sha1_file.con commit Merge branch 'jk/cvsimport-quoting' into maint-2.10 (6d6e2f8)
   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
  28#ifndef O_NOATIME
  29#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  30#define O_NOATIME 01000000
  31#else
  32#define O_NOATIME 0
  33#endif
  34#endif
  35
  36#define SZ_FMT PRIuMAX
  37static inline uintmax_t sz_fmt(size_t s) { return s; }
  38
  39const unsigned char null_sha1[20];
  40const struct object_id null_oid;
  41
  42/*
  43 * This is meant to hold a *small* number of objects that you would
  44 * want read_sha1_file() to be able to return, but yet you do not want
  45 * to write them into the object store (e.g. a browse-only
  46 * application).
  47 */
  48static struct cached_object {
  49        unsigned char sha1[20];
  50        enum object_type type;
  51        void *buf;
  52        unsigned long size;
  53} *cached_objects;
  54static int cached_object_nr, cached_object_alloc;
  55
  56static struct cached_object empty_tree = {
  57        EMPTY_TREE_SHA1_BIN_LITERAL,
  58        OBJ_TREE,
  59        "",
  60        0
  61};
  62
  63static struct cached_object *find_cached_object(const unsigned char *sha1)
  64{
  65        int i;
  66        struct cached_object *co = cached_objects;
  67
  68        for (i = 0; i < cached_object_nr; i++, co++) {
  69                if (!hashcmp(co->sha1, sha1))
  70                        return co;
  71        }
  72        if (!hashcmp(sha1, empty_tree.sha1))
  73                return &empty_tree;
  74        return NULL;
  75}
  76
  77int mkdir_in_gitdir(const char *path)
  78{
  79        if (mkdir(path, 0777)) {
  80                int saved_errno = errno;
  81                struct stat st;
  82                struct strbuf sb = STRBUF_INIT;
  83
  84                if (errno != EEXIST)
  85                        return -1;
  86                /*
  87                 * Are we looking at a path in a symlinked worktree
  88                 * whose original repository does not yet have it?
  89                 * e.g. .git/rr-cache pointing at its original
  90                 * repository in which the user hasn't performed any
  91                 * conflict resolution yet?
  92                 */
  93                if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
  94                    strbuf_readlink(&sb, path, st.st_size) ||
  95                    !is_absolute_path(sb.buf) ||
  96                    mkdir(sb.buf, 0777)) {
  97                        strbuf_release(&sb);
  98                        errno = saved_errno;
  99                        return -1;
 100                }
 101                strbuf_release(&sb);
 102        }
 103        return adjust_shared_perm(path);
 104}
 105
 106enum scld_error safe_create_leading_directories(char *path)
 107{
 108        char *next_component = path + offset_1st_component(path);
 109        enum scld_error ret = SCLD_OK;
 110
 111        while (ret == SCLD_OK && next_component) {
 112                struct stat st;
 113                char *slash = next_component, slash_character;
 114
 115                while (*slash && !is_dir_sep(*slash))
 116                        slash++;
 117
 118                if (!*slash)
 119                        break;
 120
 121                next_component = slash + 1;
 122                while (is_dir_sep(*next_component))
 123                        next_component++;
 124                if (!*next_component)
 125                        break;
 126
 127                slash_character = *slash;
 128                *slash = '\0';
 129                if (!stat(path, &st)) {
 130                        /* path exists */
 131                        if (!S_ISDIR(st.st_mode))
 132                                ret = SCLD_EXISTS;
 133                } else if (mkdir(path, 0777)) {
 134                        if (errno == EEXIST &&
 135                            !stat(path, &st) && S_ISDIR(st.st_mode))
 136                                ; /* somebody created it since we checked */
 137                        else if (errno == ENOENT)
 138                                /*
 139                                 * Either mkdir() failed because
 140                                 * somebody just pruned the containing
 141                                 * directory, or stat() failed because
 142                                 * the file that was in our way was
 143                                 * just removed.  Either way, inform
 144                                 * the caller that it might be worth
 145                                 * trying again:
 146                                 */
 147                                ret = SCLD_VANISHED;
 148                        else
 149                                ret = SCLD_FAILED;
 150                } else if (adjust_shared_perm(path)) {
 151                        ret = SCLD_PERMS;
 152                }
 153                *slash = slash_character;
 154        }
 155        return ret;
 156}
 157
 158enum scld_error safe_create_leading_directories_const(const char *path)
 159{
 160        /* path points to cache entries, so xstrdup before messing with it */
 161        char *buf = xstrdup(path);
 162        enum scld_error result = safe_create_leading_directories(buf);
 163        free(buf);
 164        return result;
 165}
 166
 167static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 168{
 169        int i;
 170        for (i = 0; i < 20; i++) {
 171                static char hex[] = "0123456789abcdef";
 172                unsigned int val = sha1[i];
 173                char *pos = pathbuf + i*2 + (i > 0);
 174                *pos++ = hex[val >> 4];
 175                *pos = hex[val & 0xf];
 176        }
 177}
 178
 179const char *sha1_file_name(const unsigned char *sha1)
 180{
 181        static char buf[PATH_MAX];
 182        const char *objdir;
 183        int len;
 184
 185        objdir = get_object_directory();
 186        len = strlen(objdir);
 187
 188        /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
 189        if (len + 43 > PATH_MAX)
 190                die("insanely long object directory %s", objdir);
 191        memcpy(buf, objdir, len);
 192        buf[len] = '/';
 193        buf[len+3] = '/';
 194        buf[len+42] = '\0';
 195        fill_sha1_path(buf + len + 1, sha1);
 196        return buf;
 197}
 198
 199/*
 200 * Return the name of the pack or index file with the specified sha1
 201 * in its filename.  *base and *name are scratch space that must be
 202 * provided by the caller.  which should be "pack" or "idx".
 203 */
 204static char *sha1_get_pack_name(const unsigned char *sha1,
 205                                struct strbuf *buf,
 206                                const char *which)
 207{
 208        strbuf_reset(buf);
 209        strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
 210                    sha1_to_hex(sha1), which);
 211        return buf->buf;
 212}
 213
 214char *sha1_pack_name(const unsigned char *sha1)
 215{
 216        static struct strbuf buf = STRBUF_INIT;
 217        return sha1_get_pack_name(sha1, &buf, "pack");
 218}
 219
 220char *sha1_pack_index_name(const unsigned char *sha1)
 221{
 222        static struct strbuf buf = STRBUF_INIT;
 223        return sha1_get_pack_name(sha1, &buf, "idx");
 224}
 225
 226struct alternate_object_database *alt_odb_list;
 227static struct alternate_object_database **alt_odb_tail;
 228
 229/*
 230 * Prepare alternate object database registry.
 231 *
 232 * The variable alt_odb_list points at the list of struct
 233 * alternate_object_database.  The elements on this list come from
 234 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 235 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 236 * whose contents is similar to that environment variable but can be
 237 * LF separated.  Its base points at a statically allocated buffer that
 238 * contains "/the/directory/corresponding/to/.git/objects/...", while
 239 * its name points just after the slash at the end of ".git/objects/"
 240 * in the example above, and has enough space to hold 40-byte hex
 241 * SHA1, an extra slash for the first level indirection, and the
 242 * terminating NUL.
 243 */
 244static int link_alt_odb_entry(const char *entry, const char *relative_base,
 245        int depth, const char *normalized_objdir)
 246{
 247        struct alternate_object_database *ent;
 248        struct alternate_object_database *alt;
 249        size_t pfxlen, entlen;
 250        struct strbuf pathbuf = STRBUF_INIT;
 251
 252        if (!is_absolute_path(entry) && relative_base) {
 253                strbuf_addstr(&pathbuf, real_path(relative_base));
 254                strbuf_addch(&pathbuf, '/');
 255        }
 256        strbuf_addstr(&pathbuf, entry);
 257
 258        normalize_path_copy(pathbuf.buf, pathbuf.buf);
 259
 260        pfxlen = strlen(pathbuf.buf);
 261
 262        /*
 263         * The trailing slash after the directory name is given by
 264         * this function at the end. Remove duplicates.
 265         */
 266        while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
 267                pfxlen -= 1;
 268
 269        entlen = st_add(pfxlen, 43); /* '/' + 2 hex + '/' + 38 hex + NUL */
 270        ent = xmalloc(st_add(sizeof(*ent), entlen));
 271        memcpy(ent->base, pathbuf.buf, pfxlen);
 272        strbuf_release(&pathbuf);
 273
 274        ent->name = ent->base + pfxlen + 1;
 275        ent->base[pfxlen + 3] = '/';
 276        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 277
 278        /* Detect cases where alternate disappeared */
 279        if (!is_directory(ent->base)) {
 280                error("object directory %s does not exist; "
 281                      "check .git/objects/info/alternates.",
 282                      ent->base);
 283                free(ent);
 284                return -1;
 285        }
 286
 287        /* Prevent the common mistake of listing the same
 288         * thing twice, or object directory itself.
 289         */
 290        for (alt = alt_odb_list; alt; alt = alt->next) {
 291                if (pfxlen == alt->name - alt->base - 1 &&
 292                    !memcmp(ent->base, alt->base, pfxlen)) {
 293                        free(ent);
 294                        return -1;
 295                }
 296        }
 297        if (!fspathcmp(ent->base, normalized_objdir)) {
 298                free(ent);
 299                return -1;
 300        }
 301
 302        /* add the alternate entry */
 303        *alt_odb_tail = ent;
 304        alt_odb_tail = &(ent->next);
 305        ent->next = NULL;
 306
 307        /* recursively add alternates */
 308        read_info_alternates(ent->base, depth + 1);
 309
 310        ent->base[pfxlen] = '/';
 311
 312        return 0;
 313}
 314
 315static void link_alt_odb_entries(const char *alt, int len, int sep,
 316                                 const char *relative_base, int depth)
 317{
 318        struct string_list entries = STRING_LIST_INIT_NODUP;
 319        char *alt_copy;
 320        int i;
 321        struct strbuf objdirbuf = STRBUF_INIT;
 322
 323        if (depth > 5) {
 324                error("%s: ignoring alternate object stores, nesting too deep.",
 325                                relative_base);
 326                return;
 327        }
 328
 329        strbuf_add_absolute_path(&objdirbuf, get_object_directory());
 330        normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
 331
 332        alt_copy = xmemdupz(alt, len);
 333        string_list_split_in_place(&entries, alt_copy, sep, -1);
 334        for (i = 0; i < entries.nr; i++) {
 335                const char *entry = entries.items[i].string;
 336                if (entry[0] == '\0' || entry[0] == '#')
 337                        continue;
 338                if (!is_absolute_path(entry) && depth) {
 339                        error("%s: ignoring relative alternate object store %s",
 340                                        relative_base, entry);
 341                } else {
 342                        link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
 343                }
 344        }
 345        string_list_clear(&entries, 0);
 346        free(alt_copy);
 347        strbuf_release(&objdirbuf);
 348}
 349
 350void read_info_alternates(const char * relative_base, int depth)
 351{
 352        char *map;
 353        size_t mapsz;
 354        struct stat st;
 355        char *path;
 356        int fd;
 357
 358        path = xstrfmt("%s/info/alternates", relative_base);
 359        fd = git_open_noatime(path);
 360        free(path);
 361        if (fd < 0)
 362                return;
 363        if (fstat(fd, &st) || (st.st_size == 0)) {
 364                close(fd);
 365                return;
 366        }
 367        mapsz = xsize_t(st.st_size);
 368        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 369        close(fd);
 370
 371        link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
 372
 373        munmap(map, mapsz);
 374}
 375
 376void add_to_alternates_file(const char *reference)
 377{
 378        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
 379        char *alts = git_pathdup("objects/info/alternates");
 380        FILE *in, *out;
 381
 382        hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
 383        out = fdopen_lock_file(lock, "w");
 384        if (!out)
 385                die_errno("unable to fdopen alternates lockfile");
 386
 387        in = fopen(alts, "r");
 388        if (in) {
 389                struct strbuf line = STRBUF_INIT;
 390                int found = 0;
 391
 392                while (strbuf_getline(&line, in) != EOF) {
 393                        if (!strcmp(reference, line.buf)) {
 394                                found = 1;
 395                                break;
 396                        }
 397                        fprintf_or_die(out, "%s\n", line.buf);
 398                }
 399
 400                strbuf_release(&line);
 401                fclose(in);
 402
 403                if (found) {
 404                        rollback_lock_file(lock);
 405                        lock = NULL;
 406                }
 407        }
 408        else if (errno != ENOENT)
 409                die_errno("unable to read alternates file");
 410
 411        if (lock) {
 412                fprintf_or_die(out, "%s\n", reference);
 413                if (commit_lock_file(lock))
 414                        die_errno("unable to move new alternates file into place");
 415                if (alt_odb_tail)
 416                        link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
 417        }
 418        free(alts);
 419}
 420
 421int foreach_alt_odb(alt_odb_fn fn, void *cb)
 422{
 423        struct alternate_object_database *ent;
 424        int r = 0;
 425
 426        prepare_alt_odb();
 427        for (ent = alt_odb_list; ent; ent = ent->next) {
 428                r = fn(ent, cb);
 429                if (r)
 430                        break;
 431        }
 432        return r;
 433}
 434
 435void prepare_alt_odb(void)
 436{
 437        const char *alt;
 438
 439        if (alt_odb_tail)
 440                return;
 441
 442        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 443        if (!alt) alt = "";
 444
 445        alt_odb_tail = &alt_odb_list;
 446        link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
 447
 448        read_info_alternates(get_object_directory(), 0);
 449}
 450
 451/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
 452static int freshen_file(const char *fn)
 453{
 454        struct utimbuf t;
 455        t.actime = t.modtime = time(NULL);
 456        return !utime(fn, &t);
 457}
 458
 459/*
 460 * All of the check_and_freshen functions return 1 if the file exists and was
 461 * freshened (if freshening was requested), 0 otherwise. If they return
 462 * 0, you should not assume that it is safe to skip a write of the object (it
 463 * either does not exist on disk, or has a stale mtime and may be subject to
 464 * pruning).
