read-cache.con commit Merge branch 'mv/pull-r-for-rebase' into maint-1.7.11 (1103996)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 */
   6#define NO_THE_INDEX_COMPATIBILITY_MACROS
   7#include "cache.h"
   8#include "cache-tree.h"
   9#include "refs.h"
  10#include "dir.h"
  11#include "tree.h"
  12#include "commit.h"
  13#include "blob.h"
  14#include "resolve-undo.h"
  15#include "strbuf.h"
  16#include "varint.h"
  17
  18static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really);
  19
  20/* Index extensions.
  21 *
  22 * The first letter should be 'A'..'Z' for extensions that are not
  23 * necessary for a correct operation (i.e. optimization data).
  24 * When new extensions are added that _needs_ to be understood in
  25 * order to correctly interpret the index file, pick character that
  26 * is outside the range, to cause the reader to abort.
  27 */
  28
  29#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  30#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  31#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  32
  33struct index_state the_index;
  34
  35static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  36{
  37        istate->cache[nr] = ce;
  38        add_name_hash(istate, ce);
  39}
  40
  41static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  42{
  43        struct cache_entry *old = istate->cache[nr];
  44
  45        remove_name_hash(old);
  46        set_index_entry(istate, nr, ce);
  47        istate->cache_changed = 1;
  48}
  49
  50void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  51{
  52        struct cache_entry *old = istate->cache[nr], *new;
  53        int namelen = strlen(new_name);
  54
  55        new = xmalloc(cache_entry_size(namelen));
  56        copy_cache_entry(new, old);
  57        new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
  58        new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
  59        memcpy(new->name, new_name, namelen + 1);
  60
  61        cache_tree_invalidate_path(istate->cache_tree, old->name);
  62        remove_index_entry_at(istate, nr);
  63        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  64}
  65
  66/*
  67 * This only updates the "non-critical" parts of the directory
  68 * cache, ie the parts that aren't tracked by GIT, and only used
  69 * to validate the cache.
  70 */
  71void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
  72{
  73        ce->ce_ctime.sec = (unsigned int)st->st_ctime;
  74        ce->ce_mtime.sec = (unsigned int)st->st_mtime;
  75        ce->ce_ctime.nsec = ST_CTIME_NSEC(*st);
  76        ce->ce_mtime.nsec = ST_MTIME_NSEC(*st);
  77        ce->ce_dev = st->st_dev;
  78        ce->ce_ino = st->st_ino;
  79        ce->ce_uid = st->st_uid;
  80        ce->ce_gid = st->st_gid;
  81        ce->ce_size = st->st_size;
  82
  83        if (assume_unchanged)
  84                ce->ce_flags |= CE_VALID;
  85
  86        if (S_ISREG(st->st_mode))
  87                ce_mark_uptodate(ce);
  88}
  89
  90static int ce_compare_data(struct cache_entry *ce, struct stat *st)
  91{
  92        int match = -1;
  93        int fd = open(ce->name, O_RDONLY);
  94
  95        if (fd >= 0) {
  96                unsigned char sha1[20];
  97                if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
  98                        match = hashcmp(sha1, ce->sha1);
  99                /* index_fd() closed the file descriptor already */
 100        }
 101        return match;
 102}
 103
 104static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
 105{
 106        int match = -1;
 107        void *buffer;
 108        unsigned long size;
 109        enum object_type type;
 110        struct strbuf sb = STRBUF_INIT;
 111
 112        if (strbuf_readlink(&sb, ce->name, expected_size))
 113                return -1;
 114
 115        buffer = read_sha1_file(ce->sha1, &type, &size);
 116        if (buffer) {
 117                if (size == sb.len)
 118                        match = memcmp(buffer, sb.buf, size);
 119                free(buffer);
 120        }
 121        strbuf_release(&sb);
 122        return match;
 123}
 124
 125static int ce_compare_gitlink(struct cache_entry *ce)
 126{
 127        unsigned char sha1[20];
 128
 129        /*
 130         * We don't actually require that the .git directory
 131         * under GITLINK directory be a valid git directory. It
 132         * might even be missing (in case nobody populated that
 133         * sub-project).
 134         *
 135         * If so, we consider it always to match.
 136         */
 137        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 138                return 0;
 139        return hashcmp(sha1, ce->sha1);
 140}
 141
 142static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
 143{
 144        switch (st->st_mode & S_IFMT) {
 145        case S_IFREG:
 146                if (ce_compare_data(ce, st))
 147                        return DATA_CHANGED;
 148                break;
 149        case S_IFLNK:
 150                if (ce_compare_link(ce, xsize_t(st->st_size)))
 151                        return DATA_CHANGED;
 152                break;
 153        case S_IFDIR:
 154                if (S_ISGITLINK(ce->ce_mode))
 155                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 156        default:
 157                return TYPE_CHANGED;
 158        }
 159        return 0;
 160}
 161
 162static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 163{
 164        unsigned int changed = 0;
 165
 166        if (ce->ce_flags & CE_REMOVE)
 167                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 168
 169        switch (ce->ce_mode & S_IFMT) {
 170        case S_IFREG:
 171                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 172                /* We consider only the owner x bit to be relevant for
 173                 * "mode changes"
 174                 */
 175                if (trust_executable_bit &&
 176                    (0100 & (ce->ce_mode ^ st->st_mode)))
 177                        changed |= MODE_CHANGED;
 178                break;
 179        case S_IFLNK:
 180                if (!S_ISLNK(st->st_mode) &&
 181                    (has_symlinks || !S_ISREG(st->st_mode)))
 182                        changed |= TYPE_CHANGED;
 183                break;
 184        case S_IFGITLINK:
 185                /* We ignore most of the st_xxx fields for gitlinks */
 186                if (!S_ISDIR(st->st_mode))
 187                        changed |= TYPE_CHANGED;
 188                else if (ce_compare_gitlink(ce))
 189                        changed |= DATA_CHANGED;
 190                return changed;
 191        default:
 192                die("internal error: ce_mode is %o", ce->ce_mode);
 193        }
 194        if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
 195                changed |= MTIME_CHANGED;
 196        if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime)
 197                changed |= CTIME_CHANGED;
 198
 199#ifdef USE_NSEC
 200        if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
 201                changed |= MTIME_CHANGED;
 202        if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
 203                changed |= CTIME_CHANGED;
 204#endif
 205
 206        if (ce->ce_uid != (unsigned int) st->st_uid ||
 207            ce->ce_gid != (unsigned int) st->st_gid)
 208                changed |= OWNER_CHANGED;
 209        if (ce->ce_ino != (unsigned int) st->st_ino)
 210                changed |= INODE_CHANGED;
 211
 212#ifdef USE_STDEV
 213        /*
 214         * st_dev breaks on network filesystems where different
 215         * clients will have different views of what "device"
 216         * the filesystem is on
 217         */
 218        if (ce->ce_dev != (unsigned int) st->st_dev)
 219                changed |= INODE_CHANGED;
 220#endif
 221
 222        if (ce->ce_size != (unsigned int) st->st_size)
 223                changed |= DATA_CHANGED;
 224
 225        /* Racily smudged entry? */
 226        if (!ce->ce_size) {
 227                if (!is_empty_blob_sha1(ce->sha1))
 228                        changed |= DATA_CHANGED;
 229        }
 230
 231        return changed;
 232}
 233
 234static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
 235{
 236        return (!S_ISGITLINK(ce->ce_mode) &&
 237                istate->timestamp.sec &&
 238#ifdef USE_NSEC
 239                 /* nanosecond timestamped files can also be racy! */
 240                (istate->timestamp.sec < ce->ce_mtime.sec ||
 241                 (istate->timestamp.sec == ce->ce_mtime.sec &&
 242                  istate->timestamp.nsec <= ce->ce_mtime.nsec))
 243#else
 244                istate->timestamp.sec <= ce->ce_mtime.sec
 245#endif
 246                 );
 247}
 248
 249int ie_match_stat(const struct index_state *istate,
 250                  struct cache_entry *ce, struct stat *st,
 251                  unsigned int options)
 252{
 253        unsigned int changed;
 254        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 255        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 256        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 257
 258        /*
 259         * If it's marked as always valid in the index, it's
 260         * valid whatever the checked-out copy says.
