fb0506ea9e4751ac45124256ca9ecb7b65e07b9c
   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 "delta.h"
  12#include "pack.h"
  13#include "blob.h"
  14#include "commit.h"
  15#include "run-command.h"
  16#include "tag.h"
  17#include "tree.h"
  18#include "tree-walk.h"
  19#include "refs.h"
  20#include "pack-revindex.h"
  21#include "sha1-lookup.h"
  22#include "bulk-checkin.h"
  23#include "streaming.h"
  24
  25#ifndef O_NOATIME
  26#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  27#define O_NOATIME 01000000
  28#else
  29#define O_NOATIME 0
  30#endif
  31#endif
  32
  33#define SZ_FMT PRIuMAX
  34static inline uintmax_t sz_fmt(size_t s) { return s; }
  35
  36const unsigned char null_sha1[20];
  37
  38/*
  39 * This is meant to hold a *small* number of objects that you would
  40 * want read_sha1_file() to be able to return, but yet you do not want
  41 * to write them into the object store (e.g. a browse-only
  42 * application).
  43 */
  44static struct cached_object {
  45        unsigned char sha1[20];
  46        enum object_type type;
  47        void *buf;
  48        unsigned long size;
  49} *cached_objects;
  50static int cached_object_nr, cached_object_alloc;
  51
  52static struct cached_object empty_tree = {
  53        EMPTY_TREE_SHA1_BIN_LITERAL,
  54        OBJ_TREE,
  55        "",
  56        0
  57};
  58
  59static struct packed_git *last_found_pack;
  60
  61static struct cached_object *find_cached_object(const unsigned char *sha1)
  62{
  63        int i;
  64        struct cached_object *co = cached_objects;
  65
  66        for (i = 0; i < cached_object_nr; i++, co++) {
  67                if (!hashcmp(co->sha1, sha1))
  68                        return co;
  69        }
  70        if (!hashcmp(sha1, empty_tree.sha1))
  71                return &empty_tree;
  72        return NULL;
  73}
  74
  75int mkdir_in_gitdir(const char *path)
  76{
  77        if (mkdir(path, 0777)) {
  78                int saved_errno = errno;
  79                struct stat st;
  80                struct strbuf sb = STRBUF_INIT;
  81
  82                if (errno != EEXIST)
  83                        return -1;
  84                /*
  85                 * Are we looking at a path in a symlinked worktree
  86                 * whose original repository does not yet have it?
  87                 * e.g. .git/rr-cache pointing at its original
  88                 * repository in which the user hasn't performed any
  89                 * conflict resolution yet?
  90                 */
  91                if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
  92                    strbuf_readlink(&sb, path, st.st_size) ||
  93                    !is_absolute_path(sb.buf) ||
  94                    mkdir(sb.buf, 0777)) {
  95                        strbuf_release(&sb);
  96                        errno = saved_errno;
  97                        return -1;
  98                }
  99                strbuf_release(&sb);
 100        }
 101        return adjust_shared_perm(path);
 102}
 103
 104int safe_create_leading_directories(char *path)
 105{
 106        char *pos = path + offset_1st_component(path);
 107        struct stat st;
 108
 109        while (pos) {
 110                pos = strchr(pos, '/');
 111                if (!pos)
 112                        break;
 113                while (*++pos == '/')
 114                        ;
 115                if (!*pos)
 116                        break;
 117                *--pos = '\0';
 118                if (!stat(path, &st)) {
 119                        /* path exists */
 120                        if (!S_ISDIR(st.st_mode)) {
 121                                *pos = '/';
 122                                return -3;
 123                        }
 124                }
 125                else if (mkdir(path, 0777)) {
 126                        *pos = '/';
 127                        return -1;
 128                }
 129                else if (adjust_shared_perm(path)) {
 130                        *pos = '/';
 131                        return -2;
 132                }
 133                *pos++ = '/';
 134        }
 135        return 0;
 136}
 137
 138int safe_create_leading_directories_const(const char *path)
 139{
 140        /* path points to cache entries, so xstrdup before messing with it */
 141        char *buf = xstrdup(path);
 142        int result = safe_create_leading_directories(buf);
 143        free(buf);
 144        return result;
 145}
 146
 147static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 148{
 149        int i;
 150        for (i = 0; i < 20; i++) {
 151                static char hex[] = "0123456789abcdef";
 152                unsigned int val = sha1[i];
 153                char *pos = pathbuf + i*2 + (i > 0);
 154                *pos++ = hex[val >> 4];
 155                *pos = hex[val & 0xf];
 156        }
 157}
 158
 159/*
 160 * NOTE! This returns a statically allocated buffer, so you have to be
 161 * careful about using it. Do an "xstrdup()" if you need to save the
 162 * filename.
 163 *
 164 * Also note that this returns the location for creating.  Reading
 165 * SHA1 file can happen from any alternate directory listed in the
 166 * DB_ENVIRONMENT environment variable if it is not found in
 167 * the primary object database.
 168 */
 169char *sha1_file_name(const unsigned char *sha1)
 170{
 171        static char buf[PATH_MAX];
 172        const char *objdir;
 173        int len;
 174
 175        objdir = get_object_directory();
 176        len = strlen(objdir);
 177
 178        /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
 179        if (len + 43 > PATH_MAX)
 180                die("insanely long object directory %s", objdir);
 181        memcpy(buf, objdir, len);
 182        buf[len] = '/';
 183        buf[len+3] = '/';
 184        buf[len+42] = '\0';
 185        fill_sha1_path(buf + len + 1, sha1);
 186        return buf;
 187}
 188
 189static char *sha1_get_pack_name(const unsigned char *sha1,
 190                                char **name, char **base, const char *which)
 191{
 192        static const char hex[] = "0123456789abcdef";
 193        char *buf;
 194        int i;
 195
 196        if (!*base) {
 197                const char *sha1_file_directory = get_object_directory();
 198                int len = strlen(sha1_file_directory);
 199                *base = xmalloc(len + 60);
 200                sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
 201                        sha1_file_directory, which);
 202                *name = *base + len + 11;
 203        }
 204
 205        buf = *name;
 206
 207        for (i = 0; i < 20; i++) {
 208                unsigned int val = *sha1++;
 209                *buf++ = hex[val >> 4];
 210                *buf++ = hex[val & 0xf];
 211        }
 212
 213        return *base;
 214}
 215
 216char *sha1_pack_name(const unsigned char *sha1)
 217{
 218        static char *name, *base;
 219
 220        return sha1_get_pack_name(sha1, &name, &base, "pack");
 221}
 222
 223char *sha1_pack_index_name(const unsigned char *sha1)
 224{
 225        static char *name, *base;
 226
 227        return sha1_get_pack_name(sha1, &name, &base, "idx");
 228}
 229
 230struct alternate_object_database *alt_odb_list;
 231static struct alternate_object_database **alt_odb_tail;
 232
 233static int git_open_noatime(const char *name);
 234
 235/*
 236 * Prepare alternate object database registry.
 237 *
 238 * The variable alt_odb_list points at the list of struct
 239 * alternate_object_database.  The elements on this list come from
 240 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 241 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 242 * whose contents is similar to that environment variable but can be
 243 * LF separated.  Its base points at a statically allocated buffer that
 244 * contains "/the/directory/corresponding/to/.git/objects/...", while
 245 * its name points just after the slash at the end of ".git/objects/"
 246 * in the example above, and has enough space to hold 40-byte hex
 247 * SHA1, an extra slash for the first level indirection, and the
 248 * terminating NUL.
 249 */
 250static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
 251{
 252        const char *objdir = get_object_directory();
 253        struct alternate_object_database *ent;
 254        struct alternate_object_database *alt;
 255        int pfxlen, entlen;
 256        struct strbuf pathbuf = STRBUF_INIT;
 257
 258        if (!is_absolute_path(entry) && relative_base) {
 259                strbuf_addstr(&pathbuf, real_path(relative_base));
 260                strbuf_addch(&pathbuf, '/');
 261        }
 262        strbuf_addstr(&pathbuf, entry);
 263
 264        normalize_path_copy(pathbuf.buf, pathbuf.buf);
 265
 266        pfxlen = strlen(pathbuf.buf);
 267
 268        /*
 269         * The trailing slash after the directory name is given by
 270         * this function at the end. Remove duplicates.
 271         */
 272        while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
 273                pfxlen -= 1;
 274
 275        entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
 276        ent = xmalloc(sizeof(*ent) + entlen);
 277        memcpy(ent->base, pathbuf.buf, pfxlen);
 278        strbuf_release(&pathbuf);
 279
 280        ent->name = ent->base + pfxlen + 1;
 281        ent->base[pfxlen + 3] = '/';
 282        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 283
 284        /* Detect cases where alternate disappeared */
 285        if (!is_directory(ent->base)) {
 286                error("object directory %s does not exist; "
 287                      "check .git/objects/info/alternates.",
 288                      ent->base);
 289                free(ent);
 290                return -1;
 291        }
 292
 293        /* Prevent the common mistake of listing the same
 294         * thing twice, or object directory itself.
