9addad68877bd27f435fc2dba4399bc544e19db6
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 *
   6 * This handles basic git sha1 object files - packing, unpacking,
   7 * creation etc.
   8 */
   9#include "cache.h"
  10#include "config.h"
  11#include "string-list.h"
  12#include "lockfile.h"
  13#include "delta.h"
  14#include "pack.h"
  15#include "blob.h"
  16#include "commit.h"
  17#include "run-command.h"
  18#include "tag.h"
  19#include "tree.h"
  20#include "tree-walk.h"
  21#include "refs.h"
  22#include "pack-revindex.h"
  23#include "sha1-lookup.h"
  24#include "bulk-checkin.h"
  25#include "repository.h"
  26#include "streaming.h"
  27#include "dir.h"
  28#include "list.h"
  29#include "mergesort.h"
  30#include "quote.h"
  31#include "packfile.h"
  32#include "fetch-object.h"
  33#include "object-store.h"
  34
  35const unsigned char null_sha1[GIT_MAX_RAWSZ];
  36const struct object_id null_oid;
  37const struct object_id empty_tree_oid = {
  38        EMPTY_TREE_SHA1_BIN_LITERAL
  39};
  40const struct object_id empty_blob_oid = {
  41        EMPTY_BLOB_SHA1_BIN_LITERAL
  42};
  43
  44static void git_hash_sha1_init(void *ctx)
  45{
  46        git_SHA1_Init((git_SHA_CTX *)ctx);
  47}
  48
  49static void git_hash_sha1_update(void *ctx, const void *data, size_t len)
  50{
  51        git_SHA1_Update((git_SHA_CTX *)ctx, data, len);
  52}
  53
  54static void git_hash_sha1_final(unsigned char *hash, void *ctx)
  55{
  56        git_SHA1_Final(hash, (git_SHA_CTX *)ctx);
  57}
  58
  59static void git_hash_unknown_init(void *ctx)
  60{
  61        die("trying to init unknown hash");
  62}
  63
  64static void git_hash_unknown_update(void *ctx, const void *data, size_t len)
  65{
  66        die("trying to update unknown hash");
  67}
  68
  69static void git_hash_unknown_final(unsigned char *hash, void *ctx)
  70{
  71        die("trying to finalize unknown hash");
  72}
  73
  74const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
  75        {
  76                NULL,
  77                0x00000000,
  78                0,
  79                0,
  80                0,
  81                git_hash_unknown_init,
  82                git_hash_unknown_update,
  83                git_hash_unknown_final,
  84                NULL,
  85                NULL,
  86        },
  87        {
  88                "sha-1",
  89                /* "sha1", big-endian */
  90                0x73686131,
  91                sizeof(git_SHA_CTX),
  92                GIT_SHA1_RAWSZ,
  93                GIT_SHA1_HEXSZ,
  94                git_hash_sha1_init,
  95                git_hash_sha1_update,
  96                git_hash_sha1_final,
  97                &empty_tree_oid,
  98                &empty_blob_oid,
  99        },
 100};
 101
 102/*
 103 * This is meant to hold a *small* number of objects that you would
 104 * want read_sha1_file() to be able to return, but yet you do not want
 105 * to write them into the object store (e.g. a browse-only
 106 * application).
 107 */
 108static struct cached_object {
 109        unsigned char sha1[20];
 110        enum object_type type;
 111        void *buf;
 112        unsigned long size;
 113} *cached_objects;
 114static int cached_object_nr, cached_object_alloc;
 115
 116static struct cached_object empty_tree = {
 117        EMPTY_TREE_SHA1_BIN_LITERAL,
 118        OBJ_TREE,
 119        "",
 120        0
 121};
 122
 123static struct cached_object *find_cached_object(const unsigned char *sha1)
 124{
 125        int i;
 126        struct cached_object *co = cached_objects;
 127
 128        for (i = 0; i < cached_object_nr; i++, co++) {
 129                if (!hashcmp(co->sha1, sha1))
 130                        return co;
 131        }
 132        if (!hashcmp(sha1, empty_tree.sha1))
 133                return &empty_tree;
 134        return NULL;
 135}
 136
 137
 138static int get_conv_flags(unsigned flags)
 139{
 140        if (flags & HASH_RENORMALIZE)
 141                return CONV_EOL_RENORMALIZE;
 142        else if (flags & HASH_WRITE_OBJECT)
 143          return global_conv_flags_eol;
 144        else
 145                return 0;
 146}
 147
 148
 149int mkdir_in_gitdir(const char *path)
 150{
 151        if (mkdir(path, 0777)) {
 152                int saved_errno = errno;
 153                struct stat st;
 154                struct strbuf sb = STRBUF_INIT;
 155
 156                if (errno != EEXIST)
 157                        return -1;
 158                /*
 159                 * Are we looking at a path in a symlinked worktree
 160                 * whose original repository does not yet have it?
 161                 * e.g. .git/rr-cache pointing at its original
 162                 * repository in which the user hasn't performed any
 163                 * conflict resolution yet?
 164                 */
 165                if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
 166                    strbuf_readlink(&sb, path, st.st_size) ||
 167                    !is_absolute_path(sb.buf) ||
 168                    mkdir(sb.buf, 0777)) {
 169                        strbuf_release(&sb);
 170                        errno = saved_errno;
 171                        return -1;
 172                }
 173                strbuf_release(&sb);
 174        }
 175        return adjust_shared_perm(path);
 176}
 177
 178enum scld_error safe_create_leading_directories(char *path)
 179{
 180        char *next_component = path + offset_1st_component(path);
 181        enum scld_error ret = SCLD_OK;
 182
 183        while (ret == SCLD_OK && next_component) {
 184                struct stat st;
 185                char *slash = next_component, slash_character;
 186
 187                while (*slash && !is_dir_sep(*slash))
 188                        slash++;
 189
 190                if (!*slash)
 191                        break;
 192
 193                next_component = slash + 1;
 194                while (is_dir_sep(*next_component))
 195                        next_component++;
 196                if (!*next_component)
 197                        break;
 198
 199                slash_character = *slash;
 200                *slash = '\0';
 201                if (!stat(path, &st)) {
 202                        /* path exists */
 203                        if (!S_ISDIR(st.st_mode)) {
 204                                errno = ENOTDIR;
 205                                ret = SCLD_EXISTS;
 206                        }
 207                } else if (mkdir(path, 0777)) {
 208                        if (errno == EEXIST &&
 209                            !stat(path, &st) && S_ISDIR(st.st_mode))
 210                                ; /* somebody created it since we checked */
 211                        else if (errno == ENOENT)
 212                                /*
 213                                 * Either mkdir() failed because
 214                                 * somebody just pruned the containing
 215                                 * directory, or stat() failed because
 216                                 * the file that was in our way was
 217                                 * just removed.  Either way, inform
 218                                 * the caller that it might be worth
 219                                 * trying again:
 220                                 */
 221                                ret = SCLD_VANISHED;
 222                        else
 223                                ret = SCLD_FAILED;
 224                } else if (adjust_shared_perm(path)) {
 225                        ret = SCLD_PERMS;
 226                }
 227                *slash = slash_character;
 228        }
 229        return ret;
 230}
 231
 232enum scld_error safe_create_leading_directories_const(const char *path)
 233{
 234        int save_errno;
 235        /* path points to cache entries, so xstrdup before messing with it */
 236        char *buf = xstrdup(path);
 237        enum scld_error result = safe_create_leading_directories(buf);
 238
 239        save_errno = errno;
 240        free(buf);
 241        errno = save_errno;
 242        return result;
 243}
 244
 245int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
 246{
 247        /*
 248         * The number of times we will try to remove empty directories
 249         * in the way of path. This is only 1 because if another
 250         * process is racily creating directories that conflict with
 251         * us, we don't want to fight against them.
 252         */
 253        int remove_directories_remaining = 1;
 254
 255        /*
 256         * The number of times that we will try to create the
 257         * directories containing path. We are willing to attempt this
 258         * more than once, because another process could be trying to
 259         * clean up empty directories at the same time as we are
 260         * trying to create them.