 261         *
 262         * skip-worktree has the same effect with higher precedence
 263         */
 264        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 265                return 0;
 266        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 267                return 0;
 268
 269        /*
 270         * Intent-to-add entries have not been added, so the index entry
 271         * by definition never matches what is in the work tree until it
 272         * actually gets added.
 273         */
 274        if (ce->ce_flags & CE_INTENT_TO_ADD)
 275                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 276
 277        changed = ce_match_stat_basic(ce, st);
 278
 279        /*
 280         * Within 1 second of this sequence:
 281         *      echo xyzzy >file && git-update-index --add file
 282         * running this command:
 283         *      echo frotz >file
 284         * would give a falsely clean cache entry.  The mtime and
 285         * length match the cache, and other stat fields do not change.
 286         *
 287         * We could detect this at update-index time (the cache entry
 288         * being registered/updated records the same time as "now")
 289         * and delay the return from git-update-index, but that would
 290         * effectively mean we can make at most one commit per second,
 291         * which is not acceptable.  Instead, we check cache entries
 292         * whose mtime are the same as the index file timestamp more
 293         * carefully than others.
 294         */
 295        if (!changed && is_racy_timestamp(istate, ce)) {
 296                if (assume_racy_is_modified)
 297                        changed |= DATA_CHANGED;
 298                else
 299                        changed |= ce_modified_check_fs(ce, st);
 300        }
 301
 302        return changed;
 303}
 304
 305int ie_modified(const struct index_state *istate,
 306                struct cache_entry *ce, struct stat *st, unsigned int options)
 307{
 308        int changed, changed_fs;
 309
 310        changed = ie_match_stat(istate, ce, st, options);
 311        if (!changed)
 312                return 0;
 313        /*
 314         * If the mode or type has changed, there's no point in trying
 315         * to refresh the entry - it's not going to match
 316         */
 317        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 318                return changed;
 319
 320        /*
 321         * Immediately after read-tree or update-index --cacheinfo,
 322         * the length field is zero, as we have never even read the
 323         * lstat(2) information once, and we cannot trust DATA_CHANGED
 324         * returned by ie_match_stat() which in turn was returned by
 325         * ce_match_stat_basic() to signal that the filesize of the
 326         * blob changed.  We have to actually go to the filesystem to
 327         * see if the contents match, and if so, should answer "unchanged".
 328         *
 329         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 330         * already has checked the actual HEAD from the filesystem in the
 331         * subproject.  If ie_match_stat() already said it is different,
 332         * then we know it is.
 333         */
 334        if ((changed & DATA_CHANGED) &&
 335            (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
 336                return changed;
 337
 338        changed_fs = ce_modified_check_fs(ce, st);
 339        if (changed_fs)
 340                return changed | changed_fs;
 341        return 0;
 342}
 343
 344int base_name_compare(const char *name1, int len1, int mode1,
 345                      const char *name2, int len2, int mode2)
 346{
 347        unsigned char c1, c2;
 348        int len = len1 < len2 ? len1 : len2;
 349        int cmp;
 350
 351        cmp = memcmp(name1, name2, len);
 352        if (cmp)
 353                return cmp;
 354        c1 = name1[len];
 355        c2 = name2[len];
 356        if (!c1 && S_ISDIR(mode1))
 357                c1 = '/';
 358        if (!c2 && S_ISDIR(mode2))
 359                c2 = '/';
 360        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 361}
 362
 363/*
 364 * df_name_compare() is identical to base_name_compare(), except it
 365 * compares conflicting directory/file entries as equal. Note that
 366 * while a directory name compares as equal to a regular file, they
 367 * then individually compare _differently_ to a filename that has
 368 * a dot after the basename (because '\0' < '.' < '/').
 369 *
 370 * This is used by routines that want to traverse the git namespace
 371 * but then handle conflicting entries together when possible.
 372 */
 373int df_name_compare(const char *name1, int len1, int mode1,
 374                    const char *name2, int len2, int mode2)
 375{
 376        int len = len1 < len2 ? len1 : len2, cmp;
 377        unsigned char c1, c2;
 378
 379        cmp = memcmp(name1, name2, len);
 380        if (cmp)
 381                return cmp;
 382        /* Directories and files compare equal (same length, same name) */
 383        if (len1 == len2)
 384                return 0;
 385        c1 = name1[len];
 386        if (!c1 && S_ISDIR(mode1))
 387                c1 = '/';
 388        c2 = name2[len];
 389        if (!c2 && S_ISDIR(mode2))
 390                c2 = '/';
 391        if (c1 == '/' && !c2)
 392                return 0;
 393        if (c2 == '/' && !c1)
 394                return 0;
 395        return c1 - c2;
 396}
 397
 398int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 399{
 400        int len1, len2, len, cmp;
 401
 402        len1 = flags1 & CE_NAMEMASK;
 403        if (CE_NAMEMASK <= len1)
 404                len1 = strlen(name1 + CE_NAMEMASK) + CE_NAMEMASK;
 405        len2 = flags2 & CE_NAMEMASK;
 406        if (CE_NAMEMASK <= len2)
 407                len2 = strlen(name2 + CE_NAMEMASK) + CE_NAMEMASK;
 408        len = len1 < len2 ? len1 : len2;
 409
 410        cmp = memcmp(name1, name2, len);
 411        if (cmp)
 412                return cmp;
 413        if (len1 < len2)
 414                return -1;
 415        if (len1 > len2)
 416                return 1;
 417
 418        /* Compare stages  */
 419        flags1 &= CE_STAGEMASK;
 420        flags2 &= CE_STAGEMASK;
 421
 422        if (flags1 < flags2)
 423                return -1;
 424        if (flags1 > flags2)
 425                return 1;
 426        return 0;
 427}
 428
 429int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 430{
 431        int first, last;
 432
 433        first = 0;
 434        last = istate->cache_nr;
 435        while (last > first) {
 436                int next = (last + first) >> 1;
 437                struct cache_entry *ce = istate->cache[next];
 438                int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
 439                if (!cmp)
 440                        return next;
 441                if (cmp < 0) {
 442                        last = next;
 443                        continue;
 444                }
 445                first = next+1;
 446        }
 447        return -first-1;
 448}
 449
 450/* Remove entry, return true if there are more entries to go.. */
 451int remove_index_entry_at(struct index_state *istate, int pos)
 452{
 453        struct cache_entry *ce = istate->cache[pos];
 454
 455        record_resolve_undo(istate, ce);
 456        remove_name_hash(ce);
 457        istate->cache_changed = 1;
 458        istate->cache_nr--;
 459        if (pos >= istate->cache_nr)
 460                return 0;
 461        memmove(istate->cache + pos,
 462                istate->cache + pos + 1,
 463                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 464        return 1;
 465}
 466
 467/*
 468 * Remove all cache ententries marked for removal, that is where
 469 * CE_REMOVE is set in ce_flags.  This is much more effective than
 470 * calling remove_index_entry_at() for each entry to be removed.
