builtin / index-pack.con commit Merge branch 'jk/peel-ref' (315ea32)
   1#include "builtin.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#include "fsck.h"
  11#include "exec_cmd.h"
  12#include "streaming.h"
  13#include "thread-utils.h"
  14
  15static const char index_pack_usage[] =
  16"git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
  17
  18struct object_entry {
  19        struct pack_idx_entry idx;
  20        unsigned long size;
  21        unsigned int hdr_size;
  22        enum object_type type;
  23        enum object_type real_type;
  24        unsigned delta_depth;
  25        int base_object_no;
  26};
  27
  28union delta_base {
  29        unsigned char sha1[20];
  30        off_t offset;
  31};
  32
  33struct base_data {
  34        struct base_data *base;
  35        struct base_data *child;
  36        struct object_entry *obj;
  37        void *data;
  38        unsigned long size;
  39        int ref_first, ref_last;
  40        int ofs_first, ofs_last;
  41};
  42
  43#if !defined(NO_PTHREADS) && defined(NO_THREAD_SAFE_PREAD)
  44/* pread() emulation is not thread-safe. Disable threading. */
  45#define NO_PTHREADS
  46#endif
  47
  48struct thread_local {
  49#ifndef NO_PTHREADS
  50        pthread_t thread;
  51#endif
  52        struct base_data *base_cache;
  53        size_t base_cache_used;
  54};
  55
  56/*
  57 * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
  58 * to memcmp() only the first 20 bytes.
  59 */
  60#define UNION_BASE_SZ   20
  61
  62#define FLAG_LINK (1u<<20)
  63#define FLAG_CHECKED (1u<<21)
  64
  65struct delta_entry {
  66        union delta_base base;
  67        int obj_no;
  68};
  69
  70static struct object_entry *objects;
  71static struct delta_entry *deltas;
  72static struct thread_local nothread_data;
  73static int nr_objects;
  74static int nr_deltas;
  75static int nr_resolved_deltas;
  76static int nr_threads;
  77
  78static int from_stdin;
  79static int strict;
  80static int verbose;
  81
  82static struct progress *progress;
  83
  84/* We always read in 4kB chunks. */
  85static unsigned char input_buffer[4096];
  86static unsigned int input_offset, input_len;
  87static off_t consumed_bytes;
  88static unsigned deepest_delta;
  89static git_SHA_CTX input_ctx;
  90static uint32_t input_crc32;
  91static int input_fd, output_fd, pack_fd;
  92
  93#ifndef NO_PTHREADS
  94
  95static struct thread_local *thread_data;
  96static int nr_dispatched;
  97static int threads_active;
  98
  99static pthread_mutex_t read_mutex;
 100#define read_lock()             lock_mutex(&read_mutex)
 101#define read_unlock()           unlock_mutex(&read_mutex)
 102
 103static pthread_mutex_t counter_mutex;
 104#define counter_lock()          lock_mutex(&counter_mutex)
 105#define counter_unlock()        unlock_mutex(&counter_mutex)
 106
 107static pthread_mutex_t work_mutex;
 108#define work_lock()             lock_mutex(&work_mutex)
 109#define work_unlock()           unlock_mutex(&work_mutex)
 110
 111static pthread_key_t key;
 112
 113static inline void lock_mutex(pthread_mutex_t *mutex)
 114{
 115        if (threads_active)
 116                pthread_mutex_lock(mutex);
 117}
 118
 119static inline void unlock_mutex(pthread_mutex_t *mutex)
 120{
 121        if (threads_active)
 122                pthread_mutex_unlock(mutex);
 123}
 124
 125/*
 126 * Mutex and conditional variable can't be statically-initialized on Windows.