 261         */
 262        int create_directories_remaining = 3;
 263
 264        /* A scratch copy of path, filled lazily if we need it: */
 265        struct strbuf path_copy = STRBUF_INIT;
 266
 267        int ret, save_errno;
 268
 269        /* Sanity check: */
 270        assert(*path);
 271
 272retry_fn:
 273        ret = fn(path, cb);
 274        save_errno = errno;
 275        if (!ret)
 276                goto out;
 277
 278        if (errno == EISDIR && remove_directories_remaining-- > 0) {
 279                /*
 280                 * A directory is in the way. Maybe it is empty; try
 281                 * to remove it:
 282                 */
 283                if (!path_copy.len)
 284                        strbuf_addstr(&path_copy, path);
 285
 286                if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
 287                        goto retry_fn;
 288        } else if (errno == ENOENT && create_directories_remaining-- > 0) {
 289                /*
 290                 * Maybe the containing directory didn't exist, or
 291                 * maybe it was just deleted by a process that is
 292                 * racing with us to clean up empty directories. Try
 293                 * to create it:
 294                 */
 295                enum scld_error scld_result;
 296
 297                if (!path_copy.len)
 298                        strbuf_addstr(&path_copy, path);
 299
 300                do {
 301                        scld_result = safe_create_leading_directories(path_copy.buf);
 302                        if (scld_result == SCLD_OK)
 303                                goto retry_fn;
 304                } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
 305        }
 306
 307out:
 308        strbuf_release(&path_copy);
 309        errno = save_errno;
 310        return ret;
 311}
 312
 313static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
 314{
 315        int i;
 316        for (i = 0; i < 20; i++) {
 317                static char hex[] = "0123456789abcdef";
 318                unsigned int val = sha1[i];
 319                strbuf_addch(buf, hex[val >> 4]);
 320                strbuf_addch(buf, hex[val & 0xf]);
 321                if (!i)
 322                        strbuf_addch(buf, '/');
 323        }
 324}
 325
 326void sha1_file_name(struct repository *r, struct strbuf *buf, const unsigned char *sha1)
 327{
 328        strbuf_addstr(buf, r->objects->objectdir);
 329        strbuf_addch(buf, '/');
 330        fill_sha1_path(buf, sha1);
 331}
 332
 333struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
 334{
 335        strbuf_setlen(&alt->scratch, alt->base_len);
 336        return &alt->scratch;
 337}
 338
 339static const char *alt_sha1_path(struct alternate_object_database *alt,
 340                                 const unsigned char *sha1)
 341{
 342        struct strbuf *buf = alt_scratch_buf(alt);
 343        fill_sha1_path(buf, sha1);
 344        return buf->buf;
 345}
 346
 347/*
 348 * Return non-zero iff the path is usable as an alternate object database.
 349 */
 350static int alt_odb_usable(struct raw_object_store *o,
 351                          struct strbuf *path,
 352                          const char *normalized_objdir)
 353{
 354        struct alternate_object_database *alt;
 355
 356        /* Detect cases where alternate disappeared */
 357        if (!is_directory(path->buf)) {
 358                error("object directory %s does not exist; "
 359                      "check .git/objects/info/alternates.",
 360                      path->buf);
 361                return 0;
 362        }
 363
 364        /*
 365         * Prevent the common mistake of listing the same
 366         * thing twice, or object directory itself.
 367         */
 368        for (alt = o->alt_odb_list; alt; alt = alt->next) {
 369                if (!fspathcmp(path->buf, alt->path))
 370                        return 0;
 371        }
 372        if (!fspathcmp(path->buf, normalized_objdir))
 373                return 0;
 374
 375        return 1;
 376}
 377
 378/*
 379 * Prepare alternate object database registry.
 380 *
 381 * The variable alt_odb_list points at the list of struct
 382 * alternate_object_database.  The elements on this list come from
 383 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 384 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 385 * whose contents is similar to that environment variable but can be
 386 * LF separated.  Its base points at a statically allocated buffer that
 387 * contains "/the/directory/corresponding/to/.git/objects/...", while
 388 * its name points just after the slash at the end of ".git/objects/"
 389 * in the example above, and has enough space to hold 40-byte hex
 390 * SHA1, an extra slash for the first level indirection, and the
 391 * terminating NUL.
 392 */
 393static void read_info_alternates(struct repository *r,
 394                                 const char *relative_base,
 395                                 int depth);
 396static int link_alt_odb_entry(struct repository *r, const char *entry,
 397        const char *relative_base, int depth, const char *normalized_objdir)
 398{
 399        struct alternate_object_database *ent;
 400        struct strbuf pathbuf = STRBUF_INIT;
 401
 402        if (!is_absolute_path(entry) && relative_base) {
 403                strbuf_realpath(&pathbuf, relative_base, 1);
 404                strbuf_addch(&pathbuf, '/');
 405        }
 406        strbuf_addstr(&pathbuf, entry);
 407
 408        if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
 409                error("unable to normalize alternate object path: %s",
 410                      pathbuf.buf);
 411                strbuf_release(&pathbuf);
 412                return -1;
 413        }
 414
 415        /*
 416         * The trailing slash after the directory name is given by
 417         * this function at the end. Remove duplicates.
 418         */
 419        while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
 420                strbuf_setlen(&pathbuf, pathbuf.len - 1);
 421
 422        if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
 423                strbuf_release(&pathbuf);
 424                return -1;
 425        }
 426
 427        ent = alloc_alt_odb(pathbuf.buf);
 428
 429        /* add the alternate entry */
 430        *r->objects->alt_odb_tail = ent;
 431        r->objects->alt_odb_tail = &(ent->next);
 432        ent->next = NULL;
 433
 434        /* recursively add alternates */
 435        read_info_alternates(r, pathbuf.buf, depth + 1);
 436
 437        strbuf_release(&pathbuf);
 438        return 0;
 439}
 440
 441static const char *parse_alt_odb_entry(const char *string,
 442                                       int sep,
 443                                       struct strbuf *out)
 444{
 445        const char *end;
 446
 447        strbuf_reset(out);
 448
 449        if (*string == '#') {
 450                /* comment; consume up to next separator */
 451                end = strchrnul(string, sep);
 452        } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
 453                /*
 454                 * quoted path; unquote_c_style has copied the
 455                 * data for us and set "end". Broken quoting (e.g.,
 456                 * an entry that doesn't end with a quote) falls
 457                 * back to the unquoted case below.
 458                 */
 459        } else {
 460                /* normal, unquoted path */
 461                end = strchrnul(string, sep);
 462                strbuf_add(out, string, end - string);
 463        }
 464
 465        if (*end)
 466                end++;
 467        return end;
 468}
 469
 470static void link_alt_odb_entries(struct repository *r, const char *alt,
 471                                 int sep, const char *relative_base, int depth)
 472{
 473        struct strbuf objdirbuf = STRBUF_INIT;
 474        struct strbuf entry = STRBUF_INIT;
 475
 476        if (!alt || !*alt)
 477                return;
 478
 479        if (depth > 5) {
 480                error("%s: ignoring alternate object stores, nesting too deep.",
 481                                relative_base);
 482                return;
 483        }
 484
 485        strbuf_add_absolute_path(&objdirbuf, r->objects->objectdir);
 486        if (strbuf_normalize_path(&objdirbuf) < 0)
 487                die("unable to normalize object directory: %s",
 488                    objdirbuf.buf);
 489
 490        while (*alt) {
 491                alt = parse_alt_odb_entry(alt, sep, &entry);
 492                if (!entry.len)
 493                        continue;
 494                link_alt_odb_entry(r, entry.buf,
 495                                   relative_base, depth, objdirbuf.buf);
 496        }
 497        strbuf_release(&entry);
 498        strbuf_release(&objdirbuf);
 499}
 500
 501static void read_info_alternates(struct repository *r,
 502                                 const char *relative_base,
 503                                 int depth)
 504{
 505        char *path;
 506        struct strbuf buf = STRBUF_INIT;
 507
 508        path = xstrfmt("%s/info/alternates", relative_base);
 509        if (strbuf_read_file(&buf, path, 1024) < 0) {
 510                warn_on_fopen_errors(path);
 511                free(path);
 512                return;
 513        }
 514
 515        link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
 516        strbuf_release(&buf);
 517        free(path);
 518}
 519
 520struct alternate_object_database *alloc_alt_odb(const char *dir)
 521{
 522        struct alternate_object_database *ent;
 523
 524        FLEX_ALLOC_STR(ent, path, dir);
 525        strbuf_init(&ent->scratch, 0);
 526        strbuf_addf(&ent->scratch, "%s/", dir);
 527        ent->base_len = ent->scratch.len;
 528
 529        return ent;
 530}
 531
 532void add_to_alternates_file(const char *reference)
 533{
 534        struct lock_file lock = LOCK_INIT;
 535        char *alts = git_pathdup("objects/info/alternates");
 536        FILE *in, *out;
 537        int found = 0;
 538
 539        hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
 540        out = fdopen_lock_file(&lock, "w");
 541        if (!out)
 542                die_errno("unable to fdopen alternates lockfile");
 543
 544        in = fopen(alts, "r");
 545        if (in) {
 546                struct strbuf line = STRBUF_INIT;
 547
 548                while (strbuf_getline(&line, in) != EOF) {
 549                        if (!strcmp(reference, line.buf)) {
 550                                found = 1;
 551                                break;
 552                        }
 553                        fprintf_or_die(out, "%s\n", line.buf);
 554                }
 555
 556                strbuf_release(&line);
 557                fclose(in);
 558        }
 559        else if (errno != ENOENT)
 560                die_errno("unable to read alternates file");
 561
 562        if (found) {
 563                rollback_lock_file(&lock);
 564        } else {
 565                fprintf_or_die(out, "%s\n", reference);
 566                if (commit_lock_file(&lock))
 567                        die_errno("unable to move new alternates file into place");
 568                if (the_repository->objects->alt_odb_tail)
 569                        link_alt_odb_entries(the_repository, reference,
 570                                             '\n', NULL, 0);
 571        }
 572        free(alts);
 573}
 574
 575void add_to_alternates_memory(const char *reference)
 576{
 577        /*
 578         * Make sure alternates are initialized, or else our entry may be
 579         * overwritten when they are.