 471 */
 472void remove_marked_cache_entries(struct index_state *istate)
 473{
 474        struct cache_entry **ce_array = istate->cache;
 475        unsigned int i, j;
 476
 477        for (i = j = 0; i < istate->cache_nr; i++) {
 478                if (ce_array[i]->ce_flags & CE_REMOVE)
 479                        remove_name_hash(ce_array[i]);
 480                else
 481                        ce_array[j++] = ce_array[i];
 482        }
 483        istate->cache_changed = 1;
 484        istate->cache_nr = j;
 485}
 486
 487int remove_file_from_index(struct index_state *istate, const char *path)
 488{
 489        int pos = index_name_pos(istate, path, strlen(path));
 490        if (pos < 0)
 491                pos = -pos-1;
 492        cache_tree_invalidate_path(istate->cache_tree, path);
 493        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 494                remove_index_entry_at(istate, pos);
 495        return 0;
 496}
 497
 498static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 499{
 500        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 501}
 502
 503static int index_name_pos_also_unmerged(struct index_state *istate,
 504        const char *path, int namelen)
 505{
 506        int pos = index_name_pos(istate, path, namelen);
 507        struct cache_entry *ce;
 508
 509        if (pos >= 0)
 510                return pos;
 511
 512        /* maybe unmerged? */
 513        pos = -1 - pos;
 514        if (pos >= istate->cache_nr ||
 515                        compare_name((ce = istate->cache[pos]), path, namelen))
 516                return -1;
 517
 518        /* order of preference: stage 2, 1, 3 */
 519        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 520                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 521                        !compare_name(ce, path, namelen))
 522                pos++;
 523        return pos;
 524}
 525
 526static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 527{
 528        int len = ce_namelen(ce);
 529        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 530}
 531
 532/*
 533 * If we add a filename that aliases in the cache, we will use the
 534 * name that we already have - but we don't want to update the same
 535 * alias twice, because that implies that there were actually two
 536 * different files with aliasing names!
 537 *
 538 * So we use the CE_ADDED flag to verify that the alias was an old
 539 * one before we accept it as
 540 */
 541static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
 542{
 543        int len;
 544        struct cache_entry *new;
 545
 546        if (alias->ce_flags & CE_ADDED)
 547                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 548
 549        /* Ok, create the new entry using the name of the existing alias */
 550        len = ce_namelen(alias);
 551        new = xcalloc(1, cache_entry_size(len));
 552        memcpy(new->name, alias->name, len);
 553        copy_cache_entry(new, ce);
 554        free(ce);
 555        return new;
 556}
 557
 558static void record_intent_to_add(struct cache_entry *ce)
 559{
 560        unsigned char sha1[20];
 561        if (write_sha1_file("", 0, blob_type, sha1))
 562                die("cannot create an empty blob in the object database");
 563        hashcpy(ce->sha1, sha1);
 564}
 565
 566int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 567{
 568        int size, namelen, was_same;
 569        mode_t st_mode = st->st_mode;
 570        struct cache_entry *ce, *alias;
 571        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 572        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 573        int pretend = flags & ADD_CACHE_PRETEND;
 574        int intent_only = flags & ADD_CACHE_INTENT;
 575        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 576                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 577
 578        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 579                return error("%s: can only add regular files, symbolic links or git-directories", path);
 580
 581        namelen = strlen(path);
 582        if (S_ISDIR(st_mode)) {
 583                while (namelen && path[namelen-1] == '/')
 584                        namelen--;
 585        }
 586        size = cache_entry_size(namelen);
 587        ce = xcalloc(1, size);
 588        memcpy(ce->name, path, namelen);
 589        ce->ce_flags = namelen;
 590        if (!intent_only)
 591                fill_stat_cache_info(ce, st);
 592        else
 593                ce->ce_flags |= CE_INTENT_TO_ADD;
 594
 595        if (trust_executable_bit && has_symlinks)
 596                ce->ce_mode = create_ce_mode(st_mode);
 597        else {
 598                /* If there is an existing entry, pick the mode bits and type
 599                 * from it, otherwise assume unexecutable regular file.
 600                 */
 601                struct cache_entry *ent;
 602                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 603
 604                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 605                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 606        }
 607
 608        /* When core.ignorecase=true, determine if a directory of the same name but differing
 609         * case already exists within the Git repository.  If it does, ensure the directory
 610         * case of the file being added to the repository matches (is folded into) the existing
 611         * entry's directory case.
 612         */
 613        if (ignore_case) {
 614                const char *startPtr = ce->name;
 615                const char *ptr = startPtr;
 616                while (*ptr) {
 617                        while (*ptr && *ptr != '/')
 618                                ++ptr;
 619                        if (*ptr == '/') {
 620                                struct cache_entry *foundce;
 621                                ++ptr;
 622                                foundce = index_name_exists(&the_index, ce->name, ptr - ce->name, ignore_case);
 623                                if (foundce) {
 624                                        memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
 625                                        startPtr = ptr;
 626                                }
 627                        }
 628                }
 629        }
 630
 631        alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 632        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 633                /* Nothing changed, really */
 634                free(ce);
 635                if (!S_ISGITLINK(alias->ce_mode))
 636                        ce_mark_uptodate(alias);
 637                alias->ce_flags |= CE_ADDED;
 638                return 0;
 639        }
 640        if (!intent_only) {
 641                if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
 642                        return error("unable to index file %s", path);
 643        } else
 644                record_intent_to_add(ce);
 645
 646        if (ignore_case && alias && different_name(ce, alias))
 647                ce = create_alias_ce(ce, alias);
 648        ce->ce_flags |= CE_ADDED;
 649
 650        /* It was suspected to be racily clean, but it turns out to be Ok */
 651        was_same = (alias &&
 652                    !ce_stage(alias) &&
 653                    !hashcmp(alias->sha1, ce->sha1) &&
 654                    ce->ce_mode == alias->ce_mode);
 655
 656        if (pretend)
 657                ;
 658        else if (add_index_entry(istate, ce, add_option))
 659                return error("unable to add %s to index",path);
 660        if (verbose && !was_same)
 661                printf("add '%s'\n", path);
 662        return 0;
 663}
 664
 665int add_file_to_index(struct index_state *istate, const char *path, int flags)
 666{
 667        struct stat st;
 668        if (lstat(path, &st))
 669                die_errno("unable to stat '%s'", path);
 670        return add_to_index(istate, path, &st, flags);
 671}
 672
 673struct cache_entry *make_cache_entry(unsigned int mode,
 674                const unsigned char *sha1, const char *path, int stage,
 675                int refresh)
 676{
 677        int size, len;
 678        struct cache_entry *ce;
 679
 680        if (!verify_path(path)) {
 681                error("Invalid path '%s'", path);
 682                return NULL;
 683        }
 684
 685        len = strlen(path);
 686        size = cache_entry_size(len);
 687        ce = xcalloc(1, size);
 688
 689        hashcpy(ce->sha1, sha1);
 690        memcpy(ce->name, path, len);
 691        ce->ce_flags = create_ce_flags(len, stage);
 692        ce->ce_mode = create_ce_mode(mode);
 693
 694        if (refresh)
 695                return refresh_cache_entry(ce, 0);
 696
 697        return ce;
 698}
 699
 700int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 701{
 702        int len = ce_namelen(a);
 703        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 704}
 705
 706int ce_path_match(const struct cache_entry *ce, const struct pathspec *pathspec)
 707{
 708        return match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, NULL);
 709}
 710
 711/*
 712 * We fundamentally don't like some paths: we don't want
 713 * dot or dot-dot anywhere, and for obvious reasons don't
 714 * want to recurse into ".git" either.
