289eff29154907dd3cd054f0986c4331d4596083
   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 "streaming.h"
  26#include "dir.h"
  27#include "mru.h"
  28#include "list.h"
  29#include "mergesort.h"
  30#include "quote.h"
  31#include "packfile.h"
  32
  33const unsigned char null_sha1[20];
  34const struct object_id null_oid;
  35const struct object_id empty_tree_oid = {
  36        EMPTY_TREE_SHA1_BIN_LITERAL
  37};
  38const struct object_id empty_blob_oid = {
  39        EMPTY_BLOB_SHA1_BIN_LITERAL
  40};
  41
  42/*
  43 * This is meant to hold a *small* number of objects that you would
  44 * want read_sha1_file() to be able to return, but yet you do not want
  45 * to write them into the object store (e.g. a browse-only
  46 * application).
  47 */
  48static struct cached_object {
  49        unsigned char sha1[20];
  50        enum object_type type;
  51        void *buf;
  52        unsigned long size;
  53} *cached_objects;
  54static int cached_object_nr, cached_object_alloc;
  55
  56static struct cached_object empty_tree = {
  57        EMPTY_TREE_SHA1_BIN_LITERAL,
  58        OBJ_TREE,
  59        "",
  60        0
  61};
  62
  63static struct cached_object *find_cached_object(const unsigned char *sha1)
  64{
  65        int i;
  66        struct cached_object *co = cached_objects;
  67
  68        for (i = 0; i < cached_object_nr; i++, co++) {
  69                if (!hashcmp(co->sha1, sha1))
  70                        return co;
  71        }
  72        if (!hashcmp(sha1, empty_tree.sha1))
  73                return &empty_tree;
  74        return NULL;
  75}
  76
  77int mkdir_in_gitdir(const char *path)
  78{
  79        if (mkdir(path, 0777)) {
  80                int saved_errno = errno;
  81                struct stat st;
  82                struct strbuf sb = STRBUF_INIT;
  83
  84                if (errno != EEXIST)
  85                        return -1;
  86                /*
  87                 * Are we looking at a path in a symlinked worktree
  88                 * whose original repository does not yet have it?
  89                 * e.g. .git/rr-cache pointing at its original
  90                 * repository in which the user hasn't performed any
  91                 * conflict resolution yet?
  92                 */
  93                if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
  94                    strbuf_readlink(&sb, path, st.st_size) ||
  95                    !is_absolute_path(sb.buf) ||
  96                    mkdir(sb.buf, 0777)) {
  97                        strbuf_release(&sb);
  98                        errno = saved_errno;
  99                        return -1;
 100                }
 101                strbuf_release(&sb);
 102        }
 103        return adjust_shared_perm(path);
 104}
 105
 106enum scld_error safe_create_leading_directories(char *path)
 107{
 108        char *next_component = path + offset_1st_component(path);
 109        enum scld_error ret = SCLD_OK;
 110
 111        while (ret == SCLD_OK && next_component) {
 112                struct stat st;
 113                char *slash = next_component, slash_character;
 114
 115                while (*slash && !is_dir_sep(*slash))
 116                        slash++;
 117
 118                if (!*slash)
 119                        break;
 120
 121                next_component = slash + 1;
 122                while (is_dir_sep(*next_component))
 123                        next_component++;
 124                if (!*next_component)
 125                        break;
 126
 127                slash_character = *slash;
 128                *slash = '\0';
 129                if (!stat(path, &st)) {
 130                        /* path exists */
 131                        if (!S_ISDIR(st.st_mode)) {
 132                                errno = ENOTDIR;
 133                                ret = SCLD_EXISTS;
 134                        }
 135                } else if (mkdir(path, 0777)) {
 136                        if (errno == EEXIST &&
 137                            !stat(path, &st) && S_ISDIR(st.st_mode))
 138                                ; /* somebody created it since we checked */
 139                        else if (errno == ENOENT)
 140                                /*
 141                                 * Either mkdir() failed because
 142                                 * somebody just pruned the containing
 143                                 * directory, or stat() failed because
 144                                 * the file that was in our way was
 145                                 * just removed.  Either way, inform
 146                                 * the caller that it might be worth
 147                                 * trying again:
 148                                 */
 149                                ret = SCLD_VANISHED;
 150                        else
 151                                ret = SCLD_FAILED;
 152                } else if (adjust_shared_perm(path)) {
 153                        ret = SCLD_PERMS;
 154                }
 155                *slash = slash_character;
 156        }
 157        return ret;
 158}
 159
 160enum scld_error safe_create_leading_directories_const(const char *path)
 161{
 162        int save_errno;
 163        /* path points to cache entries, so xstrdup before messing with it */
 164        char *buf = xstrdup(path);
 165        enum scld_error result = safe_create_leading_directories(buf);
 166
 167        save_errno = errno;
 168        free(buf);
 169        errno = save_errno;
 170        return result;
 171}
 172
 173int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
 174{
 175        /*
 176         * The number of times we will try to remove empty directories
 177         * in the way of path. This is only 1 because if another
 178         * process is racily creating directories that conflict with
 179         * us, we don't want to fight against them.
 180         */
 181        int remove_directories_remaining = 1;
 182
 183        /*
 184         * The number of times that we will try to create the
 185         * directories containing path. We are willing to attempt this
 186         * more than once, because another process could be trying to
 187         * clean up empty directories at the same time as we are
 188         * trying to create them.
 189         */
 190        int create_directories_remaining = 3;
 191
 192        /* A scratch copy of path, filled lazily if we need it: */
 193        struct strbuf path_copy = STRBUF_INIT;
 194
 195        int ret, save_errno;
 196
 197        /* Sanity check: */
 198        assert(*path);
 199
 200retry_fn:
 201        ret = fn(path, cb);
 202        save_errno = errno;
 203        if (!ret)
 204                goto out;
 205
 206        if (errno == EISDIR && remove_directories_remaining-- > 0) {
 207                /*
 208                 * A directory is in the way. Maybe it is empty; try
 209                 * to remove it:
 210                 */
 211                if (!path_copy.len)
 212                        strbuf_addstr(&path_copy, path);
 213
 214                if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
 215                        goto retry_fn;
 216        } else if (errno == ENOENT && create_directories_remaining-- > 0) {
 217                /*
 218                 * Maybe the containing directory didn't exist, or
 219                 * maybe it was just deleted by a process that is
 220                 * racing with us to clean up empty directories. Try
 221                 * to create it:
 222                 */
 223                enum scld_error scld_result;
 224
 225                if (!path_copy.len)
 226                        strbuf_addstr(&path_copy, path);
 227
 228                do {
 229                        scld_result = safe_create_leading_directories(path_copy.buf);
 230                        if (scld_result == SCLD_OK)
 231                                goto retry_fn;
 232                } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
 233        }
 234
 235out:
 236        strbuf_release(&path_copy);
 237        errno = save_errno;
 238        return ret;
 239}
 240
 241static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
 242{
 243        int i;
 244        for (i = 0; i < 20; i++) {
 245                static char hex[] = "0123456789abcdef";
 246                unsigned int val = sha1[i];
 247                strbuf_addch(buf, hex[val >> 4]);
 248                strbuf_addch(buf, hex[val & 0xf]);
 249                if (!i)
 250                        strbuf_addch(buf, '/');
 251        }
 252}
 253
 254const char *sha1_file_name(const unsigned char *sha1)
 255{
 256        static struct strbuf buf = STRBUF_INIT;
 257
 258        strbuf_reset(&buf);
 259        strbuf_addf(&buf, "%s/", get_object_directory());
 260
 261        fill_sha1_path(&buf, sha1);
 262        return buf.buf;
 263}
 264
 265struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
 266{
 267        strbuf_setlen(&alt->scratch, alt->base_len);
 268        return &alt->scratch;
 269}
 270
 271static const char *alt_sha1_path(struct alternate_object_database *alt,
 272                                 const unsigned char *sha1)
 273{
 274        struct strbuf *buf = alt_scratch_buf(alt);
 275        fill_sha1_path(buf, sha1);
 276        return buf->buf;
 277}
 278
 279struct alternate_object_database *alt_odb_list;
 280static struct alternate_object_database **alt_odb_tail;
 281
 282/*
 283 * Return non-zero iff the path is usable as an alternate object database.
 284 */
 285static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
 286{
 287        struct alternate_object_database *alt;
 288
 289        /* Detect cases where alternate disappeared */
 290        if (!is_directory(path->buf)) {
 291                error("object directory %s does not exist; "
 292                      "check .git/objects/info/alternates.",
 293                      path->buf);
 294                return 0;
 295        }
 296
 297        /*
 298         * Prevent the common mistake of listing the same
 299         * thing twice, or object directory itself.
 300         */
 301        for (alt = alt_odb_list; alt; alt = alt->next) {
 302                if (!fspathcmp(path->buf, alt->path))
 303                        return 0;
 304        }
 305        if (!fspathcmp(path->buf, normalized_objdir))
 306                return 0;
 307
 308        return 1;
 309}
 310
 311/*
 312 * Prepare alternate object database registry.
 313 *
 314 * The variable alt_odb_list points at the list of struct
 315 * alternate_object_database.  The elements on this list come from
 316 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 317 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 318 * whose contents is similar to that environment variable but can be
 319 * LF separated.  Its base points at a statically allocated buffer that
 320 * contains "/the/directory/corresponding/to/.git/objects/...", while
 321 * its name points just after the slash at the end of ".git/objects/"
 322 * in the example above, and has enough space to hold 40-byte hex
 323 * SHA1, an extra slash for the first level indirection, and the
 324 * terminating NUL.
 325 */
 326static void read_info_alternates(const char * relative_base, int depth);
 327static int link_alt_odb_entry(const char *entry, const char *relative_base,
 328        int depth, const char *normalized_objdir)
 329{
 330        struct alternate_object_database *ent;
 331        struct strbuf pathbuf = STRBUF_INIT;
 332
 333        if (!is_absolute_path(entry) && relative_base) {
 334                strbuf_realpath(&pathbuf, relative_base, 1);
 335                strbuf_addch(&pathbuf, '/');
 336        }
 337        strbuf_addstr(&pathbuf, entry);
 338
 339        if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
 340                error("unable to normalize alternate object path: %s",
 341                      pathbuf.buf);
 342                strbuf_release(&pathbuf);
 343                return -1;
 344        }
 345
 346        /*
 347         * The trailing slash after the directory name is given by
 348         * this function at the end. Remove duplicates.
 349         */
 350        while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
 351                strbuf_setlen(&pathbuf, pathbuf.len - 1);
 352
 353        if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
 354                strbuf_release(&pathbuf);
 355                return -1;
 356        }
 357
 358        ent = alloc_alt_odb(pathbuf.buf);
 359
 360        /* add the alternate entry */
 361        *alt_odb_tail = ent;
 362        alt_odb_tail = &(ent->next);
 363        ent->next = NULL;
 364
 365        /* recursively add alternates */
 366        read_info_alternates(pathbuf.buf, depth + 1);
 367
 368        strbuf_release(&pathbuf);
 369        return 0;
 370}
 371
 372static const char *parse_alt_odb_entry(const char *string,
 373                                       int sep,
 374                                       struct strbuf *out)
 375{
 376        const char *end;
 377
 378        strbuf_reset(out);
 379
 380        if (*string == '#') {
 381                /* comment; consume up to next separator */
 382                end = strchrnul(string, sep);
 383        } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
 384                /*
 385                 * quoted path; unquote_c_style has copied the
 386                 * data for us and set "end". Broken quoting (e.g.,
 387                 * an entry that doesn't end with a quote) falls
 388                 * back to the unquoted case below.
