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"config.h" 9#include"tempfile.h" 10#include"lockfile.h" 11#include"cache-tree.h" 12#include"refs.h" 13#include"dir.h" 14#include"tree.h" 15#include"commit.h" 16#include"blob.h" 17#include"resolve-undo.h" 18#include"strbuf.h" 19#include"varint.h" 20#include"split-index.h" 21#include"utf8.h" 22#include"fsmonitor.h" 23 24/* Mask for the name length in ce_flags in the on-disk index */ 25 26#define CE_NAMEMASK (0x0fff) 27 28/* Index extensions. 29 * 30 * The first letter should be 'A'..'Z' for extensions that are not 31 * necessary for a correct operation (i.e. optimization data). 32 * When new extensions are added that _needs_ to be understood in 33 * order to correctly interpret the index file, pick character that 34 * is outside the range, to cause the reader to abort. 35 */ 36 37#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) ) 38#define CACHE_EXT_TREE 0x54524545/* "TREE" */ 39#define CACHE_EXT_RESOLVE_UNDO 0x52455543/* "REUC" */ 40#define CACHE_EXT_LINK 0x6c696e6b/* "link" */ 41#define CACHE_EXT_UNTRACKED 0x554E5452/* "UNTR" */ 42#define CACHE_EXT_FSMONITOR 0x46534D4E/* "FSMN" */ 43 44/* changes that can be kept in $GIT_DIR/index (basically all extensions) */ 45#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \ 46 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \ 47 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED) 48 49struct index_state the_index; 50static const char*alternate_index_output; 51 52static voidset_index_entry(struct index_state *istate,int nr,struct cache_entry *ce) 53{ 54 istate->cache[nr] = ce; 55add_name_hash(istate, ce); 56} 57 58static voidreplace_index_entry(struct index_state *istate,int nr,struct cache_entry *ce) 59{ 60struct cache_entry *old = istate->cache[nr]; 61 62replace_index_entry_in_base(istate, old, ce); 63remove_name_hash(istate, old); 64discard_cache_entry(old); 65 ce->ce_flags &= ~CE_HASHED; 66set_index_entry(istate, nr, ce); 67 ce->ce_flags |= CE_UPDATE_IN_BASE; 68mark_fsmonitor_invalid(istate, ce); 69 istate->cache_changed |= CE_ENTRY_CHANGED; 70} 71 72voidrename_index_entry_at(struct index_state *istate,int nr,const char*new_name) 73{ 74struct cache_entry *old_entry = istate->cache[nr], *new_entry; 75int namelen =strlen(new_name); 76 77 new_entry =make_empty_cache_entry(istate, namelen); 78copy_cache_entry(new_entry, old_entry); 79 new_entry->ce_flags &= ~CE_HASHED; 80 new_entry->ce_namelen = namelen; 81 new_entry->index =0; 82memcpy(new_entry->name, new_name, namelen +1); 83 84cache_tree_invalidate_path(istate, old_entry->name); 85untracked_cache_remove_from_index(istate, old_entry->name); 86remove_index_entry_at(istate, nr); 87add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE); 88} 89 90voidfill_stat_data(struct stat_data *sd,struct stat *st) 91{ 92 sd->sd_ctime.sec = (unsigned int)st->st_ctime; 93 sd->sd_mtime.sec = (unsigned int)st->st_mtime; 94 sd->sd_ctime.nsec =ST_CTIME_NSEC(*st); 95 sd->sd_mtime.nsec =ST_MTIME_NSEC(*st); 96 sd->sd_dev = st->st_dev; 97 sd->sd_ino = st->st_ino; 98 sd->sd_uid = st->st_uid; 99 sd->sd_gid = st->st_gid; 100 sd->sd_size = st->st_size; 101} 102 103intmatch_stat_data(const struct stat_data *sd,struct stat *st) 104{ 105int changed =0; 106 107if(sd->sd_mtime.sec != (unsigned int)st->st_mtime) 108 changed |= MTIME_CHANGED; 109if(trust_ctime && check_stat && 110 sd->sd_ctime.sec != (unsigned int)st->st_ctime) 111 changed |= CTIME_CHANGED; 112 113#ifdef USE_NSEC 114if(check_stat && sd->sd_mtime.nsec !=ST_MTIME_NSEC(*st)) 115 changed |= MTIME_CHANGED; 116if(trust_ctime && check_stat && 117 sd->sd_ctime.nsec !=ST_CTIME_NSEC(*st)) 118 changed |= CTIME_CHANGED; 119#endif 120 121if(check_stat) { 122if(sd->sd_uid != (unsigned int) st->st_uid || 123 sd->sd_gid != (unsigned int) st->st_gid) 124 changed |= OWNER_CHANGED; 125if(sd->sd_ino != (unsigned int) st->st_ino) 126 changed |= INODE_CHANGED; 127} 128 129#ifdef USE_STDEV 130/* 131 * st_dev breaks on network filesystems where different 132 * clients will have different views of what "device" 133 * the filesystem is on 134 */ 135if(check_stat && sd->sd_dev != (unsigned int) st->st_dev) 136 changed |= INODE_CHANGED; 137#endif 138 139if(sd->sd_size != (unsigned int) st->st_size) 140 changed |= DATA_CHANGED; 141 142return changed; 143} 144 145/* 146 * This only updates the "non-critical" parts of the directory 147 * cache, ie the parts that aren't tracked by GIT, and only used 148 * to validate the cache. 149 */ 150voidfill_stat_cache_info(struct cache_entry *ce,struct stat *st) 151{ 152fill_stat_data(&ce->ce_stat_data, st); 153 154if(assume_unchanged) 155 ce->ce_flags |= CE_VALID; 156 157if(S_ISREG(st->st_mode)) { 158ce_mark_uptodate(ce); 159mark_fsmonitor_valid(ce); 160} 161} 162 163static intce_compare_data(const struct cache_entry *ce,struct stat *st) 164{ 165int match = -1; 166int fd =git_open_cloexec(ce->name, O_RDONLY); 167 168if(fd >=0) { 169struct object_id oid; 170if(!index_fd(&oid, fd, st, OBJ_BLOB, ce->name,0)) 171 match =oidcmp(&oid, &ce->oid); 172/* index_fd() closed the file descriptor already */ 173} 174return match; 175} 176 177static intce_compare_link(const struct cache_entry *ce,size_t expected_size) 178{ 179int match = -1; 180void*buffer; 181unsigned long size; 182enum object_type type; 183struct strbuf sb = STRBUF_INIT; 184 185if(strbuf_readlink(&sb, ce->name, expected_size)) 186return-1; 187 188 buffer =read_object_file(&ce->oid, &type, &size); 189if(buffer) { 190if(size == sb.len) 191 match =memcmp(buffer, sb.buf, size); 192free(buffer); 193} 194strbuf_release(&sb); 195return match; 196} 197 198static intce_compare_gitlink(const struct cache_entry *ce) 199{ 200struct object_id oid; 201 202/* 203 * We don't actually require that the .git directory 204 * under GITLINK directory be a valid git directory. It 205 * might even be missing (in case nobody populated that 206 * sub-project). 207 * 208 * If so, we consider it always to match. 209 */ 210if(resolve_gitlink_ref(ce->name,"HEAD", &oid) <0) 211return0; 212returnoidcmp(&oid, &ce->oid); 213} 214 215static intce_modified_check_fs(const struct cache_entry *ce,struct stat *st) 216{ 217switch(st->st_mode & S_IFMT) { 218case S_IFREG: 219if(ce_compare_data(ce, st)) 220return DATA_CHANGED; 221break; 222case S_IFLNK: 223if(ce_compare_link(ce,xsize_t(st->st_size))) 224return DATA_CHANGED; 225break; 226case S_IFDIR: 227if(S_ISGITLINK(ce->ce_mode)) 228returnce_compare_gitlink(ce) ? DATA_CHANGED :0; 229/* else fallthrough */ 230default: 231return TYPE_CHANGED; 232} 233return0; 234} 235 236static intce_match_stat_basic(const struct cache_entry *ce,struct stat *st) 237{ 238unsigned int changed =0; 239 240if(ce->ce_flags & CE_REMOVE) 241return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; 242 243switch(ce->ce_mode & S_IFMT) { 244case S_IFREG: 245 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED :0; 246/* We consider only the owner x bit to be relevant for 247 * "mode changes" 248 */ 249if(trust_executable_bit && 250(0100& (ce->ce_mode ^ st->st_mode))) 251 changed |= MODE_CHANGED; 252break; 253case S_IFLNK: 254if(!S_ISLNK(st->st_mode) && 255(has_symlinks || !S_ISREG(st->st_mode))) 256 changed |= TYPE_CHANGED; 257break; 258case S_IFGITLINK: 259/* We ignore most of the st_xxx fields for gitlinks */ 260if(!S_ISDIR(st->st_mode)) 261 changed |= TYPE_CHANGED; 262else if(ce_compare_gitlink(ce)) 263 changed |= DATA_CHANGED; 264return changed; 265default: 266die("internal error: ce_mode is%o", ce->ce_mode); 267} 268 269 changed |=match_stat_data(&ce->ce_stat_data, st); 270 271/* Racily smudged entry? */ 272if(!ce->ce_stat_data.sd_size) { 273if(!is_empty_blob_sha1(ce->oid.hash)) 274 changed |= DATA_CHANGED; 275} 276 277return changed; 278} 279 280static intis_racy_stat(const struct index_state *istate, 281const struct stat_data *sd) 282{ 283return(istate->timestamp.sec && 284#ifdef USE_NSEC 285/* nanosecond timestamped files can also be racy! */ 286(istate->timestamp.sec < sd->sd_mtime.sec || 287(istate->timestamp.sec == sd->sd_mtime.sec && 288 istate->timestamp.nsec <= sd->sd_mtime.nsec)) 289#else 290 istate->timestamp.sec <= sd->sd_mtime.sec 291#endif 292); 293} 294 295static intis_racy_timestamp(const struct index_state *istate, 296const struct cache_entry *ce) 297{ 298return(!S_ISGITLINK(ce->ce_mode) && 299is_racy_stat(istate, &ce->ce_stat_data)); 300} 301 302intmatch_stat_data_racy(const struct index_state *istate, 303const struct stat_data *sd,struct stat *st) 304{ 305if(is_racy_stat(istate, sd)) 306return MTIME_CHANGED; 307returnmatch_stat_data(sd, st); 308} 309 310intie_match_stat(struct index_state *istate, 311const struct cache_entry *ce,struct stat *st, 312unsigned int options) 313{ 314unsigned int changed; 315int ignore_valid = options & CE_MATCH_IGNORE_VALID; 316int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE; 317int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 318int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR; 319 320if(!ignore_fsmonitor) 321refresh_fsmonitor(istate); 322/* 323 * If it's marked as always valid in the index, it's 324 * valid whatever the checked-out copy says. 325 * 326 * skip-worktree has the same effect with higher precedence 327 */ 328if(!ignore_skip_worktree &&ce_skip_worktree(ce)) 329return0; 330if(!ignore_valid && (ce->ce_flags & CE_VALID)) 331return0; 332if(!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) 333return0; 334 335/* 336 * Intent-to-add entries have not been added, so the index entry 337 * by definition never matches what is in the work tree until it 338 * actually gets added. 339 */ 340if(ce_intent_to_add(ce)) 341return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED; 342 343 changed =ce_match_stat_basic(ce, st); 344 345/* 346 * Within 1 second of this sequence: 347 * echo xyzzy >file && git-update-index --add file 348 * running this command: 349 * echo frotz >file 350 * would give a falsely clean cache entry. The mtime and 351 * length match the cache, and other stat fields do not change. 