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); 64free(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 =xmalloc(cache_entry_size(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 =xcalloc(1,cache_entry_size(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 size, 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 size =cache_entry_size(namelen); 666 ce =xcalloc(1, size); 667memcpy(ce->name, path, namelen); 668 ce->ce_namelen = namelen; 669if(!intent_only) 670fill_stat_cache_info(ce, st); 671else 672 ce->ce_flags |= CE_INTENT_TO_ADD; 673 674 675if(trust_executable_bit && has_symlinks) { 676 ce->ce_mode =create_ce_mode(st_mode); 677}else{ 678/* If there is an existing entry, pick the mode bits and type 679 * from it, otherwise assume unexecutable regular file. 680 */ 681struct cache_entry *ent; 682int pos =index_name_pos_also_unmerged(istate, path, namelen); 683 684 ent = (0<= pos) ? istate->cache[pos] : NULL; 685 ce->ce_mode =ce_mode_from_stat(ent, st_mode); 686} 687 688/* When core.ignorecase=true, determine if a directory of the same name but differing 689 * case already exists within the Git repository. If it does, ensure the directory 690 * case of the file being added to the repository matches (is folded into) the existing 691 * entry's directory case. 692 */ 693if(ignore_case) { 694adjust_dirname_case(istate, ce->name); 695} 696if(!(flags & HASH_RENORMALIZE)) { 697 alias =index_file_exists(istate, ce->name, 698ce_namelen(ce), ignore_case); 699if(alias && 700!ce_stage(alias) && 701!ie_match_stat(istate, alias, st, ce_option)) { 702/* Nothing changed, really */ 703if(!S_ISGITLINK(alias->ce_mode)) 704ce_mark_uptodate(alias); 705 alias->ce_flags |= CE_ADDED; 706 707free(ce); 708return0; 709} 710} 711if(!intent_only) { 712if(index_path(&ce->oid, path, st, newflags)) { 713free(ce); 714returnerror("unable to index file%s", path); 715} 716}else 717set_object_name_for_intent_to_add_entry(ce); 718 719if(ignore_case && alias &&different_name(ce, alias)) 720 ce =create_alias_ce(istate, ce, alias); 721 ce->ce_flags |= CE_ADDED; 722 723/* It was suspected to be racily clean, but it turns out to be Ok */ 724 was_same = (alias && 725!ce_stage(alias) && 726!oidcmp(&alias->oid, &ce->oid) && 727 ce->ce_mode == alias->ce_mode); 728 729if(pretend) 730free(ce); 731else if(add_index_entry(istate, ce, add_option)) { 732free(ce); 733returnerror("unable to add%sto index", path); 734} 735if(verbose && !was_same) 736printf("add '%s'\n", path); 737return0; 738} 739 740intadd_file_to_index(struct index_state *istate,const char*path,int flags) 741{ 742struct stat st; 743if(lstat(path, &st)) 744die_errno("unable to stat '%s'", path); 745returnadd_to_index(istate, path, &st, flags); 746} 747 748struct cache_entry *make_cache_entry(unsigned int mode, 749const struct object_id *oid, 750const char*path, 751int stage, 752unsigned int refresh_options) 753{ 754int size, len; 755struct cache_entry *ce, *ret; 756 757if(!verify_path(path, mode)) { 758error("Invalid path '%s'", path); 759return NULL; 760} 761 762 len =strlen(path); 763 size =cache_entry_size(len); 764 ce =xcalloc(1, size); 765 766oidcpy(&ce->oid, oid); 767memcpy(ce->name, path, len); 768 ce->ce_flags =create_ce_flags(stage); 769 ce->ce_namelen = len; 770 ce->ce_mode =create_ce_mode(mode); 771 772 ret =refresh_cache_entry(&the_index, ce, refresh_options); 773if(ret != ce) 774free(ce); 775return ret; 776} 777 778/* 779 * Chmod an index entry with either +x or -x. 780 * 781 * Returns -1 if the chmod for the particular cache entry failed (if it's 782 * not a regular file), -2 if an invalid flip argument is passed in, 0 783 * otherwise. 784 */ 785intchmod_index_entry(struct index_state *istate,struct cache_entry *ce, 786char flip) 787{ 788if(!S_ISREG(ce->ce_mode)) 789return-1; 790switch(flip) { 791case'+': 792 ce->ce_mode |=0111; 793break; 794case'-': 795 ce->ce_mode &= ~0111; 796break; 797default: 798return-2; 799} 800cache_tree_invalidate_path(istate, ce->name); 801 ce->ce_flags |= CE_UPDATE_IN_BASE; 802mark_fsmonitor_invalid(istate, ce); 803 istate->cache_changed |= CE_ENTRY_CHANGED; 804 805return0; 806} 807 808intce_same_name(const struct cache_entry *a,const struct cache_entry *b) 809{ 810int len =ce_namelen(a); 811returnce_namelen(b) == len && !memcmp(a->name, b->name, len); 812} 813 814/* 815 * We fundamentally don't like some paths: we don't want 816 * dot or dot-dot anywhere, and for obvious reasons don't 817 * want to recurse into ".git" either. 818 * 819 * Also, we don't want double slashes or slashes at the 820 * end that can make pathnames ambiguous. 821 */ 822static intverify_dotfile(const char*rest,unsigned mode) 823{ 824/* 825 * The first character was '.', but that 826 * has already been discarded, we now test 827 * the rest. 828 */ 829 830/* "." is not allowed */ 831if(*rest =='\0'||is_dir_sep(*rest)) 832return0; 833 834switch(*rest) { 835/* 836 * ".git" followed by NUL or slash is bad. Note that we match 837 * case-insensitively here, even if ignore_case is not set. 838 * This outlaws ".GIT" everywhere out of an abundance of caution, 839 * since there's really no good reason to allow it. 840 * 841 * Once we've seen ".git", we can also find ".gitmodules", etc (also 842 * case-insensitively). 843 */ 844case'g': 845case'G': 846if(rest[1] !='i'&& rest[1] !='I') 847break; 848if(rest[2] !='t'&& rest[2] !='T') 849break; 850if(rest[3] =='\0'||is_dir_sep(rest[3])) 851return0; 852if(S_ISLNK(mode)) { 853 rest +=3; 854if(skip_iprefix(rest,"modules", &rest) && 855(*rest =='\0'||is_dir_sep(*rest))) 856return0; 857} 858break; 859case'.': 860if(rest[1] =='\0'||is_dir_sep(rest[1])) 861return0; 862} 863return1; 864} 865 866intverify_path(const char*path,unsigned mode) 867{ 868char c; 869 870if(has_dos_drive_prefix(path)) 871return0; 872 873goto inside; 874for(;;) { 875if(!c) 876return1; 877if(is_dir_sep(c)) { 878inside: 879if(protect_hfs) { 880if(is_hfs_dotgit(path)) 881return0; 882if(S_ISLNK(mode)) { 883if(is_hfs_dotgitmodules(path)) 884return0; 885} 886} 887if(protect_ntfs) { 888if(is_ntfs_dotgit(path)) 889return0; 890if(S_ISLNK(mode)) { 891if(is_ntfs_dotgitmodules(path)) 892return0; 893} 894} 895 896 c = *path++; 897if((c =='.'&& !verify_dotfile(path, mode)) || 898is_dir_sep(c) || c =='\0') 899return0; 900} 901 c = *path++; 902} 903} 904 905/* 906 * Do we have another file that has the beginning components being a 907 * proper superset of the name we're trying to add? 908 */ 909static inthas_file_name(struct index_state *istate, 910const struct cache_entry *ce,int pos,int ok_to_replace) 911{ 912int retval =0; 913int len =ce_namelen(ce); 914int stage =ce_stage(ce); 915const char*name = ce->name; 916 917while(pos < istate->cache_nr) { 918struct cache_entry *p = istate->cache[pos++]; 919 920if(len >=ce_namelen(p)) 921break; 922if(memcmp(name, p->name, len)) 923break; 924if(ce_stage(p) != stage) 925continue; 926if(p->name[len] !='/') 927continue; 928if(p->ce_flags & CE_REMOVE) 929continue; 930 retval = -1; 931if(!ok_to_replace) 932break; 933remove_index_entry_at(istate, --pos); 934} 935return retval; 936} 937 938 939/* 940 * Like strcmp(), but also return the offset of the first change. 941 * If strings are equal, return the length. 