 389                 */
 390        } else {
 391                /* normal, unquoted path */
 392                end = strchrnul(string, sep);
 393                strbuf_add(out, string, end - string);
 394        }
 395
 396        if (*end)
 397                end++;
 398        return end;
 399}
 400
 401static void link_alt_odb_entries(const char *alt, int len, int sep,
 402                                 const char *relative_base, int depth)
 403{
 404        struct strbuf objdirbuf = STRBUF_INIT;
 405        struct strbuf entry = STRBUF_INIT;
 406
 407        if (depth > 5) {
 408                error("%s: ignoring alternate object stores, nesting too deep.",
 409                                relative_base);
 410                return;
 411        }
 412
 413        strbuf_add_absolute_path(&objdirbuf, get_object_directory());
 414        if (strbuf_normalize_path(&objdirbuf) < 0)
 415                die("unable to normalize object directory: %s",
 416                    objdirbuf.buf);
 417
 418        while (*alt) {
 419                alt = parse_alt_odb_entry(alt, sep, &entry);
 420                if (!entry.len)
 421                        continue;
 422                link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
 423        }
 424        strbuf_release(&entry);
 425        strbuf_release(&objdirbuf);
 426}
 427
 428static void read_info_alternates(const char * relative_base, int depth)
 429{
 430        char *map;
 431        size_t mapsz;
 432        struct stat st;
 433        char *path;
 434        int fd;
 435
 436        path = xstrfmt("%s/info/alternates", relative_base);
 437        fd = git_open(path);
 438        free(path);
 439        if (fd < 0)
 440                return;
 441        if (fstat(fd, &st) || (st.st_size == 0)) {
 442                close(fd);
 443                return;
 444        }
 445        mapsz = xsize_t(st.st_size);
 446        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 447        close(fd);
 448
 449        link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
 450
 451        munmap(map, mapsz);
 452}
 453
 454struct alternate_object_database *alloc_alt_odb(const char *dir)
 455{
 456        struct alternate_object_database *ent;
 457
 458        FLEX_ALLOC_STR(ent, path, dir);
 459        strbuf_init(&ent->scratch, 0);
 460        strbuf_addf(&ent->scratch, "%s/", dir);
 461        ent->base_len = ent->scratch.len;
 462
 463        return ent;
 464}
 465
 466void add_to_alternates_file(const char *reference)
 467{
 468        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
 469        char *alts = git_pathdup("objects/info/alternates");
 470        FILE *in, *out;
 471
 472        hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
 473        out = fdopen_lock_file(lock, "w");
 474        if (!out)
 475                die_errno("unable to fdopen alternates lockfile");
 476
 477        in = fopen(alts, "r");
 478        if (in) {
 479                struct strbuf line = STRBUF_INIT;
 480                int found = 0;
 481
 482                while (strbuf_getline(&line, in) != EOF) {
 483                        if (!strcmp(reference, line.buf)) {
 484                                found = 1;
 485                                break;
 486                        }
 487                        fprintf_or_die(out, "%s\n", line.buf);
 488                }
 489
 490                strbuf_release(&line);
 491                fclose(in);
 492
 493                if (found) {
 494                        rollback_lock_file(lock);
 495                        lock = NULL;
 496                }
 497        }
 498        else if (errno != ENOENT)
 499                die_errno("unable to read alternates file");
 500
 501        if (lock) {
 502                fprintf_or_die(out, "%s\n", reference);
 503                if (commit_lock_file(lock))
 504                        die_errno("unable to move new alternates file into place");
 505                if (alt_odb_tail)
 506                        link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
 507        }
 508        free(alts);
 509}
 510
 511void add_to_alternates_memory(const char *reference)
 512{
 513        /*
 514         * Make sure alternates are initialized, or else our entry may be
 515         * overwritten when they are.
 516         */
 517        prepare_alt_odb();
 518
 519        link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
 520}
 521
 522/*
 523 * Compute the exact path an alternate is at and returns it. In case of
 524 * error NULL is returned and the human readable error is added to `err`
 525 * `path` may be relative and should point to $GITDIR.
 526 * `err` must not be null.
 527 */
 528char *compute_alternate_path(const char *path, struct strbuf *err)
 529{
 530        char *ref_git = NULL;
 531        const char *repo, *ref_git_s;
 532        int seen_error = 0;
 533
 534        ref_git_s = real_path_if_valid(path);
 535        if (!ref_git_s) {
 536                seen_error = 1;
 537                strbuf_addf(err, _("path '%s' does not exist"), path);
 538                goto out;
 539        } else
 540                /*
 541                 * Beware: read_gitfile(), real_path() and mkpath()
 542                 * return static buffer
 543                 */
 544                ref_git = xstrdup(ref_git_s);
 545
 546        repo = read_gitfile(ref_git);
 547        if (!repo)
 548                repo = read_gitfile(mkpath("%s/.git", ref_git));
 549        if (repo) {
 550                free(ref_git);
 551                ref_git = xstrdup(repo);
 552        }
 553
 554        if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
 555                char *ref_git_git = mkpathdup("%s/.git", ref_git);
 556                free(ref_git);
 557                ref_git = ref_git_git;
 558        } else if (!is_directory(mkpath("%s/objects", ref_git))) {
 559                struct strbuf sb = STRBUF_INIT;
 560                seen_error = 1;
 561                if (get_common_dir(&sb, ref_git)) {
 562                        strbuf_addf(err,
 563                                    _("reference repository '%s' as a linked "
 564                                      "checkout is not supported yet."),
 565                                    path);
 566                        goto out;
 567                }
 568
 569                strbuf_addf(err, _("reference repository '%s' is not a "
 570                                        "local repository."), path);
 571                goto out;
 572        }
 573
 574        if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
 575                strbuf_addf(err, _("reference repository '%s' is shallow"),
 576                            path);
 577                seen_error = 1;
 578                goto out;
 579        }
 580
 581        if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
 582                strbuf_addf(err,
 583                            _("reference repository '%s' is grafted"),
 584                            path);
 585                seen_error = 1;
 586                goto out;
 587        }
 588
 589out:
 590        if (seen_error) {
 591                FREE_AND_NULL(ref_git);
 592        }
 593
 594        return ref_git;
 595}
 596
 597int foreach_alt_odb(alt_odb_fn fn, void *cb)
 598{
 599        struct alternate_object_database *ent;
 600        int r = 0;
 601
 602        prepare_alt_odb();
 603        for (ent = alt_odb_list; ent; ent = ent->next) {
 604                r = fn(ent, cb);
 605                if (r)
 606                        break;
 607        }
 608        return r;
 609}
 610
 611void prepare_alt_odb(void)
 612{
 613        const char *alt;
 614
 615        if (alt_odb_tail)
 616                return;
 617
 618        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 619        if (!alt) alt = "";
 620
 621        alt_odb_tail = &alt_odb_list;
 622        link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
 623
 624        read_info_alternates(get_object_directory(), 0);
 625}
 626
 627/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
 628static int freshen_file(const char *fn)
 629{
 630        struct utimbuf t;
 631        t.actime = t.modtime = time(NULL);
 632        return !utime(fn, &t);
 633}
 634
 635/*
 636 * All of the check_and_freshen functions return 1 if the file exists and was
 637 * freshened (if freshening was requested), 0 otherwise. If they return
 638 * 0, you should not assume that it is safe to skip a write of the object (it
 639 * either does not exist on disk, or has a stale mtime and may be subject to
 640 * pruning).
 641 */
 642int check_and_freshen_file(const char *fn, int freshen)
 643{
 644        if (access(fn, F_OK))
 645                return 0;
 646        if (freshen && !freshen_file(fn))
 647                return 0;
 648        return 1;
 649}
 650
 651static int check_and_freshen_local(const unsigned char *sha1, int freshen)
 652{
 653        return check_and_freshen_file(sha1_file_name(sha1), freshen);
 654}
 655
 656static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
 657{
 658        struct alternate_object_database *alt;
 659        prepare_alt_odb();
 660        for (alt = alt_odb_list; alt; alt = alt->next) {
 661                const char *path = alt_sha1_path(alt, sha1);
 662                if (check_and_freshen_file(path, freshen))
 663                        return 1;
 664        }
 665        return 0;
 666}
 667
 668static int check_and_freshen(const unsigned char *sha1, int freshen)
 669{
 670        return check_and_freshen_local(sha1, freshen) ||
 671               check_and_freshen_nonlocal(sha1, freshen);
 672}
 673
 674int has_loose_object_nonlocal(const unsigned char *sha1)
 675{
 676        return check_and_freshen_nonlocal(sha1, 0);
 677}
 678
 679static int has_loose_object(const unsigned char *sha1)
 680{
 681        return check_and_freshen(sha1, 0);
 682}
 683
 684static void mmap_limit_check(size_t length)
 685{
 686        static size_t limit = 0;
 687        if (!limit) {
 688                limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
 689                if (!limit)
 690                        limit = SIZE_MAX;
 691        }
 692        if (length > limit)
 693                die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
 694                    (uintmax_t)length, (uintmax_t)limit);
 695}
 696
 697void *xmmap_gently(void *start, size_t length,
 698                  int prot, int flags, int fd, off_t offset)
 699{
 700        void *ret;
 701
 702        mmap_limit_check(length);
 703        ret = mmap(start, length, prot, flags, fd, offset);
 704        if (ret == MAP_FAILED) {
 705                if (!length)
 706                        return NULL;
 707                release_pack_memory(length);
 708                ret = mmap(start, length, prot, flags, fd, offset);
 709        }
 710        return ret;
 711}
 712
 713void *xmmap(void *start, size_t length,
 714        int prot, int flags, int fd, off_t offset)
 715{
 716        void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
 717        if (ret == MAP_FAILED)
 718                die_errno("mmap failed");
 719        return ret;
 720}
 721
 722static void mark_bad_packed_object(struct packed_git *p,
 723                                   const unsigned char *sha1)
 724{
 725        unsigned i;
 726        for (i = 0; i < p->num_bad_objects; i++)
 727                if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
 728                        return;
 729        p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
 730                                      st_mult(GIT_MAX_RAWSZ,
 731                                              st_add(p->num_bad_objects, 1)));
 732        hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
 733        p->num_bad_objects++;
 734}
 735
 736static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
 737{
 738        struct packed_git *p;
 739        unsigned i;
 740
 741        for (p = packed_git; p; p = p->next)
 742                for (i = 0; i < p->num_bad_objects; i++)
 743                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
 744                                return p;
 745        return NULL;
 746}
 747
 748/*
 749 * With an in-core object data in "map", rehash it to make sure the
 750 * object name actually matches "sha1" to detect object corruption.
 751 * With "map" == NULL, try reading the object named with "sha1" using
 752 * the streaming interface and rehash it to do the same.
 753 */
 754int check_sha1_signature(const unsigned char *sha1, void *map,
 755                         unsigned long size, const char *type)
 756{
 757        unsigned char real_sha1[20];
 758        enum object_type obj_type;
 759        struct git_istream *st;
 760        git_SHA_CTX c;
 761        char hdr[32];
 762        int hdrlen;
 763
 764        if (map) {
 765                hash_sha1_file(map, size, type, real_sha1);
 766                return hashcmp(sha1, real_sha1) ? -1 : 0;
 767        }
 768
 769        st = open_istream(sha1, &obj_type, &size, NULL);
 770        if (!st)
 771                return -1;
 772
 773        /* Generate the header */
 774        hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
 775
 776        /* Sha1.. */
 777        git_SHA1_Init(&c);
 778        git_SHA1_Update(&c, hdr, hdrlen);
 779        for (;;) {
 780                char buf[1024 * 16];
 781                ssize_t readlen = read_istream(st, buf, sizeof(buf));
 782
 783                if (readlen < 0) {
 784                        close_istream(st);
 785                        return -1;
 786                }
 787                if (!readlen)
 788                        break;
 789                git_SHA1_Update(&c, buf, readlen);
 790        }
 791        git_SHA1_Final(real_sha1, &c);
 792        close_istream(st);
 793        return hashcmp(sha1, real_sha1) ? -1 : 0;
 794}
 795
 796int git_open_cloexec(const char *name, int flags)
 797{
 798        int fd;
 799        static int o_cloexec = O_CLOEXEC;
 800
 801        fd = open(name, flags | o_cloexec);
 802        if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
 803                /* Try again w/o O_CLOEXEC: the kernel might not support it */
 804                o_cloexec &= ~O_CLOEXEC;
 805                fd = open(name, flags | o_cloexec);
 806        }
 807
 808#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
 809        {
 810                static int fd_cloexec = FD_CLOEXEC;
 811
 812                if (!o_cloexec && 0 <= fd && fd_cloexec) {
 813                        /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
 814                        int flags = fcntl(fd, F_GETFD);
 815                        if (fcntl(fd, F_SETFD, flags | fd_cloexec))
 816                                fd_cloexec = 0;
 817                }
 818        }
 819#endif
 820        return fd;
 821}
 822
 823/*
 824 * Find "sha1" as a loose object in the local repository or in an alternate.
