a49d03d6de170dbffa63b55c07bf9425b704380c
   1#include "cache.h"
   2#include "delta.h"
   3#include "pack.h"
   4#include "csum-file.h"
   5#include "blob.h"
   6#include "commit.h"
   7#include "tag.h"
   8#include "tree.h"
   9#include "progress.h"
  10
  11static const char index_pack_usage[] =
  12"git-index-pack [-v] [-o <index-file>] [{ ---keep | --keep=<msg> }] { <pack-file> | --stdin [--fix-thin] [<pack-file>] }";
  13
  14struct object_entry
  15{
  16        off_t offset;
  17        unsigned long size;
  18        unsigned int hdr_size;
  19        uint32_t crc32;
  20        enum object_type type;
  21        enum object_type real_type;
  22        unsigned char sha1[20];
  23};
  24
  25union delta_base {
  26        unsigned char sha1[20];
  27        off_t offset;
  28};
  29
  30/*
  31 * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
  32 * to memcmp() only the first 20 bytes.
  33 */
  34#define UNION_BASE_SZ   20
  35
  36struct delta_entry
  37{
  38        union delta_base base;
  39        int obj_no;
  40};
  41
  42static struct object_entry *objects;
  43static struct delta_entry *deltas;
  44static int nr_objects;
  45static int nr_deltas;
  46static int nr_resolved_deltas;
  47
  48static int from_stdin;
  49static int verbose;
  50
  51static struct progress progress;
  52
  53/* We always read in 4kB chunks. */
  54static unsigned char input_buffer[4096];
  55static unsigned int input_offset, input_len;
  56static off_t consumed_bytes;
  57static SHA_CTX input_ctx;
  58static uint32_t input_crc32;
  59static int input_fd, output_fd, pack_fd;
  60
  61/* Discard current buffer used content. */
  62static void flush(void)
  63{
  64        if (input_offset) {
  65                if (output_fd >= 0)
  66                        write_or_die(output_fd, input_buffer, input_offset);
  67                SHA1_Update(&input_ctx, input_buffer, input_offset);
  68                memmove(input_buffer, input_buffer + input_offset, input_len);
  69                input_offset = 0;
  70        }
  71}
  72
  73/*
  74 * Make sure at least "min" bytes are available in the buffer, and
  75 * return the pointer to the buffer.
  76 */
  77static void *fill(int min)
  78{
  79        if (min <= input_len)
  80                return input_buffer + input_offset;
  81        if (min > sizeof(input_buffer))
  82                die("cannot fill %d bytes", min);
  83        flush();
  84        do {
  85                int ret = xread(input_fd, input_buffer + input_len,
  86                                sizeof(input_buffer) - input_len);
  87                if (ret <= 0) {
  88                        if (!ret)
  89                                die("early EOF");
  90                        die("read error on input: %s", strerror(errno));
  91                }
  92                input_len += ret;
  93        } while (input_len < min);
  94        return input_buffer;
  95}
  96
  97static void use(int bytes)
  98{
  99        if (bytes > input_len)
 100                die("used more bytes than were available");
 101        input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
 102        input_len -= bytes;
 103        input_offset += bytes;
 104
 105        /* make sure off_t is sufficiently large not to wrap */
 106        if (consumed_bytes > consumed_bytes + bytes)
 107                die("pack too large for current definition of off_t");
 108        consumed_bytes += bytes;
 109}
 110
 111static const char *open_pack_file(const char *pack_name)
 112{
 113        if (from_stdin) {
 114                input_fd = 0;
 115                if (!pack_name) {
 116                        static char tmpfile[PATH_MAX];
 117                        snprintf(tmpfile, sizeof(tmpfile),
 118                                 "%s/tmp_pack_XXXXXX", get_object_directory());
 119                        output_fd = mkstemp(tmpfile);
 120                        pack_name = xstrdup(tmpfile);
 121                } else
 122                        output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
 123                if (output_fd < 0)
 124                        die("unable to create %s: %s\n", pack_name, strerror(errno));
 125                pack_fd = output_fd;
 126        } else {
 127                input_fd = open(pack_name, O_RDONLY);
 128                if (input_fd < 0)
 129                        die("cannot open packfile '%s': %s",
 130                            pack_name, strerror(errno));
 131                output_fd = -1;
 132                pack_fd = input_fd;
 133        }
 134        SHA1_Init(&input_ctx);
 135        return pack_name;
 136}
 137
 138static void parse_pack_header(void)
 139{
 140        struct pack_header *hdr = fill(sizeof(struct pack_header));
 141
 142        /* Header consistency check */
 143        if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
 144                die("pack signature mismatch");
 145        if (!pack_version_ok(hdr->hdr_version))
 146                die("pack version %d unsupported", ntohl(hdr->hdr_version));
 147
 148        nr_objects = ntohl(hdr->hdr_entries);
 149        use(sizeof(struct pack_header));
 150}
 151
 152static void bad_object(unsigned long offset, const char *format,
 153                       ...) NORETURN __attribute__((format (printf, 2, 3)));
 154
 155static void bad_object(unsigned long offset, const char *format, ...)
