 ce->ce_dev =ntoh_l(ondisk->dev);1330 ce->ce_ino =ntoh_l(ondisk->ino);1331 ce->ce_mode =ntoh_l(ondisk->mode);1332 ce->ce_uid =ntoh_l(ondisk->uid);1333 ce->ce_gid =ntoh_l(ondisk->gid);1334 ce->ce_size =ntoh_l(ondisk->size);1335 ce->ce_flags = flags & ~CE_NAMEMASK;1336 ce->ce_namelen = len;1337hashcpy(ce->sha1, ondisk->sha1);1338memcpy(ce->name, name, len);1339 ce->name[len] ='\0';1340return ce;1341}13421343/*1344 * Adjacent cache entries tend to share the leading paths, so it makes1345 * sense to only store the differences in later entries. In the v41346 * on-disk format of the index, each on-disk cache entry stores the1347 * number of bytes to be stripped from the end of the previous name,1348 * and the bytes to append to the result, to come up with its name.1349 */1350static unsigned longexpand_name_field(struct strbuf *name,const char*cp_)1351{1352const unsigned char*ep, *cp = (const unsigned char*)cp_;1353size_t len =decode_varint(&cp);13541355if(name->len < len)1356die("malformed name field in the index");1357strbuf_remove(name, name->len - len, len);1358for(ep = cp; *ep; ep++)1359;/* find the end */1360strbuf_add(name, cp, ep - cp);1361return(const char*)ep +1- cp_;1362}13631364static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,1365unsigned long*ent_size,1366struct strbuf *previous_name)1367{1368struct cache_entry *ce;1369size_t len;1370const char*name;1371unsigned int flags;13721373/* On-disk flags are just 16 bits */1374 flags =ntoh_s(ondisk->flags);1375 len = flags & CE_NAMEMASK;13761377if(flags & CE_EXTENDED) {1378struct ondisk_cache_entry_extended *ondisk2;1379int extended_flags;1380 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1381 extended_flags =ntoh_s(ondisk2->flags2) <<16;1382/* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1383if(extended_flags & ~CE_EXTENDED_FLAGS)1384die("Unknown index entry format%08x", extended_flags);1385 flags |= extended_flags;1386 name = ondisk2->name;1387}1388else1389 name = ondisk->name;13901391if(!previous_name) {1392/* v3 and earlier */1393if(len == CE_NAMEMASK)1394 len =strlen(name);1395 ce =cache_entry_from_ondisk(ondisk, flags, name, len);13961397*ent_size =ondisk_ce_size(ce);1398}else{1399unsigned long consumed;1400 consumed =expand_name_field(previous_name, name);1401 ce =cache_entry_from_ondisk(ondisk, flags,1402 previous_name->buf,1403 previous_name->len);14041405*ent_size = (name - ((char*)ondisk)) + consumed;1406}1407return ce;1408}14091410/* remember to discard_cache() before reading a different cache! */1411intread_index_from(struct index_state *istate,const char*path)1412{1413int fd, i;1414struct stat st;1415unsigned long src_offset;1416struct cache_header *hdr;1417void*mmap;1418size_t mmap_size;1419struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;14201421if(istate->initialized)1422return istate->cache_nr;14231424 istate->timestamp.sec =0;1425 istate->timestamp.nsec =0;1426 fd =open(path, O_RDONLY);1427if(fd <0) {1428if(errno == ENOENT)1429return0;1430die_errno("index file open failed");1431}14321433if(fstat(fd, &st))1434die_errno("cannot stat the open index");14351436 mmap_size =xsize_t(st.st_size);1437if(mmap_size <sizeof(struct cache_header) +20)1438die("index file smaller than expected");14391440 mmap =xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd,0);1441if(mmap == MAP_FAILED)1442die_errno("unable to map index file");1443close(fd);14441445 hdr = mmap;1446if(verify_hdr(hdr, mmap_size) <0)1447goto unmap;14481449 istate->version =ntohl(hdr->hdr_version);1450 