6c8e4b43dd1eadc4d7ec213d6673d6ceebbcf5d5
   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 "string-list.h"
  11#include "lockfile.h"
  12#include "delta.h"
  13#include "pack.h"
  14#include "blob.h"
  15#include "commit.h"
  16#include "run-command.h"
  17#include "tag.h"
  18#include "tree.h"
  19#include "tree-walk.h"
  20#include "refs.h"
  21#include "pack-revindex.h"
  22#include "sha1-lookup.h"
  23#include "bulk-checkin.h"
  24#include "streaming.h"
  25#include "dir.h"
  26#include "mru.h"
  27#include "list.h"
  28#include "mergesort.h"
  29
  30#ifndef O_NOATIME
  31#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  32#define O_NOATIME 01000000
  33#else
  34#define O_NOATIME 0
  35#endif
  36#endif
  37
  38#define SZ_FMT PRIuMAX
  39static inline uintmax_t sz_fmt(size_t s) { return s; }
  40
  41const unsigned char null_sha1[20];
  42const struct object_id null_oid;
  43const struct object_id empty_tree_oid = {
  44        EMPTY_TREE_SHA1_BIN_LITERAL
  45};
  46const struct object_id empty_blob_oid = {
  47        EMPTY_BLOB_SHA1_BIN_LITERAL
  48};
  49
  50/*
  51 * This is meant to hold a *small* number of objects that you would
  52 * want read_sha1_file() to be able to return, but yet you do not want
  53 * to write them into the object store (e.g. a browse-only
  54 * application).
  55 */
  56static struct cached_object {
  57        unsigned char sha1[20];
  58        enum object_type type;
  59        void *buf;
  60        unsigned long size;
  61} *cached_objects;
  62static int cached_object_nr, cached_object_alloc;
  63
  64static struct cached_object empty_tree = {
  65        EMPTY_TREE_SHA1_BIN_LITERAL,
  66        OBJ_TREE,
  67        "",
  68        0
  69};
  70
  71static struct cached_object *find_cached_object(const unsigned char *sha1)
  72{
  73        int i;
  74        struct cached_object *co = cached_objects;
  75
  76        for (i = 0; i < cached_object_nr; i++, co++) {
  77                if (!hashcmp(co->sha1, sha1))
  78                        return co;
  79        }
  80        if (!hashcmp(sha1, empty_tree.sha1))
  81                return &empty_tree;
  82        return NULL;
  83}
  84
  85int mkdir_in_gitdir(const char *path)
  86{
  87        if (mkdir(path, 0777)) {
  88                int saved_errno = errno;
  89                struct stat st;
  90                struct strbuf sb = STRBUF_INIT;
  91
  92                if (errno != EEXIST)
  93                        return -1;
  94                /*
  95                 * Are we looking at a path in a symlinked worktree
  96                 * whose original repository does not yet have it?
  97                 * e.g. .git/rr-cache pointing at its original
  98                 * repository in which the user hasn't performed any
  99                 * conflict resolution yet?
 100                 */
 101                if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
 102                    strbuf_readlink(&sb, path, st.st_size) ||
 103                    !is_absolute_path(sb.buf) ||
 104                    mkdir(sb.buf, 0777)) {
 105                        strbuf_release(&sb);
 106                        errno = saved_errno;
 107                        return -1;
 108                }
 109                strbuf_release(&sb);
 110        }
 111        return adjust_shared_perm(path);
 112}
 113
 114enum scld_error safe_create_leading_directories(char *path)
 115{
 116        char *next_component = path + offset_1st_component(path);
 117        enum scld_error ret = SCLD_OK;
 118
 119        while (ret == SCLD_OK && next_component) {
 120                struct stat st;
 121                char *slash = next_component, slash_character;
 122
 123                while (*slash && !is_dir_sep(*slash))
 124                        slash++;
 125
 126                if (!*slash)
 127                        break;
 128
 129                next_component = slash + 1;
 130                while (is_dir_sep(*next_component))
 131                        next_component++;
 132                if (!*next_component)
 133                        break;
 134
 135                slash_character = *slash;
 136                *slash = '\0';
 137                if (!stat(path, &st)) {
 138                        /* path exists */
 139                        if (!S_ISDIR(st.st_mode))
 140                                ret = SCLD_EXISTS;
 141                } else if (mkdir(path, 0777)) {
 142                        if (errno == EEXIST &&
 143                            !stat(path, &st) && S_ISDIR(st.st_mode))
 144                                ; /* somebody created it since we checked */
 145                        else if (errno == ENOENT)
 146                                /*
 147                                 * Either mkdir() failed because
 148                                 * somebody just pruned the containing
 149                                 * directory, or stat() failed because
 150                                 * the file that was in our way was
 151                                 * just removed.  Either way, inform
 152                                 * the caller that it might be worth
 153                                 * trying again:
 154                                 */
 155                                ret = SCLD_VANISHED;
 156                        else
 157                                ret = SCLD_FAILED;
 158                } else if (adjust_shared_perm(path)) {
 159                        ret = SCLD_PERMS;
 160                }
 161                *slash = slash_character;
 162        }
 163        return ret;
 164}
 165
 166enum scld_error safe_create_leading_directories_const(const char *path)
 167{
 168        /* path points to cache entries, so xstrdup before messing with it */
 169        char *buf = xstrdup(path);
 170        enum scld_error result = safe_create_leading_directories(buf);
 171        free(buf);
 172        return result;
 173}
 174
 175static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
 176{
 177        int i;
 178        for (i = 0; i < 20; i++) {
 179                static char hex[] = "0123456789abcdef";
 180                unsigned int val = sha1[i];
 181                strbuf_addch(buf, hex[val >> 4]);
 182                strbuf_addch(buf, hex[val & 0xf]);
 183                if (!i)
 184                        strbuf_addch(buf, '/');
 185        }
 186}
 187
 188const char *sha1_file_name(const unsigned char *sha1)
 189{
 190        static struct strbuf buf = STRBUF_INIT;
 191
 192        strbuf_reset(&buf);
 193        strbuf_addf(&buf, "%s/", get_object_directory());
 194
 195        fill_sha1_path(&buf, sha1);
 196        return buf.buf;
 197}
 198
 199struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
 200{
 201        strbuf_setlen(&alt->scratch, alt->base_len);
 202        return &alt->scratch;
 203}
 204
 205static const char *alt_sha1_path(struct alternate_object_database *alt,
 206                                 const unsigned char *sha1)
 207{
 208        struct strbuf *buf = alt_scratch_buf(alt);
 209        fill_sha1_path(buf, sha1);
 210        return buf->buf;
 211}
 212
 213/*
 214 * Return the name of the pack or index file with the specified sha1
 215 * in its filename.  *base and *name are scratch space that must be
 216 * provided by the caller.  which should be "pack" or "idx".
 217 */
 218static char *sha1_get_pack_name(const unsigned char *sha1,
 219                                struct strbuf *buf,
 220                                const char *which)
 221{
 222        strbuf_reset(buf);
 223        strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
 224                    sha1_to_hex(sha1), which);
 225        return buf->buf;
 226}
 227
 228char *sha1_pack_name(const unsigned char *sha1)
 229{
 230        static struct strbuf buf = STRBUF_INIT;
 231        return sha1_get_pack_name(sha1, &buf, "pack");
 232}
 233
 234char *sha1_pack_index_name(const unsigned char *sha1)
 235{
 236        static struct strbuf buf = STRBUF_INIT;
 237        return sha1_get_pack_name(sha1, &buf, "idx");
 238}
 239
 240struct alternate_object_database *alt_odb_list;
 241static struct alternate_object_database **alt_odb_tail;
 242
 243/*
 244 * Return non-zero iff the path is usable as an alternate object database.
 245 */
 246static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
 247{
 248        struct alternate_object_database *alt;
 249
 250        /* Detect cases where alternate disappeared */
 251        if (!is_directory(path->buf)) {
 252                error("object directory %s does not exist; "
 253                      "check .git/objects/info/alternates.",
 254                      path->buf);
 255                return 0;
 256        }
 257
 258        /*
 259         * Prevent the common mistake of listing the same
 260         * thing twice, or object directory itself.
 261         */
 262        for (alt = alt_odb_list; alt; alt = alt->next) {
 263                if (!fspathcmp(path->buf, alt->path))
 264                        return 0;
 265        }
 266        if (!fspathcmp(path->buf, normalized_objdir))
 267                return 0;
 268
 269        return 1;
 270}
 271
 272/*
 273 * Prepare alternate object database registry.
 274 *
 275 * The variable alt_odb_list points at the list of struct
 276 * alternate_object_database.  The elements on this list come from
 277 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 278 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 279 * whose contents is similar to that environment variable but can be
 280 * LF separated.  Its base points at a statically allocated buffer that
 281 * contains "/the/directory/corresponding/to/.git/objects/...", while
 282 * its name points just after the slash at the end of ".git/objects/"
 283 * in the example above, and has enough space to hold 40-byte hex
 284 * SHA1, an extra slash for the first level indirection, and the
 285 * terminating NUL.
 286 */
 287static int link_alt_odb_entry(const char *entry, const char *relative_base,
 288        int depth, const char *normalized_objdir)
 289{
 290        struct alternate_object_database *ent;
 291        struct strbuf pathbuf = STRBUF_INIT;
 292
 293        if (!is_absolute_path(entry) && relative_base) {
 294                strbuf_addstr(&pathbuf, real_path(relative_base));
 295                strbuf_addch(&pathbuf, '/');
 296        }
 297        strbuf_addstr(&pathbuf, entry);
 298
 299        if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
 300                error("unable to normalize alternate object path: %s",
 301                      pathbuf.buf);
 302                strbuf_release(&pathbuf);
 303                return -1;
 304        }
 305
 306        /*
 307         * The trailing slash after the directory name is given by
 308         * this function at the end. Remove duplicates.
 309         */
 310        while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
 311                strbuf_setlen(&pathbuf, pathbuf.len - 1);
 312
 313        if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
 314                strbuf_release(&pathbuf);
 315                return -1;
 316        }
 317
 318        ent = alloc_alt_odb(pathbuf.buf);
 319
 320        /* add the alternate entry */
 321        *alt_odb_tail = ent;
 322        alt_odb_tail = &(ent->next);
 323        ent->next = NULL;
 324
 325        /* recursively add alternates */
 326        read_info_alternates(pathbuf.buf, depth + 1);
 327
 328        strbuf_release(&pathbuf);
 329        return 0;
 330}
 331
 332static void link_alt_odb_entries(const char *alt, int len, int sep,
 333                                 const char *relative_base, int depth)
 334{
 335        struct string_list entries = STRING_LIST_INIT_NODUP;
 336        char *alt_copy;
 337        int i;
 338        struct strbuf objdirbuf = STRBUF_INIT;
 339
 340        if (depth > 5) {
 341                error("%s: ignoring alternate object stores, nesting too deep.",
 342                                relative_base);
 343                return;
 344        }
 345
 346        strbuf_add_absolute_path(&objdirbuf, get_object_directory());
 347        if (strbuf_normalize_path(&objdirbuf) < 0)
 348                die("unable to normalize object directory: %s",
 349                    objdirbuf.buf);
 350
 351        alt_copy = xmemdupz(alt, len);
 352        string_list_split_in_place(&entries, alt_copy, sep, -1);
 353        for (i = 0; i < entries.nr; i++) {
 354                const char *entry = entries.items[i].string;
 355                if (entry[0] == '\0' || entry[0] == '#')
 356                        continue;
 357                link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
 358        }
 359        string_list_clear(&entries, 0);
 360        free(alt_copy);
 361        strbuf_release(&objdirbuf);
 362}
 363
 364void read_info_alternates(const char * relative_base, int depth)
 365{
 366        char *map;
 367        size_t mapsz;
 368        struct stat st;
 369        char *path;
 370        int fd;
 371
 372        path = xstrfmt("%s/info/alternates", relative_base);
 373        fd = git_open(path);
 374        free(path);
 375        if (fd < 0)
 376                return;
 377        if (fstat(fd, &st) || (st.st_size == 0)) {
 378                close(fd);
 379                return;
 380        }
 381        mapsz = xsize_t(st.st_size);
 382        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 383        close(fd);
 384
 385        link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
 386
 387        munmap(map, mapsz);
 388}
 389
 390struct alternate_object_database *alloc_alt_odb(const char *dir)
 391{
 392        struct alternate_object_database *ent;
 393
 394        FLEX_ALLOC_STR(ent, path, dir);
 395        strbuf_init(&ent->scratch, 0);
 396        strbuf_addf(&ent->scratch, "%s/", dir);
 397        ent->base_len = ent->scratch.len;
 398
 399        return ent;
 400}
 401
 402void add_to_alternates_file(const char *reference)
 403{
 404        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
 405        char *alts = git_pathdup("objects/info/alternates");
 406        FILE *in, *out;
 407
 408        hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
 409        out = fdopen_lock_file(lock, "w");
 410        if (!out)
 411                die_errno("unable to fdopen alternates lockfile");
 412
 413        in = fopen(alts, "r");
 414        if (in) {
 415                struct strbuf line = STRBUF_INIT;
 416                int found = 0;
 417
 418                while (strbuf_getline(&line, in) != EOF) {
 419                        if (!strcmp(reference, line.buf)) {
 420                                found = 1;
 421                                break;
 422                        }
 423                        fprintf_or_die(out, "%s\n", line.buf);
 424                }
 425
 426                strbuf_release(&line);
 427                fclose(in);
 428
 429                if (found) {
 430                        rollback_lock_file(lock);
 431                        lock = NULL;
 432                }
 433        }
 434        else if (errno != ENOENT)
 435                die_errno("unable to read alternates file");
 436
 437        if (lock) {
 438                fprintf_or_die(out, "%s\n", reference);
 439                if (commit_lock_file(lock))
 440                        die_errno("unable to move new alternates file into place");
 441                if (alt_odb_tail)
 442                        link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
 443        }
 444        free(alts);
 445}
 446
 447void add_to_alternates_memory(const char *reference)
 448{
 449        /*
 450         * Make sure alternates are initialized, or else our entry may be
 451         * overwritten when they are.
 452         */
 453        prepare_alt_odb();
 454
 455        link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
 456}
 457
 458/*
 459 * Compute the exact path an alternate is at and returns it. In case of
 460 * error NULL is returned and the human readable error is added to `err`
 461 * `path` may be relative and should point to $GITDIR.
 462 * `err` must not be null.