352 * 353 * We could detect this at update-index time (the cache entry 354 * being registered/updated records the same time as "now") 355 * and delay the return from git-update-index, but that would 356 * effectively mean we can make at most one commit per second, 357 * which is not acceptable. Instead, we check cache entries 358 * whose mtime are the same as the index file timestamp more 359 * carefully than others. 360 */ 361if(!changed &&is_racy_timestamp(istate, ce)) { 362if(assume_racy_is_modified) 363 changed |= DATA_CHANGED; 364else 365 changed |=ce_modified_check_fs(ce, st); 366} 367 368return changed; 369} 370 371intie_modified(struct index_state *istate, 372const struct cache_entry *ce, 373struct stat *st,unsigned int options) 374{ 375int changed, changed_fs; 376 377 changed =ie_match_stat(istate, ce, st, options); 378if(!changed) 379return0; 380/* 381 * If the mode or type has changed, there's no point in trying 382 * to refresh the entry - it's not going to match 383 */ 384if(changed & (MODE_CHANGED | TYPE_CHANGED)) 385return changed; 386 387/* 388 * Immediately after read-tree or update-index --cacheinfo, 389 * the length field is zero, as we have never even read the 390 * lstat(2) information once, and we cannot trust DATA_CHANGED 391 * returned by ie_match_stat() which in turn was returned by 392 * ce_match_stat_basic() to signal that the filesize of the 393 * blob changed. We have to actually go to the filesystem to 394 * see if the contents match, and if so, should answer "unchanged". 395 * 396 * The logic does not apply to gitlinks, as ce_match_stat_basic() 397 * already has checked the actual HEAD from the filesystem in the 398 * subproject. If ie_match_stat() already said it is different, 399 * then we know it is. 400 */ 401if((changed & DATA_CHANGED) && 402(S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size !=0)) 403return changed; 404 405 changed_fs =ce_modified_check_fs(ce, st); 406if(changed_fs) 407return changed | changed_fs; 408return0; 409} 410 411intbase_name_compare(const char*name1,int len1,int mode1, 412const char*name2,int len2,int mode2) 413{ 414unsigned char c1, c2; 415int len = len1 < len2 ? len1 : len2; 416int cmp; 417 418 cmp =memcmp(name1, name2, len); 419if(cmp) 420return cmp; 421 c1 = name1[len]; 422 c2 = name2[len]; 423if(!c1 &&S_ISDIR(mode1)) 424 c1 ='/'; 425if(!c2 &&S_ISDIR(mode2)) 426 c2 ='/'; 427return(c1 < c2) ? -1: (c1 > c2) ?1:0; 428} 429 430/* 431 * df_name_compare() is identical to base_name_compare(), except it 432 * compares conflicting directory/file entries as equal. Note that 433 * while a directory name compares as equal to a regular file, they 434 * then individually compare _differently_ to a filename that has 435 * a dot after the basename (because '\0' < '.' < '/'). 436 * 437 * This is used by routines that want to traverse the git namespace 438 * but then handle conflicting entries together when possible. 439 */ 440intdf_name_compare(const char*name1,int len1,int mode1, 441const char*name2,int len2,int mode2) 442{ 443int len = len1 < len2 ? len1 : len2, cmp; 444unsigned char c1, c2; 445 446 cmp =memcmp(name1, name2, len); 447if(cmp) 448return cmp; 449/* Directories and files compare equal (same length, same name) */ 450if(len1 == len2) 451return0; 452 c1 = name1[len]; 453if(!c1 &&S_ISDIR(mode1)) 454 c1 ='/'; 455 c2 = name2[len]; 456if(!c2 &&S_ISDIR(mode2)) 457 c2 ='/'; 458if(c1 =='/'&& !c2) 459return0; 460if(c2 =='/'&& !c1) 461return0; 462return c1 - c2; 463} 464 465intname_compare(const char*name1,size_t len1,const char*name2,size_t len2) 466{ 467size_t min_len = (len1 < len2) ? len1 : len2; 468int cmp =memcmp(name1, name2, min_len); 469if(cmp) 470return cmp; 471if(len1 < len2) 472return-1; 473if(len1 > len2) 474return1; 475return0; 476} 477 478intcache_name_stage_compare(const char*name1,int len1,int stage1,const char*name2,int len2,int stage2) 479{ 480int cmp; 481 482 cmp =name_compare(name1, len1, name2, len2); 483if(cmp) 484return cmp; 485 486if(stage1 < stage2) 487return-1; 488if(stage1 > stage2) 489return1; 490return0; 491} 492 493static intindex_name_stage_pos(const struct index_state *istate,const char*name,int namelen,int stage) 494{ 495int first, last; 496 497 first =0; 498 last = istate->cache_nr; 499while(last > first) { 500int next = (last + first) >>1; 501struct cache_entry *ce = istate->cache[next]; 502int cmp =cache_name_stage_compare(name, namelen, stage, ce->name,ce_namelen(ce),ce_stage(ce)); 503if(!cmp) 504return next; 505if(cmp <0) { 506 last = next; 507continue; 508} 509 first = next+1; 510} 511return-first-1; 512} 513 514intindex_name_pos(const struct index_state *istate,const char*name,int namelen) 515{ 516returnindex_name_stage_pos(istate, name, namelen,0); 517} 518 519intremove_index_entry_at(struct index_state *istate,int pos) 520{ 521struct cache_entry *ce = istate->cache[pos]; 522 523record_resolve_undo(istate, ce); 524remove_name_hash(istate, ce); 525save_or_free_index_entry(istate, ce); 526 istate->cache_changed |= CE_ENTRY_REMOVED; 527 istate->cache_nr--; 528if(pos >= istate->cache_nr) 529return0; 530MOVE_ARRAY(istate->cache + pos, istate->cache + pos +1, 531 istate->cache_nr - pos); 532return1; 533} 534 535/* 536 * Remove all cache entries marked for removal, that is where 537 * CE_REMOVE is set in ce_flags. This is much more effective than 538 * calling remove_index_entry_at() for each entry to be removed. 539 */ 540voidremove_marked_cache_entries(struct index_state *istate) 541{ 542struct cache_entry **ce_array = istate->cache; 543unsigned int i, j; 544 545for(i = j =0; i < istate->cache_nr; i++) { 546if(ce_array[i]->ce_flags & CE_REMOVE) { 547remove_name_hash(istate, ce_array[i]); 548save_or_free_index_entry(istate, ce_array[i]); 549} 550else 551 ce_array[j++] = ce_array[i]; 552} 553if(j == istate->cache_nr) 554return; 555 istate->cache_changed |= CE_ENTRY_REMOVED; 556 istate->cache_nr = j; 557} 558 559intremove_file_from_index(struct index_state *istate,const char*path) 560{ 561int pos =index_name_pos(istate, path,strlen(path)); 562if(pos <0) 563 pos = -pos-1; 564cache_tree_invalidate_path(istate, path); 565untracked_cache_remove_from_index(istate, path); 566while(pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 567remove_index_entry_at(istate, pos); 568return0; 569} 570 571static intcompare_name(struct cache_entry *ce,const char*path,int namelen) 572{ 573return namelen !=ce_namelen(ce) ||memcmp(path, ce->name, namelen); 574} 575 576static intindex_name_pos_also_unmerged(struct index_state *istate, 577const char*path,int namelen) 578{ 579int pos =index_name_pos(istate, path, namelen); 580struct cache_entry *ce; 581 582if(pos >=0) 583return pos; 584 585/* maybe unmerged? */ 586 pos = -1- pos; 587if(pos >= istate->cache_nr || 588compare_name((ce = istate->cache[pos]), path, namelen)) 589return-1; 590 591/* order of preference: stage 2, 1, 3 */ 592if(ce_stage(ce) ==1&& pos +1< istate->cache_nr && 593ce_stage((ce = istate->cache[pos +1])) ==2&& 594!compare_name(ce, path, namelen)) 595 pos++; 596return pos; 597} 598 599static intdifferent_name(struct cache_entry *ce,struct cache_entry *alias) 600{ 601int len =ce_namelen(ce); 602returnce_namelen(alias) != len ||memcmp(ce->name, alias->name, len); 603} 604 605/* 606 * If we add a filename that aliases in the cache, we will use the 607 * name that we already have - but we don't want to update the same 608 * alias twice, because that implies that there were actually two 609 * different files with aliasing names! 610 * 611 * So we use the CE_ADDED flag to verify that the alias was an old 612 * one before we accept it as 613 */ 614static struct cache_entry *create_alias_ce(struct index_state *istate, 615struct cache_entry *ce, 616struct cache_entry *alias) 617{ 618int len; 619struct cache_entry *new_entry; 620 621if(alias->ce_flags & CE_ADDED) 622die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name); 623 624/* Ok, create the new entry using the name of the existing alias */ 625 len =ce_namelen(alias); 626 new_entry =make_empty_cache_entry(istate, len); 627memcpy(new_entry->name, alias->name, len); 628copy_cache_entry(new_entry, ce); 629save_or_free_index_entry(istate, ce); 630return new_entry; 631} 632 633voidset_object_name_for_intent_to_add_entry(struct cache_entry *ce) 634{ 635struct object_id oid; 636if(write_object_file("",0, blob_type, &oid)) 637die("cannot create an empty blob in the object database"); 638oidcpy(&ce->oid, &oid); 639} 640 641intadd_to_index(struct index_state *istate,const char*path,struct stat *st,int flags) 642{ 643int namelen, was_same; 644 mode_t st_mode = st->st_mode; 645struct cache_entry *ce, *alias = NULL; 646unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY; 647int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); 648int pretend = flags & ADD_CACHE_PRETEND; 649int intent_only = flags & ADD_CACHE_INTENT; 650int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| 651(intent_only ? ADD_CACHE_NEW_ONLY :0)); 652int newflags = HASH_WRITE_OBJECT; 653 654if(flags & HASH_RENORMALIZE) 655 newflags |= HASH_RENORMALIZE; 656 657if(!