read-cache.con commit fsck: mark strings for translation (674ba34)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 */
   6#define NO_THE_INDEX_COMPATIBILITY_MACROS
   7#include "cache.h"
   8#include "config.h"
   9#include "diff.h"
  10#include "diffcore.h"
  11#include "tempfile.h"
  12#include "lockfile.h"
  13#include "cache-tree.h"
  14#include "refs.h"
  15#include "dir.h"
  16#include "object-store.h"
  17#include "tree.h"
  18#include "commit.h"
  19#include "blob.h"
  20#include "resolve-undo.h"
  21#include "strbuf.h"
  22#include "varint.h"
  23#include "split-index.h"
  24#include "utf8.h"
  25#include "fsmonitor.h"
  26#include "thread-utils.h"
  27#include "progress.h"
  28
  29/* Mask for the name length in ce_flags in the on-disk index */
  30
  31#define CE_NAMEMASK  (0x0fff)
  32
  33/* Index extensions.
  34 *
  35 * The first letter should be 'A'..'Z' for extensions that are not
  36 * necessary for a correct operation (i.e. optimization data).
  37 * When new extensions are added that _needs_ to be understood in
  38 * order to correctly interpret the index file, pick character that
  39 * is outside the range, to cause the reader to abort.
  40 */
  41
  42#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
  43#define CACHE_EXT_TREE 0x54524545       /* "TREE" */
  44#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
  45#define CACHE_EXT_LINK 0x6c696e6b         /* "link" */
  46#define CACHE_EXT_UNTRACKED 0x554E5452    /* "UNTR" */
  47#define CACHE_EXT_FSMONITOR 0x46534D4E    /* "FSMN" */
  48#define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945  /* "EOIE" */
  49#define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
  50
  51/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
  52#define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
  53                 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
  54                 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
  55
  56
  57/*
  58 * This is an estimate of the pathname length in the index.  We use
  59 * this for V4 index files to guess the un-deltafied size of the index
  60 * in memory because of pathname deltafication.  This is not required
  61 * for V2/V3 index formats because their pathnames are not compressed.
  62 * If the initial amount of memory set aside is not sufficient, the
  63 * mem pool will allocate extra memory.
  64 */
  65#define CACHE_ENTRY_PATH_LENGTH 80
  66
  67static inline struct cache_entry *mem_pool__ce_alloc(struct mem_pool *mem_pool, size_t len)
  68{
  69        struct cache_entry *ce;
  70        ce = mem_pool_alloc(mem_pool, cache_entry_size(len));
  71        ce->mem_pool_allocated = 1;
  72        return ce;
  73}
  74
  75static inline struct cache_entry *mem_pool__ce_calloc(struct mem_pool *mem_pool, size_t len)
  76{
  77        struct cache_entry * ce;
  78        ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len));
  79        ce->mem_pool_allocated = 1;
  80        return ce;
  81}
  82
  83static struct mem_pool *find_mem_pool(struct index_state *istate)
  84{
  85        struct mem_pool **pool_ptr;
  86
  87        if (istate->split_index && istate->split_index->base)
  88                pool_ptr = &istate->split_index->base->ce_mem_pool;
  89        else
  90                pool_ptr = &istate->ce_mem_pool;
  91
  92        if (!*pool_ptr)
  93                mem_pool_init(pool_ptr, 0);
  94
  95        return *pool_ptr;
  96}
  97
  98struct index_state the_index;
  99static const char *alternate_index_output;
 100
 101static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
 102{
 103        istate->cache[nr] = ce;
 104        add_name_hash(istate, ce);
 105}
 106
 107static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
 108{
 109        struct cache_entry *old = istate->cache[nr];
 110
 111        replace_index_entry_in_base(istate, old, ce);
 112        remove_name_hash(istate, old);
 113        discard_cache_entry(old);
 114        ce->ce_flags &= ~CE_HASHED;
 115        set_index_entry(istate, nr, ce);
 116        ce->ce_flags |= CE_UPDATE_IN_BASE;
 117        mark_fsmonitor_invalid(istate, ce);
 118        istate->cache_changed |= CE_ENTRY_CHANGED;
 119}
 120
 121void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
 122{
 123        struct cache_entry *old_entry = istate->cache[nr], *new_entry;
 124        int namelen = strlen(new_name);
 125
 126        new_entry = make_empty_cache_entry(istate, namelen);
 127        copy_cache_entry(new_entry, old_entry);
 128        new_entry->ce_flags &= ~CE_HASHED;
 129        new_entry->ce_namelen = namelen;
 130        new_entry->index = 0;
 131        memcpy(new_entry->name, new_name, namelen + 1);
 132
 133        cache_tree_invalidate_path(istate, old_entry->name);
 134        untracked_cache_remove_from_index(istate, old_entry->name);
 135        remove_index_entry_at(istate, nr);
 136        add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
 137}
 138
 139void fill_stat_data(struct stat_data *sd, struct stat *st)
 140{
 141        sd->sd_ctime.sec = (unsigned int)st->st_ctime;
 142        sd->sd_mtime.sec = (unsigned int)st->st_mtime;
 143        sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
 144        sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
 145        sd->sd_dev = st->st_dev;
 146        sd->sd_ino = st->st_ino;
 147        sd->sd_uid = st->st_uid;
 148        sd->sd_gid = st->st_gid;
 149        sd->sd_size = st->st_size;
 150}
 151
 152int match_stat_data(const struct stat_data *sd, struct stat *st)
 153{
 154        int changed = 0;
 155
 156        if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
 157                changed |= MTIME_CHANGED;
 158        if (trust_ctime && check_stat &&
 159            sd->sd_ctime.sec != (unsigned int)st->st_ctime)
 160                changed |= CTIME_CHANGED;
 161
 162#ifdef USE_NSEC
 163        if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
 164                changed |= MTIME_CHANGED;
 165        if (trust_ctime && check_stat &&
 166            sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
 167                changed |= CTIME_CHANGED;
 168#endif
 169
 170        if (check_stat) {
 171                if (sd->sd_uid != (unsigned int) st->st_uid ||
 172                        sd->sd_gid != (unsigned int) st->st_gid)
 173                        changed |= OWNER_CHANGED;
 174                if (sd->sd_ino != (unsigned int) st->st_ino)
 175                        changed |= INODE_CHANGED;
 176        }
 177
 178#ifdef USE_STDEV
 179        /*
 180         * st_dev breaks on network filesystems where different
 181         * clients will have different views of what "device"
 182         * the filesystem is on
 183         */
 184        if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
 185                        changed |= INODE_CHANGED;
 186#endif
 187
 188        if (sd->sd_size != (unsigned int) st->st_size)
 189                changed |= DATA_CHANGED;
 190
 191        return changed;
 192}
 193
 194/*
 195 * This only updates the "non-critical" parts of the directory
 196 * cache, ie the parts that aren't tracked by GIT, and only used
 197 * to validate the cache.
 198 */
 199void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 200{
 201        fill_stat_data(&ce->ce_stat_data, st);
 202
 203        if (assume_unchanged)
 204                ce->ce_flags |= CE_VALID;
 205
 206        if (S_ISREG(st->st_mode)) {
 207                ce_mark_uptodate(ce);
 208                mark_fsmonitor_valid(ce);
 209        }
 210}
 211
 212static int ce_compare_data(struct index_state *istate,
 213                           const struct cache_entry *ce,
 214                           struct stat *st)
 215{
 216        int match = -1;
 217        int fd = git_open_cloexec(ce->name, O_RDONLY);
 218
 219        if (fd >= 0) {
 220                struct object_id oid;
 221                if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0))
 222                        match = !oideq(&oid, &ce->oid);
 223                /* index_fd() closed the file descriptor already */
 224        }
 225        return match;
 226}
 227
 228static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
 229{
 230        int match = -1;
 231        void *buffer;
 232        unsigned long size;
 233        enum object_type type;
 234        struct strbuf sb = STRBUF_INIT;
 235
 236        if (strbuf_readlink(&sb, ce->name, expected_size))
 237                return -1;
 238
 239        buffer = read_object_file(&ce->oid, &type, &size);
 240        if (buffer) {
 241                if (size == sb.len)
 242                        match = memcmp(buffer, sb.buf, size);
 243                free(buffer);
 244        }
 245        strbuf_release(&sb);
 246        return match;
 247}
 248
 249static int ce_compare_gitlink(const struct cache_entry *ce)
 250{
 251        struct object_id oid;
 252
 253        /*
 254         * We don't actually require that the .git directory
 255         * under GITLINK directory be a valid git directory. It
 256         * might even be missing (in case nobody populated that
 257         * sub-project).
 258         *
 259         * If so, we consider it always to match.
 260         */
 261        if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
 262                return 0;
 263        return !oideq(&oid, &ce->oid);
 264}
 265
 266static int ce_modified_check_fs(struct index_state *istate,
 267                                const struct cache_entry *ce,
 268                                struct stat *st)
 269{
 270        switch (st->st_mode & S_IFMT) {
 271        case S_IFREG:
 272                if (ce_compare_data(istate, ce, st))
 273                        return DATA_CHANGED;
 274                break;
 275        case S_IFLNK:
 276                if (ce_compare_link(ce, xsize_t(st->st_size)))
 277                        return DATA_CHANGED;
 278                break;
 279        case S_IFDIR:
 280                if (S_ISGITLINK(ce->ce_mode))
 281                        return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
 282                /* else fallthrough */
 283        default:
 284                return TYPE_CHANGED;
 285        }
 286        return 0;
 287}
 288
 289static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
 290{
 291        unsigned int changed = 0;
 292
 293        if (ce->ce_flags & CE_REMOVE)
 294                return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
 295
 296        switch (ce->ce_mode & S_IFMT) {
 297        case S_IFREG:
 298                changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
 299                /* We consider only the owner x bit to be relevant for
 300                 * "mode changes"
 301                 */
 302                if (trust_executable_bit &&
 303                    (0100 & (ce->ce_mode ^ st->st_mode)))
 304                        changed |= MODE_CHANGED;
 305                break;
 306        case S_IFLNK:
 307                if (!S_ISLNK(st->st_mode) &&
 308                    (has_symlinks || !S_ISREG(st->st_mode)))
 309                        changed |= TYPE_CHANGED;
 310                break;
 311        case S_IFGITLINK:
 312                /* We ignore most of the st_xxx fields for gitlinks */
 313                if (!S_ISDIR(st->st_mode))
 314                        changed |= TYPE_CHANGED;
 315                else if (ce_compare_gitlink(ce))
 316                        changed |= DATA_CHANGED;
 317                return changed;
 318        default:
 319                BUG("unsupported ce_mode: %o", ce->ce_mode);
 320        }
 321
 322        changed |= match_stat_data(&ce->ce_stat_data, st);
 323
 324        /* Racily smudged entry? */
 325        if (!ce->ce_stat_data.sd_size) {
 326                if (!is_empty_blob_sha1(ce->oid.hash))
 327                        changed |= DATA_CHANGED;
 328        }
 329
 330        return changed;
 331}
 332
 333static int is_racy_stat(const struct index_state *istate,
 334                        const struct stat_data *sd)
 335{
 336        return (istate->timestamp.sec &&
 337#ifdef USE_NSEC
 338                 /* nanosecond timestamped files can also be racy! */
 339                (istate->timestamp.sec < sd->sd_mtime.sec ||
 340                 (istate->timestamp.sec == sd->sd_mtime.sec &&
 341                  istate->timestamp.nsec <= sd->sd_mtime.nsec))
 342#else
 343                istate->timestamp.sec <= sd->sd_mtime.sec
 344#endif
 345                );
 346}
 347
 348int is_racy_timestamp(const struct index_state *istate,
 349                             const struct cache_entry *ce)
 350{
 351        return (!S_ISGITLINK(ce->ce_mode) &&
 352                is_racy_stat(istate, &ce->ce_stat_data));
 353}
 354
 355int match_stat_data_racy(const struct index_state *istate,
 356                         const struct stat_data *sd, struct stat *st)
 357{
 358        if (is_racy_stat(istate, sd))
 359                return MTIME_CHANGED;
 360        return match_stat_data(sd, st);
 361}
 362
 363int ie_match_stat(struct index_state *istate,
 364                  const struct cache_entry *ce, struct stat *st,
 365                  unsigned int options)
 366{
 367        unsigned int changed;
 368        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
 369        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
 370        int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
 371        int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
 372
 373        if (!ignore_fsmonitor)
 374                refresh_fsmonitor(istate);
 375        /*
 376         * If it's marked as always valid in the index, it's
 377         * valid whatever the checked-out copy says.
 378         *
 379         * skip-worktree has the same effect with higher precedence
 380         */
 381        if (!ignore_skip_worktree && ce_skip_worktree(ce))
 382                return 0;
 383        if (!ignore_valid && (ce->ce_flags & CE_VALID))
 384                return 0;
 385        if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
 386                return 0;
 387
 388        /*
 389         * Intent-to-add entries have not been added, so the index entry
 390         * by definition never matches what is in the work tree until it
 391         * actually gets added.
 392         */
 393        if (ce_intent_to_add(ce))
 394                return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
 395
 396        changed = ce_match_stat_basic(ce, st);
 397
 398        /*
 399         * Within 1 second of this sequence:
 400         *      echo xyzzy >file && git-update-index --add file
 401         * running this command:
 402         *      echo frotz >file
 403         * would give a falsely clean cache entry.  The mtime and
 404         * length match the cache, and other stat fields do not change.
 405         *
 406         * We could detect this at update-index time (the cache entry
 407         * being registered/updated records the same time as "now")
 408         * and delay the return from git-update-index, but that would
 409         * effectively mean we can make at most one commit per second,
 410         * which is not acceptable.  Instead, we check cache entries
 411         * whose mtime are the same as the index file timestamp more
 412         * carefully than others.
