sha1_file.con commit Duh. Fix transposed characters in git-pull-script (5571be7)
   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 <sys/types.h>
  10#include <dirent.h>
  11#include "cache.h"
  12#include "delta.h"
  13
  14#ifndef O_NOATIME
  15#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  16#define O_NOATIME 01000000
  17#else
  18#define O_NOATIME 0
  19#endif
  20#endif
  21
  22static unsigned int sha1_file_open_flag = O_NOATIME;
  23
  24static unsigned hexval(char c)
  25{
  26        if (c >= '0' && c <= '9')
  27                return c - '0';
  28        if (c >= 'a' && c <= 'f')
  29                return c - 'a' + 10;
  30        if (c >= 'A' && c <= 'F')
  31                return c - 'A' + 10;
  32        return ~0;
  33}
  34
  35int get_sha1_hex(const char *hex, unsigned char *sha1)
  36{
  37        int i;
  38        for (i = 0; i < 20; i++) {
  39                unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
  40                if (val & ~0xff)
  41                        return -1;
  42                *sha1++ = val;
  43                hex += 2;
  44        }
  45        return 0;
  46}
  47
  48static int get_sha1_file(const char *path, unsigned char *result)
  49{
  50        char buffer[60];
  51        int fd = open(path, O_RDONLY);
  52        int len;
  53
  54        if (fd < 0)
  55                return -1;
  56        len = read(fd, buffer, sizeof(buffer));
  57        close(fd);
  58        if (len < 40)
  59                return -1;
  60        return get_sha1_hex(buffer, result);
  61}
  62
  63static char *git_dir, *git_object_dir, *git_index_file, *git_refs_dir;
  64static void setup_git_env(void)
  65{
  66        git_dir = gitenv(GIT_DIR_ENVIRONMENT);
  67        if (!git_dir)
  68                git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
  69        git_object_dir = gitenv(DB_ENVIRONMENT);
  70        if (!git_object_dir) {
  71                git_object_dir = xmalloc(strlen(git_dir) + 9);
  72                sprintf(git_object_dir, "%s/objects", git_dir);
  73        }
  74        git_refs_dir = xmalloc(strlen(git_dir) + 6);
  75        sprintf(git_refs_dir, "%s/refs", git_dir);
  76        git_index_file = gitenv(INDEX_ENVIRONMENT);
  77        if (!git_index_file) {
  78                git_index_file = xmalloc(strlen(git_dir) + 7);
  79                sprintf(git_index_file, "%s/index", git_dir);
  80        }
  81}
  82
  83char *get_object_directory(void)
  84{
  85        if (!git_object_dir)
  86                setup_git_env();
  87        return git_object_dir;
  88}
  89
  90char *get_refs_directory(void)
  91{
  92        if (!git_refs_dir)
  93                setup_git_env();
  94        return git_refs_dir;
  95}
  96
  97char *get_index_file(void)
  98{
  99        if (!git_index_file)
 100                setup_git_env();
 101        return git_index_file;
 102}
 103
 104int get_sha1(const char *str, unsigned char *sha1)
 105{
 106        static char pathname[PATH_MAX];
 107        static const char *prefix[] = {
 108                "",
 109                "refs",
 110                "refs/tags",
 111                "refs/heads",
 112                "refs/snap",
 113                NULL
 114        };
 115        const char **p;
 116
 117        if (!get_sha1_hex(str, sha1))
 118                return 0;
 119
 120        if (!git_dir)
 121                setup_git_env();
 122        for (p = prefix; *p; p++) {
 123                snprintf(pathname, sizeof(pathname), "%s/%s/%s",
 124                         git_dir, *p, str);
 125                if (!get_sha1_file(pathname, sha1))
 126                        return 0;
 127        }
 128
 129        return -1;
 130}
 131
 132char * sha1_to_hex(const unsigned char *sha1)
 133{
 134        static char buffer[50];
 135        static const char hex[] = "0123456789abcdef";
 136        char *buf = buffer;
 137        int i;
 138
 139        for (i = 0; i < 20; i++) {
 140                unsigned int val = *sha1++;
 141                *buf++ = hex[val >> 4];
 142                *buf++ = hex[val & 0xf];
 143        }
 144        return buffer;
 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 a "strdup()" 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 *name, *base;
 172
 173        if (!base) {
 174                const char *sha1_file_directory = get_object_directory();
 175                int len = strlen(sha1_file_directory);
 176                base = xmalloc(len + 60);
 177                memcpy(base, sha1_file_directory, len);
 178                memset(base+len, 0, 60);
 179                base[len] = '/';
 180                base[len+3] = '/';
 181                name = base + len + 1;
 182        }
 183        fill_sha1_path(name, sha1);
 184        return base;
 185}
 186
 187static struct alternate_object_database {
 188        char *base;
 189        char *name;
 190} *alt_odb;
 191
 192/*
 193 * Prepare alternate object database registry.
