read-cache.con commit Merge branch 'mh/maint-parse-dirstat-fix' (8736c90)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 */
   6#define NO_THE_INDEX_COMPATIBILITY_MACROS
   7#include "cache.h"
   8#include "cache-tree.h"
   9#include "refs.h"
  10#include "dir.h"
  11#include "tree.h"
  12#include "commit.h"
  13#include "blob.h"
  14#include "resolve-undo.h"
  15#include "strbuf.h"
  16#include "varint.h"
  17
  18static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really);
  19
  20/* Mask for the name length in ce_flags in the on-disk index */
  21
  22#define CE_NAMEMASK  (0x0fff)
  23
  24/* Index extensions.
  25 *
  26 * The first letter should be 'A'..'Z' for extensions that are not
  27 * necessary for a correct operation (i.e. optimization data).
  28 * When new extensions are added that _needs_ to be understood in
  29 * order to correctly interpret the index file, pick character that
  30 * is outside the range, to cause the reader to abort.
  31 */
  32
  33#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  34#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  35#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  36
  37struct index_state the_index;
  38
  39static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  40{
  41        istate->cache[nr] = ce;
  42        add_name_hash(istate, ce);
  43}
  44
  45static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
  46{
  47        struct cache_entry *old = istate->cache[nr];
  48
  49        remove_name_hash(old);
  50        set_index_entry(istate, nr, ce);
  51        istate->cache_changed = 1;
  52}
  53
  54void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
  55{
  56        struct cache_entry *old = istate->cache[nr], *new;
  57        int namelen = strlen(new_name);
  58
  59        new = xmalloc(cache_entry_size(namelen));
  60        copy_cache_entry(new, old);
  61        new->ce_flags &= ~CE_STATE_MASK;
  62        new->ce_namelen = namelen;
  63        memcpy(new->name, new_name, namelen + 1);
  64
  65        cache_tree_invalidate_path(istate->cache_tree, old->name);
  66        remove_index_entry_at(istate, nr);
  67        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
  68}
  69
  70/*
  71 * This only updates the "non-critical" parts of the directory
  72 * cache, ie the parts that aren't tracked by GIT, and only used
  73 * to validate the cache.
  74 */
  75void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
  76{
  77        ce->ce_ctime.sec = (unsigned int)st->st_ctime;
  78        ce->ce_mtime.sec = (unsigned int)st->st_mtime;
  79        ce->ce_ctime.nsec = ST_CTIME_NSEC(*st);
  80        ce->ce_mtime.nsec = ST_MTIME_NSEC(*st);
  81        ce->ce_dev = st->st_dev;
  82        ce->ce_ino = st->st_ino;
  83        ce->ce_uid = st->st_uid;
  84        ce->ce_gid = st->st_gid;
  85        ce->ce_size = st->st_size;
  86
  87        if (assume_unchanged)
  88                ce->ce_flags |= CE_VALID;
  89
  90        if (S_ISREG(st->st_mode))
  91                ce_mark_uptodate(ce);
  92}
  93
  94static int ce_compare_data(struct cache_entry *ce, struct stat *st)
  95{
  96        int match = -1;
  97        int fd = open(ce->name, O_RDONLY);
  98
  99        if (fd >= 0) {
 100                unsigned char sha1[20];
 101                if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
 102                        match = hashcmp(sha1, ce->sha1);
 103                /* index_fd() closed the file descriptor already */
 104        }
 105        return match;
 106}
 107
 108static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
 109{
 110        int match = -1;
 111        void *buffer;
 112        unsigned long size;
 113        enum object_type type;
 114        struct strbuf sb = STRBUF_INIT;
 115
 116        if (strbuf_readlink(&sb, ce->name, expected_size))
 117                return -1;
 118
 119        buffer = read_sha1_file(ce->sha1, &type, &size);
 120        if (buffer) {
 121                if (size == sb.len)
 122                        match = memcmp(buffer, sb.buf, size);
 123                free(buffer);
 124        }
 125        strbuf_release(&sb);
 126        return match;
 127}
 128
 129static int ce_compare_gitlink(struct cache_entry *ce)
 130{
 131        unsigned char sha1[20];
 132
 133        /*
 134         * We don't actually require that the .git directory
 135         * under GITLINK directory be a valid git directory. It
 136         * might even be missing (in case nobody populated that
 137         * sub-project).
 138         *
 139         * If so, we consider it always to match.
 140         */
 141        if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
 142                return 0;
 143        return hashcmp(sha1, ce->sha1);
 144}
 145
 146static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
 147{
 148        switch (st->st_mode & S_IFMT) {
 149        case S_IFREG:
 150                if (ce_compare_data(ce, st))
 151                        return DATA_CHANGED;
 152                break;
 153        case S_IFLNK:
 154                if (ce_compare_link(ce, xsize_t(st->st_size)))
 155                        return DATA_CHANGED;
 156                break;
 157        case S_IFDIR:
 158                if (S_ISGITLINK(ce->ce_mode))
 159                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 160        default:
 161                return TYPE_CHANGED;
 162        }
 163        return 0;
 164}
 165
 166static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 167{
 168        unsigned int changed = 0;
 169
 170        if (ce->ce_flags & CE_REMOVE)
 171                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 172
 173        switch (ce->ce_mode & S_IFMT) {
 174        case S_IFREG:
 175                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 176                /* We consider only the owner x bit to be relevant for
 177                 * "mode changes"
 178                 */
 179                if (trust_executable_bit &&
 180                    (0100 & (ce->ce_mode ^ st->st_mode)))
 181                        changed |= MODE_CHANGED;
 182                break;
 183        case S_IFLNK:
 184                if (!S_ISLNK(st->st_mode) &&
 185                    (has_symlinks || !S_ISREG(st->st_mode)))
 186                        changed |= TYPE_CHANGED;
 187                break;
 188        case S_IFGITLINK:
 189                /* We ignore most of the st_xxx fields for gitlinks */
 190                if (!S_ISDIR(st->st_mode))
 191                        changed |= TYPE_CHANGED;
 192                else if (ce_compare_gitlink(ce))
 193                        changed |= DATA_CHANGED;
 194                return changed;
 195        default:
 196                die("internal error: ce_mode is %o", ce->ce_mode);
 197        }
 198        if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
 199                changed |= MTIME_CHANGED;
 200        if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime)
 201                changed |= CTIME_CHANGED;
 202
 203#ifdef USE_NSEC
 204        if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
 205                changed |= MTIME_CHANGED;
 206        if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
 207                changed |= CTIME_CHANGED;
 208#endif
 209
 210        if (ce->ce_uid != (unsigned int) st->st_uid ||
 211            ce->ce_gid != (unsigned int) st->st_gid)
 212                changed |= OWNER_CHANGED;
 213        if (ce->ce_ino != (unsigned int) st->st_ino)
 214                changed |= INODE_CHANGED;
 215
 216#ifdef USE_STDEV
 217        /*
 218         * st_dev breaks on network filesystems where different
 219         * clients will have different views of what "device"
 220         * the filesystem is on
 221         */
 222        if (ce->ce_dev != (unsigned int) st->st_dev)
 223                changed |= INODE_CHANGED;
 224#endif
 225
 226        if (ce->ce_size != (unsigned int) st->st_size)
 227                changed |= DATA_CHANGED;
 228
 229        /* Racily smudged entry? */
 230        if (!ce->ce_size) {
 231                if (!is_empty_blob_sha1(ce->sha1))
 232                        changed |= DATA_CHANGED;
 233        }
 234
 235        return changed;
 236}
 237
 238static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
 239{
 240        return (!S_ISGITLINK(ce->ce_mode) &&
 241                istate->timestamp.sec &&
 242#ifdef USE_NSEC
 243                 /* nanosecond timestamped files can also be racy! */
 244                (istate->timestamp.sec < ce->ce_mtime.sec ||
 245                 (istate->timestamp.sec == ce->ce_mtime.sec &&
 246                  istate->timestamp.nsec <= ce->ce_mtime.nsec))
 247#else
 248                istate->timestamp.sec <= ce->ce_mtime.sec
 249#endif
 250                 );
 251}
 252
 253int ie_match_stat(const struct index_state *istate,
 254                  struct cache_entry *ce, struct stat *st,
 255                  unsigned int options)
 256{
 257        unsigned int changed;
 258        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 259        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 260        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 261
 262        /*
 263         * If it's marked as always valid in the index, it's
 264         * valid whatever the checked-out copy says.