istate->cache_nr =ntohl(hdr->hdr_entries);1451 istate->cache_alloc =alloc_nr(istate->cache_nr);1452 istate->cache =xcalloc(istate->cache_alloc,sizeof(*istate->cache));1453 istate->initialized =1;14541455if(istate->version ==4)1456 previous_name = &previous_name_buf;1457else1458 previous_name = NULL;14591460 src_offset =sizeof(*hdr);1461for(i =0; i < istate->cache_nr; i++) {1462struct ondisk_cache_entry *disk_ce;1463struct cache_entry *ce;1464unsigned long consumed;14651466 disk_ce = (struct ondisk_cache_entry *)((char*)mmap + src_offset);1467 ce =create_from_disk(disk_ce, &consumed, previous_name);1468set_index_entry(istate, i, ce);14691470 src_offset += consumed;1471}1472strbuf_release(&previous_name_buf);1473 istate->timestamp.sec = st.st_mtime;1474 istate->timestamp.nsec =ST_MTIME_NSEC(st);14751476while(src_offset <= mmap_size -20-8) {1477/* After an array of active_nr index entries,1478 * there can be arbitrary number of extended1479 * sections, each of which is prefixed with1480 * extension name (4-byte) and section length1481 * in 4-byte network byte order.1482 */1483uint32_t extsize;1484memcpy(&extsize, (char*)mmap + src_offset +4,4);1485 extsize =ntohl(extsize);1486if(read_index_extension(istate,1487(const char*) mmap + src_offset,1488(char*) mmap + src_offset +8,1489 extsize) <0)1490goto unmap;1491 src_offset +=8;1492 src_offset += extsize;1493}1494munmap(mmap, mmap_size);1495return istate->cache_nr;14961497unmap:1498munmap(mmap, mmap_size);1499die("index file corrupt");1500}15011502intis_index_unborn(struct index_state *istate)1503{1504return(!istate->cache_nr && !istate->timestamp.sec);1505}15061507intdiscard_index(struct index_state *istate)1508{1509int i;15101511for(i =0; i < istate->cache_nr; i++)1512free(istate->cache[i]);1513resolve_undo_clear_index(istate);1514 istate->cache_nr =0;1515 istate->cache_changed =0;1516 istate->timestamp.sec =0;1517 istate->timestamp.nsec =0;1518free_name_hash(istate);1519cache_tree_free(&(istate->cache_tree));1520 istate->initialized =0;15211522/* no need to throw away allocated active_cache */1523return0;1524}15251526intunmerged_index(const struct index_state *istate)1527{1528int i;1529for(i =0; i < istate->cache_nr; i++) {1530if(ce_stage(istate->cache[i]))1531return1;1532}1533return0;1534}15351536#define WRITE_BUFFER_SIZE 81921537static unsigned char write_buffer[WRITE_BUFFER_SIZE];1538static unsigned long write_buffer_len;15391540static intce_write_flush(git_SHA_CTX *context,int fd)1541{1542unsigned int buffered = write_buffer_len;1543if(buffered) {1544git_SHA1_Update(context, write_buffer, buffered);1545if(write_in_full(fd, write_buffer, buffered) != buffered)1546return-1;1547 write_buffer_len =0;1548}1549return0;1550}15511552static intce_write(git_SHA_CTX *context,int fd,void*data,unsigned int len)1553{1554while(len) {1555unsigned int buffered = write_buffer_len;1556unsigned int partial = WRITE_BUFFER_SIZE - buffered;1557if(partial > len)1558 partial = len;1559memcpy(write_buffer + buffered, data, partial);1560 buffered += partial;1561if(buffered == WRITE_BUFFER_SIZE) {1562 write_buffer_len = buffered;1563if(ce_write_flush(context, fd))1564return-1;1565 buffered =0;1566}1567 write_buffer_len = buffered;1568 len -= partial;1569 data = (char*) data + partial;1570}1571return0;1572}15731574static intwrite_index_ext_header(git_SHA_CTX *context,int fd,1575unsigned int ext,unsigned int sz)1576{1577 ext =htonl(ext);1578 sz =htonl(sz);1579return((ce_write(context, fd, &ext,4) <0) ||1580(ce_write(context, fd, &sz,4) <0)) ? -1:0;1581}15821583static