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 "diff.h" 10#include "diffcore.h" 11#include "tempfile.h" 12#include "lockfile.h" 13#include "cache-tree.h" 14#include "refs.h" 15#include "dir.h" 16#include "object-store.h" 17#include "tree.h" 18#include "commit.h" 19#include "blob.h" 20#include "resolve-undo.h" 21#include "strbuf.h" 22#include "varint.h" 23#include "split-index.h" 24#include "utf8.h" 25#include "fsmonitor.h" 26#include "thread-utils.h" 27#include "progress.h" 28 29/* Mask for the name length in ce_flags in the on-disk index */ 30 31#define CE_NAMEMASK (0x0fff) 32 33/* Index extensions. 34 * 35 * The first letter should be 'A'..'Z' for extensions that are not 36 * necessary for a correct operation (i.e. optimization data). 37 * When new extensions are added that _needs_ to be understood in 38 * order to correctly interpret the index file, pick character that 39 * is outside the range, to cause the reader to abort. 40 */ 41 42#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) ) 43#define CACHE_EXT_TREE 0x54524545 /* "TREE" */ 44#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */ 45#define CACHE_EXT_LINK 0x6c696e6b /* "link" */ 46#define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */ 47#define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */ 48#define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */ 49#define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */ 50 51/* changes that can be kept in $GIT_DIR/index (basically all extensions) */ 52#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \ 53 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \ 54 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED) 55 56 57/* 58 * This is an estimate of the pathname length in the index. We use 59 * this for V4 index files to guess the un-deltafied size of the index 60 * in memory because of pathname deltafication. This is not required 61 * for V2/V3 index formats because their pathnames are not compressed. 62 * If the initial amount of memory set aside is not sufficient, the 63 * mem pool will allocate extra memory. 64 */ 65#define CACHE_ENTRY_PATH_LENGTH 80 66 67static inline struct cache_entry *mem_pool__ce_alloc(struct mem_pool *mem_pool, size_t len) 68{ 69 struct cache_entry *ce; 70 ce = mem_pool_alloc(mem_pool, cache_entry_size(len)); 71 ce->mem_pool_allocated = 1; 72 return ce; 73} 74 75static inline struct cache_entry *mem_pool__ce_calloc(struct mem_pool *mem_pool, size_t len) 76{ 77 struct cache_entry * ce; 78 ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len)); 79 ce->mem_pool_allocated = 1; 80 return ce; 81} 82 83static struct mem_pool *find_mem_pool(struct index_state *istate) 84{ 85 struct mem_pool **pool_ptr; 86 87 if (istate->split_index && istate->split_index->base) 88 pool_ptr = &istate->split_index->base->ce_mem_pool; 89 else 90 pool_ptr = &istate->ce_mem_pool; 91 92 if (!*pool_ptr) 93 mem_pool_init(pool_ptr, 0); 94 95 return *pool_ptr; 96} 97 98struct index_state the_index; 99static const char *alternate_index_output; 100 101static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 102{ 103 istate->cache[nr] = ce; 104 add_name_hash(istate, ce); 105} 106 107static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce) 108{ 109 struct cache_entry *old = istate->cache[nr]; 110 111 replace_index_entry_in_base(istate, old, ce); 112 remove_name_hash(istate, old); 113 discard_cache_entry(old); 114 ce->ce_flags &= ~CE_HASHED; 115 set_index_entry(istate, nr, ce); 116 ce->ce_flags |= CE_UPDATE_IN_BASE; 117 mark_fsmonitor_invalid(istate, ce); 118 istate->cache_changed |= CE_ENTRY_CHANGED; 119} 120 121void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name) 122{ 123 struct cache_entry *old_entry = istate->cache[nr], *new_entry; 124 int namelen = strlen(new_name); 125 126 new_entry = make_empty_cache_entry(istate, namelen); 127 copy_cache_entry(new_entry, old_entry); 128 new_entry->ce_flags &= ~CE_HASHED; 129 new_entry->ce_namelen = namelen; 130 new_entry->index = 0; 131 memcpy(new_entry->name, new_name, namelen + 1); 132 133 cache_tree_invalidate_path(istate, old_entry->name); 134 untracked_cache_remove_from_index(istate, old_entry->name); 135 remove_index_entry_at(istate, nr); 136 add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE); 137} 138 139void fill_stat_data(struct stat_data *sd, struct stat *st) 140{ 141 sd->sd_ctime.sec = (unsigned int)st->st_ctime; 142 sd->sd_mtime.sec = (unsigned int)st->st_mtime; 143 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st); 144 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st); 145 sd->sd_dev = st->st_dev; 146 sd->sd_ino = st->st_ino; 147 sd->sd_uid = st->st_uid; 148 sd->sd_gid = st->st_gid; 149 sd->sd_size = st->st_size; 150} 151 152int match_stat_data(const struct stat_data *sd, struct stat *st) 153{ 154 int changed = 0; 155 156 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime) 157 changed |= MTIME_CHANGED; 158 if (trust_ctime && check_stat && 159 sd->sd_ctime.sec != (unsigned int)st->st_ctime) 160 changed |= CTIME_CHANGED; 161 162#ifdef USE_NSEC 163 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st)) 164 changed |= MTIME_CHANGED; 165 if (trust_ctime && check_stat && 166 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st)) 167 changed |= CTIME_CHANGED; 168#endif 169 170 if (check_stat) { 171 if (sd->sd_uid != (unsigned int) st->st_uid || 172 sd->sd_gid != (unsigned int) st->st_gid) 173 changed |= OWNER_CHANGED; 174 if (sd->sd_ino != (unsigned int) st->st_ino) 175 changed |= INODE_CHANGED; 176 } 177 178#ifdef USE_STDEV 179 /* 180 * st_dev breaks on network filesystems where different 181 * clients will have different views of what "device" 182 * the filesystem is on 183 */ 184 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev) 185 changed |= INODE_CHANGED; 186#endif 187 188 if (sd->sd_size != (unsigned int) st->st_size) 189 changed |= DATA_CHANGED; 190 191 return changed; 192} 193 194/* 195 * This only updates the "non-critical" parts of the directory 196 * cache, ie the parts that aren't tracked by GIT, and only used 197 * to validate the cache. 198 */ 199void fill_stat_cache_info(struct cache_entry *ce, struct stat *st) 200{ 201 fill_stat_data(&ce->ce_stat_data, st); 202 203 if (assume_unchanged) 204 ce->ce_flags |= CE_VALID; 205 206 if (S_ISREG(st->st_mode)) { 207 ce_mark_uptodate(ce); 208 mark_fsmonitor_valid(ce); 209 } 210} 211 212static int ce_compare_data(struct index_state *istate, 213 const struct cache_entry *ce, 214 struct stat *st) 215{ 216 int match = -1; 217 int fd = git_open_cloexec(ce->name, O_RDONLY); 218 219 if (fd >= 0) { 220 struct object_id oid; 221 if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0)) 222 match = !oideq(&oid, &ce->oid); 223 /* index_fd() closed the file descriptor already */ 224 } 225 return match; 226} 227 228static int ce_compare_link(const struct cache_entry *ce, size_t expected_size) 229{ 230 int match = -1; 231 void *buffer; 232 unsigned long size; 233 enum object_type type; 234 struct strbuf sb = STRBUF_INIT; 235 236 if (strbuf_readlink(&sb, ce->name, expected_size)) 237 return -1; 238 239 buffer = read_object_file(&ce->oid, &type, &size); 240 if (buffer) { 241 if (size == sb.len) 242 match = memcmp(buffer, sb.buf, size); 243 free(buffer); 244 } 245 strbuf_release(&sb); 246 return match; 247} 248 249static int ce_compare_gitlink(const struct cache_entry *ce) 250{ 251 struct object_id oid; 252 253 /* 254 * We don't actually require that the .git directory 255 * under GITLINK directory be a valid git directory. It 256 * might even be missing (in case nobody populated that 257 * sub-project). 258 * 259 * If so, we consider it always to match. 260 */ 261 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0) 262 return 0; 263 return !oideq(&oid, &ce->oid); 264} 265 266static int ce_modified_check_fs(struct index_state *istate, 267 const struct cache_entry *ce, 268 struct stat *st) 269{ 270 switch (st->st_mode & S_IFMT) { 271 case S_IFREG: 272 if (ce_compare_data(istate, ce, st)) 273 return DATA_CHANGED; 274 break; 275 case S_IFLNK: 276 if (ce_compare_link(ce, xsize_t(st->st_size))) 277 return DATA_CHANGED; 278 break; 279 case S_IFDIR: 280 if (S_ISGITLINK(ce->ce_mode)) 281 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0; 282 /* else fallthrough */ 283 default: 284 return TYPE_CHANGED; 285 } 286 return 0; 287} 288 289static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st) 290{ 291 unsigned int changed = 0; 292 293 if (ce->ce_flags & CE_REMOVE) 294 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED; 295 296 switch (ce->ce_mode & S_IFMT) { 297 case S_IFREG: 298 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0; 299 /* We consider only the owner x bit to be relevant for 300 * "mode changes" 301 */ 302 if (trust_executable_bit && 303 (0100 & (ce->ce_mode ^ st->st_mode))) 304 changed |= MODE_CHANGED; 305 break; 306 case S_IFLNK: 307 if (!S_ISLNK(st->st_mode) && 308 (has_symlinks || !S_ISREG(st->st_mode))) 309 changed |= TYPE_CHANGED; 310 break; 311 case S_IFGITLINK: 312 /* We ignore most of the st_xxx fields for gitlinks */ 313 if (!S_ISDIR(st->st_mode)) 314 changed |= TYPE_CHANGED; 315 else if (ce_compare_gitlink(ce)) 316 changed |= DATA_CHANGED; 317 return changed; 318 default: 319 BUG("unsupported ce_mode: %o", ce->ce_mode); 320 } 321 322 changed |= match_stat_data(&ce->ce_stat_data, st); 323 324 /* Racily smudged entry? */ 325 if (!ce->ce_stat_data.sd_size) { 326 if (!is_empty_blob_sha1(ce->oid.hash)) 327 changed |= DATA_CHANGED; 328 } 329 330 return changed; 331} 332 333static int is_racy_stat(const struct index_state *istate, 334 const struct stat_data *sd) 335{ 336 return (istate->timestamp.sec && 337#ifdef USE_NSEC 338 /* nanosecond timestamped files can also be racy! */ 339 (istate->timestamp.sec < sd->sd_mtime.sec || 340 (istate->timestamp.sec == sd->sd_mtime.sec && 341 istate->timestamp.nsec <= sd->sd_mtime.nsec)) 342#else 343 istate->timestamp.sec <= sd->sd_mtime.sec 344#endif 345 ); 346} 347 348int is_racy_timestamp(const struct index_state *istate, 349 const struct cache_entry *ce) 350{ 351 return (!S_ISGITLINK(ce->ce_mode) && 352 is_racy_stat(istate, &ce->ce_stat_data)); 353} 354 355int match_stat_data_racy(const struct index_state *istate, 356 const struct stat_data *sd, struct stat *st) 357{ 358 if (is_racy_stat(istate, sd)) 359 return MTIME_CHANGED; 360 return match_stat_data(sd, st); 361} 362 363int ie_match_stat(struct index_state *istate, 364 const struct cache_entry *ce, struct stat *st, 365 unsigned int options) 366{ 367 unsigned int changed; 368 int ignore_valid = options & CE_MATCH_IGNORE_VALID; 369 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE; 370 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY; 371 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR; 372 373 if (!ignore_fsmonitor) 374 refresh_fsmonitor(istate); 375 /* 376 * If it's marked as always valid in the index, it's 377 * valid whatever the checked-out copy says. 378 * 379 * skip-worktree has the same effect with higher precedence 380 */ 381 if (!ignore_skip_worktree && ce_skip_worktree(ce)) 382 return 0; 383 if (!ignore_valid && (ce->ce_flags & CE_VALID)) 384 return 0; 385 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) 386 return 0; 387 388 /* 389 * Intent-to-add entries have not been added, so the index entry 390 * by definition never matches what is in the work tree until it 391 * actually gets added. 392 */ 393 if (ce_intent_to_add(ce)) 394 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED; 395 396 changed = ce_match_stat_basic(ce, st); 397 398 /* 399 * Within 1 second of this sequence: 400 * echo xyzzy >file && git-update-index --add file 401 * running this command: 402 * echo frotz >file 403 * would give a falsely clean cache entry. The mtime and 404 * length match the cache, and other stat fields do not change. 405 * 406 * We could detect this at update-index time (the cache entry 407 * being registered/updated records the same time as "now") 408 * and delay the return from git-update-index, but that would 409 * effectively mean we can make at most one commit per second, 410 * which is not acceptable. Instead, we check cache entries 411 * whose mtime are the same as the index file timestamp more 412 * carefully than others. 413 */ 414 if (!changed && is_racy_timestamp(istate, ce)) { 415 if (assume_racy_is_modified) 416 changed |= DATA_CHANGED; 417 else 418 changed |= ce_modified_check_fs(istate, ce, st); 419 } 420 421 return changed; 422} 423 424int ie_modified(struct index_state *istate, 425 const struct cache_entry *ce, 426 struct stat *st, unsigned int options) 427{ 428 int changed, changed_fs; 429 430 changed = ie_match_stat(istate, ce, st, options); 431 if (!changed) 432 return 0; 433 /* 434 * If the mode or type has changed, there's no point in trying 435 * to refresh the entry - it's not going to match 436 */ 437 if (changed & (MODE_CHANGED | TYPE_CHANGED)) 438 return changed; 439 440 /* 441 * Immediately after read-tree or update-index --cacheinfo, 442 * the length field is zero, as we have never even read the 443 * lstat(2) information once, and we cannot trust DATA_CHANGED 444 * returned by ie_match_stat() which in turn was returned by 445 * ce_match_stat_basic() to signal that the filesize of the 446 * blob changed. We have to actually go to the filesystem to 447 * see if the contents match, and if so, should answer "unchanged". 