 463 */
 464char *compute_alternate_path(const char *path, struct strbuf *err)
 465{
 466        char *ref_git = NULL;
 467        const char *repo, *ref_git_s;
 468        int seen_error = 0;
 469
 470        ref_git_s = real_path_if_valid(path);
 471        if (!ref_git_s) {
 472                seen_error = 1;
 473                strbuf_addf(err, _("path '%s' does not exist"), path);
 474                goto out;
 475        } else
 476                /*
 477                 * Beware: read_gitfile(), real_path() and mkpath()
 478                 * return static buffer
 479                 */
 480                ref_git = xstrdup(ref_git_s);
 481
 482        repo = read_gitfile(ref_git);
 483        if (!repo)
 484                repo = read_gitfile(mkpath("%s/.git", ref_git));
 485        if (repo) {
 486                free(ref_git);
 487                ref_git = xstrdup(repo);
 488        }
 489
 490        if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
 491                char *ref_git_git = mkpathdup("%s/.git", ref_git);
 492                free(ref_git);
 493                ref_git = ref_git_git;
 494        } else if (!is_directory(mkpath("%s/objects", ref_git))) {
 495                struct strbuf sb = STRBUF_INIT;
 496                seen_error = 1;
 497                if (get_common_dir(&sb, ref_git)) {
 498                        strbuf_addf(err,
 499                                    _("reference repository '%s' as a linked "
 500                                      "checkout is not supported yet."),
 501                                    path);
 502                        goto out;
 503                }
 504
 505                strbuf_addf(err, _("reference repository '%s' is not a "
 506                                        "local repository."), path);
 507                goto out;
 508        }
 509
 510        if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
 511                strbuf_addf(err, _("reference repository '%s' is shallow"),
 512                            path);
 513                seen_error = 1;
 514                goto out;
 515        }
 516
 517        if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
 518                strbuf_addf(err,
 519                            _("reference repository '%s' is grafted"),
 520                            path);
 521                seen_error = 1;
 522                goto out;
 523        }
 524
 525out:
 526        if (seen_error) {
 527                free(ref_git);
 528                ref_git = NULL;
 529        }
 530
 531        return ref_git;
 532}
 533
 534int foreach_alt_odb(alt_odb_fn fn, void *cb)
 535{
 536        struct alternate_object_database *ent;
 537        int r = 0;
 538
 539        prepare_alt_odb();
 540        for (ent = alt_odb_list; ent; ent = ent->next) {
 541                r = fn(ent, cb);
 542                if (r)
 543                        break;
 544        }
 545        return r;
 546}
 547
 548void prepare_alt_odb(void)
 549{
 550        const char *alt;
 551
 552        if (alt_odb_tail)
 553                return;
 554
 555        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 556        if (!alt) alt = "";
 557
 558        alt_odb_tail = &alt_odb_list;
 559        link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
 560
 561        read_info_alternates(get_object_directory(), 0);
 562}
 563
 564/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
 565static int freshen_file(const char *fn)
 566{
 567        struct utimbuf t;
 568        t.actime = t.modtime = time(NULL);
 569        return !utime(fn, &t);
 570}
 571
 572/*
 573 * All of the check_and_freshen functions return 1 if the file exists and was
 574 * freshened (if freshening was requested), 0 otherwise. If they return
 575 * 0, you should not assume that it is safe to skip a write of the object (it
 576 * either does not exist on disk, or has a stale mtime and may be subject to
 577 * pruning).
 578 */
 579static int check_and_freshen_file(const char *fn, int freshen)
 580{
 581        if (access(fn, F_OK))
 582                return 0;
 583        if (freshen && !freshen_file(fn))
 584                return 0;
 585        return 1;
 586}
 587
 588static int check_and_freshen_local(const unsigned char *sha1, int freshen)
 589{
 590        return check_and_freshen_file(sha1_file_name(sha1), freshen);
 591}
 592
 593static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
 594{
 595        struct alternate_object_database *alt;
 596        prepare_alt_odb();
 597        for (alt = alt_odb_list; alt; alt = alt->next) {
 598                const char *path = alt_sha1_path(alt, sha1);
 599                if (check_and_freshen_file(path, freshen))
 600                        return 1;
 601        }
 602        return 0;
 603}
 604
 605static int check_and_freshen(const unsigned char *sha1, int freshen)
 606{
 607        return check_and_freshen_local(sha1, freshen) ||
 608               check_and_freshen_nonlocal(sha1, freshen);
 609}
 610
 611int has_loose_object_nonlocal(const unsigned char *sha1)
 612{
 613        return check_and_freshen_nonlocal(sha1, 0);
 614}
 615
 616static int has_loose_object(const unsigned char *sha1)
 617{
 618        return check_and_freshen(sha1, 0);
 619}
 620
 621static unsigned int pack_used_ctr;
 622static unsigned int pack_mmap_calls;
 623static unsigned int peak_pack_open_windows;
 624static unsigned int pack_open_windows;
 625static unsigned int pack_open_fds;
 626static unsigned int pack_max_fds;
 627static size_t peak_pack_mapped;
 628static size_t pack_mapped;
 629struct packed_git *packed_git;
 630
 631static struct mru packed_git_mru_storage;
 632struct mru *packed_git_mru = &packed_git_mru_storage;
 633
 634void pack_report(void)
 635{
 636        fprintf(stderr,
 637                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 638                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 639                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 640                sz_fmt(getpagesize()),
 641                sz_fmt(packed_git_window_size),
 642                sz_fmt(packed_git_limit));
 643        fprintf(stderr,
 644                "pack_report: pack_used_ctr            = %10u\n"
 645                "pack_report: pack_mmap_calls          = %10u\n"
 646                "pack_report: pack_open_windows        = %10u / %10u\n"
 647                "pack_report: pack_mapped              = "
 648                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 649                pack_used_ctr,
 650                pack_mmap_calls,
 651                pack_open_windows, peak_pack_open_windows,
 652                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
 653}
 654
 655/*
 656 * Open and mmap the index file at path, perform a couple of
 657 * consistency checks, then record its information to p.  Return 0 on
 658 * success.
 659 */
 660static int check_packed_git_idx(const char *path, struct packed_git *p)
 661{
 662        void *idx_map;
 663        struct pack_idx_header *hdr;
 664        size_t idx_size;
 665        uint32_t version, nr, i, *index;
 666        int fd = git_open(path);
 667        struct stat st;
 668
 669        if (fd < 0)
 670                return -1;
 671        if (fstat(fd, &st)) {
 672                close(fd);
 673                return -1;
 674        }
 675        idx_size = xsize_t(st.st_size);
 676        if (idx_size < 4 * 256 + 20 + 20) {
 677                close(fd);
 678                return error("index file %s is too small", path);
 679        }
 680        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 681        close(fd);
 682
 683        hdr = idx_map;
 684        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 685                version = ntohl(hdr->idx_version);
 686                if (version < 2 || version > 2) {
 687                        munmap(idx_map, idx_size);
 688                        return error("index file %s is version %"PRIu32
 689                                     " and is not supported by this binary"
 690                                     " (try upgrading GIT to a newer version)",
 691                                     path, version);
 692                }
 693        } else
 694                version = 1;
 695
 696        nr = 0;
 697        index = idx_map;
 698        if (version > 1)
 699                index += 2;  /* skip index header */
 700        for (i = 0; i < 256; i++) {
 701                uint32_t n = ntohl(index[i]);
 702                if (n < nr) {
 703                        munmap(idx_map, idx_size);
 704                        return error("non-monotonic index %s", path);
 705                }
 706                nr = n;
 707        }
 708
 709        if (version == 1) {
 710                /*
 711                 * Total size:
 712                 *  - 256 index entries 4 bytes each
 713                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 714                 *  - 20-byte SHA1 of the packfile
 715                 *  - 20-byte SHA1 file checksum
 716                 */
 717                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 718                        munmap(idx_map, idx_size);
 719                        return error("wrong index v1 file size in %s", path);
 720                }
 721        } else if (version == 2) {
 722                /*
 723                 * Minimum size:
 724                 *  - 8 bytes of header
 725                 *  - 256 index entries 4 bytes each
 726                 *  - 20-byte sha1 entry * nr
 727                 *  - 4-byte crc entry * nr
 728                 *  - 4-byte offset entry * nr
 729                 *  - 20-byte SHA1 of the packfile
 730                 *  - 20-byte SHA1 file checksum
 731                 * And after the 4-byte offset table might be a
 732                 * variable sized table containing 8-byte entries
 733                 * for offsets larger than 2^31.
 734                 */
 735                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 736                unsigned long max_size = min_size;
 737                if (nr)
 738                        max_size += (nr - 1)*8;
 739                if (idx_size < min_size || idx_size > max_size) {
 740                        munmap(idx_map, idx_size);
 741                        return error("wrong index v2 file size in %s", path);
 742                }
 743                if (idx_size != min_size &&
 744                    /*
 745                     * make sure we can deal with large pack offsets.
 746                     * 31-bit signed offset won't be enough, neither
 747                     * 32-bit unsigned one will be.
 748                     */
 749                    (sizeof(off_t) <= 4)) {
 750                        munmap(idx_map, idx_size);
 751                        return error("pack too large for current definition of off_t in %s", path);
 752                }
 753        }
 754
 755        p->index_version = version;
 756        p->index_data = idx_map;
 757        p->index_size = idx_size;
 758        p->num_objects = nr;
 759        return 0;
 760}
 761
 762int open_pack_index(struct packed_git *p)
 763{
 764        char *idx_name;
 765        size_t len;
 766        int ret;
 767
 768        if (p->index_data)
 769                return 0;
 770
 771        if (!strip_suffix(p->pack_name, ".pack", &len))
 772                die("BUG: pack_name does not end in .pack");
 773        idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
 774        ret = check_packed_git_idx(idx_name, p);
 775        free(idx_name);
 776        return ret;
 777}
 778
 779static void scan_windows(struct packed_git *p,
 780        struct packed_git **lru_p,
 781        struct pack_window **lru_w,
 782        struct pack_window **lru_l)
 783{
 784        struct pack_window *w, *w_l;
 785
 786        for (w_l = NULL, w = p->windows; w; w = w->next) {
 787                if (!w->inuse_cnt) {
 788                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 789                                *lru_p = p;
 790                                *lru_w = w;
 791                                *lru_l = w_l;
 792                        }
 793                }
 794                w_l = w;
 795        }
 796}
 797
 798static int unuse_one_window(struct packed_git *current)
 799{
 800        struct packed_git *p, *lru_p = NULL;
 801        struct pack_window *lru_w = NULL, *lru_l = NULL;
 802
 803        if (current)
 804                scan_windows(current, &lru_p, &lru_w, &lru_l);
 805        for (p = packed_git; p; p = p->next)
 806                scan_windows(p, &lru_p, &lru_w, &lru_l);
 807        if (lru_p) {
 808                munmap(lru_w->base, lru_w->len);
 809                pack_mapped -= lru_w->len;
 810                if (lru_l)
 811                        lru_l->next = lru_w->next;
 812                else
 813                        lru_p->windows = lru_w->next;
 814                free(lru_w);
 815                pack_open_windows--;
 816                return 1;
 817        }
 818        return 0;
 819}
 820
 821void release_pack_memory(size_t need)
 822{
 823        size_t cur = pack_mapped;
 824        while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
 825                ; /* nothing */
 826}
 827
 828static void mmap_limit_check(size_t length)
 829{
 830        static size_t limit = 0;
 831        if (!limit) {
 832                limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
 833                if (!limit)
 834                        limit = SIZE_MAX;
 835        }
 836        if (length > limit)
 837                die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
 838                    (uintmax_t)length, (uintmax_t)limit);
 839}
 840
 841void *xmmap_gently(void *start, size_t length,
 842                  int prot, int flags, int fd, off_t offset)
 843{
 844        void *ret;
 845
 846        mmap_limit_check(length);
 847        ret = mmap(start, length, prot, flags, fd, offset);
 848        if (ret == MAP_FAILED) {
 849                if (!length)
 850                        return NULL;
 851                release_pack_memory(length);
 852                ret = mmap(start, length, prot, flags, fd, offset);
 853        }
 854        return ret;
 855}
 856
 857void *xmmap(void *start, size_t length,
 858        int prot, int flags, int fd, off_t offset)
 859{
 860        void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
 861        if (ret == MAP_FAILED)
 862                die_errno("mmap failed");
 863        return ret;
 864}
 865
 866void close_pack_windows(struct packed_git *p)
 867{
 868        while (p->windows) {
 869                struct pack_window *w = p->windows;
 870
 871                if (w->inuse_cnt)
 872                        die("pack '%s' still has open windows to it",
 873                            p->pack_name);
 874                munmap(w->base, w->len);
 875                pack_mapped -= w->len;
 876                pack_open_windows--;
 877                p->windows = w->next;
 878                free(w);
 879        }
 880}
 881
 882static int close_pack_fd(struct packed_git *p)
 883{
 884        if (p->pack_fd < 0)
 885                return 0;
 886
 887        close(p->pack_fd);
 888        pack_open_fds--;
 889        p->pack_fd = -1;
 890
 891        return 1;
 892}
 893
 894static void close_pack(struct packed_git *p)
 895{
 896        close_pack_windows(p);
 897        close_pack_fd(p);
 898        close_pack_index(p);
 899}
 900
 901void close_all_packs(void)
 902{
 903        struct packed_git *p;
 904
 905        for (p = packed_git; p; p = p->next)
 906                if (p->do_not_close)
 907                        die("BUG: want to close pack marked 'do-not-close'");
 908                else
 909                        close_pack(p);
 910}
 911
 912
 913/*
 914 * The LRU pack is the one with the oldest MRU window, preferring packs
 915 * with no used windows, or the oldest mtime if it has no windows allocated.
 916 */
 917static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
 918{
 919        struct pack_window *w, *this_mru_w;
 920        int has_windows_inuse = 0;
 921
 922        /*
 923         * Reject this pack if it has windows and the previously selected
 924         * one does not.  If this pack does not have windows, reject
 925         * it if the pack file is newer than the previously selected one.
 926         */
 927        if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
 928                return;
 929
 930        for (w = this_mru_w = p->windows; w; w = w->next) {
 931                /*
 932                 * Reject this pack if any of its windows are in use,
 933                 * but the previously selected pack did not have any
 934                 * inuse windows.  Otherwise, record that this pack
 935                 * has windows in use.
 936                 */
 937                if (w->inuse_cnt) {
 938                        if (*accept_windows_inuse)
 939                                has_windows_inuse = 1;
 940                        else
 941                                return;
 942                }
 943
 944                if (w->last_used > this_mru_w->last_used)
 945                        this_mru_w = w;
 946
 947                /*
 948                 * Reject this pack if it has windows that have been
 949                 * used more recently than the previously selected pack.