942 */ 943intstrcmp_offset(const char*s1,const char*s2,size_t*first_change) 944{ 945size_t k; 946 947if(!first_change) 948returnstrcmp(s1, s2); 949 950for(k =0; s1[k] == s2[k]; k++) 951if(s1[k] =='\0') 952break; 953 954*first_change = k; 955return(unsigned char)s1[k] - (unsigned char)s2[k]; 956} 957 958/* 959 * Do we have another file with a pathname that is a proper 960 * subset of the name we're trying to add? 961 * 962 * That is, is there another file in the index with a path 963 * that matches a sub-directory in the given entry? 964 */ 965static inthas_dir_name(struct index_state *istate, 966const struct cache_entry *ce,int pos,int ok_to_replace) 967{ 968int retval =0; 969int stage =ce_stage(ce); 970const char*name = ce->name; 971const char*slash = name +ce_namelen(ce); 972size_t len_eq_last; 973int cmp_last =0; 974 975/* 976 * We are frequently called during an iteration on a sorted 977 * list of pathnames and while building a new index. Therefore, 978 * there is a high probability that this entry will eventually 979 * be appended to the index, rather than inserted in the middle. 980 * If we can confirm that, we can avoid binary searches on the 981 * components of the pathname. 982 * 983 * Compare the entry's full path with the last path in the index. 984 */ 985if(istate->cache_nr >0) { 986 cmp_last =strcmp_offset(name, 987 istate->cache[istate->cache_nr -1]->name, 988&len_eq_last); 989if(cmp_last >0) { 990if(len_eq_last ==0) { 991/* 992 * The entry sorts AFTER the last one in the 993 * index and their paths have no common prefix, 994 * so there cannot be a F/D conflict. 995 */ 996return retval; 997}else{ 998/* 999 * The entry sorts AFTER the last one in the1000 * index, but has a common prefix. Fall through1001 * to the loop below to disect the entry's path1002 * and see where the difference is.1003 */1004}1005}else if(cmp_last ==0) {1006/*1007 * The entry exactly matches the last one in the1008 * index, but because of multiple stage and CE_REMOVE1009 * items, we fall through and let the regular search1010 * code handle it.1011 */1012}1013}10141015for(;;) {1016size_t len;10171018for(;;) {1019if(*--slash =='/')1020break;1021if(slash <= ce->name)1022return retval;1023}1024 len = slash - name;10251026if(cmp_last >0) {1027/*1028 * (len + 1) is a directory boundary (including1029 * the trailing slash). And since the loop is1030 * decrementing "slash", the first iteration is1031 * the longest directory prefix; subsequent1032 * iterations consider parent directories.1033 */10341035if(len +1<= len_eq_last) {1036/*1037 * The directory prefix (including the trailing1038 * slash) also appears as a prefix in the last1039 * entry, so the remainder cannot collide (because1040 * strcmp said the whole path was greater).1041 *1042 * EQ: last: xxx/A1043 * this: xxx/B1044 *1045 * LT: last: xxx/file_A1046 * this: xxx/file_B1047 */1048return retval;1049}10501051if(len > len_eq_last) {1052/*1053 * This part of the directory prefix (excluding1054 * the trailing slash) is longer than the known1055 * equal portions, so this sub-directory cannot1056 * collide with a file.1057 *1058 * GT: last: xxxA1059 * this: xxxB/file1060 */1061return retval;1062}10631064if(istate->cache_nr >0&&1065ce_namelen(istate->cache[istate->cache_nr -1]) > len) {1066/*1067 * The directory prefix lines up with part of1068 * a longer file or directory name, but sorts1069 * after it, so this sub-directory cannot1070 * collide with a file.1071 *1072 * last: xxx/yy-file (because '-' sorts before '/')1073 * this: xxx/yy/abc1074 */1075return retval;1076}10771078/*1079 * This is a possible collision. Fall through and1080 * let the regular search code handle it.1081 *1082 * last: xxx1083 * this: xxx/file1084 */1085}10861087 pos =index_name_stage_pos(istate, name, len, stage);1088if(pos >=0) {1089/*1090 * Found one, but not so fast. This could1091 * be a marker that says "I was here, but1092 * I am being removed". Such an entry is1093 * not a part of the resulting tree, and1094 * it is Ok to have a directory at the same1095 * path.1096 */1097if(!(istate->cache[pos]->ce_flags & CE_REMOVE)) {1098 retval = -1;1099if(!ok_to_replace)1100break;1101remove_index_entry_at(istate, pos);1102continue;1103}1104}1105else1106 pos = -pos-1;11071108/*1109 * Trivial optimization: if we find an entry that1110 * already matches the sub-directory, then we know1111 * we're ok, and we can exit.1112 */1113while(pos < istate->cache_nr) {1114struct cache_entry *p = istate->cache[pos];1115if((ce_namelen(p) <= len) ||1116(p->name[len] !='/') ||1117memcmp(p->name, name, len))1118break;/* not our subdirectory */1119if(ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))1120/*1121 * p is at the same stage as our entry, and1122 * is a subdirectory of what we are looking1123 * at, so we cannot have conflicts at our1124 * level or anything shorter.1125 */1126return retval;1127 pos++;1128}1129}1130return retval;1131}11321133/* We may be in a situation where we already have path/file and path1134 * is being added, or we already have path and path/file is being1135 * added. Either one would result in a nonsense tree that has path1136 * twice when git-write-tree tries to write it out. Prevent it.1137 *1138 * If ok-to-replace is specified, we remove the conflicting entries1139 * from the cache so the caller should recompute the insert position.1140 * When this happens, we return non-zero.1141 */1142static intcheck_file_directory_conflict(struct index_state *istate,1143const struct cache_entry *ce,1144int pos,int ok_to_replace)1145{1146int retval;11471148/*1149 * When ce is an "I am going away" entry, we allow it to be added1150 */1151if(ce->ce_flags & CE_REMOVE)1152return0;11531154/*1155 * We check if the path is a sub-path of a subsequent pathname1156 * first, since removing those will not change the position1157 * in the array.1158 */1159 retval =has_file_name(istate, ce, pos, ok_to_replace);11601161/*1162 * Then check if the path might have a clashing sub-directory1163 * before it.1164 */1165return retval +has_dir_name(istate, ce, pos, ok_to_replace);1166}11671168static intadd_index_entry_with_check(struct index_state *istate,struct cache_entry *ce,int option)1169{1170int pos;1171int ok_to_add = option & ADD_CACHE_OK_TO_ADD;1172int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;1173int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;1174int new_only = option & ADD_CACHE_NEW_ONLY;11751176if(!(option & ADD_CACHE_KEEP_CACHE_TREE))1177cache_tree_invalidate_path(istate, ce->name);11781179/*1180 * If this entry's path sorts after the last entry in the index,1181 * we can avoid searching for it.1182 */1183if(istate->cache_nr >0&&1184strcmp(ce->name, istate->cache[istate->cache_nr -1]->name) >0)1185 pos = -istate->cache_nr -1;1186else1187 pos =index_name_stage_pos(istate, ce->name,ce_namelen(ce),ce_stage(ce));11881189/* existing match? Just replace it. */1190if(pos >=0) {1191if(!new_only)1192replace_index_entry(istate, pos, ce);1193return0;1194}1195 pos = -pos-1;11961197if(!