 715 *
 716 * Also, we don't want double slashes or slashes at the
 717 * end that can make pathnames ambiguous.
 718 */
 719static int verify_dotfile(const char *rest)
 720{
 721        /*
 722         * The first character was '.', but that
 723         * has already been discarded, we now test
 724         * the rest.
 725         */
 726
 727        /* "." is not allowed */
 728        if (*rest == '\0' || is_dir_sep(*rest))
 729                return 0;
 730
 731        switch (*rest) {
 732        /*
 733         * ".git" followed by  NUL or slash is bad. This
 734         * shares the path end test with the ".." case.
 735         */
 736        case 'g':
 737                if (rest[1] != 'i')
 738                        break;
 739                if (rest[2] != 't')
 740                        break;
 741                rest += 2;
 742        /* fallthrough */
 743        case '.':
 744                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 745                        return 0;
 746        }
 747        return 1;
 748}
 749
 750int verify_path(const char *path)
 751{
 752        char c;
 753
 754        if (has_dos_drive_prefix(path))
 755                return 0;
 756
 757        goto inside;
 758        for (;;) {
 759                if (!c)
 760                        return 1;
 761                if (is_dir_sep(c)) {
 762inside:
 763                        c = *path++;
 764                        if ((c == '.' && !verify_dotfile(path)) ||
 765                            is_dir_sep(c) || c == '\0')
 766                                return 0;
 767                }
 768                c = *path++;
 769        }
 770}
 771
 772/*
 773 * Do we have another file that has the beginning components being a
 774 * proper superset of the name we're trying to add?
 775 */
 776static int has_file_name(struct index_state *istate,
 777                         const struct cache_entry *ce, int pos, int ok_to_replace)
 778{
 779        int retval = 0;
 780        int len = ce_namelen(ce);
 781        int stage = ce_stage(ce);
 782        const char *name = ce->name;
 783
 784        while (pos < istate->cache_nr) {
 785                struct cache_entry *p = istate->cache[pos++];
 786
 787                if (len >= ce_namelen(p))
 788                        break;
 789                if (memcmp(name, p->name, len))
 790                        break;
 791                if (ce_stage(p) != stage)
 792                        continue;
 793                if (p->name[len] != '/')
 794                        continue;
 795                if (p->ce_flags & CE_REMOVE)
 796                        continue;
 797                retval = -1;
 798                if (!ok_to_replace)
 799                        break;
 800                remove_index_entry_at(istate, --pos);
 801        }
 802        return retval;
 803}
 804
 805/*
 806 * Do we have another file with a pathname that is a proper
 807 * subset of the name we're trying to add?
 808 */
 809static int has_dir_name(struct index_state *istate,
 810                        const struct cache_entry *ce, int pos, int ok_to_replace)
 811{
 812        int retval = 0;
 813        int stage = ce_stage(ce);
 814        const char *name = ce->name;
 815        const char *slash = name + ce_namelen(ce);
 816
 817        for (;;) {
 818                int len;
 819
 820                for (;;) {
 821                        if (*--slash == '/')
 822                                break;
 823                        if (slash <= ce->name)
 824                                return retval;
 825                }
 826                len = slash - name;
 827
 828                pos = index_name_pos(istate, name, create_ce_flags(len, stage));
 829                if (pos >= 0) {
 830                        /*
 831                         * Found one, but not so fast.  This could
 832                         * be a marker that says "I was here, but
 833                         * I am being removed".  Such an entry is
 834                         * not a part of the resulting tree, and
 835                         * it is Ok to have a directory at the same
 836                         * path.
 837                         */
 838                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 839                                retval = -1;
 840                                if (!ok_to_replace)
 841                                        break;
 842                                remove_index_entry_at(istate, pos);
 843                                continue;
 844                        }
 845                }
 846                else
 847                        pos = -pos-1;
 848
 849                /*
 850                 * Trivial optimization: if we find an entry that
 851                 * already matches the sub-directory, then we know
 852                 * we're ok, and we can exit.
 853                 */
 854                while (pos < istate->cache_nr) {
 855                        struct cache_entry *p = istate->cache[pos];
 856                        if ((ce_namelen(p) <= len) ||
 857                            (p->name[len] != '/') ||
 858                            memcmp(p->name, name, len))
 859                                break; /* not our subdirectory */
 860                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 861                                /*
 862                                 * p is at the same stage as our entry, and
 863                                 * is a subdirectory of what we are looking
 864                                 * at, so we cannot have conflicts at our
 865                                 * level or anything shorter.
 866                                 */
 867                                return retval;
 868                        pos++;
 869                }
 870        }
 871        return retval;
 872}
 873
 874/* We may be in a situation where we already have path/file and path
 875 * is being added, or we already have path and path/file is being
 876 * added.  Either one would result in a nonsense tree that has path
 877 * twice when git-write-tree tries to write it out.  Prevent it.
 878 *
 879 * If ok-to-replace is specified, we remove the conflicting entries
 880 * from the cache so the caller should recompute the insert position.
 881 * When this happens, we return non-zero.
 882 */
 883static int check_file_directory_conflict(struct index_state *istate,
 884                                         const struct cache_entry *ce,
 885                                         int pos, int ok_to_replace)
 886{
 887        int retval;
 888
 889        /*
 890         * When ce is an "I am going away" entry, we allow it to be added
 891         */
 892        if (ce->ce_flags & CE_REMOVE)
 893                return 0;
 894
 895        /*
 896         * We check if the path is a sub-path of a subsequent pathname
 897         * first, since removing those will not change the position
 898         * in the array.
 899         */
 900        retval = has_file_name(istate, ce, pos, ok_to_replace);
 901
 902        /*
 903         * Then check if the path might have a clashing sub-directory
 904         * before it.
 905         */
 906        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 907}
 908
 909static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 910{
 911        int pos;
 912        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 913        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 914        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 915        int new_only = option & ADD_CACHE_NEW_ONLY;
 916
 917        cache_tree_invalidate_path(istate->cache_tree, ce->name);
 918        pos = index_name_pos(istate, ce->name, ce->ce_flags);
 919
 920        /* existing match? Just replace it. */
 921        if (pos >= 0) {
 922                if (!new_only)
 923                        replace_index_entry(istate, pos, ce);
 924                return 0;
 925        }
 926        pos = -pos-1;
 927
 928        /*
 929         * Inserting a merged entry ("stage 0") into the index
 930         * will always replace all non-merged entries..
 931         */
 932        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 933                while (ce_same_name(istate->cache[pos], ce)) {
 934                        ok_to_add = 1;
 935                        if (!remove_index_entry_at(istate, pos))
 936                                break;
 937                }
 938        }
 939
 940        if (!ok_to_add)
 941                return -1;
 942        if (!verify_path(ce->name))
 943                return error("Invalid path '%s'", ce->name);
 944
 945        if (!skip_df_check &&
 946            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 947                if (!ok_to_replace)
 948                        return error("'%s' appears as both a file and as a directory",
 949                                     ce->name);
 950                pos = index_name_pos(istate, ce->name, ce->ce_flags);
 951                pos = -pos-1;
 952        }
 953        return pos + 1;
 954}
 955
 956int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 957{
 958        int pos;
 959
 960        if (option & ADD_CACHE_JUST_APPEND)
 961                pos = istate->cache_nr;
 962        else {
 963                int ret;
 964                ret = add_index_entry_with_check(istate, ce, option);
 965                if (ret <= 0)
 966                        return ret;
 967                pos = ret - 1;
 968        }
 969
 970        /* Make sure the array is big enough .. */
 971        if (istate->cache_nr == istate->cache_alloc) {
 972                istate->cache_alloc = alloc_nr(istate->cache_alloc);
 973                istate->cache = xrealloc(istate->cache,
 974                                        istate->cache_alloc * sizeof(struct cache_entry *));
 975        }
 976
 977        /* Add it in.. */
 978        istate->cache_nr++;
 979        if (istate->cache_nr > pos + 1)
 980                memmove(istate->cache + pos + 1,
 981                        istate->cache + pos,
 982                        (istate->cache_nr - pos - 1) * sizeof(ce));
 983        set_index_entry(istate, pos, ce);
 984        istate->cache_changed = 1;
 985        return 0;
 986}
 987
 988/*
 989 * "refresh" does not calculate a new sha1 file or bring the
 990 * cache up-to-date for mode/content changes. But what it
 991 * _does_ do is to "re-match" the stat information of a file
 992 * with the cache, so that you can refresh the cache for a
 993 * file that hasn't been changed but where the stat entry is
 994 * out of date.