 194 * alt_odb points at an array of struct alternate_object_database.
 195 * This array is terminated with an element that has both its base
 196 * and name set to NULL.  alt_odb[n] comes from n'th non-empty
 197 * element from colon separated ALTERNATE_DB_ENVIRONMENT environment
 198 * variable, and its base points at a statically allocated buffer
 199 * that contains "/the/directory/corresponding/to/.git/objects/...",
 200 * while its name points just after the slash at the end of
 201 * ".git/objects/" in the example above, and has enough space to hold
 202 * 40-byte hex SHA1, an extra slash for the first level indirection,
 203 * and the terminating NUL.
 204 * This function allocates the alt_odb array and all the strings
 205 * pointed by base fields of the array elements with one xmalloc();
 206 * the string pool immediately follows the array.
 207 */
 208static void prepare_alt_odb(void)
 209{
 210        int pass, totlen, i;
 211        const char *cp, *last;
 212        char *op = NULL;
 213        const char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
 214
 215        /* The first pass counts how large an area to allocate to
 216         * hold the entire alt_odb structure, including array of
 217         * structs and path buffers for them.  The second pass fills
 218         * the structure and prepares the path buffers for use by
 219         * fill_sha1_path().
 220         */
 221        for (totlen = pass = 0; pass < 2; pass++) {
 222                last = alt;
 223                i = 0;
 224                do {
 225                        cp = strchr(last, ':') ? : last + strlen(last);
 226                        if (last != cp) {
 227                                /* 43 = 40-byte + 2 '/' + terminating NUL */
 228                                int pfxlen = cp - last;
 229                                int entlen = pfxlen + 43;
 230                                if (pass == 0)
 231                                        totlen += entlen;
 232                                else {
 233                                        alt_odb[i].base = op;
 234                                        alt_odb[i].name = op + pfxlen + 1;
 235                                        memcpy(op, last, pfxlen);
 236                                        op[pfxlen] = op[pfxlen + 3] = '/';
 237                                        op[entlen-1] = 0;
 238                                        op += entlen;
 239                                }
 240                                i++;
 241                        }
 242                        while (*cp && *cp == ':')
 243                                cp++;
 244                        last = cp;
 245                } while (*cp);
 246                if (pass)
 247                        break;
 248                alt_odb = xmalloc(sizeof(*alt_odb) * (i + 1) + totlen);
 249                alt_odb[i].base = alt_odb[i].name = NULL;
 250                op = (char*)(&alt_odb[i+1]);
 251        }
 252}
 253
 254static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
 255{
 256        int i;
 257        char *name = sha1_file_name(sha1);
 258
 259        if (!stat(name, st))
 260                return name;
 261        if (!alt_odb)
 262                prepare_alt_odb();
 263        for (i = 0; (name = alt_odb[i].name) != NULL; i++) {
 264                fill_sha1_path(name, sha1);
 265                if (!stat(alt_odb[i].base, st))
 266                        return alt_odb[i].base;
 267        }
 268        return NULL;
 269}
 270
 271#define PACK_MAX_SZ (1<<26)
 272static int pack_used_ctr;
 273static unsigned long pack_mapped;
 274static struct packed_git {
 275        struct packed_git *next;
 276        unsigned long index_size;
 277        unsigned long pack_size;
 278        unsigned int *index_base;
 279        void *pack_base;
 280        unsigned int pack_last_used;
 281        char pack_name[0]; /* something like ".git/objects/pack/xxxxx.pack" */
 282} *packed_git;
 283
 284struct pack_entry {
 285        unsigned int offset;
 286        unsigned char sha1[20];
 287        struct packed_git *p;
 288};
 289
 290static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
 291                                void **idx_map_)
 292{
 293        void *idx_map;
 294        unsigned int *index;
 295        unsigned long idx_size;
 296        int nr, i;
 297        int fd = open(path, O_RDONLY);