(option & ADD_CACHE_KEEP_CACHE_TREE))1198untracked_cache_add_to_index(istate, ce->name);11991200/*1201 * Inserting a merged entry ("stage 0") into the index1202 * will always replace all non-merged entries..1203 */1204if(pos < istate->cache_nr &&ce_stage(ce) ==0) {1205while(ce_same_name(istate->cache[pos], ce)) {1206 ok_to_add =1;1207if(!remove_index_entry_at(istate, pos))1208break;1209}1210}12111212if(!ok_to_add)1213return-1;1214if(!verify_path(ce->name, ce->ce_mode))1215returnerror("Invalid path '%s'", ce->name);12161217if(!skip_df_check &&1218check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {1219if(!ok_to_replace)1220returnerror("'%s' appears as both a file and as a directory",1221 ce->name);1222 pos =index_name_stage_pos(istate, ce->name,ce_namelen(ce),ce_stage(ce));1223 pos = -pos-1;1224}1225return pos +1;1226}12271228intadd_index_entry(struct index_state *istate,struct cache_entry *ce,int option)1229{1230int pos;12311232if(option & ADD_CACHE_JUST_APPEND)1233 pos = istate->cache_nr;1234else{1235int ret;1236 ret =add_index_entry_with_check(istate, ce, option);1237if(ret <=0)1238return ret;1239 pos = ret -1;1240}12411242/* Make sure the array is big enough .. */1243ALLOC_GROW(istate->cache, istate->cache_nr +1, istate->cache_alloc);12441245/* Add it in.. */1246 istate->cache_nr++;1247if(istate->cache_nr > pos +1)1248MOVE_ARRAY(istate->cache + pos +1, istate->cache + pos,1249 istate->cache_nr - pos -1);1250set_index_entry(istate, pos, ce);1251 istate->cache_changed |= CE_ENTRY_ADDED;1252return0;1253}12541255/*1256 * "refresh" does not calculate a new sha1 file or bring the1257 * cache up-to-date for mode/content changes. But what it1258 * _does_ do is to "re-match" the stat information of a file1259 * with the cache, so that you can refresh the cache for a1260 * file that hasn't been changed but where the stat entry is1261 * out of date.1262 *1263 * For example, you'd want to do this after doing a "git-read-tree",1264 * to link up the stat cache details with the proper files.1265 */1266static struct cache_entry *refresh_cache_ent(struct index_state *istate,1267struct cache_entry *ce,1268unsigned int options,int*err,1269int*changed_ret)1270{1271struct stat st;1272struct cache_entry *updated;1273int changed, size;1274int refresh = options & CE_MATCH_REFRESH;1275int ignore_valid = options & CE_MATCH_IGNORE_VALID;1276int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;1277int ignore_missing = options & CE_MATCH_IGNORE_MISSING;1278int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;12791280if(!refresh ||ce_uptodate(ce))1281return ce;12821283if(!ignore_fsmonitor)1284refresh_fsmonitor(istate);1285/*1286 * CE_VALID or CE_SKIP_WORKTREE means the user promised us1287 * that the change to the work tree does not matter and told1288 * us not to worry.1289 */1290if(!ignore_skip_worktree &&ce_skip_worktree(ce)) {1291ce_mark_uptodate(ce);1292return ce;1293}1294if(!ignore_valid && (ce->ce_flags & CE_VALID)) {1295ce_mark_uptodate(ce);1296return ce;1297}1298if(!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {1299ce_mark_uptodate(ce);1300return ce;1301}13021303if(has_symlink_leading_path(ce->name,ce_namelen(ce))) {1304if(ignore_missing)1305return ce;1306if(err)1307*err = ENOENT;1308return NULL;1309}13101311if(lstat(ce->name, &st) <0) {1312if(ignore_missing && errno == ENOENT)1313return ce;1314if(err)1315*err = errno;1316return NULL;1317}13181319 changed =ie_match_stat(istate, ce, &st, options);1320if(changed_ret)1321*changed_ret = changed;1322if(!changed) {1323/*1324 * The path is unchanged. If we were told to ignore1325 * valid bit, then we did the actual stat check and1326 * found that the entry is unmodified. If the entry1327 * is not marked VALID, this is the place to mark it1328 * valid again, under "assume unchanged" mode.1329 */1330if(ignore_valid && assume_unchanged &&1331!(ce->ce_flags & CE_VALID))1332;/* mark this one VALID again */1333else{1334/*1335 * We do not mark the index itself "modified"1336 * because CE_UPTODATE flag is in-core only;1337 * we are not going to write this change out.1338 */1339if(!S_ISGITLINK(ce->ce_mode)) {1340ce_mark_uptodate(ce);1341mark_fsmonitor_valid(ce);1342}1343return ce;1344}1345}13461347if(ie_modified(istate, ce, &st, options)) {1348if(err)1349*err = EINVAL;1350return NULL;1351}13521353 size =ce_size(ce);1354 updated =xmalloc(size);1355copy_cache_entry(updated, ce);1356memcpy(updated->name, ce->name, ce->ce_namelen +1);1357fill_stat_cache_info(updated, &st);1358/*1359 * If ignore_valid is not set, we should leave CE_VALID bit1360 * alone. Otherwise, paths marked with --no-assume-unchanged1361 * (i.e. things to be edited) will reacquire CE_VALID bit1362 * automatically, which is not really what we want.1363 */1364if(!ignore_valid && assume_unchanged &&1365!(ce->ce_flags & CE_VALID))1366 updated->ce_flags &= ~CE_VALID;13671368/* istate->cache_changed is updated in the caller */1369return updated;1370}13711372static voidshow_file(const char* fmt,const char* name,int in_porcelain,1373int* first,const char*header_msg)1374{1375if(in_porcelain && *first && header_msg) {1376printf("%s\n", header_msg);1377*first =0;1378}1379printf(fmt, name);1380}13811382intrefresh_index(struct index_state *istate,unsigned int flags,1383const struct pathspec *pathspec,1384char*seen,const char*header_msg)1385{1386int i;1387int has_errors =0;1388int really = (flags & REFRESH_REALLY) !=0;1389int allow_unmerged = (flags & REFRESH_UNMERGED) !=0;1390int quiet = (flags & REFRESH_QUIET) !=0;1391int not_new = (flags & REFRESH_IGNORE_MISSING) !=0;1392int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) !=0;1393int first =1;1394int in_porcelain = (flags & REFRESH_IN_PORCELAIN);1395unsigned int options = (CE_MATCH_REFRESH |1396(really ? CE_MATCH_IGNORE_VALID :0) |1397(not_new ? CE_MATCH_IGNORE_MISSING :0));1398const char*modified_fmt;1399const char*deleted_fmt;1400const char*typechange_fmt;1401const char*added_fmt;1402const char*unmerged_fmt;1403uint64_t start =getnanotime();14041405 modified_fmt = (in_porcelain ?"M\t%s\n":"%s: needs update\n");1406 deleted_fmt = (in_porcelain ?"D\t%s\n":"%s: needs update\n");1407 typechange_fmt = (in_porcelain ?"T\t%s\n":"%sneeds update\n");1408 added_fmt = (in_porcelain ?"A\t%s\n":"%sneeds update\n");1409 unmerged_fmt = (in_porcelain ?"U\t%s\n":"%s: needs merge\n");1410for(i =0; i < istate->cache_nr; i++) {1411struct cache_entry *ce, *new_entry;1412int cache_errno =0;1413int changed =0;1414int filtered =0;14151416 ce = istate->cache[i];1417if(ignore_submodules &&S_ISGITLINK(ce->ce_mode))1418continue;14191420if(pathspec && !ce_path_match(ce, pathspec, seen))1421 filtered =1;14221423if(ce_stage(ce)) {1424while((i < istate->cache_nr) &&1425!strcmp(istate->cache[i]->name, ce->name))1426 i++;1427 i--;1428if(allow_unmerged)1429continue;1430if(!filtered)1431show_file(unmerged_fmt, ce->name, in_porcelain,1432&first, header_msg);1433 has_errors =1;1434continue;1435}14361437if(filtered)1438continue;14391440 new_entry =refresh_cache_ent(istate, ce, options, &cache_errno, &changed);1441if(new_entry == ce)1442continue;1443if(!new_entry) {1444const char*fmt;14451446if(really && cache_errno == EINVAL) {1447/* If we are doing --really-refresh that1448 * means the index is not valid anymore.1449 */1450 ce->ce_flags &= ~CE_VALID;1451 ce->ce_flags |= CE_UPDATE_IN_BASE;1452mark_fsmonitor_invalid(istate, ce);1453 istate->cache_changed |= CE_ENTRY_CHANGED;1454}1455if(quiet)1456continue;14571458if(cache_errno == ENOENT)1459 fmt = deleted_fmt;1460else if(ce_intent_to_add(ce))1461 fmt = added_fmt;/* must be before other checks */1462else if(changed & TYPE_CHANGED)1463 fmt = typechange_fmt;1464else1465 fmt = modified_fmt;1466show_file(fmt,1467 ce->name, in_porcelain, &first, header_msg);1468 has_errors =1;1469continue;1470}14711472replace_index_entry(istate, i, new_entry);1473}1474trace_performance_since(start,"refresh index");1475return has_errors;1476}14771478struct cache_entry *refresh_cache_entry(struct index_state *istate,1479struct cache_entry *ce,1480unsigned int options)1481{1482returnrefresh_cache_ent(istate, ce, options, NULL, NULL);1483}148414851486/*****************************************************************1487 * Index File I/O1488 *****************************************************************/14891490#define INDEX_FORMAT_DEFAULT 314911492static unsigned intget_index_format_default(void)1493{1494char*envversion =getenv("GIT_INDEX_VERSION");1495char*endp;1496int value;1497unsigned int version = INDEX_FORMAT_DEFAULT;14981499if(!envversion) {1500if(!git_config_get_int("index.version", &value))1501 version = value;1502if(version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {1503warning(_("index.version set, but the value is invalid.