 950                 * If the previously selected pack had windows inuse and
 951                 * we have not encountered a window in this pack that is
 952                 * inuse, skip this check since we prefer a pack with no
 953                 * inuse windows to one that has inuse windows.
 954                 */
 955                if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
 956                    this_mru_w->last_used > (*mru_w)->last_used)
 957                        return;
 958        }
 959
 960        /*
 961         * Select this pack.
 962         */
 963        *mru_w = this_mru_w;
 964        *lru_p = p;
 965        *accept_windows_inuse = has_windows_inuse;
 966}
 967
 968static int close_one_pack(void)
 969{
 970        struct packed_git *p, *lru_p = NULL;
 971        struct pack_window *mru_w = NULL;
 972        int accept_windows_inuse = 1;
 973
 974        for (p = packed_git; p; p = p->next) {
 975                if (p->pack_fd == -1)
 976                        continue;
 977                find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
 978        }
 979
 980        if (lru_p)
 981                return close_pack_fd(lru_p);
 982
 983        return 0;
 984}
 985
 986void unuse_pack(struct pack_window **w_cursor)
 987{
 988        struct pack_window *w = *w_cursor;
 989        if (w) {
 990                w->inuse_cnt--;
 991                *w_cursor = NULL;
 992        }
 993}
 994
 995void close_pack_index(struct packed_git *p)
 996{
 997        if (p->index_data) {
 998                munmap((void *)p->index_data, p->index_size);
 999                p->index_data = NULL;
1000        }
1001}
1002
1003static unsigned int get_max_fd_limit(void)
1004{
1005#ifdef RLIMIT_NOFILE
1006        {
1007                struct rlimit lim;
1008
1009                if (!getrlimit(RLIMIT_NOFILE, &lim))
1010                        return lim.rlim_cur;
1011        }
1012#endif
1013
1014#ifdef _SC_OPEN_MAX
1015        {
1016                long open_max = sysconf(_SC_OPEN_MAX);
1017                if (0 < open_max)
1018                        return open_max;
1019                /*
1020                 * Otherwise, we got -1 for one of the two
1021                 * reasons:
1022                 *
1023                 * (1) sysconf() did not understand _SC_OPEN_MAX
1024                 *     and signaled an error with -1; or
1025                 * (2) sysconf() said there is no limit.
1026                 *
1027                 * We _could_ clear errno before calling sysconf() to
1028                 * tell these two cases apart and return a huge number
1029                 * in the latter case to let the caller cap it to a
1030                 * value that is not so selfish, but letting the
1031                 * fallback OPEN_MAX codepath take care of these cases
1032                 * is a lot simpler.
1033                 */
1034        }
1035#endif
1036
1037#ifdef OPEN_MAX
1038        return OPEN_MAX;
1039#else
1040        return 1; /* see the caller ;-) */
1041#endif
1042}
1043
1044/*
1045 * Do not call this directly as this leaks p->pack_fd on error return;
1046 * call open_packed_git() instead.
1047 */
1048static int open_packed_git_1(struct packed_git *p)
1049{
1050        struct stat st;
1051        struct pack_header hdr;
1052        unsigned char sha1[20];
1053        unsigned char *idx_sha1;
1054        long fd_flag;
1055
1056        if (!p->index_data && open_pack_index(p))
1057                return error("packfile %s index unavailable", p->pack_name);
1058
1059        if (!pack_max_fds) {
1060                unsigned int max_fds = get_max_fd_limit();
1061
1062                /* Save 3 for stdin/stdout/stderr, 22 for work */
1063                if (25 < max_fds)
1064                        pack_max_fds = max_fds - 25;
1065                else
1066                        pack_max_fds = 1;
1067        }
1068
1069        while (pack_max_fds <= pack_open_fds && close_one_pack())
1070                ; /* nothing */
1071
1072        p->pack_fd = git_open(p->pack_name);
1073        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
1074                return -1;
1075        pack_open_fds++;
1076
1077        /* If we created the struct before we had the pack we lack size. */
1078        if (!p->pack_size) {
1079                if (!S_ISREG(st.st_mode))
1080                        return error("packfile %s not a regular file", p->pack_name);
1081                p->pack_size = st.st_size;
1082        } else if (p->pack_size != st.st_size)
1083                return error("packfile %s size changed", p->pack_name);
1084
1085        /* We leave these file descriptors open with sliding mmap;
1086         * there is no point keeping them open across exec(), though.
1087         */
1088        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1089        if (fd_flag < 0)
1090                return error("cannot determine file descriptor flags");
1091        fd_flag |= FD_CLOEXEC;
1092        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1093                return error("cannot set FD_CLOEXEC");
1094
1095        /* Verify we recognize this pack file format. */
1096        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1097                return error("file %s is far too short to be a packfile", p->pack_name);
1098        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1099                return error("file %s is not a GIT packfile", p->pack_name);
1100        if (!pack_version_ok(hdr.hdr_version))
1101                return error("packfile %s is version %"PRIu32" and not"
1102                        " supported (try upgrading GIT to a newer version)",
1103                        p->pack_name, ntohl(hdr.hdr_version));
1104
1105        /* Verify the pack matches its index. */
1106        if (p->num_objects != ntohl(hdr.hdr_entries))
1107                return error("packfile %s claims to have %"PRIu32" objects"
1108                             " while index indicates %"PRIu32" objects",
1109                             p->pack_name, ntohl(hdr.hdr_entries),
1110                             p->num_objects);
1111        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1112                return error("end of packfile %s is unavailable", p->pack_name);
1113        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1114                return error("packfile %s signature is unavailable", p->pack_name);
1115        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1116        if (hashcmp(sha1, idx_sha1))
1117                return error("packfile %s does not match index", p->pack_name);
1118        return 0;
1119}
1120
1121static int open_packed_git(struct packed_git *p)
1122{
1123        if (!open_packed_git_1(p))
1124                return 0;
1125        close_pack_fd(p);
1126        return -1;
1127}
1128
1129static int in_window(struct pack_window *win, off_t offset)
1130{
1131        /* We must promise at least 20 bytes (one hash) after the
1132         * offset is available from this window, otherwise the offset
1133         * is not actually in this window and a different window (which
1134         * has that one hash excess) must be used.  This is to support
1135         * the object header and delta base parsing routines below.
1136         */
1137        off_t win_off = win->offset;
1138        return win_off <= offset
1139                && (offset + 20) <= (win_off + win->len);
1140}
1141
1142unsigned char *use_pack(struct packed_git *p,
1143                struct pack_window **w_cursor,
1144                off_t offset,
1145                unsigned long *left)
1146{
1147        struct pack_window *win = *w_cursor;
1148
1149        /* Since packfiles end in a hash of their content and it's
1150         * pointless to ask for an offset into the middle of that
1151         * hash, and the in_window function above wouldn't match
1152         * don't allow an offset too close to the end of the file.
1153         */
1154        if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1155                die("packfile %s cannot be accessed", p->pack_name);
1156        if (offset > (p->pack_size - 20))
1157                die("offset beyond end of packfile (truncated pack?)");
1158        if (offset < 0)
1159                die(_("offset before end of packfile (broken .idx?)"));
1160
1161        if (!win || !in_window(win, offset)) {
1162                if (win)
1163                        win->inuse_cnt--;
1164                for (win = p->windows; win; win = win->next) {
1165                        if (in_window(win, offset))
1166                                break;
1167                }
1168                if (!win) {
1169                        size_t window_align = packed_git_window_size / 2;
1170                        off_t len;
1171
1172                        if (p->pack_fd == -1 && open_packed_git(p))
1173                                die("packfile %s cannot be accessed", p->pack_name);
1174
1175                        win = xcalloc(1, sizeof(*win));
1176                        win->offset = (offset / window_align) * window_align;
1177                        len = p->pack_size - win->offset;
1178                        if (len > packed_git_window_size)
1179                                len = packed_git_window_size;
1180                        win->len = (size_t)len;
1181                        pack_mapped += win->len;
1182                        while (packed_git_limit < pack_mapped
1183                                && unuse_one_window(p))
1184                                ; /* nothing */
1185                        win->base = xmmap(NULL, win->len,
1186                                PROT_READ, MAP_PRIVATE,
1187                                p->pack_fd, win->offset);
1188                        if (win->base == MAP_FAILED)
1189                                die_errno("packfile %s cannot be mapped",
1190                                          p->pack_name);
1191                        if (!win->offset && win->len == p->pack_size
1192                                && !p->do_not_close)
1193                                close_pack_fd(p);
1194                        pack_mmap_calls++;
1195                        pack_open_windows++;
1196                        if (pack_mapped > peak_pack_mapped)
1197                                peak_pack_mapped = pack_mapped;
1198                        if (pack_open_windows > peak_pack_open_windows)
1199                                peak_pack_open_windows = pack_open_windows;
1200                        win->next = p->windows;
1201                        p->windows = win;
1202                }
1203        }
1204        if (win != *w_cursor) {
1205                win->last_used = pack_used_ctr++;
1206                win->inuse_cnt++;
1207                *w_cursor = win;
1208        }
1209        offset -= win->offset;
1210        if (left)
1211                *left = win->len - xsize_t(offset);
1212        return win->base + offset;
1213}
1214
1215static struct packed_git *alloc_packed_git(int extra)
1216{
1217        struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1218        memset(p, 0, sizeof(*p));
1219        p->pack_fd = -1;
1220        return p;
1221}
1222
1223static void try_to_free_pack_memory(size_t size)
1224{
1225        release_pack_memory(size);
1226}
1227
1228struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1229{
1230        static int have_set_try_to_free_routine;
1231        struct stat st;
1232        size_t alloc;
1233        struct packed_git *p;
1234
1235        if (!have_set_try_to_free_routine) {
1236                have_set_try_to_free_routine = 1;
1237                set_try_to_free_routine(try_to_free_pack_memory);
1238        }
1239
1240        /*
1241         * Make sure a corresponding .pack file exists and that
1242         * the index looks sane.
1243         */
1244        if (!strip_suffix_mem(path, &path_len, ".idx"))
1245                return NULL;
1246
1247        /*
1248         * ".pack" is long enough to hold any suffix we're adding (and
1249         * the use xsnprintf double-checks that)
1250         */
1251        alloc = st_add3(path_len, strlen(".pack"), 1);
1252        p = alloc_packed_git(alloc);
1253        memcpy(p->pack_name, path, path_len);
1254
1255        xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1256        if (!access(p->pack_name, F_OK))
1257                p->pack_keep = 1;
1258
1259        xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1260        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1261                free(p);
1262                return NULL;
1263        }
1264
1265        /* ok, it looks sane as far as we can check without
1266         * actually mapping the pack file.
1267         */
1268        p->pack_size = st.st_size;
1269        p->pack_local = local;
1270        p->mtime = st.st_mtime;
1271        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1272                hashclr(p->sha1);
1273        return p;
1274}
1275
1276struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1277{
1278        const char *path = sha1_pack_name(sha1);
1279        size_t alloc = st_add(strlen(path), 1);
1280        struct packed_git *p = alloc_packed_git(alloc);
1281
1282        memcpy(p->pack_name, path, alloc); /* includes NUL */
1283        hashcpy(p->sha1, sha1);
1284        if (check_packed_git_idx(idx_path, p)) {
1285                free(p);
1286                return NULL;
1287        }
1288
1289        return p;
1290}
1291
1292void install_packed_git(struct packed_git *pack)
1293{
1294        if (pack->pack_fd != -1)
1295                pack_open_fds++;
1296
1297        pack->next = packed_git;
1298        packed_git = pack;
1299}
1300
1301void (*report_garbage)(unsigned seen_bits, const char *path);
1302
1303static void report_helper(const struct string_list *list,
1304                          int seen_bits, int first, int last)
1305{
1306        if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1307                return;
1308
1309        for (; first < last; first++)
1310                report_garbage(seen_bits, list->items[first].string);
1311}
1312
1313static void report_pack_garbage(struct string_list *list)
1314{
1315        int i, baselen = -1, first = 0, seen_bits = 0;
1316
1317        if (!report_garbage)
1318                return;
1319
1320        string_list_sort(list);
1321
1322        for (i = 0; i < list->nr; i++) {
1323                const char *path = list->items[i].string;
1324                if (baselen != -1 &&
1325                    strncmp(path, list->items[first].string, baselen)) {
1326                        report_helper(list, seen_bits, first, i);
1327                        baselen = -1;
1328                        seen_bits = 0;
1329                }
1330                if (baselen == -1) {
1331                        const char *dot = strrchr(path, '.');
1332                        if (!dot) {
1333                                report_garbage(PACKDIR_FILE_GARBAGE, path);
1334                                continue;
1335                        }
1336                        baselen = dot - path + 1;
1337                        first = i;
1338                }
1339                if (!strcmp(path + baselen, "pack"))
1340                        seen_bits |= 1;
1341                else if (!strcmp(path + baselen, "idx"))
1342                        seen_bits |= 2;
1343        }
1344        report_helper(list, seen_bits, first, list->nr);
1345}
1346
1347static void prepare_packed_git_one(char *objdir, int local)
1348{
1349        struct strbuf path = STRBUF_INIT;
1350        size_t dirnamelen;
1351        DIR *dir;
1352        struct dirent *de;
1353        struct string_list garbage = STRING_LIST_INIT_DUP;
1354
1355        strbuf_addstr(&path, objdir);
1356        strbuf_addstr(&path, "/pack");
1357        dir = opendir(path.buf);
1358        if (!dir) {
1359                if (errno != ENOENT)
1360                        error_errno("unable to open object pack directory: %s",
1361                                    path.buf);
1362                strbuf_release(&path);
1363                return;
1364        }
1365        strbuf_addch(&path, '/');
1366        dirnamelen = path.len;
1367        while ((de = readdir(dir)) != NULL) {
1368                struct packed_git *p;
1369                size_t base_len;
1370
1371                if (is_dot_or_dotdot(de->d_name))
1372                        continue;
1373
1374                strbuf_setlen(&path, dirnamelen);
1375                strbuf_addstr(&path, de->d_name);
1376
1377                base_len = path.len;
1378                if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1379                        /* Don't reopen a pack we already have. */
1380                        for (p = packed_git; p; p = p->next) {
1381                                size_t len;
1382                                if (strip_suffix(p->pack_name, ".pack", &len) &&
1383                                    len == base_len &&
1384                                    !memcmp(p->pack_name, path.buf, len))
1385                                        break;
1386                        }
1387                        if (p == NULL &&
1388                            /*
1389                             * See if it really is a valid .idx file with
1390                             * corresponding .pack file that we can map.