448 * 449 * The logic does not apply to gitlinks, as ce_match_stat_basic() 450 * already has checked the actual HEAD from the filesystem in the 451 * subproject. If ie_match_stat() already said it is different, 452 * then we know it is. 453 */ 454 if ((changed & DATA_CHANGED) && 455 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0)) 456 return changed; 457 458 changed_fs = ce_modified_check_fs(istate, ce, st); 459 if (changed_fs) 460 return changed | changed_fs; 461 return 0; 462} 463 464int base_name_compare(const char *name1, int len1, int mode1, 465 const char *name2, int len2, int mode2) 466{ 467 unsigned char c1, c2; 468 int len = len1 < len2 ? len1 : len2; 469 int cmp; 470 471 cmp = memcmp(name1, name2, len); 472 if (cmp) 473 return cmp; 474 c1 = name1[len]; 475 c2 = name2[len]; 476 if (!c1 && S_ISDIR(mode1)) 477 c1 = '/'; 478 if (!c2 && S_ISDIR(mode2)) 479 c2 = '/'; 480 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0; 481} 482 483/* 484 * df_name_compare() is identical to base_name_compare(), except it 485 * compares conflicting directory/file entries as equal. Note that 486 * while a directory name compares as equal to a regular file, they 487 * then individually compare _differently_ to a filename that has 488 * a dot after the basename (because '\0' < '.' < '/'). 489 * 490 * This is used by routines that want to traverse the git namespace 491 * but then handle conflicting entries together when possible. 492 */ 493int df_name_compare(const char *name1, int len1, int mode1, 494 const char *name2, int len2, int mode2) 495{ 496 int len = len1 < len2 ? len1 : len2, cmp; 497 unsigned char c1, c2; 498 499 cmp = memcmp(name1, name2, len); 500 if (cmp) 501 return cmp; 502 /* Directories and files compare equal (same length, same name) */ 503 if (len1 == len2) 504 return 0; 505 c1 = name1[len]; 506 if (!c1 && S_ISDIR(mode1)) 507 c1 = '/'; 508 c2 = name2[len]; 509 if (!c2 && S_ISDIR(mode2)) 510 c2 = '/'; 511 if (c1 == '/' && !c2) 512 return 0; 513 if (c2 == '/' && !c1) 514 return 0; 515 return c1 - c2; 516} 517 518int name_compare(const char *name1, size_t len1, const char *name2, size_t len2) 519{ 520 size_t min_len = (len1 < len2) ? len1 : len2; 521 int cmp = memcmp(name1, name2, min_len); 522 if (cmp) 523 return cmp; 524 if (len1 < len2) 525 return -1; 526 if (len1 > len2) 527 return 1; 528 return 0; 529} 530 531int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2) 532{ 533 int cmp; 534 535 cmp = name_compare(name1, len1, name2, len2); 536 if (cmp) 537 return cmp; 538 539 if (stage1 < stage2) 540 return -1; 541 if (stage1 > stage2) 542 return 1; 543 return 0; 544} 545 546static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage) 547{ 548 int first, last; 549 550 first = 0; 551 last = istate->cache_nr; 552 while (last > first) { 553 int next = (last + first) >> 1; 554 struct cache_entry *ce = istate->cache[next]; 555 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce)); 556 if (!cmp) 557 return next; 558 if (cmp < 0) { 559 last = next; 560 continue; 561 } 562 first = next+1; 563 } 564 return -first-1; 565} 566 567int index_name_pos(const struct index_state *istate, const char *name, int namelen) 568{ 569 return index_name_stage_pos(istate, name, namelen, 0); 570} 571 572int remove_index_entry_at(struct index_state *istate, int pos) 573{ 574 struct cache_entry *ce = istate->cache[pos]; 575 576 record_resolve_undo(istate, ce); 577 remove_name_hash(istate, ce); 578 save_or_free_index_entry(istate, ce); 579 istate->cache_changed |= CE_ENTRY_REMOVED; 580 istate->cache_nr--; 581 if (pos >= istate->cache_nr) 582 return 0; 583 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1, 584 istate->cache_nr - pos); 585 return 1; 586} 587 588/* 589 * Remove all cache entries marked for removal, that is where 590 * CE_REMOVE is set in ce_flags. This is much more effective than 591 * calling remove_index_entry_at() for each entry to be removed. 592 */ 593void remove_marked_cache_entries(struct index_state *istate) 594{ 595 struct cache_entry **ce_array = istate->cache; 596 unsigned int i, j; 597 598 for (i = j = 0; i < istate->cache_nr; i++) { 599 if (ce_array[i]->ce_flags & CE_REMOVE) { 600 remove_name_hash(istate, ce_array[i]); 601 save_or_free_index_entry(istate, ce_array[i]); 602 } 603 else 604 ce_array[j++] = ce_array[i]; 605 } 606 if (j == istate->cache_nr) 607 return; 608 istate->cache_changed |= CE_ENTRY_REMOVED; 609 istate->cache_nr = j; 610} 611 612int remove_file_from_index(struct index_state *istate, const char *path) 613{ 614 int pos = index_name_pos(istate, path, strlen(path)); 615 if (pos < 0) 616 pos = -pos-1; 617 cache_tree_invalidate_path(istate, path); 618 untracked_cache_remove_from_index(istate, path); 619 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path)) 620 remove_index_entry_at(istate, pos); 621 return 0; 622} 623 624static int compare_name(struct cache_entry *ce, const char *path, int namelen) 625{ 626 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen); 627} 628 629static int index_name_pos_also_unmerged(struct index_state *istate, 630 const char *path, int namelen) 631{ 632 int pos = index_name_pos(istate, path, namelen); 633 struct cache_entry *ce; 634 635 if (pos >= 0) 636 return pos; 637 638 /* maybe unmerged? */ 639 pos = -1 - pos; 640 if (pos >= istate->cache_nr || 641 compare_name((ce = istate->cache[pos]), path, namelen)) 642 return -1; 643 644 /* order of preference: stage 2, 1, 3 */ 645 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr && 646 ce_stage((ce = istate->cache[pos + 1])) == 2 && 647 !compare_name(ce, path, namelen)) 648 pos++; 649 return pos; 650} 651 652static int different_name(struct cache_entry *ce, struct cache_entry *alias) 653{ 654 int len = ce_namelen(ce); 655 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len); 656} 657 658/* 659 * If we add a filename that aliases in the cache, we will use the 660 * name that we already have - but we don't want to update the same 661 * alias twice, because that implies that there were actually two 662 * different files with aliasing names! 663 * 664 * So we use the CE_ADDED flag to verify that the alias was an old 665 * one before we accept it as 666 */ 667static struct cache_entry *create_alias_ce(struct index_state *istate, 668 struct cache_entry *ce, 669 struct cache_entry *alias) 670{ 671 int len; 672 struct cache_entry *new_entry; 673 674 if (alias->ce_flags & CE_ADDED) 675 die(_("will not add file alias '%s' ('%s' already exists in index)"), 676 ce->name, alias->name); 677 678 /* Ok, create the new entry using the name of the existing alias */ 679 len = ce_namelen(alias); 680 new_entry = make_empty_cache_entry(istate, len); 681 memcpy(new_entry->name, alias->name, len); 682 copy_cache_entry(new_entry, ce); 683 save_or_free_index_entry(istate, ce); 684 return new_entry; 685} 686 687void set_object_name_for_intent_to_add_entry(struct cache_entry *ce) 688{ 689 struct object_id oid; 690 if (write_object_file("", 0, blob_type, &oid)) 691 die(_("cannot create an empty blob in the object database")); 692 oidcpy(&ce->oid, &oid); 693} 694 695int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags) 696{ 697 int namelen, was_same; 698 mode_t st_mode = st->st_mode; 699 struct cache_entry *ce, *alias = NULL; 700 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY; 701 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND); 702 int pretend = flags & ADD_CACHE_PRETEND; 703 int intent_only = flags & ADD_CACHE_INTENT; 704 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE| 705 (intent_only ? ADD_CACHE_NEW_ONLY : 0)); 706 int newflags = HASH_WRITE_OBJECT; 707 708 if (flags & HASH_RENORMALIZE) 709 newflags |= HASH_RENORMALIZE; 710 711 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode)) 712 return error(_("%s: can only add regular files, symbolic links or git-directories"), path); 713 714 namelen = strlen(path); 715 if (S_ISDIR(st_mode)) { 716 while (namelen && path[namelen-1] == '/') 717 namelen--; 718 } 719 ce = make_empty_cache_entry(istate, namelen); 720 memcpy(ce->name, path, namelen); 721 ce->ce_namelen = namelen; 722 if (!intent_only) 723 fill_stat_cache_info(ce, st); 724 else 725 ce->ce_flags |= CE_INTENT_TO_ADD; 726 727 728 if (trust_executable_bit && has_symlinks) { 729 ce->ce_mode = create_ce_mode(st_mode); 730 } else { 731 /* If there is an existing entry, pick the mode bits and type 732 * from it, otherwise assume unexecutable regular file. 733 */ 734 struct cache_entry *ent; 735 int pos = index_name_pos_also_unmerged(istate, path, namelen); 736 737 ent = (0 <= pos) ? istate->cache[pos] : NULL; 738 ce->ce_mode = ce_mode_from_stat(ent, st_mode); 739 } 740 741 /* When core.ignorecase=true, determine if a directory of the same name but differing 742 * case already exists within the Git repository. If it does, ensure the directory 743 * case of the file being added to the repository matches (is folded into) the existing 744 * entry's directory case. 745 */ 746 if (ignore_case) { 747 adjust_dirname_case(istate, ce->name); 748 } 749 if (!(flags & HASH_RENORMALIZE)) { 750 alias = index_file_exists(istate, ce->name, 751 ce_namelen(ce), ignore_case); 752 if (alias && 753 !ce_stage(alias) && 754 !ie_match_stat(istate, alias, st, ce_option)) { 755 /* Nothing changed, really */ 756 if (!S_ISGITLINK(alias->ce_mode)) 757 ce_mark_uptodate(alias); 758 alias->ce_flags |= CE_ADDED; 759 760 discard_cache_entry(ce); 761 return 0; 762 } 763 } 764 if (!intent_only) { 765 if (index_path(istate, &ce->oid, path, st, newflags)) { 766 discard_cache_entry(ce); 767 return error(_("unable to index file '%s'"), path); 768 } 769 } else 770 set_object_name_for_intent_to_add_entry(ce); 771 772 if (ignore_case && alias && different_name(ce, alias)) 773 ce = create_alias_ce(istate, ce, alias); 774 ce->ce_flags |= CE_ADDED; 775 776 /* It was suspected to be racily clean, but it turns out to be Ok */ 777 was_same = (alias && 778 !ce_stage(alias) && 779 oideq(&alias->oid, &ce->oid) && 780 ce->ce_mode == alias->ce_mode); 781 782 if (pretend) 783 discard_cache_entry(ce); 784 else if (add_index_entry(istate, ce, add_option)) { 785 discard_cache_entry(ce); 786 return error(_("unable to add '%s' to index"), path); 787 } 788 if (verbose && !was_same) 789 printf("add '%s'\n", path); 790 return 0; 791} 792 793int add_file_to_index(struct index_state *istate, const char *path, int flags) 794{ 795 struct stat st; 796 if (lstat(path, &st)) 797 die_errno(_("unable to stat '%s'"), path); 798 return add_to_index(istate, path, &st, flags); 799} 800 801struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len) 802{ 803 return mem_pool__ce_calloc(find_mem_pool(istate), len); 804} 805 806struct cache_entry *make_empty_transient_cache_entry(size_t len) 807{ 808 return xcalloc(1, cache_entry_size(len)); 809} 810 811struct cache_entry *make_cache_entry(struct index_state *istate, 812 unsigned int mode, 813 const struct object_id *oid, 814 const char *path, 815 int stage, 816 unsigned int refresh_options) 817{ 818 struct cache_entry *ce, *ret; 819 int len; 820 821 if (!verify_path(path, mode)) { 822 error(_("invalid path '%s'"), path); 823 return NULL; 824 } 825 826 len = strlen(path); 827 ce = make_empty_cache_entry(istate, len); 828 829 oidcpy(&ce->oid, oid); 830 memcpy(ce->name, path, len); 831 ce->ce_flags = create_ce_flags(stage); 832 ce->ce_namelen = len; 833 ce->ce_mode = create_ce_mode(mode); 834 835 ret = refresh_cache_entry(istate, ce, refresh_options); 836 if (ret != ce) 837 discard_cache_entry(ce); 838 return ret; 839} 840 841struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid, 842 const char *path, int stage) 843{ 844 struct cache_entry *ce; 845 int len; 846 847 if (!verify_path(path, mode)) { 848 error(_("invalid path '%s'"), path); 849 return NULL; 850 } 851 852 len = strlen(path); 853 ce = make_empty_transient_cache_entry(len); 854 855 oidcpy(&ce->oid, oid); 856 memcpy(ce->name, path, len); 857 ce->ce_flags = create_ce_flags(stage); 858 ce->ce_namelen = len; 859 ce->ce_mode = create_ce_mode(mode); 860 861 return ce; 862} 863 864/* 865 * Chmod an index entry with either +x or -x. 866 * 867 * Returns -1 if the chmod for the particular cache entry failed (if it's 868 * not a regular file), -2 if an invalid flip argument is passed in, 0 869 * otherwise. 