\n"1504"Using version%i"), INDEX_FORMAT_DEFAULT);1505return INDEX_FORMAT_DEFAULT;1506}1507return version;1508}15091510 version =strtoul(envversion, &endp,10);1511if(*endp ||1512 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {1513warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"1514"Using version%i"), INDEX_FORMAT_DEFAULT);1515 version = INDEX_FORMAT_DEFAULT;1516}1517return version;1518}15191520/*1521 * dev/ino/uid/gid/size are also just tracked to the low 32 bits1522 * Again - this is just a (very strong in practice) heuristic that1523 * the inode hasn't changed.1524 *1525 * We save the fields in big-endian order to allow using the1526 * index file over NFS transparently.1527 */1528struct ondisk_cache_entry {1529struct cache_time ctime;1530struct cache_time mtime;1531uint32_t dev;1532uint32_t ino;1533uint32_t mode;1534uint32_t uid;1535uint32_t gid;1536uint32_t size;1537unsigned char sha1[20];1538uint16_t flags;1539char name[FLEX_ARRAY];/* more */1540};15411542/*1543 * This struct is used when CE_EXTENDED bit is 11544 * The struct must match ondisk_cache_entry exactly from1545 * ctime till flags1546 */1547struct ondisk_cache_entry_extended {1548struct cache_time ctime;1549struct cache_time mtime;1550uint32_t dev;1551uint32_t ino;1552uint32_t mode;1553uint32_t uid;1554uint32_t gid;1555uint32_t size;1556unsigned char sha1[20];1557uint16_t flags;1558uint16_t flags2;1559char name[FLEX_ARRAY];/* more */1560};15611562/* These are only used for v3 or lower */1563#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)1564#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)1565#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)1566#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)1567#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \1568 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \1569 ondisk_cache_entry_size(ce_namelen(ce)))15701571/* Allow fsck to force verification of the index checksum. */1572int verify_index_checksum;15731574/* Allow fsck to force verification of the cache entry order. */1575int verify_ce_order;15761577static intverify_hdr(struct cache_header *hdr,unsigned long size)1578{1579 git_hash_ctx c;1580unsigned char hash[GIT_MAX_RAWSZ];1581int hdr_version;15821583if(hdr->hdr_signature !=htonl(CACHE_SIGNATURE))1584returnerror("bad signature");1585 hdr_version =ntohl(hdr->hdr_version);1586if(hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)1587returnerror("bad index version%d", hdr_version);15881589if(!verify_index_checksum)1590return0;15911592 the_hash_algo->init_fn(&c);1593 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);1594 the_hash_algo->final_fn(hash, &c);1595if(hashcmp(hash, (unsigned char*)hdr + size - the_hash_algo->rawsz))1596returnerror("bad index file sha1 signature");1597return0;1598}15991600static intread_index_extension(struct index_state *istate,1601const char*ext,void*data,unsigned long sz)1602{1603switch(CACHE_EXT(ext)) {1604case CACHE_EXT_TREE:1605 istate->cache_tree =cache_tree_read(data, sz);1606break;1607case CACHE_EXT_RESOLVE_UNDO:1608 istate->resolve_undo =resolve_undo_read(data, sz);1609break;1610case CACHE_EXT_LINK:1611if(read_link_extension(istate, data, sz))1612return-1;1613break;1614case CACHE_EXT_UNTRACKED:1615 istate->untracked =read_untracked_extension(data, sz);1616break;1617case CACHE_EXT_FSMONITOR:1618read_fsmonitor_extension(istate, data, sz);1619break;1620default:1621if(*ext <'A'||'Z'< *ext)1622returnerror("index uses %.4s extension, which we do not understand",1623 ext);1624fprintf(stderr,"ignoring %.4s extension\n", ext);1625break;1626}1627return0;1628}16291630inthold_locked_index(struct lock_file *lk,int lock_flags)1631{1632returnhold_lock_file_for_update(lk,get_index_file(), lock_flags);1633}16341635intread_index(struct index_state *istate)1636{1637returnread_index_from(istate,get_index_file(),get_git_dir());1638}16391640static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,1641unsigned int flags,1642const char*name,1643size_t len)1644{1645struct cache_entry *ce =xmalloc(cache_entry_size(len));16461647 ce->ce_stat_data.sd_ctime.sec =get_be32(&ondisk->ctime.sec);1648 ce->ce_stat_data.sd_mtime.sec =get_be32(&ondisk->mtime.sec);1649 ce->ce_stat_data.sd_ctime.nsec =get_be32(&ondisk->ctime.nsec);1650 ce->ce_stat_data.sd_mtime.nsec =get_be32(&ondisk->mtime.nsec);1651 ce->ce_stat_data.sd_dev =get_be32(&ondisk->dev);1652 ce->ce_stat_data.sd_ino =get_be32(&ondisk->ino);1653 ce->ce_mode =get_be32(&ondisk->mode);1654 ce->ce_stat_data.sd_uid =get_be32(&ondisk->uid);1655 ce->ce_stat_data.sd_gid =get_be32(&ondisk->gid);1656 ce->ce_stat_data.sd_size =get_be32(&ondisk->size);1657 ce->ce_flags = flags & ~CE_NAMEMASK;1658 ce->ce_namelen = len;1659 ce->index =0;1660hashcpy(ce->oid.hash, ondisk->sha1);1661memcpy(ce->name, name, len);1662 ce->name[len] ='\0';1663return ce;1664}16651666/*1667 * Adjacent cache entries tend to share the leading paths, so it makes1668 * sense to only store the differences in later entries. In the v41669 * on-disk format of the index, each on-disk cache entry stores the1670 * number of bytes to be stripped from the end of the previous name,1671 * and the bytes to append to the result, to come up with its name.1672 */1673static unsigned longexpand_name_field(struct strbuf *name,const char*cp_)1674{1675const unsigned char*ep, *cp = (const unsigned char*)cp_;1676size_t len =decode_varint(&cp);16771678if(name->len < len)1679die("malformed name field in the index");1680strbuf_remove(name, name->len - len, len);1681for(ep = cp; *ep; ep++)1682;/* find the end */1683strbuf_add(name, cp, ep - cp);1684return(const char*)ep +1- cp_;1685}16861687static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,1688unsigned long*ent_size,1689struct strbuf *previous_name)1690{1691struct cache_entry *ce;1692size_t len;1693const char*name;1694unsigned int flags;16951696/* On-disk flags are just 16 bits */1697 flags =get_be16(&ondisk->flags);1698 len = flags & CE_NAMEMASK;16991700if(flags & CE_EXTENDED) {1701struct ondisk_cache_entry_extended *ondisk2;1702int extended_flags;1703 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1704 extended_flags =get_be16(&ondisk2->flags2) <<16;1705/* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1706if(extended_flags & ~CE_EXTENDED_FLAGS)1707die("Unknown index entry format%08x", extended_flags);1708 flags |= extended_flags;1709 name = ondisk2->name;1710}1711else1712 name = ondisk->name;17131714if(!previous_name) {1715/* v3 and earlier */1716if(len == CE_NAMEMASK)1717 len =strlen(name);1718 ce =cache_entry_from_ondisk(ondisk, flags, name, len);17191720*ent_size =ondisk_ce_size(ce);1721}else{1722unsigned long consumed;1723 consumed =expand_name_field(previous_name, name);1724 ce =cache_entry_from_ondisk(ondisk, flags,1725 previous_name->buf,1726 previous_name->len);17271728*ent_size = (name - ((char*)ondisk)) + consumed;1729}1730return ce;1731}17321733static voidcheck_ce_order(struct index_state *istate)1734{1735unsigned int i;17361737if(!verify_ce_order)1738return;17391740for(i =1; i < istate->cache_nr; i++) {1741struct cache_entry *ce = istate->cache[i -1];1742struct cache_entry *next_ce = istate->cache[i];1743int name_compare =strcmp(ce->name, next_ce->name);17441745if(0< name_compare)1746die("unordered stage entries in index");1747if(!name_compare) {1748if(!ce_stage(ce))1749die("multiple stage entries for merged file '%s'",1750 ce->name);1751if(ce_stage(ce) >ce_stage(next_ce))1752die("unordered stage entries for '%s'",1753 ce->name);1754}1755}1756}17571758static voidtweak_untracked_cache(struct index_state *istate)1759{1760switch(git_config_get_untracked_cache()) {1761case-1:/* keep: do nothing */1762break;1763case0:/* false */1764remove_untracked_cache(istate);1765break;1766case1:/* true */1767add_untracked_cache(istate);1768break;1769default:/* unknown value: do nothing */1770break;1771}1772}17731774static voidtweak_split_index(struct index_state *istate)1775{1776switch(git_config_get_split_index()) {1777case-1:/* unset: do nothing */1778break;1779case0:/* false */1780remove_split_index(istate);1781break;1782case1:/* true */1783add_split_index(istate);1784break;1785default:/* unknown value: do nothing */1786break;1787}1788}17891790static voidpost_read_index_from(struct index_state *istate)1791{1792check_ce_order(istate);1793tweak_untracked_cache(istate);1794tweak_split_index(istate);1795tweak_fsmonitor(istate);1796}17971798/* remember to discard_cache() before reading a different cache! */1799intdo_read_index(struct index_state *istate,const char*path,int must_exist)1800{1801int fd, i;1802struct stat st;1803unsigned long src_offset;1804struct cache_header *hdr;1805void*mmap;1806size_t mmap_size;1807struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;18081809if(istate->initialized)1810return istate->cache_nr;18111812 istate->timestamp.sec =0;1813 istate->timestamp.nsec =0;1814 fd =open(path, O_RDONLY);1815if(fd <0) {1816if(!must_exist && errno == ENOENT)1817return0;1818die_errno("%s: index file open failed", path);1819}18201821if(fstat(fd, &st))1822die_errno("cannot stat the open index");18231824 mmap_size =xsize_t(st.st_size);1825if(mmap_size <sizeof(struct cache_header) + the_hash_algo->rawsz)1826die("index file smaller than expected");18271828 mmap =xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd,0);1829if(mmap == MAP_FAILED)1830die_errno("unable to map index file");1831close(fd);18321833 hdr = mmap;1834if(verify_hdr(hdr, mmap_size) <0)1835goto unmap;18361837hashcpy(istate->oid.hash, (const unsigned char*)hdr + mmap_size - the_hash_algo->rawsz);1838 istate->version =ntohl(hdr->hdr_version);1839 istate->cache_nr =ntohl(hdr->hdr_entries);1840 istate->cache_alloc =alloc_nr(istate->cache_nr);1841 istate->cache =xcalloc(istate->cache_alloc,sizeof(*istate->cache));1842 istate->initialized =1;18431844if(istate->version ==4)1845 previous_name = &previous_name_buf;1846else1847 previous_name = NULL;18481849 src_offset =sizeof(*hdr);1850for(i =0; i < istate->cache_nr; i++) {1851struct ondisk_cache_entry *disk_ce;1852struct cache_entry *ce;1853unsigned long consumed;18541855 disk_ce = (struct ondisk_cache_entry *)((char*)mmap + src_offset);1856 ce =create_from_disk(disk_ce, &consumed, previous_name);1857set_index_entry(istate, i, ce);18581859 src_offset += consumed;1860}1861strbuf_release(&previous_name_buf);1862 istate->timestamp.sec = st.st_mtime;1863 istate->timestamp.nsec =ST_MTIME_NSEC(st);18641865while(src_offset <= mmap_size - the_hash_algo->rawsz -8) {1866/* After an array of active_nr index entries,1867 * there can be arbitrary number of extended1868 * sections, each of which is prefixed with1869 * extension name (4-byte) and section length1870 * in 4-byte network byte order.1871 */1872uint32_t extsize;1873memcpy(&extsize, (char*)mmap + src_offset +4,4);1874 extsize =ntohl(extsize);1875if(read_index_extension(istate,1876(const char*) mmap + src_offset,1877(char*) mmap + src_offset +8,1878 extsize) <0)1879goto unmap;1880 src_offset +=8;1881 src_offset += extsize;1882}1883munmap(mmap, mmap_size);1884return istate->cache_nr;18851886unmap:1887munmap(mmap, mmap_size);1888die("index file corrupt");1889}18901891/*1892 * Signal that the shared index is used by updating its mtime.1893 *1894 * This way, shared index can be removed if they have not been used1895 * for some time.1896 */1897static voidfreshen_shared_index(const char*shared_index,int warn)1898{1899if(!check_and_freshen_file(shared_index,1) && warn)1900warning("could not freshen shared index '%s'", shared_index);1901}19021903intread_index_from(struct index_state *istate,const char*path,1904const char*gitdir)1905{1906uint64_t start =getnanotime();1907struct split_index *split_index;1908int ret;1909char*base_oid_hex;1910char*base_path;19111912/* istate->initialized covers both .git/index and .git/sharedindex.xxx */1913if(istate->initialized)1914return istate->cache_nr;19151916 ret =do_read_index(istate, path,0);1917trace_performance_since(start,"read cache%s", path);19181919 split_index = istate->split_index;1920if(!split_index ||is_null_oid(&split_index->base_oid)) {1921post_read_index_from(istate);1922return ret;1923}19241925if(split_index->base)1926discard_index(split_index->base);1927else1928 split_index->base =xcalloc(1,sizeof(*split_index->base));19291930 base_oid_hex =oid_to_hex(&split_index->base_oid);1931 base_path =xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);1932 ret =do_read_index(split_index->base, base_path,1);1933if(oidcmp(&split_index->base_oid, &split_index->base->oid))1934die("broken index, expect%sin%s, got%s",1935 base_oid_hex, base_path,1936oid_to_hex(&split_index->base->oid));19371938freshen_shared_index(base_path,0);1939merge_base_index(istate);1940post_read_index_from(istate);1941trace_performance_since(start,"read cache%s", base_path);1942free(base_path);1943return ret;1944}19451946intis_index_unborn(struct index_state *istate)1947{1948return(!istate->cache_nr && !istate->timestamp.sec);1949}19501951intdiscard_index(struct index_state *istate)1952{1953int i;19541955for(i =0; i < istate->cache_nr; i++) {1956if(istate->cache[i]->index &&1957 istate->split_index &&1958 istate->split_index->base &&1959 istate->cache[i]->index <= istate->split_index->base->cache_nr &&1960 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index -1])1961continue;1962free(istate->cache[i]);1963}1964resolve_undo_clear_index(istate);1965 istate->cache_nr =0;1966 istate->cache_changed =0;1967 istate->timestamp.sec =0;1968 istate->timestamp.nsec =0;1969free_name_hash(istate);1970cache_tree_free(&(istate->cache_tree));1971 istate->initialized =0;1972FREE_AND_NULL(istate->cache);1973 istate->cache_alloc =0;1974discard_split_index(istate);1975free_untracked_cache(istate->untracked);1976 istate->untracked = NULL;1977return0;1978}19791980intunmerged_index(const struct index_state *istate)1981{1982int i;1983for(i =0; i < istate->cache_nr; i++) {1984if(ce_stage(istate->cache[i]))1985return1;1986}1987return0;1988}19891990#define WRITE_BUFFER_SIZE 81921991static unsigned char write_buffer[WRITE_BUFFER_SIZE];1992static unsigned long write_buffer_len;19931994static intce_write_flush(git_hash_ctx *context,int fd)1995{1996unsigned int buffered = write_buffer_len;1997if(buffered) {1998 the_hash_algo->update_fn(context, write_buffer, buffered);1999if(write_in_full(fd, write_buffer, buffered) <0)2000return-1;2001 write_buffer_len =0;2002}2003return0;2004}20052006static intce_write(git_hash_ctx *context,int fd,void*data,unsigned int len)2007{2008while(len) {2009unsigned int buffered = write_buffer_len;2010unsigned int partial = WRITE_BUFFER_SIZE - buffered;2011if(partial > len)2012 partial = len;2013memcpy(write_buffer + buffered, data, partial);2014 buffered += partial;2015if(buffered == WRITE_BUFFER_SIZE) {2016 write_buffer_len = buffered;2017if(ce_write_flush(context, fd))2018return-1;2019 buffered =0;2020}2021 write_buffer_len = buffered;2022 len -= partial;2023 data = (char*) data + partial;2024}2025return0;2026}20272028static intwrite_index_ext_header(git_hash_ctx *context,int fd,2029unsigned int ext,unsigned int sz)2030{2031 ext =htonl(ext);2032 sz =htonl(sz);2033return((ce_write(context, fd, &ext,4) <0) ||2034(ce_write(context, fd, &sz,4) <0)) ? -1:0;2035}20362037static intce_flush(git_hash_ctx *context,int fd,unsigned char*hash)2038{2039unsigned int left = write_buffer_len;20402041if(left) {2042 write_buffer_len =0;2043 the_hash_algo->update_fn(context, write_buffer, left);2044}20452046/* Flush first if not enough space for hash signature */2047if(left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {2048if(write_in_full(fd, write_buffer, left) <0)2049return-1;2050 left =0;2051}20522053/* Append the hash signature at the end */2054 the_hash_algo->final_fn(write_buffer + left, context);2055hashcpy(hash, write_buffer + left);2056 left += the_hash_algo->rawsz;2057return(write_in_full(fd, write_buffer, left) <0) ? -1:0;2058}20592060static voidce_smudge_racily_clean_entry(struct cache_entry *ce)2061{2062/*2063 * The only thing we care about in this function is to smudge the2064 * falsely clean entry due to touch-update-touch race, so we leave2065 * everything else as they are. We are called for entries whose2066 * ce_stat_data.sd_mtime match the index file mtime.2067 *2068 * Note that this actually does not do much for gitlinks, for2069 * which ce_match_stat_basic() always goes to the actual2070 * contents. The caller checks with is_racy_timestamp() which2071 * always says "no" for gitlinks, so we are not called for them ;-)2072 */2073struct stat st;20742075if(lstat(ce->name, &st) <0)2076return;2077if(ce_match_stat_basic(ce, &st))2078return;2079if(ce_modified_check_fs(ce, &st)) {2080/* This is "racily clean"; smudge it. Note that this2081 * is a tricky code. At first glance, it may appear2082 * that it can break with this sequence:2083 *2084 * $ echo xyzzy >frotz2085 * $ git-update-index --add frotz2086 * $ : >frotz2087 * $ sleep 32088 * $ echo filfre >nitfol2089 * $ git-update-index --add nitfol2090 *2091 * but it does not. When the second update-index runs,2092 * it notices that the entry "frotz" has the same timestamp2093 * as index, and if we were to smudge it by resetting its2094 * size to zero here, then the object name recorded2095 * in index is the 6-byte file but the cached stat information2096 * becomes zero --- which would then match what we would2097 * obtain from the filesystem next time we stat("frotz").2098 *2099 * However, the second update-index, before calling2100 * this function, notices that the cached size is 62101 * bytes and what is on the filesystem is an empty2102 * file, and never calls us, so the cached size information2103 * for "frotz" stays 6 which does not match the filesystem.2104 */2105 ce->ce_stat_data.sd_size =0;2106}2107}21082109/* Copy miscellaneous fields but not the name */2110static voidcopy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,2111struct cache_entry *ce)2112{2113short flags;21142115 ondisk->ctime.sec =htonl(ce->ce_stat_data.sd_ctime.sec);2116 ondisk->mtime.sec =htonl(ce->ce_stat_data.sd_mtime.sec);2117 ondisk->ctime.nsec =htonl(ce->ce_stat_data.sd_ctime.nsec);2118 ondisk->mtime.nsec =htonl(ce->ce_stat_data.sd_mtime.nsec);2119 ondisk->dev =htonl(ce->ce_stat_data.sd_dev);2120 ondisk->ino =htonl(ce->ce_stat_data.sd_ino);2121 ondisk->mode =htonl(ce->ce_mode);2122 ondisk->uid =htonl(ce->ce_stat_data.sd_uid);2123 ondisk->gid =htonl(ce->ce_stat_data.sd_gid);2124 ondisk->size =htonl(ce->ce_stat_data.sd_size);2125hashcpy(ondisk->sha1, ce->oid.hash);21262127 flags = ce->ce_flags & ~CE_NAMEMASK;2128 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK :ce_namelen(ce));2129 ondisk->flags =htons(flags);2130if(ce->ce_flags & CE_EXTENDED) {2131struct ondisk_cache_entry_extended *ondisk2;2132 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;2133 ondisk2->flags2 =htons((ce->ce_flags & CE_EXTENDED_FLAGS) >>16);2134}2135}21362137static intce_write_entry(git_hash_ctx *c,int fd,struct cache_entry *ce,2138struct strbuf *previous_name,struct ondisk_cache_entry *ondisk)2139{2140int size;2141int result;2142unsigned int saved_namelen;2143int stripped_name =0;2144static unsigned char padding[8] = {0x00};21452146if(ce->ce_flags & CE_STRIP_NAME) {2147 saved_namelen =ce_namelen(ce);2148 ce->ce_namelen =0;2149 stripped_name =1;2150}21512152if(ce->ce_flags & CE_EXTENDED)2153 size =offsetof(struct ondisk_cache_entry_extended, name);2154else2155 size =offsetof(struct ondisk_cache_entry, name);21562157if(!previous_name) {2158int len =ce_namelen(ce);2159copy_cache_entry_to_ondisk(ondisk, ce);2160 result =ce_write(c, fd, ondisk, size);2161if(!result)2162 result =ce_write(c, fd, ce->name, len);2163if(!result)2164 result =ce_write(c, fd, padding,align_padding_size(size, len));2165}else{2166int common, to_remove, prefix_size;2167unsigned char to_remove_vi[16];2168for(common =0;2169(ce->name[common] &&2170 common < previous_name->len &&2171 ce->name[common] == previous_name->buf[common]);2172 common++)2173;/* still matching */2174 to_remove = previous_name->len - common;2175 prefix_size =encode_varint(to_remove, to_remove_vi);21762177copy_cache_entry_to_ondisk(ondisk, ce);2178 result =ce_write(c, fd, ondisk, size);2179if(!result)2180 result =ce_write(c, fd, to_remove_vi, prefix_size);2181if(!result)2182 result =ce_write(c, fd, ce->name + common,ce_namelen(ce) - common);2183if(!result)2184 result =ce_write(c, fd, padding,1);21852186strbuf_splice(previous_name, common, to_remove,2187 ce->name + common,ce_namelen(ce) - common);2188}2189if(stripped_name) {2190 ce->ce_namelen = saved_namelen;2191 ce->ce_flags &= ~CE_STRIP_NAME;2192}21932194return result;2195}21962197/*2198 * This function verifies if index_state has the correct sha1 of the2199 * index file. Don't die if we have any other failure, just return 0.2200 */2201static intverify_index_from(const struct index_state *istate,const char*path)2202{2203int fd;2204 ssize_t n;2205struct stat st;2206unsigned char hash[GIT_MAX_RAWSZ];22072208if(!istate->initialized)2209return0;22102211 fd =open(path, O_RDONLY);2212if(fd <0)2213return0;22142215if(fstat(fd, &st))2216goto