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 658returnerror("%s: can only add regular files, symbolic links or git-directories", path); 659 660 namelen =strlen(path); 661if(S_ISDIR(st_mode)) { 662while(namelen && path[namelen-1] =='/') 663 namelen--; 664} 665 ce =make_empty_cache_entry(istate, namelen); 666memcpy(ce->name, path, namelen); 667 ce->ce_namelen = namelen; 668if(!intent_only) 669fill_stat_cache_info(ce, st); 670else 671 ce->ce_flags |= CE_INTENT_TO_ADD; 672 673 674if(trust_executable_bit && has_symlinks) { 675 ce->ce_mode =create_ce_mode(st_mode); 676}else{ 677/* If there is an existing entry, pick the mode bits and type 678 * from it, otherwise assume unexecutable regular file. 679 */ 680struct cache_entry *ent; 681int pos =index_name_pos_also_unmerged(istate, path, namelen); 682 683 ent = (0<= pos) ? istate->cache[pos] : NULL; 684 ce->ce_mode =ce_mode_from_stat(ent, st_mode); 685} 686 687/* When core.ignorecase=true, determine if a directory of the same name but differing 688 * case already exists within the Git repository. If it does, ensure the directory 689 * case of the file being added to the repository matches (is folded into) the existing 690 * entry's directory case. 691 */ 692if(ignore_case) { 693adjust_dirname_case(istate, ce->name); 694} 695if(!(flags & HASH_RENORMALIZE)) { 696 alias =index_file_exists(istate, ce->name, 697ce_namelen(ce), ignore_case); 698if(alias && 699!ce_stage(alias) && 700!ie_match_stat(istate, alias, st, ce_option)) { 701/* Nothing changed, really */ 702if(!S_ISGITLINK(alias->ce_mode)) 703ce_mark_uptodate(alias); 704 alias->ce_flags |= CE_ADDED; 705 706discard_cache_entry(ce); 707return0; 708} 709} 710if(!intent_only) { 711if(index_path(&ce->oid, path, st, newflags)) { 712discard_cache_entry(ce); 713returnerror("unable to index file%s", path); 714} 715}else 716set_object_name_for_intent_to_add_entry(ce); 717 718if(ignore_case && alias &&different_name(ce, alias)) 719 ce =create_alias_ce(istate, ce, alias); 720 ce->ce_flags |= CE_ADDED; 721 722/* It was suspected to be racily clean, but it turns out to be Ok */ 723 was_same = (alias && 724!ce_stage(alias) && 725!oidcmp(&alias->oid, &ce->oid) && 726 ce->ce_mode == alias->ce_mode); 727 728if(pretend) 729discard_cache_entry(ce); 730else if(add_index_entry(istate, ce, add_option)) { 731discard_cache_entry(ce); 732returnerror("unable to add%sto index", path); 733} 734if(verbose && !was_same) 735printf("add '%s'\n", path); 736return0; 737} 738 739intadd_file_to_index(struct index_state *istate,const char*path,int flags) 740{ 741struct stat st; 742if(lstat(path, &st)) 743die_errno("unable to stat '%s'", path); 744returnadd_to_index(istate, path, &st, flags); 745} 746 747struct cache_entry *make_empty_cache_entry(struct index_state *istate,size_t len) 748{ 749returnxcalloc(1,cache_entry_size(len)); 750} 751 752struct cache_entry *make_empty_transient_cache_entry(size_t len) 753{ 754returnxcalloc(1,cache_entry_size(len)); 755} 756 757struct cache_entry *make_cache_entry(struct index_state *istate, 758unsigned int mode, 759const struct object_id *oid, 760const char*path, 761int stage, 762unsigned int refresh_options) 763{ 764struct cache_entry *ce, *ret; 765int len; 766 767if(!verify_path(path, mode)) { 768error("Invalid path '%s'", path); 769return NULL; 770} 771 772 len =strlen(path); 773 ce =make_empty_cache_entry(istate, len); 774 775oidcpy(&ce->oid, oid); 776memcpy(ce->name, path, len); 777 ce->ce_flags =create_ce_flags(stage); 778 ce->ce_namelen = len; 779 ce->ce_mode =create_ce_mode(mode); 780 781 ret =refresh_cache_entry(&the_index, ce, refresh_options); 782if(ret != ce) 783discard_cache_entry(ce); 784return ret; 785} 786 787struct cache_entry *make_transient_cache_entry(unsigned int mode,const struct object_id *oid, 788const char*path,int stage) 789{ 790struct cache_entry *ce; 791int len; 792 793if(!verify_path(path, mode)) { 794error("Invalid path '%s'", path); 795return NULL; 796} 797 798 len =strlen(path); 799 ce =make_empty_transient_cache_entry(len); 800 801oidcpy(&ce->oid, oid); 802memcpy(ce->name, path, len); 803 ce->ce_flags =create_ce_flags(stage); 804 ce->ce_namelen = len; 805 ce->ce_mode =create_ce_mode(mode); 806 807return ce; 808} 809 810/* 811 * Chmod an index entry with either +x or -x. 812 * 813 * Returns -1 if the chmod for the particular cache entry failed (if it's 814 * not a regular file), -2 if an invalid flip argument is passed in, 0 815 * otherwise. 816 */ 817intchmod_index_entry(struct index_state *istate,struct cache_entry *ce, 818char flip) 819{ 820if(!S_ISREG(ce->ce_mode)) 821return-1; 822switch(flip) { 823case'+': 824 ce->ce_mode |=0111; 825break; 826case'-': 827 ce->ce_mode &= ~0111; 828break; 829default: 830return-2; 831} 832cache_tree_invalidate_path(istate, ce->name); 833 ce->ce_flags |= CE_UPDATE_IN_BASE; 834mark_fsmonitor_invalid(istate, ce); 835 istate->cache_changed |= CE_ENTRY_CHANGED; 836 837return0; 838} 839 840intce_same_name(const struct cache_entry *a,const struct cache_entry *b) 841{ 842int len =ce_namelen(a); 843returnce_namelen(b) == len && !memcmp(a->name, b->name, len); 844} 845 846/* 847 * We fundamentally don't like some paths: we don't want 848 * dot or dot-dot anywhere, and for obvious reasons don't 849 * want to recurse into ".git" either. 850 * 851 * Also, we don't want double slashes or slashes at the 852 * end that can make pathnames ambiguous. 853 */ 854static intverify_dotfile(const char*rest,unsigned mode) 855{ 856/* 857 * The first character was '.', but that 858 * has already been discarded, we now test 859 * the rest. 860 */ 861 862/* "." is not allowed */ 863if(*rest =='\0'||is_dir_sep(*rest)) 864return0; 865 866switch(*rest) { 867/* 868 * ".git" followed by NUL or slash is bad. Note that we match 869 * case-insensitively here, even if ignore_case is not set. 870 * This outlaws ".GIT" everywhere out of an abundance of caution, 871 * since there's really no good reason to allow it. 872 * 873 * Once we've seen ".git", we can also find ".gitmodules", etc (also 874 * case-insensitively). 875 */ 876case'g': 877case'G': 878if(rest[1] !='i'&& rest[1] !='I') 879break; 880if(rest[2] !='t'&& rest[2] !='T') 881break; 882if(rest[3] =='\0'||is_dir_sep(rest[3])) 883return0; 884if(S_ISLNK(mode)) { 885 rest +=3; 886if(skip_iprefix(rest,"modules", &rest) && 887(*rest =='\0'||is_dir_sep(*rest))) 888return0; 889} 890break; 891case'.': 892if(rest[1] =='\0'||is_dir_sep(rest[1])) 893return0; 894} 895return1; 896} 897 898intverify_path(const char*path,unsigned mode) 899{ 900char c; 901 902if(has_dos_drive_prefix(path)) 903return0; 904 905goto inside; 906for(;;) { 907if(!c) 908return1; 909if(is_dir_sep(c)) { 910inside: 911if(protect_hfs) { 912if(is_hfs_dotgit(path)) 913return0; 914if(S_ISLNK(mode)) { 915if(is_hfs_dotgitmodules(path)) 916return0; 917} 918} 919if(protect_ntfs) { 920if(is_ntfs_dotgit(path)) 921return0; 922if(S_ISLNK(mode)) { 923if(is_ntfs_dotgitmodules(path)) 924return0; 925} 926} 927 928 c = *path++; 929if((c =='.'&& !verify_dotfile(path, mode)) || 930is_dir_sep(c) || c =='\0') 931return0; 932} 933 c = *path++; 934} 935} 936 937/* 938 * Do we have another file that has the beginning components being a 939 * proper superset of the name we're trying to add? 940 */ 941static inthas_file_name(struct index_state *istate, 942const struct cache_entry *ce,int pos,int ok_to_replace) 943{ 944int retval =0; 945int len =ce_namelen(ce); 946int stage =ce_stage(ce); 947const char*name = ce->name; 948 949while(pos < istate->cache_nr) { 950struct cache_entry *p = istate->cache[pos++]; 951 952if(len >=ce_namelen(p)) 953break; 954if(memcmp(name, p->name, len)) 955break; 956if(ce_stage(p) != stage) 957continue; 958if(p->name[len] !='/') 959continue; 960if(p->ce_flags & CE_REMOVE) 961continue; 962 retval = -1; 963if(!ok_to_replace) 964break; 965remove_index_entry_at(istate, --pos); 966} 967return retval; 968} 969 970 971/* 972 * Like strcmp(), but also return the offset of the first change. 973 * If strings are equal, return the length. 974 */ 975intstrcmp_offset(const char*s1,const char*s2,size_t*first_change) 976{ 977size_t k; 978 979if(!first_change) 980returnstrcmp(s1, s2); 981 982for(k =0; s1[k] == s2[k]; k++) 983if(s1[k] =='\0') 984break; 985 986*first_change = k; 987return(unsigned char)s1[k] - (unsigned char)s2[k]; 988} 989 990/* 991 * Do we have another file with a pathname that is a proper 992 * subset of the name we're trying to add? 993 * 994 * That is, is there another file in the index with a path 995 * that matches a sub-directory in the given entry? 996 */ 997static inthas_dir_name(struct index_state *istate, 998const struct cache_entry *ce,int pos,int ok_to_replace) 999{1000int retval =0;1001int stage =ce_stage(ce);1002const char*name = ce->name;1003const char*slash = name +ce_namelen(ce);1004size_t len_eq_last;1005int cmp_last =0;10061007/*1008 * We are frequently called during an iteration on a sorted1009 * list of pathnames and while building a new index. Therefore,1010 * there is a high probability that this entry will eventually1011 * be appended to the index, rather than inserted in the middle.1012 * If we can confirm that, we can avoid binary searches on the1013 * components of the pathname.1014 *1015 * Compare the entry's full path with the last path in the index.1016 */1017if(istate->cache_nr >0) {1018 cmp_last =strcmp_offset(name,1019 istate->cache[istate->cache_nr -1]->name,1020&len_eq_last);1021if(cmp_last >0) {1022if(len_eq_last ==0) {1023/*1024 * The entry sorts AFTER the last one in the1025 * index and their paths have no common prefix,1026 * so there cannot be a F/D conflict.1027 */1028return retval;1029}else{1030/*1031 * The entry sorts AFTER the last one in the1032 * index, but has a common prefix. Fall through1033 * to the loop below to disect the entry's path1034 * and see where the difference is.1035 */1036}1037}else if(cmp_last ==0) {1038/*1039 * The entry exactly matches the last one in the1040 * index, but because of multiple stage and CE_REMOVE1041 * items, we fall through and let the regular search1042 * code handle it.1043 */1044}1045}10461047for(;;) {1048size_t len;10491050for(;;) {1051if(*--slash =='/')1052break;1053if(slash <= ce->name)1054return retval;1055}1056 len = slash - name;10571058if(cmp_last >0) {1059/*1060 * (len + 1) is a directory boundary (including1061 * the trailing slash). And since the loop is1062 * decrementing "slash", the first iteration is1063 * the longest directory prefix; subsequent1064 * iterations