 580         */
 581        prepare_alt_odb(the_repository);
 582
 583        link_alt_odb_entries(the_repository, reference,
 584                             '\n', NULL, 0);
 585}
 586
 587/*
 588 * Compute the exact path an alternate is at and returns it. In case of
 589 * error NULL is returned and the human readable error is added to `err`
 590 * `path` may be relative and should point to $GITDIR.
 591 * `err` must not be null.
 592 */
 593char *compute_alternate_path(const char *path, struct strbuf *err)
 594{
 595        char *ref_git = NULL;
 596        const char *repo, *ref_git_s;
 597        int seen_error = 0;
 598
 599        ref_git_s = real_path_if_valid(path);
 600        if (!ref_git_s) {
 601                seen_error = 1;
 602                strbuf_addf(err, _("path '%s' does not exist"), path);
 603                goto out;
 604        } else
 605                /*
 606                 * Beware: read_gitfile(), real_path() and mkpath()
 607                 * return static buffer
 608                 */
 609                ref_git = xstrdup(ref_git_s);
 610
 611        repo = read_gitfile(ref_git);
 612        if (!repo)
 613                repo = read_gitfile(mkpath("%s/.git", ref_git));
 614        if (repo) {
 615                free(ref_git);
 616                ref_git = xstrdup(repo);
 617        }
 618
 619        if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
 620                char *ref_git_git = mkpathdup("%s/.git", ref_git);
 621                free(ref_git);
 622                ref_git = ref_git_git;
 623        } else if (!is_directory(mkpath("%s/objects", ref_git))) {
 624                struct strbuf sb = STRBUF_INIT;
 625                seen_error = 1;
 626                if (get_common_dir(&sb, ref_git)) {
 627                        strbuf_addf(err,
 628                                    _("reference repository '%s' as a linked "
 629                                      "checkout is not supported yet."),
 630                                    path);
 631                        goto out;
 632                }
 633
 634                strbuf_addf(err, _("reference repository '%s' is not a "
 635                                        "local repository."), path);
 636                goto out;
 637        }
 638
 639        if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
 640                strbuf_addf(err, _("reference repository '%s' is shallow"),
 641                            path);
 642                seen_error = 1;
 643                goto out;
 644        }
 645
 646        if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
 647                strbuf_addf(err,
 648                            _("reference repository '%s' is grafted"),
 649                            path);
 650                seen_error = 1;
 651                goto out;
 652        }
 653
 654out:
 655        if (seen_error) {
 656                FREE_AND_NULL(ref_git);
 657        }
 658
 659        return ref_git;
 660}
 661
 662int foreach_alt_odb(alt_odb_fn fn, void *cb)
 663{
 664        struct alternate_object_database *ent;
 665        int r = 0;
 666
 667        prepare_alt_odb(the_repository);
 668        for (ent = the_repository->objects->alt_odb_list; ent; ent = ent->next) {
 669                r = fn(ent, cb);
 670                if (r)
 671                        break;
 672        }
 673        return r;
 674}
 675
 676void prepare_alt_odb(struct repository *r)
 677{
 678        if (r->objects->alt_odb_tail)
 679                return;
 680
 681        r->objects->alt_odb_tail = &r->objects->alt_odb_list;
 682        link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
 683
 684        read_info_alternates(r, r->objects->objectdir, 0);
 685}
 686
 687/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
 688static int freshen_file(const char *fn)
 689{
 690        struct utimbuf t;
 691        t.actime = t.modtime = time(NULL);
 692        return !utime(fn, &t);
 693}
 694
 695/*
 696 * All of the check_and_freshen functions return 1 if the file exists and was
 697 * freshened (if freshening was requested), 0 otherwise. If they return
 698 * 0, you should not assume that it is safe to skip a write of the object (it
 699 * either does not exist on disk, or has a stale mtime and may be subject to
 700 * pruning).
 701 */
 702int check_and_freshen_file(const char *fn, int freshen)
 703{
 704        if (access(fn, F_OK))
 705                return 0;
 706        if (freshen && !freshen_file(fn))
 707                return 0;
 708        return 1;
 709}
 710
 711static int check_and_freshen_local(const unsigned char *sha1, int freshen)
 712{
 713        static struct strbuf buf = STRBUF_INIT;
 714
 715        strbuf_reset(&buf);
 716        sha1_file_name(the_repository, &buf, sha1);
 717
 718        return check_and_freshen_file(buf.buf, freshen);
 719}
 720
 721static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
 722{
 723        struct alternate_object_database *alt;
 724        prepare_alt_odb(the_repository);
 725        for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
 726                const char *path = alt_sha1_path(alt, sha1);
 727                if (check_and_freshen_file(path, freshen))
 728                        return 1;
 729        }
 730        return 0;
 731}
 732
 733static int check_and_freshen(const unsigned char *sha1, int freshen)
 734{
 735        return check_and_freshen_local(sha1, freshen) ||
 736               check_and_freshen_nonlocal(sha1, freshen);
 737}
 738
 739int has_loose_object_nonlocal(const unsigned char *sha1)
 740{
 741        return check_and_freshen_nonlocal(sha1, 0);
 742}
 743
 744static int has_loose_object(const unsigned char *sha1)
 745{
 746        return check_and_freshen(sha1, 0);
 747}
 748
 749static void mmap_limit_check(size_t length)
 750{
 751        static size_t limit = 0;
 752        if (!limit) {
 753                limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
 754                if (!limit)
 755                        limit = SIZE_MAX;
 756        }
 757        if (length > limit)
 758                die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
 759                    (uintmax_t)length, (uintmax_t)limit);
 760}
 761
 762void *xmmap_gently(void *start, size_t length,
 763                  int prot, int flags, int fd, off_t offset)
 764{
 765        void *ret;
 766
 767        mmap_limit_check(length);
 768        ret = mmap(start, length, prot, flags, fd, offset);
 769        if (ret == MAP_FAILED) {
 770                if (!length)
 771                        return NULL;
 772                release_pack_memory(length);
 773                ret = mmap(start, length, prot, flags, fd, offset);
 774        }
 775        return ret;
 776}
 777
 778void *xmmap(void *start, size_t length,
 779        int prot, int flags, int fd, off_t offset)
 780{
 781        void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
 782        if (ret == MAP_FAILED)
 783                die_errno("mmap failed");
 784        return ret;
 785}
 786
 787/*
 788 * With an in-core object data in "map", rehash it to make sure the
 789 * object name actually matches "sha1" to detect object corruption.
 790 * With "map" == NULL, try reading the object named with "sha1" using
 791 * the streaming interface and rehash it to do the same.
 792 */
 793int check_sha1_signature(const unsigned char *sha1, void *map,
 794                         unsigned long size, const char *type)
 795{
 796        unsigned char real_sha1[20];
 797        enum object_type obj_type;
 798        struct git_istream *st;
 799        git_SHA_CTX c;
 800        char hdr[32];
 801        int hdrlen;
 802
 803        if (map) {
 804                hash_sha1_file(map, size, type, real_sha1);
 805                return hashcmp(sha1, real_sha1) ? -1 : 0;
 806        }
 807
 808        st = open_istream(sha1, &obj_type, &size, NULL);
 809        if (!st)
 810                return -1;
 811
 812        /* Generate the header */
 813        hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
 814
 815        /* Sha1.. */
 816        git_SHA1_Init(&c);
 817        git_SHA1_Update(&c, hdr, hdrlen);
 818        for (;;) {
 819                char buf[1024 * 16];
 820                ssize_t readlen = read_istream(st, buf, sizeof(buf));
 821
 822                if (readlen < 0) {
 823                        close_istream(st);
 824                        return -1;
 825                }
 826                if (!readlen)
 827                        break;
 828                git_SHA1_Update(&c, buf, readlen);
 829        }
 830        git_SHA1_Final(real_sha1, &c);
 831        close_istream(st);
 832        return hashcmp(sha1, real_sha1) ? -1 : 0;
 833}
 834
 835int git_open_cloexec(const char *name, int flags)
 836{
 837        int fd;
 838        static int o_cloexec = O_CLOEXEC;
 839
 840        fd = open(name, flags | o_cloexec);
 841        if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
 842                /* Try again w/o O_CLOEXEC: the kernel might not support it */
 843                o_cloexec &= ~O_CLOEXEC;
 844                fd = open(name, flags | o_cloexec);
 845        }
 846
 847#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
 848        {
 849                static int fd_cloexec = FD_CLOEXEC;
 850
 851                if (!o_cloexec && 0 <= fd && fd_cloexec) {
 852                        /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
 853                        int flags = fcntl(fd, F_GETFD);
 854                        if (fcntl(fd, F_SETFD, flags | fd_cloexec))
 855                                fd_cloexec = 0;
 856                }
 857        }
 858#endif
 859        return fd;
 860}
 861
 862/*
 863 * Find "sha1" as a loose object in the local repository or in an alternate.