 265         *
 266         * skip-worktree has the same effect with higher precedence
 267         */
 268        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 269                return 0;
 270        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 271                return 0;
 272
 273        /*
 274         * Intent-to-add entries have not been added, so the index entry
 275         * by definition never matches what is in the work tree until it
 276         * actually gets added.
 277         */
 278        if (ce->ce_flags & CE_INTENT_TO_ADD)
 279                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 280
 281        changed = ce_match_stat_basic(ce, st);
 282
 283        /*
 284         * Within 1 second of this sequence:
 285         *      echo xyzzy >file && git-update-index --add file
 286         * running this command:
 287         *      echo frotz >file
 288         * would give a falsely clean cache entry.  The mtime and
 289         * length match the cache, and other stat fields do not change.
 290         *
 291         * We could detect this at update-index time (the cache entry
 292         * being registered/updated records the same time as "now")
 293         * and delay the return from git-update-index, but that would
 294         * effectively mean we can make at most one commit per second,
 295         * which is not acceptable.  Instead, we check cache entries
 296         * whose mtime are the same as the index file timestamp more
 297         * carefully than others.
 298         */
 299        if (!changed && is_racy_timestamp(istate, ce)) {
 300                if (assume_racy_is_modified)
 301                        changed |= DATA_CHANGED;
 302                else
 303                        changed |= ce_modified_check_fs(ce, st);
 304        }
 305
 306        return changed;
 307}
 308
 309int ie_modified(const struct index_state *istate,
 310                struct cache_entry *ce, struct stat *st, unsigned int options)
 311{
 312        int changed, changed_fs;
 313
 314        changed = ie_match_stat(istate, ce, st, options);
 315        if (!changed)
 316                return 0;
 317        /*
 318         * If the mode or type has changed, there's no point in trying
 319         * to refresh the entry - it's not going to match
 320         */
 321        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 322                return changed;
 323
 324        /*
 325         * Immediately after read-tree or update-index --cacheinfo,
 326         * the length field is zero, as we have never even read the
 327         * lstat(2) information once, and we cannot trust DATA_CHANGED
 328         * returned by ie_match_stat() which in turn was returned by
 329         * ce_match_stat_basic() to signal that the filesize of the
 330         * blob changed.  We have to actually go to the filesystem to
 331         * see if the contents match, and if so, should answer "unchanged".
 332         *
 333         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 334         * already has checked the actual HEAD from the filesystem in the
 335         * subproject.  If ie_match_stat() already said it is different,
 336         * then we know it is.
 337         */
 338        if ((changed & DATA_CHANGED) &&
 339            (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
 340                return changed;
 341
 342        changed_fs = ce_modified_check_fs(ce, st);
 343        if (changed_fs)
 344                return changed | changed_fs;
 345        return 0;
 346}
 347
 348int base_name_compare(const char *name1, int len1, int mode1,
 349                      const char *name2, int len2, int mode2)
 350{
 351        unsigned char c1, c2;
 352        int len = len1 < len2 ? len1 : len2;
 353        int cmp;
 354
 355        cmp = memcmp(name1, name2, len);
 356        if (cmp)
 357                return cmp;
 358        c1 = name1[len];
 359        c2 = name2[len];
 360        if (!c1 && S_ISDIR(mode1))
 361                c1 = '/';
 362        if (!c2 && S_ISDIR(mode2))
 363                c2 = '/';
 364        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 365}
 366
 367/*
 368 * df_name_compare() is identical to base_name_compare(), except it
 369 * compares conflicting directory/file entries as equal. Note that
 370 * while a directory name compares as equal to a regular file, they
 371 * then individually compare _differently_ to a filename that has
 372 * a dot after the basename (because '\0' < '.' < '/').
 373 *
 374 * This is used by routines that want to traverse the git namespace
 375 * but then handle conflicting entries together when possible.
 376 */
 377int df_name_compare(const char *name1, int len1, int mode1,
 378                    const char *name2, int len2, int mode2)
 379{
 380        int len = len1 < len2 ? len1 : len2, cmp;
 381        unsigned char c1, c2;
 382
 383        cmp = memcmp(name1, name2, len);
 384        if (cmp)
 385                return cmp;
 386        /* Directories and files compare equal (same length, same name) */
 387        if (len1 == len2)
 388                return 0;
 389        c1 = name1[len];
 390        if (!c1 && S_ISDIR(mode1))
 391                c1 = '/';
 392        c2 = name2[len];
 393        if (!c2 && S_ISDIR(mode2))
 394                c2 = '/';
 395        if (c1 == '/' && !c2)
 396                return 0;
 397        if (c2 == '/' && !c1)
 398                return 0;
 399        return c1 - c2;
 400}
 401
 402int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 403{
 404        int len = len1 < len2 ? len1 : len2;
 405        int cmp;
 406
 407        cmp = memcmp(name1, name2, len);
 408        if (cmp)
 409                return cmp;
 410        if (len1 < len2)
 411                return -1;
 412        if (len1 > len2)
 413                return 1;
 414
 415        if (stage1 < stage2)
 416                return -1;
 417        if (stage1 > stage2)
 418                return 1;
 419        return 0;
 420}
 421
 422int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
 423{
 424        return cache_name_stage_compare(name1, len1, 0, name2, len2, 0);
 425}
 426
 427static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 428{
 429        int first, last;
 430
 431        first = 0;
 432        last = istate->cache_nr;
 433        while (last > first) {
 434                int next = (last + first) >> 1;
 435                struct cache_entry *ce = istate->cache[next];
 436                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 437                if (!cmp)
 438                        return next;
 439                if (cmp < 0) {
 440                        last = next;
 441                        continue;
 442                }
 443                first = next+1;
 444        }
 445        return -first-1;
 446}
 447
 448int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 449{
 450        return index_name_stage_pos(istate, name, namelen, 0);
 451}
 452
 453/* Remove entry, return true if there are more entries to go.. */
 454int remove_index_entry_at(struct index_state *istate, int pos)
 455{
 456        struct cache_entry *ce = istate->cache[pos];
 457
 458        record_resolve_undo(istate, ce);
 459        remove_name_hash(ce);
 460        istate->cache_changed = 1;
 461        istate->cache_nr--;
 462        if (pos >= istate->cache_nr)
 463                return 0;
 464        memmove(istate->cache + pos,
 465                istate->cache + pos + 1,
 466                (istate->cache_nr - pos) * sizeof(struct cache_entry *));
 467        return 1;
 468}
 469
 470/*
 471 * Remove all cache ententries marked for removal, that is where
 472 * CE_REMOVE is set in ce_flags.  This is much more effective than
 473 * calling remove_index_entry_at() for each entry to be removed.