 127 */
 128static void init_thread(void)
 129{
 130        init_recursive_mutex(&read_mutex);
 131        pthread_mutex_init(&counter_mutex, NULL);
 132        pthread_mutex_init(&work_mutex, NULL);
 133        pthread_key_create(&key, NULL);
 134        thread_data = xcalloc(nr_threads, sizeof(*thread_data));
 135        threads_active = 1;
 136}
 137
 138static void cleanup_thread(void)
 139{
 140        if (!threads_active)
 141                return;
 142        threads_active = 0;
 143        pthread_mutex_destroy(&read_mutex);
 144        pthread_mutex_destroy(&counter_mutex);
 145        pthread_mutex_destroy(&work_mutex);
 146        pthread_key_delete(key);
 147        free(thread_data);
 148}
 149
 150#else
 151
 152#define read_lock()
 153#define read_unlock()
 154
 155#define counter_lock()
 156#define counter_unlock()
 157
 158#define work_lock()
 159#define work_unlock()
 160
 161#endif
 162
 163
 164static int mark_link(struct object *obj, int type, void *data)
 165{
 166        if (!obj)
 167                return -1;
 168
 169        if (type != OBJ_ANY && obj->type != type)
 170                die(_("object type mismatch at %s"), sha1_to_hex(obj->sha1));
 171
 172        obj->flags |= FLAG_LINK;
 173        return 0;
 174}
 175
 176/* The content of each linked object must have been checked
 177   or it must be already present in the object database */
 178static void check_object(struct object *obj)
 179{
 180        if (!obj)
 181                return;
 182
 183        if (!(obj->flags & FLAG_LINK))
 184                return;
 185
 186        if (!(obj->flags & FLAG_CHECKED)) {
 187                unsigned long size;
 188                int type = sha1_object_info(obj->sha1, &size);
 189                if (type != obj->type || type <= 0)
 190                        die(_("object of unexpected type"));
 191                obj->flags |= FLAG_CHECKED;
 192                return;
 193        }
 194}
 195
 196static void check_objects(void)
 197{
 198        unsigned i, max;
 199
 200        max = get_max_object_index();
 201        for (i = 0; i < max; i++)
 202                check_object(get_indexed_object(i));
 203}
 204
 205
 206/* Discard current buffer used content. */
 207static void flush(void)
 208{
 209        if (input_offset) {
 210                if (output_fd >= 0)
 211                        write_or_die(output_fd, input_buffer, input_offset);
 212                git_SHA1_Update(&input_ctx, input_buffer, input_offset);
 213                memmove(input_buffer, input_buffer + input_offset, input_len);
 214                input_offset = 0;
 215        }
 216}
 217
 218/*
 219 * Make sure at least "min" bytes are available in the buffer, and
 220 * return the pointer to the buffer.
 221 */
 222static void *fill(int min)
 223{
 224        if (min <= input_len)
 225                return input_buffer + input_offset;
 226        if (min > sizeof(input_buffer))
 227                die(Q_("cannot fill %d byte",
 228                       "cannot fill %d bytes",
 229                       min),
 230                    min);
 231        flush();
 232        do {
 233                ssize_t ret = xread(input_fd, input_buffer + input_len,
 234                                sizeof(input_buffer) - input_len);
 235                if (ret <= 0) {
 236                        if (!ret)
 237                                die(_("early EOF"));
 238                        die_errno(_("read error on input"));
 239                }
 240                input_len += ret;
 241                if (from_stdin)
 242                        display_throughput(progress, consumed_bytes + input_len);
 243        } while (input_len < min);
 244        return input_buffer;
 245}
 246
 247static void use(int bytes)
 248{
 249        if (bytes > input_len)
 250                die(_("used more bytes than were available"));
 251        input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
 252        input_len -= bytes;
 253        input_offset += bytes;
 254
 255        /* make sure off_t is sufficiently large not to wrap */
 256        if (signed_add_overflows(consumed_bytes, bytes))
 257                die(_("pack too large for current definition of off_t"));
 258        consumed_bytes += bytes;
 259}
 260
 261static const char *open_pack_file(const char *pack_name)
 262{
 263        if (from_stdin) {
 264                input_fd = 0;
 265                if (!pack_name) {
 266                        static char tmp_file[PATH_MAX];
 267                        output_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
 268                                                "pack/tmp_pack_XXXXXX");
 269                        pack_name = xstrdup(tmp_file);
 270                } else
 271                        output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
 272                if (output_fd < 0)
 273                        die_errno(_("unable to create '%s'"), pack_name);
 274                pack_fd = output_fd;
 275        } else {
 276                input_fd = open(pack_name, O_RDONLY);
 277                if (input_fd < 0)
 278                        die_errno(_("cannot open packfile '%s'"), pack_name);
 279                output_fd = -1;
 280                pack_fd = input_fd;
 281        }
 282        git_SHA1_Init(&input_ctx);
 283        return pack_name;
 284}
 285
 286static void parse_pack_header(void)
 287{
 288        struct pack_header *hdr = fill(sizeof(struct pack_header));
 289
 290        /* Header consistency check */
 291        if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
 292                die(_("pack signature mismatch"));
 293        if (!pack_version_ok(hdr->hdr_version))
 294                die(_("pack version %"PRIu32" unsupported"),
 295                        ntohl(hdr->hdr_version));
 296
 297        nr_objects = ntohl(hdr->hdr_entries);
 298        use(sizeof(struct pack_header));
 299}
 300
 301static NORETURN void bad_object(unsigned long offset, const char *format,
 302                       ...) __attribute__((format (printf, 2, 3)));
 303
 304static NORETURN void bad_object(unsigned long offset, const char *format, ...)