 465 */
 466static int check_and_freshen_file(const char *fn, int freshen)
 467{
 468        if (access(fn, F_OK))
 469                return 0;
 470        if (freshen && !freshen_file(fn))
 471                return 0;
 472        return 1;
 473}
 474
 475static int check_and_freshen_local(const unsigned char *sha1, int freshen)
 476{
 477        return check_and_freshen_file(sha1_file_name(sha1), freshen);
 478}
 479
 480static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
 481{
 482        struct alternate_object_database *alt;
 483        prepare_alt_odb();
 484        for (alt = alt_odb_list; alt; alt = alt->next) {
 485                fill_sha1_path(alt->name, sha1);
 486                if (check_and_freshen_file(alt->base, freshen))
 487                        return 1;
 488        }
 489        return 0;
 490}
 491
 492static int check_and_freshen(const unsigned char *sha1, int freshen)
 493{
 494        return check_and_freshen_local(sha1, freshen) ||
 495               check_and_freshen_nonlocal(sha1, freshen);
 496}
 497
 498int has_loose_object_nonlocal(const unsigned char *sha1)
 499{
 500        return check_and_freshen_nonlocal(sha1, 0);
 501}
 502
 503static int has_loose_object(const unsigned char *sha1)
 504{
 505        return check_and_freshen(sha1, 0);
 506}
 507
 508static unsigned int pack_used_ctr;
 509static unsigned int pack_mmap_calls;
 510static unsigned int peak_pack_open_windows;
 511static unsigned int pack_open_windows;
 512static unsigned int pack_open_fds;
 513static unsigned int pack_max_fds;
 514static size_t peak_pack_mapped;
 515static size_t pack_mapped;
 516struct packed_git *packed_git;
 517
 518static struct mru packed_git_mru_storage;
 519struct mru *packed_git_mru = &packed_git_mru_storage;
 520
 521void pack_report(void)
 522{
 523        fprintf(stderr,
 524                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 525                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 526                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 527                sz_fmt(getpagesize()),
 528                sz_fmt(packed_git_window_size),
 529                sz_fmt(packed_git_limit));
 530        fprintf(stderr,
 531                "pack_report: pack_used_ctr            = %10u\n"
 532                "pack_report: pack_mmap_calls          = %10u\n"
 533                "pack_report: pack_open_windows        = %10u / %10u\n"
 534                "pack_report: pack_mapped              = "
 535                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 536                pack_used_ctr,
 537                pack_mmap_calls,
 538                pack_open_windows, peak_pack_open_windows,
 539                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
 540}
 541
 542/*
 543 * Open and mmap the index file at path, perform a couple of
 544 * consistency checks, then record its information to p.  Return 0 on
 545 * success.
 546 */
 547static int check_packed_git_idx(const char *path, struct packed_git *p)
 548{
 549        void *idx_map;
 550        struct pack_idx_header *hdr;
 551        size_t idx_size;
 552        uint32_t version, nr, i, *index;
 553        int fd = git_open_noatime(path);
 554        struct stat st;
 555
 556        if (fd < 0)
 557                return -1;
 558        if (fstat(fd, &st)) {
 559                close(fd);
 560                return -1;
 561        }
 562        idx_size = xsize_t(st.st_size);
 563        if (idx_size < 4 * 256 + 20 + 20) {
 564                close(fd);
 565                return error("index file %s is too small", path);
 566        }
 567        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 568        close(fd);
 569
 570        hdr = idx_map;
 571        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 572                version = ntohl(hdr->idx_version);
 573                if (version < 2 || version > 2) {
 574                        munmap(idx_map, idx_size);
 575                        return error("index file %s is version %"PRIu32
 576                                     " and is not supported by this binary"
 577                                     " (try upgrading GIT to a newer version)",
 578                                     path, version);
 579                }
 580        } else
 581                version = 1;
 582
 583        nr = 0;
 584        index = idx_map;
 585        if (version > 1)
 586                index += 2;  /* skip index header */
 587        for (i = 0; i < 256; i++) {
 588                uint32_t n = ntohl(index[i]);
 589                if (n < nr) {
 590                        munmap(idx_map, idx_size);
 591                        return error("non-monotonic index %s", path);
 592                }
 593                nr = n;
 594        }
 595
 596        if (version == 1) {
 597                /*
 598                 * Total size:
 599                 *  - 256 index entries 4 bytes each
 600                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 601                 *  - 20-byte SHA1 of the packfile
 602                 *  - 20-byte SHA1 file checksum
 603                 */
 604                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 605                        munmap(idx_map, idx_size);
 606                        return error("wrong index v1 file size in %s", path);
 607                }
 608        } else if (version == 2) {
 609                /*
 610                 * Minimum size:
 611                 *  - 8 bytes of header
 612                 *  - 256 index entries 4 bytes each
 613                 *  - 20-byte sha1 entry * nr
 614                 *  - 4-byte crc entry * nr
 615                 *  - 4-byte offset entry * nr
 616                 *  - 20-byte SHA1 of the packfile
 617                 *  - 20-byte SHA1 file checksum
 618                 * And after the 4-byte offset table might be a
 619                 * variable sized table containing 8-byte entries
 620                 * for offsets larger than 2^31.
 621                 */
 622                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 623                unsigned long max_size = min_size;
 624                if (nr)
 625                        max_size += (nr - 1)*8;
 626                if (idx_size < min_size || idx_size > max_size) {
 627                        munmap(idx_map, idx_size);
 628                        return error("wrong index v2 file size in %s", path);
 629                }
 630                if (idx_size != min_size &&
 631                    /*
 632                     * make sure we can deal with large pack offsets.
 633                     * 31-bit signed offset won't be enough, neither
 634                     * 32-bit unsigned one will be.
 635                     */
 636                    (sizeof(off_t) <= 4)) {
 637                        munmap(idx_map, idx_size);
 638                        return error("pack too large for current definition of off_t in %s", path);
 639                }
 640        }
 641
 642        p->index_version = version;
 643        p->index_data = idx_map;
 644        p->index_size = idx_size;
 645        p->num_objects = nr;
 646        return 0;
 647}
 648
 649int open_pack_index(struct packed_git *p)
 650{
 651        char *idx_name;
 652        size_t len;
 653        int ret;
 654
 655        if (p->index_data)
 656                return 0;
 657
 658        if (!strip_suffix(p->pack_name, ".pack", &len))
 659                die("BUG: pack_name does not end in .pack");
 660        idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
 661        ret = check_packed_git_idx(idx_name, p);
 662        free(idx_name);
 663        return ret;
 664}
 665
 666static void scan_windows(struct packed_git *p,
 667        struct packed_git **lru_p,
 668        struct pack_window **lru_w,
 669        struct pack_window **lru_l)
 670{
 671        struct pack_window *w, *w_l;
 672
 673        for (w_l = NULL, w = p->windows; w; w = w->next) {
 674                if (!w->inuse_cnt) {
 675                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 676                                *lru_p = p;
 677                                *lru_w = w;
 678                                *lru_l = w_l;
 679                        }
 680                }
 681                w_l = w;
 682        }
 683}
 684
 685static int unuse_one_window(struct packed_git *current)
 686{
 687        struct packed_git *p, *lru_p = NULL;
 688        struct pack_window *lru_w = NULL, *lru_l = NULL;
 689
 690        if (current)
 691                scan_windows(current, &lru_p, &lru_w, &lru_l);
 692        for (p = packed_git; p; p = p->next)
 693                scan_windows(p, &lru_p, &lru_w, &lru_l);
 694        if (lru_p) {
 695                munmap(lru_w->base, lru_w->len);
 696                pack_mapped -= lru_w->len;
 697                if (lru_l)
 698                        lru_l->next = lru_w->next;
 699                else
 700                        lru_p->windows = lru_w->next;
 701                free(lru_w);
 702                pack_open_windows--;
 703                return 1;
 704        }
 705        return 0;
 706}
 707
 708void release_pack_memory(size_t need)
 709{
 710        size_t cur = pack_mapped;
 711        while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
 712                ; /* nothing */
 713}
 714
 715static void mmap_limit_check(size_t length)
 716{
 717        static size_t limit = 0;
 718        if (!limit) {
 719                limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
 720                if (!limit)
 721                        limit = SIZE_MAX;
 722        }
 723        if (length > limit)
 724                die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
 725                    (uintmax_t)length, (uintmax_t)limit);
 726}
 727
 728void *xmmap_gently(void *start, size_t length,
 729                  int prot, int flags, int fd, off_t offset)
 730{
 731        void *ret;
 732
 733        mmap_limit_check(length);
 734        ret = mmap(start, length, prot, flags, fd, offset);
 735        if (ret == MAP_FAILED) {
 736                if (!length)
 737                        return NULL;
 738                release_pack_memory(length);
 739                ret = mmap(start, length, prot, flags, fd, offset);
 740        }
 741        return ret;
 742}
 743
 744void *xmmap(void *start, size_t length,
 745        int prot, int flags, int fd, off_t offset)
 746{
 747        void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
 748        if (ret == MAP_FAILED)
 749                die_errno("mmap failed");
 750        return ret;
 751}
 752
 753void close_pack_windows(struct packed_git *p)
 754{
 755        while (p->windows) {
 756                struct pack_window *w = p->windows;
 757
 758                if (w->inuse_cnt)
 759                        die("pack '%s' still has open windows to it",
 760                            p->pack_name);
 761                munmap(w->base, w->len);
 762                pack_mapped -= w->len;
 763                pack_open_windows--;
 764                p->windows = w->next;
 765                free(w);
 766        }
 767}
 768
 769static int close_pack_fd(struct packed_git *p)
 770{
 771        if (p->pack_fd < 0)
 772                return 0;
 773
 774        close(p->pack_fd);
 775        pack_open_fds--;
 776        p->pack_fd = -1;
 777
 778        return 1;
 779}
 780
 781static void close_pack(struct packed_git *p)
 782{
 783        close_pack_windows(p);
 784        close_pack_fd(p);
 785        close_pack_index(p);
 786}
 787
 788void close_all_packs(void)
 789{
 790        struct packed_git *p;
 791
 792        for (p = packed_git; p; p = p->next)
 793                if (p->do_not_close)
 794                        die("BUG: want to close pack marked 'do-not-close'");
 795                else
 796                        close_pack(p);
 797}
 798
 799
 800/*
 801 * The LRU pack is the one with the oldest MRU window, preferring packs
 802 * with no used windows, or the oldest mtime if it has no windows allocated.
 803 */
 804static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
 805{
 806        struct pack_window *w, *this_mru_w;
 807        int has_windows_inuse = 0;
 808
 809        /*
 810         * Reject this pack if it has windows and the previously selected
 811         * one does not.  If this pack does not have windows, reject
 812         * it if the pack file is newer than the previously selected one.
 813         */
 814        if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
 815                return;
 816
 817        for (w = this_mru_w = p->windows; w; w = w->next) {
 818                /*
 819                 * Reject this pack if any of its windows are in use,
 820                 * but the previously selected pack did not have any
 821                 * inuse windows.  Otherwise, record that this pack
 822                 * has windows in use.
 823                 */
 824                if (w->inuse_cnt) {
 825                        if (*accept_windows_inuse)
 826                                has_windows_inuse = 1;
 827                        else
 828                                return;
 829                }
 830
 831                if (w->last_used > this_mru_w->last_used)
 832                        this_mru_w = w;
 833
 834                /*
 835                 * Reject this pack if it has windows that have been
 836                 * used more recently than the previously selected pack.
 837                 * If the previously selected pack had windows inuse and
 838                 * we have not encountered a window in this pack that is
 839                 * inuse, skip this check since we prefer a pack with no
 840                 * inuse windows to one that has inuse windows.
 841                 */
 842                if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
 843                    this_mru_w->last_used > (*mru_w)->last_used)
 844                        return;
 845        }
 846
 847        /*
 848         * Select this pack.
 849         */
 850        *mru_w = this_mru_w;
 851        *lru_p = p;
 852        *accept_windows_inuse = has_windows_inuse;
 853}
 854
 855static int close_one_pack(void)
 856{
 857        struct packed_git *p, *lru_p = NULL;
 858        struct pack_window *mru_w = NULL;
 859        int accept_windows_inuse = 1;
 860
 861        for (p = packed_git; p; p = p->next) {
 862                if (p->pack_fd == -1)
 863                        continue;
 864                find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
 865        }
 866
 867        if (lru_p)
 868                return close_pack_fd(lru_p);
 869
 870        return 0;
 871}
 872
 873void unuse_pack(struct pack_window **w_cursor)
 874{
 875        struct pack_window *w = *w_cursor;
 876        if (w) {
 877                w->inuse_cnt--;
 878                *w_cursor = NULL;
 879        }
 880}
 881
 882void close_pack_index(struct packed_git *p)
 883{
 884        if (p->index_data) {
 885                munmap((void *)p->index_data, p->index_size);
 886                p->index_data = NULL;
 887        }
 888}
 889
 890static unsigned int get_max_fd_limit(void)
 891{
 892#ifdef RLIMIT_NOFILE
 893        {
 894                struct rlimit lim;
 895
 896                if (!getrlimit(RLIMIT_NOFILE, &lim))
 897                        return lim.rlim_cur;
 898        }
 899#endif
 900
 901#ifdef _SC_OPEN_MAX
 902        {
 903                long open_max = sysconf(_SC_OPEN_MAX);
 904                if (0 < open_max)
 905                        return open_max;
 906                /*
 907                 * Otherwise, we got -1 for one of the two
 908                 * reasons:
 909                 *
 910                 * (1) sysconf() did not understand _SC_OPEN_MAX
 911                 *     and signaled an error with -1; or
 912                 * (2) sysconf() said there is no limit.
 913                 *
 914                 * We _could_ clear errno before calling sysconf() to
 915                 * tell these two cases apart and return a huge number
 916                 * in the latter case to let the caller cap it to a
 917                 * value that is not so selfish, but letting the
 918                 * fallback OPEN_MAX codepath take care of these cases
 919                 * is a lot simpler.