 156{
 157        va_list params;
 158        char buf[1024];
 159
 160        va_start(params, format);
 161        vsnprintf(buf, sizeof(buf), format, params);
 162        va_end(params);
 163        die("pack has bad object at offset %lu: %s", offset, buf);
 164}
 165
 166static void *unpack_entry_data(unsigned long offset, unsigned long size)
 167{
 168        z_stream stream;
 169        void *buf = xmalloc(size);
 170
 171        memset(&stream, 0, sizeof(stream));
 172        stream.next_out = buf;
 173        stream.avail_out = size;
 174        stream.next_in = fill(1);
 175        stream.avail_in = input_len;
 176        inflateInit(&stream);
 177
 178        for (;;) {
 179                int ret = inflate(&stream, 0);
 180                use(input_len - stream.avail_in);
 181                if (stream.total_out == size && ret == Z_STREAM_END)
 182                        break;
 183                if (ret != Z_OK)
 184                        bad_object(offset, "inflate returned %d", ret);
 185                stream.next_in = fill(1);
 186                stream.avail_in = input_len;
 187        }
 188        inflateEnd(&stream);
 189        return buf;
 190}
 191
 192static void *unpack_raw_entry(struct object_entry *obj, union delta_base *delta_base)
 193{
 194        unsigned char *p, c;
 195        unsigned long size;
 196        off_t base_offset;
 197        unsigned shift;
 198        void *data;
 199
 200        obj->offset = consumed_bytes;
 201        input_crc32 = crc32(0, Z_NULL, 0);
 202
 203        p = fill(1);
 204        c = *p;
 205        use(1);
 206        obj->type = (c >> 4) & 7;
 207        size = (c & 15);
 208        shift = 4;
 209        while (c & 0x80) {
 210                p = fill(1);
 211                c = *p;
 212                use(1);
 213                size += (c & 0x7fUL) << shift;
 214                shift += 7;
 215        }
 216        obj->size = size;
 217
 218        switch (obj->type) {
 219        case OBJ_REF_DELTA:
 220                hashcpy(delta_base->sha1, fill(20));
 221                use(20);
 222                break;
 223        case OBJ_OFS_DELTA:
 224                memset(delta_base, 0, sizeof(*delta_base));
 225                p = fill(1);
 226                c = *p;
 227                use(1);
 228                base_offset = c & 127;
 229                while (c & 128) {
 230                        base_offset += 1;
 231                        if (!base_offset || MSB(base_offset, 7))
 232                                bad_object(obj->offset, "offset value overflow for delta base object");
 233                        p = fill(1);
 234                        c = *p;
 235                        use(1);
 236                        base_offset = (base_offset << 7) + (c & 127);
 237                }
 238                delta_base->offset = obj->offset - base_offset;
 239                if (delta_base->offset >= obj->offset)
 240                        bad_object(obj->offset, "delta base offset is out of bound");
 241                break;
 242        case OBJ_COMMIT:
 243        case OBJ_TREE:
 244        case OBJ_BLOB:
 245        case OBJ_TAG:
 246                break;
 247        default:
 248                bad_object(obj->offset, "unknown object type %d", obj->type);
 249        }
 250        obj->hdr_size = consumed_bytes - obj->offset;
 251
 252        data = unpack_entry_data(obj->offset, obj->size);
 253        obj->crc32 = input_crc32;
 254        return data;
 255}
 256
 257static void *get_data_from_pack(struct object_entry *obj)
 258{
 259        unsigned long from = obj[0].offset + obj[0].hdr_size;
 260        unsigned long len = obj[1].offset - from;
 261        unsigned long rdy = 0;
 262        unsigned char *src, *data;
 263        z_stream stream;
 264        int st;
 265
 266        src = xmalloc(len);
 267        data = src;
 268        do {
 269                ssize_t n = pread(pack_fd, data + rdy, len - rdy, from + rdy);
 270                if (n <= 0)
 271                        die("cannot pread pack file: %s", strerror(errno));
 272                rdy += n;
 273        } while (rdy < len);
 274        data = xmalloc(obj->size);
 275        memset(&stream, 0, sizeof(stream));
 276        stream.next_out = data;
 277        stream.avail_out = obj->size;
 278        stream.next_in = src;
 279        stream.avail_in = len;
 280        inflateInit(&stream);
 281        while ((st = inflate(&stream, Z_FINISH)) == Z_OK);
 282        inflateEnd(&stream);
 283        if (st != Z_STREAM_END || stream.total_out != obj->size)
 284                die("serious inflate inconsistency");
 285        free(src);
 286        return data;
 287}
 288