870 */ 871int chmod_index_entry(struct index_state *istate, struct cache_entry *ce, 872 char flip) 873{ 874 if (!S_ISREG(ce->ce_mode)) 875 return -1; 876 switch (flip) { 877 case '+': 878 ce->ce_mode |= 0111; 879 break; 880 case '-': 881 ce->ce_mode &= ~0111; 882 break; 883 default: 884 return -2; 885 } 886 cache_tree_invalidate_path(istate, ce->name); 887 ce->ce_flags |= CE_UPDATE_IN_BASE; 888 mark_fsmonitor_invalid(istate, ce); 889 istate->cache_changed |= CE_ENTRY_CHANGED; 890 891 return 0; 892} 893 894int ce_same_name(const struct cache_entry *a, const struct cache_entry *b) 895{ 896 int len = ce_namelen(a); 897 return ce_namelen(b) == len && !memcmp(a->name, b->name, len); 898} 899 900/* 901 * We fundamentally don't like some paths: we don't want 902 * dot or dot-dot anywhere, and for obvious reasons don't 903 * want to recurse into ".git" either. 904 * 905 * Also, we don't want double slashes or slashes at the 906 * end that can make pathnames ambiguous. 907 */ 908static int verify_dotfile(const char *rest, unsigned mode) 909{ 910 /* 911 * The first character was '.', but that 912 * has already been discarded, we now test 913 * the rest. 914 */ 915 916 /* "." is not allowed */ 917 if (*rest == '\0' || is_dir_sep(*rest)) 918 return 0; 919 920 switch (*rest) { 921 /* 922 * ".git" followed by NUL or slash is bad. Note that we match 923 * case-insensitively here, even if ignore_case is not set. 924 * This outlaws ".GIT" everywhere out of an abundance of caution, 925 * since there's really no good reason to allow it. 926 * 927 * Once we've seen ".git", we can also find ".gitmodules", etc (also 928 * case-insensitively). 929 */ 930 case 'g': 931 case 'G': 932 if (rest[1] != 'i' && rest[1] != 'I') 933 break; 934 if (rest[2] != 't' && rest[2] != 'T') 935 break; 936 if (rest[3] == '\0' || is_dir_sep(rest[3])) 937 return 0; 938 if (S_ISLNK(mode)) { 939 rest += 3; 940 if (skip_iprefix(rest, "modules", &rest) && 941 (*rest == '\0' || is_dir_sep(*rest))) 942 return 0; 943 } 944 break; 945 case '.': 946 if (rest[1] == '\0' || is_dir_sep(rest[1])) 947 return 0; 948 } 949 return 1; 950} 951 952int verify_path(const char *path, unsigned mode) 953{ 954 char c; 955 956 if (has_dos_drive_prefix(path)) 957 return 0; 958 959 goto inside; 960 for (;;) { 961 if (!c) 962 return 1; 963 if (is_dir_sep(c)) { 964inside: 965 if (protect_hfs) { 966 if (is_hfs_dotgit(path)) 967 return 0; 968 if (S_ISLNK(mode)) { 969 if (is_hfs_dotgitmodules(path)) 970 return 0; 971 } 972 } 973 if (protect_ntfs) { 974 if (is_ntfs_dotgit(path)) 975 return 0; 976 if (S_ISLNK(mode)) { 977 if (is_ntfs_dotgitmodules(path)) 978 return 0; 979 } 980 } 981 982 c = *path++; 983 if ((c == '.' && !verify_dotfile(path, mode)) || 984 is_dir_sep(c) || c == '\0') 985 return 0; 986 } 987 c = *path++; 988 } 989} 990 991/* 992 * Do we have another file that has the beginning components being a 993 * proper superset of the name we're trying to add? 994 */ 995static int has_file_name(struct index_state *istate, 996 const struct cache_entry *ce, int pos, int ok_to_replace) 997{ 998 int retval = 0; 999 int len = ce_namelen(ce);1000 int stage = ce_stage(ce);1001 const char *name = ce->name;10021003 while (pos < istate->cache_nr) {1004 struct cache_entry *p = istate->cache[pos++];10051006 if (len >= ce_namelen(p))1007 break;1008 if (memcmp(name, p->name, len))1009 break;1010 if (ce_stage(p) != stage)1011 continue;1012 if (p->name[len] != '/')1013 continue;1014 if (p->ce_flags & CE_REMOVE)1015 continue;1016 retval = -1;1017 if (!ok_to_replace)1018 break;1019 remove_index_entry_at(istate, --pos);1020 }1021 return retval;1022}102310241025/*1026 * Like strcmp(), but also return the offset of the first change.1027 * If strings are equal, return the length.1028 */1029int strcmp_offset(const char *s1, const char *s2, size_t *first_change)1030{1031 size_t k;10321033 if (!first_change)1034 return strcmp(s1, s2);10351036 for (k = 0; s1[k] == s2[k]; k++)1037 if (s1[k] == '\0')1038 break;10391040 *first_change = k;1041 return (unsigned char)s1[k] - (unsigned char)s2[k];1042}10431044/*1045 * Do we have another file with a pathname that is a proper1046 * subset of the name we're trying to add?1047 *1048 * That is, is there another file in the index with a path1049 * that matches a sub-directory in the given entry?1050 */1051static int has_dir_name(struct index_state *istate,1052 const struct cache_entry *ce, int pos, int ok_to_replace)1053{1054 int retval = 0;1055 int stage = ce_stage(ce);1056 const char *name = ce->name;1057 const char *slash = name + ce_namelen(ce);1058 size_t len_eq_last;1059 int cmp_last = 0;10601061 /*1062 * We are frequently called during an iteration on a sorted1063 * list of pathnames and while building a new index. Therefore,1064 * there is a high probability that this entry will eventually1065 * be appended to the index, rather than inserted in the middle.1066 * If we can confirm that, we can avoid binary searches on the1067 * components of the pathname.1068 *1069 * Compare the entry's full path with the last path in the index.1070 */1071 if (istate->cache_nr > 0) {1072 cmp_last = strcmp_offset(name,1073 istate->cache[istate->cache_nr - 1]->name,1074 &len_eq_last);1075 if (cmp_last > 0) {1076 if (len_eq_last == 0) {1077 /*1078 * The entry sorts AFTER the last one in the1079 * index and their paths have no common prefix,1080 * so there cannot be a F/D conflict.1081 */1082 return retval;1083 } else {1084 /*1085 * The entry sorts AFTER the last one in the1086 * index, but has a common prefix. Fall through1087 * to the loop below to disect the entry's path1088 * and see where the difference is.1089 */1090 }1091 } else if (cmp_last == 0) {1092 /*1093 * The entry exactly matches the last one in the1094 * index, but because of multiple stage and CE_REMOVE1095 * items, we fall through and let the regular search1096 * code handle it.1097 */1098 }1099 }11001101 for (;;) {1102 size_t len;11031104 for (;;) {1105 if (*--slash == '/')1106 break;1107 if (slash <= ce->name)1108 return retval;1109 }1110 len = slash - name;11111112 if (cmp_last > 0) {1113 /*1114 * (len + 1) is a directory boundary (including1115 * the trailing slash). And since the loop is1116 * decrementing "slash", the first iteration is1117 * the longest directory prefix; subsequent1118 * iterations consider parent directories.1119 */11201121 if (len + 1 <= len_eq_last) {1122 /*1123 * The directory prefix (including the trailing1124 * slash) also appears as a prefix in the last1125 * entry, so the remainder cannot collide (because1126 * strcmp said the whole path was greater).1127 *1128 * EQ: last: xxx/A1129 * this: xxx/B1130 *1131 * LT: last: xxx/file_A1132 * this: xxx/file_B1133 */1134 return retval;1135 }11361137 if (len > len_eq_last) {1138 /*1139 * This part of the directory prefix (excluding1140 * the trailing slash) is longer than the known1141 * equal portions, so this sub-directory cannot1142 * collide with a file.1143 *1144 * GT: last: xxxA1145 * this: xxxB/file1146 */1147 return retval;1148 }11491150 if (istate->cache_nr > 0 &&1151 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {1152 /*1153 * The directory prefix lines up with part of1154 * a longer file or directory name, but sorts1155 * after it, so this sub-directory cannot1156 * collide with a file.1157 *1158 * last: xxx/yy-file (because '-' sorts before '/')1159 * this: xxx/yy/abc1160 */1161 return retval;1162 }11631164 /*1165 * This is a possible collision. Fall through and1166 * let the regular search code handle it.1167 *1168 * last: xxx1169 * this: xxx/file1170 */1171 }11721173 pos = index_name_stage_pos(istate, name, len, stage);1174 if (pos >= 0) {1175 /*1176 * Found one, but not so fast. This could1177 * be a marker that says "I was here, but1178 * I am being removed". Such an entry is1179 * not a part of the resulting tree, and1180 * it is Ok to have a directory at the same1181 * path.1182 */1183 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {1184 retval = -1;1185 if (!ok_to_replace)1186 break;1187 remove_index_entry_at(istate, pos);1188 continue;1189 }1190 }1191 else1192 pos = -pos-1;11931194 /*1195 * Trivial optimization: if we find an entry that1196 * already matches the sub-directory, then we know1197 * we're ok, and we can exit.1198 */1199 while (pos < istate->cache_nr) {1200 struct cache_entry *p = istate->cache[pos];1201 if ((ce_namelen(p) <= len) ||1202 (p->name[len] != '/') ||1203 memcmp(p->name, name, len))1204 break; /* not our subdirectory */1205 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))1206 /*1207 * p is at the same stage as our entry, and1208 * is a subdirectory of what we are looking1209 * at, so we cannot have conflicts at our1210 * level or anything shorter.1211 */1212 return retval;1213 pos++;1214 }1215 }1216 return retval;1217}12181219/* We may be in a situation where we already have path/file and path1220 * is being added, or we already have path and path/file is being1221 * added. Either one would result in a nonsense tree that has path1222 * twice when git-write-tree tries to write it out. Prevent it.1223 *1224 * If ok-to-replace is specified, we remove the conflicting entries1225 * from the cache so the caller should recompute the insert position.1226 * When this happens, we return non-zero.1227 */1228static int check_file_directory_conflict(struct index_state *istate,1229 const struct cache_entry *ce,1230 int pos, int ok_to_replace)1231{1232 int retval;12331234 /*1235 * When ce is an "I am going away" entry, we allow it to be added1236 */1237 if (ce->ce_flags & CE_REMOVE)1238 return 0;12391240 /*1241 * We check if the path is a sub-path of a subsequent pathname1242 * first, since removing those will not change the position1243 * in the array.1244 */1245 retval = has_file_name(istate, ce, pos, ok_to_replace);12461247 /*1248 * Then check if the path might have a clashing sub-directory1249 * before it.1250 */1251 return retval + has_dir_name(istate, ce, pos, ok_to_replace);1252}12531254static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)1255{1256 int pos;1257 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;1258 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;1259 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;1260 int new_only = option & ADD_CACHE_NEW_ONLY;12611262 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))1263 cache_tree_invalidate_path(istate, ce->name);12641265 /*1266 * If this entry's path sorts after the last entry in the index,1267 * we can avoid searching for it.1268 */1269 if (istate->cache_nr > 0 &&1270 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)1271 pos = -istate->cache_nr - 1;1272 else1273 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));12741275 /* existing match? Just replace it. */1276 if (pos >= 0) {1277 if (!new_only)1278 replace_index_entry(istate, pos, ce);1279 return 0;1280 }1281 pos = -pos-1;12821283 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))1284 untracked_cache_add_to_index(istate, ce->name);12851286 /*1287 * Inserting a merged entry ("stage 0") into the index1288 * will always replace all non-merged entries..1289 */1290 if (pos < istate->cache_nr && ce_stage(ce) == 0) {1291 while (ce_same_name(istate->cache[pos], ce)) {1292 ok_to_add = 1;1293 if (!remove_index_entry_at(istate, pos))1294 break;1295 }1296 }12971298 if (!ok_to_add)1299 return -1;1300 if (!verify_path(ce->name, ce->ce_mode))1301 return error(_("invalid path '%s'"), ce->name);13021303 if (!skip_df_check &&1304 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {1305 if (!ok_to_replace)1306 return error(_("'%s' appears as both a file and as a directory"),1307 ce->name);1308 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));1309 pos = -pos-1;1310 }1311 return pos + 1;1312}13131314int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)1315{1316 int pos;13171318 if (option & ADD_CACHE_JUST_APPEND)1319 pos = istate->cache_nr;1320 else {1321 int ret;1322 ret = add_index_entry_with_check(istate, ce, option);1323 if (ret <= 0)1324 return ret;1325 pos = ret - 1;1326 }13271328 /* Make sure the array is big enough .. */1329 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);13301331 /* Add it in.. */1332 istate->cache_nr++;1333 if (istate->cache_nr > pos + 1)1334 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,1335 istate->cache_nr - pos - 1);1336 set_index_entry(istate, pos, ce);1337 istate->cache_changed |= CE_ENTRY_ADDED;1338 return 0;1339}13401341/*1342 * "refresh" does not calculate a new sha1 file or bring the1343 * cache up-to-date for mode/content changes. But what it1344 * _does_ do is to "re-match" the stat information of a file1345 * with the cache, so that you can refresh the cache for a1346 * file that hasn't been changed but where the stat entry is1347 * out of date.1348 *1349 * For example, you'd want to do this after doing a "git-read-tree",1350 * to link up the stat cache details with the proper files.1351 */1352static struct cache_entry *refresh_cache_ent(struct index_state *istate,1353 struct cache_entry *ce,1354 unsigned int options, int *err,1355 int *changed_ret)1356{1357 struct stat st;1358 struct cache_entry *updated;1359 int changed;1360 int refresh = options & CE_MATCH_REFRESH;1361 int ignore_valid = options & CE_MATCH_IGNORE_VALID;1362 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;1363 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;1364 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;13651366 if (!refresh || ce_uptodate(ce))1367 return ce;13681369 if (!ignore_fsmonitor)1370 refresh_fsmonitor(istate);1371 /*1372 * CE_VALID or CE_SKIP_WORKTREE means the user promised us1373 * that the change to the work tree does not matter and told1374 * us not to worry.1375 */1376 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {1377 ce_mark_uptodate(ce);1378 return ce;1379 }1380 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {1381 ce_mark_uptodate(ce);1382 return ce;1383 }1384 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {1385 ce_mark_uptodate(ce);1386 return ce;1387 }13881389 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {1390 if (ignore_missing)1391 return ce;1392 if (err)1393 *err = ENOENT;1394 return NULL;1395 }13961397 if (lstat(ce->name, &st) < 0) {1398 if (ignore_missing && errno == ENOENT)1399 return ce;1400 if (err)1401 *err = errno;1402 return NULL;1403 }14041405 changed = ie_match_stat(istate, ce, &st, options);1406 if (changed_ret)1407 *changed_ret = changed;1408 if (!changed) {1409 /*1410 * The path is unchanged. If we were told to ignore1411 * valid bit, then we did the actual stat check and1412 * found that the entry is unmodified. If the entry1413 * is not marked VALID, this is the place to mark it1414 * valid again, under "assume unchanged" mode.1415 */1416 if (ignore_valid && assume_unchanged &&1417 !