 920                 */
 921        }
 922#endif
 923
 924#ifdef OPEN_MAX
 925        return OPEN_MAX;
 926#else
 927        return 1; /* see the caller ;-) */
 928#endif
 929}
 930
 931/*
 932 * Do not call this directly as this leaks p->pack_fd on error return;
 933 * call open_packed_git() instead.
 934 */
 935static int open_packed_git_1(struct packed_git *p)
 936{
 937        struct stat st;
 938        struct pack_header hdr;
 939        unsigned char sha1[20];
 940        unsigned char *idx_sha1;
 941        long fd_flag;
 942
 943        if (!p->index_data && open_pack_index(p))
 944                return error("packfile %s index unavailable", p->pack_name);
 945
 946        if (!pack_max_fds) {
 947                unsigned int max_fds = get_max_fd_limit();
 948
 949                /* Save 3 for stdin/stdout/stderr, 22 for work */
 950                if (25 < max_fds)
 951                        pack_max_fds = max_fds - 25;
 952                else
 953                        pack_max_fds = 1;
 954        }
 955
 956        while (pack_max_fds <= pack_open_fds && close_one_pack())
 957                ; /* nothing */
 958
 959        p->pack_fd = git_open_noatime(p->pack_name);
 960        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 961                return -1;
 962        pack_open_fds++;
 963
 964        /* If we created the struct before we had the pack we lack size. */
 965        if (!p->pack_size) {
 966                if (!S_ISREG(st.st_mode))
 967                        return error("packfile %s not a regular file", p->pack_name);
 968                p->pack_size = st.st_size;
 969        } else if (p->pack_size != st.st_size)
 970                return error("packfile %s size changed", p->pack_name);
 971
 972        /* We leave these file descriptors open with sliding mmap;
 973         * there is no point keeping them open across exec(), though.
 974         */
 975        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 976        if (fd_flag < 0)
 977                return error("cannot determine file descriptor flags");
 978        fd_flag |= FD_CLOEXEC;
 979        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 980                return error("cannot set FD_CLOEXEC");
 981
 982        /* Verify we recognize this pack file format. */
 983        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 984                return error("file %s is far too short to be a packfile", p->pack_name);
 985        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 986                return error("file %s is not a GIT packfile", p->pack_name);
 987        if (!pack_version_ok(hdr.hdr_version))
 988                return error("packfile %s is version %"PRIu32" and not"
 989                        " supported (try upgrading GIT to a newer version)",
 990                        p->pack_name, ntohl(hdr.hdr_version));
 991
 992        /* Verify the pack matches its index. */
 993        if (p->num_objects != ntohl(hdr.hdr_entries))
 994                return error("packfile %s claims to have %"PRIu32" objects"
 995                             " while index indicates %"PRIu32" objects",
 996                             p->pack_name, ntohl(hdr.hdr_entries),
 997                             p->num_objects);
 998        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 999                return error("end of packfile %s is unavailable", p->pack_name);
1000        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1001                return error("packfile %s signature is unavailable", p->pack_name);
1002        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1003        if (hashcmp(sha1, idx_sha1))
1004                return error("packfile %s does not match index", p->pack_name);
1005        return 0;
1006}
1007
1008static int open_packed_git(struct packed_git *p)
1009{
1010        if (!open_packed_git_1(p))
1011                return 0;
1012        close_pack_fd(p);
1013        return -1;
1014}
1015
1016static int in_window(struct pack_window *win, off_t offset)
1017{
1018        /* We must promise at least 20 bytes (one hash) after the
1019         * offset is available from this window, otherwise the offset
1020         * is not actually in this window and a different window (which
1021         * has that one hash excess) must be used.  This is to support
1022         * the object header and delta base parsing routines below.
1023         */
1024        off_t win_off = win->offset;
1025        return win_off <= offset
1026                && (offset + 20) <= (win_off + win->len);
1027}
1028
1029unsigned char *use_pack(struct packed_git *p,
1030                struct pack_window **w_cursor,
1031                off_t offset,
1032                unsigned long *left)
1033{
1034        struct pack_window *win = *w_cursor;
1035
1036        /* Since packfiles end in a hash of their content and it's
1037         * pointless to ask for an offset into the middle of that
1038         * hash, and the in_window function above wouldn't match
1039         * don't allow an offset too close to the end of the file.
1040         */
1041        if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1042                die("packfile %s cannot be accessed", p->pack_name);
1043        if (offset > (p->pack_size - 20))
1044                die("offset beyond end of packfile (truncated pack?)");
1045        if (offset < 0)
1046                die(_("offset before end of packfile (broken .idx?)"));
1047
1048        if (!win || !in_window(win, offset)) {
1049                if (win)
1050                        win->inuse_cnt--;
1051                for (win = p->windows; win; win = win->next) {
1052                        if (in_window(win, offset))
1053                                break;
1054                }
1055                if (!win) {
1056                        size_t window_align = packed_git_window_size / 2;
1057                        off_t len;
1058
1059                        if (p->pack_fd == -1 && open_packed_git(p))
1060                                die("packfile %s cannot be accessed", p->pack_name);
1061
1062                        win = xcalloc(1, sizeof(*win));
1063                        win->offset = (offset / window_align) * window_align;
1064                        len = p->pack_size - win->offset;
1065                        if (len > packed_git_window_size)
1066                                len = packed_git_window_size;
1067                        win->len = (size_t)len;
1068                        pack_mapped += win->len;
1069                        while (packed_git_limit < pack_mapped
1070                                && unuse_one_window(p))
1071                                ; /* nothing */
1072                        win->base = xmmap(NULL, win->len,
1073                                PROT_READ, MAP_PRIVATE,
1074                                p->pack_fd, win->offset);
1075                        if (win->base == MAP_FAILED)
1076                                die_errno("packfile %s cannot be mapped",
1077                                          p->pack_name);
1078                        if (!win->offset && win->len == p->pack_size
1079                                && !p->do_not_close)
1080                                close_pack_fd(p);
1081                        pack_mmap_calls++;
1082                        pack_open_windows++;
1083                        if (pack_mapped > peak_pack_mapped)
1084                                peak_pack_mapped = pack_mapped;
1085                        if (pack_open_windows > peak_pack_open_windows)
1086                                peak_pack_open_windows = pack_open_windows;
1087                        win->next = p->windows;
1088                        p->windows = win;
1089                }
1090        }
1091        if (win != *w_cursor) {
1092                win->last_used = pack_used_ctr++;
1093                win->inuse_cnt++;
1094                *w_cursor = win;
1095        }
1096        offset -= win->offset;
1097        if (left)
1098                *left = win->len - xsize_t(offset);
1099        return win->base + offset;
1100}
1101
1102static struct packed_git *alloc_packed_git(int extra)
1103{
1104        struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1105        memset(p, 0, sizeof(*p));
1106        p->pack_fd = -1;
1107        return p;
1108}
1109
1110static void try_to_free_pack_memory(size_t size)
1111{
1112        release_pack_memory(size);
1113}
1114
1115struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1116{
1117        static int have_set_try_to_free_routine;
1118        struct stat st;
1119        size_t alloc;
1120        struct packed_git *p;
1121
1122        if (!have_set_try_to_free_routine) {
1123                have_set_try_to_free_routine = 1;
1124                set_try_to_free_routine(try_to_free_pack_memory);
1125        }
1126
1127        /*
1128         * Make sure a corresponding .pack file exists and that
1129         * the index looks sane.
1130         */
1131        if (!strip_suffix_mem(path, &path_len, ".idx"))
1132                return NULL;
1133
1134        /*
1135         * ".pack" is long enough to hold any suffix we're adding (and
1136         * the use xsnprintf double-checks that)
1137         */
1138        alloc = st_add3(path_len, strlen(".pack"), 1);
1139        p = alloc_packed_git(alloc);
1140        memcpy(p->pack_name, path, path_len);
1141
1142        xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1143        if (!access(p->pack_name, F_OK))
1144                p->pack_keep = 1;
1145
1146        xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1147        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1148                free(p);
1149                return NULL;
1150        }
1151
1152        /* ok, it looks sane as far as we can check without
1153         * actually mapping the pack file.
1154         */
1155        p->pack_size = st.st_size;
1156        p->pack_local = local;
1157        p->mtime = st.st_mtime;
1158        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1159                hashclr(p->sha1);
1160        return p;
1161}
1162
1163struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1164{
1165        const char *path = sha1_pack_name(sha1);
1166        size_t alloc = st_add(strlen(path), 1);
1167        struct packed_git *p = alloc_packed_git(alloc);
1168
1169        memcpy(p->pack_name, path, alloc); /* includes NUL */
1170        hashcpy(p->sha1, sha1);
1171        if (check_packed_git_idx(idx_path, p)) {
1172                free(p);
1173                return NULL;
1174        }
1175
1176        return p;
1177}
1178
1179void install_packed_git(struct packed_git *pack)
1180{
1181        if (pack->pack_fd != -1)
1182                pack_open_fds++;
1183
1184        pack->next = packed_git;
1185        packed_git = pack;
1186}
1187
1188void (*report_garbage)(unsigned seen_bits, const char *path);
1189
1190static void report_helper(const struct string_list *list,
1191                          int seen_bits, int first, int last)
1192{
1193        if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1194                return;
1195
1196        for (; first < last; first++)
1197                report_garbage(seen_bits, list->items[first].string);
1198}
1199
1200static void report_pack_garbage(struct string_list *list)
1201{
1202        int i, baselen = -1, first = 0, seen_bits = 0;
1203
1204        if (!report_garbage)
1205                return;
1206
1207        string_list_sort(list);
1208
1209        for (i = 0; i < list->nr; i++) {
1210                const char *path = list->items[i].string;
1211                if (baselen != -1 &&
1212                    strncmp(path, list->items[first].string, baselen)) {
1213                        report_helper(list, seen_bits, first, i);
1214                        baselen = -1;
1215                        seen_bits = 0;
1216                }
1217                if (baselen == -1) {
1218                        const char *dot = strrchr(path, '.');
1219                        if (!dot) {
1220                                report_garbage(PACKDIR_FILE_GARBAGE, path);
1221                                continue;
1222                        }
1223                        baselen = dot - path + 1;
1224                        first = i;
1225                }
1226                if (!strcmp(path + baselen, "pack"))
1227                        seen_bits |= 1;
1228                else if (!strcmp(path + baselen, "idx"))
1229                        seen_bits |= 2;
1230        }
1231        report_helper(list, seen_bits, first, list->nr);
1232}
1233
1234static void prepare_packed_git_one(char *objdir, int local)
1235{
1236        struct strbuf path = STRBUF_INIT;
1237        size_t dirnamelen;
1238        DIR *dir;
1239        struct dirent *de;
1240        struct string_list garbage = STRING_LIST_INIT_DUP;
1241
1242        strbuf_addstr(&path, objdir);
1243        strbuf_addstr(&path, "/pack");
1244        dir = opendir(path.buf);
1245        if (!dir) {
1246                if (errno != ENOENT)
1247                        error_errno("unable to open object pack directory: %s",
1248                                    path.buf);
1249                strbuf_release(&path);
1250                return;
1251        }
1252        strbuf_addch(&path, '/');
1253        dirnamelen = path.len;
1254        while ((de = readdir(dir)) != NULL) {
1255                struct packed_git *p;
1256                size_t base_len;
1257
1258                if (is_dot_or_dotdot(de->d_name))
1259                        continue;
1260
1261                strbuf_setlen(&path, dirnamelen);
1262                strbuf_addstr(&path, de->d_name);
1263
1264                base_len = path.len;
1265                if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1266                        /* Don't reopen a pack we already have. */
1267                        for (p = packed_git; p; p = p->next) {
1268                                size_t len;
1269                                if (strip_suffix(p->pack_name, ".pack", &len) &&
1270                                    len == base_len &&
1271                                    !memcmp(p->pack_name, path.buf, len))
1272                                        break;
1273                        }
1274                        if (p == NULL &&
1275                            /*
1276                             * See if it really is a valid .idx file with
1277                             * corresponding .pack file that we can map.
1278                             */
1279                            (p = add_packed_git(path.buf, path.len, local)) != NULL)
1280                                install_packed_git(p);
1281                }
1282
1283                if (!report_garbage)
1284                        continue;
1285
1286                if (ends_with(de->d_name, ".idx") ||
1287                    ends_with(de->d_name, ".pack") ||
1288                    ends_with(de->d_name, ".bitmap") ||
1289                    ends_with(de->d_name, ".keep"))
1290                        string_list_append(&garbage, path.buf);
1291                else
1292                        report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1293        }
1294        closedir(dir);
1295        report_pack_garbage(&garbage);
1296        string_list_clear(&garbage, 0);
1297        strbuf_release(&path);
1298}
1299
1300static int sort_pack(const void *a_, const void *b_)
1301{
1302        struct packed_git *a = *((struct packed_git **)a_);
1303        struct packed_git *b = *((struct packed_git **)b_);
1304        int st;
1305
1306        /*
1307         * Local packs tend to contain objects specific to our
1308         * variant of the project than remote ones.  In addition,
1309         * remote ones could be on a network mounted filesystem.
1310         * Favor local ones for these reasons.
1311         */
1312        st = a->pack_local - b->pack_local;
1313        if (st)
1314                return -st;
1315
1316        /*
1317         * Younger packs tend to contain more recent objects,
1318         * and more recent objects tend to get accessed more
1319         * often.