 289static int find_delta(const union delta_base *base)
 290{
 291        int first = 0, last = nr_deltas;
 292
 293        while (first < last) {
 294                int next = (first + last) / 2;
 295                struct delta_entry *delta = &deltas[next];
 296                int cmp;
 297
 298                cmp = memcmp(base, &delta->base, UNION_BASE_SZ);
 299                if (!cmp)
 300                        return next;
 301                if (cmp < 0) {
 302                        last = next;
 303                        continue;
 304                }
 305                first = next+1;
 306        }
 307        return -first-1;
 308}
 309
 310static int find_delta_children(const union delta_base *base,
 311                               int *first_index, int *last_index)
 312{
 313        int first = find_delta(base);
 314        int last = first;
 315        int end = nr_deltas - 1;
 316
 317        if (first < 0)
 318                return -1;
 319        while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
 320                --first;
 321        while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
 322                ++last;
 323        *first_index = first;
 324        *last_index = last;
 325        return 0;
 326}
 327
 328static void sha1_object(const void *data, unsigned long size,
 329                        enum object_type type, unsigned char *sha1)
 330{
 331        hash_sha1_file(data, size, typename(type), sha1);
 332        if (has_sha1_file(sha1)) {
 333                void *has_data;
 334                enum object_type has_type;
 335                unsigned long has_size;
 336                has_data = read_sha1_file(sha1, &has_type, &has_size);
 337                if (!has_data)
 338                        die("cannot read existing object %s", sha1_to_hex(sha1));
 339                if (size != has_size || type != has_type ||
 340                    memcmp(data, has_data, size) != 0)
 341                        die("SHA1 COLLISION FOUND WITH %s !", sha1_to_hex(sha1));
 342                free(has_data);
 343        }
 344}
 345
 346static void resolve_delta(struct object_entry *delta_obj, void *base_data,
 347                          unsigned long base_size, enum object_type type)
 348{
 349        void *delta_data;
 350        unsigned long delta_size;
 351        void *result;
 352        unsigned long result_size;
 353        union delta_base delta_base;
 354        int j, first, last;
 355
 356        delta_obj->real_type = type;
 357        delta_data = get_data_from_pack(delta_obj);
 358        delta_size = delta_obj->size;
 359        result = patch_delta(base_data, base_size, delta_data, delta_size,
 360                             &result_size);
 361        free(delta_data);
 362        if (!result)
 363                bad_object(delta_obj->offset, "failed to apply delta");
 364        sha1_object(result, result_size, type, delta_obj->sha1);
 365        nr_resolved_deltas++;
 366
 367        hashcpy(delta_base.sha1, delta_obj->sha1);
 368        if (!find_delta_children(&delta_base, &first, &last)) {
 369                for (j = first; j <= last; j++) {
 370                        struct object_entry *child = objects + deltas[j].obj_no;
 371                        if (child->real_type == OBJ_REF_DELTA)
 372                                resolve_delta(child, result, result_size, type);
 373                }
 374        }
 375
 376        memset(&delta_base, 0, sizeof(delta_base));
 377        delta_base.offset = delta_obj->offset;
 378        if (!find_delta_children(&delta_base, &first, &last)) {
 379                for (j = first; j <= last; j++) {
 380                        struct object_entry *child = objects + deltas[j].obj_no;
 381                        if (child->real_type == OBJ_OFS_DELTA)
 382                                resolve_delta(child, result, result_size, type);
 383                }
 384        }
 385
 386        free(result);
 387}
 388
 389static int compare_delta_entry(const void *a, const void *b)
 390{
 391        const struct delta_entry *delta_a = a;
 392        const struct delta_entry *delta_b = b;
 393        return memcmp(&delta_a->base, &delta_b->base, UNION_BASE_SZ);
 394}
 395
 396/* Parse all objects and return the pack content SHA1 hash */
 397static void parse_pack_objects(unsigned char *sha1)
 398{
 399        int i;
 400        struct delta_entry *delta = deltas;
 401        void *data;
 402        struct stat st;
 403
 404        /*
 405         * First pass:
 406         * - find locations of all objects;
 407         * - calculate SHA1 of all non-delta objects;
 408         * - remember base (SHA1 or offset) for all deltas.