 474 */
 475void remove_marked_cache_entries(struct index_state *istate)
 476{
 477        struct cache_entry **ce_array = istate->cache;
 478        unsigned int i, j;
 479
 480        for (i = j = 0; i < istate->cache_nr; i++) {
 481                if (ce_array[i]->ce_flags & CE_REMOVE)
 482                        remove_name_hash(ce_array[i]);
 483                else
 484                        ce_array[j++] = ce_array[i];
 485        }
 486        istate->cache_changed = 1;
 487        istate->cache_nr = j;
 488}
 489
 490int remove_file_from_index(struct index_state *istate, const char *path)
 491{
 492        int pos = index_name_pos(istate, path, strlen(path));
 493        if (pos < 0)
 494                pos = -pos-1;
 495        cache_tree_invalidate_path(istate->cache_tree, path);
 496        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 497                remove_index_entry_at(istate, pos);
 498        return 0;
 499}
 500
 501static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 502{
 503        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 504}
 505
 506static int index_name_pos_also_unmerged(struct index_state *istate,
 507        const char *path, int namelen)
 508{
 509        int pos = index_name_pos(istate, path, namelen);
 510        struct cache_entry *ce;
 511
 512        if (pos >= 0)
 513                return pos;
 514
 515        /* maybe unmerged? */
 516        pos = -1 - pos;
 517        if (pos >= istate->cache_nr ||
 518                        compare_name((ce = istate->cache[pos]), path, namelen))
 519                return -1;
 520
 521        /* order of preference: stage 2, 1, 3 */
 522        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 523                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 524                        !compare_name(ce, path, namelen))
 525                pos++;
 526        return pos;
 527}
 528
 529static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 530{
 531        int len = ce_namelen(ce);
 532        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 533}
 534
 535/*
 536 * If we add a filename that aliases in the cache, we will use the
 537 * name that we already have - but we don't want to update the same
 538 * alias twice, because that implies that there were actually two
 539 * different files with aliasing names!
 540 *
 541 * So we use the CE_ADDED flag to verify that the alias was an old
 542 * one before we accept it as
 543 */
 544static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
 545{
 546        int len;
 547        struct cache_entry *new;
 548
 549        if (alias->ce_flags & CE_ADDED)
 550                die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
 551
 552        /* Ok, create the new entry using the name of the existing alias */
 553        len = ce_namelen(alias);
 554        new = xcalloc(1, cache_entry_size(len));
 555        memcpy(new->name, alias->name, len);
 556        copy_cache_entry(new, ce);
 557        free(ce);
 558        return new;
 559}
 560
 561static void record_intent_to_add(struct cache_entry *ce)
 562{
 563        unsigned char sha1[20];
 564        if (write_sha1_file("", 0, blob_type, sha1))
 565                die("cannot create an empty blob in the object database");
 566        hashcpy(ce->sha1, sha1);
 567}
 568
 569int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 570{
 571        int size, namelen, was_same;
 572        mode_t st_mode = st->st_mode;
 573        struct cache_entry *ce, *alias;
 574        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 575        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 576        int pretend = flags & ADD_CACHE_PRETEND;
 577        int intent_only = flags & ADD_CACHE_INTENT;
 578        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 579                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 580
 581        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 582                return error("%s: can only add regular files, symbolic links or git-directories", path);
 583
 584        namelen = strlen(path);
 585        if (S_ISDIR(st_mode)) {
 586                while (namelen && path[namelen-1] == '/')
 587                        namelen--;
 588        }
 589        size = cache_entry_size(namelen);
 590        ce = xcalloc(1, size);
 591        memcpy(ce->name, path, namelen);
 592        ce->ce_namelen = namelen;
 593        if (!intent_only)
 594                fill_stat_cache_info(ce, st);
 595        else
 596                ce->ce_flags |= CE_INTENT_TO_ADD;
 597
 598        if (trust_executable_bit && has_symlinks)
 599                ce->ce_mode = create_ce_mode(st_mode);
 600        else {
 601                /* If there is an existing entry, pick the mode bits and type
 602                 * from it, otherwise assume unexecutable regular file.
 603                 */
 604                struct cache_entry *ent;
 605                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 606
 607                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 608                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 609        }
 610
 611        /* When core.ignorecase=true, determine if a directory of the same name but differing
 612         * case already exists within the Git repository.  If it does, ensure the directory
 613         * case of the file being added to the repository matches (is folded into) the existing
 614         * entry's directory case.
 615         */
 616        if (ignore_case) {
 617                const char *startPtr = ce->name;
 618                const char *ptr = startPtr;
 619                while (*ptr) {
 620                        while (*ptr && *ptr != '/')
 621                                ++ptr;
 622                        if (*ptr == '/') {
 623                                struct cache_entry *foundce;
 624                                ++ptr;
 625                                foundce = index_name_exists(&the_index, ce->name, ptr - ce->name, ignore_case);
 626                                if (foundce) {
 627                                        memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
 628                                        startPtr = ptr;
 629                                }
 630                        }
 631                }
 632        }
 633
 634        alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
 635        if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
 636                /* Nothing changed, really */
 637                free(ce);
 638                if (!S_ISGITLINK(alias->ce_mode))
 639                        ce_mark_uptodate(alias);
 640                alias->ce_flags |= CE_ADDED;
 641                return 0;
 642        }
 643        if (!intent_only) {
 644                if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
 645                        return error("unable to index file %s", path);
 646        } else
 647                record_intent_to_add(ce);
 648
 649        if (ignore_case && alias && different_name(ce, alias))
 650                ce = create_alias_ce(ce, alias);
 651        ce->ce_flags |= CE_ADDED;
 652
 653        /* It was suspected to be racily clean, but it turns out to be Ok */
 654        was_same = (alias &&
 655                    !ce_stage(alias) &&
 656                    !hashcmp(alias->sha1, ce->sha1) &&
 657                    ce->ce_mode == alias->ce_mode);
 658
 659        if (pretend)
 660                ;
 661        else if (add_index_entry(istate, ce, add_option))
 662                return error("unable to add %s to index",path);
 663        if (verbose && !was_same)
 664                printf("add '%s'\n", path);
 665        return 0;
 666}
 667
 668int add_file_to_index(struct index_state *istate, const char *path, int flags)
 669{
 670        struct stat st;
 671        if (lstat(path, &st))
 672                die_errno("unable to stat '%s'", path);
 673        return add_to_index(istate, path, &st, flags);
 674}
 675
 676struct cache_entry *make_cache_entry(unsigned int mode,
 677                const unsigned char *sha1, const char *path, int stage,
 678                int refresh)
 679{
 680        int size, len;
 681        struct cache_entry *ce;
 682
 683        if (!verify_path(path)) {
 684                error("Invalid path '%s'", path);
 685                return NULL;
 686        }
 687
 688        len = strlen(path);
 689        size = cache_entry_size(len);
 690        ce = xcalloc(1, size);
 691
 692        hashcpy(ce->sha1, sha1);
 693        memcpy(ce->name, path, len);
 694        ce->ce_flags = create_ce_flags(stage);
 695        ce->ce_namelen = len;
 696        ce->ce_mode = create_ce_mode(mode);
 697
 698        if (refresh)
 699                return refresh_cache_entry(ce, 0);
 700
 701        return ce;
 702}
 703
 704int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 705{
 706        int len = ce_namelen(a);
 707        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 708}
 709
 710int ce_path_match(const struct cache_entry *ce, const struct pathspec *pathspec)
 711{
 712        return match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, NULL);
 713}
 714
 715/*
 716 * We fundamentally don't like some paths: we don't want
 717 * dot or dot-dot anywhere, and for obvious reasons don't
 718 * want to recurse into ".git" either.
 719 *
 720 * Also, we don't want double slashes or slashes at the
 721 * end that can make pathnames ambiguous.