 864 * Returns 0 on success, negative on failure.
 865 *
 866 * The "path" out-parameter will give the path of the object we found (if any).
 867 * Note that it may point to static storage and is only valid until another
 868 * call to sha1_file_name(), etc.
 869 */
 870static int stat_sha1_file(struct repository *r, const unsigned char *sha1,
 871                          struct stat *st, const char **path)
 872{
 873        struct alternate_object_database *alt;
 874        static struct strbuf buf = STRBUF_INIT;
 875
 876        strbuf_reset(&buf);
 877        sha1_file_name(r, &buf, sha1);
 878        *path = buf.buf;
 879
 880        if (!lstat(*path, st))
 881                return 0;
 882
 883        prepare_alt_odb(r);
 884        errno = ENOENT;
 885        for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
 886                *path = alt_sha1_path(alt, sha1);
 887                if (!lstat(*path, st))
 888                        return 0;
 889        }
 890
 891        return -1;
 892}
 893
 894/*
 895 * Like stat_sha1_file(), but actually open the object and return the
 896 * descriptor. See the caveats on the "path" parameter above.
 897 */
 898static int open_sha1_file(struct repository *r,
 899                          const unsigned char *sha1, const char **path)
 900{
 901        int fd;
 902        struct alternate_object_database *alt;
 903        int most_interesting_errno;
 904        static struct strbuf buf = STRBUF_INIT;
 905
 906        strbuf_reset(&buf);
 907        sha1_file_name(r, &buf, sha1);
 908        *path = buf.buf;
 909
 910        fd = git_open(*path);
 911        if (fd >= 0)
 912                return fd;
 913        most_interesting_errno = errno;
 914
 915        prepare_alt_odb(r);
 916        for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
 917                *path = alt_sha1_path(alt, sha1);
 918                fd = git_open(*path);
 919                if (fd >= 0)
 920                        return fd;
 921                if (most_interesting_errno == ENOENT)
 922                        most_interesting_errno = errno;
 923        }
 924        errno = most_interesting_errno;
 925        return -1;
 926}
 927
 928/*
 929 * Map the loose object at "path" if it is not NULL, or the path found by
 930 * searching for a loose object named "sha1".
 931 */
 932static void *map_sha1_file_1(struct repository *r, const char *path,
 933                             const unsigned char *sha1, unsigned long *size)
 934{
 935        void *map;
 936        int fd;
 937
 938        if (path)
 939                fd = git_open(path);
 940        else
 941                fd = open_sha1_file(r, sha1, &path);
 942        map = NULL;
 943        if (fd >= 0) {
 944                struct stat st;
 945
 946                if (!fstat(fd, &st)) {
 947                        *size = xsize_t(st.st_size);
 948                        if (!*size) {
 949                                /* mmap() is forbidden on empty files */
 950                                error("object file %s is empty", path);
 951                                return NULL;
 952                        }
 953                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 954                }
 955                close(fd);
 956        }
 957        return map;
 958}
 959
 960void *map_sha1_file(struct repository *r,
 961                    const unsigned char *sha1, unsigned long *size)
 962{
 963        return map_sha1_file_1(r, NULL, sha1, size);
 964}
 965
 966static int unpack_sha1_short_header(git_zstream *stream,
 967                                    unsigned char *map, unsigned long mapsize,
 968                                    void *buffer, unsigned long bufsiz)
 969{
 970        /* Get the data stream */
 971        memset(stream, 0, sizeof(*stream));
 972        stream->next_in = map;
 973        stream->avail_in = mapsize;
 974        stream->next_out = buffer;
 975        stream->avail_out = bufsiz;
 976
 977        git_inflate_init(stream);
 978        return git_inflate(stream, 0);
 979}
 980
 981int unpack_sha1_header(git_zstream *stream,
 982                       unsigned char *map, unsigned long mapsize,
 983                       void *buffer, unsigned long bufsiz)
 984{
 985        int status = unpack_sha1_short_header(stream, map, mapsize,
 986                                              buffer, bufsiz);
 987
 988        if (status < Z_OK)
 989                return status;
 990
 991        /* Make sure we have the terminating NUL */
 992        if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
 993                return -1;
 994        return 0;
 995}
 996
 997static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
 998                                        unsigned long mapsize, void *buffer,
 999                                        unsigned long bufsiz, struct strbuf *header)
1000{
1001        int status;
1002
1003        status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1004        if (status < Z_OK)
1005                return -1;
1006
1007        /*
1008         * Check if entire header is unpacked in the first iteration.
1009         */
1010        if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1011                return 0;
1012
1013        /*
1014         * buffer[0..bufsiz] was not large enough.  Copy the partial
1015         * result out to header, and then append the result of further
1016         * reading the stream.
1017         */
1018        strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1019        stream->next_out = buffer;
1020        stream->avail_out = bufsiz;
1021
1022        do {
1023                status = git_inflate(stream, 0);
1024                strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1025                if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1026                        return 0;
1027                stream->next_out = buffer;
1028                stream->avail_out = bufsiz;
1029        } while (status != Z_STREAM_END);
1030        return -1;
1031}
1032
1033static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1034{
1035        int bytes = strlen(buffer) + 1;
1036        unsigned char *buf = xmallocz(size);
1037        unsigned long n;
1038        int status = Z_OK;
1039
1040        n = stream->total_out - bytes;
1041        if (n > size)
1042                n = size;
1043        memcpy(buf, (char *) buffer + bytes, n);
1044        bytes = n;
1045        if (bytes <= size) {
1046                /*
1047                 * The above condition must be (bytes <= size), not
1048                 * (bytes < size).  In other words, even though we
1049                 * expect no more output and set avail_out to zero,
1050                 * the input zlib stream may have bytes that express
1051                 * "this concludes the stream", and we *do* want to
1052                 * eat that input.
1053                 *
1054                 * Otherwise we would not be able to test that we
1055                 * consumed all the input to reach the expected size;
1056                 * we also want to check that zlib tells us that all
1057                 * went well with status == Z_STREAM_END at the end.
1058                 */
1059                stream->next_out = buf + bytes;
1060                stream->avail_out = size - bytes;
1061                while (status == Z_OK)
1062                        status = git_inflate(stream, Z_FINISH);
1063        }
1064        if (status == Z_STREAM_END && !stream->avail_in) {
1065                git_inflate_end(stream);
1066                return buf;
1067        }
1068
1069        if (status < 0)
1070                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1071        else if (stream->avail_in)
1072                error("garbage at end of loose object '%s'",
1073                      sha1_to_hex(sha1));
1074        free(buf);
1075        return NULL;
1076}
1077
1078/*
1079 * We used to just use "sscanf()", but that's actually way
1080 * too permissive for what we want to check. So do an anal
1081 * object header parse by hand.
1082 */
1083static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1084                               unsigned int flags)
1085{
1086        const char *type_buf = hdr;
1087        unsigned long size;
1088        int type, type_len = 0;
1089
1090        /*
1091         * The type can be of any size but is followed by
1092         * a space.
1093         */
1094        for (;;) {
1095                char c = *hdr++;
1096                if (!c)
1097                        return -1;
1098                if (c == ' ')
1099                        break;
1100                type_len++;
1101        }
1102
1103        type = type_from_string_gently(type_buf, type_len, 1);
1104        if (oi->typename)
1105                strbuf_add(oi->typename, type_buf, type_len);
1106        /*
1107         * Set type to 0 if its an unknown object and
1108         * we're obtaining the type using '--allow-unknown-type'
1109         * option.
1110         */
1111        if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1112                type = 0;
1113        else if (type < 0)
1114                die("invalid object type");
1115        if (oi->typep)
1116                *oi->typep = type;
1117
1118        /*
1119         * The length must follow immediately, and be in canonical
1120         * decimal format (ie "010" is not valid).