1320         */
1321        if (a->mtime < b->mtime)
1322                return 1;
1323        else if (a->mtime == b->mtime)
1324                return 0;
1325        return -1;
1326}
1327
1328static void rearrange_packed_git(void)
1329{
1330        struct packed_git **ary, *p;
1331        int i, n;
1332
1333        for (n = 0, p = packed_git; p; p = p->next)
1334                n++;
1335        if (n < 2)
1336                return;
1337
1338        /* prepare an array of packed_git for easier sorting */
1339        ary = xcalloc(n, sizeof(struct packed_git *));
1340        for (n = 0, p = packed_git; p; p = p->next)
1341                ary[n++] = p;
1342
1343        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1344
1345        /* link them back again */
1346        for (i = 0; i < n - 1; i++)
1347                ary[i]->next = ary[i + 1];
1348        ary[n - 1]->next = NULL;
1349        packed_git = ary[0];
1350
1351        free(ary);
1352}
1353
1354static void prepare_packed_git_mru(void)
1355{
1356        struct packed_git *p;
1357
1358        mru_clear(packed_git_mru);
1359        for (p = packed_git; p; p = p->next)
1360                mru_append(packed_git_mru, p);
1361}
1362
1363static int prepare_packed_git_run_once = 0;
1364void prepare_packed_git(void)
1365{
1366        struct alternate_object_database *alt;
1367
1368        if (prepare_packed_git_run_once)
1369                return;
1370        prepare_packed_git_one(get_object_directory(), 1);
1371        prepare_alt_odb();
1372        for (alt = alt_odb_list; alt; alt = alt->next) {
1373                alt->name[-1] = 0;
1374                prepare_packed_git_one(alt->base, 0);
1375                alt->name[-1] = '/';
1376        }
1377        rearrange_packed_git();
1378        prepare_packed_git_mru();
1379        prepare_packed_git_run_once = 1;
1380}
1381
1382void reprepare_packed_git(void)
1383{
1384        prepare_packed_git_run_once = 0;
1385        prepare_packed_git();
1386}
1387
1388static void mark_bad_packed_object(struct packed_git *p,
1389                                   const unsigned char *sha1)
1390{
1391        unsigned i;
1392        for (i = 0; i < p->num_bad_objects; i++)
1393                if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1394                        return;
1395        p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1396                                      st_mult(GIT_SHA1_RAWSZ,
1397                                              st_add(p->num_bad_objects, 1)));
1398        hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1399        p->num_bad_objects++;
1400}
1401
1402static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1403{
1404        struct packed_git *p;
1405        unsigned i;
1406
1407        for (p = packed_git; p; p = p->next)
1408                for (i = 0; i < p->num_bad_objects; i++)
1409                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1410                                return p;
1411        return NULL;
1412}
1413
1414/*
1415 * With an in-core object data in "map", rehash it to make sure the
1416 * object name actually matches "sha1" to detect object corruption.
1417 * With "map" == NULL, try reading the object named with "sha1" using
1418 * the streaming interface and rehash it to do the same.
1419 */
1420int check_sha1_signature(const unsigned char *sha1, void *map,
1421                         unsigned long size, const char *type)
1422{
1423        unsigned char real_sha1[20];
1424        enum object_type obj_type;
1425        struct git_istream *st;
1426        git_SHA_CTX c;
1427        char hdr[32];
1428        int hdrlen;
1429
1430        if (map) {
1431                hash_sha1_file(map, size, type, real_sha1);
1432                return hashcmp(sha1, real_sha1) ? -1 : 0;
1433        }
1434
1435        st = open_istream(sha1, &obj_type, &size, NULL);
1436        if (!st)
1437                return -1;
1438
1439        /* Generate the header */
1440        hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1441
1442        /* Sha1.. */
1443        git_SHA1_Init(&c);
1444        git_SHA1_Update(&c, hdr, hdrlen);
1445        for (;;) {
1446                char buf[1024 * 16];
1447                ssize_t readlen = read_istream(st, buf, sizeof(buf));
1448
1449                if (readlen < 0) {
1450                        close_istream(st);
1451                        return -1;
1452                }
1453                if (!readlen)
1454                        break;
1455                git_SHA1_Update(&c, buf, readlen);
1456        }
1457        git_SHA1_Final(real_sha1, &c);
1458        close_istream(st);
1459        return hashcmp(sha1, real_sha1) ? -1 : 0;
1460}
1461
1462int git_open_noatime(const char *name)
1463{
1464        static int sha1_file_open_flag = O_NOATIME;
1465
1466        for (;;) {
1467                int fd;
1468
1469                errno = 0;
1470                fd = open(name, O_RDONLY | sha1_file_open_flag);
1471                if (fd >= 0)
1472                        return fd;
1473
1474                /* Might the failure be due to O_NOATIME? */
1475                if (errno != ENOENT && sha1_file_open_flag) {
1476                        sha1_file_open_flag = 0;
1477                        continue;
1478                }
1479
1480                return -1;
1481        }
1482}
1483
1484static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1485{
1486        struct alternate_object_database *alt;
1487
1488        if (!lstat(sha1_file_name(sha1), st))
1489                return 0;
1490
1491        prepare_alt_odb();
1492        errno = ENOENT;
1493        for (alt = alt_odb_list; alt; alt = alt->next) {
1494                fill_sha1_path(alt->name, sha1);
1495                if (!lstat(alt->base, st))
1496                        return 0;
1497        }
1498
1499        return -1;
1500}
1501
1502static int open_sha1_file(const unsigned char *sha1)
1503{
1504        int fd;
1505        struct alternate_object_database *alt;
1506        int most_interesting_errno;
1507
1508        fd = git_open_noatime(sha1_file_name(sha1));
1509        if (fd >= 0)
1510                return fd;
1511        most_interesting_errno = errno;
1512
1513        prepare_alt_odb();
1514        for (alt = alt_odb_list; alt; alt = alt->next) {
1515                fill_sha1_path(alt->name, sha1);
1516                fd = git_open_noatime(alt->base);
1517                if (fd >= 0)
1518                        return fd;
1519                if (most_interesting_errno == ENOENT)
1520                        most_interesting_errno = errno;
1521        }
1522        errno = most_interesting_errno;
1523        return -1;
1524}
1525
1526void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1527{
1528        void *map;
1529        int fd;
1530
1531        fd = open_sha1_file(sha1);
1532        map = NULL;
1533        if (fd >= 0) {
1534                struct stat st;
1535
1536                if (!fstat(fd, &st)) {
1537                        *size = xsize_t(st.st_size);
1538                        if (!*size) {
1539                                /* mmap() is forbidden on empty files */
1540                                error("object file %s is empty", sha1_file_name(sha1));
1541                                return NULL;
1542                        }
1543                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1544                }
1545                close(fd);
1546        }
1547        return map;
1548}
1549
1550unsigned long unpack_object_header_buffer(const unsigned char *buf,
1551                unsigned long len, enum object_type *type, unsigned long *sizep)
1552{
1553        unsigned shift;
1554        unsigned long size, c;
1555        unsigned long used = 0;
1556
1557        c = buf[used++];
1558        *type = (c >> 4) & 7;
1559        size = c & 15;
1560        shift = 4;
1561        while (c & 0x80) {
1562                if (len <= used || bitsizeof(long) <= shift) {
1563                        error("bad object header");
1564                        size = used = 0;
1565                        break;
1566                }
1567                c = buf[used++];
1568                size += (c & 0x7f) << shift;
1569                shift += 7;
1570        }
1571        *sizep = size;
1572        return used;
1573}
1574
1575static int unpack_sha1_short_header(git_zstream *stream,
1576                                    unsigned char *map, unsigned long mapsize,
1577                                    void *buffer, unsigned long bufsiz)
1578{
1579        /* Get the data stream */
1580        memset(stream, 0, sizeof(*stream));
1581        stream->next_in = map;
1582        stream->avail_in = mapsize;
1583        stream->next_out = buffer;
1584        stream->avail_out = bufsiz;
1585
1586        git_inflate_init(stream);
1587        return git_inflate(stream, 0);
1588}
1589
1590int unpack_sha1_header(git_zstream *stream,
1591                       unsigned char *map, unsigned long mapsize,
1592                       void *buffer, unsigned long bufsiz)
1593{
1594        int status = unpack_sha1_short_header(stream, map, mapsize,
1595                                              buffer, bufsiz);
1596
1597        if (status < Z_OK)
1598                return status;
1599
1600        /* Make sure we have the terminating NUL */
1601        if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1602                return -1;
1603        return 0;
1604}
1605
1606static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1607                                        unsigned long mapsize, void *buffer,
1608                                        unsigned long bufsiz, struct strbuf *header)
1609{
1610        int status;
1611
1612        status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1613        if (status < Z_OK)
1614                return -1;
1615
1616        /*
1617         * Check if entire header is unpacked in the first iteration.
1618         */
1619        if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1620                return 0;
1621
1622        /*
1623         * buffer[0..bufsiz] was not large enough.  Copy the partial
1624         * result out to header, and then append the result of further
1625         * reading the stream.
1626         */
1627        strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1628        stream->next_out = buffer;
1629        stream->avail_out = bufsiz;
1630
1631        do {
1632                status = git_inflate(stream, 0);
1633                strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1634                if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1635                        return 0;
1636                stream->next_out = buffer;
1637                stream->avail_out = bufsiz;
1638        } while (status != Z_STREAM_END);
1639        return -1;
1640}
1641
1642static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1643{
1644        int bytes = strlen(buffer) + 1;
1645        unsigned char *buf = xmallocz(size);
1646        unsigned long n;
1647        int status = Z_OK;
1648
1649        n = stream->total_out - bytes;
1650        if (n > size)
1651                n = size;
1652        memcpy(buf, (char *) buffer + bytes, n);
1653        bytes = n;
1654        if (bytes <= size) {
1655                /*
1656                 * The above condition must be (bytes <= size), not
1657                 * (bytes < size).  In other words, even though we
1658                 * expect no more output and set avail_out to zero,
1659                 * the input zlib stream may have bytes that express
1660                 * "this concludes the stream", and we *do* want to
1661                 * eat that input.
1662                 *
1663                 * Otherwise we would not be able to test that we
1664                 * consumed all the input to reach the expected size;
1665                 * we also want to check that zlib tells us that all
1666                 * went well with status == Z_STREAM_END at the end.
1667                 */
1668                stream->next_out = buf + bytes;
1669                stream->avail_out = size - bytes;
1670                while (status == Z_OK)
1671                        status = git_inflate(stream, Z_FINISH);
1672        }
1673        if (status == Z_STREAM_END && !stream->avail_in) {
1674                git_inflate_end(stream);
1675                return buf;
1676        }
1677
1678        if (status < 0)
1679                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1680        else if (stream->avail_in)
1681                error("garbage at end of loose object '%s'",
1682                      sha1_to_hex(sha1));
1683        free(buf);
1684        return NULL;
1685}
1686
1687/*
1688 * We used to just use "sscanf()", but that's actually way
1689 * too permissive for what we want to check. So do an anal
1690 * object header parse by hand.
1691 */
1692static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1693                               unsigned int flags)
1694{
1695        const char *type_buf = hdr;
1696        unsigned long size;
1697        int type, type_len = 0;
1698
1699        /*
1700         * The type can be of any size but is followed by
1701         * a space.
1702         */
1703        for (;;) {
1704                char c = *hdr++;
1705                if (!c)
1706                        return -1;
1707                if (c == ' ')
1708                        break;
1709                type_len++;
1710        }
1711
1712        type = type_from_string_gently(type_buf, type_len, 1);
1713        if (oi->typename)
1714                strbuf_add(oi->typename, type_buf, type_len);
1715        /*
1716         * Set type to 0 if its an unknown object and
1717         * we're obtaining the type using '--allow-unknown-type'
1718         * option.
1719         */
1720        if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1721                type = 0;
1722        else if (type < 0)
1723                die("invalid object type");
1724        if (oi->typep)
1725                *oi->typep = type;
1726
1727        /*
1728         * The length must follow immediately, and be in canonical
1729         * decimal format (ie "010" is not valid).
1730         */
1731        size = *hdr++ - '0';
1732        if (size > 9)
1733                return -1;
1734        if (size) {
1735                for (;;) {
1736                        unsigned long c = *hdr - '0';
1737                        if (c > 9)
1738                                break;
1739                        hdr++;
1740                        size = size * 10 + c;
1741                }
1742        }
1743
1744        if (oi->sizep)
1745                *oi->sizep = size;
1746
1747        /*
1748         * The length must be followed by a zero byte
1749         */
1750        return *hdr ? -1 : type;
1751}
1752
1753int parse_sha1_header(const char *hdr, unsigned long *sizep)
1754{
1755        struct object_info oi;
1756
1757        oi.sizep = sizep;
1758        oi.typename = NULL;
1759        oi.typep = NULL;
1760        return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
1761}
1762
1763static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1764{
1765        int ret;
1766        git_zstream stream;
1767        char hdr[8192];
1768
1769        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1770        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1771                return NULL;
1772
1773        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1774}
1775
1776unsigned long get_size_from_delta(struct packed_git *p,
1777                                  struct pack_window **w_curs,
1778                                  off_t curpos)
1779{
1780        const unsigned char *data;
1781        unsigned char delta_head[20], *in;
1782        git_zstream stream;
1783        int st;
1784
1785        memset(&stream, 0, sizeof(stream));
1786        stream.next_out = delta_head;
1787        stream.avail_out = sizeof(delta_head);
1788
1789        git_inflate_init(&stream);
1790        do {
1791                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1792                stream.next_in = in;
1793                st = git_inflate(&stream, Z_FINISH);
1794                curpos += stream.next_in - in;
1795        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1796                 stream.total_out < sizeof(delta_head));
1797        git_inflate_end(&stream);
1798        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1799                error("delta data unpack-initial failed");
1800                return 0;
1801        }
1802
1803        /* Examine the initial part of the delta to figure out
1804         * the result size.
1805         */
1806        data = delta_head;
1807
1808        /* ignore base size */
1809        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1810
1811        /* Read the result size */
1812        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1813}
1814
1815static off_t get_delta_base(struct packed_git *p,
1816                                    struct pack_window **w_curs,
1817                                    off_t *curpos,
1818                                    enum object_type type,
1819                                    off_t delta_obj_offset)
1820{
1821        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1822        off_t base_offset;
1823
1824        /* use_pack() assured us we have [base_info, base_info + 20)
1825         * as a range that we can look at without walking off the
1826         * end of the mapped window.  Its actually the hash size
1827         * that is assured.  An OFS_DELTA longer than the hash size
1828         * is stupid, as then a REF_DELTA would be smaller to store.