 409         */
 410        if (verbose) {
 411                fprintf(stderr, "Indexing %d objects.\n", nr_objects);
 412                start_progress(&progress, "", nr_objects);
 413        }
 414        for (i = 0; i < nr_objects; i++) {
 415                struct object_entry *obj = &objects[i];
 416                data = unpack_raw_entry(obj, &delta->base);
 417                obj->real_type = obj->type;
 418                if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) {
 419                        nr_deltas++;
 420                        delta->obj_no = i;
 421                        delta++;
 422                } else
 423                        sha1_object(data, obj->size, obj->type, obj->sha1);
 424                free(data);
 425                if (verbose)
 426                        display_progress(&progress, i+1);
 427        }
 428        objects[i].offset = consumed_bytes;
 429        if (verbose)
 430                stop_progress(&progress);
 431
 432        /* Check pack integrity */
 433        flush();
 434        SHA1_Final(sha1, &input_ctx);
 435        if (hashcmp(fill(20), sha1))
 436                die("pack is corrupted (SHA1 mismatch)");
 437        use(20);
 438
 439        /* If input_fd is a file, we should have reached its end now. */
 440        if (fstat(input_fd, &st))
 441                die("cannot fstat packfile: %s", strerror(errno));
 442        if (S_ISREG(st.st_mode) &&
 443                        lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
 444                die("pack has junk at the end");
 445
 446        if (!nr_deltas)
 447                return;
 448
 449        /* Sort deltas by base SHA1/offset for fast searching */
 450        qsort(deltas, nr_deltas, sizeof(struct delta_entry),
 451              compare_delta_entry);
 452
 453        /*
 454         * Second pass:
 455         * - for all non-delta objects, look if it is used as a base for
 456         *   deltas;
 457         * - if used as a base, uncompress the object and apply all deltas,
 458         *   recursively checking if the resulting object is used as a base
 459         *   for some more deltas.
 460         */
 461        if (verbose) {
 462                fprintf(stderr, "Resolving %d deltas.\n", nr_deltas);
 463                start_progress(&progress, "", nr_deltas);
 464        }
 465        for (i = 0; i < nr_objects; i++) {
 466                struct object_entry *obj = &objects[i];
 467                union delta_base base;
 468                int j, ref, ref_first, ref_last, ofs, ofs_first, ofs_last;
 469
 470                if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA)
 471                        continue;
 472                hashcpy(base.sha1, obj->sha1);
 473                ref = !find_delta_children(&base, &ref_first, &ref_last);
 474                memset(&base, 0, sizeof(base));
 475                base.offset = obj->offset;
 476                ofs = !find_delta_children(&base, &ofs_first, &ofs_last);
 477                if (!ref && !ofs)
 478                        continue;
 479                data = get_data_from_pack(obj);
 480                if (ref)
 481                        for (j = ref_first; j <= ref_last; j++) {
 482                                struct object_entry *child = objects + deltas[j].obj_no;
 483                                if (child->real_type == OBJ_REF_DELTA)
 484                                        resolve_delta(child, data,
 485                                                      obj->size, obj->type);
 486                        }
 487                if (ofs)
 488                        for (j = ofs_first; j <= ofs_last; j++) {
 489                                struct object_entry *child = objects + deltas[j].obj_no;
 490                                if (child->real_type == OBJ_OFS_DELTA)
 491                                        resolve_delta(child, data,
 492                                                      obj->size, obj->type);
 493                        }
 494                free(data);
 495                if (verbose)
 496                        display_progress(&progress, nr_resolved_deltas);
 497        }
 498}
 499
 500static int write_compressed(int fd, void *in, unsigned int size, uint32_t *obj_crc)
 501{
 502        z_stream stream;
 503        unsigned long maxsize;
 504        void *out;
 505
 506        memset(&stream, 0, sizeof(stream));
 507        deflateInit(&stream, zlib_compression_level);
 508        maxsize = deflateBound(&stream, size);
 509        out = xmalloc(maxsize);
 510
 511        /* Compress it */
 512        stream.next_in = in;
 513        stream.avail_in = size;
 514        stream.next_out = out;
 515        stream.avail_out = maxsize;
 516        while (deflate(&stream, Z_FINISH) == Z_OK);
 517        deflateEnd(&stream);
 518
 519        size = stream.total_out;
 520        write_or_die(fd, out, size);
 521        *obj_crc = crc32(*obj_crc, out, size);
 522        free(out);
 523        return size;
 524}
 525
 526static void append_obj_to_pack(const unsigned char *sha1, void *buf,
 527                               unsigned long size, enum object_type type)
 528{
 529        struct object_entry *obj = &objects[nr_objects++];