 995 *
 996 * For example, you'd want to do this after doing a "git-read-tree",
 997 * to link up the stat cache details with the proper files.
 998 */
 999static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1000                                             struct cache_entry *ce,
1001                                             unsigned int options, int *err,
1002                                             int *changed_ret)
1003{
1004        struct stat st;
1005        struct cache_entry *updated;
1006        int changed, size;
1007        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1008        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1009
1010        if (ce_uptodate(ce))
1011                return ce;
1012
1013        /*
1014         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1015         * that the change to the work tree does not matter and told
1016         * us not to worry.
1017         */
1018        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1019                ce_mark_uptodate(ce);
1020                return ce;
1021        }
1022        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1023                ce_mark_uptodate(ce);
1024                return ce;
1025        }
1026
1027        if (lstat(ce->name, &st) < 0) {
1028                if (err)
1029                        *err = errno;
1030                return NULL;
1031        }
1032
1033        changed = ie_match_stat(istate, ce, &st, options);
1034        if (changed_ret)
1035                *changed_ret = changed;
1036        if (!changed) {
1037                /*
1038                 * The path is unchanged.  If we were told to ignore
1039                 * valid bit, then we did the actual stat check and
1040                 * found that the entry is unmodified.  If the entry
1041                 * is not marked VALID, this is the place to mark it
1042                 * valid again, under "assume unchanged" mode.
1043                 */
1044                if (ignore_valid && assume_unchanged &&
1045                    !(ce->ce_flags & CE_VALID))
1046                        ; /* mark this one VALID again */
1047                else {
1048                        /*
1049                         * We do not mark the index itself "modified"
1050                         * because CE_UPTODATE flag is in-core only;
1051                         * we are not going to write this change out.
1052                         */
1053                        if (!S_ISGITLINK(ce->ce_mode))
1054                                ce_mark_uptodate(ce);
1055                        return ce;
1056                }
1057        }
1058
1059        if (ie_modified(istate, ce, &st, options)) {
1060                if (err)
1061                        *err = EINVAL;
1062                return NULL;
1063        }
1064
1065        size = ce_size(ce);
1066        updated = xmalloc(size);
1067        memcpy(updated, ce, size);
1068        fill_stat_cache_info(updated, &st);
1069        /*
1070         * If ignore_valid is not set, we should leave CE_VALID bit
1071         * alone.  Otherwise, paths marked with --no-assume-unchanged
1072         * (i.e. things to be edited) will reacquire CE_VALID bit
1073         * automatically, which is not really what we want.
1074         */
1075        if (!ignore_valid && assume_unchanged &&
1076            !(ce->ce_flags & CE_VALID))
1077                updated->ce_flags &= ~CE_VALID;
1078
1079        return updated;
1080}
1081
1082static void show_file(const char * fmt, const char * name, int in_porcelain,
1083                      int * first, const char *header_msg)
1084{
1085        if (in_porcelain && *first && header_msg) {
1086                printf("%s\n", header_msg);
1087                *first = 0;
1088        }
1089        printf(fmt, name);
1090}
1091
1092int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1093                  char *seen, const char *header_msg)
1094{
1095        int i;
1096        int has_errors = 0;
1097        int really = (flags & REFRESH_REALLY) != 0;
1098        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1099        int quiet = (flags & REFRESH_QUIET) != 0;
1100        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1101        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1102        int first = 1;
1103        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1104        unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1105        const char *modified_fmt;
1106        const char *deleted_fmt;
1107        const char *typechange_fmt;
1108        const char *added_fmt;
1109        const char *unmerged_fmt;
1110
1111        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1112        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1113        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1114        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1115        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1116        for (i = 0; i < istate->cache_nr; i++) {
1117                struct cache_entry *ce, *new;
1118                int cache_errno = 0;
1119                int changed = 0;
1120                int filtered = 0;
1121
1122                ce = istate->cache[i];
1123                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1124                        continue;
1125
1126                if (pathspec &&
1127                    !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1128                        filtered = 1;
1129
1130                if (ce_stage(ce)) {
1131                        while ((i < istate->cache_nr) &&
1132                               ! strcmp(istate->cache[i]->name, ce->name))
1133                                i++;
1134                        i--;
1135                        if (allow_unmerged)
1136                                continue;
1137                        if (!filtered)
1138                                show_file(unmerged_fmt, ce->name, in_porcelain,
1139                                          &first, header_msg);
1140                        has_errors = 1;
1141                        continue;
1142                }
1143
1144                if (filtered)
1145                        continue;
1146
1147                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1148                if (new == ce)
1149                        continue;
1150                if (!new) {
1151                        const char *fmt;
1152
1153                        if (not_new && cache_errno == ENOENT)
1154                                continue;
1155                        if (really && cache_errno == EINVAL) {
1156                                /* If we are doing --really-refresh that
1157                                 * means the index is not valid anymore.
1158                                 */
1159                                ce->ce_flags &= ~CE_VALID;
1160                                istate->cache_changed = 1;
1161                        }
1162                        if (quiet)
1163                                continue;
1164
1165                        if (cache_errno == ENOENT)
1166                                fmt = deleted_fmt;
1167                        else if (ce->ce_flags & CE_INTENT_TO_ADD)
1168                                fmt = added_fmt; /* must be before other checks */
1169                        else if (changed & TYPE_CHANGED)
1170                                fmt = typechange_fmt;
1171                        else
1172                                fmt = modified_fmt;
1173                        show_file(fmt,
1174                                  ce->name, in_porcelain, &first, header_msg);
1175                        has_errors = 1;
1176                        continue;
1177                }
1178
1179                replace_index_entry(istate, i, new);
1180        }
1181        return has_errors;
1182}
1183
1184static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1185{
1186        return refresh_cache_ent(&the_index, ce, really, NULL, NULL);
1187}
1188
1189
1190/*****************************************************************
1191 * Index File I/O
1192 *****************************************************************/
1193
1194#define INDEX_FORMAT_DEFAULT 3
1195
1196/*
1197 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1198 * Again - this is just a (very strong in practice) heuristic that
1199 * the inode hasn't changed.
1200 *
1201 * We save the fields in big-endian order to allow using the
1202 * index file over NFS transparently.