intce_flush(git_SHA_CTX *context,int fd)1584{1585unsigned int left = write_buffer_len;15861587if(left) {1588 write_buffer_len =0;1589git_SHA1_Update(context, write_buffer, left);1590}15911592/* Flush first if not enough space for SHA1 signature */1593if(left +20> WRITE_BUFFER_SIZE) {1594if(write_in_full(fd, write_buffer, left) != left)1595return-1;1596 left =0;1597}15981599/* Append the SHA1 signature at the end */1600git_SHA1_Final(write_buffer + left, context);1601 left +=20;1602return(write_in_full(fd, write_buffer, left) != left) ? -1:0;1603}16041605static voidce_smudge_racily_clean_entry(struct cache_entry *ce)1606{1607/*1608 * The only thing we care about in this function is to smudge the1609 * falsely clean entry due to touch-update-touch race, so we leave1610 * everything else as they are. We are called for entries whose1611 * ce_mtime match the index file mtime.1612 *1613 * Note that this actually does not do much for gitlinks, for1614 * which ce_match_stat_basic() always goes to the actual1615 * contents. The caller checks with is_racy_timestamp() which1616 * always says "no" for gitlinks, so we are not called for them ;-)1617 */1618struct stat st;16191620if(lstat(ce->name, &st) <0)1621return;1622if(ce_match_stat_basic(ce, &st))1623return;1624if(ce_modified_check_fs(ce, &st)) {1625/* This is "racily clean"; smudge it. Note that this1626 * is a tricky code. At first glance, it may appear1627 * that it can break with this sequence:1628 *1629 * $ echo xyzzy >frotz1630 * $ git-update-index --add frotz1631 * $ : >frotz1632 * $ sleep 31633 * $ echo filfre >nitfol1634 * $ git-update-index --add nitfol1635 *1636 * but it does not. When the second update-index runs,1637 * it notices that the entry "frotz" has the same timestamp1638 * as index, and if we were to smudge it by resetting its1639 * size to zero here, then the object name recorded1640 * in index is the 6-byte file but the cached stat information1641 * becomes zero --- which would then match what we would1642 * obtain from the filesystem next time we stat("frotz").1643 *1644 * However, the second update-index, before calling1645 * this function, notices that the cached size is 61646 * bytes and what is on the filesystem is an empty1647 * file, and never calls us, so the cached size information1648 * for "frotz" stays 6 which does not match the filesystem.1649 */1650 ce->ce_size =0;1651}1652}16531654/* Copy miscellaneous fields but not the name */1655static char*copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,1656struct cache_entry *ce)1657{1658short flags;16591660 ondisk->ctime.sec =htonl(ce->ce_ctime.sec);1661 ondisk->mtime.sec =htonl(ce->ce_mtime.sec);1662 ondisk->ctime.nsec =htonl(ce->ce_ctime.nsec);1663 ondisk->mtime.nsec =htonl(ce->ce_mtime.nsec);1664 ondisk->dev =htonl(ce->ce_dev);1665 ondisk->ino =htonl(ce->ce_ino);1666 ondisk->mode =htonl(ce->ce_mode);1667 ondisk->uid =htonl(ce->ce_uid);1668 ondisk->gid =htonl(ce->ce_gid);1669 ondisk->size =htonl(ce->ce_size);1670hashcpy(ondisk->sha1, ce->sha1);16711672 flags = ce->ce_flags;1673 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK :ce_namelen(ce));1674 ondisk->flags =htons(flags);1675if(ce->ce_flags & CE_EXTENDED) {1676struct ondisk_cache_entry_extended *ondisk2;1677 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1678 ondisk2->flags2 =htons((ce->ce_flags & CE_EXTENDED_FLAGS) >>16);1679return ondisk2->name;1680}1681else{1682return ondisk->name;1683}1684}16851686static intce_write_entry(git_SHA_CTX *c,int fd,struct cache_entry *ce,1687struct strbuf *previous_name)1688{1689int size;1690struct ondisk_cache_entry *ondisk;1691char*name;1692int result;16931694if(!previous_name) {1695 size =ondisk_ce_size(ce);1696 ondisk =xcalloc(1, size);1697 name =copy_cache_entry_to_ondisk(ondisk, ce);1698memcpy(name, ce->name,ce_namelen(ce));1699}else{1700int common, to_remove, prefix_size;1701unsigned char to_remove_vi[16];1702for(common =0;1703(ce->name[common] &&1704 common < previous_name->len &&1705 ce->name[common] == previous_name->buf[common]);1706 common++)1707;/* still matching */1708 to_remove = previous_name->len - common;1709 prefix_size =encode_varint(to_remove, to_remove_vi);17101711if(ce->ce_flags & CE_EXTENDED)1712 size =offsetof(struct ondisk_cache_entry_extended, name);1713else1714 size =offsetof(struct ondisk_cache_entry, name);1715 size += prefix_size + (ce_namelen(ce) - common +1);17161717 ondisk =xcalloc(1, size);1718 name =copy_cache_entry_to_ondisk(ondisk, ce);1719memcpy(name, to_remove_vi, prefix_size);1720memcpy(name + prefix_size, ce->name + common,ce_namelen(ce) - common);17211722strbuf_splice(previous_name, common, to_remove,1723 ce->name + common,ce_namelen(ce) - common);1724}17251726 result =ce_write(c, fd, ondisk, size);1727free(ondisk);1728return result;1729}17301731static inthas_racy_timestamp(struct index_state *istate)1732{1733int entries = istate->cache_nr;1734int i;17351736for(i =0; i < entries; i++) {1737struct cache_entry *ce = istate->cache[i];1738if(is_racy_timestamp(istate, ce))1739return1;1740}1741return0;1742}17431744/*1745 * Opportunisticly update the index but do not complain if we can't1746 */1747voidupdate_index_if_able(struct index_state *istate,struct lock_file *lockfile)1748{1749if((istate->cache_changed ||has_racy_timestamp(istate)) &&1750!write_index(istate, lockfile->fd))1751commit_locked_index(lockfile);1752else1753rollback_lock_file(lockfile);1754}17551756intwrite_index(struct index_state *istate,int newfd)1757{1758 git_SHA_CTX c;1759struct cache_header hdr;1760int i, err, removed, extended, hdr_version;1761struct cache_entry **cache = istate->cache;1762int entries = istate->cache_nr;1763struct stat st;1764struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;17651766for(i = removed = extended =0; i < entries; i++) {1767if(cache[i]->ce_flags & CE_REMOVE)1768 removed++;17691770/* reduce extended entries if possible */1771 cache[i]->ce_flags &= ~CE_EXTENDED;1772if(cache[i]->ce_flags & CE_EXTENDED_FLAGS) {1773 extended++;1774 cache[i]->ce_flags |= CE_EXTENDED;1775}1776}17771778if(!istate->version)1779 istate->version = INDEX_FORMAT_DEFAULT;17801781/* demote version 3 to version 2 when the latter suffices */1782if(istate->version ==3|| istate->version ==2)1783 istate->version = extended ?3:2;17841785 hdr_version = istate->version;17861787 hdr.hdr_signature =htonl(CACHE_SIGNATURE);1788 hdr.hdr_version =htonl(hdr_version);1789 hdr.hdr_entries =htonl(entries - removed);17901791git_SHA1_Init(&c);1792if(ce_write(&c, newfd, &hdr,sizeof(hdr)) <0)1793return-1;17941795 previous_name = (hdr_version ==4) ? &previous_name_buf : NULL;1796for(i =0; i < entries; i++) {1797struct cache_entry *ce = cache[i];1798if(ce->ce_flags & CE_REMOVE)1799continue;1800if(!ce_uptodate(ce) &&is_racy_timestamp(istate, ce))1801ce_smudge_racily_clean_entry(ce);1802if(is_null_sha1(ce->sha1))1803returnerror("cache entry has null sha1:%s", ce->name);1804if(ce_write_entry(&c, newfd, ce, previous_name) <0)1805return-1;1806}1807strbuf_release(&previous_name_buf);18081809/* Write