1391                             */
1392                            (p = add_packed_git(path.buf, path.len, local)) != NULL)
1393                                install_packed_git(p);
1394                }
1395
1396                if (!report_garbage)
1397                        continue;
1398
1399                if (ends_with(de->d_name, ".idx") ||
1400                    ends_with(de->d_name, ".pack") ||
1401                    ends_with(de->d_name, ".bitmap") ||
1402                    ends_with(de->d_name, ".keep"))
1403                        string_list_append(&garbage, path.buf);
1404                else
1405                        report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1406        }
1407        closedir(dir);
1408        report_pack_garbage(&garbage);
1409        string_list_clear(&garbage, 0);
1410        strbuf_release(&path);
1411}
1412
1413static int approximate_object_count_valid;
1414
1415/*
1416 * Give a fast, rough count of the number of objects in the repository. This
1417 * ignores loose objects completely. If you have a lot of them, then either
1418 * you should repack because your performance will be awful, or they are
1419 * all unreachable objects about to be pruned, in which case they're not really
1420 * interesting as a measure of repo size in the first place.
1421 */
1422unsigned long approximate_object_count(void)
1423{
1424        static unsigned long count;
1425        if (!approximate_object_count_valid) {
1426                struct packed_git *p;
1427
1428                prepare_packed_git();
1429                count = 0;
1430                for (p = packed_git; p; p = p->next) {
1431                        if (open_pack_index(p))
1432                                continue;
1433                        count += p->num_objects;
1434                }
1435        }
1436        return count;
1437}
1438
1439static void *get_next_packed_git(const void *p)
1440{
1441        return ((const struct packed_git *)p)->next;
1442}
1443
1444static void set_next_packed_git(void *p, void *next)
1445{
1446        ((struct packed_git *)p)->next = next;
1447}
1448
1449static int sort_pack(const void *a_, const void *b_)
1450{
1451        const struct packed_git *a = a_;
1452        const struct packed_git *b = b_;
1453        int st;
1454
1455        /*
1456         * Local packs tend to contain objects specific to our
1457         * variant of the project than remote ones.  In addition,
1458         * remote ones could be on a network mounted filesystem.
1459         * Favor local ones for these reasons.
1460         */
1461        st = a->pack_local - b->pack_local;
1462        if (st)
1463                return -st;
1464
1465        /*
1466         * Younger packs tend to contain more recent objects,
1467         * and more recent objects tend to get accessed more
1468         * often.
1469         */
1470        if (a->mtime < b->mtime)
1471                return 1;
1472        else if (a->mtime == b->mtime)
1473                return 0;
1474        return -1;
1475}
1476
1477static void rearrange_packed_git(void)
1478{
1479        packed_git = llist_mergesort(packed_git, get_next_packed_git,
1480                                     set_next_packed_git, sort_pack);
1481}
1482
1483static void prepare_packed_git_mru(void)
1484{
1485        struct packed_git *p;
1486
1487        mru_clear(packed_git_mru);
1488        for (p = packed_git; p; p = p->next)
1489                mru_append(packed_git_mru, p);
1490}
1491
1492static int prepare_packed_git_run_once = 0;
1493void prepare_packed_git(void)
1494{
1495        struct alternate_object_database *alt;
1496
1497        if (prepare_packed_git_run_once)
1498                return;
1499        prepare_packed_git_one(get_object_directory(), 1);
1500        prepare_alt_odb();
1501        for (alt = alt_odb_list; alt; alt = alt->next)
1502                prepare_packed_git_one(alt->path, 0);
1503        rearrange_packed_git();
1504        prepare_packed_git_mru();
1505        prepare_packed_git_run_once = 1;
1506}
1507
1508void reprepare_packed_git(void)
1509{
1510        approximate_object_count_valid = 0;
1511        prepare_packed_git_run_once = 0;
1512        prepare_packed_git();
1513}
1514
1515static void mark_bad_packed_object(struct packed_git *p,
1516                                   const unsigned char *sha1)
1517{
1518        unsigned i;
1519        for (i = 0; i < p->num_bad_objects; i++)
1520                if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1521                        return;
1522        p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1523                                      st_mult(GIT_SHA1_RAWSZ,
1524                                              st_add(p->num_bad_objects, 1)));
1525        hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1526        p->num_bad_objects++;
1527}
1528
1529static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1530{
1531        struct packed_git *p;
1532        unsigned i;
1533
1534        for (p = packed_git; p; p = p->next)
1535                for (i = 0; i < p->num_bad_objects; i++)
1536                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1537                                return p;
1538        return NULL;
1539}
1540
1541/*
1542 * With an in-core object data in "map", rehash it to make sure the
1543 * object name actually matches "sha1" to detect object corruption.
1544 * With "map" == NULL, try reading the object named with "sha1" using
1545 * the streaming interface and rehash it to do the same.
1546 */
1547int check_sha1_signature(const unsigned char *sha1, void *map,
1548                         unsigned long size, const char *type)
1549{
1550        unsigned char real_sha1[20];
1551        enum object_type obj_type;
1552        struct git_istream *st;
1553        git_SHA_CTX c;
1554        char hdr[32];
1555        int hdrlen;
1556
1557        if (map) {
1558                hash_sha1_file(map, size, type, real_sha1);
1559                return hashcmp(sha1, real_sha1) ? -1 : 0;
1560        }
1561
1562        st = open_istream(sha1, &obj_type, &size, NULL);
1563        if (!st)
1564                return -1;
1565
1566        /* Generate the header */
1567        hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1568
1569        /* Sha1.. */
1570        git_SHA1_Init(&c);
1571        git_SHA1_Update(&c, hdr, hdrlen);
1572        for (;;) {
1573                char buf[1024 * 16];
1574                ssize_t readlen = read_istream(st, buf, sizeof(buf));
1575
1576                if (readlen < 0) {
1577                        close_istream(st);
1578                        return -1;
1579                }
1580                if (!readlen)
1581                        break;
1582                git_SHA1_Update(&c, buf, readlen);
1583        }
1584        git_SHA1_Final(real_sha1, &c);
1585        close_istream(st);
1586        return hashcmp(sha1, real_sha1) ? -1 : 0;
1587}
1588
1589int git_open(const char *name)
1590{
1591        static int sha1_file_open_flag = O_NOATIME | O_CLOEXEC;
1592
1593        for (;;) {
1594                int fd;
1595
1596                errno = 0;
1597                fd = open(name, O_RDONLY | sha1_file_open_flag);
1598                if (fd >= 0)
1599                        return fd;
1600
1601                /* Try again w/o O_CLOEXEC: the kernel might not support it */
1602                if ((sha1_file_open_flag & O_CLOEXEC) && errno == EINVAL) {
1603                        sha1_file_open_flag &= ~O_CLOEXEC;
1604                        continue;
1605                }
1606
1607                /* Might the failure be due to O_NOATIME? */
1608                if (errno != ENOENT && (sha1_file_open_flag & O_NOATIME)) {
1609                        sha1_file_open_flag &= ~O_NOATIME;
1610                        continue;
1611                }
1612                return -1;
1613        }
1614}
1615
1616/*
1617 * Find "sha1" as a loose object in the local repository or in an alternate.
1618 * Returns 0 on success, negative on failure.
1619 *
1620 * The "path" out-parameter will give the path of the object we found (if any).
1621 * Note that it may point to static storage and is only valid until another
1622 * call to sha1_file_name(), etc.
1623 */
1624static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
1625                          const char **path)
1626{
1627        struct alternate_object_database *alt;
1628
1629        *path = sha1_file_name(sha1);
1630        if (!lstat(*path, st))
1631                return 0;
1632
1633        prepare_alt_odb();
1634        errno = ENOENT;
1635        for (alt = alt_odb_list; alt; alt = alt->next) {
1636                *path = alt_sha1_path(alt, sha1);
1637                if (!lstat(*path, st))
1638                        return 0;
1639        }
1640
1641        return -1;
1642}
1643
1644/*
1645 * Like stat_sha1_file(), but actually open the object and return the
1646 * descriptor. See the caveats on the "path" parameter above.
1647 */
1648static int open_sha1_file(const unsigned char *sha1, const char **path)
1649{
1650        int fd;
1651        struct alternate_object_database *alt;
1652        int most_interesting_errno;
1653
1654        *path = sha1_file_name(sha1);
1655        fd = git_open(*path);
1656        if (fd >= 0)
1657                return fd;
1658        most_interesting_errno = errno;
1659
1660        prepare_alt_odb();
1661        for (alt = alt_odb_list; alt; alt = alt->next) {
1662                *path = alt_sha1_path(alt, sha1);
1663                fd = git_open(*path);
1664                if (fd >= 0)
1665                        return fd;
1666                if (most_interesting_errno == ENOENT)
1667                        most_interesting_errno = errno;
1668        }
1669        errno = most_interesting_errno;
1670        return -1;
1671}
1672
1673void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1674{
1675        const char *path;
1676        void *map;
1677        int fd;
1678
1679        fd = open_sha1_file(sha1, &path);
1680        map = NULL;
1681        if (fd >= 0) {
1682                struct stat st;
1683
1684                if (!fstat(fd, &st)) {
1685                        *size = xsize_t(st.st_size);
1686                        if (!*size) {
1687                                /* mmap() is forbidden on empty files */
1688                                error("object file %s is empty", path);
1689                                return NULL;
1690                        }
1691                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1692                }
1693                close(fd);
1694        }
1695        return map;
1696}
1697
1698unsigned long unpack_object_header_buffer(const unsigned char *buf,
1699                unsigned long len, enum object_type *type, unsigned long *sizep)
1700{
1701        unsigned shift;
1702        unsigned long size, c;
1703        unsigned long used = 0;
1704
1705        c = buf[used++];
1706        *type = (c >> 4) & 7;
1707        size = c & 15;
1708        shift = 4;
1709        while (c & 0x80) {
1710                if (len <= used || bitsizeof(long) <= shift) {
1711                        error("bad object header");
1712                        size = used = 0;
1713                        break;
1714                }
1715                c = buf[used++];
1716                size += (c & 0x7f) << shift;
1717                shift += 7;
1718        }
1719        *sizep = size;
1720        return used;
1721}
1722
1723static int unpack_sha1_short_header(git_zstream *stream,
1724                                    unsigned char *map, unsigned long mapsize,
1725                                    void *buffer, unsigned long bufsiz)
1726{
1727        /* Get the data stream */
1728        memset(stream, 0, sizeof(*stream));
1729        stream->next_in = map;
1730        stream->avail_in = mapsize;
1731        stream->next_out = buffer;
1732        stream->avail_out = bufsiz;
1733
1734        git_inflate_init(stream);
1735        return git_inflate(stream, 0);
1736}
1737
1738int unpack_sha1_header(git_zstream *stream,
1739                       unsigned char *map, unsigned long mapsize,
1740                       void *buffer, unsigned long bufsiz)
1741{
1742        int status = unpack_sha1_short_header(stream, map, mapsize,
1743                                              buffer, bufsiz);
1744
1745        if (status < Z_OK)
1746                return status;
1747
1748        /* Make sure we have the terminating NUL */
1749        if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1750                return -1;
1751        return 0;
1752}
1753
1754static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1755                                        unsigned long mapsize, void *buffer,
1756                                        unsigned long bufsiz, struct strbuf *header)
1757{
1758        int status;
1759
1760        status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1761        if (status < Z_OK)
1762                return -1;
1763
1764        /*
1765         * Check if entire header is unpacked in the first iteration.
1766         */
1767        if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1768                return 0;
1769
1770        /*
1771         * buffer[0..bufsiz] was not large enough.  Copy the partial
1772         * result out to header, and then append the result of further
1773         * reading the stream.
1774         */
1775        strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1776        stream->next_out = buffer;
1777        stream->avail_out = bufsiz;
1778
1779        do {
1780                status = git_inflate(stream, 0);
1781                strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1782                if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1783                        return 0;
1784                stream->next_out = buffer;
1785                stream->avail_out = bufsiz;
1786        } while (status != Z_STREAM_END);
1787        return -1;
1788}
1789
1790static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1791{
1792        int bytes = strlen(buffer) + 1;
1793        unsigned char *buf = xmallocz(size);
1794        unsigned long n;
1795        int status = Z_OK;
1796
1797        n = stream->total_out - bytes;
1798        if (n > size)
1799                n = size;
1800        memcpy(buf, (char *) buffer + bytes, n);
1801        bytes = n;
1802        if (bytes <= size) {
1803                /*
1804                 * The above condition must be (bytes <= size), not
1805                 * (bytes < size).  In other words, even though we
1806                 * expect no more output and set avail_out to zero,
1807                 * the input zlib stream may have bytes that express
1808                 * "this concludes the stream", and we *do* want to
1809                 * eat that input.
1810                 *
1811                 * Otherwise we would not be able to test that we
1812                 * consumed all the input to reach the expected size;
1813                 * we also want to check that zlib tells us that all
1814                 * went well with status == Z_STREAM_END at the end.
1815                 */
1816                stream->next_out = buf + bytes;
1817                stream->avail_out = size - bytes;
1818                while (status == Z_OK)
1819                        status = git_inflate(stream, Z_FINISH);
1820        }
1821        if (status == Z_STREAM_END && !stream->avail_in) {
1822                git_inflate_end(stream);
1823                return buf;
1824        }
1825
1826        if (status < 0)
1827                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1828        else if (stream->avail_in)
1829                error("garbage at end of loose object '%s'",
1830                      sha1_to_hex(sha1));
1831        free(buf);
1832        return NULL;
1833}
1834
1835/*
1836 * We used to just use "sscanf()", but that's actually way
1837 * too permissive for what we want to check. So do an anal
1838 * object header parse by hand.
1839 */
1840static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1841                               unsigned int flags)
1842{
1843        const char *type_buf = hdr;
1844        unsigned long size;
1845        int type, type_len = 0;
1846
1847        /*
1848         * The type can be of any size but is followed by
1849         * a space.
1850         */
1851        for (;;) {
1852                char c = *hdr++;
1853                if (!c)
1854                        return -1;
1855                if (c == ' ')
1856                        break;
1857                type_len++;
1858        }
1859
1860        type = type_from_string_gently(type_buf, type_len, 1);
1861        if (oi->typename)
1862                strbuf_add(oi->typename, type_buf, type_len);
1863        /*
1864         * Set type to 0 if its an unknown object and
1865         * we're obtaining the type using '--allow-unknown-type'
1866         * option.
1867         */
1868        if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1869                type = 0;
1870        else if (type < 0)
1871                die("invalid object type");
1872        if (oi->typep)
1873                *oi->typep = type;
1874
1875        /*
1876         * The length must follow immediately, and be in canonical
1877         * decimal format (ie "010" is not valid).