1121         */
1122        size = *hdr++ - '0';
1123        if (size > 9)
1124                return -1;
1125        if (size) {
1126                for (;;) {
1127                        unsigned long c = *hdr - '0';
1128                        if (c > 9)
1129                                break;
1130                        hdr++;
1131                        size = size * 10 + c;
1132                }
1133        }
1134
1135        if (oi->sizep)
1136                *oi->sizep = size;
1137
1138        /*
1139         * The length must be followed by a zero byte
1140         */
1141        return *hdr ? -1 : type;
1142}
1143
1144int parse_sha1_header(const char *hdr, unsigned long *sizep)
1145{
1146        struct object_info oi = OBJECT_INFO_INIT;
1147
1148        oi.sizep = sizep;
1149        return parse_sha1_header_extended(hdr, &oi, 0);
1150}
1151
1152#define sha1_loose_object_info(r, s, o, f) sha1_loose_object_info_##r(s, o, f)
1153static int sha1_loose_object_info_the_repository(const unsigned char *sha1,
1154                                                 struct object_info *oi,
1155                                                 int flags)
1156{
1157        int status = 0;
1158        unsigned long mapsize;
1159        void *map;
1160        git_zstream stream;
1161        char hdr[32];
1162        struct strbuf hdrbuf = STRBUF_INIT;
1163        unsigned long size_scratch;
1164
1165        if (oi->delta_base_sha1)
1166                hashclr(oi->delta_base_sha1);
1167
1168        /*
1169         * If we don't care about type or size, then we don't
1170         * need to look inside the object at all. Note that we
1171         * do not optimize out the stat call, even if the
1172         * caller doesn't care about the disk-size, since our
1173         * return value implicitly indicates whether the
1174         * object even exists.
1175         */
1176        if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1177                const char *path;
1178                struct stat st;
1179                if (stat_sha1_file(the_repository, sha1, &st, &path) < 0)
1180                        return -1;
1181                if (oi->disk_sizep)
1182                        *oi->disk_sizep = st.st_size;
1183                return 0;
1184        }
1185
1186        map = map_sha1_file(the_repository, sha1, &mapsize);
1187        if (!map)
1188                return -1;
1189
1190        if (!oi->sizep)
1191                oi->sizep = &size_scratch;
1192
1193        if (oi->disk_sizep)
1194                *oi->disk_sizep = mapsize;
1195        if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1196                if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1197                        status = error("unable to unpack %s header with --allow-unknown-type",
1198                                       sha1_to_hex(sha1));
1199        } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1200                status = error("unable to unpack %s header",
1201                               sha1_to_hex(sha1));
1202        if (status < 0)
1203                ; /* Do nothing */
1204        else if (hdrbuf.len) {
1205                if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1206                        status = error("unable to parse %s header with --allow-unknown-type",
1207                                       sha1_to_hex(sha1));
1208        } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1209                status = error("unable to parse %s header", sha1_to_hex(sha1));
1210
1211        if (status >= 0 && oi->contentp) {
1212                *oi->contentp = unpack_sha1_rest(&stream, hdr,
1213                                                 *oi->sizep, sha1);
1214                if (!*oi->contentp) {
1215                        git_inflate_end(&stream);
1216                        status = -1;
1217                }
1218        } else
1219                git_inflate_end(&stream);
1220
1221        munmap(map, mapsize);
1222        if (status && oi->typep)
1223                *oi->typep = status;
1224        if (oi->sizep == &size_scratch)
1225                oi->sizep = NULL;
1226        strbuf_release(&hdrbuf);
1227        oi->whence = OI_LOOSE;
1228        return (status < 0) ? status : 0;
1229}
1230
1231int fetch_if_missing = 1;
1232
1233int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1234{
1235        static struct object_info blank_oi = OBJECT_INFO_INIT;
1236        struct pack_entry e;
1237        int rtype;
1238        const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1239                                    lookup_replace_object(sha1) :
1240                                    sha1;
1241        int already_retried = 0;
1242
1243        if (is_null_sha1(real))
1244                return -1;
1245
1246        if (!oi)
1247                oi = &blank_oi;
1248
1249        if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1250                struct cached_object *co = find_cached_object(real);
1251                if (co) {
1252                        if (oi->typep)
1253                                *(oi->typep) = co->type;
1254                        if (oi->sizep)
1255                                *(oi->sizep) = co->size;
1256                        if (oi->disk_sizep)
1257                                *(oi->disk_sizep) = 0;
1258                        if (oi->delta_base_sha1)
1259                                hashclr(oi->delta_base_sha1);
1260                        if (oi->typename)
1261                                strbuf_addstr(oi->typename, typename(co->type));
1262                        if (oi->contentp)
1263                                *oi->contentp = xmemdupz(co->buf, co->size);
1264                        oi->whence = OI_CACHED;
1265                        return 0;
1266                }
1267        }
1268
1269        while (1) {
1270                if (find_pack_entry(real, &e))
1271                        break;
1272
1273                /* Most likely it's a loose object. */
1274                if (!sha1_loose_object_info(the_repository, real, oi, flags))
1275                        return 0;
1276
1277                /* Not a loose object; someone else may have just packed it. */
1278                reprepare_packed_git();
1279                if (find_pack_entry(real, &e))
1280                        break;
1281
1282                /* Check if it is a missing object */
1283                if (fetch_if_missing && repository_format_partial_clone &&
1284                    !already_retried) {
1285                        /*
1286                         * TODO Investigate haveing fetch_object() return
1287                         * TODO error/success and stopping the music here.
1288                         */
1289                        fetch_object(repository_format_partial_clone, real);
1290                        already_retried = 1;
1291                        continue;
1292                }
1293
1294                return -1;
1295        }
1296
1297        if (oi == &blank_oi)
1298                /*
1299                 * We know that the caller doesn't actually need the
1300                 * information below, so return early.
1301                 */
1302                return 0;
1303        rtype = packed_object_info(e.p, e.offset, oi);
1304        if (rtype < 0) {
1305                mark_bad_packed_object(e.p, real);
1306                return sha1_object_info_extended(real, oi, 0);
1307        } else if (oi->whence == OI_PACKED) {
1308                oi->u.packed.offset = e.offset;
1309                oi->u.packed.pack = e.p;
1310                oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1311                                         rtype == OBJ_OFS_DELTA);
1312        }
1313
1314        return 0;
1315}
1316
1317/* returns enum object_type or negative */
1318int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1319{
1320        enum object_type type;
1321        struct object_info oi = OBJECT_INFO_INIT;
1322
1323        oi.typep = &type;
1324        oi.sizep = sizep;
1325        if (sha1_object_info_extended(sha1, &oi,
1326                                      OBJECT_INFO_LOOKUP_REPLACE) < 0)
1327                return -1;
1328        return type;
1329}
1330
1331static void *read_object(const unsigned char *sha1, enum object_type *type,
1332                         unsigned long *size)
1333{
1334        struct object_info oi = OBJECT_INFO_INIT;
1335        void *content;
1336        oi.typep = type;
1337        oi.sizep = size;
1338        oi.contentp = &content;
1339
1340        if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1341                return NULL;
1342        return content;
1343}
1344
1345int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1346                      unsigned char *sha1)
1347{
1348        struct cached_object *co;
1349
1350        hash_sha1_file(buf, len, typename(type), sha1);
1351        if (has_sha1_file(sha1) || find_cached_object(sha1))
1352                return 0;
1353        ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1354        co = &cached_objects[cached_object_nr++];
1355        co->size = len;
1356        co->type = type;
1357        co->buf = xmalloc(len);
1358        memcpy(co->buf, buf, len);
1359        hashcpy(co->sha1, sha1);
1360        return 0;
1361}
1362
1363/*
1364 * This function dies on corrupt objects; the callers who want to
1365 * deal with them should arrange to call read_object() and give error
1366 * messages themselves.