 722 */
 723static int verify_dotfile(const char *rest)
 724{
 725        /*
 726         * The first character was '.', but that
 727         * has already been discarded, we now test
 728         * the rest.
 729         */
 730
 731        /* "." is not allowed */
 732        if (*rest == '\0' || is_dir_sep(*rest))
 733                return 0;
 734
 735        switch (*rest) {
 736        /*
 737         * ".git" followed by  NUL or slash is bad. This
 738         * shares the path end test with the ".." case.
 739         */
 740        case 'g':
 741                if (rest[1] != 'i')
 742                        break;
 743                if (rest[2] != 't')
 744                        break;
 745                rest += 2;
 746        /* fallthrough */
 747        case '.':
 748                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 749                        return 0;
 750        }
 751        return 1;
 752}
 753
 754int verify_path(const char *path)
 755{
 756        char c;
 757
 758        if (has_dos_drive_prefix(path))
 759                return 0;
 760
 761        goto inside;
 762        for (;;) {
 763                if (!c)
 764                        return 1;
 765                if (is_dir_sep(c)) {
 766inside:
 767                        c = *path++;
 768                        if ((c == '.' && !verify_dotfile(path)) ||
 769                            is_dir_sep(c) || c == '\0')
 770                                return 0;
 771                }
 772                c = *path++;
 773        }
 774}
 775
 776/*
 777 * Do we have another file that has the beginning components being a
 778 * proper superset of the name we're trying to add?
 779 */
 780static int has_file_name(struct index_state *istate,
 781                         const struct cache_entry *ce, int pos, int ok_to_replace)
 782{
 783        int retval = 0;
 784        int len = ce_namelen(ce);
 785        int stage = ce_stage(ce);
 786        const char *name = ce->name;
 787
 788        while (pos < istate->cache_nr) {
 789                struct cache_entry *p = istate->cache[pos++];
 790
 791                if (len >= ce_namelen(p))
 792                        break;
 793                if (memcmp(name, p->name, len))
 794                        break;
 795                if (ce_stage(p) != stage)
 796                        continue;
 797                if (p->name[len] != '/')
 798                        continue;
 799                if (p->ce_flags & CE_REMOVE)
 800                        continue;
 801                retval = -1;
 802                if (!ok_to_replace)
 803                        break;
 804                remove_index_entry_at(istate, --pos);
 805        }
 806        return retval;
 807}
 808
 809/*
 810 * Do we have another file with a pathname that is a proper
 811 * subset of the name we're trying to add?
 812 */
 813static int has_dir_name(struct index_state *istate,
 814                        const struct cache_entry *ce, int pos, int ok_to_replace)
 815{
 816        int retval = 0;
 817        int stage = ce_stage(ce);
 818        const char *name = ce->name;
 819        const char *slash = name + ce_namelen(ce);
 820
 821        for (;;) {
 822                int len;
 823
 824                for (;;) {
 825                        if (*--slash == '/')
 826                                break;
 827                        if (slash <= ce->name)
 828                                return retval;
 829                }
 830                len = slash - name;
 831
 832                pos = index_name_stage_pos(istate, name, len, stage);
 833                if (pos >= 0) {
 834                        /*
 835                         * Found one, but not so fast.  This could
 836                         * be a marker that says "I was here, but
 837                         * I am being removed".  Such an entry is
 838                         * not a part of the resulting tree, and
 839                         * it is Ok to have a directory at the same
 840                         * path.
 841                         */
 842                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
 843                                retval = -1;
 844                                if (!ok_to_replace)
 845                                        break;
 846                                remove_index_entry_at(istate, pos);
 847                                continue;
 848                        }
 849                }
 850                else
 851                        pos = -pos-1;
 852
 853                /*
 854                 * Trivial optimization: if we find an entry that
 855                 * already matches the sub-directory, then we know
 856                 * we're ok, and we can exit.
 857                 */
 858                while (pos < istate->cache_nr) {
 859                        struct cache_entry *p = istate->cache[pos];
 860                        if ((ce_namelen(p) <= len) ||
 861                            (p->name[len] != '/') ||
 862                            memcmp(p->name, name, len))
 863                                break; /* not our subdirectory */
 864                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
 865                                /*
 866                                 * p is at the same stage as our entry, and
 867                                 * is a subdirectory of what we are looking
 868                                 * at, so we cannot have conflicts at our
 869                                 * level or anything shorter.
 870                                 */
 871                                return retval;
 872                        pos++;
 873                }
 874        }
 875        return retval;
 876}
 877
 878/* We may be in a situation where we already have path/file and path
 879 * is being added, or we already have path and path/file is being
 880 * added.  Either one would result in a nonsense tree that has path
 881 * twice when git-write-tree tries to write it out.  Prevent it.
 882 *
 883 * If ok-to-replace is specified, we remove the conflicting entries
 884 * from the cache so the caller should recompute the insert position.
 885 * When this happens, we return non-zero.
 886 */
 887static int check_file_directory_conflict(struct index_state *istate,
 888                                         const struct cache_entry *ce,
 889                                         int pos, int ok_to_replace)
 890{
 891        int retval;
 892
 893        /*
 894         * When ce is an "I am going away" entry, we allow it to be added
 895         */
 896        if (ce->ce_flags & CE_REMOVE)
 897                return 0;
 898
 899        /*
 900         * We check if the path is a sub-path of a subsequent pathname
 901         * first, since removing those will not change the position
 902         * in the array.
 903         */
 904        retval = has_file_name(istate, ce, pos, ok_to_replace);
 905
 906        /*
 907         * Then check if the path might have a clashing sub-directory
 908         * before it.
 909         */
 910        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
 911}
 912
 913static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
 914{
 915        int pos;
 916        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 917        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 918        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 919        int new_only = option & ADD_CACHE_NEW_ONLY;
 920
 921        cache_tree_invalidate_path(istate->cache_tree, ce->name);
 922        pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 923
 924        /* existing match? Just replace it. */
 925        if (pos >= 0) {
 926                if (!new_only)
 927                        replace_index_entry(istate, pos, ce);
 928                return 0;
 929        }
 930        pos = -pos-1;
 931
 932        /*
 933         * Inserting a merged entry ("stage 0") into the index
 934         * will always replace all non-merged entries..
 935         */
 936        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
 937                while (ce_same_name(istate->cache[pos], ce)) {
 938                        ok_to_add = 1;
 939                        if (!remove_index_entry_at(istate, pos))
 940                                break;
 941                }
 942        }
 943
 944        if (!ok_to_add)
 945                return -1;
 946        if (!verify_path(ce->name))
 947                return error("Invalid path '%s'", ce->name);
 948
 949        if (!skip_df_check &&
 950            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
 951                if (!ok_to_replace)
 952                        return error("'%s' appears as both a file and as a directory",
 953                                     ce->name);
 954                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
 955                pos = -pos-1;
 956        }
 957        return pos + 1;
 958}
 959
 960int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
 961{
 962        int pos;
 963
 964        if (option & ADD_CACHE_JUST_APPEND)
 965                pos = istate->cache_nr;
 966        else {
 967                int ret;
 968                ret = add_index_entry_with_check(istate, ce, option);
 969                if (ret <= 0)
 970                        return ret;
 971                pos = ret - 1;
 972        }
 973
 974        /* Make sure the array is big enough .. */
 975        if (istate->cache_nr == istate->cache_alloc) {
 976                istate->cache_alloc = alloc_nr(istate->cache_alloc);
 977                istate->cache = xrealloc(istate->cache,
 978                                        istate->cache_alloc * sizeof(struct cache_entry *));
 979        }
 980
 981        /* Add it in.. */
 982        istate->cache_nr++;
 983        if (istate->cache_nr > pos + 1)
 984                memmove(istate->cache + pos + 1,
 985                        istate->cache + pos,
 986                        (istate->cache_nr - pos - 1) * sizeof(ce));
 987        set_index_entry(istate, pos, ce);
 988        istate->cache_changed = 1;
 989        return 0;
 990}
 991
 992/*
 993 * "refresh" does not calculate a new sha1 file or bring the
 994 * cache up-to-date for mode/content changes. But what it
 995 * _does_ do is to "re-match" the stat information of a file
 996 * with the cache, so that you can refresh the cache for a
 997 * file that hasn't been changed but where the stat entry is
 998 * out of date.