1878         */
1879        size = *hdr++ - '0';
1880        if (size > 9)
1881                return -1;
1882        if (size) {
1883                for (;;) {
1884                        unsigned long c = *hdr - '0';
1885                        if (c > 9)
1886                                break;
1887                        hdr++;
1888                        size = size * 10 + c;
1889                }
1890        }
1891
1892        if (oi->sizep)
1893                *oi->sizep = size;
1894
1895        /*
1896         * The length must be followed by a zero byte
1897         */
1898        return *hdr ? -1 : type;
1899}
1900
1901int parse_sha1_header(const char *hdr, unsigned long *sizep)
1902{
1903        struct object_info oi = OBJECT_INFO_INIT;
1904
1905        oi.sizep = sizep;
1906        return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1907}
1908
1909static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1910{
1911        int ret;
1912        git_zstream stream;
1913        char hdr[8192];
1914
1915        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1916        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1917                return NULL;
1918
1919        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1920}
1921
1922unsigned long get_size_from_delta(struct packed_git *p,
1923                                  struct pack_window **w_curs,
1924                                  off_t curpos)
1925{
1926        const unsigned char *data;
1927        unsigned char delta_head[20], *in;
1928        git_zstream stream;
1929        int st;
1930
1931        memset(&stream, 0, sizeof(stream));
1932        stream.next_out = delta_head;
1933        stream.avail_out = sizeof(delta_head);
1934
1935        git_inflate_init(&stream);
1936        do {
1937                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1938                stream.next_in = in;
1939                st = git_inflate(&stream, Z_FINISH);
1940                curpos += stream.next_in - in;
1941        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1942                 stream.total_out < sizeof(delta_head));
1943        git_inflate_end(&stream);
1944        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1945                error("delta data unpack-initial failed");
1946                return 0;
1947        }
1948
1949        /* Examine the initial part of the delta to figure out
1950         * the result size.
1951         */
1952        data = delta_head;
1953
1954        /* ignore base size */
1955        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1956
1957        /* Read the result size */
1958        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1959}
1960
1961static off_t get_delta_base(struct packed_git *p,
1962                                    struct pack_window **w_curs,
1963                                    off_t *curpos,
1964                                    enum object_type type,
1965                                    off_t delta_obj_offset)
1966{
1967        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1968        off_t base_offset;
1969
1970        /* use_pack() assured us we have [base_info, base_info + 20)
1971         * as a range that we can look at without walking off the
1972         * end of the mapped window.  Its actually the hash size
1973         * that is assured.  An OFS_DELTA longer than the hash size
1974         * is stupid, as then a REF_DELTA would be smaller to store.
1975         */
1976        if (type == OBJ_OFS_DELTA) {
1977                unsigned used = 0;
1978                unsigned char c = base_info[used++];
1979                base_offset = c & 127;
1980                while (c & 128) {
1981                        base_offset += 1;
1982                        if (!base_offset || MSB(base_offset, 7))
1983                                return 0;  /* overflow */
1984                        c = base_info[used++];
1985                        base_offset = (base_offset << 7) + (c & 127);
1986                }
1987                base_offset = delta_obj_offset - base_offset;
1988                if (base_offset <= 0 || base_offset >= delta_obj_offset)
1989                        return 0;  /* out of bound */
1990                *curpos += used;
1991        } else if (type == OBJ_REF_DELTA) {
1992                /* The base entry _must_ be in the same pack */
1993                base_offset = find_pack_entry_one(base_info, p);
1994                *curpos += 20;
1995        } else
1996                die("I am totally screwed");
1997        return base_offset;
1998}
1999
2000/*
2001 * Like get_delta_base above, but we return the sha1 instead of the pack
2002 * offset. This means it is cheaper for REF deltas (we do not have to do
2003 * the final object lookup), but more expensive for OFS deltas (we
2004 * have to load the revidx to convert the offset back into a sha1).
2005 */
2006static const unsigned char *get_delta_base_sha1(struct packed_git *p,
2007                                                struct pack_window **w_curs,
2008                                                off_t curpos,
2009                                                enum object_type type,
2010                                                off_t delta_obj_offset)
2011{
2012        if (type == OBJ_REF_DELTA) {
2013                unsigned char *base = use_pack(p, w_curs, curpos, NULL);
2014                return base;
2015        } else if (type == OBJ_OFS_DELTA) {
2016                struct revindex_entry *revidx;
2017                off_t base_offset = get_delta_base(p, w_curs, &curpos,
2018                                                   type, delta_obj_offset);
2019
2020                if (!base_offset)
2021                        return NULL;
2022
2023                revidx = find_pack_revindex(p, base_offset);
2024                if (!revidx)
2025                        return NULL;
2026
2027                return nth_packed_object_sha1(p, revidx->nr);
2028        } else
2029                return NULL;
2030}
2031
2032int unpack_object_header(struct packed_git *p,
2033                         struct pack_window **w_curs,
2034                         off_t *curpos,
2035                         unsigned long *sizep)
2036{
2037        unsigned char *base;
2038        unsigned long left;
2039        unsigned long used;
2040        enum object_type type;
2041
2042        /* use_pack() assures us we have [base, base + 20) available
2043         * as a range that we can look at.  (Its actually the hash
2044         * size that is assured.)  With our object header encoding
2045         * the maximum deflated object size is 2^137, which is just
2046         * insane, so we know won't exceed what we have been given.
2047         */
2048        base = use_pack(p, w_curs, *curpos, &left);
2049        used = unpack_object_header_buffer(base, left, &type, sizep);
2050        if (!used) {
2051                type = OBJ_BAD;
2052        } else
2053                *curpos += used;
2054
2055        return type;
2056}
2057
2058static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
2059{
2060        int type;
2061        struct revindex_entry *revidx;
2062        const unsigned char *sha1;
2063        revidx = find_pack_revindex(p, obj_offset);
2064        if (!revidx)
2065                return OBJ_BAD;
2066        sha1 = nth_packed_object_sha1(p, revidx->nr);
2067        mark_bad_packed_object(p, sha1);
2068        type = sha1_object_info(sha1, NULL);
2069        if (type <= OBJ_NONE)
2070                return OBJ_BAD;
2071        return type;
2072}
2073
2074#define POI_STACK_PREALLOC 64
2075
2076static enum object_type packed_to_object_type(struct packed_git *p,
2077                                              off_t obj_offset,
2078                                              enum object_type type,
2079                                              struct pack_window **w_curs,
2080                                              off_t curpos)
2081{
2082        off_t small_poi_stack[POI_STACK_PREALLOC];
2083        off_t *poi_stack = small_poi_stack;
2084        int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
2085
2086        while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2087                off_t base_offset;
2088                unsigned long size;
2089                /* Push the object we're going to leave behind */
2090                if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2091                        poi_stack_alloc = alloc_nr(poi_stack_nr);
2092                        ALLOC_ARRAY(poi_stack, poi_stack_alloc);
2093                        memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2094                } else {
2095                        ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2096                }
2097                poi_stack[poi_stack_nr++] = obj_offset;
2098                /* If parsing the base offset fails, just unwind */
2099                base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
2100                if (!base_offset)
2101                        goto unwind;
2102                curpos = obj_offset = base_offset;
2103                type = unpack_object_header(p, w_curs, &curpos, &size);
2104                if (type <= OBJ_NONE) {
2105                        /* If getting the base itself fails, we first
2106                         * retry the base, otherwise unwind */
2107                        type = retry_bad_packed_offset(p, base_offset);
2108                        if (type > OBJ_NONE)
2109                                goto out;
2110                        goto unwind;
2111                }
2112        }
2113
2114        switch (type) {
2115        case OBJ_BAD:
2116        case OBJ_COMMIT:
2117        case OBJ_TREE:
2118        case OBJ_BLOB:
2119        case OBJ_TAG:
2120                break;
2121        default:
2122                error("unknown object type %i at offset %"PRIuMAX" in %s",
2123                      type, (uintmax_t)obj_offset, p->pack_name);
2124                type = OBJ_BAD;
2125        }
2126
2127out:
2128        if (poi_stack != small_poi_stack)
2129                free(poi_stack);
2130        return type;
2131
2132unwind:
2133        while (poi_stack_nr) {
2134                obj_offset = poi_stack[--poi_stack_nr];
2135                type = retry_bad_packed_offset(p, obj_offset);
2136                if (type > OBJ_NONE)
2137                        goto out;
2138        }
2139        type = OBJ_BAD;
2140        goto out;
2141}
2142
2143int packed_object_info(struct packed_git *p, off_t obj_offset,
2144                       struct object_info *oi)
2145{
2146        struct pack_window *w_curs = NULL;
2147        unsigned long size;
2148        off_t curpos = obj_offset;
2149        enum object_type type;
2150
2151        /*
2152         * We always get the representation type, but only convert it to
2153         * a "real" type later if the caller is interested.
2154         */
2155        type = unpack_object_header(p, &w_curs, &curpos, &size);
2156
2157        if (oi->sizep) {
2158                if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2159                        off_t tmp_pos = curpos;
2160                        off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2161                                                           type, obj_offset);
2162                        if (!base_offset) {
2163                                type = OBJ_BAD;
2164                                goto out;
2165                        }
2166                        *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2167                        if (*oi->sizep == 0) {
2168                                type = OBJ_BAD;
2169                                goto out;
2170                        }
2171                } else {
2172                        *oi->sizep = size;
2173                }
2174        }
2175
2176        if (oi->disk_sizep) {
2177                struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2178                *oi->disk_sizep = revidx[1].offset - obj_offset;
2179        }
2180
2181        if (oi->typep) {
2182                *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2183                if (*oi->typep < 0) {
2184                        type = OBJ_BAD;
2185                        goto out;
2186                }
2187        }
2188
2189        if (oi->delta_base_sha1) {
2190                if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2191                        const unsigned char *base;
2192
2193                        base = get_delta_base_sha1(p, &w_curs, curpos,
2194                                                   type, obj_offset);
2195                        if (!base) {
2196                                type = OBJ_BAD;
2197                                goto out;
2198                        }
2199
2200                        hashcpy(oi->delta_base_sha1, base);
2201                } else
2202                        hashclr(oi->delta_base_sha1);
2203        }
2204
2205out:
2206        unuse_pack(&w_curs);
2207        return type;
2208}
2209
2210static void *unpack_compressed_entry(struct packed_git *p,
2211                                    struct pack_window **w_curs,
2212                                    off_t curpos,
2213                                    unsigned long size)
2214{
2215        int st;
2216        git_zstream stream;
2217        unsigned char *buffer, *in;
2218
2219        buffer = xmallocz_gently(size);
2220        if (!buffer)
2221                return NULL;
2222        memset(&stream, 0, sizeof(stream));
2223        stream.next_out = buffer;
2224        stream.avail_out = size + 1;
2225
2226        git_inflate_init(&stream);
2227        do {
2228                in = use_pack(p, w_curs, curpos, &stream.avail_in);
2229                stream.next_in = in;
2230                st = git_inflate(&stream, Z_FINISH);
2231                if (!stream.avail_out)
2232                        break; /* the payload is larger than it should be */
2233                curpos += stream.next_in - in;
2234        } while (st == Z_OK || st == Z_BUF_ERROR);
2235        git_inflate_end(&stream);
2236        if ((st != Z_STREAM_END) || stream.total_out != size) {
2237                free(buffer);
2238                return NULL;
2239        }
2240
2241        return buffer;
2242}
2243
2244static struct hashmap delta_base_cache;
2245static size_t delta_base_cached;
2246
2247static LIST_HEAD(delta_base_cache_lru);
2248
2249struct delta_base_cache_key {
2250        struct packed_git *p;
2251        off_t base_offset;
2252};
2253
2254struct delta_base_cache_entry {
2255        struct hashmap hash;
2256        struct delta_base_cache_key key;
2257        struct list_head lru;
2258        void *data;
2259        unsigned long size;
2260        enum object_type type;
2261};
2262
2263static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
2264{
2265        unsigned int hash;
2266
2267        hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
2268        hash += (hash >> 8) + (hash >> 16);
2269        return hash;
2270}
2271
2272static struct delta_base_cache_entry *
2273get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2274{
2275        struct hashmap_entry entry;
2276        struct delta_base_cache_key key;
2277
2278        if (!delta_base_cache.cmpfn)
2279                return NULL;
2280
2281        hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2282        key.p = p;
2283        key.base_offset = base_offset;
2284        return hashmap_get(&delta_base_cache, &entry, &key);
2285}
2286
2287static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2288                                   const struct delta_base_cache_key *b)
2289{
2290        return a->p == b->p && a->base_offset == b->base_offset;
2291}
2292
2293static int delta_base_cache_hash_cmp(const void *va, const void *vb,
2294                                     const void *vkey)
2295{
2296        const struct delta_base_cache_entry *a = va, *b = vb;
2297        const struct delta_base_cache_key *key = vkey;
2298        if (key)
2299                return !delta_base_cache_key_eq(&a->key, key);
2300        else
2301                return !delta_base_cache_key_eq(&a->key, &b->key);
2302}
2303
2304static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2305{
2306        return !!get_delta_base_cache_entry(p, base_offset);
2307}
2308
2309/*
2310 * Remove the entry from the cache, but do _not_ free the associated
2311 * entry data. The caller takes ownership of the "data" buffer, and
2312 * should copy out any fields it wants before detaching.