 413         */
 414        if (!changed && is_racy_timestamp(istate, ce)) {
 415                if (assume_racy_is_modified)
 416                        changed |= DATA_CHANGED;
 417                else
 418                        changed |= ce_modified_check_fs(istate, ce, st);
 419        }
 420
 421        return changed;
 422}
 423
 424int ie_modified(struct index_state *istate,
 425                const struct cache_entry *ce,
 426                struct stat *st, unsigned int options)
 427{
 428        int changed, changed_fs;
 429
 430        changed = ie_match_stat(istate, ce, st, options);
 431        if (!changed)
 432                return 0;
 433        /*
 434         * If the mode or type has changed, there's no point in trying
 435         * to refresh the entry - it's not going to match
 436         */
 437        if (changed & (MODE_CHANGED | TYPE_CHANGED))
 438                return changed;
 439
 440        /*
 441         * Immediately after read-tree or update-index --cacheinfo,
 442         * the length field is zero, as we have never even read the
 443         * lstat(2) information once, and we cannot trust DATA_CHANGED
 444         * returned by ie_match_stat() which in turn was returned by
 445         * ce_match_stat_basic() to signal that the filesize of the
 446         * blob changed.  We have to actually go to the filesystem to
 447         * see if the contents match, and if so, should answer "unchanged".
 448         *
 449         * The logic does not apply to gitlinks, as ce_match_stat_basic()
 450         * already has checked the actual HEAD from the filesystem in the
 451         * subproject.  If ie_match_stat() already said it is different,
 452         * then we know it is.
 453         */
 454        if ((changed & DATA_CHANGED) &&
 455            (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
 456                return changed;
 457
 458        changed_fs = ce_modified_check_fs(istate, ce, st);
 459        if (changed_fs)
 460                return changed | changed_fs;
 461        return 0;
 462}
 463
 464int base_name_compare(const char *name1, int len1, int mode1,
 465                      const char *name2, int len2, int mode2)
 466{
 467        unsigned char c1, c2;
 468        int len = len1 < len2 ? len1 : len2;
 469        int cmp;
 470
 471        cmp = memcmp(name1, name2, len);
 472        if (cmp)
 473                return cmp;
 474        c1 = name1[len];
 475        c2 = name2[len];
 476        if (!c1 && S_ISDIR(mode1))
 477                c1 = '/';
 478        if (!c2 && S_ISDIR(mode2))
 479                c2 = '/';
 480        return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 481}
 482
 483/*
 484 * df_name_compare() is identical to base_name_compare(), except it
 485 * compares conflicting directory/file entries as equal. Note that
 486 * while a directory name compares as equal to a regular file, they
 487 * then individually compare _differently_ to a filename that has
 488 * a dot after the basename (because '\0' < '.' < '/').
 489 *
 490 * This is used by routines that want to traverse the git namespace
 491 * but then handle conflicting entries together when possible.
 492 */
 493int df_name_compare(const char *name1, int len1, int mode1,
 494                    const char *name2, int len2, int mode2)
 495{
 496        int len = len1 < len2 ? len1 : len2, cmp;
 497        unsigned char c1, c2;
 498
 499        cmp = memcmp(name1, name2, len);
 500        if (cmp)
 501                return cmp;
 502        /* Directories and files compare equal (same length, same name) */
 503        if (len1 == len2)
 504                return 0;
 505        c1 = name1[len];
 506        if (!c1 && S_ISDIR(mode1))
 507                c1 = '/';
 508        c2 = name2[len];
 509        if (!c2 && S_ISDIR(mode2))
 510                c2 = '/';
 511        if (c1 == '/' && !c2)
 512                return 0;
 513        if (c2 == '/' && !c1)
 514                return 0;
 515        return c1 - c2;
 516}
 517
 518int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
 519{
 520        size_t min_len = (len1 < len2) ? len1 : len2;
 521        int cmp = memcmp(name1, name2, min_len);
 522        if (cmp)
 523                return cmp;
 524        if (len1 < len2)
 525                return -1;
 526        if (len1 > len2)
 527                return 1;
 528        return 0;
 529}
 530
 531int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
 532{
 533        int cmp;
 534
 535        cmp = name_compare(name1, len1, name2, len2);
 536        if (cmp)
 537                return cmp;
 538
 539        if (stage1 < stage2)
 540                return -1;
 541        if (stage1 > stage2)
 542                return 1;
 543        return 0;
 544}
 545
 546static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
 547{
 548        int first, last;
 549
 550        first = 0;
 551        last = istate->cache_nr;
 552        while (last > first) {
 553                int next = (last + first) >> 1;
 554                struct cache_entry *ce = istate->cache[next];
 555                int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
 556                if (!cmp)
 557                        return next;
 558                if (cmp < 0) {
 559                        last = next;
 560                        continue;
 561                }
 562                first = next+1;
 563        }
 564        return -first-1;
 565}
 566
 567int index_name_pos(const struct index_state *istate, const char *name, int namelen)
 568{
 569        return index_name_stage_pos(istate, name, namelen, 0);
 570}
 571
 572int remove_index_entry_at(struct index_state *istate, int pos)
 573{
 574        struct cache_entry *ce = istate->cache[pos];
 575
 576        record_resolve_undo(istate, ce);
 577        remove_name_hash(istate, ce);
 578        save_or_free_index_entry(istate, ce);
 579        istate->cache_changed |= CE_ENTRY_REMOVED;
 580        istate->cache_nr--;
 581        if (pos >= istate->cache_nr)
 582                return 0;
 583        MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
 584                   istate->cache_nr - pos);
 585        return 1;
 586}
 587
 588/*
 589 * Remove all cache entries marked for removal, that is where
 590 * CE_REMOVE is set in ce_flags.  This is much more effective than
 591 * calling remove_index_entry_at() for each entry to be removed.
 592 */
 593void remove_marked_cache_entries(struct index_state *istate)
 594{
 595        struct cache_entry **ce_array = istate->cache;
 596        unsigned int i, j;
 597
 598        for (i = j = 0; i < istate->cache_nr; i++) {
 599                if (ce_array[i]->ce_flags & CE_REMOVE) {
 600                        remove_name_hash(istate, ce_array[i]);
 601                        save_or_free_index_entry(istate, ce_array[i]);
 602                }
 603                else
 604                        ce_array[j++] = ce_array[i];
 605        }
 606        if (j == istate->cache_nr)
 607                return;
 608        istate->cache_changed |= CE_ENTRY_REMOVED;
 609        istate->cache_nr = j;
 610}
 611
 612int remove_file_from_index(struct index_state *istate, const char *path)
 613{
 614        int pos = index_name_pos(istate, path, strlen(path));
 615        if (pos < 0)
 616                pos = -pos-1;
 617        cache_tree_invalidate_path(istate, path);
 618        untracked_cache_remove_from_index(istate, path);
 619        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
 620                remove_index_entry_at(istate, pos);
 621        return 0;
 622}
 623
 624static int compare_name(struct cache_entry *ce, const char *path, int namelen)
 625{
 626        return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
 627}
 628
 629static int index_name_pos_also_unmerged(struct index_state *istate,
 630        const char *path, int namelen)
 631{
 632        int pos = index_name_pos(istate, path, namelen);
 633        struct cache_entry *ce;
 634
 635        if (pos >= 0)
 636                return pos;
 637
 638        /* maybe unmerged? */
 639        pos = -1 - pos;
 640        if (pos >= istate->cache_nr ||
 641                        compare_name((ce = istate->cache[pos]), path, namelen))
 642                return -1;
 643
 644        /* order of preference: stage 2, 1, 3 */
 645        if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
 646                        ce_stage((ce = istate->cache[pos + 1])) == 2 &&
 647                        !compare_name(ce, path, namelen))
 648                pos++;
 649        return pos;
 650}
 651
 652static int different_name(struct cache_entry *ce, struct cache_entry *alias)
 653{
 654        int len = ce_namelen(ce);
 655        return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
 656}
 657
 658/*
 659 * If we add a filename that aliases in the cache, we will use the
 660 * name that we already have - but we don't want to update the same
 661 * alias twice, because that implies that there were actually two
 662 * different files with aliasing names!
 663 *
 664 * So we use the CE_ADDED flag to verify that the alias was an old
 665 * one before we accept it as
 666 */
 667static struct cache_entry *create_alias_ce(struct index_state *istate,
 668                                           struct cache_entry *ce,
 669                                           struct cache_entry *alias)
 670{
 671        int len;
 672        struct cache_entry *new_entry;
 673
 674        if (alias->ce_flags & CE_ADDED)
 675                die(_("will not add file alias '%s' ('%s' already exists in index)"),
 676                    ce->name, alias->name);
 677
 678        /* Ok, create the new entry using the name of the existing alias */
 679        len = ce_namelen(alias);
 680        new_entry = make_empty_cache_entry(istate, len);
 681        memcpy(new_entry->name, alias->name, len);
 682        copy_cache_entry(new_entry, ce);
 683        save_or_free_index_entry(istate, ce);
 684        return new_entry;
 685}
 686
 687void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
 688{
 689        struct object_id oid;
 690        if (write_object_file("", 0, blob_type, &oid))
 691                die(_("cannot create an empty blob in the object database"));
 692        oidcpy(&ce->oid, &oid);
 693}
 694
 695int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
 696{
 697        int namelen, was_same;
 698        mode_t st_mode = st->st_mode;
 699        struct cache_entry *ce, *alias = NULL;
 700        unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
 701        int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
 702        int pretend = flags & ADD_CACHE_PRETEND;
 703        int intent_only = flags & ADD_CACHE_INTENT;
 704        int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
 705                          (intent_only ? ADD_CACHE_NEW_ONLY : 0));
 706        int newflags = HASH_WRITE_OBJECT;
 707
 708        if (flags & HASH_RENORMALIZE)
 709                newflags |= HASH_RENORMALIZE;
 710
 711        if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
 712                return error(_("%s: can only add regular files, symbolic links or git-directories"), path);
 713
 714        namelen = strlen(path);
 715        if (S_ISDIR(st_mode)) {
 716                while (namelen && path[namelen-1] == '/')
 717                        namelen--;
 718        }
 719        ce = make_empty_cache_entry(istate, namelen);
 720        memcpy(ce->name, path, namelen);
 721        ce->ce_namelen = namelen;
 722        if (!intent_only)
 723                fill_stat_cache_info(ce, st);
 724        else
 725                ce->ce_flags |= CE_INTENT_TO_ADD;
 726
 727
 728        if (trust_executable_bit && has_symlinks) {
 729                ce->ce_mode = create_ce_mode(st_mode);
 730        } else {
 731                /* If there is an existing entry, pick the mode bits and type
 732                 * from it, otherwise assume unexecutable regular file.
 733                 */
 734                struct cache_entry *ent;
 735                int pos = index_name_pos_also_unmerged(istate, path, namelen);
 736
 737                ent = (0 <= pos) ? istate->cache[pos] : NULL;
 738                ce->ce_mode = ce_mode_from_stat(ent, st_mode);
 739        }
 740
 741        /* When core.ignorecase=true, determine if a directory of the same name but differing
 742         * case already exists within the Git repository.  If it does, ensure the directory
 743         * case of the file being added to the repository matches (is folded into) the existing
 744         * entry's directory case.
 745         */
 746        if (ignore_case) {
 747                adjust_dirname_case(istate, ce->name);
 748        }
 749        if (!(flags & HASH_RENORMALIZE)) {
 750                alias = index_file_exists(istate, ce->name,
 751                                          ce_namelen(ce), ignore_case);
 752                if (alias &&
 753                    !ce_stage(alias) &&
 754                    !ie_match_stat(istate, alias, st, ce_option)) {
 755                        /* Nothing changed, really */
 756                        if (!S_ISGITLINK(alias->ce_mode))
 757                                ce_mark_uptodate(alias);
 758                        alias->ce_flags |= CE_ADDED;
 759
 760                        discard_cache_entry(ce);
 761                        return 0;
 762                }
 763        }
 764        if (!intent_only) {
 765                if (index_path(istate, &ce->oid, path, st, newflags)) {
 766                        discard_cache_entry(ce);
 767                        return error(_("unable to index file '%s'"), path);
 768                }
 769        } else
 770                set_object_name_for_intent_to_add_entry(ce);
 771
 772        if (ignore_case && alias && different_name(ce, alias))
 773                ce = create_alias_ce(istate, ce, alias);
 774        ce->ce_flags |= CE_ADDED;
 775
 776        /* It was suspected to be racily clean, but it turns out to be Ok */
 777        was_same = (alias &&
 778                    !ce_stage(alias) &&
 779                    oideq(&alias->oid, &ce->oid) &&
 780                    ce->ce_mode == alias->ce_mode);
 781
 782        if (pretend)
 783                discard_cache_entry(ce);
 784        else if (add_index_entry(istate, ce, add_option)) {
 785                discard_cache_entry(ce);
 786                return error(_("unable to add '%s' to index"), path);
 787        }
 788        if (verbose && !was_same)
 789                printf("add '%s'\n", path);
 790        return 0;
 791}
 792
 793int add_file_to_index(struct index_state *istate, const char *path, int flags)
 794{
 795        struct stat st;
 796        if (lstat(path, &st))
 797                die_errno(_("unable to stat '%s'"), path);
 798        return add_to_index(istate, path, &st, flags);
 799}
 800
 801struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
 802{
 803        return mem_pool__ce_calloc(find_mem_pool(istate), len);
 804}
 805
 806struct cache_entry *make_empty_transient_cache_entry(size_t len)
 807{
 808        return xcalloc(1, cache_entry_size(len));
 809}
 810
 811struct cache_entry *make_cache_entry(struct index_state *istate,
 812                                     unsigned int mode,
 813                                     const struct object_id *oid,
 814                                     const char *path,
 815                                     int stage,
 816                                     unsigned int refresh_options)
 817{
 818        struct cache_entry *ce, *ret;
 819        int len;
 820
 821        if (!verify_path(path, mode)) {
 822                error(_("invalid path '%s'"), path);
 823                return NULL;
 824        }
 825
 826        len = strlen(path);
 827        ce = make_empty_cache_entry(istate, len);
 828
 829        oidcpy(&ce->oid, oid);
 830        memcpy(ce->name, path, len);
 831        ce->ce_flags = create_ce_flags(stage);
 832        ce->ce_namelen = len;
 833        ce->ce_mode = create_ce_mode(mode);
 834
 835        ret = refresh_cache_entry(istate, ce, refresh_options);
 836        if (ret != ce)
 837                discard_cache_entry(ce);
 838        return ret;
 839}
 840
 841struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid,
 842                                               const char *path, int stage)
 843{
 844        struct cache_entry *ce;
 845        int len;
 846
 847        if (!verify_path(path, mode)) {
 848                error(_("invalid path '%s'"), path);
 849                return NULL;
 850        }
 851
 852        len = strlen(path);
 853        ce = make_empty_transient_cache_entry(len);
 854
 855        oidcpy(&ce->oid, oid);
 856        memcpy(ce->name, path, len);
 857        ce->ce_flags = create_ce_flags(stage);
 858        ce->ce_namelen = len;
 859        ce->ce_mode = create_ce_mode(mode);
 860
 861        return ce;
 862}
 863
 864/*
 865 * Chmod an index entry with either +x or -x.