 305{
 306        va_list params;
 307        char buf[1024];
 308
 309        va_start(params, format);
 310        vsnprintf(buf, sizeof(buf), format, params);
 311        va_end(params);
 312        die(_("pack has bad object at offset %lu: %s"), offset, buf);
 313}
 314
 315static inline struct thread_local *get_thread_data(void)
 316{
 317#ifndef NO_PTHREADS
 318        if (threads_active)
 319                return pthread_getspecific(key);
 320        assert(!threads_active &&
 321               "This should only be reached when all threads are gone");
 322#endif
 323        return &nothread_data;
 324}
 325
 326#ifndef NO_PTHREADS
 327static void set_thread_data(struct thread_local *data)
 328{
 329        if (threads_active)
 330                pthread_setspecific(key, data);
 331}
 332#endif
 333
 334static struct base_data *alloc_base_data(void)
 335{
 336        struct base_data *base = xmalloc(sizeof(struct base_data));
 337        memset(base, 0, sizeof(*base));
 338        base->ref_last = -1;
 339        base->ofs_last = -1;
 340        return base;
 341}
 342
 343static void free_base_data(struct base_data *c)
 344{
 345        if (c->data) {
 346                free(c->data);
 347                c->data = NULL;
 348                get_thread_data()->base_cache_used -= c->size;
 349        }
 350}
 351
 352static void prune_base_data(struct base_data *retain)
 353{
 354        struct base_data *b;
 355        struct thread_local *data = get_thread_data();
 356        for (b = data->base_cache;
 357             data->base_cache_used > delta_base_cache_limit && b;
 358             b = b->child) {
 359                if (b->data && b != retain)
 360                        free_base_data(b);
 361        }
 362}
 363
 364static void link_base_data(struct base_data *base, struct base_data *c)
 365{
 366        if (base)
 367                base->child = c;
 368        else
 369                get_thread_data()->base_cache = c;
 370
 371        c->base = base;
 372        c->child = NULL;
 373        if (c->data)
 374                get_thread_data()->base_cache_used += c->size;
 375        prune_base_data(c);
 376}
 377
 378static void unlink_base_data(struct base_data *c)
 379{
 380        struct base_data *base = c->base;
 381        if (base)
 382                base->child = NULL;
 383        else
 384                get_thread_data()->base_cache = NULL;
 385        free_base_data(c);
 386}
 387
 388static int is_delta_type(enum object_type type)
 389{
 390        return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
 391}
 392
 393static void *unpack_entry_data(unsigned long offset, unsigned long size,
 394                               enum object_type type, unsigned char *sha1)
 395{
 396        static char fixed_buf[8192];
 397        int status;
 398        git_zstream stream;
 399        void *buf;
 400        git_SHA_CTX c;
 401        char hdr[32];
 402        int hdrlen;
 403
 404        if (!is_delta_type(type)) {
 405                hdrlen = sprintf(hdr, "%s %lu", typename(type), size) + 1;
 406                git_SHA1_Init(&c);
 407                git_SHA1_Update(&c, hdr, hdrlen);
 408        } else
 409                sha1 = NULL;
 410        if (type == OBJ_BLOB && size > big_file_threshold)
 411                buf = fixed_buf;
 412        else
 413                buf = xmalloc(size);
 414
 415        memset(&stream, 0, sizeof(stream));
 416        git_inflate_init(&stream);
 417        stream.next_out = buf;
 418        stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
 419
 420        do {
 421                unsigned char *last_out = stream.next_out;
 422                stream.next_in = fill(1);
 423                stream.avail_in = input_len;
 424                status = git_inflate(&stream, 0);
 425                use(input_len - stream.avail_in);
 426                if (sha1)
 427                        git_SHA1_Update(&c, last_out, stream.next_out - last_out);
 428                if (buf == fixed_buf) {
 429                        stream.next_out = buf;
 430                        stream.avail_out = sizeof(fixed_buf);
 431                }
 432        } while (status == Z_OK);
 433        if (stream.total_out != size || status != Z_STREAM_END)
 434                bad_object(offset, _("inflate returned %d"), status);
 435        git_inflate_end(&stream);
 436        if (sha1)
 437                git_SHA1_Final(sha1, &c);
 438        return buf == fixed_buf ? NULL : buf;
 439}
 440
 441static void *unpack_raw_entry(struct object_entry *obj,
 442                              union delta_base *delta_base,
 443                              unsigned char *sha1)
 444{
 445        unsigned char *p;
 446        unsigned long size, c;
 447        off_t base_offset;
 448        unsigned shift;
 449        void *data;
 450
 451        obj->idx.offset = consumed_bytes;
 452        input_crc32 = crc32(0, NULL, 0);
 453
 454        p = fill(1);
 455        c = *p;
 456        use(1);
 457        obj->type = (c >> 4) & 7;
 458        size = (c & 15);
 459        shift = 4;