2313 */
2314static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2315{
2316        hashmap_remove(&delta_base_cache, ent, &ent->key);
2317        list_del(&ent->lru);
2318        delta_base_cached -= ent->size;
2319        free(ent);
2320}
2321
2322static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2323        unsigned long *base_size, enum object_type *type)
2324{
2325        struct delta_base_cache_entry *ent;
2326
2327        ent = get_delta_base_cache_entry(p, base_offset);
2328        if (!ent)
2329                return unpack_entry(p, base_offset, type, base_size);
2330
2331        *type = ent->type;
2332        *base_size = ent->size;
2333        return xmemdupz(ent->data, ent->size);
2334}
2335
2336static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2337{
2338        free(ent->data);
2339        detach_delta_base_cache_entry(ent);
2340}
2341
2342void clear_delta_base_cache(void)
2343{
2344        struct hashmap_iter iter;
2345        struct delta_base_cache_entry *entry;
2346        for (entry = hashmap_iter_first(&delta_base_cache, &iter);
2347             entry;
2348             entry = hashmap_iter_next(&iter)) {
2349                release_delta_base_cache(entry);
2350        }
2351}
2352
2353static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2354        void *base, unsigned long base_size, enum object_type type)
2355{
2356        struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
2357        struct list_head *lru, *tmp;
2358
2359        delta_base_cached += base_size;
2360
2361        list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2362                struct delta_base_cache_entry *f =
2363                        list_entry(lru, struct delta_base_cache_entry, lru);
2364                if (delta_base_cached <= delta_base_cache_limit)
2365                        break;
2366                release_delta_base_cache(f);
2367        }
2368
2369        ent->key.p = p;
2370        ent->key.base_offset = base_offset;
2371        ent->type = type;
2372        ent->data = base;
2373        ent->size = base_size;
2374        list_add_tail(&ent->lru, &delta_base_cache_lru);
2375
2376        if (!delta_base_cache.cmpfn)
2377                hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, 0);
2378        hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2379        hashmap_add(&delta_base_cache, ent);
2380}
2381
2382static void *read_object(const unsigned char *sha1, enum object_type *type,
2383                         unsigned long *size);
2384
2385static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2386{
2387        static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2388        trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2389                         p->pack_name, (uintmax_t)obj_offset);
2390}
2391
2392int do_check_packed_object_crc;
2393
2394#define UNPACK_ENTRY_STACK_PREALLOC 64
2395struct unpack_entry_stack_ent {
2396        off_t obj_offset;
2397        off_t curpos;
2398        unsigned long size;
2399};
2400
2401void *unpack_entry(struct packed_git *p, off_t obj_offset,
2402                   enum object_type *final_type, unsigned long *final_size)
2403{
2404        struct pack_window *w_curs = NULL;
2405        off_t curpos = obj_offset;
2406        void *data = NULL;
2407        unsigned long size;
2408        enum object_type type;
2409        struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2410        struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2411        int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2412        int base_from_cache = 0;
2413
2414        write_pack_access_log(p, obj_offset);
2415
2416        /* PHASE 1: drill down to the innermost base object */
2417        for (;;) {
2418                off_t base_offset;
2419                int i;
2420                struct delta_base_cache_entry *ent;
2421
2422                ent = get_delta_base_cache_entry(p, curpos);
2423                if (ent) {
2424                        type = ent->type;
2425                        data = ent->data;
2426                        size = ent->size;
2427                        detach_delta_base_cache_entry(ent);
2428                        base_from_cache = 1;
2429                        break;
2430                }
2431
2432                if (do_check_packed_object_crc && p->index_version > 1) {
2433                        struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2434                        off_t len = revidx[1].offset - obj_offset;
2435                        if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2436                                const unsigned char *sha1 =
2437                                        nth_packed_object_sha1(p, revidx->nr);
2438                                error("bad packed object CRC for %s",
2439                                      sha1_to_hex(sha1));
2440                                mark_bad_packed_object(p, sha1);
2441                                unuse_pack(&w_curs);
2442                                return NULL;
2443                        }
2444                }
2445
2446                type = unpack_object_header(p, &w_curs, &curpos, &size);
2447                if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2448                        break;
2449
2450                base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2451                if (!base_offset) {
2452                        error("failed to validate delta base reference "
2453                              "at offset %"PRIuMAX" from %s",
2454                              (uintmax_t)curpos, p->pack_name);
2455                        /* bail to phase 2, in hopes of recovery */
2456                        data = NULL;
2457                        break;
2458                }
2459
2460                /* push object, proceed to base */
2461                if (delta_stack_nr >= delta_stack_alloc
2462                    && delta_stack == small_delta_stack) {
2463                        delta_stack_alloc = alloc_nr(delta_stack_nr);
2464                        ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2465                        memcpy(delta_stack, small_delta_stack,
2466                               sizeof(*delta_stack)*delta_stack_nr);
2467                } else {
2468                        ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2469                }
2470                i = delta_stack_nr++;
2471                delta_stack[i].obj_offset = obj_offset;
2472                delta_stack[i].curpos = curpos;
2473                delta_stack[i].size = size;
2474
2475                curpos = obj_offset = base_offset;
2476        }
2477
2478        /* PHASE 2: handle the base */
2479        switch (type) {
2480        case OBJ_OFS_DELTA:
2481        case OBJ_REF_DELTA:
2482                if (data)
2483                        die("BUG: unpack_entry: left loop at a valid delta");
2484                break;
2485        case OBJ_COMMIT:
2486        case OBJ_TREE:
2487        case OBJ_BLOB:
2488        case OBJ_TAG:
2489                if (!base_from_cache)
2490                        data = unpack_compressed_entry(p, &w_curs, curpos, size);
2491                break;
2492        default:
2493                data = NULL;
2494                error("unknown object type %i at offset %"PRIuMAX" in %s",
2495                      type, (uintmax_t)obj_offset, p->pack_name);
2496        }
2497
2498        /* PHASE 3: apply deltas in order */
2499
2500        /* invariants:
2501         *   'data' holds the base data, or NULL if there was corruption
2502         */
2503        while (delta_stack_nr) {
2504                void *delta_data;
2505                void *base = data;
2506                unsigned long delta_size, base_size = size;
2507                int i;
2508
2509                data = NULL;
2510
2511                if (base)
2512                        add_delta_base_cache(p, obj_offset, base, base_size, type);
2513
2514                if (!base) {
2515                        /*
2516                         * We're probably in deep shit, but let's try to fetch
2517                         * the required base anyway from another pack or loose.
2518                         * This is costly but should happen only in the presence
2519                         * of a corrupted pack, and is better than failing outright.
2520                         */
2521                        struct revindex_entry *revidx;
2522                        const unsigned char *base_sha1;
2523                        revidx = find_pack_revindex(p, obj_offset);
2524                        if (revidx) {
2525                                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2526                                error("failed to read delta base object %s"
2527                                      " at offset %"PRIuMAX" from %s",
2528                                      sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2529                                      p->pack_name);
2530                                mark_bad_packed_object(p, base_sha1);
2531                                base = read_object(base_sha1, &type, &base_size);
2532                        }
2533                }
2534
2535                i = --delta_stack_nr;
2536                obj_offset = delta_stack[i].obj_offset;
2537                curpos = delta_stack[i].curpos;
2538                delta_size = delta_stack[i].size;
2539
2540                if (!base)
2541                        continue;
2542
2543                delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2544
2545                if (!delta_data) {
2546                        error("failed to unpack compressed delta "
2547                              "at offset %"PRIuMAX" from %s",
2548                              (uintmax_t)curpos, p->pack_name);
2549                        data = NULL;
2550                        continue;
2551                }
2552
2553                data = patch_delta(base, base_size,
2554                                   delta_data, delta_size,
2555                                   &size);
2556
2557                /*
2558                 * We could not apply the delta; warn the user, but keep going.
2559                 * Our failure will be noticed either in the next iteration of
2560                 * the loop, or if this is the final delta, in the caller when
2561                 * we return NULL. Those code paths will take care of making
2562                 * a more explicit warning and retrying with another copy of
2563                 * the object.
2564                 */
2565                if (!data)
2566                        error("failed to apply delta");
2567
2568                free(delta_data);
2569        }
2570
2571        *final_type = type;
2572        *final_size = size;
2573
2574        unuse_pack(&w_curs);
2575
2576        if (delta_stack != small_delta_stack)
2577                free(delta_stack);
2578
2579        return data;
2580}
2581
2582const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2583                                            uint32_t n)
2584{
2585        const unsigned char *index = p->index_data;
2586        if (!index) {
2587                if (open_pack_index(p))
2588                        return NULL;
2589                index = p->index_data;
2590        }
2591        if (n >= p->num_objects)
2592                return NULL;
2593        index += 4 * 256;
2594        if (p->index_version == 1) {
2595                return index + 24 * n + 4;
2596        } else {
2597                index += 8;
2598                return index + 20 * n;
2599        }
2600}
2601
2602void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2603{
2604        const unsigned char *ptr = vptr;
2605        const unsigned char *start = p->index_data;
2606        const unsigned char *end = start + p->index_size;
2607        if (ptr < start)
2608                die(_("offset before start of pack index for %s (corrupt index?)"),
2609                    p->pack_name);
2610        /* No need to check for underflow; .idx files must be at least 8 bytes */
2611        if (ptr >= end - 8)
2612                die(_("offset beyond end of pack index for %s (truncated index?)"),
2613                    p->pack_name);
2614}
2615
2616off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2617{
2618        const unsigned char *index = p->index_data;
2619        index += 4 * 256;
2620        if (p->index_version == 1) {
2621                return ntohl(*((uint32_t *)(index + 24 * n)));
2622        } else {
2623                uint32_t off;
2624                index += 8 + p->num_objects * (20 + 4);
2625                off = ntohl(*((uint32_t *)(index + 4 * n)));
2626                if (!(off & 0x80000000))
2627                        return off;
2628                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2629                check_pack_index_ptr(p, index);
2630                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2631                                   ntohl(*((uint32_t *)(index + 4)));
2632        }
2633}
2634
2635off_t find_pack_entry_one(const unsigned char *sha1,
2636                                  struct packed_git *p)
2637{
2638        const uint32_t *level1_ofs = p->index_data;
2639        const unsigned char *index = p->index_data;
2640        unsigned hi, lo, stride;
2641        static int use_lookup = -1;
2642        static int debug_lookup = -1;
2643
2644        if (debug_lookup < 0)
2645                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2646
2647        if (!index) {
2648                if (open_pack_index(p))
2649                        return 0;
2650                level1_ofs = p->index_data;
2651                index = p->index_data;
2652        }
2653        if (p->index_version > 1) {
2654                level1_ofs += 2;
2655                index += 8;
2656        }
2657        index += 4 * 256;
2658        hi = ntohl(level1_ofs[*sha1]);
2659        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2660        if (p->index_version > 1) {
2661                stride = 20;
2662        } else {
2663                stride = 24;
2664                index += 4;
2665        }
2666
2667        if (debug_lookup)
2668                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2669                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2670
2671        if (use_lookup < 0)
2672                use_lookup = !!getenv("GIT_USE_LOOKUP");
2673        if (use_lookup) {
2674                int pos = sha1_entry_pos(index, stride, 0,
2675                                         lo, hi, p->num_objects, sha1);
2676                if (pos < 0)
2677                        return 0;
2678                return nth_packed_object_offset(p, pos);
2679        }
2680
2681        do {
2682                unsigned mi = (lo + hi) / 2;
2683                int cmp = hashcmp(index + mi * stride, sha1);
2684
2685                if (debug_lookup)
2686                        printf("lo %u hi %u rg %u mi %u\n",
2687                               lo, hi, hi - lo, mi);
2688                if (!cmp)
2689                        return nth_packed_object_offset(p, mi);
2690                if (cmp > 0)
2691                        hi = mi;
2692                else
2693                        lo = mi+1;
2694        } while (lo < hi);
2695        return 0;
2696}
2697
2698int is_pack_valid(struct packed_git *p)
2699{
2700        /* An already open pack is known to be valid. */
2701        if (p->pack_fd != -1)
2702                return 1;
2703
2704        /* If the pack has one window completely covering the
2705         * file size, the pack is known to be valid even if
2706         * the descriptor is not currently open.
2707         */
2708        if (p->windows) {
2709                struct pack_window *w = p->windows;
2710
2711                if (!w->offset && w->len == p->pack_size)
2712                        return 1;
2713        }
2714
2715        /* Force the pack to open to prove its valid. */
2716        return !open_packed_git(p);
2717}
2718
2719static int fill_pack_entry(const unsigned char *sha1,
2720                           struct pack_entry *e,
2721                           struct packed_git *p)
2722{
2723        off_t offset;
2724
2725        if (p->num_bad_objects) {
2726                unsigned i;
2727                for (i = 0; i < p->num_bad_objects; i++)
2728                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2729                                return 0;
2730        }
2731
2732        offset = find_pack_entry_one(sha1, p);
2733        if (!offset)
2734                return 0;
2735
2736        /*
2737         * We are about to tell the caller where they can locate the
2738         * requested object.  We better make sure the packfile is
2739         * still here and can be accessed before supplying that
2740         * answer, as it may have been deleted since the index was
2741         * loaded!
2742         */
2743        if (!is_pack_valid(p))
2744                return 0;
2745        e->offset = offset;
2746        e->p = p;
2747        hashcpy(e->sha1, sha1);
2748        return 1;
2749}
2750
2751/*
2752 * Iff a pack file contains the object named by sha1, return true and
2753 * store its location to e.
2754 */
2755static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2756{
2757        struct mru_entry *p;
2758
2759        prepare_packed_git();
2760        if (!packed_git)
2761                return 0;
2762
2763        for (p = packed_git_mru->head; p; p = p->next) {
2764                if (fill_pack_entry(sha1, e, p->item)) {
2765                        mru_mark(packed_git_mru, p);
2766                        return 1;
2767                }
2768        }
2769        return 0;
2770}
2771
2772struct packed_git *find_sha1_pack(const unsigned char *sha1,
2773                                  struct packed_git *packs)
2774{
2775        struct packed_git *p;
2776
2777        for (p = packs; p; p = p->next) {
2778                if (find_pack_entry_one(sha1, p))
2779                        return p;
2780        }
2781        return NULL;
2782
2783}
2784
2785static int sha1_loose_object_info(const unsigned char *sha1,
2786                                  struct object_info *oi,
2787                                  int flags)
2788{
2789        int status = 0;
2790        unsigned long mapsize;
2791        void *map;
2792        git_zstream stream;
2793        char hdr[32];
2794        struct strbuf hdrbuf = STRBUF_INIT;
2795
2796        if (oi->delta_base_sha1)
2797                hashclr(oi->delta_base_sha1);
2798
2799        /*
2800         * If we don't care about type or size, then we don't
2801         * need to look inside the object at all. Note that we
2802         * do not optimize out the stat call, even if the
2803         * caller doesn't care about the disk-size, since our
2804         * return value implicitly indicates whether the
2805         * object even exists.