 298        struct stat st;
 299        if (fd < 0)
 300                return -1;
 301        if (fstat(fd, &st)) {
 302                close(fd);
 303                return -1;
 304        }
 305        idx_size = st.st_size;
 306        idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 307        close(fd);
 308        if (idx_map == MAP_FAILED)
 309                return -1;
 310
 311        index = idx_map;
 312
 313        /* check index map */
 314        if (idx_size < 4*256 + 20)
 315                return error("index file too small");
 316        nr = 0;
 317        for (i = 0; i < 256; i++) {
 318                unsigned int n = ntohl(index[i]);
 319                if (n < nr)
 320                        return error("non-monotonic index");
 321                nr = n;
 322        }
 323
 324        /*
 325         * Total size:
 326         *  - 256 index entries 4 bytes each
 327         *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 328         *  - 20-byte SHA1 of the packfile
 329         *  - 20-byte SHA1 file checksum
 330         */
 331        if (idx_size != 4*256 + nr * 24 + 20 + 20)
 332                return error("wrong index file size");
 333
 334        *idx_map_ = idx_map;
 335        *idx_size_ = idx_size;
 336        return 0;
 337}
 338
 339static void unuse_one_packed_git(void)
 340{
 341        /* NOTYET */
 342}
 343
 344static int use_packed_git(struct packed_git *p)
 345{
 346        if (!p->pack_base) {
 347                int fd;
 348                struct stat st;
 349                void *map;
 350
 351                pack_mapped += p->pack_size;
 352                while (PACK_MAX_SZ < pack_mapped)
 353                        unuse_one_packed_git();
 354                fd = open(p->pack_name, O_RDONLY);
 355                if (fd < 0)
 356                        return -1;
 357                if (fstat(fd, &st)) {
 358                        close(fd);
 359                        return -1;
 360                }
 361                if (st.st_size != p->pack_size)
 362                        return -1;
 363                map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
 364                close(fd);
 365                if (map == MAP_FAILED)
 366                        return -1;
 367                p->pack_base = map;
 368        }
 369        p->pack_last_used = pack_used_ctr++;
 370        return 0;
 371}
 372
 373static struct packed_git *add_packed_git(char *path, int path_len)
 374{
 375        struct stat st;
 376        struct packed_git *p;
 377        unsigned long idx_size;
 378        void *idx_map;
 379
 380        if (check_packed_git_idx(path, &idx_size, &idx_map))
 381                return NULL;
 382
 383        /* do we have a corresponding .pack file? */
 384        strcpy(path + path_len - 4, ".pack");
 385        if (stat(path, &st) || !S_ISREG(st.st_mode)) {
 386                munmap(idx_map, idx_size);
 387                return NULL;
 388        }
 389        /* ok, it looks sane as far as we can check without
 390         * actually mapping the pack file.
 391         */
 392        p = xmalloc(sizeof(*p) + path_len + 2);
 393        strcpy(p->pack_name, path);
 394        p->index_size = idx_size;
 395        p->pack_size = st.st_size;
 396        p->index_base = idx_map;
 397        p->next = NULL;
 398        p->pack_last_used = 0;
 399        return p;
 400}
 401
 402static void prepare_packed_git_one(char *objdir)
 403{
 404        char path[PATH_MAX];
 405        int len;
 406        DIR *dir;
 407        struct dirent *de;
 408
 409        sprintf(path, "%s/pack", objdir);
 410        len = strlen(path);
 411        dir = opendir(path);
 412        if (!dir)
 413                return;
 414        path[len++] = '/';
 415        while ((de = readdir(dir)) != NULL) {
 416                int namelen = strlen(de->d_name);
 417                struct packed_git *p;
 418
 419                if (strcmp(de->d_name + namelen - 4, ".idx"))
 420                        continue;
 421
 422                /* we have .idx.  Is it a file we can map? */
 423                strcpy(path + len, de->d_name);
 424                p = add_packed_git(path, len + namelen);
 425                if (!p)
 426                        continue;
 427                p->next = packed_git;
 428                packed_git = p;
 429        }
 430}
 431
 432static void prepare_packed_git(void)
 433{
 434        int i;