1203 */
1204struct ondisk_cache_entry {
1205        struct cache_time ctime;
1206        struct cache_time mtime;
1207        unsigned int dev;
1208        unsigned int ino;
1209        unsigned int mode;
1210        unsigned int uid;
1211        unsigned int gid;
1212        unsigned int size;
1213        unsigned char sha1[20];
1214        unsigned short flags;
1215        char name[FLEX_ARRAY]; /* more */
1216};
1217
1218/*
1219 * This struct is used when CE_EXTENDED bit is 1
1220 * The struct must match ondisk_cache_entry exactly from
1221 * ctime till flags
1222 */
1223struct ondisk_cache_entry_extended {
1224        struct cache_time ctime;
1225        struct cache_time mtime;
1226        unsigned int dev;
1227        unsigned int ino;
1228        unsigned int mode;
1229        unsigned int uid;
1230        unsigned int gid;
1231        unsigned int size;
1232        unsigned char sha1[20];
1233        unsigned short flags;
1234        unsigned short flags2;
1235        char name[FLEX_ARRAY]; /* more */
1236};
1237
1238/* These are only used for v3 or lower */
1239#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1240#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1241#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1242#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1243                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1244                            ondisk_cache_entry_size(ce_namelen(ce)))
1245
1246static int verify_hdr(struct cache_header *hdr, unsigned long size)
1247{
1248        git_SHA_CTX c;
1249        unsigned char sha1[20];
1250        int hdr_version;
1251
1252        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1253                return error("bad signature");
1254        hdr_version = ntohl(hdr->hdr_version);
1255        if (hdr_version < 2 || 4 < hdr_version)
1256                return error("bad index version %d", hdr_version);
1257        git_SHA1_Init(&c);
1258        git_SHA1_Update(&c, hdr, size - 20);
1259        git_SHA1_Final(sha1, &c);
1260        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1261                return error("bad index file sha1 signature");
1262        return 0;
1263}
1264
1265static int read_index_extension(struct index_state *istate,
1266                                const char *ext, void *data, unsigned long sz)
1267{
1268        switch (CACHE_EXT(ext)) {
1269        case CACHE_EXT_TREE:
1270                istate->cache_tree = cache_tree_read(data, sz);
1271                break;
1272        case CACHE_EXT_RESOLVE_UNDO:
1273                istate->resolve_undo = resolve_undo_read(data, sz);
1274                break;
1275        default:
1276                if (*ext < 'A' || 'Z' < *ext)
1277                        return error("index uses %.4s extension, which we do not understand",
1278                                     ext);
1279                fprintf(stderr, "ignoring %.4s extension\n", ext);
1280                break;
1281        }
1282        return 0;
1283}
1284
1285int read_index(struct index_state *istate)
1286{
1287        return read_index_from(istate, get_index_file());
1288}
1289
1290#ifndef NEEDS_ALIGNED_ACCESS
1291#define ntoh_s(var) ntohs(var)
1292#define ntoh_l(var) ntohl(var)
1293#else
1294static inline uint16_t ntoh_s_force_align(void *p)
1295{
1296        uint16_t x;
1297        memcpy(&x, p, sizeof(x));
1298        return ntohs(x);
1299}
1300static inline uint32_t ntoh_l_force_align(void *p)
1301{
1302        uint32_t x;
1303        memcpy(&x, p, sizeof(x));
1304        return ntohl(x);
1305}
1306#define ntoh_s(var) ntoh_s_force_align(&(var))
1307#define ntoh_l(var) ntoh_l_force_align(&(var))
1308#endif
1309
1310static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1311                                                   unsigned int flags,
1312                                                   const char *name,
1313                                                   size_t len)
1314{
1315        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1316
1317        ce->ce_ctime.sec = ntoh_l(ondisk->ctime.sec);
1318        ce->ce_mtime.sec = ntoh_l(ondisk->mtime.sec);
1319        ce->ce_ctime.nsec = ntoh_l(ondisk->ctime.nsec);
1320        ce->ce_mtime.nsec = ntoh_l(ondisk->mtime.nsec);
1321        ce->ce_dev   = ntoh_l(ondisk->dev);
1322        ce->ce_ino   = ntoh_l(ondisk->ino);
1323        ce->ce_mode  = ntoh_l(ondisk->mode);
1324        ce->ce_uid   = ntoh_l(ondisk->uid);
1325        ce->ce_gid   = ntoh_l(ondisk->gid);
1326        ce->ce_size  = ntoh_l(ondisk->size);
1327        ce->ce_flags = flags;
1328        hashcpy(ce->sha1, ondisk->sha1);
1329        memcpy(ce->name, name, len);
1330        ce->name[len] = '\0';
1331        return ce;
1332}
1333
1334/*
1335 * Adjacent cache entries tend to share the leading paths, so it makes
1336 * sense to only store the differences in later entries.  In the v4
1337 * on-disk format of the index, each on-disk cache entry stores the
1338 * number of bytes to be stripped from the end of the previous name,
1339 * and the bytes to append to the result, to come up with its name.
1340 */
1341static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1342{
1343        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1344        size_t len = decode_varint(&cp);
1345
1346        if (name->len < len)
1347                die("malformed name field in the index");
1348        strbuf_remove(name, name->len - len, len);
1349        for (ep = cp; *ep; ep++)
1350                ; /* find the end */
1351        strbuf_add(name, cp, ep - cp);
1352        return (const char *)ep + 1 - cp_;
1353}
1354
1355static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1356                                            unsigned long *ent_size,
1357                                            struct strbuf *previous_name)
1358{
1359        struct cache_entry *ce;
1360        size_t len;
1361        const char *name;
1362        unsigned int flags;
1363
1364        /* On-disk flags are just 16 bits */
1365        flags = ntoh_s(ondisk->flags);
1366        len = flags & CE_NAMEMASK;
1367
1368        if (flags & CE_EXTENDED) {
1369                struct ondisk_cache_entry_extended *ondisk2;
1370                int extended_flags;
1371                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1372                extended_flags = ntoh_s(ondisk2->flags2) << 16;
1373                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1374                if (extended_flags & ~CE_EXTENDED_FLAGS)
1375                        die("Unknown index entry format %08x", extended_flags);
1376                flags |= extended_flags;
1377                name = ondisk2->name;
1378        }
1379        else
1380                name = ondisk->name;
1381
1382        if (!previous_name) {
1383                /* v3 and earlier */
1384                if (len == CE_NAMEMASK)
1385                        len = strlen(name);
1386                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1387
1388                *ent_size = ondisk_ce_size(ce);
1389        } else {
1390                unsigned long consumed;
1391                consumed = expand_name_field(previous_name, name);
1392                ce = cache_entry_from_ondisk(ondisk, flags,
1393                                             previous_name->buf,
1394                                             previous_name->len);
1395
1396                *ent_size = (name - ((char *)ondisk)) + consumed;
1397        }
1398        return ce;
1399}
1400
1401/* remember to discard_cache() before reading a different cache! */
1402int read_index_from(struct index_state *istate, const char *path)
1403{
1404        int fd, i;
1405        struct stat st;
1406        unsigned long src_offset;
1407        struct cache_header *hdr;
1408        void *mmap;
1409        size_t mmap_size;
1410        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1411
1412        errno = EBUSY;
1413        if (istate->initialized)
1414                return istate->cache_nr;
1415
1416        errno = ENOENT;
1417        istate->timestamp.sec = 0;
1418        istate->timestamp.nsec = 0;
1419        fd = open(path, O_RDONLY);
1420        if (fd < 0) {
1421                if (errno == ENOENT)
1422                        return 0;
1423                die_errno("index file open failed");
1424        }
1425
1426        if (fstat(fd, &st))
1427                die_errno("cannot stat the open index");
1428
1429        errno = EINVAL;
1430        mmap_size = xsize_t(st.st_size);
1431        if (mmap_size < sizeof(struct cache_header) + 20)
1432                die("index file smaller than expected");
1433
1434        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1435        close(fd);
1436        if (mmap == MAP_FAILED)
1437                die_errno("unable to map index file");
1438
1439        hdr = mmap;
1440        if (verify_hdr(hdr, mmap_size) < 0)
1441                goto unmap;
1442
1443        istate->version = ntohl(hdr->hdr_version);
1444        istate->cache_nr = ntohl(hdr->hdr_entries);
1445        istate->cache_alloc = alloc_nr(istate->cache_nr);
1446        istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1447        istate->initialized = 1;
1448
1449        if (istate->version == 4)
1450                previous_name = &previous_name_buf;
1451        else
1452                previous_name = NULL;
1453
1454        src_offset = sizeof(*hdr);
1455        for (i = 0; i < istate->cache_nr; i++) {
1456                struct ondisk_cache_entry *disk_ce;
1457                struct cache_entry *ce;
1458                unsigned long consumed;
1459
1460                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1461                ce = create_from_disk(disk_ce, &consumed, previous_name);
1462                set_index_entry(istate, i, ce);
1463
1464                src_offset += consumed;
1465        }
1466        strbuf_release(&previous_name_buf);
1467        istate->timestamp.sec = st.st_mtime;
1468        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1469
1470        while (src_offset <= mmap_size - 20 - 8) {
1471                /* After an array of active_nr index entries,
1472                 * there can be arbitrary number of extended
1473                 * sections, each of which is prefixed with
1474                 * extension name (4-byte) and section length
1475                 * in 4-byte network byte order.