2806         */
2807        if (!oi->typep && !oi->typename && !oi->sizep) {
2808                const char *path;
2809                struct stat st;
2810                if (stat_sha1_file(sha1, &st, &path) < 0)
2811                        return -1;
2812                if (oi->disk_sizep)
2813                        *oi->disk_sizep = st.st_size;
2814                return 0;
2815        }
2816
2817        map = map_sha1_file(sha1, &mapsize);
2818        if (!map)
2819                return -1;
2820        if (oi->disk_sizep)
2821                *oi->disk_sizep = mapsize;
2822        if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2823                if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2824                        status = error("unable to unpack %s header with --allow-unknown-type",
2825                                       sha1_to_hex(sha1));
2826        } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2827                status = error("unable to unpack %s header",
2828                               sha1_to_hex(sha1));
2829        if (status < 0)
2830                ; /* Do nothing */
2831        else if (hdrbuf.len) {
2832                if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2833                        status = error("unable to parse %s header with --allow-unknown-type",
2834                                       sha1_to_hex(sha1));
2835        } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2836                status = error("unable to parse %s header", sha1_to_hex(sha1));
2837        git_inflate_end(&stream);
2838        munmap(map, mapsize);
2839        if (status && oi->typep)
2840                *oi->typep = status;
2841        strbuf_release(&hdrbuf);
2842        return 0;
2843}
2844
2845int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2846{
2847        struct cached_object *co;
2848        struct pack_entry e;
2849        int rtype;
2850        enum object_type real_type;
2851        const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2852
2853        co = find_cached_object(real);
2854        if (co) {
2855                if (oi->typep)
2856                        *(oi->typep) = co->type;
2857                if (oi->sizep)
2858                        *(oi->sizep) = co->size;
2859                if (oi->disk_sizep)
2860                        *(oi->disk_sizep) = 0;
2861                if (oi->delta_base_sha1)
2862                        hashclr(oi->delta_base_sha1);
2863                if (oi->typename)
2864                        strbuf_addstr(oi->typename, typename(co->type));
2865                oi->whence = OI_CACHED;
2866                return 0;
2867        }
2868
2869        if (!find_pack_entry(real, &e)) {
2870                /* Most likely it's a loose object. */
2871                if (!sha1_loose_object_info(real, oi, flags)) {
2872                        oi->whence = OI_LOOSE;
2873                        return 0;
2874                }
2875
2876                /* Not a loose object; someone else may have just packed it. */
2877                reprepare_packed_git();
2878                if (!find_pack_entry(real, &e))
2879                        return -1;
2880        }
2881
2882        /*
2883         * packed_object_info() does not follow the delta chain to
2884         * find out the real type, unless it is given oi->typep.
2885         */
2886        if (oi->typename && !oi->typep)
2887                oi->typep = &real_type;
2888
2889        rtype = packed_object_info(e.p, e.offset, oi);
2890        if (rtype < 0) {
2891                mark_bad_packed_object(e.p, real);
2892                if (oi->typep == &real_type)
2893                        oi->typep = NULL;
2894                return sha1_object_info_extended(real, oi, 0);
2895        } else if (in_delta_base_cache(e.p, e.offset)) {
2896                oi->whence = OI_DBCACHED;
2897        } else {
2898                oi->whence = OI_PACKED;
2899                oi->u.packed.offset = e.offset;
2900                oi->u.packed.pack = e.p;
2901                oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2902                                         rtype == OBJ_OFS_DELTA);
2903        }
2904        if (oi->typename)
2905                strbuf_addstr(oi->typename, typename(*oi->typep));
2906        if (oi->typep == &real_type)
2907                oi->typep = NULL;
2908
2909        return 0;
2910}
2911
2912/* returns enum object_type or negative */
2913int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2914{
2915        enum object_type type;
2916        struct object_info oi = OBJECT_INFO_INIT;
2917
2918        oi.typep = &type;
2919        oi.sizep = sizep;
2920        if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2921                return -1;
2922        return type;
2923}
2924
2925static void *read_packed_sha1(const unsigned char *sha1,
2926                              enum object_type *type, unsigned long *size)
2927{
2928        struct pack_entry e;
2929        void *data;
2930
2931        if (!find_pack_entry(sha1, &e))
2932                return NULL;
2933        data = cache_or_unpack_entry(e.p, e.offset, size, type);
2934        if (!data) {
2935                /*
2936                 * We're probably in deep shit, but let's try to fetch
2937                 * the required object anyway from another pack or loose.
2938                 * This should happen only in the presence of a corrupted
2939                 * pack, and is better than failing outright.
2940                 */
2941                error("failed to read object %s at offset %"PRIuMAX" from %s",
2942                      sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2943                mark_bad_packed_object(e.p, sha1);
2944                data = read_object(sha1, type, size);
2945        }
2946        return data;
2947}
2948
2949int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2950                      unsigned char *sha1)
2951{
2952        struct cached_object *co;
2953
2954        hash_sha1_file(buf, len, typename(type), sha1);
2955        if (has_sha1_file(sha1) || find_cached_object(sha1))
2956                return 0;
2957        ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2958        co = &cached_objects[cached_object_nr++];
2959        co->size = len;
2960        co->type = type;
2961        co->buf = xmalloc(len);
2962        memcpy(co->buf, buf, len);
2963        hashcpy(co->sha1, sha1);
2964        return 0;
2965}
2966
2967static void *read_object(const unsigned char *sha1, enum object_type *type,
2968                         unsigned long *size)
2969{
2970        unsigned long mapsize;
2971        void *map, *buf;
2972        struct cached_object *co;
2973
2974        co = find_cached_object(sha1);
2975        if (co) {
2976                *type = co->type;
2977                *size = co->size;
2978                return xmemdupz(co->buf, co->size);
2979        }
2980
2981        buf = read_packed_sha1(sha1, type, size);
2982        if (buf)
2983                return buf;
2984        map = map_sha1_file(sha1, &mapsize);
2985        if (map) {
2986                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2987                munmap(map, mapsize);
2988                return buf;
2989        }
2990        reprepare_packed_git();
2991        return read_packed_sha1(sha1, type, size);
2992}
2993
2994/*
2995 * This function dies on corrupt objects; the callers who want to
2996 * deal with them should arrange to call read_object() and give error
2997 * messages themselves.
2998 */
2999void *read_sha1_file_extended(const unsigned char *sha1,
3000                              enum object_type *type,
3001                              unsigned long *size,
3002                              unsigned flag)
3003{
3004        void *data;
3005        const struct packed_git *p;
3006        const char *path;
3007        struct stat st;
3008        const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
3009
3010        errno = 0;
3011        data = read_object(repl, type, size);
3012        if (data)
3013                return data;
3014
3015        if (errno && errno != ENOENT)
3016                die_errno("failed to read object %s", sha1_to_hex(sha1));
3017
3018        /* die if we replaced an object with one that does not exist */
3019        if (repl != sha1)
3020                die("replacement %s not found for %s",
3021                    sha1_to_hex(repl), sha1_to_hex(sha1));
3022
3023        if (!stat_sha1_file(repl, &st, &path))
3024                die("loose object %s (stored in %s) is corrupt",
3025                    sha1_to_hex(repl), path);
3026
3027        if ((p = has_packed_and_bad(repl)) != NULL)
3028                die("packed object %s (stored in %s) is corrupt",
3029                    sha1_to_hex(repl), p->pack_name);
3030
3031        return NULL;
3032}
3033
3034void *read_object_with_reference(const unsigned char *sha1,
3035                                 const char *required_type_name,
3036                                 unsigned long *size,
3037                                 unsigned char *actual_sha1_return)
3038{
3039        enum object_type type, required_type;
3040        void *buffer;
3041        unsigned long isize;
3042        unsigned char actual_sha1[20];
3043
3044        required_type = type_from_string(required_type_name);
3045        hashcpy(actual_sha1, sha1);
3046        while (1) {
3047                int ref_length = -1;
3048                const char *ref_type = NULL;
3049
3050                buffer = read_sha1_file(actual_sha1, &type, &isize);
3051                if (!buffer)
3052                        return NULL;
3053                if (type == required_type) {
3054                        *size = isize;
3055                        if (actual_sha1_return)
3056                                hashcpy(actual_sha1_return, actual_sha1);
3057                        return buffer;
3058                }
3059                /* Handle references */
3060                else if (type == OBJ_COMMIT)
3061                        ref_type = "tree ";
3062                else if (type == OBJ_TAG)
3063                        ref_type = "object ";
3064                else {
3065                        free(buffer);
3066                        return NULL;
3067                }
3068                ref_length = strlen(ref_type);
3069
3070                if (ref_length + 40 > isize ||
3071                    memcmp(buffer, ref_type, ref_length) ||
3072                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
3073                        free(buffer);
3074                        return NULL;
3075                }
3076                free(buffer);
3077                /* Now we have the ID of the referred-to object in
3078                 * actual_sha1.  Check again. */
3079        }
3080}
3081
3082static void write_sha1_file_prepare(const void *buf, unsigned long len,
3083                                    const char *type, unsigned char *sha1,
3084                                    char *hdr, int *hdrlen)
3085{
3086        git_SHA_CTX c;
3087
3088        /* Generate the header */
3089        *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
3090
3091        /* Sha1.. */
3092        git_SHA1_Init(&c);
3093        git_SHA1_Update(&c, hdr, *hdrlen);
3094        git_SHA1_Update(&c, buf, len);
3095        git_SHA1_Final(sha1, &c);
3096}
3097
3098/*
3099 * Move the just written object into its final resting place.
3100 */
3101int finalize_object_file(const char *tmpfile, const char *filename)
3102{
3103        int ret = 0;
3104
3105        if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
3106                goto try_rename;
3107        else if (link(tmpfile, filename))
3108                ret = errno;
3109
3110        /*
3111         * Coda hack - coda doesn't like cross-directory links,
3112         * so we fall back to a rename, which will mean that it
3113         * won't be able to check collisions, but that's not a
3114         * big deal.
3115         *
3116         * The same holds for FAT formatted media.
3117         *
3118         * When this succeeds, we just return.  We have nothing
3119         * left to unlink.
3120         */
3121        if (ret && ret != EEXIST) {
3122        try_rename:
3123                if (!rename(tmpfile, filename))
3124                        goto out;
3125                ret = errno;
3126        }
3127        unlink_or_warn(tmpfile);
3128        if (ret) {
3129                if (ret != EEXIST) {
3130                        return error_errno("unable to write sha1 filename %s", filename);
3131                }
3132                /* FIXME!!! Collision check here ? */
3133        }
3134
3135out:
3136        if (adjust_shared_perm(filename))
3137                return error("unable to set permission to '%s'", filename);
3138        return 0;
3139}
3140
3141static int write_buffer(int fd, const void *buf, size_t len)
3142{
3143        if (write_in_full(fd, buf, len) < 0)
3144                return error_errno("file write error");
3145        return 0;
3146}
3147
3148int hash_sha1_file(const void *buf, unsigned long len, const char *type,
3149                   unsigned char *sha1)
3150{
3151        char hdr[32];
3152        int hdrlen = sizeof(hdr);
3153        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3154        return 0;
3155}
3156
3157/* Finalize a file on disk, and close it. */
3158static void close_sha1_file(int fd)
3159{
3160        if (fsync_object_files)
3161                fsync_or_die(fd, "sha1 file");
3162        if (close(fd) != 0)
3163                die_errno("error when closing sha1 file");
3164}
3165
3166/* Size of directory component, including the ending '/' */
3167static inline int directory_size(const char *filename)
3168{
3169        const char *s = strrchr(filename, '/');
3170        if (!s)
3171                return 0;
3172        return s - filename + 1;
3173}
3174
3175/*
3176 * This creates a temporary file in the same directory as the final
3177 * 'filename'
3178 *
3179 * We want to avoid cross-directory filename renames, because those
3180 * can have problems on various filesystems (FAT, NFS, Coda).
3181 */
3182static int create_tmpfile(struct strbuf *tmp, const char *filename)
3183{
3184        int fd, dirlen = directory_size(filename);
3185
3186        strbuf_reset(tmp);
3187        strbuf_add(tmp, filename, dirlen);
3188        strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3189        fd = git_mkstemp_mode(tmp->buf, 0444);
3190        if (fd < 0 && dirlen && errno == ENOENT) {
3191                /*
3192                 * Make sure the directory exists; note that the contents
3193                 * of the buffer are undefined after mkstemp returns an
3194                 * error, so we have to rewrite the whole buffer from
3195                 * scratch.
3196                 */
3197                strbuf_reset(tmp);
3198                strbuf_add(tmp, filename, dirlen - 1);
3199                if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3200                        return -1;
3201                if (adjust_shared_perm(tmp->buf))
3202                        return -1;
3203
3204                /* Try again */
3205                strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3206                fd = git_mkstemp_mode(tmp->buf, 0444);
3207        }
3208        return fd;
3209}
3210
3211static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3212                              const void *buf, unsigned long len, time_t mtime)
3213{
3214        int fd, ret;
3215        unsigned char compressed[4096];
3216        git_zstream stream;
3217        git_SHA_CTX c;
3218        unsigned char parano_sha1[20];
3219        static struct strbuf tmp_file = STRBUF_INIT;
3220        const char *filename = sha1_file_name(sha1);
3221
3222        fd = create_tmpfile(&tmp_file, filename);
3223        if (fd < 0) {
3224                if (errno == EACCES)
3225                        return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3226                else
3227                        return error_errno("unable to create temporary file");
3228        }
3229
3230        /* Set it up */
3231        git_deflate_init(&stream, zlib_compression_level);
3232        stream.next_out = compressed;
3233        stream.avail_out = sizeof(compressed);
3234        git_SHA1_Init(&c);
3235
3236        /* First header.. */
3237        stream.next_in = (unsigned char *)hdr;
3238        stream.avail_in = hdrlen;
3239        while (git_deflate(&stream, 0) == Z_OK)
3240                ; /* nothing */
3241        git_SHA1_Update(&c, hdr, hdrlen);
3242
3243        /* Then the data itself.. */
3244        stream.next_in = (void *)buf;
3245        stream.avail_in = len;
3246        do {
3247                unsigned char *in0 = stream.next_in;
3248                ret = git_deflate(&stream, Z_FINISH);
3249                git_SHA1_Update(&c, in0, stream.next_in - in0);
3250                if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3251                        die("unable to write sha1 file");
3252                stream.next_out = compressed;
3253                stream.avail_out = sizeof(compressed);
3254        } while (ret == Z_OK);
3255
3256        if (ret != Z_STREAM_END)
3257                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3258        ret = git_deflate_end_gently(&stream);
3259        if (ret != Z_OK)
3260                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3261        git_SHA1_Final(parano_sha1, &c);
3262        if (hashcmp(sha1, parano_sha1) != 0)
3263                die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3264
3265        close_sha1_file(fd);
3266
3267        if (mtime) {
3268                struct utimbuf utb;
3269                utb.actime = mtime;
3270                utb.modtime = mtime;
3271                if (utime(tmp_file.buf, &utb) < 0)
3272                        warning_errno("failed utime() on %s", tmp_file.buf);
3273        }
3274
3275        return finalize_object_file(tmp_file.buf, filename);
3276}
3277
3278static int freshen_loose_object(const unsigned char *sha1)
3279{
3280        return check_and_freshen(sha1, 1);
3281}
3282
3283static int freshen_packed_object(const unsigned char *sha1)
3284{
3285        struct pack_entry e;
3286        if (!find_pack_entry(sha1, &e))
3287                return 0;
3288        if (e.p->freshened)
3289                return 1;
3290        if (!freshen_file(e.p->pack_name))
3291                return 0;
3292        e.p->freshened = 1;
3293        return 1;
3294}
3295
3296int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3297{
3298        char hdr[32];
3299        int hdrlen = sizeof(hdr);
3300
3301        /* Normally if we have it in the pack then we do not bother writing
3302         * it out into .git/objects/??/?{38} file.