 435        static int run_once = 0;
 436
 437        if (run_once++)
 438                return;
 439
 440        prepare_packed_git_one(get_object_directory());
 441        if (!alt_odb)
 442                prepare_alt_odb();
 443        for (i = 0; alt_odb[i].base != NULL; i++) {
 444                alt_odb[i].name[0] = 0;
 445                prepare_packed_git_one(alt_odb[i].base);
 446        }
 447}
 448
 449int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
 450{
 451        char header[100];
 452        unsigned char real_sha1[20];
 453        SHA_CTX c;
 454
 455        SHA1_Init(&c);
 456        SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
 457        SHA1_Update(&c, map, size);
 458        SHA1_Final(real_sha1, &c);
 459        return memcmp(sha1, real_sha1, 20) ? -1 : 0;
 460}
 461
 462static void *map_sha1_file_internal(const unsigned char *sha1,
 463                                    unsigned long *size,
 464                                    int say_error)
 465{
 466        struct stat st;
 467        void *map;
 468        int fd;
 469        char *filename = find_sha1_file(sha1, &st);
 470
 471        if (!filename) {
 472                if (say_error)
 473                        error("cannot map sha1 file %s", sha1_to_hex(sha1));
 474                return NULL;
 475        }
 476
 477        fd = open(filename, O_RDONLY | sha1_file_open_flag);
 478        if (fd < 0) {
 479                /* See if it works without O_NOATIME */
 480                switch (sha1_file_open_flag) {
 481                default:
 482                        fd = open(filename, O_RDONLY);
 483                        if (fd >= 0)
 484                                break;
 485                /* Fallthrough */
 486                case 0:
 487                        if (say_error)
 488                                perror(filename);
 489                        return NULL;
 490                }
 491
 492                /* If it failed once, it will probably fail again.
 493                 * Stop using O_NOATIME
 494                 */
 495                sha1_file_open_flag = 0;
 496        }
 497        map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
 498        close(fd);
 499        if (-1 == (int)(long)map)
 500                return NULL;
 501        *size = st.st_size;
 502        return map;
 503}
 504
 505void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
 506{
 507        return map_sha1_file_internal(sha1, size, 1);
 508}
 509
 510int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
 511{
 512        /* Get the data stream */
 513        memset(stream, 0, sizeof(*stream));
 514        stream->next_in = map;
 515        stream->avail_in = mapsize;
 516        stream->next_out = buffer;
 517        stream->avail_out = size;
 518
 519        inflateInit(stream);
 520        return inflate(stream, 0);
 521}
 522
 523void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
 524{
 525        int bytes = strlen(buffer) + 1;
 526        unsigned char *buf = xmalloc(1+size);
 527
 528        memcpy(buf, buffer + bytes, stream->total_out - bytes);
 529        bytes = stream->total_out - bytes;
 530        if (bytes < size) {
 531                stream->next_out = buf + bytes;
 532                stream->avail_out = size - bytes;
 533                while (inflate(stream, Z_FINISH) == Z_OK)
 534                        /* nothing */;
 535        }
 536        buf[size] = 0;
 537        inflateEnd(stream);
 538        return buf;
 539}
 540
 541/*
 542 * We used to just use "sscanf()", but that's actually way
 543 * too permissive for what we want to check. So do an anal
 544 * object header parse by hand.
 545 */
 546int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
 547{
 548        int i;
 549        unsigned long size;
 550
 551        /*
 552         * The type can be at most ten bytes (including the 
 553         * terminating '\0' that we add), and is followed by
 554         * a space. 
 555         */
 556        i = 10;
 557        for (;;) {
 558                char c = *hdr++;
 559                if (c == ' ')
 560                        break;
 561                if (!--i)
 562                        return -1;
 563                *type++ = c;
 564        }
 565        *type = 0;
 566
 567        /*
 568         * The length must follow immediately, and be in canonical
 569         * decimal format (ie "010" is not valid).