 530        unsigned char header[10];
 531        unsigned long s = size;
 532        int n = 0;
 533        unsigned char c = (type << 4) | (s & 15);
 534        s >>= 4;
 535        while (s) {
 536                header[n++] = c | 0x80;
 537                c = s & 0x7f;
 538                s >>= 7;
 539        }
 540        header[n++] = c;
 541        write_or_die(output_fd, header, n);
 542        obj[0].crc32 = crc32(0, Z_NULL, 0);
 543        obj[0].crc32 = crc32(obj[0].crc32, header, n);
 544        obj[1].offset = obj[0].offset + n;
 545        obj[1].offset += write_compressed(output_fd, buf, size, &obj[0].crc32);
 546        hashcpy(obj->sha1, sha1);
 547}
 548
 549static int delta_pos_compare(const void *_a, const void *_b)
 550{
 551        struct delta_entry *a = *(struct delta_entry **)_a;
 552        struct delta_entry *b = *(struct delta_entry **)_b;
 553        return a->obj_no - b->obj_no;
 554}
 555
 556static void fix_unresolved_deltas(int nr_unresolved)
 557{
 558        struct delta_entry **sorted_by_pos;
 559        int i, n = 0;
 560
 561        /*
 562         * Since many unresolved deltas may well be themselves base objects
 563         * for more unresolved deltas, we really want to include the
 564         * smallest number of base objects that would cover as much delta
 565         * as possible by picking the
 566         * trunc deltas first, allowing for other deltas to resolve without
 567         * additional base objects.  Since most base objects are to be found
 568         * before deltas depending on them, a good heuristic is to start
 569         * resolving deltas in the same order as their position in the pack.
 570         */
 571        sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
 572        for (i = 0; i < nr_deltas; i++) {
 573                if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
 574                        continue;
 575                sorted_by_pos[n++] = &deltas[i];
 576        }
 577        qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
 578
 579        for (i = 0; i < n; i++) {
 580                struct delta_entry *d = sorted_by_pos[i];
 581                void *data;
 582                unsigned long size;
 583                enum object_type type;
 584                int j, first, last;
 585
 586                if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
 587                        continue;
 588                data = read_sha1_file(d->base.sha1, &type, &size);
 589                if (!data)
 590                        continue;
 591
 592                find_delta_children(&d->base, &first, &last);
 593                for (j = first; j <= last; j++) {
 594                        struct object_entry *child = objects + deltas[j].obj_no;
 595                        if (child->real_type == OBJ_REF_DELTA)
 596                                resolve_delta(child, data, size, type);
 597                }
 598
 599                if (check_sha1_signature(d->base.sha1, data, size, typename(type)))
 600                        die("local object %s is corrupt", sha1_to_hex(d->base.sha1));
 601                append_obj_to_pack(d->base.sha1, data, size, type);
 602                free(data);
 603                if (verbose)
 604                        display_progress(&progress, nr_resolved_deltas);
 605        }
 606        free(sorted_by_pos);
 607}
 608
 609static void readjust_pack_header_and_sha1(unsigned char *sha1)
 610{
 611        struct pack_header hdr;
 612        SHA_CTX ctx;
 613        int size;
 614
 615        /* Rewrite pack header with updated object number */
 616        if (lseek(output_fd, 0, SEEK_SET) != 0)
 617                die("cannot seek back: %s", strerror(errno));
 618        if (read_in_full(output_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 619                die("cannot read pack header back: %s", strerror(errno));
 620        hdr.hdr_entries = htonl(nr_objects);
 621        if (lseek(output_fd, 0, SEEK_SET) != 0)
 622                die("cannot seek back: %s", strerror(errno));
 623        write_or_die(output_fd, &hdr, sizeof(hdr));
 624        if (lseek(output_fd, 0, SEEK_SET) != 0)
 625                die("cannot seek back: %s", strerror(errno));
 626
 627        /* Recompute and store the new pack's SHA1 */
 628        SHA1_Init(&ctx);
 629        do {
 630                unsigned char *buf[4096];
 631                size = xread(output_fd, buf, sizeof(buf));
 632                if (size < 0)
 633                        die("cannot read pack data back: %s", strerror(errno));
 634                SHA1_Update(&ctx, buf, size);
 635        } while (size > 0);
 636        SHA1_Final(sha1, &ctx);
 637        write_or_die(output_fd, sha1, 20);
 638}
 639
 640static uint32_t index_default_version = 1;
 641static uint32_t index_off32_limit = 0x7fffffff;
 642
 643static int sha1_compare(const void *_a, const void *_b)
 644{
 645        struct object_entry *a = *(struct object_entry **)_a;
 646        struct object_entry *b = *(struct object_entry **)_b;
 647        return hashcmp(a->sha1, b->sha1);
 648}
 649
 650/*
 651 * On entry *sha1 contains the pack content SHA1 hash, on exit it is
 652 * the SHA1 hash of sorted object names.