1829         */
1830        if (type == OBJ_OFS_DELTA) {
1831                unsigned used = 0;
1832                unsigned char c = base_info[used++];
1833                base_offset = c & 127;
1834                while (c & 128) {
1835                        base_offset += 1;
1836                        if (!base_offset || MSB(base_offset, 7))
1837                                return 0;  /* overflow */
1838                        c = base_info[used++];
1839                        base_offset = (base_offset << 7) + (c & 127);
1840                }
1841                base_offset = delta_obj_offset - base_offset;
1842                if (base_offset <= 0 || base_offset >= delta_obj_offset)
1843                        return 0;  /* out of bound */
1844                *curpos += used;
1845        } else if (type == OBJ_REF_DELTA) {
1846                /* The base entry _must_ be in the same pack */
1847                base_offset = find_pack_entry_one(base_info, p);
1848                *curpos += 20;
1849        } else
1850                die("I am totally screwed");
1851        return base_offset;
1852}
1853
1854/*
1855 * Like get_delta_base above, but we return the sha1 instead of the pack
1856 * offset. This means it is cheaper for REF deltas (we do not have to do
1857 * the final object lookup), but more expensive for OFS deltas (we
1858 * have to load the revidx to convert the offset back into a sha1).
1859 */
1860static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1861                                                struct pack_window **w_curs,
1862                                                off_t curpos,
1863                                                enum object_type type,
1864                                                off_t delta_obj_offset)
1865{
1866        if (type == OBJ_REF_DELTA) {
1867                unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1868                return base;
1869        } else if (type == OBJ_OFS_DELTA) {
1870                struct revindex_entry *revidx;
1871                off_t base_offset = get_delta_base(p, w_curs, &curpos,
1872                                                   type, delta_obj_offset);
1873
1874                if (!base_offset)
1875                        return NULL;
1876
1877                revidx = find_pack_revindex(p, base_offset);
1878                if (!revidx)
1879                        return NULL;
1880
1881                return nth_packed_object_sha1(p, revidx->nr);
1882        } else
1883                return NULL;
1884}
1885
1886int unpack_object_header(struct packed_git *p,
1887                         struct pack_window **w_curs,
1888                         off_t *curpos,
1889                         unsigned long *sizep)
1890{
1891        unsigned char *base;
1892        unsigned long left;
1893        unsigned long used;
1894        enum object_type type;
1895
1896        /* use_pack() assures us we have [base, base + 20) available
1897         * as a range that we can look at.  (Its actually the hash
1898         * size that is assured.)  With our object header encoding
1899         * the maximum deflated object size is 2^137, which is just
1900         * insane, so we know won't exceed what we have been given.
1901         */
1902        base = use_pack(p, w_curs, *curpos, &left);
1903        used = unpack_object_header_buffer(base, left, &type, sizep);
1904        if (!used) {
1905                type = OBJ_BAD;
1906        } else
1907                *curpos += used;
1908
1909        return type;
1910}
1911
1912static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1913{
1914        int type;
1915        struct revindex_entry *revidx;
1916        const unsigned char *sha1;
1917        revidx = find_pack_revindex(p, obj_offset);
1918        if (!revidx)
1919                return OBJ_BAD;
1920        sha1 = nth_packed_object_sha1(p, revidx->nr);
1921        mark_bad_packed_object(p, sha1);
1922        type = sha1_object_info(sha1, NULL);
1923        if (type <= OBJ_NONE)
1924                return OBJ_BAD;
1925        return type;
1926}
1927
1928#define POI_STACK_PREALLOC 64
1929
1930static enum object_type packed_to_object_type(struct packed_git *p,
1931                                              off_t obj_offset,
1932                                              enum object_type type,
1933                                              struct pack_window **w_curs,
1934                                              off_t curpos)
1935{
1936        off_t small_poi_stack[POI_STACK_PREALLOC];
1937        off_t *poi_stack = small_poi_stack;
1938        int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1939
1940        while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1941                off_t base_offset;
1942                unsigned long size;
1943                /* Push the object we're going to leave behind */
1944                if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1945                        poi_stack_alloc = alloc_nr(poi_stack_nr);
1946                        ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1947                        memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1948                } else {
1949                        ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1950                }
1951                poi_stack[poi_stack_nr++] = obj_offset;
1952                /* If parsing the base offset fails, just unwind */
1953                base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1954                if (!base_offset)
1955                        goto unwind;
1956                curpos = obj_offset = base_offset;
1957                type = unpack_object_header(p, w_curs, &curpos, &size);
1958                if (type <= OBJ_NONE) {
1959                        /* If getting the base itself fails, we first
1960                         * retry the base, otherwise unwind */
1961                        type = retry_bad_packed_offset(p, base_offset);
1962                        if (type > OBJ_NONE)
1963                                goto out;
1964                        goto unwind;
1965                }
1966        }
1967
1968        switch (type) {
1969        case OBJ_BAD:
1970        case OBJ_COMMIT:
1971        case OBJ_TREE:
1972        case OBJ_BLOB:
1973        case OBJ_TAG:
1974                break;
1975        default:
1976                error("unknown object type %i at offset %"PRIuMAX" in %s",
1977                      type, (uintmax_t)obj_offset, p->pack_name);
1978                type = OBJ_BAD;
1979        }
1980
1981out:
1982        if (poi_stack != small_poi_stack)
1983                free(poi_stack);
1984        return type;
1985
1986unwind:
1987        while (poi_stack_nr) {
1988                obj_offset = poi_stack[--poi_stack_nr];
1989                type = retry_bad_packed_offset(p, obj_offset);
1990                if (type > OBJ_NONE)
1991                        goto out;
1992        }
1993        type = OBJ_BAD;
1994        goto out;
1995}
1996
1997static int packed_object_info(struct packed_git *p, off_t obj_offset,
1998                              struct object_info *oi)
1999{
2000        struct pack_window *w_curs = NULL;
2001        unsigned long size;
2002        off_t curpos = obj_offset;
2003        enum object_type type;
2004
2005        /*
2006         * We always get the representation type, but only convert it to
2007         * a "real" type later if the caller is interested.
2008         */
2009        type = unpack_object_header(p, &w_curs, &curpos, &size);
2010
2011        if (oi->sizep) {
2012                if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2013                        off_t tmp_pos = curpos;
2014                        off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2015                                                           type, obj_offset);
2016                        if (!base_offset) {
2017                                type = OBJ_BAD;
2018                                goto out;
2019                        }
2020                        *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2021                        if (*oi->sizep == 0) {
2022                                type = OBJ_BAD;
2023                                goto out;
2024                        }
2025                } else {
2026                        *oi->sizep = size;
2027                }
2028        }
2029
2030        if (oi->disk_sizep) {
2031                struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2032                *oi->disk_sizep = revidx[1].offset - obj_offset;
2033        }
2034
2035        if (oi->typep) {
2036                *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2037                if (*oi->typep < 0) {
2038                        type = OBJ_BAD;
2039                        goto out;
2040                }
2041        }
2042
2043        if (oi->delta_base_sha1) {
2044                if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2045                        const unsigned char *base;
2046
2047                        base = get_delta_base_sha1(p, &w_curs, curpos,
2048                                                   type, obj_offset);
2049                        if (!base) {
2050                                type = OBJ_BAD;
2051                                goto out;
2052                        }
2053
2054                        hashcpy(oi->delta_base_sha1, base);
2055                } else
2056                        hashclr(oi->delta_base_sha1);
2057        }
2058
2059out:
2060        unuse_pack(&w_curs);
2061        return type;
2062}
2063
2064static void *unpack_compressed_entry(struct packed_git *p,
2065                                    struct pack_window **w_curs,
2066                                    off_t curpos,
2067                                    unsigned long size)
2068{
2069        int st;
2070        git_zstream stream;
2071        unsigned char *buffer, *in;
2072
2073        buffer = xmallocz_gently(size);
2074        if (!buffer)
2075                return NULL;
2076        memset(&stream, 0, sizeof(stream));
2077        stream.next_out = buffer;
2078        stream.avail_out = size + 1;
2079
2080        git_inflate_init(&stream);
2081        do {
2082                in = use_pack(p, w_curs, curpos, &stream.avail_in);
2083                stream.next_in = in;
2084                st = git_inflate(&stream, Z_FINISH);
2085                if (!stream.avail_out)
2086                        break; /* the payload is larger than it should be */
2087                curpos += stream.next_in - in;
2088        } while (st == Z_OK || st == Z_BUF_ERROR);
2089        git_inflate_end(&stream);
2090        if ((st != Z_STREAM_END) || stream.total_out != size) {
2091                free(buffer);
2092                return NULL;
2093        }
2094
2095        return buffer;
2096}
2097
2098#define MAX_DELTA_CACHE (256)
2099
2100static size_t delta_base_cached;
2101
2102static struct delta_base_cache_lru_list {
2103        struct delta_base_cache_lru_list *prev;
2104        struct delta_base_cache_lru_list *next;
2105} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2106
2107static struct delta_base_cache_entry {
2108        struct delta_base_cache_lru_list lru;
2109        void *data;
2110        struct packed_git *p;
2111        off_t base_offset;
2112        unsigned long size;
2113        enum object_type type;
2114} delta_base_cache[MAX_DELTA_CACHE];
2115
2116static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2117{
2118        unsigned long hash;
2119
2120        hash = (unsigned long)(intptr_t)p + (unsigned long)base_offset;
2121        hash += (hash >> 8) + (hash >> 16);
2122        return hash % MAX_DELTA_CACHE;
2123}
2124
2125static struct delta_base_cache_entry *
2126get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2127{
2128        unsigned long hash = pack_entry_hash(p, base_offset);
2129        return delta_base_cache + hash;
2130}
2131
2132static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2133                                     struct packed_git *p, off_t base_offset)
2134{
2135        return (ent->data && ent->p == p && ent->base_offset == base_offset);
2136}
2137
2138static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2139{
2140        struct delta_base_cache_entry *ent;
2141        ent = get_delta_base_cache_entry(p, base_offset);
2142        return eq_delta_base_cache_entry(ent, p, base_offset);
2143}
2144
2145static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2146{
2147        ent->data = NULL;
2148        ent->lru.next->prev = ent->lru.prev;
2149        ent->lru.prev->next = ent->lru.next;
2150        delta_base_cached -= ent->size;
2151}
2152
2153static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2154        unsigned long *base_size, enum object_type *type, int keep_cache)
2155{
2156        struct delta_base_cache_entry *ent;
2157        void *ret;
2158
2159        ent = get_delta_base_cache_entry(p, base_offset);
2160
2161        if (!eq_delta_base_cache_entry(ent, p, base_offset))
2162                return unpack_entry(p, base_offset, type, base_size);
2163
2164        ret = ent->data;
2165
2166        if (!keep_cache)
2167                clear_delta_base_cache_entry(ent);
2168        else
2169                ret = xmemdupz(ent->data, ent->size);
2170        *type = ent->type;
2171        *base_size = ent->size;
2172        return ret;
2173}
2174
2175static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2176{
2177        if (ent->data) {
2178                free(ent->data);
2179                ent->data = NULL;
2180                ent->lru.next->prev = ent->lru.prev;
2181                ent->lru.prev->next = ent->lru.next;
2182                delta_base_cached -= ent->size;
2183        }
2184}
2185
2186void clear_delta_base_cache(void)
2187{
2188        unsigned long p;
2189        for (p = 0; p < MAX_DELTA_CACHE; p++)
2190                release_delta_base_cache(&delta_base_cache[p]);
2191}
2192
2193static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2194        void *base, unsigned long base_size, enum object_type type)
2195{
2196        unsigned long hash = pack_entry_hash(p, base_offset);
2197        struct delta_base_cache_entry *ent = delta_base_cache + hash;
2198        struct delta_base_cache_lru_list *lru;
2199
2200        release_delta_base_cache(ent);
2201        delta_base_cached += base_size;
2202
2203        for (lru = delta_base_cache_lru.next;
2204             delta_base_cached > delta_base_cache_limit
2205             && lru != &delta_base_cache_lru;
2206             lru = lru->next) {
2207                struct delta_base_cache_entry *f = (void *)lru;
2208                if (f->type == OBJ_BLOB)
2209                        release_delta_base_cache(f);
2210        }
2211        for (lru = delta_base_cache_lru.next;
2212             delta_base_cached > delta_base_cache_limit
2213             && lru != &delta_base_cache_lru;
2214             lru = lru->next) {
2215                struct delta_base_cache_entry *f = (void *)lru;
2216                release_delta_base_cache(f);
2217        }
2218
2219        ent->p = p;
2220        ent->base_offset = base_offset;
2221        ent->type = type;
2222        ent->data = base;
2223        ent->size = base_size;
2224        ent->lru.next = &delta_base_cache_lru;
2225        ent->lru.prev = delta_base_cache_lru.prev;
2226        delta_base_cache_lru.prev->next = &ent->lru;
2227        delta_base_cache_lru.prev = &ent->lru;
2228}
2229
2230static void *read_object(const unsigned char *sha1, enum object_type *type,
2231                         unsigned long *size);
2232
2233static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2234{
2235        static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2236        trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2237                         p->pack_name, (uintmax_t)obj_offset);
2238}
2239
2240int do_check_packed_object_crc;
2241
2242#define UNPACK_ENTRY_STACK_PREALLOC 64
2243struct unpack_entry_stack_ent {
2244        off_t obj_offset;
2245        off_t curpos;
2246        unsigned long size;
2247};
2248
2249void *unpack_entry(struct packed_git *p, off_t obj_offset,
2250                   enum object_type *final_type, unsigned long *final_size)
2251{
2252        struct pack_window *w_curs = NULL;
2253        off_t curpos = obj_offset;
2254        void *data = NULL;
2255        unsigned long size;
2256        enum object_type type;
2257        struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2258        struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2259        int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2260        int base_from_cache = 0;
2261
2262        write_pack_access_log(p, obj_offset);
2263
2264        /* PHASE 1: drill down to the innermost base object */
2265        for (;;) {
2266                off_t base_offset;
2267                int i;
2268                struct delta_base_cache_entry *ent;
2269
2270                ent = get_delta_base_cache_entry(p, curpos);
2271                if (eq_delta_base_cache_entry(ent, p, curpos)) {
2272                        type = ent->type;
2273                        data = ent->data;
2274                        size = ent->size;
2275                        clear_delta_base_cache_entry(ent);
2276                        base_from_cache = 1;
2277                        break;
2278                }
2279
2280                if (do_check_packed_object_crc && p->index_version > 1) {
2281                        struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2282                        off_t len = revidx[1].offset - obj_offset;
2283                        if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2284                                const unsigned char *sha1 =
2285                                        nth_packed_object_sha1(p, revidx->nr);
2286                                error("bad packed object CRC for %s",
2287                                      sha1_to_hex(sha1));
2288                                mark_bad_packed_object(p, sha1);
2289                                unuse_pack(&w_curs);
2290                                return NULL;
2291                        }
2292                }
2293
2294                type = unpack_object_header(p, &w_curs, &curpos, &size);
2295                if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2296                        break;
2297
2298                base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2299                if (!base_offset) {
2300                        error("failed to validate delta base reference "
2301                              "at offset %"PRIuMAX" from %s",
2302                              (uintmax_t)curpos, p->pack_name);
2303                        /* bail to phase 2, in hopes of recovery */
2304                        data = NULL;
2305                        break;
2306                }
2307
2308                /* push object, proceed to base */
2309                if (delta_stack_nr >= delta_stack_alloc
2310                    && delta_stack == small_delta_stack) {
2311                        delta_stack_alloc = alloc_nr(delta_stack_nr);
2312                        ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2313                        memcpy(delta_stack, small_delta_stack,
2314                               sizeof(*delta_stack)*delta_stack_nr);
2315                } else {
2316                        ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2317                }
2318                i = delta_stack_nr++;
2319                delta_stack[i].obj_offset = obj_offset;
2320                delta_stack[i].curpos = curpos;
2321                delta_stack[i].size = size;
2322
2323                curpos = obj_offset = base_offset;
2324        }
2325
2326        /* PHASE 2: handle the base */
2327        switch (type) {
2328        case OBJ_OFS_DELTA:
2329        case OBJ_REF_DELTA:
2330                if (data)
2331                        die("BUG: unpack_entry: left loop at a valid delta");
2332                break;
2333        case OBJ_COMMIT:
2334        case OBJ_TREE:
2335        case OBJ_BLOB:
2336        case OBJ_TAG:
2337                if (!base_from_cache)
2338                        data = unpack_compressed_entry(p, &w_curs, curpos, size);
2339                break;
2340        default:
2341                data = NULL;
2342                error("unknown object type %i at offset %"PRIuMAX" in %s",
2343                      type, (uintmax_t)obj_offset, p->pack_name);
2344        }
2345
2346        /* PHASE 3: apply deltas in order */
2347
2348        /* invariants:
2349         *   'data' holds the base data, or NULL if there was corruption
2350         */
2351        while (delta_stack_nr) {
2352                void *delta_data;
2353                void *base = data;
2354                unsigned long delta_size, base_size = size;
2355                int i;
2356
2357                data = NULL;
2358
2359                if (base)
2360                        add_delta_base_cache(p, obj_offset, base, base_size, type);
2361
2362                if (!base) {
2363                        /*
2364                         * We're probably in deep shit, but let's try to fetch
2365                         * the required base anyway from another pack or loose.