 570         */
 571        size = *hdr++ - '0';
 572        if (size > 9)
 573                return -1;
 574        if (size) {
 575                for (;;) {
 576                        unsigned long c = *hdr - '0';
 577                        if (c > 9)
 578                                break;
 579                        hdr++;
 580                        size = size * 10 + c;
 581                }
 582        }
 583        *sizep = size;
 584
 585        /*
 586         * The length must be followed by a zero byte
 587         */
 588        return *hdr ? -1 : 0;
 589}
 590
 591void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
 592{
 593        int ret;
 594        z_stream stream;
 595        char hdr[8192];
 596
 597        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
 598        if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
 599                return NULL;
 600
 601        return unpack_sha1_rest(&stream, hdr, *size);
 602}
 603
 604static int packed_delta_info(unsigned char *base_sha1,
 605                             unsigned long delta_size,
 606                             unsigned long left,
 607                             char *type,
 608                             unsigned long *sizep)
 609{
 610        unsigned char *data;
 611        unsigned char delta_head[64];
 612        int i;
 613        unsigned char cmd;
 614        unsigned long data_size, result_size, base_size, verify_base_size;
 615        z_stream stream;
 616        int st;
 617
 618        if (left < 20)
 619                die("truncated pack file");
 620        if (sha1_object_info(base_sha1, type, &base_size))
 621                die("cannot get info for delta-pack base");
 622
 623        data = base_sha1 + 20;
 624        data_size = left - 20;
 625
 626        memset(&stream, 0, sizeof(stream));
 627
 628        stream.next_in = data;
 629        stream.avail_in = data_size;
 630        stream.next_out = delta_head;
 631        stream.avail_out = sizeof(delta_head);
 632
 633        inflateInit(&stream);
 634        st = inflate(&stream, Z_FINISH);
 635        inflateEnd(&stream);
 636        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
 637                die("delta data unpack-initial failed");
 638
 639        /* Examine the initial part of the delta to figure out
 640         * the result size.  Verify the base size while we are at it.
 641         */
 642        data = delta_head;
 643        verify_base_size = i = 0;
 644        cmd = *data++;
 645        while (cmd) {
 646                if (cmd & 1)
 647                        verify_base_size |= *data++ << i;
 648                i += 8;
 649                cmd >>= 1;
 650        }
 651
 652        /* Read the result size */
 653        result_size = i = 0;
 654        cmd = *data++;
 655        while (cmd) {
 656                if (cmd & 1)
 657                        result_size |= *data++ << i;
 658                i += 8;
 659                cmd >>= 1;
 660        }
 661        if (verify_base_size != base_size)
 662                die("delta base size mismatch");
 663
 664        *sizep = result_size;
 665        return 0;
 666}
 667
 668static int packed_object_info(struct pack_entry *entry,
 669                              char *type, unsigned long *sizep)
 670{
 671        struct packed_git *p = entry->p;
 672        unsigned long offset, size, left;
 673        unsigned char *pack;
 674
 675        offset = entry->offset;
 676        if (p->pack_size - 5 < offset)
 677                die("object offset outside of pack file");
 678
 679        if (use_packed_git(p))
 680                die("cannot map packed file");
 681
 682        pack = p->pack_base + offset;
 683        size = (pack[1] << 24) + (pack[2] << 16) + (pack[3] << 8) + pack[4];
 684        left = p->pack_size - offset - 5;
 685        switch (*pack) {
 686        case 'D':
 687                return packed_delta_info(pack+5, size, left, type, sizep);
 688                break;
 689        case 'C':
 690                strcpy(type, "commit");
 691                break;
 692        case 'T':
 693                strcpy(type, "tree");
 694                break;
 695        case 'B':
 696                strcpy(type, "blob");
 697                break;
 698        case 'G':
 699                strcpy(type, "tag");
 700                break;
 701        default:
 702                die("corrupted pack file");
 703        }
 704        *sizep = size;
 705        return 0;
 706}
 707
 708/* forward declaration for a mutually recursive function */
 709static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
 710
 711static void *unpack_delta_entry(unsigned char *base_sha1,
 712                                unsigned long delta_size,
 713                                unsigned long left,
 714                                char *type,
 715                                unsigned long *sizep)
 716{
 717        void *data, *delta_data, *result, *base;
 718        unsigned long data_size, result_size, base_size;
 719        z_stream stream;
 720        int st;
 721
 722        if (left < 20)