 825 * Returns 0 on success, negative on failure.
 826 *
 827 * The "path" out-parameter will give the path of the object we found (if any).
 828 * Note that it may point to static storage and is only valid until another
 829 * call to sha1_file_name(), etc.
 830 */
 831static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
 832                          const char **path)
 833{
 834        struct alternate_object_database *alt;
 835
 836        *path = sha1_file_name(sha1);
 837        if (!lstat(*path, st))
 838                return 0;
 839
 840        prepare_alt_odb();
 841        errno = ENOENT;
 842        for (alt = alt_odb_list; alt; alt = alt->next) {
 843                *path = alt_sha1_path(alt, sha1);
 844                if (!lstat(*path, st))
 845                        return 0;
 846        }
 847
 848        return -1;
 849}
 850
 851/*
 852 * Like stat_sha1_file(), but actually open the object and return the
 853 * descriptor. See the caveats on the "path" parameter above.
 854 */
 855static int open_sha1_file(const unsigned char *sha1, const char **path)
 856{
 857        int fd;
 858        struct alternate_object_database *alt;
 859        int most_interesting_errno;
 860
 861        *path = sha1_file_name(sha1);
 862        fd = git_open(*path);
 863        if (fd >= 0)
 864                return fd;
 865        most_interesting_errno = errno;
 866
 867        prepare_alt_odb();
 868        for (alt = alt_odb_list; alt; alt = alt->next) {
 869                *path = alt_sha1_path(alt, sha1);
 870                fd = git_open(*path);
 871                if (fd >= 0)
 872                        return fd;
 873                if (most_interesting_errno == ENOENT)
 874                        most_interesting_errno = errno;
 875        }
 876        errno = most_interesting_errno;
 877        return -1;
 878}
 879
 880/*
 881 * Map the loose object at "path" if it is not NULL, or the path found by
 882 * searching for a loose object named "sha1".
 883 */
 884static void *map_sha1_file_1(const char *path,
 885                             const unsigned char *sha1,
 886                             unsigned long *size)
 887{
 888        void *map;
 889        int fd;
 890
 891        if (path)
 892                fd = git_open(path);
 893        else
 894                fd = open_sha1_file(sha1, &path);
 895        map = NULL;
 896        if (fd >= 0) {
 897                struct stat st;
 898
 899                if (!fstat(fd, &st)) {
 900                        *size = xsize_t(st.st_size);
 901                        if (!*size) {
 902                                /* mmap() is forbidden on empty files */
 903                                error("object file %s is empty", path);
 904                                return NULL;
 905                        }
 906                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 907                }
 908                close(fd);
 909        }
 910        return map;
 911}
 912
 913void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
 914{
 915        return map_sha1_file_1(NULL, sha1, size);
 916}
 917
 918unsigned long unpack_object_header_buffer(const unsigned char *buf,
 919                unsigned long len, enum object_type *type, unsigned long *sizep)
 920{
 921        unsigned shift;
 922        unsigned long size, c;
 923        unsigned long used = 0;
 924
 925        c = buf[used++];
 926        *type = (c >> 4) & 7;
 927        size = c & 15;
 928        shift = 4;
 929        while (c & 0x80) {
 930                if (len <= used || bitsizeof(long) <= shift) {
 931                        error("bad object header");
 932                        size = used = 0;
 933                        break;
 934                }
 935                c = buf[used++];
 936                size += (c & 0x7f) << shift;
 937                shift += 7;
 938        }
 939        *sizep = size;
 940        return used;
 941}
 942
 943static int unpack_sha1_short_header(git_zstream *stream,
 944                                    unsigned char *map, unsigned long mapsize,
 945                                    void *buffer, unsigned long bufsiz)
 946{
 947        /* Get the data stream */
 948        memset(stream, 0, sizeof(*stream));
 949        stream->next_in = map;
 950        stream->avail_in = mapsize;
 951        stream->next_out = buffer;
 952        stream->avail_out = bufsiz;
 953
 954        git_inflate_init(stream);
 955        return git_inflate(stream, 0);
 956}
 957
 958int unpack_sha1_header(git_zstream *stream,
 959                       unsigned char *map, unsigned long mapsize,
 960                       void *buffer, unsigned long bufsiz)
 961{
 962        int status = unpack_sha1_short_header(stream, map, mapsize,
 963                                              buffer, bufsiz);
 964
 965        if (status < Z_OK)
 966                return status;
 967
 968        /* Make sure we have the terminating NUL */
 969        if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
 970                return -1;
 971        return 0;
 972}
 973
 974static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
 975                                        unsigned long mapsize, void *buffer,
 976                                        unsigned long bufsiz, struct strbuf *header)
 977{
 978        int status;
 979
 980        status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
 981        if (status < Z_OK)
 982                return -1;
 983
 984        /*
 985         * Check if entire header is unpacked in the first iteration.
 986         */
 987        if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
 988                return 0;
 989
 990        /*
 991         * buffer[0..bufsiz] was not large enough.  Copy the partial
 992         * result out to header, and then append the result of further
 993         * reading the stream.
 994         */
 995        strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
 996        stream->next_out = buffer;
 997        stream->avail_out = bufsiz;
 998
 999        do {
1000                status = git_inflate(stream, 0);
1001                strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1002                if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1003                        return 0;
1004                stream->next_out = buffer;
1005                stream->avail_out = bufsiz;
1006        } while (status != Z_STREAM_END);
1007        return -1;
1008}
1009
1010static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1011{
1012        int bytes = strlen(buffer) + 1;
1013        unsigned char *buf = xmallocz(size);
1014        unsigned long n;
1015        int status = Z_OK;
1016
1017        n = stream->total_out - bytes;
1018        if (n > size)
1019                n = size;
1020        memcpy(buf, (char *) buffer + bytes, n);
1021        bytes = n;
1022        if (bytes <= size) {
1023                /*
1024                 * The above condition must be (bytes <= size), not
1025                 * (bytes < size).  In other words, even though we
1026                 * expect no more output and set avail_out to zero,
1027                 * the input zlib stream may have bytes that express
1028                 * "this concludes the stream", and we *do* want to
1029                 * eat that input.
1030                 *
1031                 * Otherwise we would not be able to test that we
1032                 * consumed all the input to reach the expected size;
1033                 * we also want to check that zlib tells us that all
1034                 * went well with status == Z_STREAM_END at the end.
1035                 */
1036                stream->next_out = buf + bytes;
1037                stream->avail_out = size - bytes;
1038                while (status == Z_OK)
1039                        status = git_inflate(stream, Z_FINISH);
1040        }
1041        if (status == Z_STREAM_END && !stream->avail_in) {
1042                git_inflate_end(stream);
1043                return buf;
1044        }
1045
1046        if (status < 0)
1047                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1048        else if (stream->avail_in)
1049                error("garbage at end of loose object '%s'",
1050                      sha1_to_hex(sha1));
1051        free(buf);
1052        return NULL;
1053}
1054
1055/*
1056 * We used to just use "sscanf()", but that's actually way
1057 * too permissive for what we want to check. So do an anal
1058 * object header parse by hand.
1059 */
1060static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1061                               unsigned int flags)
1062{
1063        const char *type_buf = hdr;
1064        unsigned long size;
1065        int type, type_len = 0;
1066
1067        /*
1068         * The type can be of any size but is followed by
1069         * a space.
1070         */
1071        for (;;) {
1072                char c = *hdr++;
1073                if (!c)
1074                        return -1;
1075                if (c == ' ')
1076                        break;
1077                type_len++;
1078        }
1079
1080        type = type_from_string_gently(type_buf, type_len, 1);
1081        if (oi->typename)
1082                strbuf_add(oi->typename, type_buf, type_len);
1083        /*
1084         * Set type to 0 if its an unknown object and
1085         * we're obtaining the type using '--allow-unknown-type'
1086         * option.
1087         */
1088        if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1089                type = 0;
1090        else if (type < 0)
1091                die("invalid object type");
1092        if (oi->typep)
1093                *oi->typep = type;
1094
1095        /*
1096         * The length must follow immediately, and be in canonical
1097         * decimal format (ie "010" is not valid).
1098         */
1099        size = *hdr++ - '0';
1100        if (size > 9)
1101                return -1;
1102        if (size) {
1103                for (;;) {
1104                        unsigned long c = *hdr - '0';
1105                        if (c > 9)
1106                                break;
1107                        hdr++;
1108                        size = size * 10 + c;
1109                }
1110        }
1111
1112        if (oi->sizep)
1113                *oi->sizep = size;
1114
1115        /*
1116         * The length must be followed by a zero byte
1117         */
1118        return *hdr ? -1 : type;
1119}
1120
1121int parse_sha1_header(const char *hdr, unsigned long *sizep)
1122{
1123        struct object_info oi = OBJECT_INFO_INIT;
1124
1125        oi.sizep = sizep;
1126        return parse_sha1_header_extended(hdr, &oi, 0);
1127}
1128
1129unsigned long get_size_from_delta(struct packed_git *p,
1130                                  struct pack_window **w_curs,
1131                                  off_t curpos)
1132{
1133        const unsigned char *data;
1134        unsigned char delta_head[20], *in;
1135        git_zstream stream;
1136        int st;
1137
1138        memset(&stream, 0, sizeof(stream));
1139        stream.next_out = delta_head;
1140        stream.avail_out = sizeof(delta_head);
1141
1142        git_inflate_init(&stream);
1143        do {
1144                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1145                stream.next_in = in;
1146                st = git_inflate(&stream, Z_FINISH);
1147                curpos += stream.next_in - in;
1148        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1149                 stream.total_out < sizeof(delta_head));
1150        git_inflate_end(&stream);
1151        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1152                error("delta data unpack-initial failed");
1153                return 0;
1154        }
1155
1156        /* Examine the initial part of the delta to figure out
1157         * the result size.
1158         */
1159        data = delta_head;
1160
1161        /* ignore base size */
1162        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1163
1164        /* Read the result size */
1165        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1166}
1167
1168static off_t get_delta_base(struct packed_git *p,
1169                                    struct pack_window **w_curs,
1170                                    off_t *curpos,
1171                                    enum object_type type,
1172                                    off_t delta_obj_offset)
1173{
1174        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1175        off_t base_offset;
1176
1177        /* use_pack() assured us we have [base_info, base_info + 20)
1178         * as a range that we can look at without walking off the
1179         * end of the mapped window.  Its actually the hash size
1180         * that is assured.  An OFS_DELTA longer than the hash size
1181         * is stupid, as then a REF_DELTA would be smaller to store.