(ce->ce_flags & CE_VALID))1418 ; /* mark this one VALID again */1419 else {1420 /*1421 * We do not mark the index itself "modified"1422 * because CE_UPTODATE flag is in-core only;1423 * we are not going to write this change out.1424 */1425 if (!S_ISGITLINK(ce->ce_mode)) {1426 ce_mark_uptodate(ce);1427 mark_fsmonitor_valid(ce);1428 }1429 return ce;1430 }1431 }14321433 if (ie_modified(istate, ce, &st, options)) {1434 if (err)1435 *err = EINVAL;1436 return NULL;1437 }14381439 updated = make_empty_cache_entry(istate, ce_namelen(ce));1440 copy_cache_entry(updated, ce);1441 memcpy(updated->name, ce->name, ce->ce_namelen + 1);1442 fill_stat_cache_info(updated, &st);1443 /*1444 * If ignore_valid is not set, we should leave CE_VALID bit1445 * alone. Otherwise, paths marked with --no-assume-unchanged1446 * (i.e. things to be edited) will reacquire CE_VALID bit1447 * automatically, which is not really what we want.1448 */1449 if (!ignore_valid && assume_unchanged &&1450 !(ce->ce_flags & CE_VALID))1451 updated->ce_flags &= ~CE_VALID;14521453 /* istate->cache_changed is updated in the caller */1454 return updated;1455}14561457static void show_file(const char * fmt, const char * name, int in_porcelain,1458 int * first, const char *header_msg)1459{1460 if (in_porcelain && *first && header_msg) {1461 printf("%s\n", header_msg);1462 *first = 0;1463 }1464 printf(fmt, name);1465}14661467int refresh_index(struct index_state *istate, unsigned int flags,1468 const struct pathspec *pathspec,1469 char *seen, const char *header_msg)1470{1471 int i;1472 int has_errors = 0;1473 int really = (flags & REFRESH_REALLY) != 0;1474 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;1475 int quiet = (flags & REFRESH_QUIET) != 0;1476 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;1477 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;1478 int first = 1;1479 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);1480 unsigned int options = (CE_MATCH_REFRESH |1481 (really ? CE_MATCH_IGNORE_VALID : 0) |1482 (not_new ? CE_MATCH_IGNORE_MISSING : 0));1483 const char *modified_fmt;1484 const char *deleted_fmt;1485 const char *typechange_fmt;1486 const char *added_fmt;1487 const char *unmerged_fmt;1488 struct progress *progress = NULL;14891490 if (flags & REFRESH_PROGRESS && isatty(2))1491 progress = start_delayed_progress(_("Refresh index"),1492 istate->cache_nr);14931494 trace_performance_enter();1495 modified_fmt = in_porcelain ? "M\t%s\n" : "%s: needs update\n";1496 deleted_fmt = in_porcelain ? "D\t%s\n" : "%s: needs update\n";1497 typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";1498 added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n";1499 unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";1500 /*1501 * Use the multi-threaded preload_index() to refresh most of the1502 * cache entries quickly then in the single threaded loop below,1503 * we only have to do the special cases that are left.1504 */1505 preload_index(istate, pathspec, 0);1506 for (i = 0; i < istate->cache_nr; i++) {1507 struct cache_entry *ce, *new_entry;1508 int cache_errno = 0;1509 int changed = 0;1510 int filtered = 0;15111512 ce = istate->cache[i];1513 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))1514 continue;15151516 if (pathspec && !ce_path_match(istate, ce, pathspec, seen))1517 filtered = 1;15181519 if (ce_stage(ce)) {1520 while ((i < istate->cache_nr) &&1521 ! strcmp(istate->cache[i]->name, ce->name))1522 i++;1523 i--;1524 if (allow_unmerged)1525 continue;1526 if (!filtered)1527 show_file(unmerged_fmt, ce->name, in_porcelain,1528 &first, header_msg);1529 has_errors = 1;1530 continue;1531 }15321533 if (filtered)1534 continue;15351536 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);1537 if (new_entry == ce)1538 continue;1539 if (progress)1540 display_progress(progress, i);1541 if (!new_entry) {1542 const char *fmt;15431544 if (really && cache_errno == EINVAL) {1545 /* If we are doing --really-refresh that1546 * means the index is not valid anymore.1547 */1548 ce->ce_flags &= ~CE_VALID;1549 ce->ce_flags |= CE_UPDATE_IN_BASE;1550 mark_fsmonitor_invalid(istate, ce);1551 istate->cache_changed |= CE_ENTRY_CHANGED;1552 }1553 if (quiet)1554 continue;15551556 if (cache_errno == ENOENT)1557 fmt = deleted_fmt;1558 else if (ce_intent_to_add(ce))1559 fmt = added_fmt; /* must be before other checks */1560 else if (changed & TYPE_CHANGED)1561 fmt = typechange_fmt;1562 else1563 fmt = modified_fmt;1564 show_file(fmt,1565 ce->name, in_porcelain, &first, header_msg);1566 has_errors = 1;1567 continue;1568 }15691570 replace_index_entry(istate, i, new_entry);1571 }1572 if (progress) {1573 display_progress(progress, istate->cache_nr);1574 stop_progress(&progress);1575 }1576 trace_performance_leave("refresh index");1577 return has_errors;1578}15791580struct cache_entry *refresh_cache_entry(struct index_state *istate,1581 struct cache_entry *ce,1582 unsigned int options)1583{1584 return refresh_cache_ent(istate, ce, options, NULL, NULL);1585}158615871588/*****************************************************************1589 * Index File I/O1590 *****************************************************************/15911592#define INDEX_FORMAT_DEFAULT 315931594static unsigned int get_index_format_default(void)1595{1596 char *envversion = getenv("GIT_INDEX_VERSION");1597 char *endp;1598 int value;1599 unsigned int version = INDEX_FORMAT_DEFAULT;16001601 if (!envversion) {1602 if (!git_config_get_int("index.version", &value))1603 version = value;1604 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {1605 warning(_("index.version set, but the value is invalid.\n"1606 "Using version %i"), INDEX_FORMAT_DEFAULT);1607 return INDEX_FORMAT_DEFAULT;1608 }1609 return version;1610 }16111612 version = strtoul(envversion, &endp, 10);1613 if (*endp ||1614 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {1615 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"1616 "Using version %i"), INDEX_FORMAT_DEFAULT);1617 version = INDEX_FORMAT_DEFAULT;1618 }1619 return version;1620}16211622/*1623 * dev/ino/uid/gid/size are also just tracked to the low 32 bits1624 * Again - this is just a (very strong in practice) heuristic that1625 * the inode hasn't changed.1626 *1627 * We save the fields in big-endian order to allow using the1628 * index file over NFS transparently.1629 */1630struct ondisk_cache_entry {1631 struct cache_time ctime;1632 struct cache_time mtime;1633 uint32_t dev;1634 uint32_t ino;1635 uint32_t mode;1636 uint32_t uid;1637 uint32_t gid;1638 uint32_t size;1639 unsigned char sha1[20];1640 uint16_t flags;1641 char name[FLEX_ARRAY]; /* more */1642};16431644/*1645 * This struct is used when CE_EXTENDED bit is 11646 * The struct must match ondisk_cache_entry exactly from1647 * ctime till flags1648 */1649struct ondisk_cache_entry_extended {1650 struct cache_time ctime;1651 struct cache_time mtime;1652 uint32_t dev;1653 uint32_t ino;1654 uint32_t mode;1655 uint32_t uid;1656 uint32_t gid;1657 uint32_t size;1658 unsigned char sha1[20];1659 uint16_t flags;1660 uint16_t flags2;1661 char name[FLEX_ARRAY]; /* more */1662};16631664/* These are only used for v3 or lower */1665#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)1666#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)1667#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)1668#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)1669#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \1670 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \1671 ondisk_cache_entry_size(ce_namelen(ce)))16721673/* Allow fsck to force verification of the index checksum. */1674int verify_index_checksum;16751676/* Allow fsck to force verification of the cache entry order. */1677int verify_ce_order;16781679static int verify_hdr(const struct cache_header *hdr, unsigned long size)1680{1681 git_hash_ctx c;1682 unsigned char hash[GIT_MAX_RAWSZ];1683 int hdr_version;16841685 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))1686 return error(_("bad signature 0x%08x"), hdr->hdr_signature);1687 hdr_version = ntohl(hdr->hdr_version);1688 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)1689 return error(_("bad index version %d"), hdr_version);16901691 if (!verify_index_checksum)1692 return 0;16931694 the_hash_algo->init_fn(&c);1695 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);1696 the_hash_algo->final_fn(hash, &c);1697 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))1698 return error(_("bad index file sha1 signature"));1699 return 0;1700}17011702static int read_index_extension(struct index_state *istate,1703 const char *ext, const char *data, unsigned long sz)1704{1705 switch (CACHE_EXT(ext)) {1706 case CACHE_EXT_TREE:1707 istate->cache_tree = cache_tree_read(data, sz);1708 break;1709 case CACHE_EXT_RESOLVE_UNDO:1710 istate->resolve_undo = resolve_undo_read(data, sz);1711 break;1712 case CACHE_EXT_LINK:1713 if (read_link_extension(istate, data, sz))1714 return -1;1715 break;1716 case CACHE_EXT_UNTRACKED:1717 istate->untracked = read_untracked_extension(data, sz);1718 break;1719 case CACHE_EXT_FSMONITOR:1720 read_fsmonitor_extension(istate, data, sz);1721 break;1722 case CACHE_EXT_ENDOFINDEXENTRIES:1723 case CACHE_EXT_INDEXENTRYOFFSETTABLE:1724 /* already handled in do_read_index() */1725 break;1726 default:1727 if (*ext < 'A' || 'Z' < *ext)1728 return error(_("index uses %.4s extension, which we do not understand"),1729 ext);1730 fprintf_ln(stderr, _("ignoring %.4s extension"), ext);1731 break;1732 }1733 return 0;1734}17351736int read_index(struct index_state *istate)1737{1738 return read_index_from(istate, get_index_file(), get_git_dir());1739}17401741static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,1742 unsigned int version,1743 struct ondisk_cache_entry *ondisk,1744 unsigned long *ent_size,1745 const struct cache_entry *previous_ce)1746{1747 struct cache_entry *ce;1748 size_t len;1749 const char *name;1750 unsigned int flags;1751 size_t copy_len = 0;1752 /*1753 * Adjacent cache entries tend to share the leading paths, so it makes1754 * sense to only store the differences in later entries. In the v41755 * on-disk format of the index, each on-disk cache entry stores the1756 * number of bytes to be stripped from the end of the previous name,1757 * and the bytes to append to the result, to come up with its name.1758 */1759 int expand_name_field = version == 4;17601761 /* On-disk flags are just 16 bits */1762 flags = get_be16(&ondisk->flags);1763 len = flags & CE_NAMEMASK;17641765 if (flags & CE_EXTENDED) {1766 struct ondisk_cache_entry_extended *ondisk2;1767 int extended_flags;1768 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;1769 extended_flags = get_be16(&ondisk2->flags2) << 16;1770 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */1771 if (extended_flags & ~CE_EXTENDED_FLAGS)1772 die(_("unknown index entry format 0x%08x"), extended_flags);1773 flags |= extended_flags;1774 name = ondisk2->name;1775 }1776 else1777 name = ondisk->name;17781779 if (expand_name_field) {1780 const unsigned char *cp = (const unsigned char *)name;1781 size_t strip_len, previous_len;17821783 /* If we're at the begining of a block, ignore the previous name */1784 strip_len = decode_varint(&cp);1785 if (previous_ce) {1786 previous_len = previous_ce->ce_namelen;1787 if (previous_len < strip_len)1788 die(_("malformed name field in the index, near path '%s'"),1789 previous_ce->name);1790 copy_len = previous_len - strip_len;1791 }1792 name = (const char *)cp;1793 }17941795 if (len == CE_NAMEMASK) {1796 len = strlen(name);1797 if (expand_name_field)1798 len += copy_len;1799 }18001801 ce = mem_pool__ce_alloc(ce_mem_pool, len);18021803 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);1804 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);1805 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);1806 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);1807 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);1808 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);1809 ce->ce_mode = get_be32(&ondisk->mode);1810 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);1811 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);1812 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);1813 ce->ce_flags = flags & ~CE_NAMEMASK;1814 ce->ce_namelen = len;1815 ce->index = 0;1816 hashcpy(ce->oid.hash, ondisk->sha1);18171818 if (expand_name_field) {1819 if (copy_len)1820 memcpy(ce->name, previous_ce->name, copy_len);1821 memcpy(ce->name + copy_len, name, len + 1 - copy_len);1822 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;1823 } else {1824 memcpy(ce->name, name, len + 1);1825 *ent_size = ondisk_ce_size(ce);1826 }1827 return ce;1828}18291830static void check_ce_order(struct index_state *istate)1831{1832 unsigned int i;18331834 if (!verify_ce_order)1835 return;18361837 for (i = 1; i < istate->cache_nr; i++) {1838 struct cache_entry *ce = istate->cache[i - 1];1839 struct cache_entry *next_ce = istate->cache[i];1840 int name_compare = strcmp(ce->name, next_ce->name);18411842 if (0 < name_compare)1843 die(_("unordered stage entries in index"));1844 if (!name_compare) {1845 if (!ce_stage(ce))1846 die(_("multiple stage entries for merged file '%s'"),1847 ce->name);1848 if (ce_stage(ce) > ce_stage(next_ce))1849 die(_("unordered stage entries for '%s'"),1850 ce->name);1851 }1852 }1853}18541855static void tweak_untracked_cache(struct index_state *istate)1856{1857 switch (git_config_get_untracked_cache()) {1858 case -1: /* keep: do nothing */1859 break;1860 case 0: /* false */1861 remove_untracked_cache(istate);1862 break;1863 case 1: /* true */1864 add_untracked_cache(istate);1865 break;1866 default: /* unknown value: do nothing */1867 break;1868 }1869}18701871static void tweak_split_index(struct index_state *istate)1872{1873 switch (git_config_get_split_index()) {1874 case -1: /* unset: do nothing */1875 break;1876 case 0: /* false */1877 remove_split_index(istate);1878 break;1879 case 1: /* true */1880 add_split_index(istate);1881 break;1882 default: /* unknown value: do nothing */1883 break;1884 }1885}18861887static void post_read_index_from(struct index_state *istate)1888{1889 check_ce_order(istate);1890 tweak_untracked_cache(istate);1891 tweak_split_index(istate);1892 tweak_fsmonitor(istate);1893}18941895static size_t estimate_cache_size_from_compressed(unsigned int entries)1896{1897 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);1898}18991900static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)1901{1902 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);19031904 /*1905 * Account for potential alignment differences.1906 */1907 per_entry += align_padding_size(sizeof(struct cache_entry), -sizeof(struct ondisk_cache_entry));1908 return ondisk_size + entries * per_entry;1909}19101911struct index_entry_offset1912{1913 /* starting byte offset into index file, count of index entries in this block */1914 int offset, nr;1915};19161917struct index_entry_offset_table1918{1919 int nr;1920 struct index_entry_offset entries[FLEX_ARRAY];1921};19221923static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);1924static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);19251926static size_t read_eoie_extension(const char *mmap, size_t mmap_size);1927static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);19281929struct load_index_extensions1930{1931 pthread_t pthread;1932 struct index_state *istate;1933 const char *mmap;1934 size_t mmap_size;1935 unsigned long src_offset;1936};19371938static void *load_index_extensions(void *_data)1939{1940 struct load_index_extensions *p = _data;1941 unsigned long src_offset = p->src_offset;19421943 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {1944 /* After an array of active_nr index entries,1945 * there can be arbitrary number of extended1946 * sections, each of which is prefixed with1947 * extension name (4-byte) and section length1948 * in 4-byte network byte order.1949 */1950 uint32_t extsize = get_be32(p->mmap + src_offset + 4);1951 if (read_index_extension(p->istate,1952 p->mmap + src_offset,1953 p->mmap + src_offset + 8,1954 extsize) < 0) {1955 munmap((void *)p->mmap, p->mmap_size);1956 die(_("index file corrupt"));1957 }1958 src_offset += 8;1959 src_offset += extsize;1960 }19611962 return NULL;1963}19641965/*1966 * A helper function that will load the specified range of cache entries1967 * from the memory mapped file and add them to the given index.1968 */1969static unsigned long load_cache_entry_block(struct index_state *istate,1970 struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,1971 unsigned long start_offset, const struct cache_entry *previous_ce)1972{1973 int i;1974 unsigned long src_offset = start_offset;19751976 for (i = offset; i < offset + nr; i++) {1977 struct ondisk_cache_entry *disk_ce;1978 struct cache_entry *ce;1979 unsigned long consumed;19801981 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);1982 ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce);1983 set_index_entry(istate, i, ce);19841985 src_offset += consumed;1986 previous_ce = ce;1987 }1988 return src_offset - start_offset;1989}19901991static unsigned long load_all_cache_entries(struct index_state *istate,1992 const char *mmap, size_t mmap_size, unsigned long src_offset)1993{1994 unsigned long consumed;19951996 if (istate->version == 4) {1997 mem_pool_init(&istate->ce_mem_pool,1998 estimate_cache_size_from_compressed(istate->cache_nr));1999 } else {2000 mem_pool_init(&istate->ce_mem_pool,2001 estimate_cache_size(mmap_size, istate->cache_nr));2002 }20032004 consumed = load_cache_entry_block(istate, istate->ce_mem_pool,2005 0, istate->cache_nr, mmap, src_offset, NULL);2006 return consumed;2007}20082009/*2010 * Mostly randomly chosen maximum thread counts: we2011 * cap the parallelism to online_cpus() threads, and we want2012 * to have at least 10000 cache entries per thread for it to2013 * be worth starting a thread.2014 */20152016#define THREAD_COST (10000)20172018struct load_cache_entries_thread_data2019{2020 pthread_t pthread;2021 struct index_state *istate;2022 struct mem_pool *ce_mem_pool;2023 int offset;2024 const char *mmap;2025 struct index_entry_offset_table *ieot;2026 int ieot_start; /* starting index into the ieot array */2027 int ieot_blocks; /* count of ieot entries to process */2028 unsigned long consumed; /* return # of bytes in index file processed */2029};20302031/*2032 * A thread proc to run the load_cache_entries() computation2033 * across multiple background threads.2034 */2035static void *load_cache_entries_thread(void *_data)2036{2037 struct load_cache_entries_thread_data *p = _data;2038 int i;20392040 /* iterate across all ieot blocks assigned to this thread */2041 for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {2042 p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,2043 p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);2044 p->offset += p->ieot->entries[i].nr;2045 }2046 return NULL;2047}20482049static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,2050 unsigned long src_offset, int nr_threads, struct index_entry_offset_table *ieot)2051{2052 int i, offset, ieot_blocks, ieot_start, err;2053 struct load_cache_entries_thread_data *data;2054 unsigned long consumed = 0;20552056 /* a little sanity checking */2057 if (istate->name_hash_initialized)2058 BUG("the name hash isn't thread safe");20592060 mem_pool_init(&istate->ce_mem_pool, 0);20612062 /* ensure we have no more threads than we have blocks to process */2063 if (nr_threads > ieot->nr)2064 nr_threads = ieot->nr;2065 data = xcalloc(nr_threads, sizeof(*data));20662067 offset = ieot_start = 0;2068 ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);2069 for (i = 0; i < nr_threads; i++) {2070 struct load_cache_entries_thread_data *p = &data[i];2071 int nr, j;20722073 if (ieot_start + ieot_blocks > ieot->nr)2074 ieot_blocks = ieot->nr - ieot_start;20752076 p->istate = istate;2077 p->offset = offset;2078 p->mmap = mmap;2079 p->ieot = ieot;2080 p->ieot_start = ieot_start;2081 p->ieot_blocks = ieot_blocks;20822083 /* create a mem_pool for each thread */2084 nr = 0;2085 for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)2086 nr += p->ieot->entries[j].nr;2087 if (istate->version == 4) {2088 mem_pool_init(&p->ce_mem_pool,2089 estimate_cache_size_from_compressed(nr));2090 } else {2091 mem_pool_init(&p->ce_mem_pool,2092 estimate_cache_size(mmap_size, nr));2093 }20942095 err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);2096 if (err)2097 die(_("unable to create load_cache_entries thread: %s"), strerror(err));20982099 /* increment by the number of cache entries in the ieot block being processed */2100 for (j = 0; j < ieot_blocks; j++)2101 offset += ieot->entries[ieot_start + j].nr;2102 ieot_start += ieot_blocks;2103 }21042105 for (i = 0; i < nr_threads; i++) {2106 struct load_cache_entries_thread_data *p = &data[i];21072108 err = pthread_join(p->pthread, NULL);2109 if (err)2110 die(_("unable to join load_cache_entries thread: %s"), strerror(err));2111 mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);2112 consumed += p->consumed;2113 }21142115 free(data);21162117 return consumed;2118}21192120/* remember to discard_cache() before reading a different cache! */2121int do_read_index(struct index_state *istate, const char *path, int must_exist)2122{2123 int fd;2124 struct stat st;2125 unsigned long src_offset;2126 const struct cache_header *hdr;2127 const char *mmap;2128 size_t mmap_size;2129 struct load_index_extensions p;2130 size_t extension_offset = 0;2131 int nr_threads, cpus;2132 struct index_entry_offset_table *ieot = NULL;21332134 if (istate->initialized)2135 return istate->cache_nr;21362137 istate->timestamp.sec = 0;2138 istate->timestamp.nsec = 0;2139 fd = open(path, O_RDONLY);2140 if (fd < 0) {2141 if (!must_exist && errno == ENOENT)2142 return 0;2143 die_errno(_("%s: index file open failed"), path);2144 }21452146 if (fstat(fd, &st))2147 die_errno(_("%s: cannot stat the open index"), path);21482149 mmap_size = xsize_t(st.st_size);2150 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)2151 die(_("%s: index file smaller than expected"), path);21522153 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);2154 if (mmap == MAP_FAILED)2155 die_errno(_("%s: unable to map index file"), path);2156 close(fd);21572158 hdr = (const struct cache_header *)mmap;2159 if (verify_hdr(hdr, mmap_size) < 0)2160 goto unmap;21612162 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);2163 istate->version = ntohl(hdr->hdr_version);2164 istate->cache_nr = ntohl(hdr->hdr_entries);2165 istate->cache_alloc = alloc_nr(istate->cache_nr);2166 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));2167 istate->initialized = 1;21682169 p.istate = istate;2170 p.mmap = mmap;2171 p.mmap_size = mmap_size;21722173 src_offset = sizeof(*hdr);21742175 if (git_config_get_index_threads(&nr_threads))2176 nr_threads = 1;21772178 /* TODO: does creating more threads than cores help? */2179 if (!nr_threads) {2180 nr_threads = istate->cache_nr / THREAD_COST;2181 cpus = online_cpus();2182 if (nr_threads > cpus)2183 nr_threads = cpus;2184 }21852186 if (!HAVE_THREADS)2187 nr_threads = 1;21882189 if (nr_threads > 1) {2190 extension_offset = read_eoie_extension(mmap, mmap_size);2191 if (extension_offset) {2192 int err;21932194 p.src_offset = extension_offset;2195 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);2196 if (err)2197 die(_("unable to create load_index_extensions thread: %s"), strerror(err));21982199 nr_threads--;2200 }2201 }22022203 /*2204 * Locate and read the index entry offset table so that we can use it2205 * to multi-thread the reading of the cache entries.2206 */2207 if (extension_offset && nr_threads > 1)2208 ieot = read_ieot_extension(mmap, mmap_size, extension_offset);22092210 if (ieot) {2211 src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, src_offset, nr_threads, ieot);2212 free(ieot);2213 } else {2214 src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);2215 }22162217 