 723                die("truncated pack file");
 724        data = base_sha1 + 20;
 725        data_size = left - 20;
 726        delta_data = xmalloc(delta_size);
 727
 728        memset(&stream, 0, sizeof(stream));
 729
 730        stream.next_in = data;
 731        stream.avail_in = data_size;
 732        stream.next_out = delta_data;
 733        stream.avail_out = delta_size;
 734
 735        inflateInit(&stream);
 736        st = inflate(&stream, Z_FINISH);
 737        inflateEnd(&stream);
 738        if ((st != Z_STREAM_END) || stream.total_out != delta_size)
 739                die("delta data unpack failed");
 740
 741        /* This may recursively unpack the base, which is what we want */
 742        base = read_sha1_file(base_sha1, type, &base_size);
 743        if (!base)
 744                die("failed to read delta-pack base object %s",
 745                    sha1_to_hex(base_sha1));
 746        result = patch_delta(base, base_size,
 747                             delta_data, delta_size,
 748                             &result_size);
 749        if (!result)
 750                die("failed to apply delta");
 751        free(delta_data);
 752        free(base);
 753        *sizep = result_size;
 754        return result;
 755}
 756
 757static void *unpack_non_delta_entry(unsigned char *data,
 758                                    unsigned long size,
 759                                    unsigned long left)
 760{
 761        int st;
 762        z_stream stream;
 763        char *buffer;
 764
 765        buffer = xmalloc(size + 1);
 766        buffer[size] = 0;
 767        memset(&stream, 0, sizeof(stream));
 768        stream.next_in = data;
 769        stream.avail_in = left;
 770        stream.next_out = buffer;
 771        stream.avail_out = size;
 772
 773        inflateInit(&stream);
 774        st = inflate(&stream, Z_FINISH);
 775        inflateEnd(&stream);
 776        if ((st != Z_STREAM_END) || stream.total_out != size) {
 777                free(buffer);
 778                return NULL;
 779        }
 780
 781        return buffer;
 782}
 783
 784static void *unpack_entry(struct pack_entry *entry,
 785                          char *type, unsigned long *sizep)
 786{
 787        struct packed_git *p = entry->p;
 788        unsigned long offset, size, left;
 789        unsigned char *pack;
 790
 791        offset = entry->offset;
 792        if (p->pack_size - 5 < offset)
 793                die("object offset outside of pack file");
 794
 795        if (use_packed_git(p))
 796                die("cannot map packed file");
 797
 798        pack = p->pack_base + offset;
 799        size = (pack[1] << 24) + (pack[2] << 16) + (pack[3] << 8) + pack[4];
 800        left = p->pack_size - offset - 5;
 801        switch (*pack) {
 802        case 'D':
 803                return unpack_delta_entry(pack+5, size, left, type, sizep);
 804        case 'C':
 805                strcpy(type, "commit");
 806                break;
 807        case 'T':
 808                strcpy(type, "tree");
 809                break;
 810        case 'B':
 811                strcpy(type, "blob");
 812                break;
 813        case 'G':
 814                strcpy(type, "tag");
 815                break;
 816        default:
 817                die("corrupted pack file");
 818        }
 819        *sizep = size;
 820        return unpack_non_delta_entry(pack+5, size, left);
 821}
 822
 823static int find_pack_entry_1(const unsigned char *sha1,
 824                             struct pack_entry *e, struct packed_git *p)
 825{
 826        int *level1_ofs = p->index_base;
 827        int hi = ntohl(level1_ofs[*sha1]);
 828        int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
 829        void *index = p->index_base + 256;
 830
 831        do {
 832                int mi = (lo + hi) / 2;
 833                int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
 834                if (!cmp) {
 835                        e->offset = ntohl(*((int*)(index + 24 * mi)));
 836                        memcpy(e->sha1, sha1, 20);
 837                        e->p = p;
 838                        return 1;
 839                }
 840                if (cmp > 0)
 841                        hi = mi;
 842                else
 843                        lo = mi+1;
 844        } while (lo < hi);
 845        return 0;
 846}
 847
 848static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
 849{
 850        struct packed_git *p;
 851        prepare_packed_git();
 852
 853        for (p = packed_git; p; p = p->next) {
 854                if (find_pack_entry_1(sha1, e, p))
 855                        return 1;
 856        }
 857        return 0;
 858}
 859
 860int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
 861{
 862        int status;
 863        unsigned long mapsize, size;
 864        void *map;
 865        z_stream stream;
 866        char hdr[128];
 867
 868        map = map_sha1_file_internal(sha1, &mapsize, 0);
 869        if (!map) {
 870                struct pack_entry e;
 871
 872                if (!find_pack_entry(sha1, &e))