 653 */
 654static const char *write_index_file(const char *index_name, unsigned char *sha1)
 655{
 656        struct sha1file *f;
 657        struct object_entry **sorted_by_sha, **list, **last;
 658        uint32_t array[256];
 659        int i, fd;
 660        SHA_CTX ctx;
 661        uint32_t index_version;
 662
 663        if (nr_objects) {
 664                sorted_by_sha =
 665                        xcalloc(nr_objects, sizeof(struct object_entry *));
 666                list = sorted_by_sha;
 667                last = sorted_by_sha + nr_objects;
 668                for (i = 0; i < nr_objects; ++i)
 669                        sorted_by_sha[i] = &objects[i];
 670                qsort(sorted_by_sha, nr_objects, sizeof(sorted_by_sha[0]),
 671                      sha1_compare);
 672        }
 673        else
 674                sorted_by_sha = list = last = NULL;
 675
 676        if (!index_name) {
 677                static char tmpfile[PATH_MAX];
 678                snprintf(tmpfile, sizeof(tmpfile),
 679                         "%s/tmp_idx_XXXXXX", get_object_directory());
 680                fd = mkstemp(tmpfile);
 681                index_name = xstrdup(tmpfile);
 682        } else {
 683                unlink(index_name);
 684                fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
 685        }
 686        if (fd < 0)
 687                die("unable to create %s: %s", index_name, strerror(errno));
 688        f = sha1fd(fd, index_name);
 689
 690        /* if last object's offset is >= 2^31 we should use index V2 */
 691        index_version = (objects[nr_objects-1].offset >> 31) ? 2 : index_default_version;
 692
 693        /* index versions 2 and above need a header */
 694        if (index_version >= 2) {
 695                struct pack_idx_header hdr;
 696                hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
 697                hdr.idx_version = htonl(index_version);
 698                sha1write(f, &hdr, sizeof(hdr));
 699        }
 700
 701        /*
 702         * Write the first-level table (the list is sorted,
 703         * but we use a 256-entry lookup to be able to avoid
 704         * having to do eight extra binary search iterations).
 705         */
 706        for (i = 0; i < 256; i++) {
 707                struct object_entry **next = list;
 708                while (next < last) {
 709                        struct object_entry *obj = *next;
 710                        if (obj->sha1[0] != i)
 711                                break;
 712                        next++;
 713                }
 714                array[i] = htonl(next - sorted_by_sha);
 715                list = next;
 716        }
 717        sha1write(f, array, 256 * 4);
 718
 719        /* compute the SHA1 hash of sorted object names. */
 720        SHA1_Init(&ctx);
 721
 722        /*
 723         * Write the actual SHA1 entries..
 724         */
 725        list = sorted_by_sha;
 726        for (i = 0; i < nr_objects; i++) {
 727                struct object_entry *obj = *list++;
 728                if (index_version < 2) {
 729                        uint32_t offset = htonl(obj->offset);
 730                        sha1write(f, &offset, 4);
 731                }
 732                sha1write(f, obj->sha1, 20);
 733                SHA1_Update(&ctx, obj->sha1, 20);
 734        }
 735
 736        if (index_version >= 2) {
 737                unsigned int nr_large_offset = 0;
 738
 739                /* write the crc32 table */
 740                list = sorted_by_sha;
 741                for (i = 0; i < nr_objects; i++) {
 742                        struct object_entry *obj = *list++;
 743                        uint32_t crc32_val = htonl(obj->crc32);
 744                        sha1write(f, &crc32_val, 4);
 745                }
 746
 747                /* write the 32-bit offset table */
 748                list = sorted_by_sha;
 749                for (i = 0; i < nr_objects; i++) {
 750                        struct object_entry *obj = *list++;
 751                        uint32_t offset = (obj->offset <= index_off32_limit) ?