istate->timestamp.sec = st.st_mtime;2218 istate->timestamp.nsec = ST_MTIME_NSEC(st);22192220 /* if we created a thread, join it otherwise load the extensions on the primary thread */2221 if (extension_offset) {2222 int ret = pthread_join(p.pthread, NULL);2223 if (ret)2224 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));2225 } else {2226 p.src_offset = src_offset;2227 load_index_extensions(&p);2228 }2229 munmap((void *)mmap, mmap_size);2230 return istate->cache_nr;22312232unmap:2233 munmap((void *)mmap, mmap_size);2234 die(_("index file corrupt"));2235}22362237/*2238 * Signal that the shared index is used by updating its mtime.2239 *2240 * This way, shared index can be removed if they have not been used2241 * for some time.2242 */2243static void freshen_shared_index(const char *shared_index, int warn)2244{2245 if (!check_and_freshen_file(shared_index, 1) && warn)2246 warning(_("could not freshen shared index '%s'"), shared_index);2247}22482249int read_index_from(struct index_state *istate, const char *path,2250 const char *gitdir)2251{2252 struct split_index *split_index;2253 int ret;2254 char *base_oid_hex;2255 char *base_path;22562257 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */2258 if (istate->initialized)2259 return istate->cache_nr;22602261 trace_performance_enter();2262 ret = do_read_index(istate, path, 0);2263 trace_performance_leave("read cache %s", path);22642265 split_index = istate->split_index;2266 if (!split_index || is_null_oid(&split_index->base_oid)) {2267 post_read_index_from(istate);2268 return ret;2269 }22702271 trace_performance_enter();2272 if (split_index->base)2273 discard_index(split_index->base);2274 else2275 split_index->base = xcalloc(1, sizeof(*split_index->base));22762277 base_oid_hex = oid_to_hex(&split_index->base_oid);2278 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);2279 ret = do_read_index(split_index->base, base_path, 1);2280 if (!oideq(&split_index->base_oid, &split_index->base->oid))2281 die(_("broken index, expect %s in %s, got %s"),2282 base_oid_hex, base_path,2283 oid_to_hex(&split_index->base->oid));22842285 freshen_shared_index(base_path, 0);2286 merge_base_index(istate);2287 post_read_index_from(istate);2288 trace_performance_leave("read cache %s", base_path);2289 free(base_path);2290 return ret;2291}22922293int is_index_unborn(struct index_state *istate)2294{2295 return (!istate->cache_nr && !istate->timestamp.sec);2296}22972298int discard_index(struct index_state *istate)2299{2300 /*2301 * Cache entries in istate->cache[] should have been allocated2302 * from the memory pool associated with this index, or from an2303 * associated split_index. There is no need to free individual2304 * cache entries. validate_cache_entries can detect when this2305 * assertion does not hold.2306 */2307 validate_cache_entries(istate);23082309 resolve_undo_clear_index(istate);2310 istate->cache_nr = 0;2311 istate->cache_changed = 0;2312 istate->timestamp.sec = 0;2313 istate->timestamp.nsec = 0;2314 free_name_hash(istate);2315 cache_tree_free(&(istate->cache_tree));2316 istate->initialized = 0;2317 FREE_AND_NULL(istate->cache);2318 istate->cache_alloc = 0;2319 discard_split_index(istate);2320 free_untracked_cache(istate->untracked);2321 istate->untracked = NULL;23222323 if (istate->ce_mem_pool) {2324 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());2325 istate->ce_mem_pool = NULL;2326 }23272328 return 0;2329}23302331/*2332 * Validate the cache entries of this index.2333 * All cache entries associated with this index2334 * should have been allocated by the memory pool2335 * associated with this index, or by a referenced2336 * split index.2337 */2338void validate_cache_entries(const struct index_state *istate)2339{2340 int i;23412342 if (!should_validate_cache_entries() ||!istate || !istate->initialized)2343 return;23442345 for (i = 0; i < istate->cache_nr; i++) {2346 if (!istate) {2347 BUG("cache entry is not allocated from expected memory pool");2348 } else if (!istate->ce_mem_pool ||2349 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {2350 if (!istate->split_index ||2351 !istate->split_index->base ||2352 !istate->split_index->base->ce_mem_pool ||2353 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {2354 BUG("cache entry is not allocated from expected memory pool");2355 }2356 }2357 }23582359 if (istate->split_index)2360 validate_cache_entries(istate->split_index->base);2361}23622363int unmerged_index(const struct index_state *istate)2364{2365 int i;2366 for (i = 0; i < istate->cache_nr; i++) {2367 if (ce_stage(istate->cache[i]))2368 return 1;2369 }2370 return 0;2371}23722373int index_has_changes(struct index_state *istate,2374 struct tree *tree,2375 struct strbuf *sb)2376{2377 struct object_id cmp;2378 int i;23792380 if (istate != &the_index) {2381 BUG("index_has_changes cannot yet accept istate != &the_index; do_diff_cache needs updating first.");2382 }2383 if (tree)2384 cmp = tree->object.oid;2385 if (tree || !get_oid_tree("HEAD", &cmp)) {2386 struct diff_options opt;23872388 repo_diff_setup(the_repository, &opt);2389 opt.flags.exit_with_status = 1;2390 if (!sb)2391 opt.flags.quick = 1;2392 do_diff_cache(&cmp, &opt);2393 diffcore_std(&opt);2394 for (i = 0; sb && i < diff_queued_diff.nr; i++) {2395 if (i)2396 strbuf_addch(sb, ' ');2397 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);2398 }2399 diff_flush(&opt);2400 return opt.flags.has_changes != 0;2401 } else {2402 for (i = 0; sb && i < istate->cache_nr; i++) {2403 if (i)2404 strbuf_addch(sb, ' ');2405 strbuf_addstr(sb, istate->cache[i]->name);2406 }2407 return !!istate->cache_nr;2408 }2409}24102411#define WRITE_BUFFER_SIZE 81922412static unsigned char write_buffer[WRITE_BUFFER_SIZE];2413static unsigned long write_buffer_len;24142415static int ce_write_flush(git_hash_ctx *context, int fd)2416{2417 unsigned int buffered = write_buffer_len;2418 if (buffered) {2419 the_hash_algo->update_fn(context, write_buffer, buffered);2420 if (write_in_full(fd, write_buffer, buffered) < 0)2421 return -1;2422 write_buffer_len = 0;2423 }2424 return 0;2425}24262427static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)2428{2429 while (len) {2430 unsigned int buffered = write_buffer_len;2431 unsigned int partial = WRITE_BUFFER_SIZE - buffered;2432 if (partial > len)2433 partial = len;2434 memcpy(write_buffer + buffered, data, partial);2435 buffered += partial;2436 if (buffered == WRITE_BUFFER_SIZE) {2437 write_buffer_len = buffered;2438 if (ce_write_flush(context, fd))2439 return -1;2440 buffered = 0;2441 }2442 write_buffer_len = buffered;2443 len -= partial;2444 data = (char *) data + partial;2445 }2446 return 0;2447}24482449static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,2450 int fd, unsigned int ext, unsigned int sz)2451{2452 ext = htonl(ext);2453 sz = htonl(sz);2454 if (eoie_context) {2455 the_hash_algo->update_fn(eoie_context, &ext, 4);2456 the_hash_algo->update_fn(eoie_context, &sz, 4);2457 }2458 return ((ce_write(context, fd, &ext, 4) < 0) ||2459 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;2460}24612462static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)2463{2464 unsigned int left = write_buffer_len;24652466 if (left) {2467 write_buffer_len = 0;2468 the_hash_algo->update_fn(context, write_buffer, left);2469 }24702471 /* Flush first if not enough space for hash signature */2472 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {2473 if (write_in_full(fd, write_buffer, left) < 0)2474 return -1;2475 left = 0;2476 }24772478 /* Append the hash signature at the end */2479 the_hash_algo->final_fn(write_buffer + left, context);2480 hashcpy(hash, write_buffer + left);2481 left += the_hash_algo->rawsz;2482 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;2483}24842485static void ce_smudge_racily_clean_entry(struct index_state *istate,2486 struct cache_entry *ce)2487{2488 /*2489 * The only thing we care about in this function is to smudge the2490 * falsely clean entry due to touch-update-touch race, so we leave2491 * everything else as they are. We are called for entries whose2492 * ce_stat_data.sd_mtime match the index file mtime.2493 *2494 * Note that this actually does not do much for gitlinks, for2495 * which ce_match_stat_basic() always goes to the actual2496 * contents. The caller checks with is_racy_timestamp() which2497 * always says "no" for gitlinks, so we are not called for them ;-)2498 */2499 struct stat st;25002501 if (lstat(ce->name, &st) < 0)2502 return;2503 if (ce_match_stat_basic(ce, &st))2504 return;2505 if (ce_modified_check_fs(istate, ce, &st)) {2506 /* This is "racily clean"; smudge it. Note that this2507 * is a tricky code. At first glance, it may appear2508 * that it can break with this sequence:2509 *2510 * $ echo xyzzy >frotz2511 * $ git-update-index --add frotz2512 * $ : >frotz2513 * $ sleep 32514 * $ echo filfre >nitfol2515 * $ git-update-index --add nitfol2516 *2517 * but it does not. When the second update-index runs,2518 * it notices that the entry "frotz" has the same timestamp2519 * as index, and if we were to smudge it by resetting its2520 * size to zero here, then the object name recorded2521 * in index is the 6-byte file but the cached stat information2522 * becomes zero --- which would then match what we would2523 * obtain from the filesystem next time we stat("frotz").2524 *2525 * However, the second update-index, before calling2526 * this function, notices that the cached size is 62527 * bytes and what is on the filesystem is an empty2528 * file, and never calls us, so the cached size information2529 * for "frotz" stays 6 which does not match the filesystem.2530 */2531 ce->ce_stat_data.sd_size = 0;2532 }2533}25342535/* Copy miscellaneous fields but not the name */2536static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,2537 struct cache_entry *ce)2538{2539 short flags;25402541 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);2542 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);2543 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);2544 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);2545 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);2546 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);2547 ondisk->mode = htonl(ce->ce_mode);2548 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);2549 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);2550 ondisk->size = htonl(ce->ce_stat_data.sd_size);2551 hashcpy(ondisk->sha1, ce->oid.hash);25522553 flags = ce->ce_flags & ~CE_NAMEMASK;2554 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));2555 ondisk->flags = htons(flags);2556 if (ce->ce_flags & CE_EXTENDED) {2557 struct ondisk_cache_entry_extended *ondisk2;2558 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;2559 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);2560 }2561}25622563static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,2564 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)2565{2566 int size;2567 int result;2568 unsigned int saved_namelen;2569 int stripped_name = 0;2570 static unsigned char padding[8] = { 0x00 };25712572 if (ce->ce_flags & CE_STRIP_NAME) {2573 saved_namelen = ce_namelen(ce);2574 ce->ce_namelen = 0;2575 stripped_name = 1;2576 }25772578 if (ce->ce_flags & CE_EXTENDED)2579 size = offsetof(struct ondisk_cache_entry_extended, name);2580 else2581 size = offsetof(struct ondisk_cache_entry, name);25822583 if (!previous_name) {2584 int len = ce_namelen(ce);2585 copy_cache_entry_to_ondisk(ondisk, ce);2586 result = ce_write(c, fd, ondisk, size);2587 if (!result)2588 result = ce_write(c, fd, ce->name, len);2589 if (!result)2590 result = ce_write(c, fd, padding, align_padding_size(size, len));2591 } else {2592 int common, to_remove, prefix_size;2593 unsigned char to_remove_vi[16];2594 for (common = 0;2595 (ce->name[common] &&2596 common < previous_name->len &&2597 ce->name[common] == previous_name->buf[common]);2598 common++)2599 ; /* still matching */2600 to_remove = previous_name->len - common;2601 prefix_size = encode_varint(to_remove, to_remove_vi);26022603 copy_cache_entry_to_ondisk(ondisk, ce);2604 result = ce_write(c, fd, ondisk, size);2605 if (!result)2606 result = ce_write(c, fd, to_remove_vi, prefix_size);2607 if (!result)2608 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);2609 if (!result)2610 result = ce_write(c, fd, padding, 1);26112612 strbuf_splice(previous_name, common, to_remove,2613 ce->name + common, ce_namelen(ce) - common);2614 }2615 if (stripped_name) {2616 ce->ce_namelen = saved_namelen;2617 ce->ce_flags &= ~CE_STRIP_NAME;2618 }26192620 return result;2621}26222623/*2624 * This function verifies if index_state has the correct sha1 of the2625 * index file. Don't die if we have any other failure, just return 0.2626 */2627static int verify_index_from(const struct index_state *istate, const char *path)2628{2629 int fd;2630 ssize_t n;2631 struct stat st;2632 unsigned char