 873                        return error("unable to find %s", sha1_to_hex(sha1));
 874                if (!packed_object_info(&e, type, sizep))
 875                        return 0;
 876                /* sheesh */
 877                map = unpack_entry(&e, type, sizep);
 878                free(map);
 879                return (map == NULL) ? 0 : -1;
 880        }
 881        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
 882                status = error("unable to unpack %s header",
 883                               sha1_to_hex(sha1));
 884        if (parse_sha1_header(hdr, type, &size) < 0)
 885                status = error("unable to parse %s header", sha1_to_hex(sha1));
 886        else {
 887                status = 0;
 888                *sizep = size;
 889        }
 890        inflateEnd(&stream);
 891        munmap(map, mapsize);
 892        return status;
 893}
 894
 895static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
 896{
 897        struct pack_entry e;
 898
 899        if (!find_pack_entry(sha1, &e)) {
 900                error("cannot read sha1_file for %s", sha1_to_hex(sha1));
 901                return NULL;
 902        }
 903        return unpack_entry(&e, type, size);
 904}
 905
 906void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
 907{
 908        unsigned long mapsize;
 909        void *map, *buf;
 910
 911        map = map_sha1_file_internal(sha1, &mapsize, 0);
 912        if (map) {
 913                buf = unpack_sha1_file(map, mapsize, type, size);
 914                munmap(map, mapsize);
 915                return buf;
 916        }
 917        return read_packed_sha1(sha1, type, size);
 918}
 919
 920void *read_object_with_reference(const unsigned char *sha1,
 921                                 const char *required_type,
 922                                 unsigned long *size,
 923                                 unsigned char *actual_sha1_return)
 924{
 925        char type[20];
 926        void *buffer;
 927        unsigned long isize;
 928        unsigned char actual_sha1[20];
 929
 930        memcpy(actual_sha1, sha1, 20);
 931        while (1) {
 932                int ref_length = -1;
 933                const char *ref_type = NULL;
 934
 935                buffer = read_sha1_file(actual_sha1, type, &isize);
 936                if (!buffer)
 937                        return NULL;
 938                if (!strcmp(type, required_type)) {
 939                        *size = isize;
 940                        if (actual_sha1_return)
 941                                memcpy(actual_sha1_return, actual_sha1, 20);
 942                        return buffer;
 943                }
 944                /* Handle references */
 945                else if (!strcmp(type, "commit"))
 946                        ref_type = "tree ";
 947                else if (!strcmp(type, "tag"))
 948                        ref_type = "object ";
 949                else {
 950                        free(buffer);
 951                        return NULL;
 952                }
 953                ref_length = strlen(ref_type);
 954
 955                if (memcmp(buffer, ref_type, ref_length) ||
 956                    get_sha1_hex(buffer + ref_length, actual_sha1)) {
 957                        free(buffer);
 958                        return NULL;
 959                }
 960                /* Now we have the ID of the referred-to object in
 961                 * actual_sha1.  Check again. */
 962        }
 963}
 964
 965static char *write_sha1_file_prepare(void *buf,
 966                                     unsigned long len,
 967                                     const char *type,
 968                                     unsigned char *sha1,
 969                                     unsigned char *hdr,
 970                                     int *hdrlen)
 971{
 972        SHA_CTX c;
 973
 974        /* Generate the header */
 975        *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
 976
 977        /* Sha1.. */
 978        SHA1_Init(&c);
 979        SHA1_Update(&c, hdr, *hdrlen);
 980        SHA1_Update(&c, buf, len);
 981        SHA1_Final(sha1, &c);
 982
 983        return sha1_file_name(sha1);
 984}
 985
 986int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
 987{
 988        int size;
 989        unsigned char *compressed;
 990        z_stream stream;
 991        unsigned char sha1[20];
 992        char *filename;
 993        static char tmpfile[PATH_MAX];
 994        unsigned char hdr[50];
 995        int fd, hdrlen, ret;
 996
 997        /* Normally if we have it in the pack then we do not bother writing
 998         * it out into .git/objects/??/?{38} file.
 999         */
1000        filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1001        if (returnsha1)
1002                memcpy(returnsha1, sha1, 20);
1003        if (has_sha1_file(sha1))
1004                return 0;
1005        fd = open(filename, O_RDONLY);
1006        if (fd >= 0) {
1007                /*
1008                 * FIXME!!! We might do collision checking here, but we'd
1009                 * need to uncompress the old file and check it. Later.