consider parent directories.1065 */10661067if(len +1<= len_eq_last) {1068/*1069 * The directory prefix (including the trailing1070 * slash) also appears as a prefix in the last1071 * entry, so the remainder cannot collide (because1072 * strcmp said the whole path was greater).1073 *1074 * EQ: last: xxx/A1075 * this: xxx/B1076 *1077 * LT: last: xxx/file_A1078 * this: xxx/file_B1079 */1080return retval;1081}10821083if(len > len_eq_last) {1084/*1085 * This part of the directory prefix (excluding1086 * the trailing slash) is longer than the known1087 * equal portions, so this sub-directory cannot1088 * collide with a file.1089 *1090 * GT: last: xxxA1091 * this: xxxB/file1092 */1093return retval;1094}10951096if(istate->cache_nr >0&&1097ce_namelen(istate->cache[istate->cache_nr -1]) > len) {1098/*1099 * The directory prefix lines up with part of1100 * a longer file or directory name, but sorts1101 * after it, so this sub-directory cannot1102 * collide with a file.1103 *1104 * last: xxx/yy-file (because '-' sorts before '/')1105 * this: xxx/yy/abc1106 */1107return retval;1108}11091110/*1111 * This is a possible collision. Fall through and1112 * let the regular search code handle it.1113 *1114 * last: xxx1115 * this: xxx/file1116 */1117}11181119 pos =index_name_stage_pos(istate, name, len, stage);1120if(pos >=0) {1121/*1122 * Found one, but not so fast. This could1123 * be a marker that says "I was here, but1124 * I am being removed". Such an entry is1125 * not a part of the resulting tree, and1126 * it is Ok to have a directory at the same1127 * path.1128 */1129if(!(istate->cache[pos]->ce_flags & CE_REMOVE)) {1130 retval = -1;1131if(!ok_to_replace)1132break;1133remove_index_entry_at(istate, pos);1134continue;1135}1136}1137else1138 pos = -pos-1;11391140/*1141 * Trivial optimization: if we find an entry that1142 * already matches the sub-directory, then we know1143 * we're ok, and we can exit.1144 */1145while(pos < istate->cache_nr) {1146struct cache_entry *p = istate->cache[pos];1147if((ce_namelen(p) <= len) ||1148(p->name[len] !='/') ||1149memcmp(p->name, name, len))1150break;/* not our subdirectory */1151if(ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))1152/*1153 * p is at the same stage as our entry, and1154 * is a subdirectory of what we are looking1155 * at, so we cannot have conflicts at our1156 * level or anything shorter.1157 */1158return retval;1159 pos++;1160}1161}1162return retval;1163}11641165/* We may be in a situation where we already have path/file and path1166 * is being added, or we already have path and path/file is being1167 * added. Either one would result in a nonsense tree that has path1168 * twice when git-write-tree tries to write it out. Prevent it.1169 *1170 * If ok-to-replace is specified, we remove the conflicting entries1171 * from the cache so the caller should recompute the insert position.1172 * When this happens, we return non-zero.1173 */1174static intcheck_file_directory_conflict(struct index_state *istate,1175const struct cache_entry *ce,1176int pos,int ok_to_replace)1177{1178int retval;11791180/*1181 * When ce is an "I am going away" entry, we allow it to be added1182 */1183if(ce->ce_flags & CE_REMOVE)1184return0;11851186/*1187 * We check if the path is a sub-path of a subsequent pathname1188 * first, since removing those will not change the position1189 * in the array.1190 */1191 retval =has_file_name(istate, ce, pos, ok_to_replace);11921193/*1194 * Then check if the path might have a clashing sub-directory1195 * before it.1196 */1197return retval +has_dir_name(istate, ce, pos, ok_to_replace);1198}11991200static intadd_index_entry_with_check(struct index_state *istate,struct cache_entry *ce,int option)1201{1202int pos;1203int ok_to_add = option & ADD_CACHE_OK_TO_ADD;1204int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;1205int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;1206int new_only = option & ADD_CACHE_NEW_ONLY;12071208if(!(option & ADD_CACHE_KEEP_CACHE_TREE))1209cache_tree_invalidate_path(istate, ce->name);12101211/*1212 * If this entry's path sorts after the last entry in the index,1213 * we can avoid searching for it.1214 */1215if(istate->cache_nr >0&&1216strcmp(ce->name, istate->cache[istate->cache_nr -1]->name) >0)1217 pos = -istate->cache_nr -1;1218else1219 pos =index_name_stage_pos(istate, ce->name,ce_namelen(ce),ce_stage(ce));12201221/* existing match? Just replace it. */1222if(pos >=0) {1223if(!new_only)1224replace_index_entry(istate, pos, ce);1225return0;1226}1227 pos = -pos-1;12281229if(!(option & ADD_CACHE_KEEP_CACHE_TREE))1230untracked_cache_add_to_index(istate, ce->name);12311232/*1233 * Inserting a merged entry ("stage 0") into the index1234 * will always replace all non-merged entries..1235 */1236if(pos < istate->cache_nr &&ce_stage(ce) ==0) {1237while(ce_same_name(istate->cache[pos], ce)) {1238 ok_to_add =1;1239if(!remove_index_entry_at(istate, pos))1240break;1241}1242}12431244if(!ok_to_add)1245return-1;1246if(!verify_path(ce->name, ce->ce_mode))1247returnerror("Invalid path '%s'", ce->name);12481249if(!skip_df_check &&1250check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {1251if(!ok_to_replace)1252returnerror("'%s' appears as both a file and as a directory",1253 ce->name);1254 pos =index_name_stage_pos(istate, ce->name,ce_namelen(ce),ce_stage(ce));1255 pos = -pos-1;1256}1257return pos +1;1258}12591260intadd_index_entry(struct index_state *istate,struct cache_entry *ce,int option)1261{1262int pos;12631264if(option & ADD_CACHE_JUST_APPEND)1265 pos = istate->cache_nr;1266else{1267int ret;1268 ret =add_index_entry_with_check(istate, ce, option);1269if(ret <=0)1270return ret;1271 pos = ret -1;1272}12731274/* Make sure the array is big enough .. */1275ALLOC_GROW(istate->cache, istate->cache_nr +1, istate->cache_alloc);12761277/* Add it in.. */1278 istate->cache_nr++;1279if(istate->cache_nr > pos +1)1280MOVE_ARRAY(istate->cache + pos +1, istate->cache + pos,1281 istate->cache_nr - pos -1);1282set_index_entry(istate, pos, ce);1283 istate->cache_changed |= CE_ENTRY_ADDED;1284return0;1285}12861287/*1288 * "refresh" does not calculate a new sha1 file or bring the1289 * cache up-to-date for mode/content changes. But what it1290 * _does_ do is to "re-match" the stat information of a file1291 * with the cache, so that you can refresh the cache for a1292 * file that hasn't been changed but where the stat entry is1293 * out of date.1294 *1295 * For example, you'd want to do this after doing a "git-read-tree",1296 * to link up the stat cache details with the proper files.1297 */1298static struct cache_entry *refresh_cache_ent(struct index_state *istate,1299struct cache_entry *ce,1300unsigned int options,int*err,1301int*changed_ret)1302{1303struct stat st;1304struct cache_entry *updated;1305int changed;1306int refresh = options & CE_MATCH_REFRESH;1307int ignore_valid = options & CE_MATCH_IGNORE_VALID;1308int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;1309int ignore_missing = options & CE_MATCH_IGNORE_MISSING;1310int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;13111312if(!refresh ||ce_uptodate(ce))1313return ce;13141315if(!ignore_fsmonitor)1316refresh_fsmonitor(istate);1317/*1318 * CE_VALID or CE_SKIP_WORKTREE means the user promised us1319 * that the change to the work tree does not matter and told1320 * us not to worry.1321 */1322if(!ignore_skip_worktree &&ce_skip_worktree(ce)) {1323ce_mark_uptodate(ce);1324return ce;1325}1326if(!ignore_valid && (ce->ce_flags & CE_VALID)) {1327ce_mark_uptodate(ce);1328return ce;1329}1330if(!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {1331ce_mark_uptodate(ce);1332return ce;1333}13341335if(has_symlink_leading_path(ce->name,ce_namelen(ce))) {1336if(ignore_missing)1337return ce;1338if(err)1339*err = ENOENT;1340return NULL;1341}13421343if(lstat(ce->name, &st) <0) {1344if(ignore_missing && errno == ENOENT)1345return ce;1346if(err)1347*err = errno;1348return NULL;1349}13501351 changed =ie_match_stat(istate, ce, &st, options);1352if(changed_ret)1353*changed_ret = changed;1354if(!changed) {1355/*1356 * The path is unchanged. If we were told to ignore1357 * valid bit, then we did the actual stat check and1358 * found that the entry is unmodified. If the entry1359 * is not marked VALID, this is the place to mark it1360 * valid again, under "assume unchanged" mode.1361 */1362if(ignore_valid && assume_unchanged &&1363!(ce->ce_flags & CE_VALID))1364;/* mark this one VALID again */1365else{1366/*1367 * We do not mark the index itself "modified"1368 * because CE_UPTODATE flag is in-core only;1369 * we are not going to write this change out.1370 */1371if(!S_ISGITLINK(ce->ce_mode)) {1372ce_mark_uptodate(ce);1373mark_fsmonitor_valid(ce);1374}1375return ce;1376}1377}13781379if(ie_modified(istate, ce, &st, options)) {1380if(err)1381*err = EINVAL;1382return NULL;1383}13841385 updated =make_empty_cache_entry(istate,ce_namelen(ce));1386copy_cache_entry(updated, ce);1387memcpy(updated->name, ce->name, ce->ce_namelen +1);1388fill_stat_cache_info(updated, &st);1389/*1390 * If ignore_valid is not set, we should leave CE_VALID bit1391 * alone. Otherwise, paths marked with --no-assume-unchanged1392 * (i.e. things to be edited) will reacquire CE_VALID bit1393 * automatically, which is not really what we want.1394 */1395if(!ignore_valid && assume_unchanged &&1396!(ce->ce_flags & CE_VALID))1397 updated->ce_flags &= ~CE_VALID;13981399/* istate->cache_changed is updated in the caller */1400return updated;1401}14021403static voidshow_file(const char* fmt,const char* name,int in_porcelain,1404int* first,const char*header_msg)1405{1406if(in_porcelain && *first && header_msg) {1407printf("%s\n", header_msg);1408*first =0;1409}1410printf(fmt, name);1411}14121413intrefresh_index(struct index_state *istate,unsigned int flags,1414const struct pathspec *pathspec,1415char*seen,const char*header_msg)1416{1417int i;1418int has_errors =0;1419int really = (flags & REFRESH_REALLY) !