 752                                obj->offset : (0x80000000 | nr_large_offset++);
 753                        offset = htonl(offset);
 754                        sha1write(f, &offset, 4);
 755                }
 756
 757                /* write the large offset table */
 758                list = sorted_by_sha;
 759                while (nr_large_offset) {
 760                        struct object_entry *obj = *list++;
 761                        uint64_t offset = obj->offset;
 762                        if (offset > index_off32_limit) {
 763                                uint32_t split[2];
 764                                split[0]        = htonl(offset >> 32);
 765                                split[1] = htonl(offset & 0xffffffff);
 766                                sha1write(f, split, 8);
 767                                nr_large_offset--;
 768                        }
 769                }
 770        }
 771
 772        sha1write(f, sha1, 20);
 773        sha1close(f, NULL, 1);
 774        free(sorted_by_sha);
 775        SHA1_Final(sha1, &ctx);
 776        return index_name;
 777}
 778
 779static void final(const char *final_pack_name, const char *curr_pack_name,
 780                  const char *final_index_name, const char *curr_index_name,
 781                  const char *keep_name, const char *keep_msg,
 782                  unsigned char *sha1)
 783{
 784        const char *report = "pack";
 785        char name[PATH_MAX];
 786        int err;
 787
 788        if (!from_stdin) {
 789                close(input_fd);
 790        } else {
 791                err = close(output_fd);
 792                if (err)
 793                        die("error while closing pack file: %s", strerror(errno));
 794                chmod(curr_pack_name, 0444);
 795        }
 796
 797        if (keep_msg) {
 798                int keep_fd, keep_msg_len = strlen(keep_msg);
 799                if (!keep_name) {
 800                        snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 801                                 get_object_directory(), sha1_to_hex(sha1));
 802                        keep_name = name;
 803                }
 804                keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
 805                if (keep_fd < 0) {
 806                        if (errno != EEXIST)
 807                                die("cannot write keep file");
 808                } else {
 809                        if (keep_msg_len > 0) {
 810                                write_or_die(keep_fd, keep_msg, keep_msg_len);
 811                                write_or_die(keep_fd, "\n", 1);
 812                        }
 813                        close(keep_fd);
 814                        report = "keep";
 815                }
 816        }
 817
 818        if (final_pack_name != curr_pack_name) {
 819                if (!final_pack_name) {
 820                        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 821                                 get_object_directory(), sha1_to_hex(sha1));
 822                        final_pack_name = name;
 823                }
 824                if (move_temp_to_file(curr_pack_name, final_pack_name))
 825                        die("cannot store pack file");
 826        }
 827
 828        chmod(curr_index_name, 0444);
 829        if (final_index_name != curr_index_name) {
 830                if (!final_index_name) {
 831                        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 832                                 get_object_directory(), sha1_to_hex(sha1));
 833                        final_index_name = name;
 834                }
 835                if (move_temp_to_file(curr_index_name, final_index_name))
 836                        die("cannot store index file");
 837        }
 838
 839        if (!from_stdin) {
 840                printf("%s\n", sha1_to_hex(sha1));
 841        } else {
 842                char buf[48];
 843                int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
 844                                   report, sha1_to_hex(sha1));
 845                write_or_die(1, buf, len);
 846
 847                /*
 848                 * Let's just mimic git-unpack-objects here and write
 849                 * the last part of the input buffer to stdout.