1476                 */
1477                uint32_t extsize;
1478                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1479                extsize = ntohl(extsize);
1480                if (read_index_extension(istate,
1481                                         (const char *) mmap + src_offset,
1482                                         (char *) mmap + src_offset + 8,
1483                                         extsize) < 0)
1484                        goto unmap;
1485                src_offset += 8;
1486                src_offset += extsize;
1487        }
1488        munmap(mmap, mmap_size);
1489        return istate->cache_nr;
1490
1491unmap:
1492        munmap(mmap, mmap_size);
1493        errno = EINVAL;
1494        die("index file corrupt");
1495}
1496
1497int is_index_unborn(struct index_state *istate)
1498{
1499        return (!istate->cache_nr && !istate->timestamp.sec);
1500}
1501
1502int discard_index(struct index_state *istate)
1503{
1504        int i;
1505
1506        for (i = 0; i < istate->cache_nr; i++)
1507                free(istate->cache[i]);
1508        resolve_undo_clear_index(istate);
1509        istate->cache_nr = 0;
1510        istate->cache_changed = 0;
1511        istate->timestamp.sec = 0;
1512        istate->timestamp.nsec = 0;
1513        istate->name_hash_initialized = 0;
1514        free_hash(&istate->name_hash);
1515        cache_tree_free(&(istate->cache_tree));
1516        istate->initialized = 0;
1517
1518        /* no need to throw away allocated active_cache */
1519        return 0;
1520}
1521
1522int unmerged_index(const struct index_state *istate)
1523{
1524        int i;
1525        for (i = 0; i < istate->cache_nr; i++) {
1526                if (ce_stage(istate->cache[i]))
1527                        return 1;
1528        }
1529        return 0;
1530}
1531
1532#define WRITE_BUFFER_SIZE 8192
1533static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1534static unsigned long write_buffer_len;
1535
1536static int ce_write_flush(git_SHA_CTX *context, int fd)
1537{
1538        unsigned int buffered = write_buffer_len;
1539        if (buffered) {
1540                git_SHA1_Update(context, write_buffer, buffered);
1541                if (write_in_full(fd, write_buffer, buffered) != buffered)
1542                        return -1;
1543                write_buffer_len = 0;
1544        }
1545        return 0;
1546}
1547
1548static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1549{
1550        while (len) {
1551                unsigned int buffered = write_buffer_len;
1552                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1553                if (partial > len)
1554                        partial = len;
1555                memcpy(write_buffer + buffered, data, partial);
1556                buffered += partial;
1557                if (buffered == WRITE_BUFFER_SIZE) {
1558                        write_buffer_len = buffered;
1559                        if (ce_write_flush(context, fd))
1560                                return -1;
1561                        buffered = 0;
1562                }
1563                write_buffer_len = buffered;
1564                len -= partial;
1565                data = (char *) data + partial;
1566        }
1567        return 0;
1568}
1569
1570static int write_index_ext_header(git_SHA_CTX *context, int fd,
1571                                  unsigned int ext, unsigned int sz)
1572{
1573        ext = htonl(ext);
1574        sz = htonl(sz);
1575        return ((ce_write(context, fd, &ext, 4) < 0) ||
1576                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1577}
1578
1579static int ce_flush(git_SHA_CTX *context, int fd)
1580{
1581        unsigned int left = write_buffer_len;
1582
1583        if (left) {
1584                write_buffer_len = 0;
1585                git_SHA1_Update(context, write_buffer, left);
1586        }
1587
1588        /* Flush first if not enough space for SHA1 signature */
1589        if (left + 20 > WRITE_BUFFER_SIZE) {
1590                if (write_in_full(fd, write_buffer, left) != left)
1591                        return -1;
1592                left = 0;
1593        }
1594
1595        /* Append the SHA1 signature at the end */
1596        git_SHA1_Final(write_buffer + left, context);
1597        left += 20;
1598        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1599}
1600
1601static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1602{
1603        /*
1604         * The only thing we care about in this function is to smudge the
1605         * falsely clean entry due to touch-update-touch race, so we leave
1606         * everything else as they are.  We are called for entries whose
1607         * ce_mtime match the index file mtime.
1608         *
1609         * Note that this actually does not do much for gitlinks, for
1610         * which ce_match_stat_basic() always goes to the actual
1611         * contents.  The caller checks with is_racy_timestamp() which
1612         * always says "no" for gitlinks, so we are not called for them ;-)
1613         */
1614        struct stat st;
1615
1616        if (lstat(ce->name, &st) < 0)
1617                return;
1618        if (ce_match_stat_basic(ce, &st))
1619                return;
1620        if (ce_modified_check_fs(ce, &st)) {
1621                /* This is "racily clean"; smudge it.  Note that this
1622                 * is a tricky code.  At first glance, it may appear
1623                 * that it can break with this sequence:
1624                 *
1625                 * $ echo xyzzy >frotz
1626                 * $ git-update-index --add frotz
1627                 * $ : >frotz
1628                 * $ sleep 3
1629                 * $ echo filfre >nitfol
1630                 * $ git-update-index --add nitfol
1631                 *
1632                 * but it does not.  When the second update-index runs,
1633                 * it notices that the entry "frotz" has the same timestamp
1634                 * as index, and if we were to smudge it by resetting its
1635                 * size to zero here, then the object name recorded
1636                 * in index is the 6-byte file but the cached stat information
1637                 * becomes zero --- which would then match what we would
1638                 * obtain from the filesystem next time we stat("frotz").
1639                 *
1640                 * However, the second update-index, before calling
1641                 * this function, notices that the cached size is 6
1642                 * bytes and what is on the filesystem is an empty
1643                 * file, and never calls us, so the cached size information
1644                 * for "frotz" stays 6 which does not match the filesystem.