out;22172218if(st.st_size <sizeof(struct cache_header) + the_hash_algo->rawsz)2219goto out;22202221 n =pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);2222if(n != the_hash_algo->rawsz)2223goto out;22242225if(hashcmp(istate->oid.hash, hash))2226goto out;22272228close(fd);2229return1;22302231out:2232close(fd);2233return0;2234}22352236static intverify_index(const struct index_state *istate)2237{2238returnverify_index_from(istate,get_index_file());2239}22402241static inthas_racy_timestamp(struct index_state *istate)2242{2243int entries = istate->cache_nr;2244int i;22452246for(i =0; i < entries; i++) {2247struct cache_entry *ce = istate->cache[i];2248if(is_racy_timestamp(istate, ce))2249return1;2250}2251return0;2252}22532254voidupdate_index_if_able(struct index_state *istate,struct lock_file *lockfile)2255{2256if((istate->cache_changed ||has_racy_timestamp(istate)) &&2257verify_index(istate))2258write_locked_index(istate, lockfile, COMMIT_LOCK);2259else2260rollback_lock_file(lockfile);2261}22622263/*2264 * On success, `tempfile` is closed. If it is the temporary file2265 * of a `struct lock_file`, we will therefore effectively perform2266 * a 'close_lock_file_gently()`. Since that is an implementation2267 * detail of lockfiles, callers of `do_write_index()` should not2268 * rely on it.2269 */2270static intdo_write_index(struct index_state *istate,struct tempfile *tempfile,2271int strip_extensions)2272{2273uint64_t start =getnanotime();2274int newfd = tempfile->fd;2275 git_hash_ctx c;2276struct cache_header hdr;2277int i, err =0, removed, extended, hdr_version;2278struct cache_entry **cache = istate->cache;2279int entries = istate->cache_nr;2280struct stat st;2281struct ondisk_cache_entry_extended ondisk;2282struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;2283int drop_cache_tree = istate->drop_cache_tree;22842285for(i = removed = extended =0; i < entries; i++) {2286if(cache[i]->ce_flags & CE_REMOVE)2287 removed++;22882289/* reduce extended entries if possible */2290 cache[i]->ce_flags &= ~CE_EXTENDED;2291if(cache[i]->ce_flags & CE_EXTENDED_FLAGS) {2292 extended++;2293 cache[i]->ce_flags |= CE_EXTENDED;2294}2295}22962297if(!istate->version) {2298 istate->version =get_index_format_default();2299if(git_env_bool("GIT_TEST_SPLIT_INDEX",0))2300init_split_index(istate);2301}23022303/* demote version 3 to version 2 when the latter suffices */2304if(istate->version ==3|| istate->version ==2)2305 istate->version = extended ?3:2;23062307 hdr_version = istate->version;23082309 hdr.hdr_signature =htonl(CACHE_SIGNATURE);2310 hdr.hdr_version =htonl(hdr_version);2311 hdr.hdr_entries =htonl(entries - removed);23122313 the_hash_algo->init_fn(&c);2314if(ce_write(&c, newfd, &hdr,sizeof(hdr)) <0)2315return-1;23162317 previous_name = (hdr_version ==4) ? &previous_name_buf : NULL;23182319for(i =0; i < entries; i++) {2320struct cache_entry *ce = cache[i];2321if(ce->ce_flags & CE_REMOVE)2322continue;2323if(!ce_uptodate(ce) &&is_racy_timestamp(istate, ce))2324ce_smudge_racily_clean_entry(ce);2325if(is_null_oid(&ce->oid)) {2326static const char msg[] ="cache entry has null sha1:%s";2327static int allow = -1;23282329if(allow <0)2330 allow =git_env_bool("GIT_ALLOW_NULL_SHA1",0);2331if(allow)2332warning(msg, ce->name);2333else2334 err =error(msg, ce->name);23352336 drop_cache_tree =1;2337}2338if(ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) <0)2339 err = -1;23402341if(err)2342break;2343}2344strbuf_release(&previous_name_buf);23452346if(err)2347return err;23482349/* Write extension data here */2350if(!strip_extensions && istate->split_index) {2351struct strbuf sb = STRBUF_INIT;23522353 err =write_link_extension(&sb, istate) <0||2354write_index_ext_header(&c, newfd, CACHE_EXT_LINK,2355 sb.len) <0||2356ce_write(&c, newfd, sb.buf, sb.len) <0;2357strbuf_release(&sb);2358if(err)2359return-1;2360}2361if(!strip_extensions && !drop_cache_tree && istate->cache_tree) {2362struct strbuf sb = STRBUF_INIT;23632364cache_tree_write(&sb, istate->cache_tree);2365 err =write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) <02366||ce_write(&c, newfd, sb.buf, sb.len) <0;2367strbuf_release(&sb);2368if(err)2369return-1;2370}2371if(!strip_extensions && istate->resolve_undo) {2372struct strbuf sb = STRBUF_INIT;23732374resolve_undo_write(&sb, istate->resolve_undo);2375 err =write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,2376 sb.len) <02377||ce_write(&c, newfd, sb.buf, sb.len) <0;2378strbuf_release(&sb);2379if(err)2380return-1;2381}2382if(!strip_extensions && istate->untracked) {2383struct strbuf sb = STRBUF_INIT;23842385write_untracked_extension(&sb, istate->untracked);2386 err =write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,2387 sb.len) <0||2388ce_write(&c, newfd, sb.buf, sb.len) <0;2389strbuf_release(&sb);2390if(err)2391return-1;2392}2393if(!strip_extensions && istate->fsmonitor_last_update) {2394struct strbuf sb = STRBUF_INIT;23952396write_fsmonitor_extension(&sb, istate);2397 err =write_index_ext_header(&c, newfd, CACHE_EXT_FSMONITOR, sb.len) <02398||ce_write(&c, newfd, sb.buf, sb.len) <0;2399strbuf_release(&sb);2400if(err)2401return-1;2402}24032404if(ce_flush(&c, newfd, istate->oid.hash))2405return-1;2406if(close_tempfile_gently(tempfile)) {2407error(_("could not close '%s'"), tempfile->filename.buf);2408return-1;2409}2410if(stat(tempfile->filename.buf, &st))2411return-1;2412 istate->timestamp.sec = (unsigned int)st.st_mtime;2413 istate->timestamp.nsec =ST_MTIME_NSEC(st);2414trace_performance_since(start,"write index, changed mask =%x", istate->cache_changed);2415return0;2416}24172418voidset_alternate_index_output(const char*name)2419{2420 alternate_index_output = name;2421}24222423static intcommit_locked_index(struct lock_file *lk)2424{2425if(alternate_index_output)2426returncommit_lock_file_to(lk, alternate_index_output);2427else2428returncommit_lock_file(lk);2429}24302431static intdo_write_locked_index(struct index_state *istate,struct lock_file *lock,2432unsigned flags)2433{2434int ret =do_write_index(istate, lock->tempfile,0);2435if(ret)2436return ret;2437if(flags & COMMIT_LOCK)2438returncommit_locked_index(lock);2439returnclose_lock_file_gently(lock);2440}24412442static intwrite_split_index(struct index_state *istate,2443struct lock_file *lock,2444unsigned flags)2445{2446int ret;2447prepare_to_write_split_index(istate);2448 ret =do_write_locked_index(istate, lock, flags);2449finish_writing_split_index(istate);2450return ret;2451}24522453static const char*shared_index_expire ="2.weeks.ago";24542455static unsigned longget_shared_index_expire_date(void)2456{2457static unsigned long shared_index_expire_date;2458static int shared_index_expire_date_prepared;24592460if(!shared_index_expire_date_prepared) {2461git_config_get_expiry("splitindex.sharedindexexpire",2462&shared_index_expire);2463 shared_index_expire_date =approxidate(shared_index_expire);2464 shared_index_expire_date_prepared =1;2465}24662467return shared_index_expire_date;2468}24692470static intshould_delete_shared_index(const char*shared_index_path)2471{2472struct stat st;2473unsigned long expiration;24742475/* Check timestamp */2476 expiration =get_shared_index_expire_date();2477if(!expiration)2478return0;2479if(stat(shared_index_path, &st))2480returnerror_errno(_("could not stat '%s'"), shared_index_path);2481if(st.st_mtime > expiration)2482return0;24832484return1;2485}24862487static