hash[GIT_MAX_RAWSZ];26332634 if (!istate->initialized)2635 return 0;26362637 fd = open(path, O_RDONLY);2638 if (fd < 0)2639 return 0;26402641 if (fstat(fd, &st))2642 goto out;26432644 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)2645 goto out;26462647 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);2648 if (n != the_hash_algo->rawsz)2649 goto out;26502651 if (!hasheq(istate->oid.hash, hash))2652 goto out;26532654 close(fd);2655 return 1;26562657out:2658 close(fd);2659 return 0;2660}26612662static int verify_index(const struct index_state *istate)2663{2664 return verify_index_from(istate, get_index_file());2665}26662667static int has_racy_timestamp(struct index_state *istate)2668{2669 int entries = istate->cache_nr;2670 int i;26712672 for (i = 0; i < entries; i++) {2673 struct cache_entry *ce = istate->cache[i];2674 if (is_racy_timestamp(istate, ce))2675 return 1;2676 }2677 return 0;2678}26792680void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)2681{2682 if ((istate->cache_changed || has_racy_timestamp(istate)) &&2683 verify_index(istate))2684 write_locked_index(istate, lockfile, COMMIT_LOCK);2685 else2686 rollback_lock_file(lockfile);2687}26882689static int record_eoie(void)2690{2691 int val;26922693 if (!git_config_get_bool("index.recordendofindexentries", &val))2694 return val;26952696 /*2697 * As a convenience, the end of index entries extension2698 * used for threading is written by default if the user2699 * explicitly requested threaded index reads.2700 */2701 return !git_config_get_index_threads(&val) && val != 1;2702}27032704static int record_ieot(void)2705{2706 int val;27072708 if (!git_config_get_bool("index.recordoffsettable", &val))2709 return val;27102711 /*2712 * As a convenience, the offset table used for threading is2713 * written by default if the user explicitly requested2714 * threaded index reads.2715 */2716 return !git_config_get_index_threads(&val) && val != 1;2717}27182719/*2720 * On success, `tempfile` is closed. If it is the temporary file2721 * of a `struct lock_file`, we will therefore effectively perform2722 * a 'close_lock_file_gently()`. Since that is an implementation2723 * detail of lockfiles, callers of `do_write_index()` should not2724 * rely on it.2725 */2726static int do_write_index(struct index_state *istate, struct tempfile *tempfile,2727 int strip_extensions)2728{2729 uint64_t start = getnanotime();2730 int newfd = tempfile->fd;2731 git_hash_ctx c, eoie_c;2732 struct cache_header hdr;2733 int i, err = 0, removed, extended, hdr_version;2734 struct cache_entry **cache = istate->cache;2735 int entries = istate->cache_nr;2736 struct stat st;2737 struct ondisk_cache_entry_extended ondisk;2738 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;2739 int drop_cache_tree = istate->drop_cache_tree;2740 off_t offset;2741 int ieot_entries = 1;2742 struct index_entry_offset_table *ieot = NULL;2743 int nr, nr_threads;27442745 for (i = removed = extended = 0; i < entries; i++) {2746 if (cache[i]->ce_flags & CE_REMOVE)2747 removed++;27482749 /* reduce extended entries if possible */2750 cache[i]->ce_flags &= ~CE_EXTENDED;2751 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {2752 extended++;2753 cache[i]->ce_flags |= CE_EXTENDED;2754 }2755 }27562757 if (!istate->version) {2758 istate->version = get_index_format_default();2759 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))2760 init_split_index(istate);2761 }27622763 /* demote version 3 to version 2 when the latter suffices */2764 if (istate->version == 3 || istate->version == 2)2765 istate->version = extended ? 3 : 2;27662767 hdr_version = istate->version;27682769 hdr.hdr_signature = htonl(CACHE_SIGNATURE);2770 hdr.hdr_version = htonl(hdr_version);2771 hdr.hdr_entries = htonl(entries - removed);27722773 the_hash_algo->init_fn(&c);2774 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)2775 return -1;27762777 if (!HAVE_THREADS || git_config_get_index_threads(&nr_threads))2778 nr_threads = 1;27792780 if (nr_threads != 1 && record_ieot()) {2781 int ieot_blocks, cpus;27822783 /*2784 * ensure default number of ieot blocks maps evenly to the2785 * default number of threads that will process them leaving2786 * room for the thread to load the index extensions.2787 */2788 if (!nr_threads) {2789 ieot_blocks = istate->cache_nr / THREAD_COST;2790 cpus = online_cpus();2791 if (ieot_blocks > cpus - 1)2792 ieot_blocks = cpus - 1;2793 } else {2794 ieot_blocks = nr_threads;2795 if (ieot_blocks > istate->cache_nr)2796 ieot_blocks = istate->cache_nr;2797 }27982799 /*2800 * no reason to write out the IEOT extension if we don't2801 * have enough blocks to utilize multi-threading2802 */2803 if (ieot_blocks > 1) {2804 ieot = xcalloc(1, sizeof(struct index_entry_offset_table)2805 + (ieot_blocks * sizeof(struct index_entry_offset)));2806 ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);2807 }2808 }28092810 offset = lseek(newfd, 0, SEEK_CUR);2811 if (offset < 0) {2812 free(ieot);2813 return -1;2814 }2815 offset += write_buffer_len;2816 nr = 0;2817 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;28182819 for (i = 0; i < entries; i++) {2820 struct cache_entry *ce = cache[i];2821 if (ce->ce_flags & CE_REMOVE)2822 continue;2823 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))2824 ce_smudge_racily_clean_entry(istate, ce);2825 if (is_null_oid(&ce->oid)) {2826 static const char msg[] = "cache entry has null sha1: %s";2827 static int allow = -1;28282829 if (allow < 0)2830 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);2831 if (allow)2832 warning(msg, ce->name);2833 else2834 err = error(msg, ce->name);28352836 drop_cache_tree = 1;2837 }2838 if (ieot && i && (i % ieot_entries == 0)) {2839 ieot->entries[ieot->nr].nr = nr;2840 ieot->entries[ieot->nr].offset = offset;2841 ieot->nr++;2842 /*2843 * If we have a V4 index, set the first byte to an invalid2844 * character to ensure there is nothing common with the previous2845 * entry2846 */2847 if (previous_name)2848 previous_name->buf[0] = 0;2849 nr = 0;2850 offset = lseek(newfd, 0, SEEK_CUR);2851 if (offset < 0) {2852 free(ieot);2853 return -1;2854 }2855 offset += write_buffer_len;2856 }2857 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)2858 err = -1;28592860 if (err)2861 break;2862 nr++;2863 }2864 if (ieot && nr) {2865 ieot->entries[ieot->nr].nr = nr;2866 ieot->entries[ieot->nr].offset = offset;2867 ieot->nr++;2868 }2869 strbuf_release(&previous_name_buf);28702871 if (err) {2872 free(ieot);2873 return err;2874 }28752876 /* Write extension data here */2877 offset = lseek(newfd, 0, SEEK_CUR);2878 if (offset < 0) {2879 free(ieot);2880 return -1;2881 }2882 offset += write_buffer_len;2883 the_hash_algo->init_fn(&eoie_c);28842885 /*2886 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we2887 * can minimize the number of extensions we have to scan through to2888 * find it during load. Write it out regardless of the2889 * strip_extensions parameter as we need it when loading the shared2890 * index.2891 */2892 if (ieot) {2893 struct strbuf sb = STRBUF_INIT;28942895 write_ieot_extension(&sb, ieot);2896 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 02897 || ce_write(&c, newfd, sb.buf, sb.len) < 0;2898 strbuf_release(&sb);2899 free(ieot);2900 if (err)2901 return -1;2902 }29032904 if (!strip_extensions && istate->split_index) {2905 struct strbuf sb = STRBUF_INIT;29062907 err = write_link_extension(&sb, istate) < 0 ||2908 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,2909 sb.len) < 0 ||2910 ce_write(&c, newfd, sb.buf, sb.len) < 0;2911 strbuf_release(&sb);2912 if (err)2913 return -1;2914 }2915 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {2916 struct strbuf sb = STRBUF_INIT;29172918 cache_tree_write(&sb, istate->cache_tree);2919 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 02920 || ce_write(&c, newfd, sb.buf, sb.len) < 0;2921 strbuf_release(&sb);2922 if (err)2923 return -1;2924 }2925 if (!strip_extensions && istate->resolve_undo) {2926 struct strbuf sb = STRBUF_INIT;29272928 resolve_undo_write(&sb, istate->resolve_undo);2929 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,2930 sb.len) < 02931 || ce_write(&c, newfd, sb.buf, sb.len) < 0;2932 strbuf_release(&sb);2933 if (err)2934 return -1;2935 }2936 if (!strip_extensions && istate->untracked) {2937 struct strbuf sb = STRBUF_INIT;29382939 write_untracked_extension(&sb, istate->untracked);2940 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,2941 sb.len) < 0 ||2942 ce_write(&c, newfd, sb.buf, sb.len) < 0;2943 strbuf_release(&sb);2944 if (err)2945 return -1;2946 }2947 if (!strip_extensions && istate->fsmonitor_last_update) {2948 struct strbuf sb = STRBUF_INIT;29492950 write_fsmonitor_extension(&sb, istate);2951 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 02952 || ce_write(&c, newfd, sb.buf, sb.len) < 0;2953 strbuf_release(&sb);2954 if (err)2955 return -1;2956 }29572958 /*2959 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA12960 * so that it can be found and processed before all the index entries are2961 * read. Write it out regardless of the strip_extensions parameter as we need it2962 * when loading the shared index.2963 */2964 if (offset && record_eoie()) {2965 struct strbuf sb = STRBUF_INIT;29662967 write_eoie_extension(&sb, &eoie_c, offset);2968 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 02969 || ce_write(&c, newfd, sb.buf, sb.len) < 0;2970 strbuf_release(&sb);2971 if (err)2972 return -1;2973 }29742975 if (ce_flush(&c, newfd, istate->oid.hash))2976 return -1;2977 if (close_tempfile_gently(tempfile)) {2978 error(_("could not close '%s'"), tempfile->filename.buf);2979 return -1;2980 }2981 if (stat(tempfile->filename.buf, &st))2982 return -1;2983 istate->timestamp.sec = (unsigned int)st.st_mtime;2984 istate->timestamp.nsec = ST_MTIME_NSEC(st);2985 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);2986 return 0;2987}29882989void set_alternate_index_output(const char *name)2990{2991 alternate_index_output = name;2992}29932994static int commit_locked_index(struct lock_file *lk)2995{2996 if (alternate_index_output)2997 return commit_lock_file_to(lk, alternate_index_output);2998 else2999 return commit_lock_file(lk);3000}30013002static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,3003 unsigned flags)3004{3005 int ret = do_write_index(istate, lock->tempfile, 0);3006 if (ret)3007 return ret;3008 if (flags & COMMIT_LOCK)3009 return commit_locked_index(lock);3010 return close_lock_file_gently(lock);3011}30123013static int write_split_index(struct index_state *istate,3014 struct lock_file *lock,3015 unsigned flags)3016{3017 int ret;3018 prepare_to_write_split_index(istate);3019 ret = do_write_locked_index(istate, lock, flags);3020 finish_writing_split_index(istate);3021 return ret;3022}30233024static const char *shared_index_expire = "2.weeks.ago";30253026static unsigned long get_shared_index_expire_date(void)3027{3028 static unsigned long shared_index_expire_date;3029 static int shared_index_expire_date_prepared;30303031 if (!shared_index_expire_date_prepared) {3032 git_config_get_expiry("splitindex.sharedindexexpire",3033 &shared_index_expire);3034 shared_index_expire_date = approxidate(shared_index_expire);3035 shared_index_expire_date_prepared = 1;3036 }30373038 return shared_index_expire_date;3039}30403041static int should_delete_shared_index(const char *shared_index_path)3042{3043 struct stat st;3044 unsigned long expiration;30453046 /* Check timestamp */3047 expiration = get_shared_index_expire_date();3048 if (!expiration)3049 return 0;3050 if (stat(shared_index_path, &st))3051 return error_errno(_("could not stat '%s'"), shared_index_path);3052 if (st.st_mtime > expiration)3053 return 0;30543055 return 1;3056}30573058static int clean_shared_index_files(const char *current_hex)3059{3060 struct dirent *de;3061 DIR *dir = opendir(get_git_dir());30623063 if (!dir)3064 return error_errno(_("unable to open git dir: %s"), get_git_dir());30653066 while ((de = readdir(dir)) != NULL) {3067 const char *sha1_hex;3068 const char *shared_index_path;3069 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))3070 continue;3071 if (!strcmp(sha1_hex, current_hex))3072 continue;3073 shared_index_path = git_path("%s", de->d_name);3074 if (should_delete_shared_index(shared_index_path) > 0 &&3075 unlink(shared_index_path))3076 warning_errno(_("unable to unlink: %s"), shared_index_path);3077 }3078 closedir(dir);30793080 return 0;3081}30823083static int write_shared_index(struct index_state *istate,3084 struct tempfile **temp)3085{3086 struct split_index *si = istate->split_index;3087 int ret;30883089 move_cache_to_base_index(istate);3090 ret = do_write_index(si->base, *temp, 1);3091 if (ret)3092 return ret;3093 ret = adjust_shared_perm(get_tempfile_path(*temp));3094 if (ret) {3095 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));3096 return ret;3097 }3098 ret = rename_tempfile(temp,3099 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));3100 