 460        while (c & 0x80) {
 461                p = fill(1);
 462                c = *p;
 463                use(1);
 464                size += (c & 0x7f) << shift;
 465                shift += 7;
 466        }
 467        obj->size = size;
 468
 469        switch (obj->type) {
 470        case OBJ_REF_DELTA:
 471                hashcpy(delta_base->sha1, fill(20));
 472                use(20);
 473                break;
 474        case OBJ_OFS_DELTA:
 475                memset(delta_base, 0, sizeof(*delta_base));
 476                p = fill(1);
 477                c = *p;
 478                use(1);
 479                base_offset = c & 127;
 480                while (c & 128) {
 481                        base_offset += 1;
 482                        if (!base_offset || MSB(base_offset, 7))
 483                                bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
 484                        p = fill(1);
 485                        c = *p;
 486                        use(1);
 487                        base_offset = (base_offset << 7) + (c & 127);
 488                }
 489                delta_base->offset = obj->idx.offset - base_offset;
 490                if (delta_base->offset <= 0 || delta_base->offset >= obj->idx.offset)
 491                        bad_object(obj->idx.offset, _("delta base offset is out of bound"));
 492                break;
 493        case OBJ_COMMIT:
 494        case OBJ_TREE:
 495        case OBJ_BLOB:
 496        case OBJ_TAG:
 497                break;
 498        default:
 499                bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
 500        }
 501        obj->hdr_size = consumed_bytes - obj->idx.offset;
 502
 503        data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, sha1);
 504        obj->idx.crc32 = input_crc32;
 505        return data;
 506}
 507
 508static void *unpack_data(struct object_entry *obj,
 509                         int (*consume)(const unsigned char *, unsigned long, void *),
 510                         void *cb_data)
 511{
 512        off_t from = obj[0].idx.offset + obj[0].hdr_size;
 513        unsigned long len = obj[1].idx.offset - from;
 514        unsigned char *data, *inbuf;
 515        git_zstream stream;
 516        int status;
 517
 518        data = xmalloc(consume ? 64*1024 : obj->size);
 519        inbuf = xmalloc((len < 64*1024) ? len : 64*1024);
 520
 521        memset(&stream, 0, sizeof(stream));
 522        git_inflate_init(&stream);
 523        stream.next_out = data;
 524        stream.avail_out = consume ? 64*1024 : obj->size;
 525
 526        do {
 527                ssize_t n = (len < 64*1024) ? len : 64*1024;
 528                n = pread(pack_fd, inbuf, n, from);
 529                if (n < 0)
 530                        die_errno(_("cannot pread pack file"));
 531                if (!n)
 532                        die(Q_("premature end of pack file, %lu byte missing",
 533                               "premature end of pack file, %lu bytes missing",
 534                               len),
 535                            len);
 536                from += n;
 537                len -= n;
 538                stream.next_in = inbuf;
 539                stream.avail_in = n;
 540                if (!consume)
 541                        status = git_inflate(&stream, 0);
 542                else {
 543                        do {
 544                                status = git_inflate(&stream, 0);
 545                                if (consume(data, stream.next_out - data, cb_data)) {
 546                                        free(inbuf);
 547                                        free(data);
 548                                        return NULL;
 549                                }
 550                                stream.next_out = data;
 551                                stream.avail_out = 64*1024;
 552                        } while (status == Z_OK && stream.avail_in);
 553                }
 554        } while (len && status == Z_OK && !stream.avail_in);
 555
 556        /* This has been inflated OK when first encountered, so... */
 557        if (status != Z_STREAM_END || stream.total_out != obj->size)
 558                die(_("serious inflate inconsistency"));
 559
 560        git_inflate_end(&stream);
 561        free(inbuf);
 562        if (consume) {
 563                free(data);
 564                data = NULL;
 565        }
 566        return data;
 567}
 568
 569static void *get_data_from_pack(struct object_entry *obj)
 570{
 571        return unpack_data(obj, NULL, NULL);
 572}
 573
 574static int compare_delta_bases(const union delta_base *base1,
 575                               const union delta_base *base2,
 576                               enum object_type type1,
 577                               enum object_type type2)
 578{
 579        int cmp = type1 - type2;
 580        if (cmp)
 581                return cmp;
 582        return memcmp(base1, base2, UNION_BASE_SZ);
 583}
 584
 585static int find_delta(const union delta_base *base, enum object_type type)