intclean_shared_index_files(const char*current_hex)2488{2489struct dirent *de;2490DIR*dir =opendir(get_git_dir());24912492if(!dir)2493returnerror_errno(_("unable to open git dir:%s"),get_git_dir());24942495while((de =readdir(dir)) != NULL) {2496const char*sha1_hex;2497const char*shared_index_path;2498if(!skip_prefix(de->d_name,"sharedindex.", &sha1_hex))2499continue;2500if(!strcmp(sha1_hex, current_hex))2501continue;2502 shared_index_path =git_path("%s", de->d_name);2503if(should_delete_shared_index(shared_index_path) >0&&2504unlink(shared_index_path))2505warning_errno(_("unable to unlink:%s"), shared_index_path);2506}2507closedir(dir);25082509return0;2510}25112512static intwrite_shared_index(struct index_state *istate,2513struct tempfile **temp)2514{2515struct split_index *si = istate->split_index;2516int ret;25172518move_cache_to_base_index(istate);2519 ret =do_write_index(si->base, *temp,1);2520if(ret)2521return ret;2522 ret =adjust_shared_perm(get_tempfile_path(*temp));2523if(ret) {2524error("cannot fix permission bits on%s",get_tempfile_path(*temp));2525return ret;2526}2527 ret =rename_tempfile(temp,2528git_path("sharedindex.%s",oid_to_hex(&si->base->oid)));2529if(!ret) {2530oidcpy(&si->base_oid, &si->base->oid);2531clean_shared_index_files(oid_to_hex(&si->base->oid));2532}25332534return ret;2535}25362537static const int default_max_percent_split_change =20;25382539static inttoo_many_not_shared_entries(struct index_state *istate)2540{2541int i, not_shared =0;2542int max_split =git_config_get_max_percent_split_change();25432544switch(max_split) {2545case-1:2546/* not or badly configured: use the default value */2547 max_split = default_max_percent_split_change;2548break;2549case0:2550return1;/* 0% means always write a new shared index */2551case100:2552return0;/* 100% means never write a new shared index */2553default:2554break;/* just use the configured value */2555}25562557/* Count not shared entries */2558for(i =0; i < istate->cache_nr; i++) {2559struct cache_entry *ce = istate->cache[i];2560if(!ce->index)2561 not_shared++;2562}25632564return(int64_t)istate->cache_nr * max_split < (int64_t)not_shared *100;2565}25662567intwrite_locked_index(struct index_state *istate,struct lock_file *lock,2568unsigned flags)2569{2570int new_shared_index, ret;2571struct split_index *si = istate->split_index;25722573if((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {2574if(flags & COMMIT_LOCK)2575rollback_lock_file(lock);2576return0;2577}25782579if(istate->fsmonitor_last_update)2580fill_fsmonitor_bitmap(istate);25812582if(!si || alternate_index_output ||2583(istate->cache_changed & ~EXTMASK)) {2584if(si)2585oidclr(&si->base_oid);2586 ret =do_write_locked_index(istate, lock, flags);2587goto out;2588}25892590if(git_env_bool("GIT_TEST_SPLIT_INDEX",0)) {2591int v = si->base_oid.hash[0];2592if((v &15) <6)2593 istate->cache_changed |= SPLIT_INDEX_ORDERED;2594}2595if(too_many_not_shared_entries(istate))2596 istate->cache_changed |= SPLIT_INDEX_ORDERED;25972598 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;25992600if(new_shared_index) {2601struct tempfile *temp;2602int saved_errno;26032604 temp =mks_tempfile(git_path("sharedindex_XXXXXX"));2605if(!temp) {2606oidclr(&si->base_oid);2607 ret =do_write_locked_index(istate, lock, flags);2608goto out;2609}2610 ret =write_shared_index(istate, &temp);26112612 saved_errno = errno;2613if(is_tempfile_active(temp))2614delete_tempfile(&temp);2615 errno = saved_errno;26162617if(ret)2618goto out;2619}26202621 ret =write_split_index(istate, lock, flags);26222623/* Freshen the shared index only if the split-index was written */2624if(!ret && !new_shared_index) {2625const char*shared_index =git_path("sharedindex.%s",2626oid_to_hex(&si->base_oid));2627freshen_shared_index(shared_index,1);2628}26292630out:2631if(flags & COMMIT_LOCK)2632rollback_lock_file(lock);2633return ret;2634}26352636/*2637 * Read the index file that is potentially unmerged into given2638 * index_state, dropping any unmerged entries. Returns true if2639 * the index is unmerged. Callers who want to refuse to work2640 * from an unmerged state can call this and check its return value,2641 * instead of calling read_cache().2642 */2643intread_index_unmerged(struct index_state *istate)2644{2645int i;2646int unmerged =0;26472648read_index(istate);2649for(i =0; i < istate->cache_nr; i++) {2650struct cache_entry *ce = istate->cache[i];2651struct cache_entry *new_ce;2652int size, len;26532654if(!ce_stage(ce))2655continue;2656 unmerged =1;2657 len =ce_namelen(ce);2658 size =cache_entry_size(len);2659 new_ce =xcalloc(1, size);2660memcpy(new_ce->name, ce->name, len);2661 new_ce->ce_flags =create_ce_flags(0) | CE_CONFLICTED;2662 new_ce->ce_namelen = len;2663 new_ce->ce_mode = ce->ce_mode;2664if(add_index_entry(istate, new_ce,0))2665returnerror("%s: cannot drop to stage #0",2666 new_ce->name);2667}2668return unmerged;2669}26702671/*2672 * Returns 1 if the path is an "other" path with respect to2673 * the index; that is, the path is not mentioned in the index at all,2674 * either as a file, a directory with some files in the index,2675 * or as an unmerged entry.2676 *2677 * We helpfully remove a trailing "/" from directories so that2678 * the output of read_directory can be used as-is.2679 */2680intindex_name_is_other(const struct index_state *istate,const char*name,2681int namelen)2682{2683int pos;2684if(namelen && name[namelen -1] =='/')2685 namelen--;2686 pos =index_name_pos(istate, name, namelen);2687if(0<= pos)2688return0;/* exact match */2689 pos = -pos -1;2690if(pos < istate->cache_nr) {2691struct cache_entry *ce = istate->cache[pos];2692if(ce_namelen(ce) == namelen &&2693!memcmp(ce->name, name, namelen))2694return0;/* Yup, this one exists unmerged */2695}2696return1;2697}26982699void*read_blob_data_from_index(const struct index_state *istate,2700const char*path,unsigned long*size)2701{2702int pos, len;2703unsigned long sz;2704enum object_type type;2705void*data;27062707 len =strlen(path);2708 pos =index_name_pos(istate, path, len);2709if(pos <0) {2710/*2711 * We might be in the middle of a merge, in which2712 * case we would read stage #2 (ours).2713 */2714int i;2715for(i = -pos -1;2716(pos <0&& i < istate->cache_nr &&2717!strcmp(istate->cache[i]->name, path));2718 i++)2719if(ce_stage(istate->cache[i]) ==2)2720 pos = i;2721}2722if(pos <0)2723return NULL;2724 data =read_object_file(&istate->cache[pos]->oid, &type, &sz);2725if(!data || type != OBJ_BLOB) {2726free(data);2727return NULL;2728}2729if(size)2730*size = sz;2731return data;2732}27332734voidstat_validity_clear(struct stat_validity *sv)2735{2736FREE_AND_NULL(sv->sd);2737}27382739intstat_validity_check(struct stat_validity *sv,const char*path)2740{2741struct stat st;27422743if(stat(path, &st) <0)2744return sv->sd == NULL;2745if(!sv->sd)2746return0;2747returnS_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);2748}27492750voidstat_validity_update(struct stat_validity *sv,int fd)2751{2752struct stat st;27532754if(fstat(fd, &st) <0|| !S_ISREG(st.st_mode))2755stat_validity_clear(sv);2756else{2757if(!sv->sd)2758 sv->sd =xcalloc(1,sizeof(struct stat_data));2759fill_stat_data(sv->sd, &st);2760}2761}27622763voidmove_index_extensions(struct index_state *dst,struct index_state *src)2764{2765 dst->untracked = src->untracked;2766 src->untracked = NULL;2767}