1367 */
1368void *read_sha1_file_extended(const unsigned char *sha1,
1369                              enum object_type *type,
1370                              unsigned long *size,
1371                              int lookup_replace)
1372{
1373        void *data;
1374        const struct packed_git *p;
1375        const char *path;
1376        struct stat st;
1377        const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1378                                                   : sha1;
1379
1380        errno = 0;
1381        data = read_object(repl, type, size);
1382        if (data)
1383                return data;
1384
1385        if (errno && errno != ENOENT)
1386                die_errno("failed to read object %s", sha1_to_hex(sha1));
1387
1388        /* die if we replaced an object with one that does not exist */
1389        if (repl != sha1)
1390                die("replacement %s not found for %s",
1391                    sha1_to_hex(repl), sha1_to_hex(sha1));
1392
1393        if (!stat_sha1_file(the_repository, repl, &st, &path))
1394                die("loose object %s (stored in %s) is corrupt",
1395                    sha1_to_hex(repl), path);
1396
1397        if ((p = has_packed_and_bad(repl)) != NULL)
1398                die("packed object %s (stored in %s) is corrupt",
1399                    sha1_to_hex(repl), p->pack_name);
1400
1401        return NULL;
1402}
1403
1404void *read_object_with_reference(const unsigned char *sha1,
1405                                 const char *required_type_name,
1406                                 unsigned long *size,
1407                                 unsigned char *actual_sha1_return)
1408{
1409        enum object_type type, required_type;
1410        void *buffer;
1411        unsigned long isize;
1412        unsigned char actual_sha1[20];
1413
1414        required_type = type_from_string(required_type_name);
1415        hashcpy(actual_sha1, sha1);
1416        while (1) {
1417                int ref_length = -1;
1418                const char *ref_type = NULL;
1419
1420                buffer = read_sha1_file(actual_sha1, &type, &isize);
1421                if (!buffer)
1422                        return NULL;
1423                if (type == required_type) {
1424                        *size = isize;
1425                        if (actual_sha1_return)
1426                                hashcpy(actual_sha1_return, actual_sha1);
1427                        return buffer;
1428                }
1429                /* Handle references */
1430                else if (type == OBJ_COMMIT)
1431                        ref_type = "tree ";
1432                else if (type == OBJ_TAG)
1433                        ref_type = "object ";
1434                else {
1435                        free(buffer);
1436                        return NULL;
1437                }
1438                ref_length = strlen(ref_type);
1439
1440                if (ref_length + 40 > isize ||
1441                    memcmp(buffer, ref_type, ref_length) ||
1442                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1443                        free(buffer);
1444                        return NULL;
1445                }
1446                free(buffer);
1447                /* Now we have the ID of the referred-to object in
1448                 * actual_sha1.  Check again. */
1449        }
1450}
1451
1452static void write_sha1_file_prepare(const void *buf, unsigned long len,
1453                                    const char *type, unsigned char *sha1,
1454                                    char *hdr, int *hdrlen)
1455{
1456        git_SHA_CTX c;
1457
1458        /* Generate the header */
1459        *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1460
1461        /* Sha1.. */
1462        git_SHA1_Init(&c);
1463        git_SHA1_Update(&c, hdr, *hdrlen);
1464        git_SHA1_Update(&c, buf, len);
1465        git_SHA1_Final(sha1, &c);
1466}
1467
1468/*
1469 * Move the just written object into its final resting place.
1470 */
1471int finalize_object_file(const char *tmpfile, const char *filename)
1472{
1473        int ret = 0;
1474
1475        if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1476                goto try_rename;
1477        else if (link(tmpfile, filename))
1478                ret = errno;
1479
1480        /*
1481         * Coda hack - coda doesn't like cross-directory links,
1482         * so we fall back to a rename, which will mean that it
1483         * won't be able to check collisions, but that's not a
1484         * big deal.
1485         *
1486         * The same holds for FAT formatted media.
1487         *
1488         * When this succeeds, we just return.  We have nothing
1489         * left to unlink.
1490         */
1491        if (ret && ret != EEXIST) {
1492        try_rename:
1493                if (!rename(tmpfile, filename))
1494                        goto out;
1495                ret = errno;
1496        }
1497        unlink_or_warn(tmpfile);
1498        if (ret) {
1499                if (ret != EEXIST) {
1500                        return error_errno("unable to write sha1 filename %s", filename);
1501                }
1502                /* FIXME!!! Collision check here ? */
1503        }
1504
1505out:
1506        if (adjust_shared_perm(filename))
1507                return error("unable to set permission to '%s'", filename);
1508        return 0;
1509}
1510
1511static int write_buffer(int fd, const void *buf, size_t len)
1512{
1513        if (write_in_full(fd, buf, len) < 0)
1514                return error_errno("file write error");
1515        return 0;
1516}
1517
1518int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1519                   unsigned char *sha1)
1520{
1521        char hdr[32];
1522        int hdrlen = sizeof(hdr);
1523        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1524        return 0;
1525}
1526
1527/* Finalize a file on disk, and close it. */
1528static void close_sha1_file(int fd)
1529{
1530        if (fsync_object_files)
1531                fsync_or_die(fd, "sha1 file");
1532        if (close(fd) != 0)
1533                die_errno("error when closing sha1 file");
1534}
1535
1536/* Size of directory component, including the ending '/' */
1537static inline int directory_size(const char *filename)
1538{
1539        const char *s = strrchr(filename, '/');
1540        if (!s)
1541                return 0;
1542        return s - filename + 1;
1543}
1544
1545/*
1546 * This creates a temporary file in the same directory as the final
1547 * 'filename'
1548 *
1549 * We want to avoid cross-directory filename renames, because those
1550 * can have problems on various filesystems (FAT, NFS, Coda).
1551 */
1552static int create_tmpfile(struct strbuf *tmp, const char *filename)
1553{
1554        int fd, dirlen = directory_size(filename);
1555
1556        strbuf_reset(tmp);
1557        strbuf_add(tmp, filename, dirlen);
1558        strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1559        fd = git_mkstemp_mode(tmp->buf, 0444);
1560        if (fd < 0 && dirlen && errno == ENOENT) {
1561                /*
1562                 * Make sure the directory exists; note that the contents
1563                 * of the buffer are undefined after mkstemp returns an
1564                 * error, so we have to rewrite the whole buffer from
1565                 * scratch.
1566                 */
1567                strbuf_reset(tmp);
1568                strbuf_add(tmp, filename, dirlen - 1);
1569                if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1570                        return -1;
1571                if (adjust_shared_perm(tmp->buf))
1572                        return -1;
1573
1574                /* Try again */
1575                strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1576                fd = git_mkstemp_mode(tmp->buf, 0444);
1577        }
1578        return fd;
1579}
1580
1581static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1582                              const void *buf, unsigned long len, time_t mtime)
1583{
1584        int fd, ret;
1585        unsigned char compressed[4096];
1586        git_zstream stream;
1587        git_SHA_CTX c;
1588        unsigned char parano_sha1[20];
1589        static struct strbuf tmp_file = STRBUF_INIT;
1590        static struct strbuf filename = STRBUF_INIT;
1591
1592        strbuf_reset(&filename);
1593        sha1_file_name(the_repository, &filename, sha1);
1594
1595        fd = create_tmpfile(&tmp_file, filename.buf);
1596        if (fd < 0) {
1597                if (errno == EACCES)
1598                        return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1599                else
1600                        return error_errno("unable to create temporary file");
1601        }
1602
1603        /* Set it up */
1604        git_deflate_init(&stream, zlib_compression_level);
1605        stream.next_out = compressed;
1606        stream.avail_out = sizeof(compressed);
1607        git_SHA1_Init(&c);
1608
1609        /* First header.. */
1610        stream.next_in = (unsigned char *)hdr;
1611        stream.avail_in = hdrlen;
1612        while (git_deflate(&stream, 0) == Z_OK)
1613                ; /* nothing */
1614        git_SHA1_Update(&c, hdr, hdrlen);
1615
1616        /* Then the data itself.. */
1617        stream.next_in = (void *)buf;
1618        stream.avail_in = len;
1619        do {
1620                unsigned char *in0 = stream.next_in;
1621                ret = git_deflate(&stream, Z_FINISH);
1622                git_SHA1_Update(&c, in0, stream.next_in - in0);
1623                if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1624                        die("unable to write sha1 file");
1625                stream.next_out = compressed;
1626                stream.avail_out = sizeof(compressed);
1627        } while (ret == Z_OK);
1628
1629        if (ret != Z_STREAM_END)
1630                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1631        ret = git_deflate_end_gently(&stream);
1632        if (ret != Z_OK)
1633                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1634        git_SHA1_Final(parano_sha1, &c);
1635        if (hashcmp(sha1, parano_sha1) != 0)
1636                die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1637
1638        close_sha1_file(fd);
1639
1640        if (mtime) {
1641                struct utimbuf utb;
1642                utb.actime = mtime;
1643                utb.modtime = mtime;
1644                if (utime(tmp_file.buf, &utb) < 0)
1645                        warning_errno("failed utime() on %s", tmp_file.buf);
1646        }
1647
1648        return finalize_object_file(tmp_file.buf, filename.buf);
1649}
1650
1651static int freshen_loose_object(const unsigned char *sha1)
1652{
1653        return check_and_freshen(sha1, 1);
1654}
1655
1656static int freshen_packed_object(const unsigned char *sha1)
1657{
1658        struct pack_entry e;
1659        if (!find_pack_entry(sha1, &e))
1660                return 0;
1661        if (e.p->freshened)
1662                return 1;
1663        if (!freshen_file(e.p->pack_name))
1664                return 0;
1665        e.p->freshened = 1;
1666        return 1;
1667}
1668
1669int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1670{
1671        char hdr[32];
1672        int hdrlen = sizeof(hdr);
1673
1674        /* Normally if we have it in the pack then we do not bother writing
1675         * it out into .git/objects/??/?{38} file.