1182         */
1183        if (type == OBJ_OFS_DELTA) {
1184                unsigned used = 0;
1185                unsigned char c = base_info[used++];
1186                base_offset = c & 127;
1187                while (c & 128) {
1188                        base_offset += 1;
1189                        if (!base_offset || MSB(base_offset, 7))
1190                                return 0;  /* overflow */
1191                        c = base_info[used++];
1192                        base_offset = (base_offset << 7) + (c & 127);
1193                }
1194                base_offset = delta_obj_offset - base_offset;
1195                if (base_offset <= 0 || base_offset >= delta_obj_offset)
1196                        return 0;  /* out of bound */
1197                *curpos += used;
1198        } else if (type == OBJ_REF_DELTA) {
1199                /* The base entry _must_ be in the same pack */
1200                base_offset = find_pack_entry_one(base_info, p);
1201                *curpos += 20;
1202        } else
1203                die("I am totally screwed");
1204        return base_offset;
1205}
1206
1207/*
1208 * Like get_delta_base above, but we return the sha1 instead of the pack
1209 * offset. This means it is cheaper for REF deltas (we do not have to do
1210 * the final object lookup), but more expensive for OFS deltas (we
1211 * have to load the revidx to convert the offset back into a sha1).
1212 */
1213static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1214                                                struct pack_window **w_curs,
1215                                                off_t curpos,
1216                                                enum object_type type,
1217                                                off_t delta_obj_offset)
1218{
1219        if (type == OBJ_REF_DELTA) {
1220                unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1221                return base;
1222        } else if (type == OBJ_OFS_DELTA) {
1223                struct revindex_entry *revidx;
1224                off_t base_offset = get_delta_base(p, w_curs, &curpos,
1225                                                   type, delta_obj_offset);
1226
1227                if (!base_offset)
1228                        return NULL;
1229
1230                revidx = find_pack_revindex(p, base_offset);
1231                if (!revidx)
1232                        return NULL;
1233
1234                return nth_packed_object_sha1(p, revidx->nr);
1235        } else
1236                return NULL;
1237}
1238
1239int unpack_object_header(struct packed_git *p,
1240                         struct pack_window **w_curs,
1241                         off_t *curpos,
1242                         unsigned long *sizep)
1243{
1244        unsigned char *base;
1245        unsigned long left;
1246        unsigned long used;
1247        enum object_type type;
1248
1249        /* use_pack() assures us we have [base, base + 20) available
1250         * as a range that we can look at.  (Its actually the hash
1251         * size that is assured.)  With our object header encoding
1252         * the maximum deflated object size is 2^137, which is just
1253         * insane, so we know won't exceed what we have been given.
1254         */
1255        base = use_pack(p, w_curs, *curpos, &left);
1256        used = unpack_object_header_buffer(base, left, &type, sizep);
1257        if (!used) {
1258                type = OBJ_BAD;
1259        } else
1260                *curpos += used;
1261
1262        return type;
1263}
1264
1265static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1266{
1267        int type;
1268        struct revindex_entry *revidx;
1269        const unsigned char *sha1;
1270        revidx = find_pack_revindex(p, obj_offset);
1271        if (!revidx)
1272                return OBJ_BAD;
1273        sha1 = nth_packed_object_sha1(p, revidx->nr);
1274        mark_bad_packed_object(p, sha1);
1275        type = sha1_object_info(sha1, NULL);
1276        if (type <= OBJ_NONE)
1277                return OBJ_BAD;
1278        return type;
1279}
1280
1281#define POI_STACK_PREALLOC 64
1282
1283static enum object_type packed_to_object_type(struct packed_git *p,
1284                                              off_t obj_offset,
1285                                              enum object_type type,
1286                                              struct pack_window **w_curs,
1287                                              off_t curpos)
1288{
1289        off_t small_poi_stack[POI_STACK_PREALLOC];
1290        off_t *poi_stack = small_poi_stack;
1291        int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1292
1293        while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1294                off_t base_offset;
1295                unsigned long size;
1296                /* Push the object we're going to leave behind */
1297                if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1298                        poi_stack_alloc = alloc_nr(poi_stack_nr);
1299                        ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1300                        memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1301                } else {
1302                        ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1303                }
1304                poi_stack[poi_stack_nr++] = obj_offset;
1305                /* If parsing the base offset fails, just unwind */
1306                base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1307                if (!base_offset)
1308                        goto unwind;
1309                curpos = obj_offset = base_offset;
1310                type = unpack_object_header(p, w_curs, &curpos, &size);
1311                if (type <= OBJ_NONE) {
1312                        /* If getting the base itself fails, we first
1313                         * retry the base, otherwise unwind */
1314                        type = retry_bad_packed_offset(p, base_offset);
1315                        if (type > OBJ_NONE)
1316                                goto out;
1317                        goto unwind;
1318                }
1319        }
1320
1321        switch (type) {
1322        case OBJ_BAD:
1323        case OBJ_COMMIT:
1324        case OBJ_TREE:
1325        case OBJ_BLOB:
1326        case OBJ_TAG:
1327                break;
1328        default:
1329                error("unknown object type %i at offset %"PRIuMAX" in %s",
1330                      type, (uintmax_t)obj_offset, p->pack_name);
1331                type = OBJ_BAD;
1332        }
1333
1334out:
1335        if (poi_stack != small_poi_stack)
1336                free(poi_stack);
1337        return type;
1338
1339unwind:
1340        while (poi_stack_nr) {
1341                obj_offset = poi_stack[--poi_stack_nr];
1342                type = retry_bad_packed_offset(p, obj_offset);
1343                if (type > OBJ_NONE)
1344                        goto out;
1345        }
1346        type = OBJ_BAD;
1347        goto out;
1348}
1349
1350static struct hashmap delta_base_cache;
1351static size_t delta_base_cached;
1352
1353static LIST_HEAD(delta_base_cache_lru);
1354
1355struct delta_base_cache_key {
1356        struct packed_git *p;
1357        off_t base_offset;
1358};
1359
1360struct delta_base_cache_entry {
1361        struct hashmap hash;
1362        struct delta_base_cache_key key;
1363        struct list_head lru;
1364        void *data;
1365        unsigned long size;
1366        enum object_type type;
1367};
1368
1369static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1370{
1371        unsigned int hash;
1372
1373        hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1374        hash += (hash >> 8) + (hash >> 16);
1375        return hash;
1376}
1377
1378static struct delta_base_cache_entry *
1379get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1380{
1381        struct hashmap_entry entry;
1382        struct delta_base_cache_key key;
1383
1384        if (!delta_base_cache.cmpfn)
1385                return NULL;
1386
1387        hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1388        key.p = p;
1389        key.base_offset = base_offset;
1390        return hashmap_get(&delta_base_cache, &entry, &key);
1391}
1392
1393static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1394                                   const struct delta_base_cache_key *b)
1395{
1396        return a->p == b->p && a->base_offset == b->base_offset;
1397}
1398
1399static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1400                                     const void *va, const void *vb,
1401                                     const void *vkey)
1402{
1403        const struct delta_base_cache_entry *a = va, *b = vb;
1404        const struct delta_base_cache_key *key = vkey;
1405        if (key)
1406                return !delta_base_cache_key_eq(&a->key, key);
1407        else
1408                return !delta_base_cache_key_eq(&a->key, &b->key);
1409}
1410
1411static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1412{
1413        return !!get_delta_base_cache_entry(p, base_offset);
1414}
1415
1416/*
1417 * Remove the entry from the cache, but do _not_ free the associated
1418 * entry data. The caller takes ownership of the "data" buffer, and
1419 * should copy out any fields it wants before detaching.
1420 */
1421static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1422{
1423        hashmap_remove(&delta_base_cache, ent, &ent->key);
1424        list_del(&ent->lru);
1425        delta_base_cached -= ent->size;
1426        free(ent);
1427}
1428
1429static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1430        unsigned long *base_size, enum object_type *type)
1431{
1432        struct delta_base_cache_entry *ent;
1433
1434        ent = get_delta_base_cache_entry(p, base_offset);
1435        if (!ent)
1436                return unpack_entry(p, base_offset, type, base_size);
1437
1438        if (type)
1439                *type = ent->type;
1440        if (base_size)
1441                *base_size = ent->size;
1442        return xmemdupz(ent->data, ent->size);
1443}
1444
1445static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1446{
1447        free(ent->data);
1448        detach_delta_base_cache_entry(ent);
1449}
1450
1451void clear_delta_base_cache(void)
1452{
1453        struct list_head *lru, *tmp;
1454        list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1455                struct delta_base_cache_entry *entry =
1456                        list_entry(lru, struct delta_base_cache_entry, lru);
1457                release_delta_base_cache(entry);
1458        }
1459}
1460
1461static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1462        void *base, unsigned long base_size, enum object_type type)
1463{
1464        struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1465        struct list_head *lru, *tmp;
1466
1467        delta_base_cached += base_size;
1468
1469        list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1470                struct delta_base_cache_entry *f =
1471                        list_entry(lru, struct delta_base_cache_entry, lru);
1472                if (delta_base_cached <= delta_base_cache_limit)
1473                        break;
1474                release_delta_base_cache(f);
1475        }
1476
1477        ent->key.p = p;
1478        ent->key.base_offset = base_offset;
1479        ent->type = type;
1480        ent->data = base;
1481        ent->size = base_size;
1482        list_add_tail(&ent->lru, &delta_base_cache_lru);
1483
1484        if (!delta_base_cache.cmpfn)
1485                hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1486        hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1487        hashmap_add(&delta_base_cache, ent);
1488}
1489
1490int packed_object_info(struct packed_git *p, off_t obj_offset,
1491                       struct object_info *oi)
1492{
1493        struct pack_window *w_curs = NULL;
1494        unsigned long size;
1495        off_t curpos = obj_offset;
1496        enum object_type type;
1497
1498        /*
1499         * We always get the representation type, but only convert it to
1500         * a "real" type later if the caller is interested.