extension data here */1810if(istate->cache_tree) {1811struct strbuf sb = STRBUF_INIT;18121813cache_tree_write(&sb, istate->cache_tree);1814 err =write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) <01815||ce_write(&c, newfd, sb.buf, sb.len) <0;1816strbuf_release(&sb);1817if(err)1818return-1;1819}1820if(istate->resolve_undo) {1821struct strbuf sb = STRBUF_INIT;18221823resolve_undo_write(&sb, istate->resolve_undo);1824 err =write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,1825 sb.len) <01826||ce_write(&c, newfd, sb.buf, sb.len) <0;1827strbuf_release(&sb);1828if(err)1829return-1;1830}18311832if(ce_flush(&c, newfd) ||fstat(newfd, &st))1833return-1;1834 istate->timestamp.sec = (unsigned int)st.st_mtime;1835 istate->timestamp.nsec =ST_MTIME_NSEC(st);1836return0;1837}18381839/*1840 * Read the index file that is potentially unmerged into given1841 * index_state, dropping any unmerged entries. Returns true if1842 * the index is unmerged. Callers who want to refuse to work1843 * from an unmerged state can call this and check its return value,1844 * instead of calling read_cache().1845 */1846intread_index_unmerged(struct index_state *istate)1847{1848int i;1849int unmerged =0;18501851read_index(istate);1852for(i =0; i < istate->cache_nr; i++) {1853struct cache_entry *ce = istate->cache[i];1854struct cache_entry *new_ce;1855int size, len;18561857if(!ce_stage(ce))1858continue;1859 unmerged =1;1860 len =ce_namelen(ce);1861 size =cache_entry_size(len);1862 new_ce =xcalloc(1, size);1863memcpy(new_ce->name, ce->name, len);1864 new_ce->ce_flags =create_ce_flags(0) | CE_CONFLICTED;1865 new_ce->ce_namelen = len;1866 new_ce->ce_mode = ce->ce_mode;1867if(add_index_entry(istate, new_ce,0))1868returnerror("%s: cannot drop to stage #0",1869 ce->name);1870 i =index_name_pos(istate, new_ce->name, len);1871}1872return unmerged;1873}18741875/*1876 * Returns 1 if the path is an "other" path with respect to1877 * the index; that is, the path is not mentioned in the index at all,1878 * either as a file, a directory with some files in the index,1879 * or as an unmerged entry.1880 *1881 * We helpfully remove a trailing "/" from directories so that1882 * the output of read_directory can be used as-is.1883 */1884intindex_name_is_other(const struct index_state *istate,const char*name,1885int namelen)1886{1887int pos;1888if(namelen && name[namelen -1] =='/')1889 namelen--;1890 pos =index_name_pos(istate, name, namelen);1891if(0<= pos)1892return0;/* exact match */1893 pos = -pos -1;1894if(pos < istate->cache_nr) {1895struct cache_entry *ce = istate->cache[pos];1896if(ce_namelen(ce) == namelen &&1897!memcmp(ce->name, name, namelen))1898return0;/* Yup, this one exists unmerged */1899}1900return1;1901}19021903void*read_blob_data_from_index(struct index_state *istate,const char*path,unsigned long*size)1904{1905int pos, len;1906unsigned long sz;1907enum object_type type;1908void*data;19091910 len =strlen(path);1911 pos =index_name_pos(istate, path, len);1912if(pos <0) {1913/*1914 * We might be in the middle of a merge, in which1915 * case we would read stage #2 (ours).1916 */1917int i;1918for(i = -pos -1;1919(pos <0&& i < istate->cache_nr &&1920!strcmp(istate->cache[i]->name, path));1921 i++)1922if(ce_stage(istate->cache[i]) ==2)1923 pos = i;1924}1925if(pos <0)1926return NULL;1927 data =read_sha1_file(istate->cache[pos]->sha1, &type, &sz);1928if(!data || type != OBJ_BLOB) {1929free(data);1930return NULL;1931}1932if(size)1933*size = sz;1934return data;1935}