 866 *
 867 * Returns -1 if the chmod for the particular cache entry failed (if it's
 868 * not a regular file), -2 if an invalid flip argument is passed in, 0
 869 * otherwise.
 870 */
 871int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
 872                      char flip)
 873{
 874        if (!S_ISREG(ce->ce_mode))
 875                return -1;
 876        switch (flip) {
 877        case '+':
 878                ce->ce_mode |= 0111;
 879                break;
 880        case '-':
 881                ce->ce_mode &= ~0111;
 882                break;
 883        default:
 884                return -2;
 885        }
 886        cache_tree_invalidate_path(istate, ce->name);
 887        ce->ce_flags |= CE_UPDATE_IN_BASE;
 888        mark_fsmonitor_invalid(istate, ce);
 889        istate->cache_changed |= CE_ENTRY_CHANGED;
 890
 891        return 0;
 892}
 893
 894int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
 895{
 896        int len = ce_namelen(a);
 897        return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 898}
 899
 900/*
 901 * We fundamentally don't like some paths: we don't want
 902 * dot or dot-dot anywhere, and for obvious reasons don't
 903 * want to recurse into ".git" either.
 904 *
 905 * Also, we don't want double slashes or slashes at the
 906 * end that can make pathnames ambiguous.
 907 */
 908static int verify_dotfile(const char *rest, unsigned mode)
 909{
 910        /*
 911         * The first character was '.', but that
 912         * has already been discarded, we now test
 913         * the rest.
 914         */
 915
 916        /* "." is not allowed */
 917        if (*rest == '\0' || is_dir_sep(*rest))
 918                return 0;
 919
 920        switch (*rest) {
 921        /*
 922         * ".git" followed by NUL or slash is bad. Note that we match
 923         * case-insensitively here, even if ignore_case is not set.
 924         * This outlaws ".GIT" everywhere out of an abundance of caution,
 925         * since there's really no good reason to allow it.
 926         *
 927         * Once we've seen ".git", we can also find ".gitmodules", etc (also
 928         * case-insensitively).
 929         */
 930        case 'g':
 931        case 'G':
 932                if (rest[1] != 'i' && rest[1] != 'I')
 933                        break;
 934                if (rest[2] != 't' && rest[2] != 'T')
 935                        break;
 936                if (rest[3] == '\0' || is_dir_sep(rest[3]))
 937                        return 0;
 938                if (S_ISLNK(mode)) {
 939                        rest += 3;
 940                        if (skip_iprefix(rest, "modules", &rest) &&
 941                            (*rest == '\0' || is_dir_sep(*rest)))
 942                                return 0;
 943                }
 944                break;
 945        case '.':
 946                if (rest[1] == '\0' || is_dir_sep(rest[1]))
 947                        return 0;
 948        }
 949        return 1;
 950}
 951
 952int verify_path(const char *path, unsigned mode)
 953{
 954        char c;
 955
 956        if (has_dos_drive_prefix(path))
 957                return 0;
 958
 959        goto inside;
 960        for (;;) {
 961                if (!c)
 962                        return 1;
 963                if (is_dir_sep(c)) {
 964inside:
 965                        if (protect_hfs) {
 966                                if (is_hfs_dotgit(path))
 967                                        return 0;
 968                                if (S_ISLNK(mode)) {
 969                                        if (is_hfs_dotgitmodules(path))
 970                                                return 0;
 971                                }
 972                        }
 973                        if (protect_ntfs) {
 974                                if (is_ntfs_dotgit(path))
 975                                        return 0;
 976                                if (S_ISLNK(mode)) {
 977                                        if (is_ntfs_dotgitmodules(path))
 978                                                return 0;
 979                                }
 980                        }
 981
 982                        c = *path++;
 983                        if ((c == '.' && !verify_dotfile(path, mode)) ||
 984                            is_dir_sep(c) || c == '\0')
 985                                return 0;
 986                }
 987                c = *path++;
 988        }
 989}
 990
 991/*
 992 * Do we have another file that has the beginning components being a
 993 * proper superset of the name we're trying to add?
 994 */
 995static int has_file_name(struct index_state *istate,
 996                         const struct cache_entry *ce, int pos, int ok_to_replace)
 997{
 998        int retval = 0;
 999        int len = ce_namelen(ce);
1000        int stage = ce_stage(ce);
1001        const char *name = ce->name;
1002
1003        while (pos < istate->cache_nr) {
1004                struct cache_entry *p = istate->cache[pos++];
1005
1006                if (len >= ce_namelen(p))
1007                        break;
1008                if (memcmp(name, p->name, len))
1009                        break;
1010                if (ce_stage(p) != stage)
1011                        continue;
1012                if (p->name[len] != '/')
1013                        continue;
1014                if (p->ce_flags & CE_REMOVE)
1015                        continue;
1016                retval = -1;
1017                if (!ok_to_replace)
1018                        break;
1019                remove_index_entry_at(istate, --pos);
1020        }
1021        return retval;
1022}
1023
1024
1025/*
1026 * Like strcmp(), but also return the offset of the first change.
1027 * If strings are equal, return the length.
1028 */
1029int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1030{
1031        size_t k;
1032
1033        if (!first_change)
1034                return strcmp(s1, s2);
1035
1036        for (k = 0; s1[k] == s2[k]; k++)
1037                if (s1[k] == '\0')
1038                        break;
1039
1040        *first_change = k;
1041        return (unsigned char)s1[k] - (unsigned char)s2[k];
1042}
1043
1044/*
1045 * Do we have another file with a pathname that is a proper
1046 * subset of the name we're trying to add?
1047 *
1048 * That is, is there another file in the index with a path
1049 * that matches a sub-directory in the given entry?
1050 */
1051static int has_dir_name(struct index_state *istate,
1052                        const struct cache_entry *ce, int pos, int ok_to_replace)
1053{
1054        int retval = 0;
1055        int stage = ce_stage(ce);
1056        const char *name = ce->name;
1057        const char *slash = name + ce_namelen(ce);
1058        size_t len_eq_last;
1059        int cmp_last = 0;
1060
1061        /*
1062         * We are frequently called during an iteration on a sorted
1063         * list of pathnames and while building a new index.  Therefore,
1064         * there is a high probability that this entry will eventually
1065         * be appended to the index, rather than inserted in the middle.
1066         * If we can confirm that, we can avoid binary searches on the
1067         * components of the pathname.
1068         *
1069         * Compare the entry's full path with the last path in the index.
1070         */
1071        if (istate->cache_nr > 0) {
1072                cmp_last = strcmp_offset(name,
1073                        istate->cache[istate->cache_nr - 1]->name,
1074                        &len_eq_last);
1075                if (cmp_last > 0) {
1076                        if (len_eq_last == 0) {
1077                                /*
1078                                 * The entry sorts AFTER the last one in the
1079                                 * index and their paths have no common prefix,
1080                                 * so there cannot be a F/D conflict.
1081                                 */
1082                                return retval;
1083                        } else {
1084                                /*
1085                                 * The entry sorts AFTER the last one in the
1086                                 * index, but has a common prefix.  Fall through
1087                                 * to the loop below to disect the entry's path
1088                                 * and see where the difference is.
1089                                 */
1090                        }
1091                } else if (cmp_last == 0) {
1092                        /*
1093                         * The entry exactly matches the last one in the
1094                         * index, but because of multiple stage and CE_REMOVE
1095                         * items, we fall through and let the regular search
1096                         * code handle it.
1097                         */
1098                }
1099        }
1100
1101        for (;;) {
1102                size_t len;
1103
1104                for (;;) {
1105                        if (*--slash == '/')
1106                                break;
1107                        if (slash <= ce->name)
1108                                return retval;
1109                }
1110                len = slash - name;
1111
1112                if (cmp_last > 0) {
1113                        /*
1114                         * (len + 1) is a directory boundary (including
1115                         * the trailing slash).  And since the loop is
1116                         * decrementing "slash", the first iteration is
1117                         * the longest directory prefix; subsequent
1118                         * iterations consider parent directories.
1119                         */
1120
1121                        if (len + 1 <= len_eq_last) {
1122                                /*
1123                                 * The directory prefix (including the trailing
1124                                 * slash) also appears as a prefix in the last
1125                                 * entry, so the remainder cannot collide (because
1126                                 * strcmp said the whole path was greater).
1127                                 *
1128                                 * EQ: last: xxx/A
1129                                 *     this: xxx/B
1130                                 *
1131                                 * LT: last: xxx/file_A
1132                                 *     this: xxx/file_B
1133                                 */
1134                                return retval;
1135                        }
1136
1137                        if (len > len_eq_last) {
1138                                /*
1139                                 * This part of the directory prefix (excluding
1140                                 * the trailing slash) is longer than the known
1141                                 * equal portions, so this sub-directory cannot
1142                                 * collide with a file.
1143                                 *
1144                                 * GT: last: xxxA
1145                                 *     this: xxxB/file
1146                                 */
1147                                return retval;
1148                        }
1149
1150                        if (istate->cache_nr > 0 &&
1151                                ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1152                                /*
1153                                 * The directory prefix lines up with part of
1154                                 * a longer file or directory name, but sorts
1155                                 * after it, so this sub-directory cannot
1156                                 * collide with a file.
1157                                 *
1158                                 * last: xxx/yy-file (because '-' sorts before '/')
1159                                 * this: xxx/yy/abc
1160                                 */
1161                                return retval;
1162                        }
1163
1164                        /*
1165                         * This is a possible collision. Fall through and
1166                         * let the regular search code handle it.
1167                         *
1168                         * last: xxx
1169                         * this: xxx/file
1170                         */
1171                }
1172
1173                pos = index_name_stage_pos(istate, name, len, stage);
1174                if (pos >= 0) {
1175                        /*
1176                         * Found one, but not so fast.  This could
1177                         * be a marker that says "I was here, but
1178                         * I am being removed".  Such an entry is
1179                         * not a part of the resulting tree, and
1180                         * it is Ok to have a directory at the same
1181                         * path.
1182                         */
1183                        if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1184                                retval = -1;
1185                                if (!ok_to_replace)
1186                                        break;
1187                                remove_index_entry_at(istate, pos);
1188                                continue;
1189                        }
1190                }
1191                else
1192                        pos = -pos-1;
1193
1194                /*
1195                 * Trivial optimization: if we find an entry that
1196                 * already matches the sub-directory, then we know
1197                 * we're ok, and we can exit.
1198                 */
1199                while (pos < istate->cache_nr) {
1200                        struct cache_entry *p = istate->cache[pos];
1201                        if ((ce_namelen(p) <= len) ||
1202                            (p->name[len] != '/') ||
1203                            memcmp(p->name, name, len))
1204                                break; /* not our subdirectory */
1205                        if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1206                                /*
1207                                 * p is at the same stage as our entry, and
1208                                 * is a subdirectory of what we are looking
1209                                 * at, so we cannot have conflicts at our
1210                                 * level or anything shorter.
1211                                 */
1212                                return retval;
1213                        pos++;
1214                }
1215        }
1216        return retval;
1217}
1218
1219/* We may be in a situation where we already have path/file and path
1220 * is being added, or we already have path and path/file is being
1221 * added.  Either one would result in a nonsense tree that has path
1222 * twice when git-write-tree tries to write it out.  Prevent it.
1223 *
1224 * If ok-to-replace is specified, we remove the conflicting entries
1225 * from the cache so the caller should recompute the insert position.
1226 * When this happens, we return non-zero.
1227 */
1228static int check_file_directory_conflict(struct index_state *istate,
1229                                         const struct cache_entry *ce,
1230                                         int pos, int ok_to_replace)
1231{
1232        int retval;
1233
1234        /*
1235         * When ce is an "I am going away" entry, we allow it to be added
1236         */
1237        if (ce->ce_flags & CE_REMOVE)
1238                return 0;
1239
1240        /*
1241         * We check if the path is a sub-path of a subsequent pathname
1242         * first, since removing those will not change the position
1243         * in the array.
1244         */
1245        retval = has_file_name(istate, ce, pos, ok_to_replace);
1246
1247        /*
1248         * Then check if the path might have a clashing sub-directory
1249         * before it.
1250         */
1251        return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1252}
1253
1254static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1255{
1256        int pos;
1257        int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1258        int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1259        int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1260        int new_only = option & ADD_CACHE_NEW_ONLY;
1261
1262        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1263                cache_tree_invalidate_path(istate, ce->name);
1264
1265        /*
1266         * If this entry's path sorts after the last entry in the index,
1267         * we can avoid searching for it.
1268         */
1269        if (istate->cache_nr > 0 &&
1270                strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1271                pos = -istate->cache_nr - 1;
1272        else
1273                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1274
1275        /* existing match? Just replace it. */
1276        if (pos >= 0) {
1277                if (!new_only)
1278                        replace_index_entry(istate, pos, ce);
1279                return 0;
1280        }
1281        pos = -pos-1;
1282
1283        if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1284                untracked_cache_add_to_index(istate, ce->name);
1285
1286        /*
1287         * Inserting a merged entry ("stage 0") into the index
1288         * will always replace all non-merged entries..