1501         */
1502        if (oi->contentp) {
1503                *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
1504                                                      &type);
1505                if (!*oi->contentp)
1506                        type = OBJ_BAD;
1507        } else {
1508                type = unpack_object_header(p, &w_curs, &curpos, &size);
1509        }
1510
1511        if (!oi->contentp && oi->sizep) {
1512                if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1513                        off_t tmp_pos = curpos;
1514                        off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1515                                                           type, obj_offset);
1516                        if (!base_offset) {
1517                                type = OBJ_BAD;
1518                                goto out;
1519                        }
1520                        *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1521                        if (*oi->sizep == 0) {
1522                                type = OBJ_BAD;
1523                                goto out;
1524                        }
1525                } else {
1526                        *oi->sizep = size;
1527                }
1528        }
1529
1530        if (oi->disk_sizep) {
1531                struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1532                *oi->disk_sizep = revidx[1].offset - obj_offset;
1533        }
1534
1535        if (oi->typep || oi->typename) {
1536                enum object_type ptot;
1537                ptot = packed_to_object_type(p, obj_offset, type, &w_curs,
1538                                             curpos);
1539                if (oi->typep)
1540                        *oi->typep = ptot;
1541                if (oi->typename) {
1542                        const char *tn = typename(ptot);
1543                        if (tn)
1544                                strbuf_addstr(oi->typename, tn);
1545                }
1546                if (ptot < 0) {
1547                        type = OBJ_BAD;
1548                        goto out;
1549                }
1550        }
1551
1552        if (oi->delta_base_sha1) {
1553                if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1554                        const unsigned char *base;
1555
1556                        base = get_delta_base_sha1(p, &w_curs, curpos,
1557                                                   type, obj_offset);
1558                        if (!base) {
1559                                type = OBJ_BAD;
1560                                goto out;
1561                        }
1562
1563                        hashcpy(oi->delta_base_sha1, base);
1564                } else
1565                        hashclr(oi->delta_base_sha1);
1566        }
1567
1568        oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1569                                                          OI_PACKED;
1570
1571out:
1572        unuse_pack(&w_curs);
1573        return type;
1574}
1575
1576static void *unpack_compressed_entry(struct packed_git *p,
1577                                    struct pack_window **w_curs,
1578                                    off_t curpos,
1579                                    unsigned long size)
1580{
1581        int st;
1582        git_zstream stream;
1583        unsigned char *buffer, *in;
1584
1585        buffer = xmallocz_gently(size);
1586        if (!buffer)
1587                return NULL;
1588        memset(&stream, 0, sizeof(stream));
1589        stream.next_out = buffer;
1590        stream.avail_out = size + 1;
1591
1592        git_inflate_init(&stream);
1593        do {
1594                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1595                stream.next_in = in;
1596                st = git_inflate(&stream, Z_FINISH);
1597                if (!stream.avail_out)
1598                        break; /* the payload is larger than it should be */
1599                curpos += stream.next_in - in;
1600        } while (st == Z_OK || st == Z_BUF_ERROR);
1601        git_inflate_end(&stream);
1602        if ((st != Z_STREAM_END) || stream.total_out != size) {
1603                free(buffer);
1604                return NULL;
1605        }
1606
1607        return buffer;
1608}
1609
1610static void *read_object(const unsigned char *sha1, enum object_type *type,
1611                         unsigned long *size);
1612
1613static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1614{
1615        static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1616        trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1617                         p->pack_name, (uintmax_t)obj_offset);
1618}
1619
1620int do_check_packed_object_crc;
1621
1622#define UNPACK_ENTRY_STACK_PREALLOC 64
1623struct unpack_entry_stack_ent {
1624        off_t obj_offset;
1625        off_t curpos;
1626        unsigned long size;
1627};
1628
1629void *unpack_entry(struct packed_git *p, off_t obj_offset,
1630                   enum object_type *final_type, unsigned long *final_size)
1631{
1632        struct pack_window *w_curs = NULL;
1633        off_t curpos = obj_offset;
1634        void *data = NULL;
1635        unsigned long size;
1636        enum object_type type;
1637        struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1638        struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1639        int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1640        int base_from_cache = 0;
1641
1642        write_pack_access_log(p, obj_offset);
1643
1644        /* PHASE 1: drill down to the innermost base object */
1645        for (;;) {
1646                off_t base_offset;
1647                int i;
1648                struct delta_base_cache_entry *ent;
1649
1650                ent = get_delta_base_cache_entry(p, curpos);
1651                if (ent) {
1652                        type = ent->type;
1653                        data = ent->data;
1654                        size = ent->size;
1655                        detach_delta_base_cache_entry(ent);
1656                        base_from_cache = 1;
1657                        break;
1658                }
1659
1660                if (do_check_packed_object_crc && p->index_version > 1) {
1661                        struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1662                        off_t len = revidx[1].offset - obj_offset;
1663                        if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1664                                const unsigned char *sha1 =
1665                                        nth_packed_object_sha1(p, revidx->nr);
1666                                error("bad packed object CRC for %s",
1667                                      sha1_to_hex(sha1));
1668                                mark_bad_packed_object(p, sha1);
1669                                data = NULL;
1670                                goto out;
1671                        }
1672                }
1673
1674                type = unpack_object_header(p, &w_curs, &curpos, &size);
1675                if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1676                        break;
1677
1678                base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1679                if (!base_offset) {
1680                        error("failed to validate delta base reference "
1681                              "at offset %"PRIuMAX" from %s",
1682                              (uintmax_t)curpos, p->pack_name);
1683                        /* bail to phase 2, in hopes of recovery */
1684                        data = NULL;
1685                        break;
1686                }
1687
1688                /* push object, proceed to base */
1689                if (delta_stack_nr >= delta_stack_alloc
1690                    && delta_stack == small_delta_stack) {
1691                        delta_stack_alloc = alloc_nr(delta_stack_nr);
1692                        ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1693                        memcpy(delta_stack, small_delta_stack,
1694                               sizeof(*delta_stack)*delta_stack_nr);
1695                } else {
1696                        ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1697                }
1698                i = delta_stack_nr++;
1699                delta_stack[i].obj_offset = obj_offset;
1700                delta_stack[i].curpos = curpos;
1701                delta_stack[i].size = size;
1702
1703                curpos = obj_offset = base_offset;
1704        }
1705
1706        /* PHASE 2: handle the base */
1707        switch (type) {
1708        case OBJ_OFS_DELTA:
1709        case OBJ_REF_DELTA:
1710                if (data)
1711                        die("BUG: unpack_entry: left loop at a valid delta");
1712                break;
1713        case OBJ_COMMIT:
1714        case OBJ_TREE:
1715        case OBJ_BLOB:
1716        case OBJ_TAG:
1717                if (!base_from_cache)
1718                        data = unpack_compressed_entry(p, &w_curs, curpos, size);
1719                break;
1720        default:
1721                data = NULL;
1722                error("unknown object type %i at offset %"PRIuMAX" in %s",
1723                      type, (uintmax_t)obj_offset, p->pack_name);
1724        }
1725
1726        /* PHASE 3: apply deltas in order */
1727
1728        /* invariants:
1729         *   'data' holds the base data, or NULL if there was corruption
1730         */
1731        while (delta_stack_nr) {
1732                void *delta_data;
1733                void *base = data;
1734                void *external_base = NULL;
1735                unsigned long delta_size, base_size = size;
1736                int i;
1737
1738                data = NULL;
1739
1740                if (base)
1741                        add_delta_base_cache(p, obj_offset, base, base_size, type);
1742
1743                if (!base) {
1744                        /*
1745                         * We're probably in deep shit, but let's try to fetch
1746                         * the required base anyway from another pack or loose.
1747                         * This is costly but should happen only in the presence
1748                         * of a corrupted pack, and is better than failing outright.
1749                         */
1750                        struct revindex_entry *revidx;
1751                        const unsigned char *base_sha1;
1752                        revidx = find_pack_revindex(p, obj_offset);
1753                        if (revidx) {
1754                                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1755                                error("failed to read delta base object %s"
1756                                      " at offset %"PRIuMAX" from %s",
1757                                      sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
1758                                      p->pack_name);
1759                                mark_bad_packed_object(p, base_sha1);
1760                                base = read_object(base_sha1, &type, &base_size);
1761                                external_base = base;
1762                        }
1763                }
1764
1765                i = --delta_stack_nr;
1766                obj_offset = delta_stack[i].obj_offset;
1767                curpos = delta_stack[i].curpos;
1768                delta_size = delta_stack[i].size;
1769
1770                if (!base)
1771                        continue;
1772
1773                delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1774
1775                if (!delta_data) {
1776                        error("failed to unpack compressed delta "
1777                              "at offset %"PRIuMAX" from %s",
1778                              (uintmax_t)curpos, p->pack_name);
1779                        data = NULL;
1780                        free(external_base);
1781                        continue;
1782                }
1783
1784                data = patch_delta(base, base_size,
1785                                   delta_data, delta_size,
1786                                   &size);
1787
1788                /*
1789                 * We could not apply the delta; warn the user, but keep going.
1790                 * Our failure will be noticed either in the next iteration of
1791                 * the loop, or if this is the final delta, in the caller when
1792                 * we return NULL. Those code paths will take care of making
1793                 * a more explicit warning and retrying with another copy of
1794                 * the object.
1795                 */
1796                if (!data)
1797                        error("failed to apply delta");
1798
1799                free(delta_data);
1800                free(external_base);
1801        }
1802
1803        if (final_type)
1804                *final_type = type;
1805        if (final_size)
1806                *final_size = size;
1807
1808out:
1809        unuse_pack(&w_curs);
1810
1811        if (delta_stack != small_delta_stack)
1812                free(delta_stack);
1813
1814        return data;
1815}
1816
1817const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1818                                            uint32_t n)
1819{
1820        const unsigned char *index = p->index_data;
1821        if (!index) {
1822                if (open_pack_index(p))
1823                        return NULL;
1824                index = p->index_data;
1825        }
1826        if (n >= p->num_objects)
1827                return NULL;
1828        index += 4 * 256;
1829        if (p->index_version == 1) {
1830                return index + 24 * n + 4;
1831        } else {
1832                index += 8;
1833                return index + 20 * n;
1834        }
1835}
1836
1837const struct object_id *nth_packed_object_oid(struct object_id *oid,
1838                                              struct packed_git *p,
1839                                              uint32_t n)
1840{
1841        const unsigned char *hash = nth_packed_object_sha1(p, n);
1842        if (!hash)
1843                return NULL;
1844        hashcpy(oid->hash, hash);
1845        return oid;
1846}
1847
1848void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1849{
1850        const unsigned char *ptr = vptr;
1851        const unsigned char *start = p->index_data;
1852        const unsigned char *end = start + p->index_size;
1853        if (ptr < start)
1854                die(_("offset before start of pack index for %s (corrupt index?)"),
1855                    p->pack_name);
1856        /* No need to check for underflow; .idx files must be at least 8 bytes */
1857        if (ptr >= end - 8)
1858                die(_("offset beyond end of pack index for %s (truncated index?)"),
1859                    p->pack_name);
1860}
1861
1862off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1863{
1864        const unsigned char *index = p->index_data;
1865        index += 4 * 256;
1866        if (p->index_version == 1) {
1867                return ntohl(*((uint32_t *)(index + 24 * n)));
1868        } else {
1869                uint32_t off;
1870                index += 8 + p->num_objects * (20 + 4);
1871                off = ntohl(*((uint32_t *)(index + 4 * n)));
1872                if (!(off & 0x80000000))
1873                        return off;
1874                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1875                check_pack_index_ptr(p, index);
1876                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1877                                   ntohl(*((uint32_t *)(index + 4)));
1878        }
1879}
1880
1881off_t find_pack_entry_one(const unsigned char *sha1,
1882                                  struct packed_git *p)
1883{
1884        const uint32_t *level1_ofs = p->index_data;
1885        const unsigned char *index = p->index_data;
1886        unsigned hi, lo, stride;
1887        static int debug_lookup = -1;
1888
1889        if (debug_lookup < 0)
1890                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1891
1892        if (!index) {
1893                if (open_pack_index(p))
1894                        return 0;
1895                level1_ofs = p->index_data;
1896                index = p->index_data;
1897        }
1898        if (p->index_version > 1) {
1899                level1_ofs += 2;
1900                index += 8;
1901        }
1902        index += 4 * 256;
1903        hi = ntohl(level1_ofs[*sha1]);
1904        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1905        if (p->index_version > 1) {
1906                stride = 20;
1907        } else {
1908                stride = 24;
1909                index += 4;
1910        }
1911
1912        if (debug_lookup)
1913                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1914                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1915
1916        while (lo < hi) {
1917                unsigned mi = (lo + hi) / 2;
1918                int cmp = hashcmp(index + mi * stride, sha1);
1919
1920                if (debug_lookup)
1921                        printf("lo %u hi %u rg %u mi %u\n",
1922                               lo, hi, hi - lo, mi);
1923                if (!cmp)
1924                        return nth_packed_object_offset(p, mi);
1925                if (cmp > 0)
1926                        hi = mi;
1927                else
1928                        lo = mi+1;
1929        }
1930        return 0;
1931}
1932
1933int is_pack_valid(struct packed_git *p)
1934{
1935        /* An already open pack is known to be valid. */
1936        if (p->pack_fd != -1)
1937                return 1;
1938
1939        /* If the pack has one window completely covering the
1940         * file size, the pack is known to be valid even if
1941         * the descriptor is not currently open.
1942         */
1943        if (p->windows) {
1944                struct pack_window *w = p->windows;
1945
1946                if (!w->offset && w->len == p->pack_size)
1947                        return 1;
1948        }
1949
1950        /* Force the pack to open to prove its valid. */
1951        return !open_packed_git(p);
1952}
1953
1954static int fill_pack_entry(const unsigned char *sha1,
1955                           struct pack_entry *e,
1956                           struct packed_git *p)
1957{
1958        off_t offset;
1959
1960        if (p->num_bad_objects) {
1961                unsigned i;
1962                for (i = 0; i < p->num_bad_objects; i++)
1963                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1964                                return 0;
1965        }
1966
1967        offset = find_pack_entry_one(sha1, p);
1968        if (!offset)
1969                return 0;
1970
1971        /*
1972         * We are about to tell the caller where they can locate the
1973         * requested object.  We better make sure the packfile is
1974         * still here and can be accessed before supplying that
1975         * answer, as it may have been deleted since the index was
1976         * loaded!