=0;1420int allow_unmerged = (flags & REFRESH_UNMERGED) !=0;1421int quiet = (flags & REFRESH_QUIET) !=0;1422int not_new = (flags & REFRESH_IGNORE_MISSING) !=0;1423int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) !=0;1424int first =1;1425int in_porcelain = (flags & REFRESH_IN_PORCELAIN);1426unsigned int options = (CE_MATCH_REFRESH |1427(really ? CE_MATCH_IGNORE_VALID :0) |1428(not_new ? CE_MATCH_IGNORE_MISSING :0));1429const char*modified_fmt;1430const char*deleted_fmt;1431const char*typechange_fmt;1432const char*added_fmt;1433const char*unmerged_fmt;1434uint64_t start =getnanotime();14351436 modified_fmt = (in_porcelain ?"M\t%s\n":"%s: needs update\n");1437 deleted_fmt = (in_porcelain ?"D\t%s\n":"%s: needs update\n");1438 typechange_fmt = (in_porcelain ?"T\t%s\n":"%sneeds update\n");1439 added_fmt = (in_porcelain ?"A\t%s\n":"%sneeds update\n");1440 unmerged_fmt = (in_porcelain ?"U\t%s\n":"%s: needs merge\n");1441for(i =0; i < istate->cache_nr; i++) {1442struct cache_entry *ce, *new_entry;1443int cache_errno =0;1444int changed =0;1445int filtered =0;14461447 ce = istate->cache[i];1448if(ignore_submodules &&S_ISGITLINK(ce->ce_mode))1449continue;14501451if(pathspec && !ce_path_match(ce, pathspec, seen))1452 filtered =1;14531454if(ce_stage(ce)) {1455while((i < istate->cache_nr) &&1456!strcmp(istate->cache[i]->name, ce->name))1457 i++;1458 i--;1459if(allow_unmerged)1460continue;1461if(!filtered)1462show_file(unmerged_fmt, ce->name, in_porcelain,1463&first, header_msg);1464 has_errors =1;1465continue;1466}14671468if(filtered)1469continue;14701471 new_entry =refresh_cache_ent(istate, ce, options, &cache_errno, &changed);1472if(new_entry == ce)1473continue;1474if(!new_entry) {1475const char*fmt;14761477if(really && cache_errno == EINVAL) {1478/* If we are doing --really-refresh that1479 * means the index is not valid anymore.1480 */1481 ce->ce_flags &= ~CE_VALID;1482 ce->ce_flags |= CE_UPDATE_IN_BASE;1483mark_fsmonitor_invalid(istate, ce);1484 istate->cache_changed |= CE_ENTRY_CHANGED;1485}1486if(quiet)1487continue;14881489if(cache_errno == ENOENT)1490 fmt = deleted_fmt;1491else if(ce_intent_to_add(ce))1492 fmt = added_fmt;/* must be before other checks */1493else if(changed & TYPE_CHANGED)1494 fmt = typechange_fmt;1495else1496 fmt = modified_fmt;1497show_file(fmt,1498 ce->name, in_porcelain, &first, header_msg);1499 has_errors =1;1500continue;1501}15021503replace_index_entry(istate, i, new_entry);1504}1505trace_performance_since(start,"refresh index");1506return has_errors;1507}15081509struct cache_entry *refresh_cache_entry(struct index_state *istate,1510struct cache_entry *ce,1511unsigned int options)1512{1513returnrefresh_cache_ent(istate, ce, options, NULL, NULL);1514}151515161517/*****************************************************************1518 * Index File I/O1519 *****************************************************************/15201521#define INDEX_FORMAT_DEFAULT 315221523static unsigned intget_index_format_default(void)1524{1525char*envversion =getenv("GIT_INDEX_VERSION");1526char*endp;1527int value;1528unsigned int version = INDEX_FORMAT_DEFAULT;15291530if(!envversion) {1531if(!git_config_get_int("index.version", &value))1532 version = value;1533if(version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {1534warning(_("index.version set, but the value is invalid.\n"1535"Using version%i"), INDEX_FORMAT_DEFAULT);1536return INDEX_FORMAT_DEFAULT;1537}1538return version;1539}15401541 version =strtoul(envversion, &endp,10);1542if(*endp ||1543 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {1544warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"1545"Using version%i"), INDEX_FORMAT_DEFAULT);1546 version = INDEX_FORMAT_DEFAULT;1547}1548return version;1549}15501551/*1552 * dev/ino/uid/gid/size are also just tracked to the low 32 bits1553 * Again - this is just a (very strong in practice) heuristic that1554 * the inode hasn't changed.1555 *1556 * We save the fields in big-endian order to allow using the1557 * index file over NFS transparently.1558 */1559struct ondisk_cache_entry {1560struct cache_time ctime;1561struct cache_time mtime;1562uint32_t dev;1563uint32_t ino;1564uint32_t mode;1565uint32_t uid;1566uint32_t gid;1567uint32_t size;1568unsigned char sha1[20];1569uint16_t flags;1570char name[FLEX_ARRAY];/* more */1571};15721573/*1574 * This struct is used when CE_EXTENDED bit is 11575 * The struct must match ondisk_cache_entry exactly from1576 * ctime till flags1577 */1578struct ondisk_cache_entry_extended {1579struct cache_time ctime;1580struct cache_time mtime;1581uint32_t dev;1582uint32_t ino;1583uint32_t mode;1584uint32_t uid;1585uint32_t gid;1586uint32_t size;1587unsigned char sha1[20];1588uint16_t flags;1589uint16_t flags2;1590char name[FLEX_ARRAY];/* more */1591};15921593/* These are only used for v3 or lower */1594#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)1595#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)1596#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)1597#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)1598#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \1599 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \1600 ondisk_cache_entry_size(ce_namelen(ce)))16011602/* Allow fsck to force verification of the index checksum. */1603int verify_index_checksum;16041605/* Allow fsck to force verification of the cache entry order. */1606int verify_ce_order;16071608static intverify_hdr(struct cache_header *hdr,unsigned long size)1609{1610 git_hash_ctx c;1611unsigned char hash[GIT_MAX_RAWSZ];1612int hdr_version;16131614if(hdr->hdr_signature !=htonl(CACHE_SIGNATURE))1615returnerror("bad signature");1616 hdr_version =ntohl(hdr->hdr_version);1617if(hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)1618returnerror("bad index version%d", hdr_version);16191620if(!verify_index_checksum)1621return0;16221623 the_hash_algo->init_fn(&c);1624 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);1625 the_hash_algo->final_fn(hash, &c);1626if(hashcmp(hash, (unsigned char*)hdr + size - the_hash_algo->rawsz))1627returnerror("bad index file sha1 signature");1628return0;1629}16301631static intread_index_extension(struct index_state *istate,1632const char*ext,void*data,unsigned long sz)1633{1634switch(CACHE_EXT(ext)) {1635case CACHE_EXT_TREE:1636 istate->cache_tree =cache_tree_read(data, sz);1637break;1638case CACHE_EXT_RESOLVE_UNDO:1639 istate->resolve_undo =resolve_undo_read(data, sz);1640break;1641case CACHE_EXT_LINK:1642if(read_link_extension(istate, data, sz))1643return-1;1644break;1645case CACHE_EXT_UNTRACKED:1646 istate->untracked =read_untracked_extension(data, sz);1647break;1648case CACHE_EXT_FSMONITOR:1649read_fsmonitor_extension(istate, data, sz);1650break;1651default:1652if(*ext <'A'||'Z'< *ext)1653returnerror("index uses %.4s extension, which we do not understand",1654 ext);1655fprintf(stderr,"ignoring %.4s extension\n", ext);1656break;1657}1658return0;1659}16601661inthold_locked_index(struct lock_file *lk,int lock_flags)1662{1663returnhold_lock_file_for_update(lk,get_index_file(), lock_flags);1664}16651666intread_index(struct index_state *istate)1667{1668returnread_index_from(istate,get_index_file(),get_git_dir());1669}16701671static struct cache_entry *cache_entry_from_ondisk(struct index_state *istate,1672struct ondisk_cache_entry *ondisk,1673unsigned int flags,1674const char*name,1675size_t len)1676{1677struct cache_entry *ce =make_empty_cache_entry(istate, len);16781679 ce->ce_stat_data.sd_ctime.sec =get_be32(&ondisk->ctime.sec);1680 ce->ce_stat_data.sd_mtime.sec =get_be32(&ondisk->mtime.sec);1681 ce->ce_stat_data.sd_ctime.nsec =get_be32(&ondisk->ctime.nsec);1682 ce->ce_stat_data.sd_mtime.nsec =get_be32(&ondisk->mtime.nsec);1683 ce->ce_stat_data.sd_dev =get_be32(&ondisk->dev);1684 ce->ce_stat_data.sd_ino =get_be32(&ondisk->ino);1685 ce->ce_mode =get_be32(&ondisk->mode);1686 ce->ce_stat_data.sd_uid =get_be32(&ondisk->uid);1687 ce->ce_stat_data.sd_gid =get_be32(&ondisk->gid);1688 ce->ce_stat_data.sd_size =get_be32(&ondisk->size);1689 ce->ce_flags = flags & ~CE_NAMEMASK;1690 ce->ce_namelen = len;1691 ce->index =0;1692hashcpy(ce->oid.hash, ondisk->sha1);1693memcpy(ce->name, name, len);1694 ce->name[len] ='\0';1695return ce;1696}16971698/*1699 * Adjacent cache entries tend to share the leading paths, so it makes1700 * sense to only store the differences in later entries. In the v41701 * on-disk format of the index, each on-disk cache entry stores the1702 * number of bytes to be stripped from the end of the previous name,1703 * and the bytes to append to the result, to come up with its name.1704 */1705static unsigned longexpand_name_field(struct strbuf *name,const char*cp_)1706{1707const unsigned char*ep, *cp = (const unsigned char*)cp_;1708size_t len =decode_varint(&cp);17091710if(name->len < len)1711die("malformed name field in the index");1712strbuf_remove(name, name->len - len, len);1713for(ep = cp; *ep; ep++)1714;/* find the end */1715strbuf_add(name, cp, ep - cp);1716return(const char*)ep +1- cp_;1717}17181719static struct cache_entry *create_from_disk(struct index_state *istate,1720struct ondisk_cache_entry *ondisk,1721unsigned long*ent_size,1722struct strbuf *previous_name)1723{1724struct cache_entry *ce;1725size_t len;1726const char*name;1727unsigned int flags;17281729/* On-disk flags are just 16 bits */1730 flags =get_be16(&ondisk->flags);1731 len = flags & CE_NAMEMASK;17321733if(flags & CE_EXTENDED) {1734struct ondisk_cache_entry_extended *ondisk2;1735int extended_flags;1736 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1737 extended_flags =get_be16(&ondisk2->flags2) <<16;1738/* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1739if(extended_flags & ~CE_EXTENDED_FLAGS)1740die("Unknown index entry format%08x", extended_flags);1741 