 295         */
 296        for (alt = alt_odb_list; alt; alt = alt->next) {
 297                if (!memcmp(ent->base, alt->base, pfxlen)) {
 298                        free(ent);
 299                        return -1;
 300                }
 301        }
 302        if (!strcmp(ent->base, objdir)) {
 303                free(ent);
 304                return -1;
 305        }
 306
 307        /* add the alternate entry */
 308        *alt_odb_tail = ent;
 309        alt_odb_tail = &(ent->next);
 310        ent->next = NULL;
 311
 312        /* recursively add alternates */
 313        read_info_alternates(ent->base, depth + 1);
 314
 315        ent->base[pfxlen] = '/';
 316
 317        return 0;
 318}
 319
 320static void link_alt_odb_entries(const char *alt, int len, int sep,
 321                                 const char *relative_base, int depth)
 322{
 323        struct string_list entries = STRING_LIST_INIT_NODUP;
 324        char *alt_copy;
 325        int i;
 326
 327        if (depth > 5) {
 328                error("%s: ignoring alternate object stores, nesting too deep.",
 329                                relative_base);
 330                return;
 331        }
 332
 333        alt_copy = xmemdupz(alt, len);
 334        string_list_split_in_place(&entries, alt_copy, sep, -1);
 335        for (i = 0; i < entries.nr; i++) {
 336                const char *entry = entries.items[i].string;
 337                if (entry[0] == '\0' || entry[0] == '#')
 338                        continue;
 339                if (!is_absolute_path(entry) && depth) {
 340                        error("%s: ignoring relative alternate object store %s",
 341                                        relative_base, entry);
 342                } else {
 343                        link_alt_odb_entry(entry, relative_base, depth);
 344                }
 345        }
 346        string_list_clear(&entries, 0);
 347        free(alt_copy);
 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        const char alt_file_name[] = "info/alternates";
 356        /* Given that relative_base is no longer than PATH_MAX,
 357           ensure that "path" has enough space to append "/", the
 358           file name, "info/alternates", and a trailing NUL.  */
 359        char path[PATH_MAX + 1 + sizeof alt_file_name];
 360        int fd;
 361
 362        sprintf(path, "%s/%s", relative_base, alt_file_name);
 363        fd = git_open_noatime(path);
 364        if (fd < 0)
 365                return;
 366        if (fstat(fd, &st) || (st.st_size == 0)) {
 367                close(fd);
 368                return;
 369        }
 370        mapsz = xsize_t(st.st_size);
 371        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 372        close(fd);
 373
 374        link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
 375
 376        munmap(map, mapsz);
 377}
 378
 379void add_to_alternates_file(const char *reference)
 380{
 381        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
 382        int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
 383        char *alt = mkpath("%s\n", reference);
 384        write_or_die(fd, alt, strlen(alt));
 385        if (commit_lock_file(lock))
 386                die("could not close alternates file");
 387        if (alt_odb_tail)
 388                link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
 389}
 390
 391void foreach_alt_odb(alt_odb_fn fn, void *cb)
 392{
 393        struct alternate_object_database *ent;
 394
 395        prepare_alt_odb();
 396        for (ent = alt_odb_list; ent; ent = ent->next)
 397                if (fn(ent, cb))
 398                        return;
 399}
 400
 401void prepare_alt_odb(void)
 402{
 403        const char *alt;
 404
 405        if (alt_odb_tail)
 406                return;
 407
 408        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 409        if (!alt) alt = "";
 410
 411        alt_odb_tail = &alt_odb_list;
 412        link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
 413
 414        read_info_alternates(get_object_directory(), 0);
 415}
 416
 417static int has_loose_object_local(const unsigned char *sha1)
 418{
 419        char *name = sha1_file_name(sha1);
 420        return !access(name, F_OK);
 421}
 422
 423int has_loose_object_nonlocal(const unsigned char *sha1)
 424{
 425        struct alternate_object_database *alt;
 426        prepare_alt_odb();
 427        for (alt = alt_odb_list; alt; alt = alt->next) {
 428                fill_sha1_path(alt->name, sha1);
 429                if (!access(alt->base, F_OK))
 430                        return 1;
 431        }
 432        return 0;
 433}
 434
 435static int has_loose_object(const unsigned char *sha1)
 436{
 437        return has_loose_object_local(sha1) ||
 438               has_loose_object_nonlocal(sha1);
 439}
 440
 441static unsigned int pack_used_ctr;
 442static unsigned int pack_mmap_calls;
 443static unsigned int peak_pack_open_windows;
 444static unsigned int pack_open_windows;
 445static unsigned int pack_open_fds;
 446static unsigned int pack_max_fds;
 447static size_t peak_pack_mapped;
 448static size_t pack_mapped;
 449struct packed_git *packed_git;
 450
 451void pack_report(void)
 452{
 453        fprintf(stderr,
 454                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 455                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 456                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 457                sz_fmt(getpagesize()),
 458                sz_fmt(packed_git_window_size),
 459                sz_fmt(packed_git_limit));
 460        fprintf(stderr,
 461                "pack_report: pack_used_ctr            = %10u\n"
 462                "pack_report: pack_mmap_calls          = %10u\n"
 463                "pack_report: pack_open_windows        = %10u / %10u\n"
 464                "pack_report: pack_mapped              = "
 465                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 466                pack_used_ctr,
 467                pack_mmap_calls,
 468                pack_open_windows, peak_pack_open_windows,
 469                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
 470}
 471
 472static int check_packed_git_idx(const char *path,  struct packed_git *p)
 473{
 474        void *idx_map;
 475        struct pack_idx_header *hdr;
 476        size_t idx_size;
 477        uint32_t version, nr, i, *index;
 478        int fd = git_open_noatime(path);
 479        struct stat st;
 480
 481        if (fd < 0)
 482                return -1;
 483        if (fstat(fd, &st)) {
 484                close(fd);
 485                return -1;
 486        }
 487        idx_size = xsize_t(st.st_size);
 488        if (idx_size < 4 * 256 + 20 + 20) {
 489                close(fd);
 490                return error("index file %s is too small", path);
 491        }
 492        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 493        close(fd);
 494
 495        hdr = idx_map;
 496        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 497                version = ntohl(hdr->idx_version);
 498                if (version < 2 || version > 2) {
 499                        munmap(idx_map, idx_size);
 500                        return error("index file %s is version %"PRIu32
 501                                     " and is not supported by this binary"
 502                                     " (try upgrading GIT to a newer version)",
 503                                     path, version);
 504                }
 505        } else
 506                version = 1;
 507
 508        nr = 0;
 509        index = idx_map;
 510        if (version > 1)
 511                index += 2;  /* skip index header */
 512        for (i = 0; i < 256; i++) {
 513                uint32_t n = ntohl(index[i]);
 514                if (n < nr) {
 515                        munmap(idx_map, idx_size);
 516                        return error("non-monotonic index %s", path);
 517                }
 518                nr = n;
 519        }
 520
 521        if (version == 1) {
 522                /*
 523                 * Total size:
 524                 *  - 256 index entries 4 bytes each
 525                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 526                 *  - 20-byte SHA1 of the packfile
 527                 *  - 20-byte SHA1 file checksum
 528                 */
 529                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 530                        munmap(idx_map, idx_size);
 531                        return error("wrong index v1 file size in %s", path);
 532                }
 533        } else if (version == 2) {
 534                /*
 535                 * Minimum size:
 536                 *  - 8 bytes of header
 537                 *  - 256 index entries 4 bytes each
 538                 *  - 20-byte sha1 entry * nr
 539                 *  - 4-byte crc entry * nr
 540                 *  - 4-byte offset entry * nr
 541                 *  - 20-byte SHA1 of the packfile
 542                 *  - 20-byte SHA1 file checksum
 543                 * And after the 4-byte offset table might be a
 544                 * variable sized table containing 8-byte entries
 545                 * for offsets larger than 2^31.
 546                 */
 547                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 548                unsigned long max_size = min_size;
 549                if (nr)
 550                        max_size += (nr - 1)*8;
 551                if (idx_size < min_size || idx_size > max_size) {
 552                        munmap(idx_map, idx_size);
 553                        return error("wrong index v2 file size in %s", path);
 554                }
 555                if (idx_size != min_size &&
 556                    /*
 557                     * make sure we can deal with large pack offsets.
 558                     * 31-bit signed offset won't be enough, neither
 559                     * 32-bit unsigned one will be.
 560                     */
 561                    (sizeof(off_t) <= 4)) {
 562                        munmap(idx_map, idx_size);
 563                        return error("pack too large for current definition of off_t in %s", path);
 564                }
 565        }
 566
 567        p->index_version = version;
 568        p->index_data = idx_map;
 569        p->index_size = idx_size;
 570        p->num_objects = nr;
 571        return 0;
 572}
 573
 574int open_pack_index(struct packed_git *p)
 575{
 576        char *idx_name;
 577        int ret;
 578
 579        if (p->index_data)
 580                return 0;
 581
 582        idx_name = xstrdup(p->pack_name);
 583        strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
 584        ret = check_packed_git_idx(idx_name, p);
 585        free(idx_name);
 586        return ret;
 587}
 588
 589static void scan_windows(struct packed_git *p,
 590        struct packed_git **lru_p,
 591        struct pack_window **lru_w,
 592        struct pack_window **lru_l)
 593{
 594        struct pack_window *w, *w_l;
 595
 596        for (w_l = NULL, w = p->windows; w; w = w->next) {
 597                if (!w->inuse_cnt) {
 598                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 599                                *lru_p = p;
 600                                *lru_w = w;
 601                                *lru_l = w_l;
 602                        }
 603                }
 604                w_l = w;
 605        }
 606}
 607
 608static int unuse_one_window(struct packed_git *current, int keep_fd)
 609{
 610        struct packed_git *p, *lru_p = NULL;
 611        struct pack_window *lru_w = NULL, *lru_l = NULL;
 612
 613        if (current)
 614                scan_windows(current, &lru_p, &lru_w, &lru_l);
 615        for (p = packed_git; p; p = p->next)
 616                scan_windows(p, &lru_p, &lru_w, &lru_l);
 617        if (lru_p) {
 618                munmap(lru_w->base, lru_w->len);
 619                pack_mapped -= lru_w->len;
 620                if (lru_l)
 621                        lru_l->next = lru_w->next;
 622                else {
 623                        lru_p->windows = lru_w->next;
 624                        if (!lru_p->windows && lru_p->pack_fd != -1
 625                                && lru_p->pack_fd != keep_fd) {
 626                                close(lru_p->pack_fd);
 627                                pack_open_fds--;
 628                                lru_p->pack_fd = -1;
 629                        }
 630                }
 631                free(lru_w);
 632                pack_open_windows--;
 633                return 1;
 634        }
 635        return 0;
 636}
 637
 638void release_pack_memory(size_t need, int fd)
 639{
 640        size_t cur = pack_mapped;
 641        while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
 642                ; /* nothing */
 643}
 644
 645void *xmmap(void *start, size_t length,
 646        int prot, int flags, int fd, off_t offset)
 647{
 648        void *ret = mmap(start, length, prot, flags, fd, offset);
 649        if (ret == MAP_FAILED) {
 650                if (!length)
 651                        return NULL;
 652                release_pack_memory(length, fd);
 653                ret = mmap(start, length, prot, flags, fd, offset);
 654                if (ret == MAP_FAILED)
 655                        die_errno("Out of memory? mmap failed");
 656        }
 657        return ret;
 658}
 659
 660void close_pack_windows(struct packed_git *p)
 661{
 662        while (p->windows) {
 663                struct pack_window *w = p->windows;
 664
 665                if (w->inuse_cnt)
 666                        die("pack '%s' still has open windows to it",
 667                            p->pack_name);
 668                munmap(w->base, w->len);
 669                pack_mapped -= w->len;
 670                pack_open_windows--;
 671                p->windows = w->next;
 672                free(w);
 673        }
 674}
 675
 676void unuse_pack(struct pack_window **w_cursor)
 677{
 678        struct pack_window *w = *w_cursor;
 679        if (w) {
 680                w->inuse_cnt--;
 681                *w_cursor = NULL;
 682        }
 683}
 684
 685void close_pack_index(struct packed_git *p)
 686{
 687        if (p->index_data) {
 688                munmap((void *)p->index_data, p->index_size);
 689                p->index_data = NULL;
 690        }
 691}
 692
 693/*
 694 * This is used by git-repack in case a newly created pack happens to
 695 * contain the same set of objects as an existing one.  In that case
 696 * the resulting file might be different even if its name would be the
 697 * same.  It is best to close any reference to the old pack before it is
 698 * replaced on disk.  Of course no index pointers nor windows for given pack
 699 * must subsist at this point.  If ever objects from this pack are requested
 700 * again, the new version of the pack will be reinitialized through
 701 * reprepare_packed_git().
 702 */
 703void free_pack_by_name(const char *pack_name)
 704{
 705        struct packed_git *p, **pp = &packed_git;
 706
 707        while (*pp) {
 708                p = *pp;
 709                if (strcmp(pack_name, p->pack_name) == 0) {
 710                        clear_delta_base_cache();
 711                        close_pack_windows(p);
 712                        if (p->pack_fd != -1) {
 713                                close(p->pack_fd);
 714                                pack_open_fds--;
 715                        }
 716                        close_pack_index(p);
 717                        free(p->bad_object_sha1);
 718                        *pp = p->next;
 719                        if (last_found_pack == p)
 720                                last_found_pack = NULL;
 721                        free(p);
 722                        return;
 723                }
 724                pp = &p->next;
 725        }
 726}
 727
 728static unsigned int get_max_fd_limit(void)
 729{
 730#ifdef RLIMIT_NOFILE
 731        struct rlimit lim;
 732
 733        if (getrlimit(RLIMIT_NOFILE, &lim))
 734                die_errno("cannot get RLIMIT_NOFILE");
 735
 736        return lim.rlim_cur;
 737#elif defined(_SC_OPEN_MAX)
 738        return sysconf(_SC_OPEN_MAX);
 739#elif defined(OPEN_MAX)
 740        return OPEN_MAX;
 741#else
 742        return 1; /* see the caller ;-) */
 743#endif
 744}
 745
 746/*
 747 * Do not call this directly as this leaks p->pack_fd on error return;
 748 * call open_packed_git() instead.