3303         */
3304        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3305        if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3306                return 0;
3307        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3308}
3309
3310int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3311                             unsigned char *sha1, unsigned flags)
3312{
3313        char *header;
3314        int hdrlen, status = 0;
3315
3316        /* type string, SP, %lu of the length plus NUL must fit this */
3317        hdrlen = strlen(type) + 32;
3318        header = xmalloc(hdrlen);
3319        write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3320
3321        if (!(flags & HASH_WRITE_OBJECT))
3322                goto cleanup;
3323        if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3324                goto cleanup;
3325        status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3326
3327cleanup:
3328        free(header);
3329        return status;
3330}
3331
3332int force_object_loose(const unsigned char *sha1, time_t mtime)
3333{
3334        void *buf;
3335        unsigned long len;
3336        enum object_type type;
3337        char hdr[32];
3338        int hdrlen;
3339        int ret;
3340
3341        if (has_loose_object(sha1))
3342                return 0;
3343        buf = read_packed_sha1(sha1, &type, &len);
3344        if (!buf)
3345                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3346        hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3347        ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3348        free(buf);
3349
3350        return ret;
3351}
3352
3353int has_pack_index(const unsigned char *sha1)
3354{
3355        struct stat st;
3356        if (stat(sha1_pack_index_name(sha1), &st))
3357                return 0;
3358        return 1;
3359}
3360
3361int has_sha1_pack(const unsigned char *sha1)
3362{
3363        struct pack_entry e;
3364        return find_pack_entry(sha1, &e);
3365}
3366
3367int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3368{
3369        struct pack_entry e;
3370
3371        if (find_pack_entry(sha1, &e))
3372                return 1;
3373        if (has_loose_object(sha1))
3374                return 1;
3375        if (flags & HAS_SHA1_QUICK)
3376                return 0;
3377        reprepare_packed_git();
3378        return find_pack_entry(sha1, &e);
3379}
3380
3381int has_object_file(const struct object_id *oid)
3382{
3383        return has_sha1_file(oid->hash);
3384}
3385
3386int has_object_file_with_flags(const struct object_id *oid, int flags)
3387{
3388        return has_sha1_file_with_flags(oid->hash, flags);
3389}
3390
3391static void check_tree(const void *buf, size_t size)
3392{
3393        struct tree_desc desc;
3394        struct name_entry entry;
3395
3396        init_tree_desc(&desc, buf, size);
3397        while (tree_entry(&desc, &entry))
3398                /* do nothing
3399                 * tree_entry() will die() on malformed entries */
3400                ;
3401}
3402
3403static void check_commit(const void *buf, size_t size)
3404{
3405        struct commit c;
3406        memset(&c, 0, sizeof(c));
3407        if (parse_commit_buffer(&c, buf, size))
3408                die("corrupt commit");
3409}
3410
3411static void check_tag(const void *buf, size_t size)
3412{
3413        struct tag t;
3414        memset(&t, 0, sizeof(t));
3415        if (parse_tag_buffer(&t, buf, size))
3416                die("corrupt tag");
3417}
3418
3419static int index_mem(unsigned char *sha1, void *buf, size_t size,
3420                     enum object_type type,
3421                     const char *path, unsigned flags)
3422{
3423        int ret, re_allocated = 0;
3424        int write_object = flags & HASH_WRITE_OBJECT;
3425
3426        if (!type)
3427                type = OBJ_BLOB;
3428
3429        /*
3430         * Convert blobs to git internal format
3431         */
3432        if ((type == OBJ_BLOB) && path) {
3433                struct strbuf nbuf = STRBUF_INIT;
3434                if (convert_to_git(path, buf, size, &nbuf,
3435                                   write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3436                        buf = strbuf_detach(&nbuf, &size);
3437                        re_allocated = 1;
3438                }
3439        }
3440        if (flags & HASH_FORMAT_CHECK) {
3441                if (type == OBJ_TREE)
3442                        check_tree(buf, size);
3443                if (type == OBJ_COMMIT)
3444                        check_commit(buf, size);
3445                if (type == OBJ_TAG)
3446                        check_tag(buf, size);
3447        }
3448
3449        if (write_object)
3450                ret = write_sha1_file(buf, size, typename(type), sha1);
3451        else
3452                ret = hash_sha1_file(buf, size, typename(type), sha1);
3453        if (re_allocated)
3454                free(buf);
3455        return ret;
3456}
3457
3458static int index_stream_convert_blob(unsigned char *sha1, int fd,
3459                                     const char *path, unsigned flags)
3460{
3461        int ret;
3462        const int write_object = flags & HASH_WRITE_OBJECT;
3463        struct strbuf sbuf = STRBUF_INIT;
3464
3465        assert(path);
3466        assert(would_convert_to_git_filter_fd(path));
3467
3468        convert_to_git_filter_fd(path, fd, &sbuf,
3469                                 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3470
3471        if (write_object)
3472                ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3473                                      sha1);
3474        else
3475                ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3476                                     sha1);
3477        strbuf_release(&sbuf);
3478        return ret;
3479}
3480
3481static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3482                      const char *path, unsigned flags)
3483{
3484        struct strbuf sbuf = STRBUF_INIT;
3485        int ret;
3486
3487        if (strbuf_read(&sbuf, fd, 4096) >= 0)
3488                ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3489        else
3490                ret = -1;
3491        strbuf_release(&sbuf);
3492        return ret;
3493}
3494
3495#define SMALL_FILE_SIZE (32*1024)
3496
3497static int index_core(unsigned char *sha1, int fd, size_t size,
3498                      enum object_type type, const char *path,
3499                      unsigned flags)
3500{
3501        int ret;
3502
3503        if (!size) {
3504                ret = index_mem(sha1, "", size, type, path, flags);
3505        } else if (size <= SMALL_FILE_SIZE) {
3506                char *buf = xmalloc(size);
3507                if (size == read_in_full(fd, buf, size))
3508                        ret = index_mem(sha1, buf, size, type, path, flags);
3509                else
3510                        ret = error_errno("short read");
3511                free(buf);
3512        } else {
3513                void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3514                ret = index_mem(sha1, buf, size, type, path, flags);
3515                munmap(buf, size);
3516        }
3517        return ret;
3518}
3519
3520/*
3521 * This creates one packfile per large blob unless bulk-checkin
3522 * machinery is "plugged".
3523 *
3524 * This also bypasses the usual "convert-to-git" dance, and that is on
3525 * purpose. We could write a streaming version of the converting
3526 * functions and insert that before feeding the data to fast-import
3527 * (or equivalent in-core API described above). However, that is
3528 * somewhat complicated, as we do not know the size of the filter
3529 * result, which we need to know beforehand when writing a git object.
3530 * Since the primary motivation for trying to stream from the working
3531 * tree file and to avoid mmaping it in core is to deal with large
3532 * binary blobs, they generally do not want to get any conversion, and
3533 * callers should avoid this code path when filters are requested.
3534 */
3535static int index_stream(unsigned char *sha1, int fd, size_t size,
3536                        enum object_type type, const char *path,
3537                        unsigned flags)
3538{
3539        return index_bulk_checkin(sha1, fd, size, type, path, flags);
3540}
3541
3542int index_fd(unsigned char *sha1, int fd, struct stat *st,
3543             enum object_type type, const char *path, unsigned flags)
3544{
3545        int ret;
3546
3547        /*
3548         * Call xsize_t() only when needed to avoid potentially unnecessary
3549         * die() for large files.
3550         */
3551        if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3552                ret = index_stream_convert_blob(sha1, fd, path, flags);
3553        else if (!S_ISREG(st->st_mode))
3554                ret = index_pipe(sha1, fd, type, path, flags);
3555        else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3556                 (path && would_convert_to_git(path)))
3557                ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3558                                 flags);
3559        else
3560                ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3561                                   flags);
3562        close(fd);
3563        return ret;
3564}
3565
3566int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3567{
3568        int fd;
3569        struct strbuf sb = STRBUF_INIT;
3570
3571        switch (st->st_mode & S_IFMT) {
3572        case S_IFREG:
3573                fd = open(path, O_RDONLY);
3574                if (fd < 0)
3575                        return error_errno("open(\"%s\")", path);
3576                if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3577                        return error("%s: failed to insert into database",
3578                                     path);
3579                break;
3580        case S_IFLNK:
3581                if (strbuf_readlink(&sb, path, st->st_size))
3582                        return error_errno("readlink(\"%s\")", path);
3583                if (!(flags & HASH_WRITE_OBJECT))
3584                        hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3585                else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3586                        return error("%s: failed to insert into database",
3587                                     path);
3588                strbuf_release(&sb);
3589                break;
3590        case S_IFDIR:
3591                return resolve_gitlink_ref(path, "HEAD", sha1);
3592        default:
3593                return error("%s: unsupported file type", path);
3594        }
3595        return 0;
3596}
3597
3598int read_pack_header(int fd, struct pack_header *header)
3599{
3600        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3601                /* "eof before pack header was fully read" */
3602                return PH_ERROR_EOF;
3603
3604        if (header->hdr_signature != htonl(PACK_SIGNATURE))
3605                /* "protocol error (pack signature mismatch detected)" */
3606                return PH_ERROR_PACK_SIGNATURE;
3607        if (!pack_version_ok(header->hdr_version))
3608                /* "protocol error (pack version unsupported)" */
3609                return PH_ERROR_PROTOCOL;
3610        return 0;
3611}
3612
3613void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3614{
3615        enum object_type type = sha1_object_info(sha1, NULL);
3616        if (type < 0)
3617                die("%s is not a valid object", sha1_to_hex(sha1));
3618        if (type != expect)
3619                die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3620                    typename(expect));
3621}
3622
3623static int for_each_file_in_obj_subdir(int subdir_nr,
3624                                       struct strbuf *path,
3625                                       each_loose_object_fn obj_cb,
3626                                       each_loose_cruft_fn cruft_cb,
3627                                       each_loose_subdir_fn subdir_cb,
3628                                       void *data)
3629{
3630        size_t baselen = path->len;
3631        DIR *dir = opendir(path->buf);
3632        struct dirent *de;
3633        int r = 0;
3634
3635        if (!dir) {
3636                if (errno == ENOENT)
3637                        return 0;
3638                return error_errno("unable to open %s", path->buf);
3639        }
3640
3641        while ((de = readdir(dir))) {
3642                if (is_dot_or_dotdot(de->d_name))
3643                        continue;
3644
3645                strbuf_setlen(path, baselen);
3646                strbuf_addf(path, "/%s", de->d_name);
3647
3648                if (strlen(de->d_name) == 38)  {
3649                        char hex[41];
3650                        unsigned char sha1[20];
3651
3652                        snprintf(hex, sizeof(hex), "%02x%s",
3653                                 subdir_nr, de->d_name);
3654                        if (!get_sha1_hex(hex, sha1)) {
3655                                if (obj_cb) {
3656                                        r = obj_cb(sha1, path->buf, data);
3657                                        if (r)
3658                                                break;
3659                                }
3660                                continue;
3661                        }
3662                }
3663
3664                if (cruft_cb) {
3665                        r = cruft_cb(de->d_name, path->buf, data);
3666                        if (r)
3667                                break;
3668                }
3669        }
3670        closedir(dir);
3671
3672        strbuf_setlen(path, baselen);
3673        if (!r && subdir_cb)
3674                r = subdir_cb(subdir_nr, path->buf, data);
3675
3676        return r;
3677}
3678
3679int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3680                            each_loose_object_fn obj_cb,
3681                            each_loose_cruft_fn cruft_cb,
3682                            each_loose_subdir_fn subdir_cb,
3683                            void *data)
3684{
3685        size_t baselen = path->len;
3686        int r = 0;
3687        int i;
3688
3689        for (i = 0; i < 256; i++) {
3690                strbuf_addf(path, "/%02x", i);
3691                r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3692                                                subdir_cb, data);
3693                strbuf_setlen(path, baselen);
3694                if (r)
3695                        break;
3696        }
3697
3698        return r;
3699}
3700
3701int for_each_loose_file_in_objdir(const char *path,
3702                                  each_loose_object_fn obj_cb,
3703                                  each_loose_cruft_fn cruft_cb,
3704                                  each_loose_subdir_fn subdir_cb,
3705                                  void *data)
3706{
3707        struct strbuf buf = STRBUF_INIT;
3708        int r;
3709
3710        strbuf_addstr(&buf, path);
3711        r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3712                                              subdir_cb, data);
3713        strbuf_release(&buf);
3714
3715        return r;
3716}
3717
3718struct loose_alt_odb_data {
3719        each_loose_object_fn *cb;
3720        void *data;
3721};
3722
3723static int loose_from_alt_odb(struct alternate_object_database *alt,
3724                              void *vdata)
3725{
3726        struct loose_alt_odb_data *data = vdata;
3727        struct strbuf buf = STRBUF_INIT;
3728        int r;
3729
3730        strbuf_addstr(&buf, alt->path);
3731        r = for_each_loose_file_in_objdir_buf(&buf,
3732                                              data->cb, NULL, NULL,
3733                                              data->data);
3734        strbuf_release(&buf);
3735        return r;
3736}
3737
3738int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3739{
3740        struct loose_alt_odb_data alt;
3741        int r;
3742
3743        r = for_each_loose_file_in_objdir(get_object_directory(),
3744                                          cb, NULL, NULL, data);
3745        if (r)
3746                return r;
3747
3748        if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3749                return 0;
3750
3751        alt.cb = cb;
3752        alt.data = data;
3753        return foreach_alt_odb(loose_from_alt_odb, &alt);
3754}
3755
3756static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3757{
3758        uint32_t i;
3759        int r = 0;
3760
3761        for (i = 0; i < p->num_objects; i++) {
3762                const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3763
3764                if (!sha1)
3765                        return error("unable to get sha1 of object %u in %s",
3766                                     i, p->pack_name);
3767
3768                r = cb(sha1, p, i, data);
3769                if (r)
3770                        break;
3771        }
3772        return r;
3773}
3774
3775int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3776{
3777        struct packed_git *p;
3778        int r = 0;
3779        int pack_errors = 0;
3780
3781        prepare_packed_git();
3782        for (p = packed_git; p; p = p->next) {
3783                if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3784                        continue;
3785                if (open_pack_index(p)) {
3786                        pack_errors = 1;
3787                        continue;
3788                }
3789                r = for_each_object_in_pack(p, cb, data);
3790                if (r)
3791                        break;
3792        }
3793        return r ? r : pack_errors;
3794}