1977         */
1978        if (!is_pack_valid(p))
1979                return 0;
1980        e->offset = offset;
1981        e->p = p;
1982        hashcpy(e->sha1, sha1);
1983        return 1;
1984}
1985
1986/*
1987 * Iff a pack file contains the object named by sha1, return true and
1988 * store its location to e.
1989 */
1990static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1991{
1992        struct mru_entry *p;
1993
1994        prepare_packed_git();
1995        if (!packed_git)
1996                return 0;
1997
1998        for (p = packed_git_mru->head; p; p = p->next) {
1999                if (fill_pack_entry(sha1, e, p->item)) {
2000                        mru_mark(packed_git_mru, p);
2001                        return 1;
2002                }
2003        }
2004        return 0;
2005}
2006
2007struct packed_git *find_sha1_pack(const unsigned char *sha1,
2008                                  struct packed_git *packs)
2009{
2010        struct packed_git *p;
2011
2012        for (p = packs; p; p = p->next) {
2013                if (find_pack_entry_one(sha1, p))
2014                        return p;
2015        }
2016        return NULL;
2017
2018}
2019
2020static int sha1_loose_object_info(const unsigned char *sha1,
2021                                  struct object_info *oi,
2022                                  int flags)
2023{
2024        int status = 0;
2025        unsigned long mapsize;
2026        void *map;
2027        git_zstream stream;
2028        char hdr[32];
2029        struct strbuf hdrbuf = STRBUF_INIT;
2030        unsigned long size_scratch;
2031
2032        if (oi->delta_base_sha1)
2033                hashclr(oi->delta_base_sha1);
2034
2035        /*
2036         * If we don't care about type or size, then we don't
2037         * need to look inside the object at all. Note that we
2038         * do not optimize out the stat call, even if the
2039         * caller doesn't care about the disk-size, since our
2040         * return value implicitly indicates whether the
2041         * object even exists.
2042         */
2043        if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
2044                const char *path;
2045                struct stat st;
2046                if (stat_sha1_file(sha1, &st, &path) < 0)
2047                        return -1;
2048                if (oi->disk_sizep)
2049                        *oi->disk_sizep = st.st_size;
2050                return 0;
2051        }
2052
2053        map = map_sha1_file(sha1, &mapsize);
2054        if (!map)
2055                return -1;
2056
2057        if (!oi->sizep)
2058                oi->sizep = &size_scratch;
2059
2060        if (oi->disk_sizep)
2061                *oi->disk_sizep = mapsize;
2062        if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
2063                if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2064                        status = error("unable to unpack %s header with --allow-unknown-type",
2065                                       sha1_to_hex(sha1));
2066        } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2067                status = error("unable to unpack %s header",
2068                               sha1_to_hex(sha1));
2069        if (status < 0)
2070                ; /* Do nothing */
2071        else if (hdrbuf.len) {
2072                if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2073                        status = error("unable to parse %s header with --allow-unknown-type",
2074                                       sha1_to_hex(sha1));
2075        } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2076                status = error("unable to parse %s header", sha1_to_hex(sha1));
2077
2078        if (status >= 0 && oi->contentp)
2079                *oi->contentp = unpack_sha1_rest(&stream, hdr,
2080                                                 *oi->sizep, sha1);
2081        else
2082                git_inflate_end(&stream);
2083
2084        munmap(map, mapsize);
2085        if (status && oi->typep)
2086                *oi->typep = status;
2087        if (oi->sizep == &size_scratch)
2088                oi->sizep = NULL;
2089        strbuf_release(&hdrbuf);
2090        oi->whence = OI_LOOSE;
2091        return (status < 0) ? status : 0;
2092}
2093
2094int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2095{
2096        static struct object_info blank_oi = OBJECT_INFO_INIT;
2097        struct pack_entry e;
2098        int rtype;
2099        const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
2100                                    lookup_replace_object(sha1) :
2101                                    sha1;
2102
2103        if (!oi)
2104                oi = &blank_oi;
2105
2106        if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
2107                struct cached_object *co = find_cached_object(real);
2108                if (co) {
2109                        if (oi->typep)
2110                                *(oi->typep) = co->type;
2111                        if (oi->sizep)
2112                                *(oi->sizep) = co->size;
2113                        if (oi->disk_sizep)
2114                                *(oi->disk_sizep) = 0;
2115                        if (oi->delta_base_sha1)
2116                                hashclr(oi->delta_base_sha1);
2117                        if (oi->typename)
2118                                strbuf_addstr(oi->typename, typename(co->type));
2119                        if (oi->contentp)
2120                                *oi->contentp = xmemdupz(co->buf, co->size);
2121                        oi->whence = OI_CACHED;
2122                        return 0;
2123                }
2124        }
2125
2126        if (!find_pack_entry(real, &e)) {
2127                /* Most likely it's a loose object. */
2128                if (!sha1_loose_object_info(real, oi, flags))
2129                        return 0;
2130
2131                /* Not a loose object; someone else may have just packed it. */
2132                if (flags & OBJECT_INFO_QUICK) {
2133                        return -1;
2134                } else {
2135                        reprepare_packed_git();
2136                        if (!find_pack_entry(real, &e))
2137                                return -1;
2138                }
2139        }
2140
2141        if (oi == &blank_oi)
2142                /*
2143                 * We know that the caller doesn't actually need the
2144                 * information below, so return early.
2145                 */
2146                return 0;
2147
2148        rtype = packed_object_info(e.p, e.offset, oi);
2149        if (rtype < 0) {
2150                mark_bad_packed_object(e.p, real);
2151                return sha1_object_info_extended(real, oi, 0);
2152        } else if (oi->whence == OI_PACKED) {
2153                oi->u.packed.offset = e.offset;
2154                oi->u.packed.pack = e.p;
2155                oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2156                                         rtype == OBJ_OFS_DELTA);
2157        }
2158
2159        return 0;
2160}
2161
2162/* returns enum object_type or negative */
2163int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2164{
2165        enum object_type type;
2166        struct object_info oi = OBJECT_INFO_INIT;
2167
2168        oi.typep = &type;
2169        oi.sizep = sizep;
2170        if (sha1_object_info_extended(sha1, &oi,
2171                                      OBJECT_INFO_LOOKUP_REPLACE) < 0)
2172                return -1;
2173        return type;
2174}
2175
2176int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2177                      unsigned char *sha1)
2178{
2179        struct cached_object *co;
2180
2181        hash_sha1_file(buf, len, typename(type), sha1);
2182        if (has_sha1_file(sha1) || find_cached_object(sha1))
2183                return 0;
2184        ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2185        co = &cached_objects[cached_object_nr++];
2186        co->size = len;
2187        co->type = type;
2188        co->buf = xmalloc(len);
2189        memcpy(co->buf, buf, len);
2190        hashcpy(co->sha1, sha1);
2191        return 0;
2192}
2193
2194static void *read_object(const unsigned char *sha1, enum object_type *type,
2195                         unsigned long *size)
2196{
2197        struct object_info oi = OBJECT_INFO_INIT;
2198        void *content;
2199        oi.typep = type;
2200        oi.sizep = size;
2201        oi.contentp = &content;
2202
2203        if (sha1_object_info_extended(sha1, &oi, 0) < 0)
2204                return NULL;
2205        return content;
2206}
2207
2208/*
2209 * This function dies on corrupt objects; the callers who want to
2210 * deal with them should arrange to call read_object() and give error
2211 * messages themselves.
2212 */
2213void *read_sha1_file_extended(const unsigned char *sha1,
2214                              enum object_type *type,
2215                              unsigned long *size,
2216                              int lookup_replace)
2217{
2218        void *data;
2219        const struct packed_git *p;
2220        const char *path;
2221        struct stat st;
2222        const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
2223                                                   : sha1;
2224
2225        errno = 0;
2226        data = read_object(repl, type, size);
2227        if (data)
2228                return data;
2229
2230        if (errno && errno != ENOENT)
2231                die_errno("failed to read object %s", sha1_to_hex(sha1));
2232
2233        /* die if we replaced an object with one that does not exist */
2234        if (repl != sha1)
2235                die("replacement %s not found for %s",
2236                    sha1_to_hex(repl), sha1_to_hex(sha1));
2237
2238        if (!stat_sha1_file(repl, &st, &path))
2239                die("loose object %s (stored in %s) is corrupt",
2240                    sha1_to_hex(repl), path);
2241
2242        if ((p = has_packed_and_bad(repl)) != NULL)
2243                die("packed object %s (stored in %s) is corrupt",
2244                    sha1_to_hex(repl), p->pack_name);
2245
2246        return NULL;
2247}
2248
2249void *read_object_with_reference(const unsigned char *sha1,
2250                                 const char *required_type_name,
2251                                 unsigned long *size,
2252                                 unsigned char *actual_sha1_return)
2253{
2254        enum object_type type, required_type;
2255        void *buffer;
2256        unsigned long isize;
2257        unsigned char actual_sha1[20];
2258
2259        required_type = type_from_string(required_type_name);
2260        hashcpy(actual_sha1, sha1);
2261        while (1) {
2262                int ref_length = -1;
2263                const char *ref_type = NULL;
2264
2265                buffer = read_sha1_file(actual_sha1, &type, &isize);
2266                if (!buffer)
2267                        return NULL;
2268                if (type == required_type) {
2269                        *size = isize;
2270                        if (actual_sha1_return)
2271                                hashcpy(actual_sha1_return, actual_sha1);
2272                        return buffer;
2273                }
2274                /* Handle references */
2275                else if (type == OBJ_COMMIT)
2276                        ref_type = "tree ";
2277                else if (type == OBJ_TAG)
2278                        ref_type = "object ";
2279                else {
2280                        free(buffer);
2281                        return NULL;
2282                }
2283                ref_length = strlen(ref_type);
2284
2285                if (ref_length + 40 > isize ||
2286                    memcmp(buffer, ref_type, ref_length) ||
2287                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2288                        free(buffer);
2289                        return NULL;
2290                }
2291                free(buffer);
2292                /* Now we have the ID of the referred-to object in
2293                 * actual_sha1.  Check again. */
2294        }
2295}
2296
2297static void write_sha1_file_prepare(const void *buf, unsigned long len,
2298                                    const char *type, unsigned char *sha1,
2299                                    char *hdr, int *hdrlen)
2300{
2301        git_SHA_CTX c;
2302
2303        /* Generate the header */
2304        *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
2305
2306        /* Sha1.. */
2307        git_SHA1_Init(&c);
2308        git_SHA1_Update(&c, hdr, *hdrlen);
2309        git_SHA1_Update(&c, buf, len);
2310        git_SHA1_Final(sha1, &c);
2311}
2312
2313/*
2314 * Move the just written object into its final resting place.
2315 */
2316int finalize_object_file(const char *tmpfile, const char *filename)
2317{
2318        int ret = 0;
2319
2320        if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2321                goto try_rename;
2322        else if (link(tmpfile, filename))
2323                ret = errno;
2324
2325        /*
2326         * Coda hack - coda doesn't like cross-directory links,
2327         * so we fall back to a rename, which will mean that it
2328         * won't be able to check collisions, but that's not a
2329         * big deal.
2330         *
2331         * The same holds for FAT formatted media.
2332         *
2333         * When this succeeds, we just return.  We have nothing
2334         * left to unlink.