1289         */
1290        if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1291                while (ce_same_name(istate->cache[pos], ce)) {
1292                        ok_to_add = 1;
1293                        if (!remove_index_entry_at(istate, pos))
1294                                break;
1295                }
1296        }
1297
1298        if (!ok_to_add)
1299                return -1;
1300        if (!verify_path(ce->name, ce->ce_mode))
1301                return error(_("invalid path '%s'"), ce->name);
1302
1303        if (!skip_df_check &&
1304            check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1305                if (!ok_to_replace)
1306                        return error(_("'%s' appears as both a file and as a directory"),
1307                                     ce->name);
1308                pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1309                pos = -pos-1;
1310        }
1311        return pos + 1;
1312}
1313
1314int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1315{
1316        int pos;
1317
1318        if (option & ADD_CACHE_JUST_APPEND)
1319                pos = istate->cache_nr;
1320        else {
1321                int ret;
1322                ret = add_index_entry_with_check(istate, ce, option);
1323                if (ret <= 0)
1324                        return ret;
1325                pos = ret - 1;
1326        }
1327
1328        /* Make sure the array is big enough .. */
1329        ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1330
1331        /* Add it in.. */
1332        istate->cache_nr++;
1333        if (istate->cache_nr > pos + 1)
1334                MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1335                           istate->cache_nr - pos - 1);
1336        set_index_entry(istate, pos, ce);
1337        istate->cache_changed |= CE_ENTRY_ADDED;
1338        return 0;
1339}
1340
1341/*
1342 * "refresh" does not calculate a new sha1 file or bring the
1343 * cache up-to-date for mode/content changes. But what it
1344 * _does_ do is to "re-match" the stat information of a file
1345 * with the cache, so that you can refresh the cache for a
1346 * file that hasn't been changed but where the stat entry is
1347 * out of date.
1348 *
1349 * For example, you'd want to do this after doing a "git-read-tree",
1350 * to link up the stat cache details with the proper files.
1351 */
1352static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1353                                             struct cache_entry *ce,
1354                                             unsigned int options, int *err,
1355                                             int *changed_ret)
1356{
1357        struct stat st;
1358        struct cache_entry *updated;
1359        int changed;
1360        int refresh = options & CE_MATCH_REFRESH;
1361        int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1362        int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1363        int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1364        int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1365
1366        if (!refresh || ce_uptodate(ce))
1367                return ce;
1368
1369        if (!ignore_fsmonitor)
1370                refresh_fsmonitor(istate);
1371        /*
1372         * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1373         * that the change to the work tree does not matter and told
1374         * us not to worry.
1375         */
1376        if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1377                ce_mark_uptodate(ce);
1378                return ce;
1379        }
1380        if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1381                ce_mark_uptodate(ce);
1382                return ce;
1383        }
1384        if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1385                ce_mark_uptodate(ce);
1386                return ce;
1387        }
1388
1389        if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1390                if (ignore_missing)
1391                        return ce;
1392                if (err)
1393                        *err = ENOENT;
1394                return NULL;
1395        }
1396
1397        if (lstat(ce->name, &st) < 0) {
1398                if (ignore_missing && errno == ENOENT)
1399                        return ce;
1400                if (err)
1401                        *err = errno;
1402                return NULL;
1403        }
1404
1405        changed = ie_match_stat(istate, ce, &st, options);
1406        if (changed_ret)
1407                *changed_ret = changed;
1408        if (!changed) {
1409                /*
1410                 * The path is unchanged.  If we were told to ignore
1411                 * valid bit, then we did the actual stat check and
1412                 * found that the entry is unmodified.  If the entry
1413                 * is not marked VALID, this is the place to mark it
1414                 * valid again, under "assume unchanged" mode.
1415                 */
1416                if (ignore_valid && assume_unchanged &&
1417                    !(ce->ce_flags & CE_VALID))
1418                        ; /* mark this one VALID again */
1419                else {
1420                        /*
1421                         * We do not mark the index itself "modified"
1422                         * because CE_UPTODATE flag is in-core only;
1423                         * we are not going to write this change out.
1424                         */
1425                        if (!S_ISGITLINK(ce->ce_mode)) {
1426                                ce_mark_uptodate(ce);
1427                                mark_fsmonitor_valid(ce);
1428                        }
1429                        return ce;
1430                }
1431        }
1432
1433        if (ie_modified(istate, ce, &st, options)) {
1434                if (err)
1435                        *err = EINVAL;
1436                return NULL;
1437        }
1438
1439        updated = make_empty_cache_entry(istate, ce_namelen(ce));
1440        copy_cache_entry(updated, ce);
1441        memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1442        fill_stat_cache_info(updated, &st);
1443        /*
1444         * If ignore_valid is not set, we should leave CE_VALID bit
1445         * alone.  Otherwise, paths marked with --no-assume-unchanged
1446         * (i.e. things to be edited) will reacquire CE_VALID bit
1447         * automatically, which is not really what we want.
1448         */
1449        if (!ignore_valid && assume_unchanged &&
1450            !(ce->ce_flags & CE_VALID))
1451                updated->ce_flags &= ~CE_VALID;
1452
1453        /* istate->cache_changed is updated in the caller */
1454        return updated;
1455}
1456
1457static void show_file(const char * fmt, const char * name, int in_porcelain,
1458                      int * first, const char *header_msg)
1459{
1460        if (in_porcelain && *first && header_msg) {
1461                printf("%s\n", header_msg);
1462                *first = 0;
1463        }
1464        printf(fmt, name);
1465}
1466
1467int refresh_index(struct index_state *istate, unsigned int flags,
1468                  const struct pathspec *pathspec,
1469                  char *seen, const char *header_msg)
1470{
1471        int i;
1472        int has_errors = 0;
1473        int really = (flags & REFRESH_REALLY) != 0;
1474        int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1475        int quiet = (flags & REFRESH_QUIET) != 0;
1476        int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1477        int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1478        int first = 1;
1479        int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1480        unsigned int options = (CE_MATCH_REFRESH |
1481                                (really ? CE_MATCH_IGNORE_VALID : 0) |
1482                                (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1483        const char *modified_fmt;
1484        const char *deleted_fmt;
1485        const char *typechange_fmt;
1486        const char *added_fmt;
1487        const char *unmerged_fmt;
1488        struct progress *progress = NULL;
1489
1490        if (flags & REFRESH_PROGRESS && isatty(2))
1491                progress = start_delayed_progress(_("Refresh index"),
1492                                                  istate->cache_nr);
1493
1494        trace_performance_enter();
1495        modified_fmt   = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1496        deleted_fmt    = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1497        typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1498        added_fmt      = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1499        unmerged_fmt   = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
1500        for (i = 0; i < istate->cache_nr; i++) {
1501                struct cache_entry *ce, *new_entry;
1502                int cache_errno = 0;
1503                int changed = 0;
1504                int filtered = 0;
1505
1506                ce = istate->cache[i];
1507                if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1508                        continue;
1509
1510                if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1511                        filtered = 1;
1512
1513                if (ce_stage(ce)) {
1514                        while ((i < istate->cache_nr) &&
1515                               ! strcmp(istate->cache[i]->name, ce->name))
1516                                i++;
1517                        i--;
1518                        if (allow_unmerged)
1519                                continue;
1520                        if (!filtered)
1521                                show_file(unmerged_fmt, ce->name, in_porcelain,
1522                                          &first, header_msg);
1523                        has_errors = 1;
1524                        continue;
1525                }
1526
1527                if (filtered)
1528                        continue;
1529
1530                new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1531                if (new_entry == ce)
1532                        continue;
1533                if (progress)
1534                        display_progress(progress, i);
1535                if (!new_entry) {
1536                        const char *fmt;
1537
1538                        if (really && cache_errno == EINVAL) {
1539                                /* If we are doing --really-refresh that
1540                                 * means the index is not valid anymore.
1541                                 */
1542                                ce->ce_flags &= ~CE_VALID;
1543                                ce->ce_flags |= CE_UPDATE_IN_BASE;
1544                                mark_fsmonitor_invalid(istate, ce);
1545                                istate->cache_changed |= CE_ENTRY_CHANGED;
1546                        }
1547                        if (quiet)
1548                                continue;
1549
1550                        if (cache_errno == ENOENT)
1551                                fmt = deleted_fmt;
1552                        else if (ce_intent_to_add(ce))
1553                                fmt = added_fmt; /* must be before other checks */
1554                        else if (changed & TYPE_CHANGED)
1555                                fmt = typechange_fmt;
1556                        else
1557                                fmt = modified_fmt;
1558                        show_file(fmt,
1559                                  ce->name, in_porcelain, &first, header_msg);
1560                        has_errors = 1;
1561                        continue;
1562                }
1563
1564                replace_index_entry(istate, i, new_entry);
1565        }
1566        if (progress) {
1567                display_progress(progress, istate->cache_nr);
1568                stop_progress(&progress);
1569        }
1570        trace_performance_leave("refresh index");
1571        return has_errors;
1572}
1573
1574struct cache_entry *refresh_cache_entry(struct index_state *istate,
1575                                        struct cache_entry *ce,
1576                                        unsigned int options)
1577{
1578        return refresh_cache_ent(istate, ce, options, NULL, NULL);
1579}
1580
1581
1582/*****************************************************************
1583 * Index File I/O
1584 *****************************************************************/
1585
1586#define INDEX_FORMAT_DEFAULT 3
1587
1588static unsigned int get_index_format_default(void)
1589{
1590        char *envversion = getenv("GIT_INDEX_VERSION");
1591        char *endp;
1592        int value;
1593        unsigned int version = INDEX_FORMAT_DEFAULT;
1594
1595        if (!envversion) {
1596                if (!git_config_get_int("index.version", &value))
1597                        version = value;
1598                if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1599                        warning(_("index.version set, but the value is invalid.\n"
1600                                  "Using version %i"), INDEX_FORMAT_DEFAULT);
1601                        return INDEX_FORMAT_DEFAULT;
1602                }
1603                return version;
1604        }
1605
1606        version = strtoul(envversion, &endp, 10);
1607        if (*endp ||
1608            version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1609                warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1610                          "Using version %i"), INDEX_FORMAT_DEFAULT);
1611                version = INDEX_FORMAT_DEFAULT;
1612        }
1613        return version;
1614}
1615
1616/*
1617 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1618 * Again - this is just a (very strong in practice) heuristic that
1619 * the inode hasn't changed.
1620 *
1621 * We save the fields in big-endian order to allow using the
1622 * index file over NFS transparently.
1623 */
1624struct ondisk_cache_entry {
1625        struct cache_time ctime;
1626        struct cache_time mtime;
1627        uint32_t dev;
1628        uint32_t ino;
1629        uint32_t mode;
1630        uint32_t uid;
1631        uint32_t gid;
1632        uint32_t size;
1633        unsigned char sha1[20];
1634        uint16_t flags;
1635        char name[FLEX_ARRAY]; /* more */
1636};
1637
1638/*
1639 * This struct is used when CE_EXTENDED bit is 1
1640 * The struct must match ondisk_cache_entry exactly from
1641 * ctime till flags
1642 */
1643struct ondisk_cache_entry_extended {
1644        struct cache_time ctime;
1645        struct cache_time mtime;
1646        uint32_t dev;
1647        uint32_t ino;
1648        uint32_t mode;
1649        uint32_t uid;
1650        uint32_t gid;
1651        uint32_t size;
1652        unsigned char sha1[20];
1653        uint16_t flags;
1654        uint16_t flags2;
1655        char name[FLEX_ARRAY]; /* more */
1656};
1657
1658/* These are only used for v3 or lower */
1659#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1660#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1661#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1662#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1663#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1664                            ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1665                            ondisk_cache_entry_size(ce_namelen(ce)))
1666
1667/* Allow fsck to force verification of the index checksum. */
1668int verify_index_checksum;
1669
1670/* Allow fsck to force verification of the cache entry order. */
1671int verify_ce_order;
1672
1673static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1674{
1675        git_hash_ctx c;
1676        unsigned char hash[GIT_MAX_RAWSZ];
1677        int hdr_version;
1678
1679        if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1680                return error(_("bad signature 0x%08x"), hdr->hdr_signature);
1681        hdr_version = ntohl(hdr->hdr_version);
1682        if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1683                return error(_("bad index version %d"), hdr_version);
1684
1685        if (!verify_index_checksum)
1686                return 0;
1687
1688        the_hash_algo->init_fn(&c);
1689        the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1690        the_hash_algo->final_fn(hash, &c);
1691        if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1692                return error(_("bad index file sha1 signature"));
1693        return 0;
1694}
1695
1696static int read_index_extension(struct index_state *istate,
1697                                const char *ext, const char *data, unsigned long sz)
1698{
1699        switch (CACHE_EXT(ext)) {
1700        case CACHE_EXT_TREE:
1701                istate->cache_tree = cache_tree_read(data, sz);
1702                break;
1703        case CACHE_EXT_RESOLVE_UNDO:
1704                istate->resolve_undo = resolve_undo_read(data, sz);
1705                break;
1706        case CACHE_EXT_LINK:
1707                if (read_link_extension(istate, data, sz))
1708                        return -1;
1709                break;
1710        case CACHE_EXT_UNTRACKED:
1711                istate->untracked = read_untracked_extension(data, sz);
1712                break;
1713        case CACHE_EXT_FSMONITOR:
1714                read_fsmonitor_extension(istate, data, sz);
1715                break;
1716        case CACHE_EXT_ENDOFINDEXENTRIES:
1717        case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1718                /* already handled in do_read_index() */
1719                break;
1720        default:
1721                if (*ext < 'A' || 'Z' < *ext)
1722                        return error(_("index uses %.4s extension, which we do not understand"),
1723                                     ext);
1724                fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
1725                break;
1726        }
1727        return 0;
1728}
1729
1730int hold_locked_index(struct lock_file *lk, int lock_flags)
1731{
1732        return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1733}
1734
1735int read_index(struct index_state *istate)
1736{
1737        return read_index_from(istate, get_index_file(), get_git_dir());
1738}
1739
1740static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,
1741                                            unsigned int version,
1742                                            struct ondisk_cache_entry *ondisk,
1743                                            unsigned long *ent_size,
1744                                            const struct cache_entry *previous_ce)
1745{
1746        struct cache_entry *ce;
1747        size_t len;
1748        const char *name;
1749        unsigned int flags;
1750        size_t copy_len;
1751        /*
1752         * Adjacent cache entries tend to share the leading paths, so it makes
1753         * sense to only store the differences in later entries.  In the v4
1754         * on-disk format of the index, each on-disk cache entry stores the
1755         * number of bytes to be stripped from the end of the previous name,
1756         * and the bytes to append to the result, to come up with its name.