1676         */
1677        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1678        if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1679                return 0;
1680        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1681}
1682
1683int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1684                             struct object_id *oid, unsigned flags)
1685{
1686        char *header;
1687        int hdrlen, status = 0;
1688
1689        /* type string, SP, %lu of the length plus NUL must fit this */
1690        hdrlen = strlen(type) + 32;
1691        header = xmalloc(hdrlen);
1692        write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1693
1694        if (!(flags & HASH_WRITE_OBJECT))
1695                goto cleanup;
1696        if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1697                goto cleanup;
1698        status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1699
1700cleanup:
1701        free(header);
1702        return status;
1703}
1704
1705int force_object_loose(const unsigned char *sha1, time_t mtime)
1706{
1707        void *buf;
1708        unsigned long len;
1709        enum object_type type;
1710        char hdr[32];
1711        int hdrlen;
1712        int ret;
1713
1714        if (has_loose_object(sha1))
1715                return 0;
1716        buf = read_object(sha1, &type, &len);
1717        if (!buf)
1718                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1719        hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1720        ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1721        free(buf);
1722
1723        return ret;
1724}
1725
1726int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1727{
1728        if (!startup_info->have_repository)
1729                return 0;
1730        return sha1_object_info_extended(sha1, NULL,
1731                                         flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1732}
1733
1734int has_object_file(const struct object_id *oid)
1735{
1736        return has_sha1_file(oid->hash);
1737}
1738
1739int has_object_file_with_flags(const struct object_id *oid, int flags)
1740{
1741        return has_sha1_file_with_flags(oid->hash, flags);
1742}
1743
1744static void check_tree(const void *buf, size_t size)
1745{
1746        struct tree_desc desc;
1747        struct name_entry entry;
1748
1749        init_tree_desc(&desc, buf, size);
1750        while (tree_entry(&desc, &entry))
1751                /* do nothing
1752                 * tree_entry() will die() on malformed entries */
1753                ;
1754}
1755
1756static void check_commit(const void *buf, size_t size)
1757{
1758        struct commit c;
1759        memset(&c, 0, sizeof(c));
1760        if (parse_commit_buffer(&c, buf, size))
1761                die("corrupt commit");
1762}
1763
1764static void check_tag(const void *buf, size_t size)
1765{
1766        struct tag t;
1767        memset(&t, 0, sizeof(t));
1768        if (parse_tag_buffer(&t, buf, size))
1769                die("corrupt tag");
1770}
1771
1772static int index_mem(struct object_id *oid, void *buf, size_t size,
1773                     enum object_type type,
1774                     const char *path, unsigned flags)
1775{
1776        int ret, re_allocated = 0;
1777        int write_object = flags & HASH_WRITE_OBJECT;
1778
1779        if (!type)
1780                type = OBJ_BLOB;
1781
1782        /*
1783         * Convert blobs to git internal format
1784         */
1785        if ((type == OBJ_BLOB) && path) {
1786                struct strbuf nbuf = STRBUF_INIT;
1787                if (convert_to_git(&the_index, path, buf, size, &nbuf,
1788                                   get_conv_flags(flags))) {
1789                        buf = strbuf_detach(&nbuf, &size);
1790                        re_allocated = 1;
1791                }
1792        }
1793        if (flags & HASH_FORMAT_CHECK) {
1794                if (type == OBJ_TREE)
1795                        check_tree(buf, size);
1796                if (type == OBJ_COMMIT)
1797                        check_commit(buf, size);
1798                if (type == OBJ_TAG)
1799                        check_tag(buf, size);
1800        }
1801
1802        if (write_object)
1803                ret = write_sha1_file(buf, size, typename(type), oid->hash);
1804        else
1805                ret = hash_sha1_file(buf, size, typename(type), oid->hash);
1806        if (re_allocated)
1807                free(buf);
1808        return ret;
1809}
1810
1811static int index_stream_convert_blob(struct object_id *oid, int fd,
1812                                     const char *path, unsigned flags)
1813{
1814        int ret;
1815        const int write_object = flags & HASH_WRITE_OBJECT;
1816        struct strbuf sbuf = STRBUF_INIT;
1817
1818        assert(path);
1819        assert(would_convert_to_git_filter_fd(path));
1820
1821        convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1822                                 get_conv_flags(flags));
1823
1824        if (write_object)
1825                ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1826                                      oid->hash);
1827        else
1828                ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1829                                     oid->hash);
1830        strbuf_release(&sbuf);
1831        return ret;
1832}
1833
1834static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1835                      const char *path, unsigned flags)
1836{
1837        struct strbuf sbuf = STRBUF_INIT;
1838        int ret;
1839
1840        if (strbuf_read(&sbuf, fd, 4096) >= 0)
1841                ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1842        else
1843                ret = -1;
1844        strbuf_release(&sbuf);
1845        return ret;
1846}
1847
1848#define SMALL_FILE_SIZE (32*1024)
1849
1850static int index_core(struct object_id *oid, int fd, size_t size,
1851                      enum object_type type, const char *path,
1852                      unsigned flags)
1853{
1854        int ret;
1855
1856        if (!size) {
1857                ret = index_mem(oid, "", size, type, path, flags);
1858        } else if (size <= SMALL_FILE_SIZE) {
1859                char *buf = xmalloc(size);
1860                ssize_t read_result = read_in_full(fd, buf, size);
1861                if (read_result < 0)
1862                        ret = error_errno("read error while indexing %s",
1863                                          path ? path : "<unknown>");
1864                else if (read_result != size)
1865                        ret = error("short read while indexing %s",
1866                                    path ? path : "<unknown>");
1867                else
1868                        ret = index_mem(oid, buf, size, type, path, flags);
1869                free(buf);
1870        } else {
1871                void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1872                ret = index_mem(oid, buf, size, type, path, flags);
1873                munmap(buf, size);
1874        }
1875        return ret;
1876}
1877
1878/*
1879 * This creates one packfile per large blob unless bulk-checkin
1880 * machinery is "plugged".
1881 *
1882 * This also bypasses the usual "convert-to-git" dance, and that is on
1883 * purpose. We could write a streaming version of the converting
1884 * functions and insert that before feeding the data to fast-import
1885 * (or equivalent in-core API described above). However, that is
1886 * somewhat complicated, as we do not know the size of the filter
1887 * result, which we need to know beforehand when writing a git object.
1888 * Since the primary motivation for trying to stream from the working
1889 * tree file and to avoid mmaping it in core is to deal with large
1890 * binary blobs, they generally do not want to get any conversion, and
1891 * callers should avoid this code path when filters are requested.
1892 */
1893static int index_stream(struct object_id *oid, int fd, size_t size,
1894                        enum object_type type, const char *path,
1895                        unsigned flags)
1896{
1897        return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1898}
1899
1900int index_fd(struct object_id *oid, int fd, struct stat *st,
1901             enum object_type type, const char *path, unsigned flags)
1902{
1903        int ret;
1904
1905        /*
1906         * Call xsize_t() only when needed to avoid potentially unnecessary
1907         * die() for large files.