 749 */
 750static int open_packed_git_1(struct packed_git *p)
 751{
 752        struct stat st;
 753        struct pack_header hdr;
 754        unsigned char sha1[20];
 755        unsigned char *idx_sha1;
 756        long fd_flag;
 757
 758        if (!p->index_data && open_pack_index(p))
 759                return error("packfile %s index unavailable", p->pack_name);
 760
 761        if (!pack_max_fds) {
 762                unsigned int max_fds = get_max_fd_limit();
 763
 764                /* Save 3 for stdin/stdout/stderr, 22 for work */
 765                if (25 < max_fds)
 766                        pack_max_fds = max_fds - 25;
 767                else
 768                        pack_max_fds = 1;
 769        }
 770
 771        while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
 772                ; /* nothing */
 773
 774        p->pack_fd = git_open_noatime(p->pack_name);
 775        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 776                return -1;
 777        pack_open_fds++;
 778
 779        /* If we created the struct before we had the pack we lack size. */
 780        if (!p->pack_size) {
 781                if (!S_ISREG(st.st_mode))
 782                        return error("packfile %s not a regular file", p->pack_name);
 783                p->pack_size = st.st_size;
 784        } else if (p->pack_size != st.st_size)
 785                return error("packfile %s size changed", p->pack_name);
 786
 787        /* We leave these file descriptors open with sliding mmap;
 788         * there is no point keeping them open across exec(), though.
 789         */
 790        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 791        if (fd_flag < 0)
 792                return error("cannot determine file descriptor flags");
 793        fd_flag |= FD_CLOEXEC;
 794        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 795                return error("cannot set FD_CLOEXEC");
 796
 797        /* Verify we recognize this pack file format. */
 798        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 799                return error("file %s is far too short to be a packfile", p->pack_name);
 800        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 801                return error("file %s is not a GIT packfile", p->pack_name);
 802        if (!pack_version_ok(hdr.hdr_version))
 803                return error("packfile %s is version %"PRIu32" and not"
 804                        " supported (try upgrading GIT to a newer version)",
 805                        p->pack_name, ntohl(hdr.hdr_version));
 806
 807        /* Verify the pack matches its index. */
 808        if (p->num_objects != ntohl(hdr.hdr_entries))
 809                return error("packfile %s claims to have %"PRIu32" objects"
 810                             " while index indicates %"PRIu32" objects",
 811                             p->pack_name, ntohl(hdr.hdr_entries),
 812                             p->num_objects);
 813        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 814                return error("end of packfile %s is unavailable", p->pack_name);
 815        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 816                return error("packfile %s signature is unavailable", p->pack_name);
 817        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 818        if (hashcmp(sha1, idx_sha1))
 819                return error("packfile %s does not match index", p->pack_name);
 820        return 0;
 821}
 822
 823static int open_packed_git(struct packed_git *p)
 824{
 825        if (!open_packed_git_1(p))
 826                return 0;
 827        if (p->pack_fd != -1) {
 828                close(p->pack_fd);
 829                pack_open_fds--;
 830                p->pack_fd = -1;
 831        }
 832        return -1;
 833}
 834
 835static int in_window(struct pack_window *win, off_t offset)
 836{
 837        /* We must promise at least 20 bytes (one hash) after the
 838         * offset is available from this window, otherwise the offset
 839         * is not actually in this window and a different window (which
 840         * has that one hash excess) must be used.  This is to support
 841         * the object header and delta base parsing routines below.
 842         */
 843        off_t win_off = win->offset;
 844        return win_off <= offset
 845                && (offset + 20) <= (win_off + win->len);
 846}
 847
 848unsigned char *use_pack(struct packed_git *p,
 849                struct pack_window **w_cursor,
 850                off_t offset,
 851                unsigned long *left)
 852{
 853        struct pack_window *win = *w_cursor;
 854
 855        /* Since packfiles end in a hash of their content and it's
 856         * pointless to ask for an offset into the middle of that
 857         * hash, and the in_window function above wouldn't match
 858         * don't allow an offset too close to the end of the file.
 859         */
 860        if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
 861                die("packfile %s cannot be accessed", p->pack_name);
 862        if (offset > (p->pack_size - 20))
 863                die("offset beyond end of packfile (truncated pack?)");
 864
 865        if (!win || !in_window(win, offset)) {
 866                if (win)
 867                        win->inuse_cnt--;
 868                for (win = p->windows; win; win = win->next) {
 869                        if (in_window(win, offset))
 870                                break;
 871                }
 872                if (!win) {
 873                        size_t window_align = packed_git_window_size / 2;
 874                        off_t len;
 875
 876                        if (p->pack_fd == -1 && open_packed_git(p))
 877                                die("packfile %s cannot be accessed", p->pack_name);
 878
 879                        win = xcalloc(1, sizeof(*win));
 880                        win->offset = (offset / window_align) * window_align;
 881                        len = p->pack_size - win->offset;
 882                        if (len > packed_git_window_size)
 883                                len = packed_git_window_size;
 884                        win->len = (size_t)len;
 885                        pack_mapped += win->len;
 886                        while (packed_git_limit < pack_mapped
 887                                && unuse_one_window(p, p->pack_fd))
 888                                ; /* nothing */
 889                        win->base = xmmap(NULL, win->len,
 890                                PROT_READ, MAP_PRIVATE,
 891                                p->pack_fd, win->offset);
 892                        if (win->base == MAP_FAILED)
 893                                die("packfile %s cannot be mapped: %s",
 894                                        p->pack_name,
 895                                        strerror(errno));
 896                        if (!win->offset && win->len == p->pack_size
 897                                && !p->do_not_close) {
 898                                close(p->pack_fd);
 899                                pack_open_fds--;
 900                                p->pack_fd = -1;
 901                        }
 902                        pack_mmap_calls++;
 903                        pack_open_windows++;
 904                        if (pack_mapped > peak_pack_mapped)
 905                                peak_pack_mapped = pack_mapped;
 906                        if (pack_open_windows > peak_pack_open_windows)
 907                                peak_pack_open_windows = pack_open_windows;
 908                        win->next = p->windows;
 909                        p->windows = win;
 910                }
 911        }
 912        if (win != *w_cursor) {
 913                win->last_used = pack_used_ctr++;
 914                win->inuse_cnt++;
 915                *w_cursor = win;
 916        }
 917        offset -= win->offset;
 918        if (left)
 919                *left = win->len - xsize_t(offset);
 920        return win->base + offset;
 921}
 922
 923static struct packed_git *alloc_packed_git(int extra)
 924{
 925        struct packed_git *p = xmalloc(sizeof(*p) + extra);
 926        memset(p, 0, sizeof(*p));
 927        p->pack_fd = -1;
 928        return p;
 929}
 930
 931static void try_to_free_pack_memory(size_t size)
 932{
 933        release_pack_memory(size, -1);
 934}
 935
 936struct packed_git *add_packed_git(const char *path, int path_len, int local)
 937{
 938        static int have_set_try_to_free_routine;
 939        struct stat st;
 940        struct packed_git *p = alloc_packed_git(path_len + 2);
 941
 942        if (!have_set_try_to_free_routine) {
 943                have_set_try_to_free_routine = 1;
 944                set_try_to_free_routine(try_to_free_pack_memory);
 945        }
 946
 947        /*
 948         * Make sure a corresponding .pack file exists and that
 949         * the index looks sane.
 950         */
 951        path_len -= strlen(".idx");
 952        if (path_len < 1) {
 953                free(p);
 954                return NULL;
 955        }
 956        memcpy(p->pack_name, path, path_len);
 957
 958        strcpy(p->pack_name + path_len, ".keep");
 959        if (!access(p->pack_name, F_OK))
 960                p->pack_keep = 1;
 961
 962        strcpy(p->pack_name + path_len, ".pack");
 963        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 964                free(p);
 965                return NULL;
 966        }
 967
 968        /* ok, it looks sane as far as we can check without
 969         * actually mapping the pack file.
 970         */
 971        p->pack_size = st.st_size;
 972        p->pack_local = local;
 973        p->mtime = st.st_mtime;
 974        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 975                hashclr(p->sha1);
 976        return p;
 977}
 978
 979struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
 980{
 981        const char *path = sha1_pack_name(sha1);
 982        struct packed_git *p = alloc_packed_git(strlen(path) + 1);
 983
 984        strcpy(p->pack_name, path);
 985        hashcpy(p->sha1, sha1);
 986        if (check_packed_git_idx(idx_path, p)) {
 987                free(p);
 988                return NULL;
 989        }
 990
 991        return p;
 992}
 993
 994void install_packed_git(struct packed_git *pack)
 995{
 996        if (pack->pack_fd != -1)
 997                pack_open_fds++;
 998
 999        pack->next = packed_git;
1000        packed_git = pack;
1001}
1002
1003static void prepare_packed_git_one(char *objdir, int local)
1004{
1005        /* Ensure that this buffer is large enough so that we can
1006           append "/pack/" without clobbering the stack even if
1007           strlen(objdir) were PATH_MAX.  */
1008        char path[PATH_MAX + 1 + 4 + 1 + 1];
1009        int len;
1010        DIR *dir;
1011        struct dirent *de;
1012
1013        sprintf(path, "%s/pack", objdir);
1014        len = strlen(path);
1015        dir = opendir(path);
1016        if (!dir) {
1017                if (errno != ENOENT)
1018                        error("unable to open object pack directory: %s: %s",
1019                              path, strerror(errno));
1020                return;
1021        }
1022        path[len++] = '/';
1023        while ((de = readdir(dir)) != NULL) {
1024                int namelen = strlen(de->d_name);
1025                struct packed_git *p;
1026
1027                if (!has_extension(de->d_name, ".idx"))
1028                        continue;
1029
1030                if (len + namelen + 1 > sizeof(path))
1031                        continue;
1032
1033                /* Don't reopen a pack we already have. */
1034                strcpy(path + len, de->d_name);
1035                for (p = packed_git; p; p = p->next) {
1036                        if (!memcmp(path, p->pack_name, len + namelen - 4))
1037                                break;
1038                }
1039                if (p)
1040                        continue;
1041                /* See if it really is a valid .idx file with corresponding
1042                 * .pack file that we can map.
1043                 */
1044                p = add_packed_git(path, len + namelen, local);
1045                if (!p)
1046                        continue;
1047                install_packed_git(p);
1048        }
1049        closedir(dir);
1050}
1051
1052static int sort_pack(const void *a_, const void *b_)
1053{
1054        struct packed_git *a = *((struct packed_git **)a_);
1055        struct packed_git *b = *((struct packed_git **)b_);
1056        int st;
1057
1058        /*
1059         * Local packs tend to contain objects specific to our
1060         * variant of the project than remote ones.  In addition,
1061         * remote ones could be on a network mounted filesystem.