2335         */
2336        if (ret && ret != EEXIST) {
2337        try_rename:
2338                if (!rename(tmpfile, filename))
2339                        goto out;
2340                ret = errno;
2341        }
2342        unlink_or_warn(tmpfile);
2343        if (ret) {
2344                if (ret != EEXIST) {
2345                        return error_errno("unable to write sha1 filename %s", filename);
2346                }
2347                /* FIXME!!! Collision check here ? */
2348        }
2349
2350out:
2351        if (adjust_shared_perm(filename))
2352                return error("unable to set permission to '%s'", filename);
2353        return 0;
2354}
2355
2356static int write_buffer(int fd, const void *buf, size_t len)
2357{
2358        if (write_in_full(fd, buf, len) < 0)
2359                return error_errno("file write error");
2360        return 0;
2361}
2362
2363int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2364                   unsigned char *sha1)
2365{
2366        char hdr[32];
2367        int hdrlen = sizeof(hdr);
2368        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2369        return 0;
2370}
2371
2372/* Finalize a file on disk, and close it. */
2373static void close_sha1_file(int fd)
2374{
2375        if (fsync_object_files)
2376                fsync_or_die(fd, "sha1 file");
2377        if (close(fd) != 0)
2378                die_errno("error when closing sha1 file");
2379}
2380
2381/* Size of directory component, including the ending '/' */
2382static inline int directory_size(const char *filename)
2383{
2384        const char *s = strrchr(filename, '/');
2385        if (!s)
2386                return 0;
2387        return s - filename + 1;
2388}
2389
2390/*
2391 * This creates a temporary file in the same directory as the final
2392 * 'filename'
2393 *
2394 * We want to avoid cross-directory filename renames, because those
2395 * can have problems on various filesystems (FAT, NFS, Coda).
2396 */
2397static int create_tmpfile(struct strbuf *tmp, const char *filename)
2398{
2399        int fd, dirlen = directory_size(filename);
2400
2401        strbuf_reset(tmp);
2402        strbuf_add(tmp, filename, dirlen);
2403        strbuf_addstr(tmp, "tmp_obj_XXXXXX");
2404        fd = git_mkstemp_mode(tmp->buf, 0444);
2405        if (fd < 0 && dirlen && errno == ENOENT) {
2406                /*
2407                 * Make sure the directory exists; note that the contents
2408                 * of the buffer are undefined after mkstemp returns an
2409                 * error, so we have to rewrite the whole buffer from
2410                 * scratch.
2411                 */
2412                strbuf_reset(tmp);
2413                strbuf_add(tmp, filename, dirlen - 1);
2414                if (mkdir(tmp->buf, 0777) && errno != EEXIST)
2415                        return -1;
2416                if (adjust_shared_perm(tmp->buf))
2417                        return -1;
2418
2419                /* Try again */
2420                strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
2421                fd = git_mkstemp_mode(tmp->buf, 0444);
2422        }
2423        return fd;
2424}
2425
2426static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2427                              const void *buf, unsigned long len, time_t mtime)
2428{
2429        int fd, ret;
2430        unsigned char compressed[4096];
2431        git_zstream stream;
2432        git_SHA_CTX c;
2433        unsigned char parano_sha1[20];
2434        static struct strbuf tmp_file = STRBUF_INIT;
2435        const char *filename = sha1_file_name(sha1);
2436
2437        fd = create_tmpfile(&tmp_file, filename);
2438        if (fd < 0) {
2439                if (errno == EACCES)
2440                        return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2441                else
2442                        return error_errno("unable to create temporary file");
2443        }
2444
2445        /* Set it up */
2446        git_deflate_init(&stream, zlib_compression_level);
2447        stream.next_out = compressed;
2448        stream.avail_out = sizeof(compressed);
2449        git_SHA1_Init(&c);
2450
2451        /* First header.. */
2452        stream.next_in = (unsigned char *)hdr;
2453        stream.avail_in = hdrlen;
2454        while (git_deflate(&stream, 0) == Z_OK)
2455                ; /* nothing */
2456        git_SHA1_Update(&c, hdr, hdrlen);
2457
2458        /* Then the data itself.. */
2459        stream.next_in = (void *)buf;
2460        stream.avail_in = len;
2461        do {
2462                unsigned char *in0 = stream.next_in;
2463                ret = git_deflate(&stream, Z_FINISH);
2464                git_SHA1_Update(&c, in0, stream.next_in - in0);
2465                if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2466                        die("unable to write sha1 file");
2467                stream.next_out = compressed;
2468                stream.avail_out = sizeof(compressed);
2469        } while (ret == Z_OK);
2470
2471        if (ret != Z_STREAM_END)
2472                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2473        ret = git_deflate_end_gently(&stream);
2474        if (ret != Z_OK)
2475                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2476        git_SHA1_Final(parano_sha1, &c);
2477        if (hashcmp(sha1, parano_sha1) != 0)
2478                die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2479
2480        close_sha1_file(fd);
2481
2482        if (mtime) {
2483                struct utimbuf utb;
2484                utb.actime = mtime;
2485                utb.modtime = mtime;
2486                if (utime(tmp_file.buf, &utb) < 0)
2487                        warning_errno("failed utime() on %s", tmp_file.buf);
2488        }
2489
2490        return finalize_object_file(tmp_file.buf, filename);
2491}
2492
2493static int freshen_loose_object(const unsigned char *sha1)
2494{
2495        return check_and_freshen(sha1, 1);
2496}
2497
2498static int freshen_packed_object(const unsigned char *sha1)
2499{
2500        struct pack_entry e;
2501        if (!find_pack_entry(sha1, &e))
2502                return 0;
2503        if (e.p->freshened)
2504                return 1;
2505        if (!freshen_file(e.p->pack_name))
2506                return 0;
2507        e.p->freshened = 1;
2508        return 1;
2509}
2510
2511int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
2512{
2513        char hdr[32];
2514        int hdrlen = sizeof(hdr);
2515
2516        /* Normally if we have it in the pack then we do not bother writing
2517         * it out into .git/objects/??/?{38} file.
2518         */
2519        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2520        if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
2521                return 0;
2522        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2523}
2524
2525int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
2526                             unsigned char *sha1, unsigned flags)
2527{
2528        char *header;
2529        int hdrlen, status = 0;
2530
2531        /* type string, SP, %lu of the length plus NUL must fit this */
2532        hdrlen = strlen(type) + 32;
2533        header = xmalloc(hdrlen);
2534        write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
2535
2536        if (!(flags & HASH_WRITE_OBJECT))
2537                goto cleanup;
2538        if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
2539                goto cleanup;
2540        status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
2541
2542cleanup:
2543        free(header);
2544        return status;
2545}
2546
2547int force_object_loose(const unsigned char *sha1, time_t mtime)
2548{
2549        void *buf;
2550        unsigned long len;
2551        enum object_type type;
2552        char hdr[32];
2553        int hdrlen;
2554        int ret;
2555
2556        if (has_loose_object(sha1))
2557                return 0;
2558        buf = read_object(sha1, &type, &len);
2559        if (!buf)
2560                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2561        hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
2562        ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2563        free(buf);
2564
2565        return ret;
2566}
2567
2568int has_pack_index(const unsigned char *sha1)
2569{
2570        struct stat st;
2571        if (stat(sha1_pack_index_name(sha1), &st))
2572                return 0;
2573        return 1;
2574}
2575
2576int has_sha1_pack(const unsigned char *sha1)
2577{
2578        struct pack_entry e;
2579        return find_pack_entry(sha1, &e);
2580}
2581
2582int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
2583{
2584        if (!startup_info->have_repository)
2585                return 0;
2586        return sha1_object_info_extended(sha1, NULL,
2587                                         flags | OBJECT_INFO_SKIP_CACHED) >= 0;
2588}
2589
2590int has_object_file(const struct object_id *oid)
2591{
2592        return has_sha1_file(oid->hash);
2593}
2594
2595int has_object_file_with_flags(const struct object_id *oid, int flags)
2596{
2597        return has_sha1_file_with_flags(oid->hash, flags);
2598}
2599
2600static void check_tree(const void *buf, size_t size)
2601{
2602        struct tree_desc desc;
2603        struct name_entry entry;
2604
2605        init_tree_desc(&desc, buf, size);
2606        while (tree_entry(&desc, &entry))
2607                /* do nothing
2608                 * tree_entry() will die() on malformed entries */
2609                ;
2610}
2611
2612static void check_commit(const void *buf, size_t size)
2613{
2614        struct commit c;
2615        memset(&c, 0, sizeof(c));
2616        if (parse_commit_buffer(&c, buf, size))
2617                die("corrupt commit");
2618}
2619
2620static void check_tag(const void *buf, size_t size)
2621{
2622        struct tag t;
2623        memset(&t, 0, sizeof(t));
2624        if (parse_tag_buffer(&t, buf, size))
2625                die("corrupt tag");
2626}
2627
2628static int index_mem(unsigned char *sha1, void *buf, size_t size,
2629                     enum object_type type,
2630                     const char *path, unsigned flags)
2631{
2632        int ret, re_allocated = 0;
2633        int write_object = flags & HASH_WRITE_OBJECT;
2634
2635        if (!type)
2636                type = OBJ_BLOB;
2637
2638        /*
2639         * Convert blobs to git internal format
2640         */
2641        if ((type == OBJ_BLOB) && path) {
2642                struct strbuf nbuf = STRBUF_INIT;
2643                if (convert_to_git(&the_index, path, buf, size, &nbuf,
2644                                   write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2645                        buf = strbuf_detach(&nbuf, &size);
2646                        re_allocated = 1;
2647                }
2648        }
2649        if (flags & HASH_FORMAT_CHECK) {
2650                if (type == OBJ_TREE)
2651                        check_tree(buf, size);
2652                if (type == OBJ_COMMIT)
2653                        check_commit(buf, size);
2654                if (type == OBJ_TAG)
2655                        check_tag(buf, size);
2656        }
2657
2658        if (write_object)
2659                ret = write_sha1_file(buf, size, typename(type), sha1);
2660        else
2661                ret = hash_sha1_file(buf, size, typename(type), sha1);
2662        if (re_allocated)
2663                free(buf);
2664        return ret;
2665}
2666
2667static int index_stream_convert_blob(unsigned char *sha1, int fd,
2668                                     const char *path, unsigned flags)
2669{
2670        int ret;
2671        const int write_object = flags & HASH_WRITE_OBJECT;
2672        struct strbuf sbuf = STRBUF_INIT;
2673
2674        assert(path);
2675        assert(would_convert_to_git_filter_fd(path));
2676
2677        convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
2678                                 write_object ? safe_crlf : SAFE_CRLF_FALSE);
2679
2680        if (write_object)
2681                ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
2682                                      sha1);
2683        else
2684                ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
2685                                     sha1);
2686        strbuf_release(&sbuf);
2687        return ret;
2688}
2689
2690static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2691                      const char *path, unsigned flags)
2692{
2693        struct strbuf sbuf = STRBUF_INIT;
2694        int ret;
2695
2696        if (strbuf_read(&sbuf, fd, 4096) >= 0)
2697                ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2698        else
2699                ret = -1;
2700        strbuf_release(&sbuf);
2701        return ret;
2702}
2703
2704#define SMALL_FILE_SIZE (32*1024)
2705
2706static int index_core(unsigned char *sha1, int fd, size_t size,
2707                      enum object_type type, const char *path,
2708                      unsigned flags)
2709{
2710        int ret;
2711
2712        if (!size) {
2713                ret = index_mem(sha1, "", size, type, path, flags);
2714        } else if (size <= SMALL_FILE_SIZE) {
2715                char *buf = xmalloc(size);
2716                if (size == read_in_full(fd, buf, size))
2717                        ret = index_mem(sha1, buf, size, type, path, flags);
2718                else
2719                        ret = error_errno("short read");
2720                free(buf);
2721        } else {
2722                void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2723                ret = index_mem(sha1, buf, size, type, path, flags);
2724                munmap(buf, size);
2725        }
2726        return ret;
2727}
2728
2729/*
2730 * This creates one packfile per large blob unless bulk-checkin
2731 * machinery is "plugged".
2732 *
2733 * This also bypasses the usual "convert-to-git" dance, and that is on
2734 * purpose. We could write a streaming version of the converting
2735 * functions and insert that before feeding the data to fast-import
2736 * (or equivalent in-core API described above). However, that is
2737 * somewhat complicated, as we do not know the size of the filter
2738 * result, which we need to know beforehand when writing a git object.