1010                 */
1011                close(fd);
1012                return 0;
1013        }
1014
1015        if (errno != ENOENT) {
1016                fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
1017                return -1;
1018        }
1019
1020        snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1021
1022        fd = mkstemp(tmpfile);
1023        if (fd < 0) {
1024                fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
1025                return -1;
1026        }
1027
1028        /* Set it up */
1029        memset(&stream, 0, sizeof(stream));
1030        deflateInit(&stream, Z_BEST_COMPRESSION);
1031        size = deflateBound(&stream, len+hdrlen);
1032        compressed = xmalloc(size);
1033
1034        /* Compress it */
1035        stream.next_out = compressed;
1036        stream.avail_out = size;
1037
1038        /* First header.. */
1039        stream.next_in = hdr;
1040        stream.avail_in = hdrlen;
1041        while (deflate(&stream, 0) == Z_OK)
1042                /* nothing */;
1043
1044        /* Then the data itself.. */
1045        stream.next_in = buf;
1046        stream.avail_in = len;
1047        while (deflate(&stream, Z_FINISH) == Z_OK)
1048                /* nothing */;
1049        deflateEnd(&stream);
1050        size = stream.total_out;
1051
1052        if (write(fd, compressed, size) != size)
1053                die("unable to write file");
1054        fchmod(fd, 0444);
1055        close(fd);
1056        free(compressed);
1057
1058        ret = link(tmpfile, filename);
1059        if (ret < 0) {
1060                ret = errno;
1061
1062                /*
1063                 * Coda hack - coda doesn't like cross-directory links,
1064                 * so we fall back to a rename, which will mean that it
1065                 * won't be able to check collisions, but that's not a
1066                 * big deal.
1067                 *
1068                 * When this succeeds, we just return 0. We have nothing
1069                 * left to unlink.
1070                 */
1071                if (ret == EXDEV && !rename(tmpfile, filename))
1072                        return 0;
1073        }
1074        unlink(tmpfile);
1075        if (ret) {
1076                if (ret != EEXIST) {
1077                        fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
1078                        return -1;
1079                }
1080                /* FIXME!!! Collision check here ? */
1081        }
1082
1083        return 0;
1084}
1085
1086int write_sha1_from_fd(const unsigned char *sha1, int fd)
1087{
1088        char *filename = sha1_file_name(sha1);
1089
1090        int local;
1091        z_stream stream;
1092        unsigned char real_sha1[20];
1093        unsigned char buf[4096];
1094        unsigned char discard[4096];
1095        int ret;
1096        SHA_CTX c;
1097
1098        local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
1099
1100        if (local < 0)
1101                return error("Couldn't open %s\n", filename);
1102
1103        memset(&stream, 0, sizeof(stream));
1104
1105        inflateInit(&stream);
1106
1107        SHA1_Init(&c);
1108
1109        do {
1110                ssize_t size;
1111                size = read(fd, buf, 4096);
1112                if (size <= 0) {
1113                        close(local);
1114                        unlink(filename);
1115                        if (!size)
1116                                return error("Connection closed?");
1117                        perror("Reading from connection");
1118                        return -1;
1119                }
1120                write(local, buf, size);
1121                stream.avail_in = size;
1122                stream.next_in = buf;
1123                do {
1124                        stream.next_out = discard;
1125                        stream.avail_out = sizeof(discard);
1126                        ret = inflate(&stream, Z_SYNC_FLUSH);
1127                        SHA1_Update(&c, discard, sizeof(discard) -
1128                                    stream.avail_out);
1129                } while (stream.avail_in && ret == Z_OK);
1130                
1131        } while (ret == Z_OK);
1132        inflateEnd(&stream);
1133
1134        close(local);
1135        SHA1_Final(real_sha1, &c);
1136        if (ret != Z_STREAM_END) {
1137                unlink(filename);
1138                return error("File %s corrupted", sha1_to_hex(sha1));
1139        }
1140        if (memcmp(sha1, real_sha1, 20)) {
1141                unlink(filename);
1142                return error("File %s has bad hash\n", sha1_to_hex(sha1));
1143        }
1144        
1145        return 0;
1146}
1147
1148int has_sha1_file(const unsigned char *sha1)
1149{
1150        struct stat st;
1151        struct pack_entry e;
1152
1153        if (find_sha1_file(sha1, &st))
1154                return 1;
1155        return find_pack_entry(sha1, &e);
1156}
1157
1158int index_fd(unsigned char *sha1, int fd, struct stat *st)
1159{
1160        unsigned long size = st->st_size;
1161        void *buf;
1162        int ret;
1163
1164        buf = "";
1165        if (size)
1166                buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1167        close(fd);
1168        if ((int)(long)buf == -1)
1169                return -1;
1170
1171        ret = write_sha1_file(buf, size, "blob", sha1);
1172        if (size)
1173                munmap(buf, size);
1174        return ret;
1175}