1062         * Favor local ones for these reasons.
1063         */
1064        st = a->pack_local - b->pack_local;
1065        if (st)
1066                return -st;
1067
1068        /*
1069         * Younger packs tend to contain more recent objects,
1070         * and more recent objects tend to get accessed more
1071         * often.
1072         */
1073        if (a->mtime < b->mtime)
1074                return 1;
1075        else if (a->mtime == b->mtime)
1076                return 0;
1077        return -1;
1078}
1079
1080static void rearrange_packed_git(void)
1081{
1082        struct packed_git **ary, *p;
1083        int i, n;
1084
1085        for (n = 0, p = packed_git; p; p = p->next)
1086                n++;
1087        if (n < 2)
1088                return;
1089
1090        /* prepare an array of packed_git for easier sorting */
1091        ary = xcalloc(n, sizeof(struct packed_git *));
1092        for (n = 0, p = packed_git; p; p = p->next)
1093                ary[n++] = p;
1094
1095        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1096
1097        /* link them back again */
1098        for (i = 0; i < n - 1; i++)
1099                ary[i]->next = ary[i + 1];
1100        ary[n - 1]->next = NULL;
1101        packed_git = ary[0];
1102
1103        free(ary);
1104}
1105
1106static int prepare_packed_git_run_once = 0;
1107void prepare_packed_git(void)
1108{
1109        struct alternate_object_database *alt;
1110
1111        if (prepare_packed_git_run_once)
1112                return;
1113        prepare_packed_git_one(get_object_directory(), 1);
1114        prepare_alt_odb();
1115        for (alt = alt_odb_list; alt; alt = alt->next) {
1116                alt->name[-1] = 0;
1117                prepare_packed_git_one(alt->base, 0);
1118                alt->name[-1] = '/';
1119        }
1120        rearrange_packed_git();
1121        prepare_packed_git_run_once = 1;
1122}
1123
1124void reprepare_packed_git(void)
1125{
1126        discard_revindex();
1127        prepare_packed_git_run_once = 0;
1128        prepare_packed_git();
1129}
1130
1131static void mark_bad_packed_object(struct packed_git *p,
1132                                   const unsigned char *sha1)
1133{
1134        unsigned i;
1135        for (i = 0; i < p->num_bad_objects; i++)
1136                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1137                        return;
1138        p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1139        hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1140        p->num_bad_objects++;
1141}
1142
1143static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1144{
1145        struct packed_git *p;
1146        unsigned i;
1147
1148        for (p = packed_git; p; p = p->next)
1149                for (i = 0; i < p->num_bad_objects; i++)
1150                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1151                                return p;
1152        return NULL;
1153}
1154
1155/*
1156 * With an in-core object data in "map", rehash it to make sure the
1157 * object name actually matches "sha1" to detect object corruption.
1158 * With "map" == NULL, try reading the object named with "sha1" using
1159 * the streaming interface and rehash it to do the same.
1160 */
1161int check_sha1_signature(const unsigned char *sha1, void *map,
1162                         unsigned long size, const char *type)
1163{
1164        unsigned char real_sha1[20];
1165        enum object_type obj_type;
1166        struct git_istream *st;
1167        git_SHA_CTX c;
1168        char hdr[32];
1169        int hdrlen;
1170
1171        if (map) {
1172                hash_sha1_file(map, size, type, real_sha1);
1173                return hashcmp(sha1, real_sha1) ? -1 : 0;
1174        }
1175
1176        st = open_istream(sha1, &obj_type, &size, NULL);
1177        if (!st)
1178                return -1;
1179
1180        /* Generate the header */
1181        hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1182
1183        /* Sha1.. */
1184        git_SHA1_Init(&c);
1185        git_SHA1_Update(&c, hdr, hdrlen);
1186        for (;;) {
1187                char buf[1024 * 16];
1188                ssize_t readlen = read_istream(st, buf, sizeof(buf));
1189
1190                if (!readlen)
1191                        break;
1192                git_SHA1_Update(&c, buf, readlen);
1193        }
1194        git_SHA1_Final(real_sha1, &c);
1195        close_istream(st);
1196        return hashcmp(sha1, real_sha1) ? -1 : 0;
1197}
1198
1199static int git_open_noatime(const char *name)
1200{
1201        static int sha1_file_open_flag = O_NOATIME;
1202
1203        for (;;) {
1204                int fd = open(name, O_RDONLY | sha1_file_open_flag);
1205                if (fd >= 0)
1206                        return fd;
1207
1208                /* Might the failure be due to O_NOATIME? */
1209                if (errno != ENOENT && sha1_file_open_flag) {
1210                        sha1_file_open_flag = 0;
1211                        continue;
1212                }
1213
1214                return -1;
1215        }
1216}
1217
1218static int open_sha1_file(const unsigned char *sha1)
1219{
1220        int fd;
1221        char *name = sha1_file_name(sha1);
1222        struct alternate_object_database *alt;
1223
1224        fd = git_open_noatime(name);
1225        if (fd >= 0)
1226                return fd;
1227
1228        prepare_alt_odb();
1229        errno = ENOENT;
1230        for (alt = alt_odb_list; alt; alt = alt->next) {
1231                name = alt->name;
1232                fill_sha1_path(name, sha1);
1233                fd = git_open_noatime(alt->base);
1234                if (fd >= 0)
1235                        return fd;
1236        }
1237        return -1;
1238}
1239
1240void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1241{
1242        void *map;
1243        int fd;
1244
1245        fd = open_sha1_file(sha1);
1246        map = NULL;
1247        if (fd >= 0) {
1248                struct stat st;
1249
1250                if (!fstat(fd, &st)) {
1251                        *size = xsize_t(st.st_size);
1252                        if (!*size) {
1253                                /* mmap() is forbidden on empty files */
1254                                error("object file %s is empty", sha1_file_name(sha1));
1255                                return NULL;
1256                        }
1257                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1258                }
1259                close(fd);
1260        }
1261        return map;
1262}
1263
1264/*
1265 * There used to be a second loose object header format which
1266 * was meant to mimic the in-pack format, allowing for direct
1267 * copy of the object data.  This format turned up not to be
1268 * really worth it and we no longer write loose objects in that
1269 * format.
1270 */
1271static int experimental_loose_object(unsigned char *map)
1272{
1273        unsigned int word;
1274
1275        /*
1276         * We must determine if the buffer contains the standard
1277         * zlib-deflated stream or the experimental format based
1278         * on the in-pack object format. Compare the header byte
1279         * for each format:
1280         *
1281         * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1282         * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1283         *
1284         * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1285         * in standard loose-object format, UNLESS it is a Git-pack
1286         * format object *exactly* 8 bytes in size when inflated.
1287         *
1288         * However, RFC1950 also specifies that the 1st 16-bit word
1289         * must be divisible by 31 - this checksum tells us our buffer
1290         * is in the standard format, giving a false positive only if
1291         * the 1st word of the Git-pack format object happens to be
1292         * divisible by 31, ie:
1293         *      ((byte0 * 256) + byte1) % 31 = 0
1294         *   =>        0ttt10000www1000 % 31 = 0
1295         *
1296         * As it happens, this case can only arise for www=3 & ttt=1
1297         * - ie, a Commit object, which would have to be 8 bytes in
1298         * size. As no Commit can be that small, we find that the
1299         * combination of these two criteria (bitmask & checksum)
1300         * can always correctly determine the buffer format.
1301         */
1302        word = (map[0] << 8) + map[1];
1303        if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1304                return 0;
1305        else
1306                return 1;
1307}
1308
1309unsigned long unpack_object_header_buffer(const unsigned char *buf,
1310                unsigned long len, enum object_type *type, unsigned long *sizep)
1311{
1312        unsigned shift;
1313        unsigned long size, c;
1314        unsigned long used = 0;
1315
1316        c = buf[used++];
1317        *type = (c >> 4) & 7;
1318        size = c & 15;
1319        shift = 4;
1320        while (c & 0x80) {
1321                if (len <= used || bitsizeof(long) <= shift) {
1322                        error("bad object header");
1323                        size = used = 0;
1324                        break;
1325                }
1326                c = buf[used++];
1327                size += (c & 0x7f) << shift;
1328                shift += 7;
1329        }
1330        *sizep = size;
1331        return used;
1332}
1333
1334int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1335{
1336        unsigned long size, used;
1337        static const char valid_loose_object_type[8] = {
1338                0, /* OBJ_EXT */
1339                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1340                0, /* "delta" and others are invalid in a loose object */
1341        };
1342        enum object_type type;
1343
1344        /* Get the data stream */
1345        memset(stream, 0, sizeof(*stream));
1346        stream->next_in = map;
1347        stream->avail_in = mapsize;
1348        stream->next_out = buffer;
1349        stream->avail_out = bufsiz;
1350
1351        if (experimental_loose_object(map)) {
1352                /*
1353                 * The old experimental format we no longer produce;
1354                 * we can still read it.
1355                 */
1356                used = unpack_object_header_buffer(map, mapsize, &type, &size);
1357                if (!used || !valid_loose_object_type[type])
1358                        return -1;
1359                map += used;
1360                mapsize -= used;
1361
1362                /* Set up the stream for the rest.. */
1363                stream->next_in = map;
1364                stream->avail_in = mapsize;
1365                git_inflate_init(stream);
1366
1367                /* And generate the fake traditional header */
1368                stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1369                                                 typename(type), size);
1370                return 0;
1371        }
1372        git_inflate_init(stream);
1373        return git_inflate(stream, 0);
1374}
1375
1376static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1377{
1378        int bytes = strlen(buffer) + 1;
1379        unsigned char *buf = xmallocz(size);
1380        unsigned long n;
1381        int status = Z_OK;
1382
1383        n = stream->total_out - bytes;
1384        if (n > size)
1385                n = size;
1386        memcpy(buf, (char *) buffer + bytes, n);
1387        bytes = n;
1388        if (bytes <= size) {
1389                /*
1390                 * The above condition must be (bytes <= size), not
1391                 * (bytes < size).  In other words, even though we
1392                 * expect no more output and set avail_out to zero,
1393                 * the input zlib stream may have bytes that express
1394                 * "this concludes the stream", and we *do* want to
1395                 * eat that input.
1396                 *
1397                 * Otherwise we would not be able to test that we
1398                 * consumed all the input to reach the expected size;
1399                 * we also want to check that zlib tells us that all
1400                 * went well with status == Z_STREAM_END at the end.
1401                 */
1402                stream->next_out = buf + bytes;
1403                stream->avail_out = size - bytes;
1404                while (status == Z_OK)
1405                        status = git_inflate(stream, Z_FINISH);
1406        }
1407        if (status == Z_STREAM_END && !stream->avail_in) {
1408                git_inflate_end(stream);
1409                return buf;
1410        }
1411
1412        if (status < 0)
1413                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1414        else if (stream->avail_in)
1415                error("garbage at end of loose object '%s'",
1416                      sha1_to_hex(sha1));
1417        free(buf);
1418        return NULL;
1419}
1420
1421/*
1422 * We used to just use "sscanf()", but that's actually way
1423 * too permissive for what we want to check. So do an anal
1424 * object header parse by hand.