2366                         * This is costly but should happen only in the presence
2367                         * of a corrupted pack, and is better than failing outright.
2368                         */
2369                        struct revindex_entry *revidx;
2370                        const unsigned char *base_sha1;
2371                        revidx = find_pack_revindex(p, obj_offset);
2372                        if (revidx) {
2373                                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2374                                error("failed to read delta base object %s"
2375                                      " at offset %"PRIuMAX" from %s",
2376                                      sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2377                                      p->pack_name);
2378                                mark_bad_packed_object(p, base_sha1);
2379                                base = read_object(base_sha1, &type, &base_size);
2380                        }
2381                }
2382
2383                i = --delta_stack_nr;
2384                obj_offset = delta_stack[i].obj_offset;
2385                curpos = delta_stack[i].curpos;
2386                delta_size = delta_stack[i].size;
2387
2388                if (!base)
2389                        continue;
2390
2391                delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2392
2393                if (!delta_data) {
2394                        error("failed to unpack compressed delta "
2395                              "at offset %"PRIuMAX" from %s",
2396                              (uintmax_t)curpos, p->pack_name);
2397                        data = NULL;
2398                        continue;
2399                }
2400
2401                data = patch_delta(base, base_size,
2402                                   delta_data, delta_size,
2403                                   &size);
2404
2405                /*
2406                 * We could not apply the delta; warn the user, but keep going.
2407                 * Our failure will be noticed either in the next iteration of
2408                 * the loop, or if this is the final delta, in the caller when
2409                 * we return NULL. Those code paths will take care of making
2410                 * a more explicit warning and retrying with another copy of
2411                 * the object.
2412                 */
2413                if (!data)
2414                        error("failed to apply delta");
2415
2416                free(delta_data);
2417        }
2418
2419        *final_type = type;
2420        *final_size = size;
2421
2422        unuse_pack(&w_curs);
2423
2424        if (delta_stack != small_delta_stack)
2425                free(delta_stack);
2426
2427        return data;
2428}
2429
2430const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2431                                            uint32_t n)
2432{
2433        const unsigned char *index = p->index_data;
2434        if (!index) {
2435                if (open_pack_index(p))
2436                        return NULL;
2437                index = p->index_data;
2438        }
2439        if (n >= p->num_objects)
2440                return NULL;
2441        index += 4 * 256;
2442        if (p->index_version == 1) {
2443                return index + 24 * n + 4;
2444        } else {
2445                index += 8;
2446                return index + 20 * n;
2447        }
2448}
2449
2450void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2451{
2452        const unsigned char *ptr = vptr;
2453        const unsigned char *start = p->index_data;
2454        const unsigned char *end = start + p->index_size;
2455        if (ptr < start)
2456                die(_("offset before start of pack index for %s (corrupt index?)"),
2457                    p->pack_name);
2458        /* No need to check for underflow; .idx files must be at least 8 bytes */
2459        if (ptr >= end - 8)
2460                die(_("offset beyond end of pack index for %s (truncated index?)"),
2461                    p->pack_name);
2462}
2463
2464off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2465{
2466        const unsigned char *index = p->index_data;
2467        index += 4 * 256;
2468        if (p->index_version == 1) {
2469                return ntohl(*((uint32_t *)(index + 24 * n)));
2470        } else {
2471                uint32_t off;
2472                index += 8 + p->num_objects * (20 + 4);
2473                off = ntohl(*((uint32_t *)(index + 4 * n)));
2474                if (!(off & 0x80000000))
2475                        return off;
2476                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2477                check_pack_index_ptr(p, index);
2478                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2479                                   ntohl(*((uint32_t *)(index + 4)));
2480        }
2481}
2482
2483off_t find_pack_entry_one(const unsigned char *sha1,
2484                                  struct packed_git *p)
2485{
2486        const uint32_t *level1_ofs = p->index_data;
2487        const unsigned char *index = p->index_data;
2488        unsigned hi, lo, stride;
2489        static int use_lookup = -1;
2490        static int debug_lookup = -1;
2491
2492        if (debug_lookup < 0)
2493                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2494
2495        if (!index) {
2496                if (open_pack_index(p))
2497                        return 0;
2498                level1_ofs = p->index_data;
2499                index = p->index_data;
2500        }
2501        if (p->index_version > 1) {
2502                level1_ofs += 2;
2503                index += 8;
2504        }
2505        index += 4 * 256;
2506        hi = ntohl(level1_ofs[*sha1]);
2507        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2508        if (p->index_version > 1) {
2509                stride = 20;
2510        } else {
2511                stride = 24;
2512                index += 4;
2513        }
2514
2515        if (debug_lookup)
2516                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2517                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2518
2519        if (use_lookup < 0)
2520                use_lookup = !!getenv("GIT_USE_LOOKUP");
2521        if (use_lookup) {
2522                int pos = sha1_entry_pos(index, stride, 0,
2523                                         lo, hi, p->num_objects, sha1);
2524                if (pos < 0)
2525                        return 0;
2526                return nth_packed_object_offset(p, pos);
2527        }
2528
2529        do {
2530                unsigned mi = (lo + hi) / 2;
2531                int cmp = hashcmp(index + mi * stride, sha1);
2532
2533                if (debug_lookup)
2534                        printf("lo %u hi %u rg %u mi %u\n",
2535                               lo, hi, hi - lo, mi);
2536                if (!cmp)
2537                        return nth_packed_object_offset(p, mi);
2538                if (cmp > 0)
2539                        hi = mi;
2540                else
2541                        lo = mi+1;
2542        } while (lo < hi);
2543        return 0;
2544}
2545
2546int is_pack_valid(struct packed_git *p)
2547{
2548        /* An already open pack is known to be valid. */
2549        if (p->pack_fd != -1)
2550                return 1;
2551
2552        /* If the pack has one window completely covering the
2553         * file size, the pack is known to be valid even if
2554         * the descriptor is not currently open.
2555         */
2556        if (p->windows) {
2557                struct pack_window *w = p->windows;
2558
2559                if (!w->offset && w->len == p->pack_size)
2560                        return 1;
2561        }
2562
2563        /* Force the pack to open to prove its valid. */
2564        return !open_packed_git(p);
2565}
2566
2567static int fill_pack_entry(const unsigned char *sha1,
2568                           struct pack_entry *e,
2569                           struct packed_git *p)
2570{
2571        off_t offset;
2572
2573        if (p->num_bad_objects) {
2574                unsigned i;
2575                for (i = 0; i < p->num_bad_objects; i++)
2576                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2577                                return 0;
2578        }
2579
2580        offset = find_pack_entry_one(sha1, p);
2581        if (!offset)
2582                return 0;
2583
2584        /*
2585         * We are about to tell the caller where they can locate the
2586         * requested object.  We better make sure the packfile is
2587         * still here and can be accessed before supplying that
2588         * answer, as it may have been deleted since the index was
2589         * loaded!
2590         */
2591        if (!is_pack_valid(p))
2592                return 0;
2593        e->offset = offset;
2594        e->p = p;
2595        hashcpy(e->sha1, sha1);
2596        return 1;
2597}
2598
2599/*
2600 * Iff a pack file contains the object named by sha1, return true and
2601 * store its location to e.
2602 */
2603static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2604{
2605        struct mru_entry *p;
2606
2607        prepare_packed_git();
2608        if (!packed_git)
2609                return 0;
2610
2611        for (p = packed_git_mru->head; p; p = p->next) {
2612                if (fill_pack_entry(sha1, e, p->item)) {
2613                        mru_mark(packed_git_mru, p);
2614                        return 1;
2615                }
2616        }
2617        return 0;
2618}
2619
2620struct packed_git *find_sha1_pack(const unsigned char *sha1,
2621                                  struct packed_git *packs)
2622{
2623        struct packed_git *p;
2624
2625        for (p = packs; p; p = p->next) {
2626                if (find_pack_entry_one(sha1, p))
2627                        return p;
2628        }
2629        return NULL;
2630
2631}
2632
2633static int sha1_loose_object_info(const unsigned char *sha1,
2634                                  struct object_info *oi,
2635                                  int flags)
2636{
2637        int status = 0;
2638        unsigned long mapsize;
2639        void *map;
2640        git_zstream stream;
2641        char hdr[32];
2642        struct strbuf hdrbuf = STRBUF_INIT;
2643
2644        if (oi->delta_base_sha1)
2645                hashclr(oi->delta_base_sha1);
2646
2647        /*
2648         * If we don't care about type or size, then we don't
2649         * need to look inside the object at all. Note that we
2650         * do not optimize out the stat call, even if the
2651         * caller doesn't care about the disk-size, since our
2652         * return value implicitly indicates whether the
2653         * object even exists.
2654         */
2655        if (!oi->typep && !oi->typename && !oi->sizep) {
2656                struct stat st;
2657                if (stat_sha1_file(sha1, &st) < 0)
2658                        return -1;
2659                if (oi->disk_sizep)
2660                        *oi->disk_sizep = st.st_size;
2661                return 0;
2662        }
2663
2664        map = map_sha1_file(sha1, &mapsize);
2665        if (!map)
2666                return -1;
2667        if (oi->disk_sizep)
2668                *oi->disk_sizep = mapsize;
2669        if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2670                if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2671                        status = error("unable to unpack %s header with --allow-unknown-type",
2672                                       sha1_to_hex(sha1));
2673        } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2674                status = error("unable to unpack %s header",
2675                               sha1_to_hex(sha1));
2676        if (status < 0)
2677                ; /* Do nothing */
2678        else if (hdrbuf.len) {
2679                if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2680                        status = error("unable to parse %s header with --allow-unknown-type",
2681                                       sha1_to_hex(sha1));
2682        } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2683                status = error("unable to parse %s header", sha1_to_hex(sha1));
2684        git_inflate_end(&stream);
2685        munmap(map, mapsize);
2686        if (status && oi->typep)
2687                *oi->typep = status;
2688        strbuf_release(&hdrbuf);
2689        return 0;
2690}
2691
2692int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2693{
2694        struct cached_object *co;
2695        struct pack_entry e;
2696        int rtype;
2697        enum object_type real_type;
2698        const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2699
2700        co = find_cached_object(real);
2701        if (co) {
2702                if (oi->typep)
2703                        *(oi->typep) = co->type;
2704                if (oi->sizep)
2705                        *(oi->sizep) = co->size;
2706                if (oi->disk_sizep)
2707                        *(oi->disk_sizep) = 0;
2708                if (oi->delta_base_sha1)
2709                        hashclr(oi->delta_base_sha1);
2710                if (oi->typename)
2711                        strbuf_addstr(oi->typename, typename(co->type));
2712                oi->whence = OI_CACHED;
2713                return 0;
2714        }
2715
2716        if (!find_pack_entry(real, &e)) {
2717                /* Most likely it's a loose object. */
2718                if (!sha1_loose_object_info(real, oi, flags)) {
2719                        oi->whence = OI_LOOSE;
2720                        return 0;
2721                }
2722
2723                /* Not a loose object; someone else may have just packed it. */
2724                reprepare_packed_git();
2725                if (!find_pack_entry(real, &e))
2726                        return -1;
2727        }
2728
2729        /*
2730         * packed_object_info() does not follow the delta chain to
2731         * find out the real type, unless it is given oi->typep.