 999 *
1000 * For example, you'd want to do this after doing a "git-read-tree",
1001 * to link up the stat cache details with the proper files.
1002 */
1003static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1004                                             struct cache_entry *ce,
1005                                             unsigned int options, int *err,
1006                                             int *changed_ret)
1007{
1008        struct stat st;
1009        struct cache_entry *updated;
1010        int changed, size;
1011        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1012        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1013
1014        if (ce_uptodate(ce))
1015                return ce;
1016
1017        /*
1018         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1019         * that the change to the work tree does not matter and told
1020         * us not to worry.
1021         */
1022        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1023                ce_mark_uptodate(ce);
1024                return ce;
1025        }
1026        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1027                ce_mark_uptodate(ce);
1028                return ce;
1029        }
1030
1031        if (lstat(ce->name, &st) < 0) {
1032                if (err)
1033                        *err = errno;
1034                return NULL;
1035        }
1036
1037        changed = ie_match_stat(istate, ce, &st, options);
1038        if (changed_ret)
1039                *changed_ret = changed;
1040        if (!changed) {
1041                /*
1042                 * The path is unchanged.  If we were told to ignore
1043                 * valid bit, then we did the actual stat check and
1044                 * found that the entry is unmodified.  If the entry
1045                 * is not marked VALID, this is the place to mark it
1046                 * valid again, under "assume unchanged" mode.
1047                 */
1048                if (ignore_valid && assume_unchanged &&
1049                    !(ce->ce_flags & CE_VALID))
1050                        ; /* mark this one VALID again */
1051                else {
1052                        /*
1053                         * We do not mark the index itself "modified"
1054                         * because CE_UPTODATE flag is in-core only;
1055                         * we are not going to write this change out.
1056                         */
1057                        if (!S_ISGITLINK(ce->ce_mode))
1058                                ce_mark_uptodate(ce);
1059                        return ce;
1060                }
1061        }
1062
1063        if (ie_modified(istate, ce, &st, options)) {
1064                if (err)
1065                        *err = EINVAL;
1066                return NULL;
1067        }
1068
1069        size = ce_size(ce);
1070        updated = xmalloc(size);
1071        memcpy(updated, ce, size);
1072        fill_stat_cache_info(updated, &st);
1073        /*
1074         * If ignore_valid is not set, we should leave CE_VALID bit
1075         * alone.  Otherwise, paths marked with --no-assume-unchanged
1076         * (i.e. things to be edited) will reacquire CE_VALID bit
1077         * automatically, which is not really what we want.
1078         */
1079        if (!ignore_valid && assume_unchanged &&
1080            !(ce->ce_flags & CE_VALID))
1081                updated->ce_flags &= ~CE_VALID;
1082
1083        return updated;
1084}
1085
1086static void show_file(const char * fmt, const char * name, int in_porcelain,
1087                      int * first, const char *header_msg)
1088{
1089        if (in_porcelain && *first && header_msg) {
1090                printf("%s\n", header_msg);
1091                *first = 0;
1092        }
1093        printf(fmt, name);
1094}
1095
1096int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1097                  char *seen, const char *header_msg)
1098{
1099        int i;
1100        int has_errors = 0;
1101        int really = (flags & REFRESH_REALLY) != 0;
1102        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1103        int quiet = (flags & REFRESH_QUIET) != 0;
1104        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1105        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1106        int first = 1;
1107        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1108        unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1109        const char *modified_fmt;
1110        const char *deleted_fmt;
1111        const char *typechange_fmt;
1112        const char *added_fmt;
1113        const char *unmerged_fmt;
1114
1115        modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1116        deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1117        typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1118        added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1119        unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1120        for (i = 0; i < istate->cache_nr; i++) {
1121                struct cache_entry *ce, *new;
1122                int cache_errno = 0;
1123                int changed = 0;
1124                int filtered = 0;
1125
1126                ce = istate->cache[i];
1127                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1128                        continue;
1129
1130                if (pathspec &&
1131                    !match_pathspec(pathspec, ce->name, ce_namelen(ce), 0, seen))
1132                        filtered = 1;
1133
1134                if (ce_stage(ce)) {
1135                        while ((i < istate->cache_nr) &&
1136                               ! strcmp(istate->cache[i]->name, ce->name))
1137                                i++;
1138                        i--;
1139                        if (allow_unmerged)
1140                                continue;
1141                        if (!filtered)
1142                                show_file(unmerged_fmt, ce->name, in_porcelain,
1143                                          &first, header_msg);
1144                        has_errors = 1;
1145                        continue;
1146                }
1147
1148                if (filtered)
1149                        continue;
1150
1151                new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1152                if (new == ce)
1153                        continue;
1154                if (!new) {
1155                        const char *fmt;
1156
1157                        if (not_new && cache_errno == ENOENT)
1158                                continue;
1159                        if (really && cache_errno == EINVAL) {
1160                                /* If we are doing --really-refresh that
1161                                 * means the index is not valid anymore.
1162                                 */
1163                                ce->ce_flags &= ~CE_VALID;
1164                                istate->cache_changed = 1;
1165                        }
1166                        if (quiet)
1167                                continue;
1168
1169                        if (cache_errno == ENOENT)
1170                                fmt = deleted_fmt;
1171                        else if (ce->ce_flags & CE_INTENT_TO_ADD)
1172                                fmt = added_fmt; /* must be before other checks */
1173                        else if (changed & TYPE_CHANGED)
1174                                fmt = typechange_fmt;
1175                        else
1176                                fmt = modified_fmt;
1177                        show_file(fmt,
1178                                  ce->name, in_porcelain, &first, header_msg);
1179                        has_errors = 1;
1180                        continue;
1181                }
1182
1183                replace_index_entry(istate, i, new);
1184        }
1185        return has_errors;
1186}
1187
1188static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1189{
1190        return refresh_cache_ent(&the_index, ce, really, NULL, NULL);
1191}
1192
1193
1194/*****************************************************************
1195 * Index File I/O
1196 *****************************************************************/
1197
1198#define INDEX_FORMAT_DEFAULT 3
1199
1200/*
1201 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1202 * Again - this is just a (very strong in practice) heuristic that
1203 * the inode hasn't changed.
1204 *
1205 * We save the fields in big-endian order to allow using the
1206 * index file over NFS transparently.