1425 */
1426int parse_sha1_header(const char *hdr, unsigned long *sizep)
1427{
1428        char type[10];
1429        int i;
1430        unsigned long size;
1431
1432        /*
1433         * The type can be at most ten bytes (including the
1434         * terminating '\0' that we add), and is followed by
1435         * a space.
1436         */
1437        i = 0;
1438        for (;;) {
1439                char c = *hdr++;
1440                if (c == ' ')
1441                        break;
1442                type[i++] = c;
1443                if (i >= sizeof(type))
1444                        return -1;
1445        }
1446        type[i] = 0;
1447
1448        /*
1449         * The length must follow immediately, and be in canonical
1450         * decimal format (ie "010" is not valid).
1451         */
1452        size = *hdr++ - '0';
1453        if (size > 9)
1454                return -1;
1455        if (size) {
1456                for (;;) {
1457                        unsigned long c = *hdr - '0';
1458                        if (c > 9)
1459                                break;
1460                        hdr++;
1461                        size = size * 10 + c;
1462                }
1463        }
1464        *sizep = size;
1465
1466        /*
1467         * The length must be followed by a zero byte
1468         */
1469        return *hdr ? -1 : type_from_string(type);
1470}
1471
1472static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1473{
1474        int ret;
1475        git_zstream stream;
1476        char hdr[8192];
1477
1478        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1479        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1480                return NULL;
1481
1482        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1483}
1484
1485unsigned long get_size_from_delta(struct packed_git *p,
1486                                  struct pack_window **w_curs,
1487                                  off_t curpos)
1488{
1489        const unsigned char *data;
1490        unsigned char delta_head[20], *in;
1491        git_zstream stream;
1492        int st;
1493
1494        memset(&stream, 0, sizeof(stream));
1495        stream.next_out = delta_head;
1496        stream.avail_out = sizeof(delta_head);
1497
1498        git_inflate_init(&stream);
1499        do {
1500                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1501                stream.next_in = in;
1502                st = git_inflate(&stream, Z_FINISH);
1503                curpos += stream.next_in - in;
1504        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1505                 stream.total_out < sizeof(delta_head));
1506        git_inflate_end(&stream);
1507        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1508                error("delta data unpack-initial failed");
1509                return 0;
1510        }
1511
1512        /* Examine the initial part of the delta to figure out
1513         * the result size.
1514         */
1515        data = delta_head;
1516
1517        /* ignore base size */
1518        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1519
1520        /* Read the result size */
1521        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1522}
1523
1524static off_t get_delta_base(struct packed_git *p,
1525                                    struct pack_window **w_curs,
1526                                    off_t *curpos,
1527                                    enum object_type type,
1528                                    off_t delta_obj_offset)
1529{
1530        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1531        off_t base_offset;
1532
1533        /* use_pack() assured us we have [base_info, base_info + 20)
1534         * as a range that we can look at without walking off the
1535         * end of the mapped window.  Its actually the hash size
1536         * that is assured.  An OFS_DELTA longer than the hash size
1537         * is stupid, as then a REF_DELTA would be smaller to store.
1538         */
1539        if (type == OBJ_OFS_DELTA) {
1540                unsigned used = 0;
1541                unsigned char c = base_info[used++];
1542                base_offset = c & 127;
1543                while (c & 128) {
1544                        base_offset += 1;
1545                        if (!base_offset || MSB(base_offset, 7))
1546                                return 0;  /* overflow */
1547                        c = base_info[used++];
1548                        base_offset = (base_offset << 7) + (c & 127);
1549                }
1550                base_offset = delta_obj_offset - base_offset;
1551                if (base_offset <= 0 || base_offset >= delta_obj_offset)
1552                        return 0;  /* out of bound */
1553                *curpos += used;
1554        } else if (type == OBJ_REF_DELTA) {
1555                /* The base entry _must_ be in the same pack */
1556                base_offset = find_pack_entry_one(base_info, p);
1557                *curpos += 20;
1558        } else
1559                die("I am totally screwed");
1560        return base_offset;
1561}
1562
1563int unpack_object_header(struct packed_git *p,
1564                         struct pack_window **w_curs,
1565                         off_t *curpos,
1566                         unsigned long *sizep)
1567{
1568        unsigned char *base;
1569        unsigned long left;
1570        unsigned long used;
1571        enum object_type type;
1572
1573        /* use_pack() assures us we have [base, base + 20) available
1574         * as a range that we can look at.  (Its actually the hash
1575         * size that is assured.)  With our object header encoding
1576         * the maximum deflated object size is 2^137, which is just
1577         * insane, so we know won't exceed what we have been given.
1578         */
1579        base = use_pack(p, w_curs, *curpos, &left);
1580        used = unpack_object_header_buffer(base, left, &type, sizep);
1581        if (!used) {
1582                type = OBJ_BAD;
1583        } else
1584                *curpos += used;
1585
1586        return type;
1587}
1588
1589static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1590{
1591        int type;
1592        struct revindex_entry *revidx;
1593        const unsigned char *sha1;
1594        revidx = find_pack_revindex(p, obj_offset);
1595        if (!revidx)
1596                return OBJ_BAD;
1597        sha1 = nth_packed_object_sha1(p, revidx->nr);
1598        mark_bad_packed_object(p, sha1);
1599        type = sha1_object_info(sha1, NULL);
1600        if (type <= OBJ_NONE)
1601                return OBJ_BAD;
1602        return type;
1603}
1604
1605
1606#define POI_STACK_PREALLOC 64
1607
1608static int packed_object_info(struct packed_git *p, off_t obj_offset,
1609                              unsigned long *sizep, int *rtype)
1610{
1611        struct pack_window *w_curs = NULL;
1612        unsigned long size;
1613        off_t curpos = obj_offset;
1614        enum object_type type;
1615        off_t small_poi_stack[POI_STACK_PREALLOC];
1616        off_t *poi_stack = small_poi_stack;
1617        int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1618
1619        type = unpack_object_header(p, &w_curs, &curpos, &size);
1620
1621        if (rtype)
1622                *rtype = type; /* representation type */
1623
1624        if (sizep) {
1625                if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1626                        off_t tmp_pos = curpos;
1627                        off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1628                                                           type, obj_offset);
1629                        if (!base_offset) {
1630                                type = OBJ_BAD;
1631                                goto out;
1632                        }
1633                        *sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1634                        if (*sizep == 0) {
1635                                type = OBJ_BAD;
1636                                goto out;
1637                        }
1638                } else {
1639                        *sizep = size;
1640                }
1641        }
1642
1643        while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1644                off_t base_offset;
1645                /* Push the object we're going to leave behind */
1646                if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1647                        poi_stack_alloc = alloc_nr(poi_stack_nr);
1648                        poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1649                        memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1650                } else {
1651                        ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1652                }
1653                poi_stack[poi_stack_nr++] = obj_offset;
1654                /* If parsing the base offset fails, just unwind */
1655                base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1656                if (!base_offset)
1657                        goto unwind;
1658                curpos = obj_offset = base_offset;
1659                type = unpack_object_header(p, &w_curs, &curpos, &size);
1660                if (type <= OBJ_NONE) {
1661                        /* If getting the base itself fails, we first
1662                         * retry the base, otherwise unwind */
1663                        type = retry_bad_packed_offset(p, base_offset);
1664                        if (type > OBJ_NONE)
1665                                goto out;
1666                        goto unwind;
1667                }
1668        }
1669
1670        switch (type) {
1671        case OBJ_BAD:
1672        case OBJ_COMMIT:
1673        case OBJ_TREE:
1674        case OBJ_BLOB:
1675        case OBJ_TAG:
1676                break;
1677        default:
1678                error("unknown object type %i at offset %"PRIuMAX" in %s",
1679                      type, (uintmax_t)obj_offset, p->pack_name);
1680                type = OBJ_BAD;
1681        }
1682
1683out:
1684        if (poi_stack != small_poi_stack)
1685                free(poi_stack);
1686        unuse_pack(&w_curs);
1687        return type;
1688
1689unwind:
1690        while (poi_stack_nr) {
1691                obj_offset = poi_stack[--poi_stack_nr];
1692                type = retry_bad_packed_offset(p, obj_offset);
1693                if (type > OBJ_NONE)
1694                        goto out;
1695        }
1696        type = OBJ_BAD;
1697        goto out;
1698}
1699
1700static void *unpack_compressed_entry(struct packed_git *p,
1701                                    struct pack_window **w_curs,
1702                                    off_t curpos,
1703                                    unsigned long size)
1704{
1705        int st;
1706        git_zstream stream;
1707        unsigned char *buffer, *in;
1708
1709        buffer = xmallocz(size);
1710        memset(&stream, 0, sizeof(stream));
1711        stream.next_out = buffer;
1712        stream.avail_out = size + 1;
1713
1714        git_inflate_init(&stream);
1715        do {
1716                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1717                stream.next_in = in;
1718                st = git_inflate(&stream, Z_FINISH);
1719                if (!stream.avail_out)
1720                        break; /* the payload is larger than it should be */
1721                curpos += stream.next_in - in;
1722        } while (st == Z_OK || st == Z_BUF_ERROR);
1723        git_inflate_end(&stream);
1724        if ((st != Z_STREAM_END) || stream.total_out != size) {
1725                free(buffer);
1726                return NULL;
1727        }
1728
1729        return buffer;
1730}
1731
1732#define MAX_DELTA_CACHE (256)
1733
1734static size_t delta_base_cached;
1735
1736static struct delta_base_cache_lru_list {
1737        struct delta_base_cache_lru_list *prev;
1738        struct delta_base_cache_lru_list *next;
1739} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1740
1741static struct delta_base_cache_entry {
1742        struct delta_base_cache_lru_list lru;
1743        void *data;
1744        struct packed_git *p;
1745        off_t base_offset;
1746        unsigned long size;
1747        enum object_type type;
1748} delta_base_cache[MAX_DELTA_CACHE];
1749
1750static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1751{
1752        unsigned long hash;
1753
1754        hash = (unsigned long)p + (unsigned long)base_offset;
1755        hash += (hash >> 8) + (hash >> 16);
1756        return hash % MAX_DELTA_CACHE;
1757}
1758
1759static struct delta_base_cache_entry *
1760get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1761{
1762        unsigned long hash = pack_entry_hash(p, base_offset);
1763        return delta_base_cache + hash;
1764}
1765
1766static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1767                                     struct packed_git *p, off_t base_offset)
1768{
1769        return (ent->data && ent->p == p && ent->base_offset == base_offset);
1770}
1771
1772static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1773{
1774        struct delta_base_cache_entry *ent;
1775        ent = get_delta_base_cache_entry(p, base_offset);
1776        return eq_delta_base_cache_entry(ent, p, base_offset);
1777}
1778
1779static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1780{
1781        ent->data = NULL;
1782        ent->lru.next->prev = ent->lru.prev;
1783        ent->lru.prev->next = ent->lru.next;
1784        delta_base_cached -= ent->size;
1785}
1786
1787static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1788        unsigned long *base_size, enum object_type *type, int keep_cache)
1789{
1790        struct delta_base_cache_entry *ent;
1791        void *ret;
1792
1793        ent = get_delta_base_cache_entry(p, base_offset);
1794
1795        if (!eq_delta_base_cache_entry(ent, p, base_offset))
1796                return unpack_entry(p, base_offset, type, base_size);
1797
1798        ret = ent->data;
1799
1800        if (!keep_cache)
1801                clear_delta_base_cache_entry(ent);
1802        else
1803                ret = xmemdupz(ent->data, ent->size);
1804        *type = ent->type;
1805        *base_size = ent->size;
1806        return ret;
1807}
1808
1809static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1810{
1811        if (ent->data) {
1812                free(ent->data);
1813                ent->data = NULL;
1814                ent->lru.next->prev = ent->lru.prev;
1815                ent->lru.prev->next = ent->lru.next;
1816                delta_base_cached -= ent->size;
1817        }
1818}
1819
1820void clear_delta_base_cache(void)
1821{
1822        unsigned long p;
1823        for (p = 0; p < MAX_DELTA_CACHE; p++)
1824                release_delta_base_cache(&delta_base_cache[p]);
1825}
1826
1827static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1828        void *base, unsigned long base_size, enum object_type type)
1829{
1830        unsigned long hash = pack_entry_hash(p, base_offset);
1831        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1832        struct delta_base_cache_lru_list *lru;
1833
1834        release_delta_base_cache(ent);
1835        delta_base_cached += base_size;
1836
1837        for (lru = delta_base_cache_lru.next;
1838             delta_base_cached > delta_base_cache_limit
1839             && lru != &delta_base_cache_lru;
1840             lru = lru->next) {
1841                struct delta_base_cache_entry *f = (void *)lru;
1842                if (f->type == OBJ_BLOB)
1843                        release_delta_base_cache(f);
1844        }
1845        for (lru = delta_base_cache_lru.next;
1846             delta_base_cached > delta_base_cache_limit
1847             && lru != &delta_base_cache_lru;
1848             lru = lru->next) {
1849                struct delta_base_cache_entry *f = (void *)lru;
1850                release_delta_base_cache(f);
1851        }
1852
1853        ent->p = p;
1854        ent->base_offset = base_offset;
1855        ent->type = type;
1856        ent->data = base;
1857        ent->size = base_size;
1858        ent->lru.next = &delta_base_cache_lru;
1859        ent->lru.prev = delta_base_cache_lru.prev;
1860        delta_base_cache_lru.prev->next = &ent->lru;
1861        delta_base_cache_lru.prev = &ent->lru;
1862}
1863
1864static void *read_object(const unsigned char *sha1, enum object_type *type,
1865                         unsigned long *size);
1866
1867static void *unpack_delta_entry(struct packed_git *p,
1868                                struct pack_window **w_curs,
1869                                off_t curpos,
1870                                unsigned long delta_size,
1871                                off_t obj_offset,
1872                                enum object_type *type,
1873                                unsigned long *sizep)
1874{
1875        void *delta_data, *result, *base;
1876        unsigned long base_size;
1877        off_t base_offset;
1878
1879        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1880        if (!base_offset) {
1881                error("failed to validate delta base reference "
1882                      "at offset %"PRIuMAX" from %s",
1883                      (uintmax_t)curpos, p->pack_name);
1884                return NULL;
1885        }
1886        unuse_pack(w_curs);
1887        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1888        if (!base) {
1889                /*
1890                 * We're probably in deep shit, but let's try to fetch
1891                 * the required base anyway from another pack or loose.