2732         */
2733        if (oi->typename && !oi->typep)
2734                oi->typep = &real_type;
2735
2736        rtype = packed_object_info(e.p, e.offset, oi);
2737        if (rtype < 0) {
2738                mark_bad_packed_object(e.p, real);
2739                if (oi->typep == &real_type)
2740                        oi->typep = NULL;
2741                return sha1_object_info_extended(real, oi, 0);
2742        } else if (in_delta_base_cache(e.p, e.offset)) {
2743                oi->whence = OI_DBCACHED;
2744        } else {
2745                oi->whence = OI_PACKED;
2746                oi->u.packed.offset = e.offset;
2747                oi->u.packed.pack = e.p;
2748                oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2749                                         rtype == OBJ_OFS_DELTA);
2750        }
2751        if (oi->typename)
2752                strbuf_addstr(oi->typename, typename(*oi->typep));
2753        if (oi->typep == &real_type)
2754                oi->typep = NULL;
2755
2756        return 0;
2757}
2758
2759/* returns enum object_type or negative */
2760int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2761{
2762        enum object_type type;
2763        struct object_info oi = {NULL};
2764
2765        oi.typep = &type;
2766        oi.sizep = sizep;
2767        if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2768                return -1;
2769        return type;
2770}
2771
2772static void *read_packed_sha1(const unsigned char *sha1,
2773                              enum object_type *type, unsigned long *size)
2774{
2775        struct pack_entry e;
2776        void *data;
2777
2778        if (!find_pack_entry(sha1, &e))
2779                return NULL;
2780        data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2781        if (!data) {
2782                /*
2783                 * We're probably in deep shit, but let's try to fetch
2784                 * the required object anyway from another pack or loose.
2785                 * This should happen only in the presence of a corrupted
2786                 * pack, and is better than failing outright.
2787                 */
2788                error("failed to read object %s at offset %"PRIuMAX" from %s",
2789                      sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2790                mark_bad_packed_object(e.p, sha1);
2791                data = read_object(sha1, type, size);
2792        }
2793        return data;
2794}
2795
2796int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2797                      unsigned char *sha1)
2798{
2799        struct cached_object *co;
2800
2801        hash_sha1_file(buf, len, typename(type), sha1);
2802        if (has_sha1_file(sha1) || find_cached_object(sha1))
2803                return 0;
2804        ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
2805        co = &cached_objects[cached_object_nr++];
2806        co->size = len;
2807        co->type = type;
2808        co->buf = xmalloc(len);
2809        memcpy(co->buf, buf, len);
2810        hashcpy(co->sha1, sha1);
2811        return 0;
2812}
2813
2814static void *read_object(const unsigned char *sha1, enum object_type *type,
2815                         unsigned long *size)
2816{
2817        unsigned long mapsize;
2818        void *map, *buf;
2819        struct cached_object *co;
2820
2821        co = find_cached_object(sha1);
2822        if (co) {
2823                *type = co->type;
2824                *size = co->size;
2825                return xmemdupz(co->buf, co->size);
2826        }
2827
2828        buf = read_packed_sha1(sha1, type, size);
2829        if (buf)
2830                return buf;
2831        map = map_sha1_file(sha1, &mapsize);
2832        if (map) {
2833                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2834                munmap(map, mapsize);
2835                return buf;
2836        }
2837        reprepare_packed_git();
2838        return read_packed_sha1(sha1, type, size);
2839}
2840
2841/*
2842 * This function dies on corrupt objects; the callers who want to
2843 * deal with them should arrange to call read_object() and give error
2844 * messages themselves.
2845 */
2846void *read_sha1_file_extended(const unsigned char *sha1,
2847                              enum object_type *type,
2848                              unsigned long *size,
2849                              unsigned flag)
2850{
2851        void *data;
2852        const struct packed_git *p;
2853        const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2854
2855        errno = 0;
2856        data = read_object(repl, type, size);
2857        if (data)
2858                return data;
2859
2860        if (errno && errno != ENOENT)
2861                die_errno("failed to read object %s", sha1_to_hex(sha1));
2862
2863        /* die if we replaced an object with one that does not exist */
2864        if (repl != sha1)
2865                die("replacement %s not found for %s",
2866                    sha1_to_hex(repl), sha1_to_hex(sha1));
2867
2868        if (has_loose_object(repl)) {
2869                const char *path = sha1_file_name(sha1);
2870
2871                die("loose object %s (stored in %s) is corrupt",
2872                    sha1_to_hex(repl), path);
2873        }
2874
2875        if ((p = has_packed_and_bad(repl)) != NULL)
2876                die("packed object %s (stored in %s) is corrupt",
2877                    sha1_to_hex(repl), p->pack_name);
2878
2879        return NULL;
2880}
2881
2882void *read_object_with_reference(const unsigned char *sha1,
2883                                 const char *required_type_name,
2884                                 unsigned long *size,
2885                                 unsigned char *actual_sha1_return)
2886{
2887        enum object_type type, required_type;
2888        void *buffer;
2889        unsigned long isize;
2890        unsigned char actual_sha1[20];
2891
2892        required_type = type_from_string(required_type_name);
2893        hashcpy(actual_sha1, sha1);
2894        while (1) {
2895                int ref_length = -1;
2896                const char *ref_type = NULL;
2897
2898                buffer = read_sha1_file(actual_sha1, &type, &isize);
2899                if (!buffer)
2900                        return NULL;
2901                if (type == required_type) {
2902                        *size = isize;
2903                        if (actual_sha1_return)
2904                                hashcpy(actual_sha1_return, actual_sha1);
2905                        return buffer;
2906                }
2907                /* Handle references */
2908                else if (type == OBJ_COMMIT)
2909                        ref_type = "tree ";
2910                else if (type == OBJ_TAG)
2911                        ref_type = "object ";
2912                else {
2913                        free(buffer);
2914                        return NULL;
2915                }
2916                ref_length = strlen(ref_type);
2917
2918                if (ref_length + 40 > isize ||
2919                    memcmp(buffer, ref_type, ref_length) ||
2920                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2921                        free(buffer);
2922                        return NULL;
2923                }
2924                free(buffer);
2925                /* Now we have the ID of the referred-to object in
2926                 * actual_sha1.  Check again. */
2927        }
2928}
2929
2930static void write_sha1_file_prepare(const void *buf, unsigned long len,
2931                                    const char *type, unsigned char *sha1,
2932                                    char *hdr, int *hdrlen)
2933{
2934        git_SHA_CTX c;
2935
2936        /* Generate the header */
2937        *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
2938
2939        /* Sha1.. */
2940        git_SHA1_Init(&c);
2941        git_SHA1_Update(&c, hdr, *hdrlen);
2942        git_SHA1_Update(&c, buf, len);
2943        git_SHA1_Final(sha1, &c);
2944}
2945
2946/*
2947 * Move the just written object into its final resting place.
2948 */
2949int finalize_object_file(const char *tmpfile, const char *filename)
2950{
2951        int ret = 0;
2952
2953        if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2954                goto try_rename;
2955        else if (link(tmpfile, filename))
2956                ret = errno;
2957
2958        /*
2959         * Coda hack - coda doesn't like cross-directory links,
2960         * so we fall back to a rename, which will mean that it
2961         * won't be able to check collisions, but that's not a
2962         * big deal.
2963         *
2964         * The same holds for FAT formatted media.
2965         *
2966         * When this succeeds, we just return.  We have nothing
2967         * left to unlink.
2968         */
2969        if (ret && ret != EEXIST) {
2970        try_rename:
2971                if (!rename(tmpfile, filename))
2972                        goto out;
2973                ret = errno;
2974        }
2975        unlink_or_warn(tmpfile);
2976        if (ret) {
2977                if (ret != EEXIST) {
2978                        return error_errno("unable to write sha1 filename %s", filename);
2979                }
2980                /* FIXME!!! Collision check here ? */
2981        }
2982
2983out:
2984        if (adjust_shared_perm(filename))
2985                return error("unable to set permission to '%s'", filename);
2986        return 0;
2987}
2988
2989static int write_buffer(int fd, const void *buf, size_t len)
2990{
2991        if (write_in_full(fd, buf, len) < 0)
2992                return error_errno("file write error");
2993        return 0;
2994}
2995
2996int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2997                   unsigned char *sha1)
2998{
2999        char hdr[32];
3000        int hdrlen = sizeof(hdr);
3001        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3002        return 0;
3003}
3004
3005/* Finalize a file on disk, and close it. */
3006static void close_sha1_file(int fd)
3007{
3008        if (fsync_object_files)
3009                fsync_or_die(fd, "sha1 file");
3010        if (close(fd) != 0)
3011                die_errno("error when closing sha1 file");
3012}
3013
3014/* Size of directory component, including the ending '/' */
3015static inline int directory_size(const char *filename)
3016{
3017        const char *s = strrchr(filename, '/');
3018        if (!s)
3019                return 0;
3020        return s - filename + 1;
3021}
3022
3023/*
3024 * This creates a temporary file in the same directory as the final
3025 * 'filename'
3026 *
3027 * We want to avoid cross-directory filename renames, because those
3028 * can have problems on various filesystems (FAT, NFS, Coda).
3029 */
3030static int create_tmpfile(struct strbuf *tmp, const char *filename)
3031{
3032        int fd, dirlen = directory_size(filename);
3033
3034        strbuf_reset(tmp);
3035        strbuf_add(tmp, filename, dirlen);
3036        strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3037        fd = git_mkstemp_mode(tmp->buf, 0444);
3038        if (fd < 0 && dirlen && errno == ENOENT) {
3039                /*
3040                 * Make sure the directory exists; note that the contents
3041                 * of the buffer are undefined after mkstemp returns an
3042                 * error, so we have to rewrite the whole buffer from
3043                 * scratch.
3044                 */
3045                strbuf_reset(tmp);
3046                strbuf_add(tmp, filename, dirlen - 1);
3047                if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3048                        return -1;
3049                if (adjust_shared_perm(tmp->buf))
3050                        return -1;
3051
3052                /* Try again */
3053                strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3054                fd = git_mkstemp_mode(tmp->buf, 0444);
3055        }
3056        return fd;
3057}
3058
3059static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3060                              const void *buf, unsigned long len, time_t mtime)
3061{
3062        int fd, ret;
3063        unsigned char compressed[4096];
3064        git_zstream stream;
3065        git_SHA_CTX c;
3066        unsigned char parano_sha1[20];
3067        static struct strbuf tmp_file = STRBUF_INIT;
3068        const char *filename = sha1_file_name(sha1);
3069
3070        fd = create_tmpfile(&tmp_file, filename);
3071        if (fd < 0) {
3072                if (errno == EACCES)
3073                        return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3074                else
3075                        return error_errno("unable to create temporary file");
3076        }
3077
3078        /* Set it up */
3079        git_deflate_init(&stream, zlib_compression_level);
3080        stream.next_out = compressed;
3081        stream.avail_out = sizeof(compressed);
3082        git_SHA1_Init(&c);
3083
3084        /* First header.. */
3085        stream.next_in = (unsigned char *)hdr;
3086        stream.avail_in = hdrlen;
3087        while (git_deflate(&stream, 0) == Z_OK)
3088                ; /* nothing */
3089        git_SHA1_Update(&c, hdr, hdrlen);
3090
3091        /* Then the data itself.. */
3092        stream.next_in = (void *)buf;
3093        stream.avail_in = len;
3094        do {
3095                unsigned char *in0 = stream.next_in;
3096                ret = git_deflate(&stream, Z_FINISH);
3097                git_SHA1_Update(&c, in0, stream.next_in - in0);
3098                if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3099                        die("unable to write sha1 file");
3100                stream.next_out = compressed;
3101                stream.avail_out = sizeof(compressed);
3102        } while (ret == Z_OK);
3103
3104        if (ret != Z_STREAM_END)
3105                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3106        ret = git_deflate_end_gently(&stream);
3107        if (ret != Z_OK)
3108                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3109        git_SHA1_Final(parano_sha1, &c);
3110        if (hashcmp(sha1, parano_sha1) != 0)
3111                die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3112
3113        close_sha1_file(fd);
3114
3115        if (mtime) {
3116                struct utimbuf utb;
3117                utb.actime = mtime;
3118                utb.modtime = mtime;
3119                if (utime(tmp_file.buf, &utb) < 0)
3120                        warning_errno("failed utime() on %s", tmp_file.buf);
3121        }
3122
3123        return finalize_object_file(tmp_file.buf, filename);
3124}
3125
3126static int freshen_loose_object(const unsigned char *sha1)
3127{
3128        return check_and_freshen(sha1, 1);
3129}
3130
3131static int freshen_packed_object(const unsigned char *sha1)
3132{
3133        struct pack_entry e;
3134        if (!find_pack_entry(sha1, &e))
3135                return 0;
3136        if (e.p->freshened)
3137                return 1;
3138        if (!freshen_file(e.p->pack_name))
3139                return 0;
3140        e.p->freshened = 1;
3141        return 1;
3142}
3143
3144int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3145{
3146        char hdr[32];
3147        int hdrlen = sizeof(hdr);
3148
3149        /* Normally if we have it in the pack then we do not bother writing
3150         * it out into .git/objects/??/?{38} file.