1757         */
1758        int expand_name_field = version == 4;
1759
1760        /* On-disk flags are just 16 bits */
1761        flags = get_be16(&ondisk->flags);
1762        len = flags & CE_NAMEMASK;
1763
1764        if (flags & CE_EXTENDED) {
1765                struct ondisk_cache_entry_extended *ondisk2;
1766                int extended_flags;
1767                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1768                extended_flags = get_be16(&ondisk2->flags2) << 16;
1769                /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1770                if (extended_flags & ~CE_EXTENDED_FLAGS)
1771                        die(_("unknown index entry format 0x%08x"), extended_flags);
1772                flags |= extended_flags;
1773                name = ondisk2->name;
1774        }
1775        else
1776                name = ondisk->name;
1777
1778        if (expand_name_field) {
1779                const unsigned char *cp = (const unsigned char *)name;
1780                size_t strip_len, previous_len;
1781
1782                /* If we're at the begining of a block, ignore the previous name */
1783                strip_len = decode_varint(&cp);
1784                if (previous_ce) {
1785                        previous_len = previous_ce->ce_namelen;
1786                        if (previous_len < strip_len)
1787                                die(_("malformed name field in the index, near path '%s'"),
1788                                        previous_ce->name);
1789                        copy_len = previous_len - strip_len;
1790                } else {
1791                        copy_len = 0;
1792                }
1793                name = (const char *)cp;
1794        }
1795
1796        if (len == CE_NAMEMASK) {
1797                len = strlen(name);
1798                if (expand_name_field)
1799                        len += copy_len;
1800        }
1801
1802        ce = mem_pool__ce_alloc(ce_mem_pool, len);
1803
1804        ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1805        ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1806        ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1807        ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1808        ce->ce_stat_data.sd_dev   = get_be32(&ondisk->dev);
1809        ce->ce_stat_data.sd_ino   = get_be32(&ondisk->ino);
1810        ce->ce_mode  = get_be32(&ondisk->mode);
1811        ce->ce_stat_data.sd_uid   = get_be32(&ondisk->uid);
1812        ce->ce_stat_data.sd_gid   = get_be32(&ondisk->gid);
1813        ce->ce_stat_data.sd_size  = get_be32(&ondisk->size);
1814        ce->ce_flags = flags & ~CE_NAMEMASK;
1815        ce->ce_namelen = len;
1816        ce->index = 0;
1817        hashcpy(ce->oid.hash, ondisk->sha1);
1818
1819        if (expand_name_field) {
1820                if (copy_len)
1821                        memcpy(ce->name, previous_ce->name, copy_len);
1822                memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1823                *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1824        } else {
1825                memcpy(ce->name, name, len + 1);
1826                *ent_size = ondisk_ce_size(ce);
1827        }
1828        return ce;
1829}
1830
1831static void check_ce_order(struct index_state *istate)
1832{
1833        unsigned int i;
1834
1835        if (!verify_ce_order)
1836                return;
1837
1838        for (i = 1; i < istate->cache_nr; i++) {
1839                struct cache_entry *ce = istate->cache[i - 1];
1840                struct cache_entry *next_ce = istate->cache[i];
1841                int name_compare = strcmp(ce->name, next_ce->name);
1842
1843                if (0 < name_compare)
1844                        die(_("unordered stage entries in index"));
1845                if (!name_compare) {
1846                        if (!ce_stage(ce))
1847                                die(_("multiple stage entries for merged file '%s'"),
1848                                    ce->name);
1849                        if (ce_stage(ce) > ce_stage(next_ce))
1850                                die(_("unordered stage entries for '%s'"),
1851                                    ce->name);
1852                }
1853        }
1854}
1855
1856static void tweak_untracked_cache(struct index_state *istate)
1857{
1858        switch (git_config_get_untracked_cache()) {
1859        case -1: /* keep: do nothing */
1860                break;
1861        case 0: /* false */
1862                remove_untracked_cache(istate);
1863                break;
1864        case 1: /* true */
1865                add_untracked_cache(istate);
1866                break;
1867        default: /* unknown value: do nothing */
1868                break;
1869        }
1870}
1871
1872static void tweak_split_index(struct index_state *istate)
1873{
1874        switch (git_config_get_split_index()) {
1875        case -1: /* unset: do nothing */
1876                break;
1877        case 0: /* false */
1878                remove_split_index(istate);
1879                break;
1880        case 1: /* true */
1881                add_split_index(istate);
1882                break;
1883        default: /* unknown value: do nothing */
1884                break;
1885        }
1886}
1887
1888static void post_read_index_from(struct index_state *istate)
1889{
1890        check_ce_order(istate);
1891        tweak_untracked_cache(istate);
1892        tweak_split_index(istate);
1893        tweak_fsmonitor(istate);
1894}
1895
1896static size_t estimate_cache_size_from_compressed(unsigned int entries)
1897{
1898        return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1899}
1900
1901static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1902{
1903        long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1904
1905        /*
1906         * Account for potential alignment differences.
1907         */
1908        per_entry += align_padding_size(sizeof(struct cache_entry), -sizeof(struct ondisk_cache_entry));
1909        return ondisk_size + entries * per_entry;
1910}
1911
1912struct index_entry_offset
1913{
1914        /* starting byte offset into index file, count of index entries in this block */
1915        int offset, nr;
1916};
1917
1918struct index_entry_offset_table
1919{
1920        int nr;
1921        struct index_entry_offset entries[FLEX_ARRAY];
1922};
1923
1924#ifndef NO_PTHREADS
1925static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1926static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
1927#endif
1928
1929static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1930static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1931
1932struct load_index_extensions
1933{
1934#ifndef NO_PTHREADS
1935        pthread_t pthread;
1936#endif
1937        struct index_state *istate;
1938        const char *mmap;
1939        size_t mmap_size;
1940        unsigned long src_offset;
1941};
1942
1943static void *load_index_extensions(void *_data)
1944{
1945        struct load_index_extensions *p = _data;
1946        unsigned long src_offset = p->src_offset;
1947
1948        while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1949                /* After an array of active_nr index entries,
1950                 * there can be arbitrary number of extended
1951                 * sections, each of which is prefixed with
1952                 * extension name (4-byte) and section length
1953                 * in 4-byte network byte order.
1954                 */
1955                uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1956                if (read_index_extension(p->istate,
1957                                         p->mmap + src_offset,
1958                                         p->mmap + src_offset + 8,
1959                                         extsize) < 0) {
1960                        munmap((void *)p->mmap, p->mmap_size);
1961                        die(_("index file corrupt"));
1962                }
1963                src_offset += 8;
1964                src_offset += extsize;
1965        }
1966
1967        return NULL;
1968}
1969
1970/*
1971 * A helper function that will load the specified range of cache entries
1972 * from the memory mapped file and add them to the given index.
1973 */
1974static unsigned long load_cache_entry_block(struct index_state *istate,
1975                        struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,
1976                        unsigned long start_offset, const struct cache_entry *previous_ce)
1977{
1978        int i;
1979        unsigned long src_offset = start_offset;
1980
1981        for (i = offset; i < offset + nr; i++) {
1982                struct ondisk_cache_entry *disk_ce;
1983                struct cache_entry *ce;
1984                unsigned long consumed;
1985
1986                disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
1987                ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce);
1988                set_index_entry(istate, i, ce);
1989
1990                src_offset += consumed;
1991                previous_ce = ce;
1992        }
1993        return src_offset - start_offset;
1994}
1995
1996static unsigned long load_all_cache_entries(struct index_state *istate,
1997                        const char *mmap, size_t mmap_size, unsigned long src_offset)
1998{
1999        unsigned long consumed;
2000
2001        if (istate->version == 4) {
2002                mem_pool_init(&istate->ce_mem_pool,
2003                                estimate_cache_size_from_compressed(istate->cache_nr));
2004        } else {
2005                mem_pool_init(&istate->ce_mem_pool,
2006                                estimate_cache_size(mmap_size, istate->cache_nr));
2007        }
2008
2009        consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
2010                                        0, istate->cache_nr, mmap, src_offset, NULL);
2011        return consumed;
2012}
2013
2014#ifndef NO_PTHREADS
2015
2016/*
2017 * Mostly randomly chosen maximum thread counts: we
2018 * cap the parallelism to online_cpus() threads, and we want
2019 * to have at least 10000 cache entries per thread for it to
2020 * be worth starting a thread.
2021 */
2022
2023#define THREAD_COST             (10000)
2024
2025struct load_cache_entries_thread_data
2026{
2027        pthread_t pthread;
2028        struct index_state *istate;
2029        struct mem_pool *ce_mem_pool;
2030        int offset;
2031        const char *mmap;
2032        struct index_entry_offset_table *ieot;
2033        int ieot_start;         /* starting index into the ieot array */
2034        int ieot_blocks;        /* count of ieot entries to process */
2035        unsigned long consumed; /* return # of bytes in index file processed */
2036};
2037
2038/*
2039 * A thread proc to run the load_cache_entries() computation
2040 * across multiple background threads.
2041 */
2042static void *load_cache_entries_thread(void *_data)
2043{
2044        struct load_cache_entries_thread_data *p = _data;
2045        int i;
2046
2047        /* iterate across all ieot blocks assigned to this thread */
2048        for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2049                p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2050                        p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2051                p->offset += p->ieot->entries[i].nr;
2052        }
2053        return NULL;
2054}
2055
2056static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,
2057                        unsigned long src_offset, int nr_threads, struct index_entry_offset_table *ieot)
2058{
2059        int i, offset, ieot_blocks, ieot_start, err;
2060        struct load_cache_entries_thread_data *data;
2061        unsigned long consumed = 0;
2062
2063        /* a little sanity checking */
2064        if (istate->name_hash_initialized)
2065                BUG("the name hash isn't thread safe");
2066
2067        mem_pool_init(&istate->ce_mem_pool, 0);
2068
2069        /* ensure we have no more threads than we have blocks to process */
2070        if (nr_threads > ieot->nr)
2071                nr_threads = ieot->nr;
2072        data = xcalloc(nr_threads, sizeof(*data));
2073
2074        offset = ieot_start = 0;
2075        ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2076        for (i = 0; i < nr_threads; i++) {
2077                struct load_cache_entries_thread_data *p = &data[i];
2078                int nr, j;
2079
2080                if (ieot_start + ieot_blocks > ieot->nr)
2081                        ieot_blocks = ieot->nr - ieot_start;
2082
2083                p->istate = istate;
2084                p->offset = offset;
2085                p->mmap = mmap;
2086                p->ieot = ieot;
2087                p->ieot_start = ieot_start;
2088                p->ieot_blocks = ieot_blocks;
2089
2090                /* create a mem_pool for each thread */
2091                nr = 0;
2092                for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2093                        nr += p->ieot->entries[j].nr;
2094                if (istate->version == 4) {
2095                        mem_pool_init(&p->ce_mem_pool,
2096                                estimate_cache_size_from_compressed(nr));
2097                } else {
2098                        mem_pool_init(&p->ce_mem_pool,
2099                                estimate_cache_size(mmap_size, nr));
2100                }
2101
2102                err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2103                if (err)
2104                        die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2105
2106                /* increment by the number of cache entries in the ieot block being processed */
2107                for (j = 0; j < ieot_blocks; j++)
2108                        offset += ieot->entries[ieot_start + j].nr;
2109                ieot_start += ieot_blocks;
2110        }
2111
2112        for (i = 0; i < nr_threads; i++) {
2113                struct load_cache_entries_thread_data *p = &data[i];
2114
2115                err = pthread_join(p->pthread, NULL);
2116                if (err)
2117                        die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2118                mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2119                consumed += p->consumed;
2120        }
2121
2122        free(data);
2123
2124        return consumed;
2125}
2126#endif
2127
2128/* remember to discard_cache() before reading a different cache! */
2129int do_read_index(struct index_state *istate, const char *path, int must_exist)
2130{
2131        int fd;
2132        struct stat st;
2133        unsigned long src_offset;
2134        const struct cache_header *hdr;
2135        const char *mmap;
2136        size_t mmap_size;
2137        struct load_index_extensions p;
2138        size_t extension_offset = 0;
2139#ifndef NO_PTHREADS
2140        int nr_threads, cpus;
2141        struct index_entry_offset_table *ieot = NULL;
2142#endif
2143
2144        if (istate->initialized)
2145                return istate->cache_nr;
2146
2147        istate->timestamp.sec = 0;
2148        istate->timestamp.nsec = 0;
2149        fd = open(path, O_RDONLY);
2150        if (fd < 0) {
2151                if (!must_exist && errno == ENOENT)
2152                        return 0;
2153                die_errno(_("%s: index file open failed"), path);
2154        }
2155
2156        if (fstat(fd, &st))
2157                die_errno(_("%s: cannot stat the open index"), path);
2158
2159        mmap_size = xsize_t(st.st_size);
2160        if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2161                die(_("%s: index file smaller than expected"), path);
2162
2163        mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2164        if (mmap == MAP_FAILED)
2165                die_errno(_("%s: unable to map index file"), path);
2166        close(fd);
2167
2168        hdr = (const struct cache_header *)mmap;
2169        if (verify_hdr(hdr, mmap_size) < 0)
2170                goto unmap;
2171
2172        hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
2173        istate->version = ntohl(hdr->hdr_version);
2174        istate->cache_nr = ntohl(hdr->hdr_entries);
2175        istate->cache_alloc = alloc_nr(istate->cache_nr);
2176        istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
2177        istate->initialized = 1;
2178
2179        p.istate = istate;
2180        p.mmap = mmap;
2181        p.mmap_size = mmap_size;
2182
2183        src_offset = sizeof(*hdr);
2184
2185#ifndef NO_PTHREADS
2186        nr_threads = git_config_get_index_threads();
2187
2188        /* TODO: does creating more threads than cores help? */
2189        if (!nr_threads) {
2190                nr_threads = istate->cache_nr / THREAD_COST;
2191                cpus = online_cpus();
2192                if (nr_threads > cpus)
2193                        nr_threads = cpus;
2194        }
2195
2196        if (nr_threads > 1) {
2197                extension_offset = read_eoie_extension(mmap, mmap_size);
2198                if (extension_offset) {
2199                        int err;
2200
2201                        p.src_offset = extension_offset;
2202                        err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2203                        if (err)
2204                                die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2205
2206                        nr_threads--;
2207                }
2208        }
2209
2210        /*
2211         * Locate and read the index entry offset table so that we can use it
2212         * to multi-thread the reading of the cache entries.