1892                 * This is costly but should happen only in the presence
1893                 * of a corrupted pack, and is better than failing outright.
1894                 */
1895                struct revindex_entry *revidx;
1896                const unsigned char *base_sha1;
1897                revidx = find_pack_revindex(p, base_offset);
1898                if (!revidx)
1899                        return NULL;
1900                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1901                error("failed to read delta base object %s"
1902                      " at offset %"PRIuMAX" from %s",
1903                      sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1904                      p->pack_name);
1905                mark_bad_packed_object(p, base_sha1);
1906                base = read_object(base_sha1, type, &base_size);
1907                if (!base)
1908                        return NULL;
1909        }
1910
1911        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1912        if (!delta_data) {
1913                error("failed to unpack compressed delta "
1914                      "at offset %"PRIuMAX" from %s",
1915                      (uintmax_t)curpos, p->pack_name);
1916                free(base);
1917                return NULL;
1918        }
1919        result = patch_delta(base, base_size,
1920                             delta_data, delta_size,
1921                             sizep);
1922        if (!result)
1923                die("failed to apply delta");
1924        free(delta_data);
1925        add_delta_base_cache(p, base_offset, base, base_size, *type);
1926        return result;
1927}
1928
1929static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1930{
1931        static FILE *log_file;
1932
1933        if (!log_file) {
1934                log_file = fopen(log_pack_access, "w");
1935                if (!log_file) {
1936                        error("cannot open pack access log '%s' for writing: %s",
1937                              log_pack_access, strerror(errno));
1938                        log_pack_access = NULL;
1939                        return;
1940                }
1941        }
1942        fprintf(log_file, "%s %"PRIuMAX"\n",
1943                p->pack_name, (uintmax_t)obj_offset);
1944        fflush(log_file);
1945}
1946
1947int do_check_packed_object_crc;
1948
1949void *unpack_entry(struct packed_git *p, off_t obj_offset,
1950                   enum object_type *type, unsigned long *sizep)
1951{
1952        struct pack_window *w_curs = NULL;
1953        off_t curpos = obj_offset;
1954        void *data;
1955
1956        if (log_pack_access)
1957                write_pack_access_log(p, obj_offset);
1958
1959        if (do_check_packed_object_crc && p->index_version > 1) {
1960                struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1961                unsigned long len = revidx[1].offset - obj_offset;
1962                if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1963                        const unsigned char *sha1 =
1964                                nth_packed_object_sha1(p, revidx->nr);
1965                        error("bad packed object CRC for %s",
1966                              sha1_to_hex(sha1));
1967                        mark_bad_packed_object(p, sha1);
1968                        unuse_pack(&w_curs);
1969                        return NULL;
1970                }
1971        }
1972
1973        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1974        switch (*type) {
1975        case OBJ_OFS_DELTA:
1976        case OBJ_REF_DELTA:
1977                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1978                                          obj_offset, type, sizep);
1979                break;
1980        case OBJ_COMMIT:
1981        case OBJ_TREE:
1982        case OBJ_BLOB:
1983        case OBJ_TAG:
1984                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1985                break;
1986        default:
1987                data = NULL;
1988                error("unknown object type %i at offset %"PRIuMAX" in %s",
1989                      *type, (uintmax_t)obj_offset, p->pack_name);
1990        }
1991        unuse_pack(&w_curs);
1992        return data;
1993}
1994
1995const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1996                                            uint32_t n)
1997{
1998        const unsigned char *index = p->index_data;
1999        if (!index) {
2000                if (open_pack_index(p))
2001                        return NULL;
2002                index = p->index_data;
2003        }
2004        if (n >= p->num_objects)
2005                return NULL;
2006        index += 4 * 256;
2007        if (p->index_version == 1) {
2008                return index + 24 * n + 4;
2009        } else {
2010                index += 8;
2011                return index + 20 * n;
2012        }
2013}
2014
2015off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2016{
2017        const unsigned char *index = p->index_data;
2018        index += 4 * 256;
2019        if (p->index_version == 1) {
2020                return ntohl(*((uint32_t *)(index + 24 * n)));
2021        } else {
2022                uint32_t off;
2023                index += 8 + p->num_objects * (20 + 4);
2024                off = ntohl(*((uint32_t *)(index + 4 * n)));
2025                if (!(off & 0x80000000))
2026                        return off;
2027                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2028                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2029                                   ntohl(*((uint32_t *)(index + 4)));
2030        }
2031}
2032
2033off_t find_pack_entry_one(const unsigned char *sha1,
2034                                  struct packed_git *p)
2035{
2036        const uint32_t *level1_ofs = p->index_data;
2037        const unsigned char *index = p->index_data;
2038        unsigned hi, lo, stride;
2039        static int use_lookup = -1;
2040        static int debug_lookup = -1;
2041
2042        if (debug_lookup < 0)
2043                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2044
2045        if (!index) {
2046                if (open_pack_index(p))
2047                        return 0;
2048                level1_ofs = p->index_data;
2049                index = p->index_data;
2050        }
2051        if (p->index_version > 1) {
2052                level1_ofs += 2;
2053                index += 8;
2054        }
2055        index += 4 * 256;
2056        hi = ntohl(level1_ofs[*sha1]);
2057        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2058        if (p->index_version > 1) {
2059                stride = 20;
2060        } else {
2061                stride = 24;
2062                index += 4;
2063        }
2064
2065        if (debug_lookup)
2066                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2067                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2068
2069        if (use_lookup < 0)
2070                use_lookup = !!getenv("GIT_USE_LOOKUP");
2071        if (use_lookup) {
2072                int pos = sha1_entry_pos(index, stride, 0,
2073                                         lo, hi, p->num_objects, sha1);
2074                if (pos < 0)
2075                        return 0;
2076                return nth_packed_object_offset(p, pos);
2077        }
2078
2079        do {
2080                unsigned mi = (lo + hi) / 2;
2081                int cmp = hashcmp(index + mi * stride, sha1);
2082
2083                if (debug_lookup)
2084                        printf("lo %u hi %u rg %u mi %u\n",
2085                               lo, hi, hi - lo, mi);
2086                if (!cmp)
2087                        return nth_packed_object_offset(p, mi);
2088                if (cmp > 0)
2089                        hi = mi;
2090                else
2091                        lo = mi+1;
2092        } while (lo < hi);
2093        return 0;
2094}
2095
2096int is_pack_valid(struct packed_git *p)
2097{
2098        /* An already open pack is known to be valid. */
2099        if (p->pack_fd != -1)
2100                return 1;
2101
2102        /* If the pack has one window completely covering the
2103         * file size, the pack is known to be valid even if
2104         * the descriptor is not currently open.
2105         */
2106        if (p->windows) {
2107                struct pack_window *w = p->windows;
2108
2109                if (!w->offset && w->len == p->pack_size)
2110                        return 1;
2111        }
2112
2113        /* Force the pack to open to prove its valid. */
2114        return !open_packed_git(p);
2115}
2116
2117static int fill_pack_entry(const unsigned char *sha1,
2118                           struct pack_entry *e,
2119                           struct packed_git *p)
2120{
2121        off_t offset;
2122
2123        if (p->num_bad_objects) {
2124                unsigned i;
2125                for (i = 0; i < p->num_bad_objects; i++)
2126                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2127                                return 0;
2128        }
2129
2130        offset = find_pack_entry_one(sha1, p);
2131        if (!offset)
2132                return 0;
2133
2134        /*
2135         * We are about to tell the caller where they can locate the
2136         * requested object.  We better make sure the packfile is
2137         * still here and can be accessed before supplying that
2138         * answer, as it may have been deleted since the index was
2139         * loaded!