 850                 */
 851                while (input_len) {
 852                        err = xwrite(1, input_buffer + input_offset, input_len);
 853                        if (err <= 0)
 854                                break;
 855                        input_len -= err;
 856                        input_offset += err;
 857                }
 858        }
 859}
 860
 861int main(int argc, char **argv)
 862{
 863        int i, fix_thin_pack = 0;
 864        const char *curr_pack, *pack_name = NULL;
 865        const char *curr_index, *index_name = NULL;
 866        const char *keep_name = NULL, *keep_msg = NULL;
 867        char *index_name_buf = NULL, *keep_name_buf = NULL;
 868        unsigned char sha1[20];
 869
 870        for (i = 1; i < argc; i++) {
 871                const char *arg = argv[i];
 872
 873                if (*arg == '-') {
 874                        if (!strcmp(arg, "--stdin")) {
 875                                from_stdin = 1;
 876                        } else if (!strcmp(arg, "--fix-thin")) {
 877                                fix_thin_pack = 1;
 878                        } else if (!strcmp(arg, "--keep")) {
 879                                keep_msg = "";
 880                        } else if (!prefixcmp(arg, "--keep=")) {
 881                                keep_msg = arg + 7;
 882                        } else if (!prefixcmp(arg, "--pack_header=")) {
 883                                struct pack_header *hdr;
 884                                char *c;
 885
 886                                hdr = (struct pack_header *)input_buffer;
 887                                hdr->hdr_signature = htonl(PACK_SIGNATURE);
 888                                hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
 889                                if (*c != ',')
 890                                        die("bad %s", arg);
 891                                hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
 892                                if (*c)
 893                                        die("bad %s", arg);
 894                                input_len = sizeof(*hdr);
 895                        } else if (!strcmp(arg, "-v")) {
 896                                verbose = 1;
 897                        } else if (!strcmp(arg, "-o")) {
 898                                if (index_name || (i+1) >= argc)
 899                                        usage(index_pack_usage);
 900                                index_name = argv[++i];
 901                        } else if (!prefixcmp(arg, "--index-version=")) {
 902                                char *c;
 903                                index_default_version = strtoul(arg + 16, &c, 10);
 904                                if (index_default_version > 2)
 905                                        die("bad %s", arg);
 906                                if (*c == ',')
 907                                        index_off32_limit = strtoul(c+1, &c, 0);
 908                                if (*c || index_off32_limit & 0x80000000)
 909                                        die("bad %s", arg);
 910                        } else
 911                                usage(index_pack_usage);
 912                        continue;
 913                }
 914
 915                if (pack_name)
 916                        usage(index_pack_usage);
 917                pack_name = arg;
 918        }
 919
 920        if (!pack_name && !from_stdin)
 921                usage(index_pack_usage);
 922        if (fix_thin_pack && !from_stdin)
 923                die("--fix-thin cannot be used without --stdin");
 924        if (!index_name && pack_name) {
 925                int len = strlen(pack_name);
 926                if (!has_extension(pack_name, ".pack"))
 927                        die("packfile name '%s' does not end with '.pack'",
 928                            pack_name);
 929                index_name_buf = xmalloc(len);
 930                memcpy(index_name_buf, pack_name, len - 5);
 931                strcpy(index_name_buf + len - 5, ".idx");
 932                index_name = index_name_buf;
 933        }
 934        if (keep_msg && !keep_name && pack_name) {
 935                int len = strlen(pack_name);
 936                if (!has_extension(pack_name, ".pack"))
 937                        die("packfile name '%s' does not end with '.pack'",
 938                            pack_name);
 939                keep_name_buf = xmalloc(len);
 940                memcpy(keep_name_buf, pack_name, len - 5);
 941                strcpy(keep_name_buf + len - 5, ".keep");
 942                keep_name = keep_name_buf;
 943        }
 944
 945        curr_pack = open_pack_file(pack_name);
 946        parse_pack_header();
 947        objects = xmalloc((nr_objects + 1) * sizeof(struct object_entry));
 948        deltas = xmalloc(nr_objects * sizeof(struct delta_entry));
 949        parse_pack_objects(sha1);
 950        if (nr_deltas == nr_resolved_deltas) {
 951                if (verbose)
 952                        stop_progress(&progress);
 953                /* Flush remaining pack final 20-byte SHA1. */
 954                flush();
 955        } else {
 956                if (fix_thin_pack) {
 957                        int nr_unresolved = nr_deltas - nr_resolved_deltas;
 958                        int nr_objects_initial = nr_objects;
 959                        if (nr_unresolved <= 0)
 960                                die("confusion beyond insanity");
 961                        objects = xrealloc(objects,
 962                                           (nr_objects + nr_unresolved + 1)
 963                                           * sizeof(*objects));
 964                        fix_unresolved_deltas(nr_unresolved);
 965                        if (verbose) {
 966                                stop_progress(&progress);
 967                                fprintf(stderr, "%d objects were added to complete this thin pack.\n",
 968                                        nr_objects - nr_objects_initial);
 969                        }
 970                        readjust_pack_header_and_sha1(sha1);
 971                }
 972                if (nr_deltas != nr_resolved_deltas)
 973                        die("pack has %d unresolved deltas",
 974                            nr_deltas - nr_resolved_deltas);
 975        }
 976        free(deltas);
 977        curr_index = write_index_file(index_name, sha1);
 978        final(pack_name, curr_pack,
 979                index_name, curr_index,
 980                keep_name, keep_msg,
 981                sha1);
 982        free(objects);
 983        free(index_name_buf);
 984        free(keep_name_buf);
 985
 986        return 0;
 987}