flags |= extended_flags;1742 name = ondisk2->name;1743}1744else1745 name = ondisk->name;17461747if(!previous_name) {1748/* v3 and earlier */1749if(len == CE_NAMEMASK)1750 len =strlen(name);1751 ce =cache_entry_from_ondisk(istate, ondisk, flags, name, len);17521753*ent_size =ondisk_ce_size(ce);1754}else{1755unsigned long consumed;1756 consumed =expand_name_field(previous_name, name);1757 ce =cache_entry_from_ondisk(istate, ondisk, flags,1758 previous_name->buf,1759 previous_name->len);17601761*ent_size = (name - ((char*)ondisk)) + consumed;1762}1763return ce;1764}17651766static voidcheck_ce_order(struct index_state *istate)1767{1768unsigned int i;17691770if(!verify_ce_order)1771return;17721773for(i =1; i < istate->cache_nr; i++) {1774struct cache_entry *ce = istate->cache[i -1];1775struct cache_entry *next_ce = istate->cache[i];1776int name_compare =strcmp(ce->name, next_ce->name);17771778if(0< name_compare)1779die("unordered stage entries in index");1780if(!name_compare) {1781if(!ce_stage(ce))1782die("multiple stage entries for merged file '%s'",1783 ce->name);1784if(ce_stage(ce) >ce_stage(next_ce))1785die("unordered stage entries for '%s'",1786 ce->name);1787}1788}1789}17901791static voidtweak_untracked_cache(struct index_state *istate)1792{1793switch(git_config_get_untracked_cache()) {1794case-1:/* keep: do nothing */1795break;1796case0:/* false */1797remove_untracked_cache(istate);1798break;1799case1:/* true */1800add_untracked_cache(istate);1801break;1802default:/* unknown value: do nothing */1803break;1804}1805}18061807static voidtweak_split_index(struct index_state *istate)1808{1809switch(git_config_get_split_index()) {1810case-1:/* unset: do nothing */1811break;1812case0:/* false */1813remove_split_index(istate);1814break;1815case1:/* true */1816add_split_index(istate);1817break;1818default:/* unknown value: do nothing */1819break;1820}1821}18221823static voidpost_read_index_from(struct index_state *istate)1824{1825check_ce_order(istate);1826tweak_untracked_cache(istate);1827tweak_split_index(istate);1828tweak_fsmonitor(istate);1829}18301831/* remember to discard_cache() before reading a different cache! */1832intdo_read_index(struct index_state *istate,const char*path,int must_exist)1833{1834int fd, i;1835struct stat st;1836unsigned long src_offset;1837struct cache_header *hdr;1838void*mmap;1839size_t mmap_size;1840struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;18411842if(istate->initialized)1843return istate->cache_nr;18441845 istate->timestamp.sec =0;1846 istate->timestamp.nsec =0;1847 fd =open(path, O_RDONLY);1848if(fd <0) {1849if(!must_exist && errno == ENOENT)1850return0;1851die_errno("%s: index file open failed", path);1852}18531854if(fstat(fd, &st))1855die_errno("cannot stat the open index");18561857 mmap_size =xsize_t(st.st_size);1858if(mmap_size <sizeof(struct cache_header) + the_hash_algo->rawsz)1859die("index file smaller than expected");18601861 mmap =xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd,0);1862if(mmap == MAP_FAILED)1863die_errno("unable to map index file");1864close(fd);18651866 hdr = mmap;1867if(verify_hdr(hdr, mmap_size) <0)1868goto unmap;18691870hashcpy(istate->oid.hash, (const unsigned char*)hdr + mmap_size - the_hash_algo->rawsz);1871 istate->version =ntohl(hdr->hdr_version);1872 istate->cache_nr =ntohl(hdr->hdr_entries);1873 istate->cache_alloc =alloc_nr(istate->cache_nr);1874 istate->cache =xcalloc(istate->cache_alloc,sizeof(*istate->cache));1875 istate->initialized =1;18761877if(istate->version ==4)1878 previous_name = &previous_name_buf;1879else1880 previous_name = NULL;18811882 src_offset =sizeof(*hdr);1883for(i =0; i < istate->cache_nr; i++) {1884struct ondisk_cache_entry *disk_ce;1885struct cache_entry *ce;1886unsigned long consumed;18871888 disk_ce = (struct ondisk_cache_entry *)((char*)mmap + src_offset);1889 ce =create_from_disk(istate, disk_ce, &consumed, previous_name);1890set_index_entry(istate, i, ce);18911892 src_offset += consumed;1893}1894strbuf_release(&previous_name_buf);1895 istate->timestamp.sec = st.st_mtime;1896 istate->timestamp.nsec =ST_MTIME_NSEC(st);18971898while(src_offset <= mmap_size - the_hash_algo->rawsz -8) {1899/* After an array of active_nr index entries,1900 * there can be arbitrary number of extended1901 * sections, each of which is prefixed with1902 * extension name (4-byte) and section length1903 * in 4-byte network byte order.1904 */1905uint32_t extsize;1906memcpy(&extsize, (char*)mmap + src_offset +4,4);1907 extsize =ntohl(extsize);1908if(read_index_extension(istate,1909(const char*) mmap + src_offset,1910(char*) mmap + src_offset +8,1911 extsize) <0)1912goto unmap;1913 src_offset +=8;1914 src_offset += extsize;1915}1916munmap(mmap, mmap_size);1917return istate->cache_nr;19181919unmap:1920munmap(mmap, mmap_size);1921die("index file corrupt");1922}19231924/*1925 * Signal that the shared index is used by updating its mtime.1926 *1927 * This way, shared index can be removed if they have not been used1928 * for some time.1929 */1930static voidfreshen_shared_index(const char*shared_index,int warn)1931{1932if(!check_and_freshen_file(shared_index,1) && warn)1933warning("could not freshen shared index '%s'", shared_index);1934}19351936intread_index_from(struct index_state *istate,const char*path,1937const char*gitdir)1938{1939uint64_t start =getnanotime();1940struct split_index *split_index;1941int ret;1942char*base_oid_hex;1943char*base_path;19441945/* istate->initialized covers both .git/index and .git/sharedindex.xxx */1946if(istate->initialized)1947return istate->cache_nr;19481949 ret =do_read_index(istate, path,0);1950trace_performance_since(start,"read cache%s", path);19511952 split_index = istate->split_index;1953if(!split_index ||is_null_oid(&split_index->base_oid)) {1954post_read_index_from(istate);1955return ret;1956}19571958if(split_index->base)1959discard_index(split_index->base);1960else1961 split_index->base =xcalloc(1,sizeof(*split_index->base));19621963 base_oid_hex =oid_to_hex(&split_index->base_oid);1964 base_path =xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);1965 ret =do_read_index(split_index->base, base_path,1);1966if(oidcmp(&split_index->base_oid, &split_index->base->oid))1967die("broken index, expect%sin%s, got%s",1968 base_oid_hex, base_path,1969oid_to_hex(&split_index->base->oid));19701971freshen_shared_index(base_path,0);1972merge_base_index(istate);1973post_read_index_from(istate);1974trace_performance_since(start,"read cache%s", base_path);1975free(base_path);1976return ret;1977}19781979intis_index_unborn(struct index_state *istate)1980{1981return(!istate->cache_nr && !istate->timestamp.sec);1982}19831984intdiscard_index(struct index_state *istate)1985{1986int i;19871988for(i =0; i < istate->cache_nr; i++) {1989if(istate->cache[i]->index &&1990 istate->split_index &&1991 istate->split_index->base &&1992 istate->cache[i]->index <= istate->split_index->base->cache_nr &&1993 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index -1])1994continue;1995discard_cache_entry(istate->cache[i]);1996}1997resolve_undo_clear_index(istate);1998 istate->cache_nr =0;1999 istate->cache_changed =0;2000 istate->timestamp.sec =0;2001 istate->timestamp.nsec =0;2002free_name_hash(istate);2003cache_tree_free(&(istate->cache_tree));2004 istate->initialized =0;2005FREE_AND_NULL(istate->cache);2006 istate->cache_alloc =0;2007discard_split_index(istate);2008free_untracked_cache(istate->untracked);2009 istate->untracked = NULL;2010return0;2011}20122013intunmerged_index(const struct index_state *istate)2014{2015int i;2016for(i =0; i < istate->cache_nr; i++) {2017if(ce_stage(istate->cache[i]))2018return1;2019}2020return0;2021}20222023#define WRITE_BUFFER_SIZE 81922024static unsigned char write_buffer[WRITE_BUFFER_SIZE];2025static unsigned long write_buffer_len;20262027static intce_write_flush(git_hash_ctx *context,int fd)2028{2029unsigned int buffered = write_buffer_len;2030if(buffered) {2031 the_hash_algo->update_fn(context, write_buffer, buffered);2032if(write_in_full(fd, write_buffer, buffered) <0)2033return-1;2034 write_buffer_len =0;2035}2036return0;2037}20382039static intce_write(git_hash_ctx *context,int fd,void*data,unsigned int len)2040{2041while(len) {2042unsigned int buffered = write_buffer_len;2043unsigned int partial = WRITE_BUFFER_SIZE - buffered;2044if(partial > len)2045 partial = len;2046memcpy(write_buffer + buffered, data, partial);2047 buffered += partial;2048if(buffered == WRITE_BUFFER_SIZE) {2049 write_buffer_len = buffered;2050if(ce_write_flush(context, fd))2051return-1;2052 buffered =0;2053}2054 write_buffer_len = buffered;2055 len -= partial;2056 data = (char*) data + partial;2057}2058return0;2059}20602061static intwrite_index_ext_header(git_hash_ctx *context,int fd,2062unsigned int ext,unsigned int sz)2063{2064 ext =htonl(ext);2065 sz =htonl(sz);2066return((ce_write(context, fd, &ext,4) <0) ||2067(ce_write(context, fd, &sz,4) <0)) ? -1:0;2068}20692070static intce_flush(git_hash_ctx *context,int fd,unsigned char*hash)2071{2072unsigned int left = write_buffer_len;20732074if(left) {2075 write_buffer_len =0;2076 the_hash_algo->update_fn(context, write_buffer, left);2077}20782079/* Flush first if not enough space for hash signature */2080if(left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {2081if(write_in_full(fd, write_buffer, left) <0)2082return-1;2083 left =0;2084}20852086/* Append the hash signature at the end */2087 the_hash_algo->final_fn(write_buffer + left, context);2088hashcpy(hash, write_buffer + left);2089 left += the_hash_algo->rawsz;2090return(write_in_full(fd, write_buffer, left) <0) ? -1:0;2091}20922093static voidce_smudge_racily_clean_entry(struct cache_entry *ce)2094{2095/*2096 * The only thing we care about in this function is to smudge the2097 * falsely clean entry due to touch-update-touch race, so we leave2098 * everything else