if (!ret) {3101 oidcpy(&si->base_oid, &si->base->oid);3102 clean_shared_index_files(oid_to_hex(&si->base->oid));3103 }31043105 return ret;3106}31073108static const int default_max_percent_split_change = 20;31093110static int too_many_not_shared_entries(struct index_state *istate)3111{3112 int i, not_shared = 0;3113 int max_split = git_config_get_max_percent_split_change();31143115 switch (max_split) {3116 case -1:3117 /* not or badly configured: use the default value */3118 max_split = default_max_percent_split_change;3119 break;3120 case 0:3121 return 1; /* 0% means always write a new shared index */3122 case 100:3123 return 0; /* 100% means never write a new shared index */3124 default:3125 break; /* just use the configured value */3126 }31273128 /* Count not shared entries */3129 for (i = 0; i < istate->cache_nr; i++) {3130 struct cache_entry *ce = istate->cache[i];3131 if (!ce->index)3132 not_shared++;3133 }31343135 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;3136}31373138int write_locked_index(struct index_state *istate, struct lock_file *lock,3139 unsigned flags)3140{3141 int new_shared_index, ret;3142 struct split_index *si = istate->split_index;31433144 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))3145 cache_tree_verify(the_repository, istate);31463147 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {3148 if (flags & COMMIT_LOCK)3149 rollback_lock_file(lock);3150 return 0;3151 }31523153 if (istate->fsmonitor_last_update)3154 fill_fsmonitor_bitmap(istate);31553156 if (!si || alternate_index_output ||3157 (istate->cache_changed & ~EXTMASK)) {3158 if (si)3159 oidclr(&si->base_oid);3160 ret = do_write_locked_index(istate, lock, flags);3161 goto out;3162 }31633164 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {3165 int v = si->base_oid.hash[0];3166 if ((v & 15) < 6)3167 istate->cache_changed |= SPLIT_INDEX_ORDERED;3168 }3169 if (too_many_not_shared_entries(istate))3170 istate->cache_changed |= SPLIT_INDEX_ORDERED;31713172 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;31733174 if (new_shared_index) {3175 struct tempfile *temp;3176 int saved_errno;31773178 /* Same initial permissions as the main .git/index file */3179 temp = mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);3180 if (!temp) {3181 oidclr(&si->base_oid);3182 ret = do_write_locked_index(istate, lock, flags);3183 goto out;3184 }3185 ret = write_shared_index(istate, &temp);31863187 saved_errno = errno;3188 if (is_tempfile_active(temp))3189 delete_tempfile(&temp);3190 errno = saved_errno;31913192 if (ret)3193 goto out;3194 }31953196 ret = write_split_index(istate, lock, flags);31973198 /* Freshen the shared index only if the split-index was written */3199 if (!ret && !new_shared_index) {3200 const char *shared_index = git_path("sharedindex.%s",3201 oid_to_hex(&si->base_oid));3202 freshen_shared_index(shared_index, 1);3203 }32043205out:3206 if (flags & COMMIT_LOCK)3207 rollback_lock_file(lock);3208 return ret;3209}32103211/*3212 * Read the index file that is potentially unmerged into given3213 * index_state, dropping any unmerged entries to stage #0 (potentially3214 * resulting in a path appearing as both a file and a directory in the3215 * index; the caller is responsible to clear out the extra entries3216 * before writing the index to a tree). Returns true if the index is3217 * unmerged. Callers who want to refuse to work from an unmerged3218 * state can call this and check its return value, instead of calling3219 * read_cache().3220 */3221int read_index_unmerged(struct index_state *istate)3222{3223 int i;3224 int unmerged = 0;32253226 read_index(istate);3227 for (i = 0; i < istate->cache_nr; i++) {3228 struct cache_entry *ce = istate->cache[i];3229 struct cache_entry *new_ce;3230 int len;32313232 if (!ce_stage(ce))3233 continue;3234 unmerged = 1;3235 len = ce_namelen(ce);3236 new_ce = make_empty_cache_entry(istate, len);3237 memcpy(new_ce->name, ce->name, len);3238 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;3239 new_ce->ce_namelen = len;3240 new_ce->ce_mode = ce->ce_mode;3241 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))3242 return error(_("%s: cannot drop to stage #0"),3243 new_ce->name);3244 }3245 return unmerged;3246}32473248/*3249 * Returns 1 if the path is an "other" path with respect to3250 * the index; that is, the path is not mentioned in the index at all,3251 * either as a file, a directory with some files in the index,3252 * or as an unmerged entry.3253 *3254 * We helpfully remove a trailing "/" from directories so that3255 * the output of read_directory can be used as-is.3256 */3257int index_name_is_other(const struct index_state *istate, const char *name,3258 int namelen)3259{3260 int pos;3261 if (namelen && name[namelen - 1] == '/')3262 namelen--;3263 pos = index_name_pos(istate, name, namelen);3264 if (0 <= pos)3265 return 0; /* exact match */3266 pos = -pos - 1;3267 if (pos < istate->cache_nr) {3268 struct cache_entry *ce = istate->cache[pos];3269 if (ce_namelen(ce) == namelen &&3270 !memcmp(ce->name, name, namelen))3271 return 0; /* Yup, this one exists unmerged */3272 }3273 return 1;3274}32753276void *read_blob_data_from_index(const struct index_state *istate,3277 const char *path, unsigned long *size)3278{3279 int pos, len;3280 unsigned long sz;3281 enum object_type type;3282 void *data;32833284 len = strlen(path);3285 pos = index_name_pos(istate, path, len);3286 if (pos < 0) {3287 /*3288 * We might be in the middle of a merge, in which3289 * case we would read stage #2 (ours).3290 */3291 int i;3292 for (i = -pos - 1;3293 (pos < 0 && i < istate->cache_nr &&3294 !strcmp(istate->cache[i]->name, path));3295 i++)3296 if (ce_stage(istate->cache[i]) == 2)3297 pos = i;3298 }3299 if (pos < 0)3300 return NULL;3301 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);3302 if (!data || type != OBJ_BLOB) {3303 free(data);3304 return NULL;3305 }3306 if (size)3307 *size = sz;3308 return data;3309}33103311void stat_validity_clear(struct stat_validity *sv)3312{3313 FREE_AND_NULL(sv->sd);3314}33153316int stat_validity_check(struct stat_validity *sv, const char *path)3317{3318 struct stat st;33193320 if (stat(path, &st) < 0)3321 return sv->sd == NULL;3322 if (!sv->sd)3323 return 0;3324 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);3325}33263327void stat_validity_update(struct stat_validity *sv, int fd)3328{3329 struct stat st;33303331 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))3332 stat_validity_clear(sv);3333 else {3334 if (!sv->sd)3335 sv->sd = xcalloc(1, sizeof(struct stat_data));3336 fill_stat_data(sv->sd, &st);3337 }3338}33393340void move_index_extensions(struct index_state *dst, struct index_state *src)3341{3342 dst->untracked = src->untracked;3343 src->untracked = NULL;3344 dst->cache_tree = src->cache_tree;3345 src->cache_tree = NULL;3346}33473348struct cache_entry *dup_cache_entry(const struct cache_entry *ce,3349 struct index_state *istate)3350{3351 unsigned int size = ce_size(ce);3352 int mem_pool_allocated;3353 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));3354 mem_pool_allocated = new_entry->mem_pool_allocated;33553356 memcpy(new_entry, ce, size);3357 new_entry->mem_pool_allocated = mem_pool_allocated;3358 return new_entry;3359}33603361void discard_cache_entry(struct cache_entry *ce)3362{3363 if (ce && should_validate_cache_entries())3364 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));33653366 if (ce && ce->mem_pool_allocated)3367 return;33683369 free(ce);3370}33713372int should_validate_cache_entries(void)3373{3374 static int validate_index_cache_entries = -1;33753376 if (validate_index_cache_entries < 0) {3377 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))3378 validate_index_cache_entries = 1;3379 else3380 validate_index_cache_entries = 0;3381 }33823383 return validate_index_cache_entries;3384}33853386#define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */3387#define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */33883389static size_t read_eoie_extension(const char *mmap, size_t mmap_size)3390{3391 /*3392 * The end of index entries (EOIE) extension is guaranteed to be last3393 * so that it can be found by scanning backwards from the EOF.3394 *3395 * "EOIE"3396 * <4-byte length>3397 * <4-byte offset>3398 * <20-byte hash>3399 */3400 const char *index, *eoie;3401 uint32_t extsize;3402 size_t offset, src_offset;3403 unsigned char hash[GIT_MAX_RAWSZ];3404 git_hash_ctx c;34053406 /* ensure we have an index big enough to contain an EOIE extension */3407 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)3408 return 0;34093410 /* validate the extension signature */3411 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;3412 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)3413 return 0;3414 index += sizeof(uint32_t);34153416 /* validate the extension size */3417 extsize = get_be32(index);3418 if (extsize != EOIE_SIZE)3419 return 0;3420 index += sizeof(uint32_t);34213422 /*3423 * Validate the offset we're going to look for the first extension3424 * signature is after the index header and before the eoie extension.3425 */3426 offset = get_be32(index);3427 if (mmap + offset < mmap + sizeof(struct cache_header))3428 return 0;3429 if (mmap + offset >= eoie)3430 return 0;3431 index += sizeof(uint32_t);34323433 /*3434 * The hash is computed over extension types and their sizes (but not3435 * their contents). E.g. if we have "TREE" extension that is N-bytes3436 * long, "REUC" extension that is M-bytes long, followed by "EOIE",3437 * then the hash would be:3438 *3439 * SHA-1("TREE" + <binary representation of N> +3440 * "REUC" + <binary representation of M>)3441 */3442 src_offset = offset;3443 the_hash_algo->init_fn(&c);3444 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {3445 /* After an array of active_nr index entries,3446 * there can be arbitrary number of extended3447 * sections, each of which is prefixed with3448 * extension name (4-byte) and section length3449 * in 4-byte network byte order.3450 */3451 uint32_t extsize;3452 memcpy(&extsize, mmap + src_offset + 4, 4);3453 extsize = ntohl(extsize);34543455 /* verify the extension size isn't so large it will wrap around */3456 if (src_offset + 8 + extsize < src_offset)3457 return 0;34583459 the_hash_algo->update_fn(&c, mmap + src_offset, 8);34603461 src_offset += 8;3462 src_offset += extsize;3463 }3464 the_hash_algo->final_fn(hash, &c);3465 if (!hasheq(hash, (const unsigned char *)index))3466 return 0;34673468 /* Validate that the extension offsets returned us back to the eoie extension. */3469 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)3470 return 0;34713472 return offset;3473}34743475static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)3476{3477 uint32_t buffer;3478 unsigned char hash[GIT_MAX_RAWSZ];34793480 /* offset */3481 put_be32(&buffer, offset);3482 strbuf_add(sb, &buffer, sizeof(uint32_t));34833484 /* hash */3485 the_hash_algo->final_fn(hash, eoie_context);3486 strbuf_add(sb, hash, the_hash_algo->rawsz);3487}34883489#define IEOT_VERSION (1)34903491static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)3492{3493 const char *index = NULL;3494 uint32_t extsize, ext_version;3495 struct index_entry_offset_table *ieot;3496 int i, nr;34973498 /* find the IEOT extension */3499 if (!offset)3500 return NULL;3501 while (offset <= mmap_size - the_hash_algo->rawsz - 8) {3502 extsize = get_be32(mmap + offset + 4);3503 if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {3504 index = mmap + offset + 4 + 4;3505 break;3506 }3507 offset += 8;3508 offset += extsize;3509 }3510 if (!index)3511 return NULL;35123513 /* validate the version is IEOT_VERSION */3514 ext_version = get_be32(index);3515 if (ext_version != IEOT_VERSION) {3516 error("invalid IEOT version %d", ext_version);3517 return NULL;3518 }3519 index += sizeof(uint32_t);35203521 /* extension size - version bytes / bytes per entry */3522 nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));3523 if (!nr) {3524 error("invalid number of IEOT entries %d", nr);3525 return NULL;3526 }3527 ieot = xmalloc(sizeof(struct index_entry_offset_table)3528 + (nr * sizeof(struct index_entry_offset)));3529 ieot->nr = nr;3530 for (i = 0; i < nr; i++) {3531 ieot->entries[i].offset = get_be32(index);3532 index += sizeof(uint32_t);3533 ieot->entries[i].nr = get_be32(index);3534 index += sizeof(uint32_t);3535 }35363537 return ieot;3538}35393540static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)3541{3542 uint32_t buffer;3543 int i;35443545 /* version */3546 put_be32(&buffer, IEOT_VERSION);3547 strbuf_add(sb, &buffer, sizeof(uint32_t));35483549 /* ieot */3550 for (i = 0; i < ieot->nr; i++) {35513552 /* offset */3553 put_be32(&buffer, ieot->entries[i].offset);3554 strbuf_add(sb, &buffer, sizeof(uint32_t));35553556 /* count */3557 put_be32(&buffer, ieot->entries[i].nr);3558 strbuf_add(sb, &buffer, sizeof(uint32_t));3559 }3560}