2213         */
2214        if (extension_offset && nr_threads > 1)
2215                ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2216
2217        if (ieot) {
2218                src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, src_offset, nr_threads, ieot);
2219                free(ieot);
2220        } else {
2221                src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2222        }
2223#else
2224        src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2225#endif
2226
2227        istate->timestamp.sec = st.st_mtime;
2228        istate->timestamp.nsec = ST_MTIME_NSEC(st);
2229
2230        /* if we created a thread, join it otherwise load the extensions on the primary thread */
2231#ifndef NO_PTHREADS
2232        if (extension_offset) {
2233                int ret = pthread_join(p.pthread, NULL);
2234                if (ret)
2235                        die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2236        }
2237#endif
2238        if (!extension_offset) {
2239                p.src_offset = src_offset;
2240                load_index_extensions(&p);
2241        }
2242        munmap((void *)mmap, mmap_size);
2243        return istate->cache_nr;
2244
2245unmap:
2246        munmap((void *)mmap, mmap_size);
2247        die(_("index file corrupt"));
2248}
2249
2250/*
2251 * Signal that the shared index is used by updating its mtime.
2252 *
2253 * This way, shared index can be removed if they have not been used
2254 * for some time.
2255 */
2256static void freshen_shared_index(const char *shared_index, int warn)
2257{
2258        if (!check_and_freshen_file(shared_index, 1) && warn)
2259                warning(_("could not freshen shared index '%s'"), shared_index);
2260}
2261
2262int read_index_from(struct index_state *istate, const char *path,
2263                    const char *gitdir)
2264{
2265        struct split_index *split_index;
2266        int ret;
2267        char *base_oid_hex;
2268        char *base_path;
2269
2270        /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2271        if (istate->initialized)
2272                return istate->cache_nr;
2273
2274        trace_performance_enter();
2275        ret = do_read_index(istate, path, 0);
2276        trace_performance_leave("read cache %s", path);
2277
2278        split_index = istate->split_index;
2279        if (!split_index || is_null_oid(&split_index->base_oid)) {
2280                post_read_index_from(istate);
2281                return ret;
2282        }
2283
2284        trace_performance_enter();
2285        if (split_index->base)
2286                discard_index(split_index->base);
2287        else
2288                split_index->base = xcalloc(1, sizeof(*split_index->base));
2289
2290        base_oid_hex = oid_to_hex(&split_index->base_oid);
2291        base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2292        ret = do_read_index(split_index->base, base_path, 1);
2293        if (!oideq(&split_index->base_oid, &split_index->base->oid))
2294                die(_("broken index, expect %s in %s, got %s"),
2295                    base_oid_hex, base_path,
2296                    oid_to_hex(&split_index->base->oid));
2297
2298        freshen_shared_index(base_path, 0);
2299        merge_base_index(istate);
2300        post_read_index_from(istate);
2301        free(base_path);
2302        trace_performance_leave("read cache %s", base_path);
2303        return ret;
2304}
2305
2306int is_index_unborn(struct index_state *istate)
2307{
2308        return (!istate->cache_nr && !istate->timestamp.sec);
2309}
2310
2311int discard_index(struct index_state *istate)
2312{
2313        /*
2314         * Cache entries in istate->cache[] should have been allocated
2315         * from the memory pool associated with this index, or from an
2316         * associated split_index. There is no need to free individual
2317         * cache entries. validate_cache_entries can detect when this
2318         * assertion does not hold.
2319         */
2320        validate_cache_entries(istate);
2321
2322        resolve_undo_clear_index(istate);
2323        istate->cache_nr = 0;
2324        istate->cache_changed = 0;
2325        istate->timestamp.sec = 0;
2326        istate->timestamp.nsec = 0;
2327        free_name_hash(istate);
2328        cache_tree_free(&(istate->cache_tree));
2329        istate->initialized = 0;
2330        FREE_AND_NULL(istate->cache);
2331        istate->cache_alloc = 0;
2332        discard_split_index(istate);
2333        free_untracked_cache(istate->untracked);
2334        istate->untracked = NULL;
2335
2336        if (istate->ce_mem_pool) {
2337                mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2338                istate->ce_mem_pool = NULL;
2339        }
2340
2341        return 0;
2342}
2343
2344/*
2345 * Validate the cache entries of this index.
2346 * All cache entries associated with this index
2347 * should have been allocated by the memory pool
2348 * associated with this index, or by a referenced
2349 * split index.
2350 */
2351void validate_cache_entries(const struct index_state *istate)
2352{
2353        int i;
2354
2355        if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2356                return;
2357
2358        for (i = 0; i < istate->cache_nr; i++) {
2359                if (!istate) {
2360                        BUG("cache entry is not allocated from expected memory pool");
2361                } else if (!istate->ce_mem_pool ||
2362                        !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2363                        if (!istate->split_index ||
2364                                !istate->split_index->base ||
2365                                !istate->split_index->base->ce_mem_pool ||
2366                                !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2367                                BUG("cache entry is not allocated from expected memory pool");
2368                        }
2369                }
2370        }
2371
2372        if (istate->split_index)
2373                validate_cache_entries(istate->split_index->base);
2374}
2375
2376int unmerged_index(const struct index_state *istate)
2377{
2378        int i;
2379        for (i = 0; i < istate->cache_nr; i++) {
2380                if (ce_stage(istate->cache[i]))
2381                        return 1;
2382        }
2383        return 0;
2384}
2385
2386int index_has_changes(struct index_state *istate,
2387                      struct tree *tree,
2388                      struct strbuf *sb)
2389{
2390        struct object_id cmp;
2391        int i;
2392
2393        if (istate != &the_index) {
2394                BUG("index_has_changes cannot yet accept istate != &the_index; do_diff_cache needs updating first.");
2395        }
2396        if (tree)
2397                cmp = tree->object.oid;
2398        if (tree || !get_oid_tree("HEAD", &cmp)) {
2399                struct diff_options opt;
2400
2401                repo_diff_setup(the_repository, &opt);
2402                opt.flags.exit_with_status = 1;
2403                if (!sb)
2404                        opt.flags.quick = 1;
2405                do_diff_cache(&cmp, &opt);
2406                diffcore_std(&opt);
2407                for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2408                        if (i)
2409                                strbuf_addch(sb, ' ');
2410                        strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2411                }
2412                diff_flush(&opt);
2413                return opt.flags.has_changes != 0;
2414        } else {
2415                for (i = 0; sb && i < istate->cache_nr; i++) {
2416                        if (i)
2417                                strbuf_addch(sb, ' ');
2418                        strbuf_addstr(sb, istate->cache[i]->name);
2419                }
2420                return !!istate->cache_nr;
2421        }
2422}
2423
2424#define WRITE_BUFFER_SIZE 8192
2425static unsigned char write_buffer[WRITE_BUFFER_SIZE];
2426static unsigned long write_buffer_len;
2427
2428static int ce_write_flush(git_hash_ctx *context, int fd)
2429{
2430        unsigned int buffered = write_buffer_len;
2431        if (buffered) {
2432                the_hash_algo->update_fn(context, write_buffer, buffered);
2433                if (write_in_full(fd, write_buffer, buffered) < 0)
2434                        return -1;
2435                write_buffer_len = 0;
2436        }
2437        return 0;
2438}
2439
2440static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
2441{
2442        while (len) {
2443                unsigned int buffered = write_buffer_len;
2444                unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2445                if (partial > len)
2446                        partial = len;
2447                memcpy(write_buffer + buffered, data, partial);
2448                buffered += partial;
2449                if (buffered == WRITE_BUFFER_SIZE) {
2450                        write_buffer_len = buffered;
2451                        if (ce_write_flush(context, fd))
2452                                return -1;
2453                        buffered = 0;
2454                }
2455                write_buffer_len = buffered;
2456                len -= partial;
2457                data = (char *) data + partial;
2458        }
2459        return 0;
2460}
2461
2462static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2463                                  int fd, unsigned int ext, unsigned int sz)
2464{
2465        ext = htonl(ext);
2466        sz = htonl(sz);
2467        if (eoie_context) {
2468                the_hash_algo->update_fn(eoie_context, &ext, 4);
2469                the_hash_algo->update_fn(eoie_context, &sz, 4);
2470        }
2471        return ((ce_write(context, fd, &ext, 4) < 0) ||
2472                (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2473}
2474
2475static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2476{
2477        unsigned int left = write_buffer_len;
2478
2479        if (left) {
2480                write_buffer_len = 0;
2481                the_hash_algo->update_fn(context, write_buffer, left);
2482        }
2483
2484        /* Flush first if not enough space for hash signature */
2485        if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2486                if (write_in_full(fd, write_buffer, left) < 0)
2487                        return -1;
2488                left = 0;
2489        }
2490
2491        /* Append the hash signature at the end */
2492        the_hash_algo->final_fn(write_buffer + left, context);
2493        hashcpy(hash, write_buffer + left);
2494        left += the_hash_algo->rawsz;
2495        return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2496}
2497
2498static void ce_smudge_racily_clean_entry(struct index_state *istate,
2499                                         struct cache_entry *ce)
2500{
2501        /*
2502         * The only thing we care about in this function is to smudge the
2503         * falsely clean entry due to touch-update-touch race, so we leave
2504         * everything else as they are.  We are called for entries whose
2505         * ce_stat_data.sd_mtime match the index file mtime.
2506         *
2507         * Note that this actually does not do much for gitlinks, for
2508         * which ce_match_stat_basic() always goes to the actual
2509         * contents.  The caller checks with is_racy_timestamp() which
2510         * always says "no" for gitlinks, so we are not called for them ;-)
2511         */
2512        struct stat st;
2513
2514        if (lstat(ce->name, &st) < 0)
2515                return;
2516        if (ce_match_stat_basic(ce, &st))
2517                return;
2518        if (ce_modified_check_fs(istate, ce, &st)) {
2519                /* This is "racily clean"; smudge it.  Note that this
2520                 * is a tricky code.  At first glance, it may appear
2521                 * that it can break with this sequence:
2522                 *
2523                 * $ echo xyzzy >frotz
2524                 * $ git-update-index --add frotz
2525                 * $ : >frotz
2526                 * $ sleep 3
2527                 * $ echo filfre >nitfol
2528                 * $ git-update-index --add nitfol
2529                 *
2530                 * but it does not.  When the second update-index runs,
2531                 * it notices that the entry "frotz" has the same timestamp
2532                 * as index, and if we were to smudge it by resetting its
2533                 * size to zero here, then the object name recorded
2534                 * in index is the 6-byte file but the cached stat information
2535                 * becomes zero --- which would then match what we would
2536                 * obtain from the filesystem next time we stat("frotz").
2537                 *
2538                 * However, the second update-index, before calling
2539                 * this function, notices that the cached size is 6
2540                 * bytes and what is on the filesystem is an empty
2541                 * file, and never calls us, so the cached size information
2542                 * for "frotz" stays 6 which does not match the filesystem.
2543                 */
2544                ce->ce_stat_data.sd_size = 0;
2545        }
2546}
2547
2548/* Copy miscellaneous fields but not the name */
2549static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2550                                       struct cache_entry *ce)
2551{
2552        short flags;
2553
2554        ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2555        ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2556        ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2557        ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2558        ondisk->dev  = htonl(ce->ce_stat_data.sd_dev);
2559        ondisk->ino  = htonl(ce->ce_stat_data.sd_ino);
2560        ondisk->mode = htonl(ce->ce_mode);
2561        ondisk->uid  = htonl(ce->ce_stat_data.sd_uid);
2562        ondisk->gid  = htonl(ce->ce_stat_data.sd_gid);
2563        ondisk->size = htonl(ce->ce_stat_data.sd_size);
2564        hashcpy(ondisk->sha1, ce->oid.hash);
2565
2566        flags = ce->ce_flags & ~CE_NAMEMASK;
2567        flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2568        ondisk->flags = htons(flags);
2569        if (ce->ce_flags & CE_EXTENDED) {
2570                struct ondisk_cache_entry_extended *ondisk2;
2571                ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2572                ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2573        }
2574}
2575
2576static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2577                          struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2578{
2579        int size;
2580        int result;
2581        unsigned int saved_namelen;
2582        int stripped_name = 0;
2583        static unsigned char padding[8] = { 0x00 };
2584
2585        if (ce->ce_flags & CE_STRIP_NAME) {
2586                saved_namelen = ce_namelen(ce);
2587                ce->ce_namelen = 0;
2588                stripped_name = 1;
2589        }
2590
2591        if (ce->ce_flags & CE_EXTENDED)
2592                size = offsetof(struct ondisk_cache_entry_extended, name);
2593        else
2594                size = offsetof(struct ondisk_cache_entry, name);
2595
2596        if (!previous_name) {
2597                int len = ce_namelen(ce);
2598                copy_cache_entry_to_ondisk(ondisk, ce);
2599                result = ce_write(c, fd, ondisk, size);
2600                if (!result)
2601                        result = ce_write(c, fd, ce->name, len);
2602                if (!result)
2603                        result = ce_write(c, fd, padding, align_padding_size(size, len));
2604        } else {
2605                int common, to_remove, prefix_size;
2606                unsigned char to_remove_vi[16];
2607                for (common = 0;
2608                     (ce->name[common] &&
2609                      common < previous_name->len &&
2610                      ce->name[common] == previous_name->buf[common]);
2611                     common++)
2612                        ; /* still matching */
2613                to_remove = previous_name->len - common;
2614                prefix_size = encode_varint(to_remove, to_remove_vi);
2615
2616                copy_cache_entry_to_ondisk(ondisk, ce);
2617                result = ce_write(c, fd, ondisk, size);
2618                if (!result)
2619                        result = ce_write(c, fd, to_remove_vi, prefix_size);
2620                if (!result)
2621                        result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2622                if (!result)
2623                        result = ce_write(c, fd, padding, 1);
2624
2625                strbuf_splice(previous_name, common, to_remove,
2626                              ce->name + common, ce_namelen(ce) - common);
2627        }
2628        if (stripped_name) {
2629                ce->ce_namelen = saved_namelen;
2630                ce->ce_flags &= ~CE_STRIP_NAME;
2631        }
2632
2633        return result;
2634}
2635
2636/*
2637 * This function verifies if index_state has the correct sha1 of the
2638 * index file.  Don't die if we have any other failure, just return 0.