2140         */
2141        if (!is_pack_valid(p)) {
2142                warning("packfile %s cannot be accessed", p->pack_name);
2143                return 0;
2144        }
2145        e->offset = offset;
2146        e->p = p;
2147        hashcpy(e->sha1, sha1);
2148        return 1;
2149}
2150
2151static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2152{
2153        struct packed_git *p;
2154
2155        prepare_packed_git();
2156        if (!packed_git)
2157                return 0;
2158
2159        if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2160                return 1;
2161
2162        for (p = packed_git; p; p = p->next) {
2163                if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2164                        continue;
2165
2166                last_found_pack = p;
2167                return 1;
2168        }
2169        return 0;
2170}
2171
2172struct packed_git *find_sha1_pack(const unsigned char *sha1,
2173                                  struct packed_git *packs)
2174{
2175        struct packed_git *p;
2176
2177        for (p = packs; p; p = p->next) {
2178                if (find_pack_entry_one(sha1, p))
2179                        return p;
2180        }
2181        return NULL;
2182
2183}
2184
2185static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2186{
2187        int status;
2188        unsigned long mapsize, size;
2189        void *map;
2190        git_zstream stream;
2191        char hdr[32];
2192
2193        map = map_sha1_file(sha1, &mapsize);
2194        if (!map)
2195                return error("unable to find %s", sha1_to_hex(sha1));
2196        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2197                status = error("unable to unpack %s header",
2198                               sha1_to_hex(sha1));
2199        else if ((status = parse_sha1_header(hdr, &size)) < 0)
2200                status = error("unable to parse %s header", sha1_to_hex(sha1));
2201        else if (sizep)
2202                *sizep = size;
2203        git_inflate_end(&stream);
2204        munmap(map, mapsize);
2205        return status;
2206}
2207
2208/* returns enum object_type or negative */
2209int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2210{
2211        struct cached_object *co;
2212        struct pack_entry e;
2213        int status, rtype;
2214
2215        co = find_cached_object(sha1);
2216        if (co) {
2217                if (oi->sizep)
2218                        *(oi->sizep) = co->size;
2219                oi->whence = OI_CACHED;
2220                return co->type;
2221        }
2222
2223        if (!find_pack_entry(sha1, &e)) {
2224                /* Most likely it's a loose object. */
2225                status = sha1_loose_object_info(sha1, oi->sizep);
2226                if (status >= 0) {
2227                        oi->whence = OI_LOOSE;
2228                        return status;
2229                }
2230
2231                /* Not a loose object; someone else may have just packed it. */
2232                reprepare_packed_git();
2233                if (!find_pack_entry(sha1, &e))
2234                        return status;
2235        }
2236
2237        status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2238        if (status < 0) {
2239                mark_bad_packed_object(e.p, sha1);
2240                status = sha1_object_info_extended(sha1, oi);
2241        } else if (in_delta_base_cache(e.p, e.offset)) {
2242                oi->whence = OI_DBCACHED;
2243        } else {
2244                oi->whence = OI_PACKED;
2245                oi->u.packed.offset = e.offset;
2246                oi->u.packed.pack = e.p;
2247                oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2248                                         rtype == OBJ_OFS_DELTA);
2249        }
2250
2251        return status;
2252}
2253
2254int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2255{
2256        struct object_info oi;
2257
2258        oi.sizep = sizep;
2259        return sha1_object_info_extended(sha1, &oi);
2260}
2261
2262static void *read_packed_sha1(const unsigned char *sha1,
2263                              enum object_type *type, unsigned long *size)
2264{
2265        struct pack_entry e;
2266        void *data;
2267
2268        if (!find_pack_entry(sha1, &e))
2269                return NULL;
2270        data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2271        if (!data) {
2272                /*
2273                 * We're probably in deep shit, but let's try to fetch
2274                 * the required object anyway from another pack or loose.
2275                 * This should happen only in the presence of a corrupted
2276                 * pack, and is better than failing outright.
2277                 */
2278                error("failed to read object %s at offset %"PRIuMAX" from %s",
2279                      sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2280                mark_bad_packed_object(e.p, sha1);
2281                data = read_object(sha1, type, size);
2282        }
2283        return data;
2284}
2285
2286int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2287                      unsigned char *sha1)
2288{
2289        struct cached_object *co;
2290
2291        hash_sha1_file(buf, len, typename(type), sha1);
2292        if (has_sha1_file(sha1) || find_cached_object(sha1))
2293                return 0;
2294        if (cached_object_alloc <= cached_object_nr) {
2295                cached_object_alloc = alloc_nr(cached_object_alloc);
2296                cached_objects = xrealloc(cached_objects,
2297                                          sizeof(*cached_objects) *
2298                                          cached_object_alloc);
2299        }
2300        co = &cached_objects[cached_object_nr++];
2301        co->size = len;
2302        co->type = type;
2303        co->buf = xmalloc(len);
2304        memcpy(co->buf, buf, len);
2305        hashcpy(co->sha1, sha1);
2306        return 0;
2307}
2308
2309static void *read_object(const unsigned char *sha1, enum object_type *type,
2310                         unsigned long *size)
2311{
2312        unsigned long mapsize;
2313        void *map, *buf;
2314        struct cached_object *co;
2315
2316        co = find_cached_object(sha1);
2317        if (co) {
2318                *type = co->type;
2319                *size = co->size;
2320                return xmemdupz(co->buf, co->size);
2321        }
2322
2323        buf = read_packed_sha1(sha1, type, size);
2324        if (buf)
2325                return buf;
2326        map = map_sha1_file(sha1, &mapsize);
2327        if (map) {
2328                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2329                munmap(map, mapsize);
2330                return buf;
2331        }
2332        reprepare_packed_git();
2333        return read_packed_sha1(sha1, type, size);
2334}
2335
2336/*
2337 * This function dies on corrupt objects; the callers who want to
2338 * deal with them should arrange to call read_object() and give error
2339 * messages themselves.
2340 */
2341void *read_sha1_file_extended(const unsigned char *sha1,
2342                              enum object_type *type,
2343                              unsigned long *size,
2344                              unsigned flag)
2345{
2346        void *data;
2347        char *path;
2348        const struct packed_git *p;
2349        const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2350                ? lookup_replace_object(sha1) : sha1;
2351
2352        errno = 0;
2353        data = read_object(repl, type, size);
2354        if (data)
2355                return data;
2356
2357        if (errno && errno != ENOENT)
2358                die_errno("failed to read object %s", sha1_to_hex(sha1));
2359
2360        /* die if we replaced an object with one that does not exist */
2361        if (repl != sha1)
2362                die("replacement %s not found for %s",
2363                    sha1_to_hex(repl), sha1_to_hex(sha1));
2364
2365        if (has_loose_object(repl)) {
2366                path = sha1_file_name(sha1);
2367                die("loose object %s (stored in %s) is corrupt",
2368                    sha1_to_hex(repl), path);
2369        }
2370
2371        if ((p = has_packed_and_bad(repl)) != NULL)
2372                die("packed object %s (stored in %s) is corrupt",
2373                    sha1_to_hex(repl), p->pack_name);
2374
2375        return NULL;
2376}
2377
2378void *read_object_with_reference(const unsigned char *sha1,
2379                                 const char *required_type_name,
2380                                 unsigned long *size,
2381                                 unsigned char *actual_sha1_return)
2382{
2383        enum object_type type, required_type;
2384        void *buffer;
2385        unsigned long isize;
2386        unsigned char actual_sha1[20];
2387
2388        required_type = type_from_string(required_type_name);
2389        hashcpy(actual_sha1, sha1);
2390        while (1) {
2391                int ref_length = -1;
2392                const char *ref_type = NULL;
2393
2394                buffer = read_sha1_file(actual_sha1, &type, &isize);
2395                if (!buffer)
2396                        return NULL;
2397                if (type == required_type) {
2398                        *size = isize;
2399                        if (actual_sha1_return)
2400                                hashcpy(actual_sha1_return, actual_sha1);
2401                        return buffer;
2402                }
2403                /* Handle references */
2404                else if (type == OBJ_COMMIT)
2405                        ref_type = "tree ";
2406                else if (type == OBJ_TAG)
2407                        ref_type = "object ";
2408                else {
2409                        free(buffer);
2410                        return NULL;
2411                }
2412                ref_length = strlen(ref_type);
2413
2414                if (ref_length + 40 > isize ||
2415                    memcmp(buffer, ref_type, ref_length) ||
2416                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2417                        free(buffer);
2418                        return NULL;
2419                }
2420                free(buffer);
2421                /* Now we have the ID of the referred-to object in
2422                 * actual_sha1.  Check again. */
2423        }
2424}
2425
2426static void write_sha1_file_prepare(const void *buf, unsigned long len,
2427                                    const char *type, unsigned char *sha1,
2428                                    char *hdr, int *hdrlen)
2429{
2430        git_SHA_CTX c;
2431
2432        /* Generate the header */
2433        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2434
2435        /* Sha1.. */
2436        git_SHA1_Init(&c);
2437        git_SHA1_Update(&c, hdr, *hdrlen);
2438        git_SHA1_Update(&c, buf, len);
2439        git_SHA1_Final(sha1, &c);
2440}
2441
2442/*
2443 * Move the just written object into its final resting place.
2444 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2445 * "moving" is only a part of what it does, when no patch between
2446 * master to pu changes the call sites of this function.
2447 */
2448int move_temp_to_file(const char *tmpfile, const char *filename)
2449{
2450        int ret = 0;
2451
2452        if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2453                goto try_rename;
2454        else if (link(tmpfile, filename))
2455                ret = errno;
2456
2457        /*
2458         * Coda hack - coda doesn't like cross-directory links,
2459         * so we fall back to a rename, which will mean that it
2460         * won't be able to check collisions, but that's not a
2461         * big deal.
2462         *
2463         * The same holds for FAT formatted media.
2464         *
2465         * When this succeeds, we just return.  We have nothing
2466         * left to unlink.
2467         */
2468        if (ret && ret != EEXIST) {
2469        try_rename:
2470                if (!rename(tmpfile, filename))
2471                        goto out;
2472                ret = errno;
2473        }
2474        unlink_or_warn(tmpfile);
2475        if (ret) {
2476                if (ret != EEXIST) {
2477                        return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2478                }
2479                /* FIXME!!! Collision check here ? */
2480        }
2481
2482out:
2483        if (adjust_shared_perm(filename))
2484                return error("unable to set permission to '%s'", filename);
2485        return 0;
2486}
2487
2488static int write_buffer(int fd, const void *buf, size_t len)
2489{
2490        if (write_in_full(fd, buf, len) < 0)
2491                return error("file write error (%s)", strerror(errno));
2492        return 0;
2493}
2494
2495int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2496                   unsigned char *sha1)
2497{
2498        char hdr[32];
2499        int hdrlen;
2500        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2501        return 0;
2502}
2503
2504/* Finalize a file on disk, and close it. */
2505static void close_sha1_file(int fd)
2506{
2507        if (fsync_object_files)
2508                fsync_or_die(fd, "sha1 file");
2509        if (close(fd) != 0)
2510                die_errno("error when closing sha1 file");
2511}
2512
2513/* Size of directory component, including the ending '/' */
2514static inline int directory_size(const char *filename)
2515{
2516        const char *s = strrchr(filename, '/');
2517        if (!s)
2518                return 0;
2519        return s - filename + 1;
2520}
2521
2522/*
2523 * This creates a temporary file in the same directory as the final
2524 * 'filename'
2525 *
2526 * We want to avoid cross-directory filename renames, because those
2527 * can have problems on various filesystems (FAT, NFS, Coda).