1908         */
1909        if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1910                ret = index_stream_convert_blob(oid, fd, path, flags);
1911        else if (!S_ISREG(st->st_mode))
1912                ret = index_pipe(oid, fd, type, path, flags);
1913        else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1914                 (path && would_convert_to_git(&the_index, path)))
1915                ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1916                                 flags);
1917        else
1918                ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1919                                   flags);
1920        close(fd);
1921        return ret;
1922}
1923
1924int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1925{
1926        int fd;
1927        struct strbuf sb = STRBUF_INIT;
1928        int rc = 0;
1929
1930        switch (st->st_mode & S_IFMT) {
1931        case S_IFREG:
1932                fd = open(path, O_RDONLY);
1933                if (fd < 0)
1934                        return error_errno("open(\"%s\")", path);
1935                if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1936                        return error("%s: failed to insert into database",
1937                                     path);
1938                break;
1939        case S_IFLNK:
1940                if (strbuf_readlink(&sb, path, st->st_size))
1941                        return error_errno("readlink(\"%s\")", path);
1942                if (!(flags & HASH_WRITE_OBJECT))
1943                        hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1944                else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1945                        rc = error("%s: failed to insert into database", path);
1946                strbuf_release(&sb);
1947                break;
1948        case S_IFDIR:
1949                return resolve_gitlink_ref(path, "HEAD", oid);
1950        default:
1951                return error("%s: unsupported file type", path);
1952        }
1953        return rc;
1954}
1955
1956int read_pack_header(int fd, struct pack_header *header)
1957{
1958        if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1959                /* "eof before pack header was fully read" */
1960                return PH_ERROR_EOF;
1961
1962        if (header->hdr_signature != htonl(PACK_SIGNATURE))
1963                /* "protocol error (pack signature mismatch detected)" */
1964                return PH_ERROR_PACK_SIGNATURE;
1965        if (!pack_version_ok(header->hdr_version))
1966                /* "protocol error (pack version unsupported)" */
1967                return PH_ERROR_PROTOCOL;
1968        return 0;
1969}
1970
1971void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1972{
1973        enum object_type type = sha1_object_info(sha1, NULL);
1974        if (type < 0)
1975                die("%s is not a valid object", sha1_to_hex(sha1));
1976        if (type != expect)
1977                die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1978                    typename(expect));
1979}
1980
1981int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1982                                struct strbuf *path,
1983                                each_loose_object_fn obj_cb,
1984                                each_loose_cruft_fn cruft_cb,
1985                                each_loose_subdir_fn subdir_cb,
1986                                void *data)
1987{
1988        size_t origlen, baselen;
1989        DIR *dir;
1990        struct dirent *de;
1991        int r = 0;
1992        struct object_id oid;
1993
1994        if (subdir_nr > 0xff)
1995                BUG("invalid loose object subdirectory: %x", subdir_nr);
1996
1997        origlen = path->len;
1998        strbuf_complete(path, '/');
1999        strbuf_addf(path, "%02x", subdir_nr);
2000
2001        dir = opendir(path->buf);
2002        if (!dir) {
2003                if (errno != ENOENT)
2004                        r = error_errno("unable to open %s", path->buf);
2005                strbuf_setlen(path, origlen);
2006                return r;
2007        }
2008
2009        oid.hash[0] = subdir_nr;
2010        strbuf_addch(path, '/');
2011        baselen = path->len;
2012
2013        while ((de = readdir(dir))) {
2014                size_t namelen;
2015                if (is_dot_or_dotdot(de->d_name))
2016                        continue;
2017
2018                namelen = strlen(de->d_name);
2019                strbuf_setlen(path, baselen);
2020                strbuf_add(path, de->d_name, namelen);
2021                if (namelen == GIT_SHA1_HEXSZ - 2 &&
2022                    !hex_to_bytes(oid.hash + 1, de->d_name,
2023                                  GIT_SHA1_RAWSZ - 1)) {
2024                        if (obj_cb) {
2025                                r = obj_cb(&oid, path->buf, data);
2026                                if (r)
2027                                        break;
2028                        }
2029                        continue;
2030                }
2031
2032                if (cruft_cb) {
2033                        r = cruft_cb(de->d_name, path->buf, data);
2034                        if (r)
2035                                break;
2036                }
2037        }
2038        closedir(dir);
2039
2040        strbuf_setlen(path, baselen - 1);
2041        if (!r && subdir_cb)
2042                r = subdir_cb(subdir_nr, path->buf, data);
2043
2044        strbuf_setlen(path, origlen);
2045
2046        return r;
2047}
2048
2049int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2050                            each_loose_object_fn obj_cb,
2051                            each_loose_cruft_fn cruft_cb,
2052                            each_loose_subdir_fn subdir_cb,
2053                            void *data)
2054{
2055        int r = 0;
2056        int i;
2057
2058        for (i = 0; i < 256; i++) {
2059                r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2060                                                subdir_cb, data);
2061                if (r)
2062                        break;
2063        }
2064
2065        return r;
2066}
2067
2068int for_each_loose_file_in_objdir(const char *path,
2069                                  each_loose_object_fn obj_cb,
2070                                  each_loose_cruft_fn cruft_cb,
2071                                  each_loose_subdir_fn subdir_cb,
2072                                  void *data)
2073{
2074        struct strbuf buf = STRBUF_INIT;
2075        int r;
2076
2077        strbuf_addstr(&buf, path);
2078        r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2079                                              subdir_cb, data);
2080        strbuf_release(&buf);
2081
2082        return r;
2083}
2084
2085struct loose_alt_odb_data {
2086        each_loose_object_fn *cb;
2087        void *data;
2088};
2089
2090static int loose_from_alt_odb(struct alternate_object_database *alt,
2091                              void *vdata)
2092{
2093        struct loose_alt_odb_data *data = vdata;
2094        struct strbuf buf = STRBUF_INIT;
2095        int r;
2096
2097        strbuf_addstr(&buf, alt->path);
2098        r = for_each_loose_file_in_objdir_buf(&buf,
2099                                              data->cb, NULL, NULL,
2100                                              data->data);
2101        strbuf_release(&buf);
2102        return r;
2103}
2104
2105int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2106{
2107        struct loose_alt_odb_data alt;
2108        int r;
2109
2110        r = for_each_loose_file_in_objdir(get_object_directory(),
2111                                          cb, NULL, NULL, data);
2112        if (r)
2113                return r;
2114
2115        if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2116                return 0;
2117
2118        alt.cb = cb;
2119        alt.data = data;
2120        return foreach_alt_odb(loose_from_alt_odb, &alt);
2121}
2122
2123static int check_stream_sha1(git_zstream *stream,
2124                             const char *hdr,
2125                             unsigned long size,
2126                             const char *path,
2127                             const unsigned char *expected_sha1)
2128{
2129        git_SHA_CTX c;
2130        unsigned char real_sha1[GIT_MAX_RAWSZ];
2131        unsigned char buf[4096];
2132        unsigned long total_read;
2133        int status = Z_OK;
2134
2135        git_SHA1_Init(&c);
2136        git_SHA1_Update(&c, hdr, stream->total_out);
2137
2138        /*
2139         * We already read some bytes into hdr, but the ones up to the NUL
2140         * do not count against the object's content size.
2141         */
2142        total_read = stream->total_out - strlen(hdr) - 1;
2143
2144        /*
2145         * This size comparison must be "<=" to read the final zlib packets;
2146         * see the comment in unpack_sha1_rest for details.
2147         */
2148        while (total_read <= size &&
2149               (status == Z_OK || status == Z_BUF_ERROR)) {
2150                stream->next_out = buf;
2151                stream->avail_out = sizeof(buf);
2152                if (size - total_read < stream->avail_out)
2153                        stream->avail_out = size - total_read;
2154                status = git_inflate(stream, Z_FINISH);
2155                git_SHA1_Update(&c, buf, stream->next_out - buf);
2156                total_read += stream->next_out - buf;
2157        }
2158        git_inflate_end(stream);
2159
2160        if (status != Z_STREAM_END) {
2161                error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2162                return -1;
2163        }
2164        if (stream->avail_in) {
2165                error("garbage at end of loose object '%s'",
2166                      sha1_to_hex(expected_sha1));
2167                return -1;
2168        }
2169
2170        git_SHA1_Final(real_sha1, &c);
2171        if (hashcmp(expected_sha1, real_sha1)) {
2172                error("sha1 mismatch for %s (expected %s)", path,
2173                      sha1_to_hex(expected_sha1));
2174                return -1;
2175        }
2176
2177        return 0;
2178}
2179
2180int read_loose_object(const char *path,
2181                      const unsigned char *expected_sha1,
2182                      enum object_type *type,
2183                      unsigned long *size,
2184                      void **contents)
2185{
2186        int ret = -1;
2187        void *map = NULL;
2188        unsigned long mapsize;
2189        git_zstream stream;
2190        char hdr[32];
2191
2192        *contents = NULL;
2193
2194        map = map_sha1_file_1(the_repository, path, NULL, &mapsize);
2195        if (!map) {
2196                error_errno("unable to mmap %s", path);
2197                goto out;
2198        }
2199
2200        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2201                error("unable to unpack header of %s", path);
2202                goto out;
2203        }
2204
2205        *type = parse_sha1_header(hdr, size);
2206        if (*type < 0) {
2207                error("unable to parse header of %s", path);
2208                git_inflate_end(&stream);
2209                goto out;
2210        }
2211
2212        if (*type == OBJ_BLOB) {
2213                if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2214                        goto out;
2215        } else {
2216                *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2217                if (!*contents) {
2218                        error("unable to unpack contents of %s", path);
2219                        git_inflate_end(&stream);
2220                        goto out;
2221                }
2222                if (check_sha1_signature(expected_sha1, *contents,
2223                                         *size, typename(*type))) {
2224                        error("sha1 mismatch for %s (expected %s)", path,
2225                              sha1_to_hex(expected_sha1));
2226                        free(*contents);
2227                        goto out;
2228                }
2229        }
2230
2231        ret = 0; /* everything checks out */
2232
2233out:
2234        if (map)
2235                munmap(map, mapsize);
2236        return ret;
2237}