1207 */
1208struct ondisk_cache_entry {
1209        struct cache_time ctime;
1210        struct cache_time mtime;
1211        unsigned int dev;
1212        unsigned int ino;
1213        unsigned int mode;
1214        unsigned int uid;
1215        unsigned int gid;
1216        unsigned int size;
1217        unsigned char sha1[20];
1218        unsigned short flags;
1219        char name[FLEX_ARRAY]; /* more */
1220};
1221
1222/*
1223 * This struct is used when CE_EXTENDED bit is 1
1224 * The struct must match ondisk_cache_entry exactly from
1225 * ctime till flags
1226 */
1227struct ondisk_cache_entry_extended {
1228        struct cache_time ctime;
1229        struct cache_time mtime;
1230        unsigned int dev;
1231        unsigned int ino;
1232        unsigned int mode;
1233        unsigned int uid;
1234        unsigned int gid;
1235        unsigned int size;
1236        unsigned char sha1[20];
1237        unsigned short flags;
1238        unsigned short flags2;
1239        char name[FLEX_ARRAY]; /* more */
1240};
1241
1242/* These are only used for v3 or lower */
1243#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1244#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1245#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1246#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1247                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1248                            ondisk_cache_entry_size(ce_namelen(ce)))
1249
1250static int verify_hdr(struct cache_header *hdr, unsigned long size)
1251{
1252        git_SHA_CTX c;
1253        unsigned char sha1[20];
1254        int hdr_version;
1255
1256        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1257                return error("bad signature");
1258        hdr_version = ntohl(hdr->hdr_version);
1259        if (hdr_version < 2 || 4 < hdr_version)
1260                return error("bad index version %d", hdr_version);
1261        git_SHA1_Init(&c);
1262        git_SHA1_Update(&c, hdr, size - 20);
1263        git_SHA1_Final(sha1, &c);
1264        if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1265                return error("bad index file sha1 signature");
1266        return 0;
1267}
1268
1269static int read_index_extension(struct index_state *istate,
1270                                const char *ext, void *data, unsigned long sz)
1271{
1272        switch (CACHE_EXT(ext)) {
1273        case CACHE_EXT_TREE:
1274                istate->cache_tree = cache_tree_read(data, sz);
1275                break;
1276        case CACHE_EXT_RESOLVE_UNDO:
1277                istate->resolve_undo = resolve_undo_read(data, sz);
1278                break;
1279        default:
1280                if (*ext < 'A' || 'Z' < *ext)
1281                        return error("index uses %.4s extension, which we do not understand",
1282                                     ext);
1283                fprintf(stderr, "ignoring %.4s extension\n", ext);
1284                break;
1285        }
1286        return 0;
1287}
1288
1289int read_index(struct index_state *istate)
1290{
1291        return read_index_from(istate, get_index_file());
1292}
1293
1294#ifndef NEEDS_ALIGNED_ACCESS
1295#define ntoh_s(var) ntohs(var)
1296#define ntoh_l(var) ntohl(var)
1297#else
1298static inline uint16_t ntoh_s_force_align(void *p)
1299{
1300        uint16_t x;
1301        memcpy(&x, p, sizeof(x));
1302        return ntohs(x);
1303}
1304static inline uint32_t ntoh_l_force_align(void *p)
1305{
1306        uint32_t x;
1307        memcpy(&x, p, sizeof(x));
1308        return ntohl(x);
1309}
1310#define ntoh_s(var) ntoh_s_force_align(&(var))
1311#define ntoh_l(var) ntoh_l_force_align(&(var))
1312#endif
1313
1314static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1315                                                   unsigned int flags,
1316                                                   const char *name,
1317                                                   size_t len)
1318{
1319        struct cache_entry *ce = xmalloc(cache_entry_size(len));
1320
1321        ce->ce_ctime.sec = ntoh_l(ondisk->ctime.sec);
1322        ce->ce_mtime.sec = ntoh_l(ondisk->mtime.sec);
1323        ce->ce_ctime.nsec = ntoh_l(ondisk->ctime.nsec);
1324        ce->ce_mtime.nsec = ntoh_l(ondisk->mtime.nsec);
1325        ce->ce_dev   = ntoh_l(ondisk->dev);
1326        ce->ce_ino   = ntoh_l(ondisk->ino);
1327        ce->ce_mode  = ntoh_l(ondisk->mode);
1328        ce->ce_uid   = ntoh_l(ondisk->uid);
1329        ce->ce_gid   = ntoh_l(ondisk->gid);
1330        ce->ce_size  = ntoh_l(ondisk->size);
1331        ce->ce_flags = flags & ~CE_NAMEMASK;
1332        ce->ce_namelen = len;
1333        hashcpy(ce->sha1, ondisk->sha1);
1334        memcpy(ce->name, name, len);
1335        ce->name[len] = '\0';
1336        return ce;
1337}
1338
1339/*
1340 * Adjacent cache entries tend to share the leading paths, so it makes
1341 * sense to only store the differences in later entries.  In the v4
1342 * on-disk format of the index, each on-disk cache entry stores the
1343 * number of bytes to be stripped from the end of the previous name,
1344 * and the bytes to append to the result, to come up with its name.
1345 */
1346static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1347{
1348        const unsigned char *ep, *cp = (const unsigned char *)cp_;
1349        size_t len = decode_varint(&cp);
1350
1351        if (name->len < len)
1352                die("malformed name field in the index");
1353        strbuf_remove(name, name->len - len, len);
1354        for (ep = cp; *ep; ep++)
1355                ; /* find the end */
1356        strbuf_add(name, cp, ep - cp);
1357        return (const char *)ep + 1 - cp_;
1358}
1359
1360static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1361                                            unsigned long *ent_size,
1362                                            struct strbuf *previous_name)
1363{
1364        struct cache_entry *ce;
1365        size_t len;
1366        const char *name;
1367        unsigned int flags;
1368
1369        /* On-disk flags are just 16 bits */
1370        flags = ntoh_s(ondisk->flags);
1371        len = flags & CE_NAMEMASK;
1372
1373        if (flags & CE_EXTENDED) {
1374                struct ondisk_cache_entry_extended *ondisk2;
1375                int extended_flags;
1376                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1377                extended_flags = ntoh_s(ondisk2->flags2) << 16;
1378                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1379                if (extended_flags & ~CE_EXTENDED_FLAGS)
1380                        die("Unknown index entry format %08x", extended_flags);
1381                flags |= extended_flags;
1382                name = ondisk2->name;
1383        }
1384        else
1385                name = ondisk->name;
1386
1387        if (!previous_name) {
1388                /* v3 and earlier */
1389                if (len == CE_NAMEMASK)
1390                        len = strlen(name);
1391                ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1392
1393                *ent_size = ondisk_ce_size(ce);
1394        } else {
1395                unsigned long consumed;
1396                consumed = expand_name_field(previous_name, name);
1397                ce = cache_entry_from_ondisk(ondisk, flags,
1398                                             previous_name->buf,
1399                                             previous_name->len);
1400
1401                *ent_size = (name - ((char *)ondisk)) + consumed;
1402        }
1403        return ce;
1404}
1405
1406/* remember to discard_cache() before reading a different cache! */
1407int read_index_from(struct index_state *istate, const char *path)
1408{
1409        int fd, i;
1410        struct stat st;
1411        unsigned long src_offset;
1412        struct cache_header *hdr;
1413        void *mmap;
1414        size_t mmap_size;
1415        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1416
1417        if (istate->initialized)
1418                return istate->cache_nr;
1419
1420        istate->timestamp.sec = 0;
1421        istate->timestamp.nsec = 0;
1422        fd = open(path, O_RDONLY);
1423        if (fd < 0) {
1424                if (errno == ENOENT)
1425                        return 0;
1426                die_errno("index file open failed");
1427        }
1428
1429        if (fstat(fd, &st))
1430                die_errno("cannot stat the open index");
1431
1432        mmap_size = xsize_t(st.st_size);
1433        if (mmap_size < sizeof(struct cache_header) + 20)
1434                die("index file smaller than expected");
1435
1436        mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1437        if (mmap == MAP_FAILED)
1438                die_errno("unable to map index file");
1439        close(fd);
1440
1441        hdr = mmap;
1442        if (verify_hdr(hdr, mmap_size) < 0)
1443                goto unmap;
1444
1445        istate->version = ntohl(hdr->hdr_version);
1446        istate->cache_nr = ntohl(hdr->hdr_entries);
1447        istate->cache_alloc = alloc_nr(istate->cache_nr);
1448        istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1449        istate->initialized = 1;
1450
1451        if (istate->version == 4)
1452                previous_name = &previous_name_buf;
1453        else
1454                previous_name = NULL;
1455
1456        src_offset = sizeof(*hdr);
1457        for (i = 0; i < istate->cache_nr; i++) {
1458                struct ondisk_cache_entry *disk_ce;
1459                struct cache_entry *ce;
1460                unsigned long consumed;
1461
1462                disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1463                ce = create_from_disk(disk_ce, &consumed, previous_name);
1464                set_index_entry(istate, i, ce);
1465
1466                src_offset += consumed;
1467        }
1468        strbuf_release(&previous_name_buf);
1469        istate->timestamp.sec = st.st_mtime;
1470        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1471
1472        while (src_offset <= mmap_size - 20 - 8) {
1473                /* After an array of active_nr index entries,
1474                 * there can be arbitrary number of extended
1475                 * sections, each of which is prefixed with
1476                 * extension name (4-byte) and section length
1477                 * in 4-byte network byte order.