2639 */
2640static int verify_index_from(const struct index_state *istate, const char *path)
2641{
2642        int fd;
2643        ssize_t n;
2644        struct stat st;
2645        unsigned char hash[GIT_MAX_RAWSZ];
2646
2647        if (!istate->initialized)
2648                return 0;
2649
2650        fd = open(path, O_RDONLY);
2651        if (fd < 0)
2652                return 0;
2653
2654        if (fstat(fd, &st))
2655                goto out;
2656
2657        if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2658                goto out;
2659
2660        n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2661        if (n != the_hash_algo->rawsz)
2662                goto out;
2663
2664        if (!hasheq(istate->oid.hash, hash))
2665                goto out;
2666
2667        close(fd);
2668        return 1;
2669
2670out:
2671        close(fd);
2672        return 0;
2673}
2674
2675static int verify_index(const struct index_state *istate)
2676{
2677        return verify_index_from(istate, get_index_file());
2678}
2679
2680static int has_racy_timestamp(struct index_state *istate)
2681{
2682        int entries = istate->cache_nr;
2683        int i;
2684
2685        for (i = 0; i < entries; i++) {
2686                struct cache_entry *ce = istate->cache[i];
2687                if (is_racy_timestamp(istate, ce))
2688                        return 1;
2689        }
2690        return 0;
2691}
2692
2693void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2694{
2695        if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2696            verify_index(istate))
2697                write_locked_index(istate, lockfile, COMMIT_LOCK);
2698        else
2699                rollback_lock_file(lockfile);
2700}
2701
2702/*
2703 * On success, `tempfile` is closed. If it is the temporary file
2704 * of a `struct lock_file`, we will therefore effectively perform
2705 * a 'close_lock_file_gently()`. Since that is an implementation
2706 * detail of lockfiles, callers of `do_write_index()` should not
2707 * rely on it.
2708 */
2709static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2710                          int strip_extensions)
2711{
2712        uint64_t start = getnanotime();
2713        int newfd = tempfile->fd;
2714        git_hash_ctx c, eoie_c;
2715        struct cache_header hdr;
2716        int i, err = 0, removed, extended, hdr_version;
2717        struct cache_entry **cache = istate->cache;
2718        int entries = istate->cache_nr;
2719        struct stat st;
2720        struct ondisk_cache_entry_extended ondisk;
2721        struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2722        int drop_cache_tree = istate->drop_cache_tree;
2723        off_t offset;
2724        int ieot_entries = 1;
2725        struct index_entry_offset_table *ieot = NULL;
2726        int nr, nr_threads;
2727
2728        for (i = removed = extended = 0; i < entries; i++) {
2729                if (cache[i]->ce_flags & CE_REMOVE)
2730                        removed++;
2731
2732                /* reduce extended entries if possible */
2733                cache[i]->ce_flags &= ~CE_EXTENDED;
2734                if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2735                        extended++;
2736                        cache[i]->ce_flags |= CE_EXTENDED;
2737                }
2738        }
2739
2740        if (!istate->version) {
2741                istate->version = get_index_format_default();
2742                if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2743                        init_split_index(istate);
2744        }
2745
2746        /* demote version 3 to version 2 when the latter suffices */
2747        if (istate->version == 3 || istate->version == 2)
2748                istate->version = extended ? 3 : 2;
2749
2750        hdr_version = istate->version;
2751
2752        hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2753        hdr.hdr_version = htonl(hdr_version);
2754        hdr.hdr_entries = htonl(entries - removed);
2755
2756        the_hash_algo->init_fn(&c);
2757        if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2758                return -1;
2759
2760#ifndef NO_PTHREADS
2761        nr_threads = git_config_get_index_threads();
2762        if (nr_threads != 1) {
2763                int ieot_blocks, cpus;
2764
2765                /*
2766                 * ensure default number of ieot blocks maps evenly to the
2767                 * default number of threads that will process them leaving
2768                 * room for the thread to load the index extensions.
2769                 */
2770                if (!nr_threads) {
2771                        ieot_blocks = istate->cache_nr / THREAD_COST;
2772                        cpus = online_cpus();
2773                        if (ieot_blocks > cpus - 1)
2774                                ieot_blocks = cpus - 1;
2775                } else {
2776                        ieot_blocks = nr_threads;
2777                        if (ieot_blocks > istate->cache_nr)
2778                                ieot_blocks = istate->cache_nr;
2779                }
2780
2781                /*
2782                 * no reason to write out the IEOT extension if we don't
2783                 * have enough blocks to utilize multi-threading
2784                 */
2785                if (ieot_blocks > 1) {
2786                        ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2787                                + (ieot_blocks * sizeof(struct index_entry_offset)));
2788                        ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2789                }
2790        }
2791#endif
2792
2793        offset = lseek(newfd, 0, SEEK_CUR);
2794        if (offset < 0) {
2795                free(ieot);
2796                return -1;
2797        }
2798        offset += write_buffer_len;
2799        nr = 0;
2800        previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2801
2802        for (i = 0; i < entries; i++) {
2803                struct cache_entry *ce = cache[i];
2804                if (ce->ce_flags & CE_REMOVE)
2805                        continue;
2806                if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2807                        ce_smudge_racily_clean_entry(istate, ce);
2808                if (is_null_oid(&ce->oid)) {
2809                        static const char msg[] = "cache entry has null sha1: %s";
2810                        static int allow = -1;
2811
2812                        if (allow < 0)
2813                                allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2814                        if (allow)
2815                                warning(msg, ce->name);
2816                        else
2817                                err = error(msg, ce->name);
2818
2819                        drop_cache_tree = 1;
2820                }
2821                if (ieot && i && (i % ieot_entries == 0)) {
2822                        ieot->entries[ieot->nr].nr = nr;
2823                        ieot->entries[ieot->nr].offset = offset;
2824                        ieot->nr++;
2825                        /*
2826                         * If we have a V4 index, set the first byte to an invalid
2827                         * character to ensure there is nothing common with the previous
2828                         * entry
2829                         */
2830                        if (previous_name)
2831                                previous_name->buf[0] = 0;
2832                        nr = 0;
2833                        offset = lseek(newfd, 0, SEEK_CUR);
2834                        if (offset < 0) {
2835                                free(ieot);
2836                                return -1;
2837                        }
2838                        offset += write_buffer_len;
2839                }
2840                if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2841                        err = -1;
2842
2843                if (err)
2844                        break;
2845                nr++;
2846        }
2847        if (ieot && nr) {
2848                ieot->entries[ieot->nr].nr = nr;
2849                ieot->entries[ieot->nr].offset = offset;
2850                ieot->nr++;
2851        }
2852        strbuf_release(&previous_name_buf);
2853
2854        if (err) {
2855                free(ieot);
2856                return err;
2857        }
2858
2859        /* Write extension data here */
2860        offset = lseek(newfd, 0, SEEK_CUR);
2861        if (offset < 0) {
2862                free(ieot);
2863                return -1;
2864        }
2865        offset += write_buffer_len;
2866        the_hash_algo->init_fn(&eoie_c);
2867
2868        /*
2869         * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2870         * can minimize the number of extensions we have to scan through to
2871         * find it during load.  Write it out regardless of the
2872         * strip_extensions parameter as we need it when loading the shared
2873         * index.
2874         */
2875#ifndef NO_PTHREADS
2876        if (ieot) {
2877                struct strbuf sb = STRBUF_INIT;
2878
2879                write_ieot_extension(&sb, ieot);
2880                err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0
2881                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2882                strbuf_release(&sb);
2883                free(ieot);
2884                if (err)
2885                        return -1;
2886        }
2887#endif
2888
2889        if (!strip_extensions && istate->split_index) {
2890                struct strbuf sb = STRBUF_INIT;
2891
2892                err = write_link_extension(&sb, istate) < 0 ||
2893                        write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
2894                                               sb.len) < 0 ||
2895                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2896                strbuf_release(&sb);
2897                if (err)
2898                        return -1;
2899        }
2900        if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2901                struct strbuf sb = STRBUF_INIT;
2902
2903                cache_tree_write(&sb, istate->cache_tree);
2904                err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
2905                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2906                strbuf_release(&sb);
2907                if (err)
2908                        return -1;
2909        }
2910        if (!strip_extensions && istate->resolve_undo) {
2911                struct strbuf sb = STRBUF_INIT;
2912
2913                resolve_undo_write(&sb, istate->resolve_undo);
2914                err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
2915                                             sb.len) < 0
2916                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2917                strbuf_release(&sb);
2918                if (err)
2919                        return -1;
2920        }
2921        if (!strip_extensions && istate->untracked) {
2922                struct strbuf sb = STRBUF_INIT;
2923
2924                write_untracked_extension(&sb, istate->untracked);
2925                err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
2926                                             sb.len) < 0 ||
2927                        ce_write(&c, newfd, sb.buf, sb.len) < 0;
2928                strbuf_release(&sb);
2929                if (err)
2930                        return -1;
2931        }
2932        if (!strip_extensions && istate->fsmonitor_last_update) {
2933                struct strbuf sb = STRBUF_INIT;
2934
2935                write_fsmonitor_extension(&sb, istate);
2936                err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2937                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2938                strbuf_release(&sb);
2939                if (err)
2940                        return -1;
2941        }
2942
2943        /*
2944         * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2945         * so that it can be found and processed before all the index entries are
2946         * read.  Write it out regardless of the strip_extensions parameter as we need it
2947         * when loading the shared index.
2948         */
2949        if (offset) {
2950                struct strbuf sb = STRBUF_INIT;
2951
2952                write_eoie_extension(&sb, &eoie_c, offset);
2953                err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
2954                        || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2955                strbuf_release(&sb);
2956                if (err)
2957                        return -1;
2958        }
2959
2960        if (ce_flush(&c, newfd, istate->oid.hash))
2961                return -1;
2962        if (close_tempfile_gently(tempfile)) {
2963                error(_("could not close '%s'"), tempfile->filename.buf);
2964                return -1;
2965        }
2966        if (stat(tempfile->filename.buf, &st))
2967                return -1;
2968        istate->timestamp.sec = (unsigned int)st.st_mtime;
2969        istate->timestamp.nsec = ST_MTIME_NSEC(st);
2970        trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
2971        return 0;
2972}
2973
2974void set_alternate_index_output(const char *name)
2975{
2976        alternate_index_output = name;
2977}
2978
2979static int commit_locked_index(struct lock_file *lk)
2980{
2981        if (alternate_index_output)
2982                return commit_lock_file_to(lk, alternate_index_output);
2983        else
2984                return commit_lock_file(lk);
2985}
2986
2987static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2988                                 unsigned flags)
2989{
2990        int ret = do_write_index(istate, lock->tempfile, 0);
2991        if (ret)
2992                return ret;
2993        if (flags & COMMIT_LOCK)
2994                return commit_locked_index(lock);
2995        return close_lock_file_gently(lock);
2996}
2997
2998static int write_split_index(struct index_state *istate,
2999                             struct lock_file *lock,
3000                             unsigned flags)
3001{
3002        int ret;
3003        prepare_to_write_split_index(istate);
3004        ret = do_write_locked_index(istate, lock, flags);
3005        finish_writing_split_index(istate);
3006        return ret;
3007}
3008
3009static const char *shared_index_expire = "2.weeks.ago";
3010
3011static unsigned long get_shared_index_expire_date(void)
3012{
3013        static unsigned long shared_index_expire_date;
3014        static int shared_index_expire_date_prepared;
3015
3016        if (!shared_index_expire_date_prepared) {
3017                git_config_get_expiry("splitindex.sharedindexexpire",
3018                                      &shared_index_expire);
3019                shared_index_expire_date = approxidate(shared_index_expire);
3020                shared_index_expire_date_prepared = 1;
3021        }
3022
3023        return shared_index_expire_date;
3024}
3025
3026static int should_delete_shared_index(const char *shared_index_path)
3027{
3028        struct stat st;
3029        unsigned long expiration;
3030
3031        /* Check timestamp */
3032        expiration = get_shared_index_expire_date();
3033        if (!expiration)
3034                return 0;
3035        if (stat(shared_index_path, &st))
3036                return error_errno(_("could not stat '%s'"), shared_index_path);
3037        if (st.st_mtime > expiration)
3038                return 0;
3039
3040        return 1;
3041}
3042
3043static int clean_shared_index_files(const char *current_hex)
3044{
3045        struct dirent *de;
3046        DIR *dir = opendir(get_git_dir());
3047
3048        if (!dir)
3049                return error_errno(_("unable to open git dir: %s"), get_git_dir());
3050
3051        while ((de = readdir(dir)) != NULL) {
3052                const char *sha1_hex;
3053                const char *shared_index_path;
3054                if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3055                        continue;
3056                if (!strcmp(sha1_hex, current_hex))
3057                        continue;
3058                shared_index_path = git_path("%s", de->d_name);
3059                if (should_delete_shared_index(shared_index_path) > 0 &&
3060                    unlink(shared_index_path))
3061                        warning_errno(_("unable to unlink: %s"), shared_index_path);
3062        }
3063        closedir(dir);
3064
3065        return 0;
3066}
3067
3068static int write_shared_index(struct index_state *istate,
3069                              struct tempfile **temp)
3070{
3071        struct split_index *si = istate->split_index;
3072        int ret;
3073
3074        move_cache_to_base_index(istate);
3075        ret = do_write_index(si->base, *temp, 1);
3076        if (ret)
3077                return ret;
3078        ret = adjust_shared_perm(get_tempfile_path(*temp));
3079        if (ret) {
3080                error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
3081                return ret;
3082        }
3083        ret = rename_tempfile(temp,
3084                              git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
3085        if (!ret) {
3086                oidcpy(&si->base_oid, &si->base->oid);
3087                clean_shared_index_files(oid_to_hex(&si->base->oid));
3088        }
3089
3090        return ret;
3091}
3092
3093static const int default_max_percent_split_change = 20;
3094
3095static int too_many_not_shared_entries(struct index_state *istate)
3096{
3097        int i, not_shared = 0;
3098        int max_split = git_config_get_max_percent_split_change();
3099
3100        switch (max_split) {
3101        case -1:
3102                /* not or badly configured: use the default value */
3103                max_split = default_max_percent_split_change;
3104                break;
3105        case 0:
3106                return 1; /* 0% means always write a new shared index */
3107        case 100:
3108                return 0; /* 100% means never write a new shared index */
3109        default:
3110                break; /* just use the configured value */
3111        }
3112
3113        /* Count not shared entries */
3114        for (i = 0; i < istate->cache_nr; i++) {
3115                struct cache_entry *ce = istate->cache[i];
3116                if (!ce->index)
3117                        not_shared++;
3118        }
3119
3120        return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3121}
3122
3123int write_locked_index(struct index_state *istate, struct lock_file *lock,
3124                       unsigned flags)
3125{
3126        int new_shared_index, ret;
3127        struct split_index *si = istate->split_index;
3128
3129        if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3130                cache_tree_verify(istate);
3131
3132        if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3133                if (flags & COMMIT_LOCK)
3134                        rollback_lock_file(lock);
3135                return 0;
3136        }
3137
3138        if (istate->fsmonitor_last_update)
3139                fill_fsmonitor_bitmap(istate);
3140
3141        if (!si || alternate_index_output ||
3142            (istate->cache_changed & ~EXTMASK)) {
3143                if (si)
3144                        oidclr(&si->base_oid);
3145                ret = do_write_locked_index(istate, lock, flags);
3146                goto out;
3147        }
3148
3149        if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3150                int v = si->base_oid.hash[0];
3151                if ((v & 15) < 6)
3152                        istate->cache_changed |= SPLIT_INDEX_ORDERED;
3153        }
3154        if (too_many_not_shared_entries(istate))
3155                istate->cache_changed |= SPLIT_INDEX_ORDERED;
3156
3157        new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3158
3159        if (new_shared_index) {
3160                struct tempfile *temp;
3161                int saved_errno;
3162
3163                temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
3164                if (!temp) {
3165                        oidclr(&si->base_oid);
3166                        ret = do_write_locked_index(istate, lock, flags);
3167                        goto out;
3168                }
3169                ret = write_shared_index(istate, &temp);
3170
3171                saved_errno = errno;
3172                if (is_tempfile_active(temp))
3173                        delete_tempfile(&temp);
3174                errno = saved_errno;
3175
3176                if (ret)
3177                        goto out;
3178        }
3179
3180        ret = write_split_index(istate, lock, flags);
3181
3182        /* Freshen the shared index only if the split-index was written */
3183        if (!ret && !new_shared_index) {
3184                const char *shared_index = git_path("sharedindex.%s",
3185                                                    oid_to_hex(&si->base_oid));
3186                freshen_shared_index(shared_index, 1);
3187        }
3188
3189out:
3190        if (flags & COMMIT_LOCK)
3191                rollback_lock_file(lock);
3192        return ret;
3193}
3194
3195/*
3196 * Read the index file that is potentially unmerged into given
3197 * index_state, dropping any unmerged entries to stage #0 (potentially
3198 * resulting in a path appearing as both a file and a directory in the
3199 * index; the caller is responsible to clear out the extra entries
3200 * before writing the index to a tree).  Returns true if the index is
3201 * unmerged.  Callers who want to refuse to work from an unmerged
3202 * state can call this and check its return value, instead of calling
3203 * read_cache().