3151         */
3152        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3153        if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3154                return 0;
3155        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3156}
3157
3158int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3159                             unsigned char *sha1, unsigned flags)
3160{
3161        char *header;
3162        int hdrlen, status = 0;
3163
3164        /* type string, SP, %lu of the length plus NUL must fit this */
3165        hdrlen = strlen(type) + 32;
3166        header = xmalloc(hdrlen);
3167        write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3168
3169        if (!(flags & HASH_WRITE_OBJECT))
3170                goto cleanup;
3171        if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3172                goto cleanup;
3173        status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3174
3175cleanup:
3176        free(header);
3177        return status;
3178}
3179
3180int force_object_loose(const unsigned char *sha1, time_t mtime)
3181{
3182        void *buf;
3183        unsigned long len;
3184        enum object_type type;
3185        char hdr[32];
3186        int hdrlen;
3187        int ret;
3188
3189        if (has_loose_object(sha1))
3190                return 0;
3191        buf = read_packed_sha1(sha1, &type, &len);
3192        if (!buf)
3193                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3194        hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3195        ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3196        free(buf);
3197
3198        return ret;
3199}
3200
3201int has_pack_index(const unsigned char *sha1)
3202{
3203        struct stat st;
3204        if (stat(sha1_pack_index_name(sha1), &st))
3205                return 0;
3206        return 1;
3207}
3208
3209int has_sha1_pack(const unsigned char *sha1)
3210{
3211        struct pack_entry e;
3212        return find_pack_entry(sha1, &e);
3213}
3214
3215int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3216{
3217        struct pack_entry e;
3218
3219        if (find_pack_entry(sha1, &e))
3220                return 1;
3221        if (has_loose_object(sha1))
3222                return 1;
3223        if (flags & HAS_SHA1_QUICK)
3224                return 0;
3225        reprepare_packed_git();
3226        return find_pack_entry(sha1, &e);
3227}
3228
3229int has_object_file(const struct object_id *oid)
3230{
3231        return has_sha1_file(oid->hash);
3232}
3233
3234int has_object_file_with_flags(const struct object_id *oid, int flags)
3235{
3236        return has_sha1_file_with_flags(oid->hash, flags);
3237}
3238
3239static void check_tree(const void *buf, size_t size)
3240{
3241        struct tree_desc desc;
3242        struct name_entry entry;
3243
3244        init_tree_desc(&desc, buf, size);
3245        while (tree_entry(&desc, &entry))
3246                /* do nothing
3247                 * tree_entry() will die() on malformed entries */
3248                ;
3249}
3250
3251static void check_commit(const void *buf, size_t size)
3252{
3253        struct commit c;
3254        memset(&c, 0, sizeof(c));
3255        if (parse_commit_buffer(&c, buf, size))
3256                die("corrupt commit");
3257}
3258
3259static void check_tag(const void *buf, size_t size)
3260{
3261        struct tag t;
3262        memset(&t, 0, sizeof(t));
3263        if (parse_tag_buffer(&t, buf, size))
3264                die("corrupt tag");
3265}
3266
3267static int index_mem(unsigned char *sha1, void *buf, size_t size,
3268                     enum object_type type,
3269                     const char *path, unsigned flags)
3270{
3271        int ret, re_allocated = 0;
3272        int write_object = flags & HASH_WRITE_OBJECT;
3273
3274        if (!type)
3275                type = OBJ_BLOB;
3276
3277        /*
3278         * Convert blobs to git internal format
3279         */
3280        if ((type == OBJ_BLOB) && path) {
3281                struct strbuf nbuf = STRBUF_INIT;
3282                if (convert_to_git(path, buf, size, &nbuf,
3283                                   write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3284                        buf = strbuf_detach(&nbuf, &size);
3285                        re_allocated = 1;
3286                }
3287        }
3288        if (flags & HASH_FORMAT_CHECK) {
3289                if (type == OBJ_TREE)
3290                        check_tree(buf, size);
3291                if (type == OBJ_COMMIT)
3292                        check_commit(buf, size);
3293                if (type == OBJ_TAG)
3294                        check_tag(buf, size);
3295        }
3296
3297        if (write_object)
3298                ret = write_sha1_file(buf, size, typename(type), sha1);
3299        else
3300                ret = hash_sha1_file(buf, size, typename(type), sha1);
3301        if (re_allocated)
3302                free(buf);
3303        return ret;
3304}
3305
3306static int index_stream_convert_blob(unsigned char *sha1, int fd,
3307                                     const char *path, unsigned flags)
3308{
3309        int ret;
3310        const int write_object = flags & HASH_WRITE_OBJECT;
3311        struct strbuf sbuf = STRBUF_INIT;
3312
3313        assert(path);
3314        assert(would_convert_to_git_filter_fd(path));
3315
3316        convert_to_git_filter_fd(path, fd, &sbuf,
3317                                 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3318
3319        if (write_object)
3320                ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3321                                      sha1);
3322        else
3323                ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3324                                     sha1);
3325        strbuf_release(&sbuf);
3326        return ret;
3327}
3328
3329static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3330                      const char *path, unsigned flags)
3331{
3332        struct strbuf sbuf = STRBUF_INIT;
3333        int ret;
3334
3335        if (strbuf_read(&sbuf, fd, 4096) >= 0)
3336                ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3337        else
3338                ret = -1;
3339        strbuf_release(&sbuf);
3340        return ret;
3341}
3342
3343#define SMALL_FILE_SIZE (32*1024)
3344
3345static int index_core(unsigned char *sha1, int fd, size_t size,
3346                      enum object_type type, const char *path,
3347                      unsigned flags)
3348{
3349        int ret;
3350
3351        if (!size) {
3352                ret = index_mem(sha1, "", size, type, path, flags);
3353        } else if (size <= SMALL_FILE_SIZE) {
3354                char *buf = xmalloc(size);
3355                if (size == read_in_full(fd, buf, size))
3356                        ret = index_mem(sha1, buf, size, type, path, flags);
3357                else
3358                        ret = error_errno("short read");
3359                free(buf);
3360        } else {
3361                void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3362                ret = index_mem(sha1, buf, size, type, path, flags);
3363                munmap(buf, size);
3364        }
3365        return ret;
3366}
3367
3368/*
3369 * This creates one packfile per large blob unless bulk-checkin
3370 * machinery is "plugged".
3371 *
3372 * This also bypasses the usual "convert-to-git" dance, and that is on
3373 * purpose. We could write a streaming version of the converting
3374 * functions and insert that before feeding the data to fast-import
3375 * (or equivalent in-core API described above). However, that is
3376 * somewhat complicated, as we do not know the size of the filter
3377 * result, which we need to know beforehand when writing a git object.
3378 * Since the primary motivation for trying to stream from the working
3379 * tree file and to avoid mmaping it in core is to deal with large
3380 * binary blobs, they generally do not want to get any conversion, and
3381 * callers should avoid this code path when filters are requested.
3382 */
3383static int index_stream(unsigned char *sha1, int fd, size_t size,
3384                        enum object_type type, const char *path,
3385                        unsigned flags)
3386{
3387        return index_bulk_checkin(sha1, fd, size, type, path, flags);
3388}
3389
3390int index_fd(unsigned char *sha1, int fd, struct stat *st,
3391             enum object_type type, const char *path, unsigned flags)
3392{
3393        int ret;
3394
3395        /*
3396         * Call xsize_t() only when needed to avoid potentially unnecessary
3397         * die() for large files.
3398         */
3399        if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3400                ret = index_stream_convert_blob(sha1, fd, path, flags);
3401        else if (!S_ISREG(st->st_mode))
3402                ret = index_pipe(sha1, fd, type, path, flags);
3403        else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3404                 (path && would_convert_to_git(path)))
3405                ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3406                                 flags);
3407        else
3408                ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3409                                   flags);
3410        close(fd);
3411        return ret;
3412}
3413
3414int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3415{
3416        int fd;
3417        struct strbuf sb = STRBUF_INIT;
3418
3419        switch (st->st_mode & S_IFMT) {
3420        case S_IFREG:
3421                fd = open(path, O_RDONLY);
3422                if (fd < 0)
3423                        return error_errno("open(\"%s\")", path);
3424                if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3425                        return error("%s: failed to insert into database",
3426                                     path);
3427                break;
3428        case S_IFLNK:
3429                if (strbuf_readlink(&sb, path, st->st_size))
3430                        return error_errno("readlink(\"%s\")", path);
3431                if (!(flags & HASH_WRITE_OBJECT))
3432                        hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3433                else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3434                        return error("%s: failed to insert into database",
3435                                     path);
3436                strbuf_release(&sb);
3437                break;
3438        case S_IFDIR:
3439                return resolve_gitlink_ref(path, "HEAD", sha1);
3440        default:
3441                return error("%s: unsupported file type", path);
3442        }
3443        return 0;
3444}
3445
3446int read_pack_header(int fd, struct pack_header *header)
3447{
3448        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3449                /* "eof before pack header was fully read" */
3450                return PH_ERROR_EOF;
3451
3452        if (header->hdr_signature != htonl(PACK_SIGNATURE))
3453                /* "protocol error (pack signature mismatch detected)" */
3454                return PH_ERROR_PACK_SIGNATURE;
3455        if (!pack_version_ok(header->hdr_version))
3456                /* "protocol error (pack version unsupported)" */
3457                return PH_ERROR_PROTOCOL;
3458        return 0;
3459}
3460
3461void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3462{
3463        enum object_type type = sha1_object_info(sha1, NULL);
3464        if (type < 0)
3465                die("%s is not a valid object", sha1_to_hex(sha1));
3466        if (type != expect)
3467                die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3468                    typename(expect));
3469}
3470
3471static int for_each_file_in_obj_subdir(int subdir_nr,
3472                                       struct strbuf *path,
3473                                       each_loose_object_fn obj_cb,
3474                                       each_loose_cruft_fn cruft_cb,
3475                                       each_loose_subdir_fn subdir_cb,
3476                                       void *data)
3477{
3478        size_t baselen = path->len;
3479        DIR *dir = opendir(path->buf);
3480        struct dirent *de;
3481        int r = 0;
3482
3483        if (!dir) {
3484                if (errno == ENOENT)
3485                        return 0;
3486                return error_errno("unable to open %s", path->buf);
3487        }
3488
3489        while ((de = readdir(dir))) {
3490                if (is_dot_or_dotdot(de->d_name))
3491                        continue;
3492
3493                strbuf_setlen(path, baselen);
3494                strbuf_addf(path, "/%s", de->d_name);
3495
3496                if (strlen(de->d_name) == 38)  {
3497                        char hex[41];
3498                        unsigned char sha1[20];
3499
3500                        snprintf(hex, sizeof(hex), "%02x%s",
3501                                 subdir_nr, de->d_name);
3502                        if (!get_sha1_hex(hex, sha1)) {
3503                                if (obj_cb) {
3504                                        r = obj_cb(sha1, path->buf, data);
3505                                        if (r)
3506                                                break;
3507                                }
3508                                continue;
3509                        }
3510                }
3511
3512                if (cruft_cb) {
3513                        r = cruft_cb(de->d_name, path->buf, data);
3514                        if (r)
3515                                break;
3516                }
3517        }
3518        closedir(dir);
3519
3520        strbuf_setlen(path, baselen);
3521        if (!r && subdir_cb)
3522                r = subdir_cb(subdir_nr, path->buf, data);
3523
3524        return r;
3525}
3526
3527int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3528                            each_loose_object_fn obj_cb,
3529                            each_loose_cruft_fn cruft_cb,
3530                            each_loose_subdir_fn subdir_cb,
3531                            void *data)
3532{
3533        size_t baselen = path->len;
3534        int r = 0;
3535        int i;
3536
3537        for (i = 0; i < 256; i++) {
3538                strbuf_addf(path, "/%02x", i);
3539                r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3540                                                subdir_cb, data);
3541                strbuf_setlen(path, baselen);
3542                if (r)
3543                        break;
3544        }
3545
3546        return r;
3547}
3548
3549int for_each_loose_file_in_objdir(const char *path,
3550                                  each_loose_object_fn obj_cb,
3551                                  each_loose_cruft_fn cruft_cb,
3552                                  each_loose_subdir_fn subdir_cb,
3553                                  void *data)
3554{
3555        struct strbuf buf = STRBUF_INIT;
3556        int r;
3557
3558        strbuf_addstr(&buf, path);
3559        r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3560                                              subdir_cb, data);
3561        strbuf_release(&buf);
3562
3563        return r;
3564}
3565
3566struct loose_alt_odb_data {
3567        each_loose_object_fn *cb;
3568        void *data;
3569};
3570
3571static int loose_from_alt_odb(struct alternate_object_database *alt,
3572                              void *vdata)
3573{
3574        struct loose_alt_odb_data *data = vdata;
3575        struct strbuf buf = STRBUF_INIT;
3576        int r;
3577
3578        /* copy base not including trailing '/' */
3579        strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3580        r = for_each_loose_file_in_objdir_buf(&buf,
3581                                              data->cb, NULL, NULL,
3582                                              data->data);
3583        strbuf_release(&buf);
3584        return r;
3585}
3586
3587int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3588{
3589        struct loose_alt_odb_data alt;
3590        int r;
3591
3592        r = for_each_loose_file_in_objdir(get_object_directory(),
3593                                          cb, NULL, NULL, data);
3594        if (r)
3595                return r;
3596
3597        if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3598                return 0;
3599
3600        alt.cb = cb;
3601        alt.data = data;
3602        return foreach_alt_odb(loose_from_alt_odb, &alt);
3603}
3604
3605static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3606{
3607        uint32_t i;
3608        int r = 0;
3609
3610        for (i = 0; i < p->num_objects; i++) {
3611                const unsigned char *sha1 = nth_packed_object_sha1(p, i);
3612
3613                if (!sha1)
3614                        return error("unable to get sha1 of object %u in %s",
3615                                     i, p->pack_name);
3616
3617                r = cb(sha1, p, i, data);
3618                if (r)
3619                        break;
3620        }
3621        return r;
3622}
3623
3624int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3625{
3626        struct packed_git *p;
3627        int r = 0;
3628        int pack_errors = 0;
3629
3630        prepare_packed_git();
3631        for (p = packed_git; p; p = p->next) {
3632                if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3633                        continue;
3634                if (open_pack_index(p)) {
3635                        pack_errors = 1;
3636                        continue;
3637                }
3638                r = for_each_object_in_pack(p, cb, data);
3639                if (r)
3640                        break;
3641        }
3642        return r ? r : pack_errors;
3643}