2739 * Since the primary motivation for trying to stream from the working
2740 * tree file and to avoid mmaping it in core is to deal with large
2741 * binary blobs, they generally do not want to get any conversion, and
2742 * callers should avoid this code path when filters are requested.
2743 */
2744static int index_stream(unsigned char *sha1, int fd, size_t size,
2745                        enum object_type type, const char *path,
2746                        unsigned flags)
2747{
2748        return index_bulk_checkin(sha1, fd, size, type, path, flags);
2749}
2750
2751int index_fd(unsigned char *sha1, int fd, struct stat *st,
2752             enum object_type type, const char *path, unsigned flags)
2753{
2754        int ret;
2755
2756        /*
2757         * Call xsize_t() only when needed to avoid potentially unnecessary
2758         * die() for large files.
2759         */
2760        if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
2761                ret = index_stream_convert_blob(sha1, fd, path, flags);
2762        else if (!S_ISREG(st->st_mode))
2763                ret = index_pipe(sha1, fd, type, path, flags);
2764        else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2765                 (path && would_convert_to_git(&the_index, path)))
2766                ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
2767                                 flags);
2768        else
2769                ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
2770                                   flags);
2771        close(fd);
2772        return ret;
2773}
2774
2775int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2776{
2777        int fd;
2778        struct strbuf sb = STRBUF_INIT;
2779
2780        switch (st->st_mode & S_IFMT) {
2781        case S_IFREG:
2782                fd = open(path, O_RDONLY);
2783                if (fd < 0)
2784                        return error_errno("open(\"%s\")", path);
2785                if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2786                        return error("%s: failed to insert into database",
2787                                     path);
2788                break;
2789        case S_IFLNK:
2790                if (strbuf_readlink(&sb, path, st->st_size))
2791                        return error_errno("readlink(\"%s\")", path);
2792                if (!(flags & HASH_WRITE_OBJECT))
2793                        hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2794                else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2795                        return error("%s: failed to insert into database",
2796                                     path);
2797                strbuf_release(&sb);
2798                break;
2799        case S_IFDIR:
2800                return resolve_gitlink_ref(path, "HEAD", sha1);
2801        default:
2802                return error("%s: unsupported file type", path);
2803        }
2804        return 0;
2805}
2806
2807int read_pack_header(int fd, struct pack_header *header)
2808{
2809        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2810                /* "eof before pack header was fully read" */
2811                return PH_ERROR_EOF;
2812
2813        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2814                /* "protocol error (pack signature mismatch detected)" */
2815                return PH_ERROR_PACK_SIGNATURE;
2816        if (!pack_version_ok(header->hdr_version))
2817                /* "protocol error (pack version unsupported)" */
2818                return PH_ERROR_PROTOCOL;
2819        return 0;
2820}
2821
2822void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2823{
2824        enum object_type type = sha1_object_info(sha1, NULL);
2825        if (type < 0)
2826                die("%s is not a valid object", sha1_to_hex(sha1));
2827        if (type != expect)
2828                die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2829                    typename(expect));
2830}
2831
2832int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2833                                struct strbuf *path,
2834                                each_loose_object_fn obj_cb,
2835                                each_loose_cruft_fn cruft_cb,
2836                                each_loose_subdir_fn subdir_cb,
2837                                void *data)
2838{
2839        size_t origlen, baselen;
2840        DIR *dir;
2841        struct dirent *de;
2842        int r = 0;
2843
2844        if (subdir_nr > 0xff)
2845                BUG("invalid loose object subdirectory: %x", subdir_nr);
2846
2847        origlen = path->len;
2848        strbuf_complete(path, '/');
2849        strbuf_addf(path, "%02x", subdir_nr);
2850        baselen = path->len;
2851
2852        dir = opendir(path->buf);
2853        if (!dir) {
2854                if (errno != ENOENT)
2855                        r = error_errno("unable to open %s", path->buf);
2856                strbuf_setlen(path, origlen);
2857                return r;
2858        }
2859
2860        while ((de = readdir(dir))) {
2861                if (is_dot_or_dotdot(de->d_name))
2862                        continue;
2863
2864                strbuf_setlen(path, baselen);
2865                strbuf_addf(path, "/%s", de->d_name);
2866
2867                if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2)  {
2868                        char hex[GIT_MAX_HEXSZ+1];
2869                        struct object_id oid;
2870
2871                        xsnprintf(hex, sizeof(hex), "%02x%s",
2872                                  subdir_nr, de->d_name);
2873                        if (!get_oid_hex(hex, &oid)) {
2874                                if (obj_cb) {
2875                                        r = obj_cb(&oid, path->buf, data);
2876                                        if (r)
2877                                                break;
2878                                }
2879                                continue;
2880                        }
2881                }
2882
2883                if (cruft_cb) {
2884                        r = cruft_cb(de->d_name, path->buf, data);
2885                        if (r)
2886                                break;
2887                }
2888        }
2889        closedir(dir);
2890
2891        strbuf_setlen(path, baselen);
2892        if (!r && subdir_cb)
2893                r = subdir_cb(subdir_nr, path->buf, data);
2894
2895        strbuf_setlen(path, origlen);
2896
2897        return r;
2898}
2899
2900int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2901                            each_loose_object_fn obj_cb,
2902                            each_loose_cruft_fn cruft_cb,
2903                            each_loose_subdir_fn subdir_cb,
2904                            void *data)
2905{
2906        int r = 0;
2907        int i;
2908
2909        for (i = 0; i < 256; i++) {
2910                r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2911                                                subdir_cb, data);
2912                if (r)
2913                        break;
2914        }
2915
2916        return r;
2917}
2918
2919int for_each_loose_file_in_objdir(const char *path,
2920                                  each_loose_object_fn obj_cb,
2921                                  each_loose_cruft_fn cruft_cb,
2922                                  each_loose_subdir_fn subdir_cb,
2923                                  void *data)
2924{
2925        struct strbuf buf = STRBUF_INIT;
2926        int r;
2927
2928        strbuf_addstr(&buf, path);
2929        r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2930                                              subdir_cb, data);
2931        strbuf_release(&buf);
2932
2933        return r;
2934}
2935
2936struct loose_alt_odb_data {
2937        each_loose_object_fn *cb;
2938        void *data;
2939};
2940
2941static int loose_from_alt_odb(struct alternate_object_database *alt,
2942                              void *vdata)
2943{
2944        struct loose_alt_odb_data *data = vdata;
2945        struct strbuf buf = STRBUF_INIT;
2946        int r;
2947
2948        strbuf_addstr(&buf, alt->path);
2949        r = for_each_loose_file_in_objdir_buf(&buf,
2950                                              data->cb, NULL, NULL,
2951                                              data->data);
2952        strbuf_release(&buf);
2953        return r;
2954}
2955
2956int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2957{
2958        struct loose_alt_odb_data alt;
2959        int r;
2960
2961        r = for_each_loose_file_in_objdir(get_object_directory(),
2962                                          cb, NULL, NULL, data);
2963        if (r)
2964                return r;
2965
2966        if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2967                return 0;
2968
2969        alt.cb = cb;
2970        alt.data = data;
2971        return foreach_alt_odb(loose_from_alt_odb, &alt);
2972}
2973
2974static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
2975{
2976        uint32_t i;
2977        int r = 0;
2978
2979        for (i = 0; i < p->num_objects; i++) {
2980                struct object_id oid;
2981
2982                if (!nth_packed_object_oid(&oid, p, i))
2983                        return error("unable to get sha1 of object %u in %s",
2984                                     i, p->pack_name);
2985
2986                r = cb(&oid, p, i, data);
2987                if (r)
2988                        break;
2989        }
2990        return r;
2991}
2992
2993int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
2994{
2995        struct packed_git *p;
2996        int r = 0;
2997        int pack_errors = 0;
2998
2999        prepare_packed_git();
3000        for (p = packed_git; p; p = p->next) {
3001                if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3002                        continue;
3003                if (open_pack_index(p)) {
3004                        pack_errors = 1;
3005                        continue;
3006                }
3007                r = for_each_object_in_pack(p, cb, data);
3008                if (r)
3009                        break;
3010        }
3011        return r ? r : pack_errors;
3012}
3013
3014static int check_stream_sha1(git_zstream *stream,
3015                             const char *hdr,
3016                             unsigned long size,
3017                             const char *path,
3018                             const unsigned char *expected_sha1)
3019{
3020        git_SHA_CTX c;
3021        unsigned char real_sha1[GIT_MAX_RAWSZ];
3022        unsigned char buf[4096];
3023        unsigned long total_read;
3024        int status = Z_OK;
3025
3026        git_SHA1_Init(&c);
3027        git_SHA1_Update(&c, hdr, stream->total_out);
3028
3029        /*
3030         * We already read some bytes into hdr, but the ones up to the NUL
3031         * do not count against the object's content size.
3032         */
3033        total_read = stream->total_out - strlen(hdr) - 1;
3034
3035        /*
3036         * This size comparison must be "<=" to read the final zlib packets;
3037         * see the comment in unpack_sha1_rest for details.
3038         */
3039        while (total_read <= size &&
3040               (status == Z_OK || status == Z_BUF_ERROR)) {
3041                stream->next_out = buf;
3042                stream->avail_out = sizeof(buf);
3043                if (size - total_read < stream->avail_out)
3044                        stream->avail_out = size - total_read;
3045                status = git_inflate(stream, Z_FINISH);
3046                git_SHA1_Update(&c, buf, stream->next_out - buf);
3047                total_read += stream->next_out - buf;
3048        }
3049        git_inflate_end(stream);
3050
3051        if (status != Z_STREAM_END) {
3052                error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
3053                return -1;
3054        }
3055        if (stream->avail_in) {
3056                error("garbage at end of loose object '%s'",
3057                      sha1_to_hex(expected_sha1));
3058                return -1;
3059        }
3060
3061        git_SHA1_Final(real_sha1, &c);
3062        if (hashcmp(expected_sha1, real_sha1)) {
3063                error("sha1 mismatch for %s (expected %s)", path,
3064                      sha1_to_hex(expected_sha1));
3065                return -1;
3066        }
3067
3068        return 0;
3069}
3070
3071int read_loose_object(const char *path,
3072                      const unsigned char *expected_sha1,
3073                      enum object_type *type,
3074                      unsigned long *size,
3075                      void **contents)
3076{
3077        int ret = -1;
3078        void *map = NULL;
3079        unsigned long mapsize;
3080        git_zstream stream;
3081        char hdr[32];
3082
3083        *contents = NULL;
3084
3085        map = map_sha1_file_1(path, NULL, &mapsize);
3086        if (!map) {
3087                error_errno("unable to mmap %s", path);
3088                goto out;
3089        }
3090
3091        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
3092                error("unable to unpack header of %s", path);
3093                goto out;
3094        }
3095
3096        *type = parse_sha1_header(hdr, size);
3097        if (*type < 0) {
3098                error("unable to parse header of %s", path);
3099                git_inflate_end(&stream);
3100                goto out;
3101        }
3102
3103        if (*type == OBJ_BLOB) {
3104                if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
3105                        goto out;
3106        } else {
3107                *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
3108                if (!*contents) {
3109                        error("unable to unpack contents of %s", path);
3110                        git_inflate_end(&stream);
3111                        goto out;
3112                }
3113                if (check_sha1_signature(expected_sha1, *contents,
3114                                         *size, typename(*type))) {
3115                        error("sha1 mismatch for %s (expected %s)", path,
3116                              sha1_to_hex(expected_sha1));
3117                        free(*contents);
3118                        goto out;
3119                }
3120        }
3121
3122        ret = 0; /* everything checks out */
3123
3124out:
3125        if (map)
3126                munmap(map, mapsize);
3127        return ret;
3128}