1478                 */
1479                uint32_t extsize;
1480                memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1481                extsize = ntohl(extsize);
1482                if (read_index_extension(istate,
1483                                         (const char *) mmap + src_offset,
1484                                         (char *) mmap + src_offset + 8,
1485                                         extsize) < 0)
1486                        goto unmap;
1487                src_offset += 8;
1488                src_offset += extsize;
1489        }
1490        munmap(mmap, mmap_size);
1491        return istate->cache_nr;
1492
1493unmap:
1494        munmap(mmap, mmap_size);
1495        die("index file corrupt");
1496}
1497
1498int is_index_unborn(struct index_state *istate)
1499{
1500        return (!istate->cache_nr && !istate->timestamp.sec);
1501}
1502
1503int discard_index(struct index_state *istate)
1504{
1505        int i;
1506
1507        for (i = 0; i < istate->cache_nr; i++)
1508                free(istate->cache[i]);
1509        resolve_undo_clear_index(istate);
1510        istate->cache_nr = 0;
1511        istate->cache_changed = 0;
1512        istate->timestamp.sec = 0;
1513        istate->timestamp.nsec = 0;
1514        istate->name_hash_initialized = 0;
1515        free_hash(&istate->name_hash);
1516        cache_tree_free(&(istate->cache_tree));
1517        istate->initialized = 0;
1518
1519        /* no need to throw away allocated active_cache */
1520        return 0;
1521}
1522
1523int unmerged_index(const struct index_state *istate)
1524{
1525        int i;
1526        for (i = 0; i < istate->cache_nr; i++) {
1527                if (ce_stage(istate->cache[i]))
1528                        return 1;
1529        }
1530        return 0;
1531}
1532
1533#define WRITE_BUFFER_SIZE 8192
1534static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1535static unsigned long write_buffer_len;
1536
1537static int ce_write_flush(git_SHA_CTX *context, int fd)
1538{
1539        unsigned int buffered = write_buffer_len;
1540        if (buffered) {
1541                git_SHA1_Update(context, write_buffer, buffered);
1542                if (write_in_full(fd, write_buffer, buffered) != buffered)
1543                        return -1;
1544                write_buffer_len = 0;
1545        }
1546        return 0;
1547}
1548
1549static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1550{
1551        while (len) {
1552                unsigned int buffered = write_buffer_len;
1553                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1554                if (partial > len)
1555                        partial = len;
1556                memcpy(write_buffer + buffered, data, partial);
1557                buffered += partial;
1558                if (buffered == WRITE_BUFFER_SIZE) {
1559                        write_buffer_len = buffered;
1560                        if (ce_write_flush(context, fd))
1561                                return -1;
1562                        buffered = 0;
1563                }
1564                write_buffer_len = buffered;
1565                len -= partial;
1566                data = (char *) data + partial;
1567        }
1568        return 0;
1569}
1570
1571static int write_index_ext_header(git_SHA_CTX *context, int fd,
1572                                  unsigned int ext, unsigned int sz)
1573{
1574        ext = htonl(ext);
1575        sz = htonl(sz);
1576        return ((ce_write(context, fd, &ext, 4) < 0) ||
1577                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1578}
1579
1580static int ce_flush(git_SHA_CTX *context, int fd)
1581{
1582        unsigned int left = write_buffer_len;
1583
1584        if (left) {
1585                write_buffer_len = 0;
1586                git_SHA1_Update(context, write_buffer, left);
1587        }
1588
1589        /* Flush first if not enough space for SHA1 signature */
1590        if (left + 20 > WRITE_BUFFER_SIZE) {
1591                if (write_in_full(fd, write_buffer, left) != left)
1592                        return -1;
1593                left = 0;
1594        }
1595
1596        /* Append the SHA1 signature at the end */
1597        git_SHA1_Final(write_buffer + left, context);
1598        left += 20;
1599        return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1600}
1601
1602static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1603{
1604        /*
1605         * The only thing we care about in this function is to smudge the
1606         * falsely clean entry due to touch-update-touch race, so we leave
1607         * everything else as they are.  We are called for entries whose
1608         * ce_mtime match the index file mtime.
1609         *
1610         * Note that this actually does not do much for gitlinks, for
1611         * which ce_match_stat_basic() always goes to the actual
1612         * contents.  The caller checks with is_racy_timestamp() which
1613         * always says "no" for gitlinks, so we are not called for them ;-)
1614         */
1615        struct stat st;
1616
1617        if (lstat(ce->name, &st) < 0)
1618                return;
1619        if (ce_match_stat_basic(ce, &st))
1620                return;
1621        if (ce_modified_check_fs(ce, &st)) {
1622                /* This is "racily clean"; smudge it.  Note that this
1623                 * is a tricky code.  At first glance, it may appear
1624                 * that it can break with this sequence:
1625                 *
1626                 * $ echo xyzzy >frotz
1627                 * $ git-update-index --add frotz
1628                 * $ : >frotz
1629                 * $ sleep 3
1630                 * $ echo filfre >nitfol
1631                 * $ git-update-index --add nitfol
1632                 *
1633                 * but it does not.  When the second update-index runs,
1634                 * it notices that the entry "frotz" has the same timestamp
1635                 * as index, and if we were to smudge it by resetting its
1636                 * size to zero here, then the object name recorded
1637                 * in index is the 6-byte file but the cached stat information
1638                 * becomes zero --- which would then match what we would
1639                 * obtain from the filesystem next time we stat("frotz").
1640                 *
1641                 * However, the second update-index, before calling
1642                 * this function, notices that the cached size is 6
1643                 * bytes and what is on the filesystem is an empty
1644                 * file, and never calls us, so the cached size information
1645                 * for "frotz" stays 6 which does not match the filesystem.