2528 */
2529static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2530{
2531        int fd, dirlen = directory_size(filename);
2532
2533        if (dirlen + 20 > bufsiz) {
2534                errno = ENAMETOOLONG;
2535                return -1;
2536        }
2537        memcpy(buffer, filename, dirlen);
2538        strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2539        fd = git_mkstemp_mode(buffer, 0444);
2540        if (fd < 0 && dirlen && errno == ENOENT) {
2541                /* Make sure the directory exists */
2542                memcpy(buffer, filename, dirlen);
2543                buffer[dirlen-1] = 0;
2544                if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2545                        return -1;
2546
2547                /* Try again */
2548                strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2549                fd = git_mkstemp_mode(buffer, 0444);
2550        }
2551        return fd;
2552}
2553
2554static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2555                              const void *buf, unsigned long len, time_t mtime)
2556{
2557        int fd, ret;
2558        unsigned char compressed[4096];
2559        git_zstream stream;
2560        git_SHA_CTX c;
2561        unsigned char parano_sha1[20];
2562        char *filename;
2563        static char tmp_file[PATH_MAX];
2564
2565        filename = sha1_file_name(sha1);
2566        fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2567        if (fd < 0) {
2568                if (errno == EACCES)
2569                        return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2570                else
2571                        return error("unable to create temporary file: %s", strerror(errno));
2572        }
2573
2574        /* Set it up */
2575        memset(&stream, 0, sizeof(stream));
2576        git_deflate_init(&stream, zlib_compression_level);
2577        stream.next_out = compressed;
2578        stream.avail_out = sizeof(compressed);
2579        git_SHA1_Init(&c);
2580
2581        /* First header.. */
2582        stream.next_in = (unsigned char *)hdr;
2583        stream.avail_in = hdrlen;
2584        while (git_deflate(&stream, 0) == Z_OK)
2585                ; /* nothing */
2586        git_SHA1_Update(&c, hdr, hdrlen);
2587
2588        /* Then the data itself.. */
2589        stream.next_in = (void *)buf;
2590        stream.avail_in = len;
2591        do {
2592                unsigned char *in0 = stream.next_in;
2593                ret = git_deflate(&stream, Z_FINISH);
2594                git_SHA1_Update(&c, in0, stream.next_in - in0);
2595                if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2596                        die("unable to write sha1 file");
2597                stream.next_out = compressed;
2598                stream.avail_out = sizeof(compressed);
2599        } while (ret == Z_OK);
2600
2601        if (ret != Z_STREAM_END)
2602                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2603        ret = git_deflate_end_gently(&stream);
2604        if (ret != Z_OK)
2605                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2606        git_SHA1_Final(parano_sha1, &c);
2607        if (hashcmp(sha1, parano_sha1) != 0)
2608                die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2609
2610        close_sha1_file(fd);
2611
2612        if (mtime) {
2613                struct utimbuf utb;
2614                utb.actime = mtime;
2615                utb.modtime = mtime;
2616                if (utime(tmp_file, &utb) < 0)
2617                        warning("failed utime() on %s: %s",
2618                                tmp_file, strerror(errno));
2619        }
2620
2621        return move_temp_to_file(tmp_file, filename);
2622}
2623
2624int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2625{
2626        unsigned char sha1[20];
2627        char hdr[32];
2628        int hdrlen;
2629
2630        /* Normally if we have it in the pack then we do not bother writing
2631         * it out into .git/objects/??/?{38} file.
2632         */
2633        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2634        if (returnsha1)
2635                hashcpy(returnsha1, sha1);
2636        if (has_sha1_file(sha1))
2637                return 0;
2638        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2639}
2640
2641int force_object_loose(const unsigned char *sha1, time_t mtime)
2642{
2643        void *buf;
2644        unsigned long len;
2645        enum object_type type;
2646        char hdr[32];
2647        int hdrlen;
2648        int ret;
2649
2650        if (has_loose_object(sha1))
2651                return 0;
2652        buf = read_packed_sha1(sha1, &type, &len);
2653        if (!buf)
2654                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2655        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2656        ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2657        free(buf);
2658
2659        return ret;
2660}
2661
2662int has_pack_index(const unsigned char *sha1)
2663{
2664        struct stat st;
2665        if (stat(sha1_pack_index_name(sha1), &st))
2666                return 0;
2667        return 1;
2668}
2669
2670int has_sha1_pack(const unsigned char *sha1)
2671{
2672        struct pack_entry e;
2673        return find_pack_entry(sha1, &e);
2674}
2675
2676int has_sha1_file(const unsigned char *sha1)
2677{
2678        struct pack_entry e;
2679
2680        if (find_pack_entry(sha1, &e))
2681                return 1;
2682        return has_loose_object(sha1);
2683}
2684
2685static void check_tree(const void *buf, size_t size)
2686{
2687        struct tree_desc desc;
2688        struct name_entry entry;
2689
2690        init_tree_desc(&desc, buf, size);
2691        while (tree_entry(&desc, &entry))
2692                /* do nothing
2693                 * tree_entry() will die() on malformed entries */
2694                ;
2695}
2696
2697static void check_commit(const void *buf, size_t size)
2698{
2699        struct commit c;
2700        memset(&c, 0, sizeof(c));
2701        if (parse_commit_buffer(&c, buf, size))
2702                die("corrupt commit");
2703}
2704
2705static void check_tag(const void *buf, size_t size)
2706{
2707        struct tag t;
2708        memset(&t, 0, sizeof(t));
2709        if (parse_tag_buffer(&t, buf, size))
2710                die("corrupt tag");
2711}
2712
2713static int index_mem(unsigned char *sha1, void *buf, size_t size,
2714                     enum object_type type,
2715                     const char *path, unsigned flags)
2716{
2717        int ret, re_allocated = 0;
2718        int write_object = flags & HASH_WRITE_OBJECT;
2719
2720        if (!type)
2721                type = OBJ_BLOB;
2722
2723        /*
2724         * Convert blobs to git internal format
2725         */
2726        if ((type == OBJ_BLOB) && path) {
2727                struct strbuf nbuf = STRBUF_INIT;
2728                if (convert_to_git(path, buf, size, &nbuf,
2729                                   write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2730                        buf = strbuf_detach(&nbuf, &size);
2731                        re_allocated = 1;
2732                }
2733        }
2734        if (flags & HASH_FORMAT_CHECK) {
2735                if (type == OBJ_TREE)
2736                        check_tree(buf, size);
2737                if (type == OBJ_COMMIT)
2738                        check_commit(buf, size);
2739                if (type == OBJ_TAG)
2740                        check_tag(buf, size);
2741        }
2742
2743        if (write_object)
2744                ret = write_sha1_file(buf, size, typename(type), sha1);
2745        else
2746                ret = hash_sha1_file(buf, size, typename(type), sha1);
2747        if (re_allocated)
2748                free(buf);
2749        return ret;
2750}
2751
2752static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2753                      const char *path, unsigned flags)
2754{
2755        struct strbuf sbuf = STRBUF_INIT;
2756        int ret;
2757
2758        if (strbuf_read(&sbuf, fd, 4096) >= 0)
2759                ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2760        else
2761                ret = -1;
2762        strbuf_release(&sbuf);
2763        return ret;
2764}
2765
2766#define SMALL_FILE_SIZE (32*1024)
2767
2768static int index_core(unsigned char *sha1, int fd, size_t size,
2769                      enum object_type type, const char *path,
2770                      unsigned flags)
2771{
2772        int ret;
2773
2774        if (!size) {
2775                ret = index_mem(sha1, NULL, size, type, path, flags);
2776        } else if (size <= SMALL_FILE_SIZE) {
2777                char *buf = xmalloc(size);
2778                if (size == read_in_full(fd, buf, size))
2779                        ret = index_mem(sha1, buf, size, type, path, flags);
2780                else
2781                        ret = error("short read %s", strerror(errno));
2782                free(buf);
2783        } else {
2784                void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2785                ret = index_mem(sha1, buf, size, type, path, flags);
2786                munmap(buf, size);
2787        }
2788        return ret;
2789}
2790
2791/*
2792 * This creates one packfile per large blob unless bulk-checkin
2793 * machinery is "plugged".
2794 *
2795 * This also bypasses the usual "convert-to-git" dance, and that is on
2796 * purpose. We could write a streaming version of the converting
2797 * functions and insert that before feeding the data to fast-import
2798 * (or equivalent in-core API described above). However, that is
2799 * somewhat complicated, as we do not know the size of the filter
2800 * result, which we need to know beforehand when writing a git object.
2801 * Since the primary motivation for trying to stream from the working
2802 * tree file and to avoid mmaping it in core is to deal with large
2803 * binary blobs, they generally do not want to get any conversion, and
2804 * callers should avoid this code path when filters are requested.
2805 */
2806static int index_stream(unsigned char *sha1, int fd, size_t size,
2807                        enum object_type type, const char *path,
2808                        unsigned flags)
2809{
2810        return index_bulk_checkin(sha1, fd, size, type, path, flags);
2811}
2812
2813int index_fd(unsigned char *sha1, int fd, struct stat *st,
2814             enum object_type type, const char *path, unsigned flags)
2815{
2816        int ret;
2817        size_t size = xsize_t(st->st_size);
2818
2819        if (!S_ISREG(st->st_mode))
2820                ret = index_pipe(sha1, fd, type, path, flags);
2821        else if (size <= big_file_threshold || type != OBJ_BLOB ||
2822                 (path && would_convert_to_git(path, NULL, 0, 0)))
2823                ret = index_core(sha1, fd, size, type, path, flags);
2824        else
2825                ret = index_stream(sha1, fd, size, type, path, flags);
2826        close(fd);
2827        return ret;
2828}
2829
2830int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2831{
2832        int fd;
2833        struct strbuf sb = STRBUF_INIT;
2834
2835        switch (st->st_mode & S_IFMT) {
2836        case S_IFREG:
2837                fd = open(path, O_RDONLY);
2838                if (fd < 0)
2839                        return error("open(\"%s\"): %s", path,
2840                                     strerror(errno));
2841                if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2842                        return error("%s: failed to insert into database",
2843                                     path);
2844                break;
2845        case S_IFLNK:
2846                if (strbuf_readlink(&sb, path, st->st_size)) {
2847                        char *errstr = strerror(errno);
2848                        return error("readlink(\"%s\"): %s", path,
2849                                     errstr);
2850                }
2851                if (!(flags & HASH_WRITE_OBJECT))
2852                        hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2853                else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2854                        return error("%s: failed to insert into database",
2855                                     path);
2856                strbuf_release(&sb);
2857                break;
2858        case S_IFDIR:
2859                return resolve_gitlink_ref(path, "HEAD", sha1);
2860        default:
2861                return error("%s: unsupported file type", path);
2862        }
2863        return 0;
2864}
2865
2866int read_pack_header(int fd, struct pack_header *header)
2867{
2868        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2869                /* "eof before pack header was fully read" */
2870                return PH_ERROR_EOF;
2871
2872        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2873                /* "protocol error (pack signature mismatch detected)" */
2874                return PH_ERROR_PACK_SIGNATURE;
2875        if (!pack_version_ok(header->hdr_version))
2876                /* "protocol error (pack version unsupported)" */
2877                return PH_ERROR_PROTOCOL;
2878        return 0;
2879}
2880
2881void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2882{
2883        enum object_type type = sha1_object_info(sha1, NULL);
2884        if (type < 0)
2885                die("%s is not a valid object", sha1_to_hex(sha1));
2886        if (type != expect)
2887                die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2888                    typename(expect));
2889}