3204 */
3205int read_index_unmerged(struct index_state *istate)
3206{
3207        int i;
3208        int unmerged = 0;
3209
3210        read_index(istate);
3211        for (i = 0; i < istate->cache_nr; i++) {
3212                struct cache_entry *ce = istate->cache[i];
3213                struct cache_entry *new_ce;
3214                int len;
3215
3216                if (!ce_stage(ce))
3217                        continue;
3218                unmerged = 1;
3219                len = ce_namelen(ce);
3220                new_ce = make_empty_cache_entry(istate, len);
3221                memcpy(new_ce->name, ce->name, len);
3222                new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3223                new_ce->ce_namelen = len;
3224                new_ce->ce_mode = ce->ce_mode;
3225                if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3226                        return error(_("%s: cannot drop to stage #0"),
3227                                     new_ce->name);
3228        }
3229        return unmerged;
3230}
3231
3232/*
3233 * Returns 1 if the path is an "other" path with respect to
3234 * the index; that is, the path is not mentioned in the index at all,
3235 * either as a file, a directory with some files in the index,
3236 * or as an unmerged entry.
3237 *
3238 * We helpfully remove a trailing "/" from directories so that
3239 * the output of read_directory can be used as-is.
3240 */
3241int index_name_is_other(const struct index_state *istate, const char *name,
3242                int namelen)
3243{
3244        int pos;
3245        if (namelen && name[namelen - 1] == '/')
3246                namelen--;
3247        pos = index_name_pos(istate, name, namelen);
3248        if (0 <= pos)
3249                return 0;       /* exact match */
3250        pos = -pos - 1;
3251        if (pos < istate->cache_nr) {
3252                struct cache_entry *ce = istate->cache[pos];
3253                if (ce_namelen(ce) == namelen &&
3254                    !memcmp(ce->name, name, namelen))
3255                        return 0; /* Yup, this one exists unmerged */
3256        }
3257        return 1;
3258}
3259
3260void *read_blob_data_from_index(const struct index_state *istate,
3261                                const char *path, unsigned long *size)
3262{
3263        int pos, len;
3264        unsigned long sz;
3265        enum object_type type;
3266        void *data;
3267
3268        len = strlen(path);
3269        pos = index_name_pos(istate, path, len);
3270        if (pos < 0) {
3271                /*
3272                 * We might be in the middle of a merge, in which
3273                 * case we would read stage #2 (ours).
3274                 */
3275                int i;
3276                for (i = -pos - 1;
3277                     (pos < 0 && i < istate->cache_nr &&
3278                      !strcmp(istate->cache[i]->name, path));
3279                     i++)
3280                        if (ce_stage(istate->cache[i]) == 2)
3281                                pos = i;
3282        }
3283        if (pos < 0)
3284                return NULL;
3285        data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
3286        if (!data || type != OBJ_BLOB) {
3287                free(data);
3288                return NULL;
3289        }
3290        if (size)
3291                *size = sz;
3292        return data;
3293}
3294
3295void stat_validity_clear(struct stat_validity *sv)
3296{
3297        FREE_AND_NULL(sv->sd);
3298}
3299
3300int stat_validity_check(struct stat_validity *sv, const char *path)
3301{
3302        struct stat st;
3303
3304        if (stat(path, &st) < 0)
3305                return sv->sd == NULL;
3306        if (!sv->sd)
3307                return 0;
3308        return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3309}
3310
3311void stat_validity_update(struct stat_validity *sv, int fd)
3312{
3313        struct stat st;
3314
3315        if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3316                stat_validity_clear(sv);
3317        else {
3318                if (!sv->sd)
3319                        sv->sd = xcalloc(1, sizeof(struct stat_data));
3320                fill_stat_data(sv->sd, &st);
3321        }
3322}
3323
3324void move_index_extensions(struct index_state *dst, struct index_state *src)
3325{
3326        dst->untracked = src->untracked;
3327        src->untracked = NULL;
3328        dst->cache_tree = src->cache_tree;
3329        src->cache_tree = NULL;
3330}
3331
3332struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3333                                    struct index_state *istate)
3334{
3335        unsigned int size = ce_size(ce);
3336        int mem_pool_allocated;
3337        struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3338        mem_pool_allocated = new_entry->mem_pool_allocated;
3339
3340        memcpy(new_entry, ce, size);
3341        new_entry->mem_pool_allocated = mem_pool_allocated;
3342        return new_entry;
3343}
3344
3345void discard_cache_entry(struct cache_entry *ce)
3346{
3347        if (ce && should_validate_cache_entries())
3348                memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3349
3350        if (ce && ce->mem_pool_allocated)
3351                return;
3352
3353        free(ce);
3354}
3355
3356int should_validate_cache_entries(void)
3357{
3358        static int validate_index_cache_entries = -1;
3359
3360        if (validate_index_cache_entries < 0) {
3361                if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3362                        validate_index_cache_entries = 1;
3363                else
3364                        validate_index_cache_entries = 0;
3365        }
3366
3367        return validate_index_cache_entries;
3368}
3369
3370#define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3371#define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3372
3373static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3374{
3375        /*
3376         * The end of index entries (EOIE) extension is guaranteed to be last
3377         * so that it can be found by scanning backwards from the EOF.
3378         *
3379         * "EOIE"
3380         * <4-byte length>
3381         * <4-byte offset>
3382         * <20-byte hash>
3383         */
3384        const char *index, *eoie;
3385        uint32_t extsize;
3386        size_t offset, src_offset;
3387        unsigned char hash[GIT_MAX_RAWSZ];
3388        git_hash_ctx c;
3389
3390        /* ensure we have an index big enough to contain an EOIE extension */
3391        if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3392                return 0;
3393
3394        /* validate the extension signature */
3395        index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3396        if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3397                return 0;
3398        index += sizeof(uint32_t);
3399
3400        /* validate the extension size */
3401        extsize = get_be32(index);
3402        if (extsize != EOIE_SIZE)
3403                return 0;
3404        index += sizeof(uint32_t);
3405
3406        /*
3407         * Validate the offset we're going to look for the first extension
3408         * signature is after the index header and before the eoie extension.
3409         */
3410        offset = get_be32(index);
3411        if (mmap + offset < mmap + sizeof(struct cache_header))
3412                return 0;
3413        if (mmap + offset >= eoie)
3414                return 0;
3415        index += sizeof(uint32_t);
3416
3417        /*
3418         * The hash is computed over extension types and their sizes (but not
3419         * their contents).  E.g. if we have "TREE" extension that is N-bytes
3420         * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3421         * then the hash would be:
3422         *
3423         * SHA-1("TREE" + <binary representation of N> +
3424         *       "REUC" + <binary representation of M>)
3425         */
3426        src_offset = offset;
3427        the_hash_algo->init_fn(&c);
3428        while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3429                /* After an array of active_nr index entries,
3430                 * there can be arbitrary number of extended
3431                 * sections, each of which is prefixed with
3432                 * extension name (4-byte) and section length
3433                 * in 4-byte network byte order.
3434                 */
3435                uint32_t extsize;
3436                memcpy(&extsize, mmap + src_offset + 4, 4);
3437                extsize = ntohl(extsize);
3438
3439                /* verify the extension size isn't so large it will wrap around */
3440                if (src_offset + 8 + extsize < src_offset)
3441                        return 0;
3442
3443                the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3444
3445                src_offset += 8;
3446                src_offset += extsize;
3447        }
3448        the_hash_algo->final_fn(hash, &c);
3449        if (!hasheq(hash, (const unsigned char *)index))
3450                return 0;
3451
3452        /* Validate that the extension offsets returned us back to the eoie extension. */
3453        if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3454                return 0;
3455
3456        return offset;
3457}
3458
3459static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3460{
3461        uint32_t buffer;
3462        unsigned char hash[GIT_MAX_RAWSZ];
3463
3464        /* offset */
3465        put_be32(&buffer, offset);
3466        strbuf_add(sb, &buffer, sizeof(uint32_t));
3467
3468        /* hash */
3469        the_hash_algo->final_fn(hash, eoie_context);
3470        strbuf_add(sb, hash, the_hash_algo->rawsz);
3471}
3472
3473#ifndef NO_PTHREADS
3474#define IEOT_VERSION    (1)
3475
3476static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3477{
3478       const char *index = NULL;
3479       uint32_t extsize, ext_version;
3480       struct index_entry_offset_table *ieot;
3481       int i, nr;
3482
3483       /* find the IEOT extension */
3484       if (!offset)
3485               return NULL;
3486       while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3487               extsize = get_be32(mmap + offset + 4);
3488               if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3489                       index = mmap + offset + 4 + 4;
3490                       break;
3491               }
3492               offset += 8;
3493               offset += extsize;
3494       }
3495       if (!index)
3496               return NULL;
3497
3498       /* validate the version is IEOT_VERSION */
3499       ext_version = get_be32(index);
3500       if (ext_version != IEOT_VERSION) {
3501               error("invalid IEOT version %d", ext_version);
3502               return NULL;
3503       }
3504       index += sizeof(uint32_t);
3505
3506       /* extension size - version bytes / bytes per entry */
3507       nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3508       if (!nr) {
3509               error("invalid number of IEOT entries %d", nr);
3510               return NULL;
3511       }
3512       ieot = xmalloc(sizeof(struct index_entry_offset_table)
3513               + (nr * sizeof(struct index_entry_offset)));
3514       ieot->nr = nr;
3515       for (i = 0; i < nr; i++) {
3516               ieot->entries[i].offset = get_be32(index);
3517               index += sizeof(uint32_t);
3518               ieot->entries[i].nr = get_be32(index);
3519               index += sizeof(uint32_t);
3520       }
3521
3522       return ieot;
3523}
3524
3525static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)
3526{
3527       uint32_t buffer;
3528       int i;
3529
3530       /* version */
3531       put_be32(&buffer, IEOT_VERSION);
3532       strbuf_add(sb, &buffer, sizeof(uint32_t));
3533
3534       /* ieot */
3535       for (i = 0; i < ieot->nr; i++) {
3536
3537               /* offset */
3538               put_be32(&buffer, ieot->entries[i].offset);
3539               strbuf_add(sb, &buffer, sizeof(uint32_t));
3540
3541               /* count */
3542               put_be32(&buffer, ieot->entries[i].nr);
3543               strbuf_add(sb, &buffer, sizeof(uint32_t));
3544       }
3545}
3546#endif