1646                 */
1647                ce->ce_size = 0;
1648        }
1649}
1650
1651/* Copy miscellaneous fields but not the name */
1652static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1653                                       struct cache_entry *ce)
1654{
1655        short flags;
1656
1657        ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1658        ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1659        ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1660        ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1661        ondisk->dev  = htonl(ce->ce_dev);
1662        ondisk->ino  = htonl(ce->ce_ino);
1663        ondisk->mode = htonl(ce->ce_mode);
1664        ondisk->uid  = htonl(ce->ce_uid);
1665        ondisk->gid  = htonl(ce->ce_gid);
1666        ondisk->size = htonl(ce->ce_size);
1667        hashcpy(ondisk->sha1, ce->sha1);
1668
1669        flags = ce->ce_flags;
1670        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1671        ondisk->flags = htons(flags);
1672        if (ce->ce_flags & CE_EXTENDED) {
1673                struct ondisk_cache_entry_extended *ondisk2;
1674                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1675                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1676                return ondisk2->name;
1677        }
1678        else {
1679                return ondisk->name;
1680        }
1681}
1682
1683static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1684                          struct strbuf *previous_name)
1685{
1686        int size;
1687        struct ondisk_cache_entry *ondisk;
1688        char *name;
1689        int result;
1690
1691        if (!previous_name) {
1692                size = ondisk_ce_size(ce);
1693                ondisk = xcalloc(1, size);
1694                name = copy_cache_entry_to_ondisk(ondisk, ce);
1695                memcpy(name, ce->name, ce_namelen(ce));
1696        } else {
1697                int common, to_remove, prefix_size;
1698                unsigned char to_remove_vi[16];
1699                for (common = 0;
1700                     (ce->name[common] &&
1701                      common < previous_name->len &&
1702                      ce->name[common] == previous_name->buf[common]);
1703                     common++)
1704                        ; /* still matching */
1705                to_remove = previous_name->len - common;
1706                prefix_size = encode_varint(to_remove, to_remove_vi);
1707
1708                if (ce->ce_flags & CE_EXTENDED)
1709                        size = offsetof(struct ondisk_cache_entry_extended, name);
1710                else
1711                        size = offsetof(struct ondisk_cache_entry, name);
1712                size += prefix_size + (ce_namelen(ce) - common + 1);
1713
1714                ondisk = xcalloc(1, size);
1715                name = copy_cache_entry_to_ondisk(ondisk, ce);
1716                memcpy(name, to_remove_vi, prefix_size);
1717                memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1718
1719                strbuf_splice(previous_name, common, to_remove,
1720                              ce->name + common, ce_namelen(ce) - common);
1721        }
1722
1723        result = ce_write(c, fd, ondisk, size);
1724        free(ondisk);
1725        return result;
1726}
1727
1728static int has_racy_timestamp(struct index_state *istate)
1729{
1730        int entries = istate->cache_nr;
1731        int i;
1732
1733        for (i = 0; i < entries; i++) {
1734                struct cache_entry *ce = istate->cache[i];
1735                if (is_racy_timestamp(istate, ce))
1736                        return 1;
1737        }
1738        return 0;
1739}
1740
1741/*
1742 * Opportunisticly update the index but do not complain if we can't
1743 */
1744void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1745{
1746        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1747            !write_index(istate, lockfile->fd))
1748                commit_locked_index(lockfile);
1749        else
1750                rollback_lock_file(lockfile);
1751}
1752
1753int write_index(struct index_state *istate, int newfd)
1754{
1755        git_SHA_CTX c;
1756        struct cache_header hdr;
1757        int i, err, removed, extended, hdr_version;
1758        struct cache_entry **cache = istate->cache;
1759        int entries = istate->cache_nr;
1760        struct stat st;
1761        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1762
1763        for (i = removed = extended = 0; i < entries; i++) {
1764                if (cache[i]->ce_flags & CE_REMOVE)
1765                        removed++;
1766
1767                /* reduce extended entries if possible */
1768                cache[i]->ce_flags &= ~CE_EXTENDED;
1769                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1770                        extended++;
1771                        cache[i]->ce_flags |= CE_EXTENDED;
1772                }
1773        }
1774
1775        if (!istate->version)
1776                istate->version = INDEX_FORMAT_DEFAULT;
1777
1778        /* demote version 3 to version 2 when the latter suffices */
1779        if (istate->version == 3 || istate->version == 2)
1780                istate->version = extended ? 3 : 2;
1781
1782        hdr_version = istate->version;
1783
1784        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1785        hdr.hdr_version = htonl(hdr_version);
1786        hdr.hdr_entries = htonl(entries - removed);
1787
1788        git_SHA1_Init(&c);
1789        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1790                return -1;
1791
1792        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1793        for (i = 0; i < entries; i++) {
1794                struct cache_entry *ce = cache[i];
1795                if (ce->ce_flags & CE_REMOVE)
1796                        continue;
1797                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1798                        ce_smudge_racily_clean_entry(ce);
1799                if (is_null_sha1(ce->sha1))
1800                        return error("cache entry has null sha1: %s", ce->name);
1801                if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1802                        return -1;
1803        }
1804        strbuf_release(&previous_name_buf);
1805
1806        /* Write extension data here */
1807        if (istate->cache_tree) {
1808                struct strbuf sb = STRBUF_INIT;
1809
1810                cache_tree_write(&sb, istate->cache_tree);
1811                err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1812                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1813                strbuf_release(&sb);
1814                if (err)
1815                        return -1;
1816        }
1817        if (istate->resolve_undo) {
1818                struct strbuf sb = STRBUF_INIT;
1819
1820                resolve_undo_write(&sb, istate->resolve_undo);
1821                err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1822                                             sb.len) < 0
1823                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1824                strbuf_release(&sb);
1825                if (err)
1826                        return -1;
1827        }
1828
1829        if (ce_flush(&c, newfd) || fstat(newfd, &st))
1830                return -1;
1831        istate->timestamp.sec = (unsigned int)st.st_mtime;
1832        istate->timestamp.nsec = ST_MTIME_NSEC(st);
1833        return 0;
1834}
1835
1836/*
1837 * Read the index file that is potentially unmerged into given
1838 * index_state, dropping any unmerged entries.  Returns true if
1839 * the index is unmerged.  Callers who want to refuse to work
1840 * from an unmerged state can call this and check its return value,
1841 * instead of calling read_cache().
1842 */
1843int read_index_unmerged(struct index_state *istate)
1844{
1845        int i;
1846        int unmerged = 0;
1847
1848        read_index(istate);
1849        for (i = 0; i < istate->cache_nr; i++) {
1850                struct cache_entry *ce = istate->cache[i];
1851                struct cache_entry *new_ce;
1852                int size, len;
1853
1854                if (!ce_stage(ce))
1855                        continue;
1856                unmerged = 1;
1857                len = ce_namelen(ce);
1858                size = cache_entry_size(len);
1859                new_ce = xcalloc(1, size);
1860                memcpy(new_ce->name, ce->name, len);
1861                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
1862                new_ce->ce_namelen = len;
1863                new_ce->ce_mode = ce->ce_mode;
1864                if (add_index_entry(istate, new_ce, 0))
1865                        return error("%s: cannot drop to stage #0",
1866                                     ce->name);
1867                i = index_name_pos(istate, new_ce->name, len);
1868        }
1869        return unmerged;
1870}
1871
1872/*
1873 * Returns 1 if the path is an "other" path with respect to
1874 * the index; that is, the path is not mentioned in the index at all,
1875 * either as a file, a directory with some files in the index,
1876 * or as an unmerged entry.
1877 *
1878 * We helpfully remove a trailing "/" from directories so that
1879 * the output of read_directory can be used as-is.
1880 */
1881int index_name_is_other(const struct index_state *istate, const char *name,
1882                int namelen)
1883{
1884        int pos;
1885        if (namelen && name[namelen - 1] == '/')
1886                namelen--;
1887        pos = index_name_pos(istate, name, namelen);
1888        if (0 <= pos)
1889                return 0;       /* exact match */
1890        pos = -pos - 1;
1891        if (pos < istate->cache_nr) {
1892                struct cache_entry *ce = istate->cache[pos];
1893                if (ce_namelen(ce) == namelen &&
1894                    !memcmp(ce->name, name, namelen))
1895                        return 0; /* Yup, this one exists unmerged */
1896        }
1897        return 1;
1898}