1645                 */
1646                ce->ce_size = 0;
1647        }
1648}
1649
1650/* Copy miscellaneous fields but not the name */
1651static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1652                                       struct cache_entry *ce)
1653{
1654        ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1655        ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1656        ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1657        ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1658        ondisk->dev  = htonl(ce->ce_dev);
1659        ondisk->ino  = htonl(ce->ce_ino);
1660        ondisk->mode = htonl(ce->ce_mode);
1661        ondisk->uid  = htonl(ce->ce_uid);
1662        ondisk->gid  = htonl(ce->ce_gid);
1663        ondisk->size = htonl(ce->ce_size);
1664        hashcpy(ondisk->sha1, ce->sha1);
1665        ondisk->flags = htons(ce->ce_flags);
1666        if (ce->ce_flags & CE_EXTENDED) {
1667                struct ondisk_cache_entry_extended *ondisk2;
1668                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1669                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1670                return ondisk2->name;
1671        }
1672        else {
1673                return ondisk->name;
1674        }
1675}
1676
1677static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1678                          struct strbuf *previous_name)
1679{
1680        int size;
1681        struct ondisk_cache_entry *ondisk;
1682        char *name;
1683        int result;
1684
1685        if (!previous_name) {
1686                size = ondisk_ce_size(ce);
1687                ondisk = xcalloc(1, size);
1688                name = copy_cache_entry_to_ondisk(ondisk, ce);
1689                memcpy(name, ce->name, ce_namelen(ce));
1690        } else {
1691                int common, to_remove, prefix_size;
1692                unsigned char to_remove_vi[16];
1693                for (common = 0;
1694                     (ce->name[common] &&
1695                      common < previous_name->len &&
1696                      ce->name[common] == previous_name->buf[common]);
1697                     common++)
1698                        ; /* still matching */
1699                to_remove = previous_name->len - common;
1700                prefix_size = encode_varint(to_remove, to_remove_vi);
1701
1702                if (ce->ce_flags & CE_EXTENDED)
1703                        size = offsetof(struct ondisk_cache_entry_extended, name);
1704                else
1705                        size = offsetof(struct ondisk_cache_entry, name);
1706                size += prefix_size + (ce_namelen(ce) - common + 1);
1707
1708                ondisk = xcalloc(1, size);
1709                name = copy_cache_entry_to_ondisk(ondisk, ce);
1710                memcpy(name, to_remove_vi, prefix_size);
1711                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1712
1713                strbuf_splice(previous_name, common, to_remove,
1714                              ce->name + common, ce_namelen(ce) - common);
1715        }
1716
1717        result = ce_write(c, fd, ondisk, size);
1718        free(ondisk);
1719        return result;
1720}
1721
1722static int has_racy_timestamp(struct index_state *istate)
1723{
1724        int entries = istate->cache_nr;
1725        int i;
1726
1727        for (i = 0; i < entries; i++) {
1728                struct cache_entry *ce = istate->cache[i];
1729                if (is_racy_timestamp(istate, ce))
1730                        return 1;
1731        }
1732        return 0;
1733}
1734
1735/*
1736 * Opportunisticly update the index but do not complain if we can't
1737 */
1738void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1739{
1740        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1741            !write_index(istate, lockfile->fd))
1742                commit_locked_index(lockfile);
1743        else
1744                rollback_lock_file(lockfile);
1745}
1746
1747int write_index(struct index_state *istate, int newfd)
1748{
1749        git_SHA_CTX c;
1750        struct cache_header hdr;
1751        int i, err, removed, extended, hdr_version;
1752        struct cache_entry **cache = istate->cache;
1753        int entries = istate->cache_nr;
1754        struct stat st;
1755        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1756
1757        for (i = removed = extended = 0; i < entries; i++) {
1758                if (cache[i]->ce_flags & CE_REMOVE)
1759                        removed++;
1760
1761                /* reduce extended entries if possible */
1762                cache[i]->ce_flags &= ~CE_EXTENDED;
1763                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1764                        extended++;
1765                        cache[i]->ce_flags |= CE_EXTENDED;
1766                }
1767        }
1768
1769        if (!istate->version)
1770                istate->version = INDEX_FORMAT_DEFAULT;
1771
1772        /* demote version 3 to version 2 when the latter suffices */
1773        if (istate->version == 3 || istate->version == 2)
1774                istate->version = extended ? 3 : 2;
1775
1776        hdr_version = istate->version;
1777
1778        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1779        hdr.hdr_version = htonl(hdr_version);
1780        hdr.hdr_entries = htonl(entries - removed);
1781
1782        git_SHA1_Init(&c);
1783        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1784                return -1;
1785
1786        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1787        for (i = 0; i < entries; i++) {
1788                struct cache_entry *ce = cache[i];
1789                if (ce->ce_flags & CE_REMOVE)
1790                        continue;
1791                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1792                        ce_smudge_racily_clean_entry(ce);
1793                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1794                        return -1;
1795        }
1796        strbuf_release(&previous_name_buf);
1797
1798        /* Write extension data here */
1799        if (istate->cache_tree) {
1800                struct strbuf sb = STRBUF_INIT;
1801
1802                cache_tree_write(&sb, istate->cache_tree);
1803                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1804                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1805                strbuf_release(&sb);
1806                if (err)
1807                        return -1;
1808        }
1809        if (istate->resolve_undo) {
1810                struct strbuf sb = STRBUF_INIT;
1811
1812                resolve_undo_write(&sb, istate->resolve_undo);
1813                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1814                                             sb.len) < 0
1815                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1816                strbuf_release(&sb);
1817                if (err)
1818                        return -1;
1819        }
1820
1821        if (ce_flush(&c, newfd) || fstat(newfd, &st))
1822                return -1;
1823        istate->timestamp.sec = (unsigned int)st.st_mtime;
1824        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1825        return 0;
1826}
1827
1828/*
1829 * Read the index file that is potentially unmerged into given
1830 * index_state, dropping any unmerged entries.  Returns true if
1831 * the index is unmerged.  Callers who want to refuse to work
1832 * from an unmerged state can call this and check its return value,
1833 * instead of calling read_cache().
1834 */
1835int read_index_unmerged(struct index_state *istate)
1836{
1837        int i;
1838        int unmerged = 0;
1839
1840        read_index(istate);
1841        for (i = 0; i < istate->cache_nr; i++) {
1842                struct cache_entry *ce = istate->cache[i];
1843                struct cache_entry *new_ce;
1844                int size, len;
1845
1846                if (!ce_stage(ce))
1847                        continue;
1848                unmerged = 1;
1849                len = strlen(ce->name);
1850                size = cache_entry_size(len);
1851                new_ce = xcalloc(1, size);
1852                memcpy(new_ce->name, ce->name, len);
1853                new_ce->ce_flags = create_ce_flags(len, 0) | CE_CONFLICTED;
1854                new_ce->ce_mode = ce->ce_mode;
1855                if (add_index_entry(istate, new_ce, 0))
1856                        return error("%s: cannot drop to stage #0",
1857                                     ce->name);
1858                i = index_name_pos(istate, new_ce->name, len);
1859        }
1860        return unmerged;
1861}
1862
1863/*
1864 * Returns 1 if the path is an "other" path with respect to
1865 * the index; that is, the path is not mentioned in the index at all,
1866 * either as a file, a directory with some files in the index,
1867 * or as an unmerged entry.
1868 *
1869 * We helpfully remove a trailing "/" from directories so that
1870 * the output of read_directory can be used as-is.
1871 */
1872int index_name_is_other(const struct index_state *istate, const char *name,
1873                int namelen)
1874{
1875        int pos;
1876        if (namelen && name[namelen - 1] == '/')
1877                namelen--;
1878        pos = index_name_pos(istate, name, namelen);
1879        if (0 <= pos)
1880                return 0;       /* exact match */
1881        pos = -pos - 1;
1882        if (pos < istate->cache_nr) {
1883                struct cache_entry *ce = istate->cache[pos];
1884                if (ce_namelen(ce) == namelen &&
1885                    !memcmp(ce->name, name, namelen))
1886                        return 0; /* Yup, this one exists unmerged */
1887        }
1888        return 1;
1889}