as they are. We are called for entries whose2099 * ce_stat_data.sd_mtime match the index file mtime.2100 *2101 * Note that this actually does not do much for gitlinks, for2102 * which ce_match_stat_basic() always goes to the actual2103 * contents. The caller checks with is_racy_timestamp() which2104 * always says "no" for gitlinks, so we are not called for them ;-)2105 */2106struct stat st;21072108if(lstat(ce->name, &st) <0)2109return;2110if(ce_match_stat_basic(ce, &st))2111return;2112if(ce_modified_check_fs(ce, &st)) {2113/* This is "racily clean"; smudge it. Note that this2114 * is a tricky code. At first glance, it may appear2115 * that it can break with this sequence:2116 *2117 * $ echo xyzzy >frotz2118 * $ git-update-index --add frotz2119 * $ : >frotz2120 * $ sleep 32121 * $ echo filfre >nitfol2122 * $ git-update-index --add nitfol2123 *2124 * but it does not. When the second update-index runs,2125 * it notices that the entry "frotz" has the same timestamp2126 * as index, and if we were to smudge it by resetting its2127 * size to zero here, then the object name recorded2128 * in index is the 6-byte file but the cached stat information2129 * becomes zero --- which would then match what we would2130 * obtain from the filesystem next time we stat("frotz").2131 *2132 * However, the second update-index, before calling2133 * this function, notices that the cached size is 62134 * bytes and what is on the filesystem is an empty2135 * file, and never calls us, so the cached size information2136 * for "frotz" stays 6 which does not match the filesystem.2137 */2138 ce->ce_stat_data.sd_size =0;2139}2140}21412142/* Copy miscellaneous fields but not the name */2143static voidcopy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,2144struct cache_entry *ce)2145{2146short flags;21472148 ondisk->ctime.sec =htonl(ce->ce_stat_data.sd_ctime.sec);2149 ondisk->mtime.sec =htonl(ce->ce_stat_data.sd_mtime.sec);2150 ondisk->ctime.nsec =htonl(ce->ce_stat_data.sd_ctime.nsec);2151 ondisk->mtime.nsec =htonl(ce->ce_stat_data.sd_mtime.nsec);2152 ondisk->dev =htonl(ce->ce_stat_data.sd_dev);2153 ondisk->ino =htonl(ce->ce_stat_data.sd_ino);2154 ondisk->mode =htonl(ce->ce_mode);2155 ondisk->uid =htonl(ce->ce_stat_data.sd_uid);2156 ondisk->gid =htonl(ce->ce_stat_data.sd_gid);2157 ondisk->size =htonl(ce->ce_stat_data.sd_size);2158hashcpy(ondisk->sha1, ce->oid.hash);21592160 flags = ce->ce_flags & ~CE_NAMEMASK;2161 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK :ce_namelen(ce));2162 ondisk->flags =htons(flags);2163if(ce->ce_flags & CE_EXTENDED) {2164struct ondisk_cache_entry_extended *ondisk2;2165 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;2166 ondisk2->flags2 =htons((ce->ce_flags & CE_EXTENDED_FLAGS) >>16);2167}2168}21692170static intce_write_entry(git_hash_ctx *c,int fd,struct cache_entry *ce,2171struct strbuf *previous_name,struct ondisk_cache_entry *ondisk)2172{2173int size;2174int result;2175unsigned int saved_namelen;2176int stripped_name =0;2177static unsigned char padding[8] = {0x00};21782179if(ce->ce_flags & CE_STRIP_NAME) {2180 saved_namelen =ce_namelen(ce);2181 ce->ce_namelen =0;2182 stripped_name =1;2183}21842185if(ce->ce_flags & CE_EXTENDED)2186 size =offsetof(struct ondisk_cache_entry_extended, name);2187else2188 size =offsetof(struct ondisk_cache_entry, name);21892190if(!previous_name) {2191int len =ce_namelen(ce);2192copy_cache_entry_to_ondisk(ondisk, ce);2193 result =ce_write(c, fd, ondisk, size);2194if(!result)2195 result =ce_write(c, fd, ce->name, len);2196if(!result)2197 result =ce_write(c, fd, padding,align_padding_size(size, len));2198}else{2199int common, to_remove, prefix_size;2200unsigned char to_remove_vi[16];2201for(common =0;2202(ce->name[common] &&2203 common < previous_name->len &&2204 ce->name[common] == previous_name->buf[common]);2205 common++)2206;/* still matching */2207 to_remove = previous_name->len - common;2208 prefix_size =encode_varint(to_remove, to_remove_vi);22092210copy_cache_entry_to_ondisk(ondisk, ce);2211 result =ce_write(c, fd, ondisk, size);2212if(!result)2213 result =ce_write(c, fd, to_remove_vi, prefix_size);2214if(!result)2215 result =ce_write(c, fd, ce->name + common,ce_namelen(ce) - common);2216if(!result)2217 result =ce_write(c, fd, padding,1);22182219strbuf_splice(previous_name, common, to_remove,2220 ce->name + common,ce_namelen(ce) - common);2221}2222if(stripped_name) {2223 ce->ce_namelen = saved_namelen;2224 ce->ce_flags &= ~CE_STRIP_NAME;2225}22262227return result;2228}22292230/*2231 * This function verifies if index_state has the correct sha1 of the2232 * index file. Don't die if we have any other failure, just return 0.2233 */2234static intverify_index_from(const struct index_state *istate,const char*path)2235{2236int fd;2237 ssize_t n;2238struct stat st;2239unsigned char hash[GIT_MAX_RAWSZ];22402241if(!istate->initialized)2242return0;22432244 fd =open(path, O_RDONLY);2245if(fd <0)2246return0;22472248if(fstat(fd, &st))2249goto out;22502251if(st.st_size <sizeof(struct cache_header) + the_hash_algo->rawsz)2252goto out;22532254 n =pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);2255if(n != the_hash_algo->rawsz)2256goto out;22572258if(hashcmp(istate->oid.hash, hash))2259goto out;22602261close(fd);2262return1;22632264out:2265close(fd);2266return0;2267}22682269static intverify_index(const struct index_state *istate)2270{2271returnverify_index_from(istate,get_index_file());2272}22732274static inthas_racy_timestamp(struct index_state *istate)2275{2276int entries = istate->cache_nr;2277int i;22782279for(i =0; i < entries; i++) {2280struct cache_entry *ce = istate->cache[i];2281if(is_racy_timestamp(istate, ce))2282return1;2283}2284return0;2285}22862287voidupdate_index_if_able(struct index_state *istate,struct lock_file *lockfile)2288{2289if((istate->cache_changed ||has_racy_timestamp(istate)) &&2290verify_index(istate))2291write_locked_index(istate, lockfile, COMMIT_LOCK);2292else2293rollback_lock_file(lockfile);2294}22952296/*2297 * On success, `tempfile` is closed. If it is the temporary file2298 * of a `struct lock_file`, we will therefore effectively perform2299 * a 'close_lock_file_gently()`. Since that is an implementation2300 * detail of lockfiles, callers of `do_write_index()` should not2301 * rely on it.2302 */2303static intdo_write_index(struct index_state *istate,struct tempfile *tempfile,2304int strip_extensions)2305{2306uint64_t start =getnanotime();2307int newfd = tempfile->fd;2308 git_hash_ctx c;2309struct cache_header hdr;2310int i, err =0, removed, extended, hdr_version;2311struct cache_entry **cache = istate->cache;2312int entries = istate->cache_nr;2313struct stat st;2314struct ondisk_cache_entry_extended ondisk;2315struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;2316int drop_cache_tree = istate->drop_cache_tree;23172318for(i = removed = extended =0; i < entries; i++) {2319if(cache[i]->ce_flags & CE_REMOVE)2320 removed++;23212322/* reduce extended entries if possible */2323 cache[i]->ce_flags &= ~CE_EXTENDED;2324if(cache[i]->ce_flags & CE_EXTENDED_FLAGS) {2325 extended++;2326 cache[i]->ce_flags |= CE_EXTENDED;2327}2328}23292330if(!istate->version) {2331 istate->version =get_index_format_default();2332if(git_env_bool("GIT_TEST_SPLIT_INDEX",0))2333init_split_index(istate);2334}23352336/* demote version 3 to version 2 when the latter suffices */2337if(istate->version ==3|| istate->version ==2)2338 istate->version = extended ?3:2;23392340 hdr_version = istate->version;23412342 hdr.hdr_signature =htonl(CACHE_SIGNATURE);2343 hdr.hdr_version =htonl(hdr_version);2344 hdr.hdr_entries =htonl(entries - removed);23452346 the_hash_algo->init_fn(&c);2347if(ce_write(&c, newfd, &hdr,sizeof(hdr)) <0)2348return-1;23492350 previous_name = (hdr_version ==4) ? &previous_name_buf : NULL;23512352for(i =0; i < entries; i++) {2353struct cache_entry *ce = cache[i];2354if(ce->ce_flags & CE_REMOVE)2355continue;2356if(!ce_uptodate(ce) &&is_racy_timestamp(istate, ce))2357ce_smudge_racily_clean_entry(ce);2358if(is_null_oid(&ce->oid)) {2359static const char msg[] ="cache entry has null sha1:%s";2360static int allow = -1;23612362if(allow <0)2363 allow =git_env_bool("GIT_ALLOW_NULL_SHA1",0);2364if(allow)2365warning(msg, ce->name);2366else2367 err =error(msg, ce->name);23682369 drop_cache_tree =1;2370}2371if(ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) <0)2372 err = -1;23732374if(err)2375break;2376}2377strbuf_release(&previous_name_buf);23782379if(err)2380return err;23812382/* Write extension data here */2383if(!strip_extensions && istate->split_index) {2384struct strbuf sb = STRBUF_INIT;23852386 err =write_link_extension(&sb, istate) <0||2387write_index_ext_header(&c, newfd, CACHE_EXT_LINK,2388 sb.len) <0||2389ce_write(&c, newfd, sb.buf, sb.len) <0;2390strbuf_release(&sb);2391if(err)2392return-1;2393}2394if(!strip_extensions && !drop_cache_tree && istate->cache_tree) {2395struct strbuf sb = STRBUF_INIT;23962397cache_tree_write(&sb, istate->cache_tree);2398 err =write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) <02399||ce_write(&c, newfd, sb.buf, sb.len) <0;2400strbuf_release(&sb);2401if(err)2402return-1;2403}2404if(!strip_extensions && istate->resolve_undo) {2405struct strbuf sb = STRBUF_INIT;24062407resolve_undo_write(&sb, istate->resolve_undo);2408 err =write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,2409 sb.len) <02410||ce_write(&c, newfd, sb.buf, sb.len) <0;2411strbuf_release(&sb);2412if(err)2413return-1;2414}2415if(!strip_extensions && istate->untracked) {2416struct strbuf sb = STRBUF_INIT;24172418write_untracked_extension(&sb, istate->untracked);2419 err =write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,2420 sb.len) <0||2421ce_write(&c, newfd, sb.buf, sb.len) <0;2422strbuf_release(&sb);2423if(err)2424return-1;2425}2426if(!strip_extensions && istate->fsmonitor_last_update) {2427struct strbuf sb = STRBUF_INIT;24282429write_fsmonitor_extension(&sb, istate);2430 err =write_index_ext_header(&c, newfd, CACHE_EXT_FSMONITOR, sb.len) <02431||ce_write(&c, newfd, sb.buf, sb.len) <0;2432strbuf_release(&sb);2433if(err)2434return-1;2435}24362437if(ce_flush(&c, newfd, istate->oid.hash))2438return-1;2439if(close_tempfile_gently(tempfile)) {2440error(_("could not close '%s'"), tempfile->filename.buf);2441return-1;2442}2443if(stat(tempfile->filename.buf, &st))2444return-1;2445 istate->timestamp.sec = (unsigned int)st.st_mtime;2446 istate->timestamp.nsec =ST_MTIME_NSEC(st);2447trace_performance_since(start,"write index, changed mask =%x", istate->cache_changed);2448return0;2449}24502451voidset_alternate_index_output(const char*name)2452{2453 alternate_index_output = name;2454}24552456static intcommit_locked_index(struct lock_file *lk)2457{2458if(alternate_index_output)2459returncommit_lock_file_to(lk, alternate_index_output);2460else2461returncommit_lock_file(lk);2462}24632464static intdo_write_locked_index(struct index_state *istate,struct lock_file *lock,2465unsigned flags)2466{2467int ret =do_write_index(istate, lock->tempfile,0);2468if(ret)2469return ret;2470if(flags & COMMIT_LOCK)2471returncommit_locked_index(lock);2472returnclose_lock_file_gently(lock);2473}24742475static intwrite_split_index(struct index_state *istate,2476struct lock_file *lock,2477unsigned flags)2478{2479int ret;2480prepare_to_write_split_index(istate);2481 ret =do_write_locked_index(istate, lock, flags);2482finish_writing_split_index(istate);2483return ret;2484}24852486static const char*shared_index_expire ="2.weeks.ago";24872488static unsigned longget_shared_index_expire_date(void)2489{2490static unsigned long shared_index_expire_date;2491static int shared_index_expire_date_prepared;24922493if(!shared_index_expire_date_prepared) {2494git_config_get_expiry("splitindex.sharedindexexpire",2495&shared_index_expire);2496 shared_index_expire_date =approxidate(shared_index_expire);2497 shared_index_expire_date_prepared =1;2498}24992500return shared_index_expire_date;2501}25022503static intshould_delete_shared_index(const char*shared_index_path)2504{2505struct stat st;2506unsigned long expiration;25072508/* Check timestamp */2509 expiration =get_shared_index_expire_date();2510if(!expiration)2511return0;2512if(stat(shared_index_path, &st))2513returnerror_errno(_("could not stat '%s'"), shared_index_path);2514if(st.st_mtime > expiration)2515return0;25162517return1;2518}25192520static intclean_shared_index_files(const char*current_hex)2521{2522struct dirent *de;2523DIR*dir =opendir(get_git_dir());25242525if(!dir)2526returnerror_errno(_("unable to open git dir:%s"),get_git_dir());25272528while((de =readdir(dir)) != NULL) {2529const char*sha1_hex;2530const char*shared_index_path;2531if(!skip_prefix(de->d_name,"sharedindex.", &sha1_hex))2532continue;2533if(!strcmp(sha1_hex, current_hex))2534continue;2535 shared_index_path =git_path("%s", de->d_name);2536if(should_delete_shared_index(shared_index_path) >0&&2537unlink(shared_index_path))2538warning_errno(_("unable to unlink:%s"), shared_index_path);2539}2540closedir(dir);25412542return0;2543}25442545static intwrite_shared_index(struct index_state *istate,2546struct tempfile **temp)2547{2548struct split_index *si = istate->split_index;2549int ret;25502551move_cache_to_base_index(istate);2552 ret =do_write_index(si->base, *temp,1);2553if(ret)2554return ret;2555 ret =adjust_shared_perm(get_tempfile_path(*temp));2556if(ret) {2557error("cannot fix permission bits on%s",get_tempfile_path(*temp));2558return ret;2559}2560 ret =rename_tempfile(temp,2561git_path("sharedindex.%s",oid_to_hex(&si->base->oid)));2562if(!ret) {2563oidcpy(&si->base_oid, &si->base->oid);2564clean_shared_index_files(oid_to_hex(&si->base->oid));2565}25662567return ret;2568}25692570static const int default_max_percent_split_change =20;25712572static inttoo_many_not_shared_entries(struct index_state *istate)2573{2574int i, not_shared =0;2575int max_split =git_config_get_max_percent_split_change();25762577switch(max_split) {2578case-1:2579/* not or badly configured: use the default value */2580 max_split = default_max_percent_split_change;2581break;2582case0:2583return1;/* 0% means always write a new shared index */2584case100:2585return0;/* 100% means never write a new shared index */2586default:2587break;/* just use the configured value */2588}25892590/* Count not shared entries */2591for(i =0; i < istate->cache_nr; i++) {2592struct cache_entry *ce = istate->cache[i];2593if(!ce->index)2594 not_shared++;2595}25962597return(int64_t)istate->cache_nr * max_split < (int64_t)not_shared *100;2598}25992600intwrite_locked_index(struct index_state *istate,struct lock_file *lock,2601unsigned flags)2602{2603int new_shared_index, ret;2604struct split_index *si = istate->split_index;26052606if((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {2607if(flags & COMMIT_LOCK)2608rollback_lock_file(lock);2609return0;2610}26112612if(istate->fsmonitor_last_update)2613fill_fsmonitor_bitmap(istate);26142615if(!si || alternate_index_output ||2616(istate->cache_changed & ~EXTMASK)) {2617if(si)2618oidclr(&si->base_oid);2619 ret =do_write_locked_index(istate, lock, flags);2620goto out;2621}26222623if(git_env_bool("GIT_TEST_SPLIT_INDEX",0)) {2624int v = si->base_oid.hash[0];2625if((v &15) <6)2626 istate->cache_changed |= SPLIT_INDEX_ORDERED;2627}2628if(too_many_not_shared_entries(istate))2629 istate->cache_changed |= SPLIT_INDEX_ORDERED;26302631 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;26322633if(new_shared_index) {2634struct tempfile *temp;2635int saved_errno;26362637 temp =mks_tempfile(git_path("sharedindex_XXXXXX"));2638if(!temp) {2639oidclr(&si->base_oid);2640 ret =do_write_locked_index(istate, lock, flags);2641goto out;2642}2643 ret =write_shared_index(istate, &temp);26442645 saved_errno = errno;2646if(is_tempfile_active(temp))2647delete_tempfile(&temp);2648 errno = saved_errno;26492650if(ret)2651goto out;2652}26532654 ret =write_split_index(istate, lock, flags);26552656/* Freshen the shared index only if the split-index was written */2657if(!ret && !new_shared_index) {2658const char*shared_index =git_path("sharedindex.%s",2659oid_to_hex(&si->base_oid));2660freshen_shared_index(shared_index,1);2661}26622663out:2664if(flags & COMMIT_LOCK)2665rollback_lock_file(lock);2666return ret;2667}26682669/*2670 * Read the index file that is potentially unmerged into given2671 * index_state, dropping any unmerged entries. Returns true if2672 * the index is unmerged. Callers who want to refuse to work2673 * from an unmerged state can call this and check its return value,2674 * instead of calling read_cache().2675 */2676intread_index_unmerged(struct index_state *istate)2677{2678int i;2679int unmerged =0;26802681read_index(istate);2682for(i =0; i < istate->cache_nr; i++) {2683struct cache_entry *ce = istate->cache[i];2684struct cache_entry *new_ce;2685int len;26862687if(!ce_stage(ce))2688continue;2689 unmerged =1;2690 len =ce_namelen(ce);2691 new_ce =make_empty_cache_entry(istate, len);2692memcpy(new_ce->name, ce->name, len);2693 new_ce->ce_flags =create_ce_flags(0) | CE_CONFLICTED;2694 new_ce->ce_namelen = len;2695 new_ce->ce_mode = ce->ce_mode;2696if(add_index_entry(istate, new_ce,0))2697returnerror("%s: cannot drop to stage #0",2698 new_ce->name);2699}2700return unmerged;2701}27022703/*2704 * Returns 1 if the path is an "other" path with respect to2705 * the index; that is, the path is not mentioned in the index at all,2706 * either as a file, a directory with some files in the index,2707 * or as an unmerged entry.2708 *2709 * We helpfully remove a trailing "/" from directories so that2710 * the output of read_directory can be used as-is.2711 */2712intindex_name_is_other(const struct index_state *istate,const char*name,2713int namelen)2714{2715int pos;2716if(namelen && name[namelen -1] =='/')2717 namelen--;2718 pos =index_name_pos(istate, name, namelen);2719if(0<= pos)2720return0;/* exact match */2721 pos = -pos -1;2722if(pos < istate->cache_nr) {2723struct cache_entry *ce = istate->cache[pos];2724if(ce_namelen(ce) == namelen &&2725!memcmp(ce->name, name, namelen))2726return0;/* Yup, this one exists unmerged */2727}2728return1;2729}27302731void*read_blob_data_from_index(const struct index_state *istate,2732const char*path,unsigned long*size)2733{2734int pos, len;2735unsigned long sz;2736enum object_type type;2737void*data;27382739 len =strlen(path);2740 pos =index_name_pos(istate, path, len);2741if(pos <0) {2742/*2743 * We might be in the middle of a merge, in which2744 * case we would read stage #2 (ours).2745 */2746int i;2747for(i = -pos -1;2748(pos <0&& i < istate->cache_nr &&2749!strcmp(istate->cache[i]->name, path));2750 i++)2751if(ce_stage(istate->cache[i]) ==2)2752 pos = i;2753}2754if(pos <0)2755return NULL;2756 data =read_object_file(&istate->cache[pos]->oid, &type, &sz);2757if(!data || type != OBJ_BLOB) {2758free(data);2759return NULL;2760}2761if(size)2762*size = sz;2763return data;2764}27652766voidstat_validity_clear(struct stat_validity *sv)2767{2768FREE_AND_NULL(sv->sd);2769}27702771intstat_validity_check(struct stat_validity *sv,const char*path)2772{2773struct stat st;27742775if(stat(path, &st) <0)2776return sv->sd == NULL;2777if(!sv->sd)2778return0;2779returnS_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);2780}27812782voidstat_validity_update(struct stat_validity *sv,int fd)2783{2784struct stat st;27852786if(fstat(fd, &st) <0|| !S_ISREG(st.st_mode))2787stat_validity_clear(sv);2788else{2789if(!sv->sd)2790 sv->sd =xcalloc(1,sizeof(struct stat_data));2791fill_stat_data(sv->sd, &st);2792}2793}27942795voidmove_index_extensions(struct index_state *dst,struct index_state *src)2796{2797 dst->untracked = src->untracked;2798 src->untracked = NULL;2799}28002801voiddiscard_cache_entry(struct cache_entry *ce)2802{2803free(ce);2804}