 586{
 587        int first = 0, last = nr_deltas;
 588
 589        while (first < last) {
 590                int next = (first + last) / 2;
 591                struct delta_entry *delta = &deltas[next];
 592                int cmp;
 593
 594                cmp = compare_delta_bases(base, &delta->base,
 595                                          type, objects[delta->obj_no].type);
 596                if (!cmp)
 597                        return next;
 598                if (cmp < 0) {
 599                        last = next;
 600                        continue;
 601                }
 602                first = next+1;
 603        }
 604        return -first-1;
 605}
 606
 607static void find_delta_children(const union delta_base *base,
 608                                int *first_index, int *last_index,
 609                                enum object_type type)
 610{
 611        int first = find_delta(base, type);
 612        int last = first;
 613        int end = nr_deltas - 1;
 614
 615        if (first < 0) {
 616                *first_index = 0;
 617                *last_index = -1;
 618                return;
 619        }
 620        while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
 621                --first;
 622        while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
 623                ++last;
 624        *first_index = first;
 625        *last_index = last;
 626}
 627
 628struct compare_data {
 629        struct object_entry *entry;
 630        struct git_istream *st;
 631        unsigned char *buf;
 632        unsigned long buf_size;
 633};
 634
 635static int compare_objects(const unsigned char *buf, unsigned long size,
 636                           void *cb_data)
 637{
 638        struct compare_data *data = cb_data;
 639
 640        if (data->buf_size < size) {
 641                free(data->buf);
 642                data->buf = xmalloc(size);
 643                data->buf_size = size;
 644        }
 645
 646        while (size) {
 647                ssize_t len = read_istream(data->st, data->buf, size);
 648                if (len == 0)
 649                        die(_("SHA1 COLLISION FOUND WITH %s !"),
 650                            sha1_to_hex(data->entry->idx.sha1));
 651                if (len < 0)
 652                        die(_("unable to read %s"),
 653                            sha1_to_hex(data->entry->idx.sha1));
 654                if (memcmp(buf, data->buf, len))
 655                        die(_("SHA1 COLLISION FOUND WITH %s !"),
 656                            sha1_to_hex(data->entry->idx.sha1));
 657                size -= len;
 658                buf += len;
 659        }
 660        return 0;
 661}
 662
 663static int check_collison(struct object_entry *entry)
 664{
 665        struct compare_data data;
 666        enum object_type type;
 667        unsigned long size;
 668
 669        if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
 670                return -1;
 671
 672        memset(&data, 0, sizeof(data));
 673        data.entry = entry;
 674        data.st = open_istream(entry->idx.sha1, &type, &size, NULL);
 675        if (!data.st)
 676                return -1;
 677        if (size != entry->size || type != entry->type)
 678                die(_("SHA1 COLLISION FOUND WITH %s !"),
 679                    sha1_to_hex(entry->idx.sha1));
 680        unpack_data(entry, compare_objects, &data);
 681        close_istream(data.st);
 682        free(data.buf);
 683        return 0;
 684}
 685
 686static void sha1_object(const void *data, struct object_entry *obj_entry,
 687                        unsigned long size, enum object_type type,
 688                        const unsigned char *sha1)
 689{
 690        void *new_data = NULL;
 691        int collision_test_needed;
 692
 693        assert(data || obj_entry);
 694
 695        read_lock();
 696        collision_test_needed = has_sha1_file(sha1);
 697        read_unlock();
 698
 699        if (collision_test_needed && !data) {
 700                read_lock();
 701                if (!check_collison(obj_entry))
 702                        collision_test_needed = 0;
 703                read_unlock();
 704        }
 705        if (collision_test_needed) {
 706                void *has_data;
 707                enum object_type has_type;
 708                unsigned long has_size;
 709                read_lock();
 710                has_type = sha1_object_info(sha1, &has_size);
 711                if (has_type != type || has_size != size)
 712                        die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
 713                has_data = read_sha1_file(sha1, &has_type, &has_size);
 714                read_unlock();
 715                if (!data)
 716                        data = new_data = get_data_from_pack(obj_entry);
 717                if (!has_data)
 718                        die(_("cannot read existing object %s"), sha1_to_hex(sha1));
 719                if (size != has_size || type != has_type ||
 720                    memcmp(data, has_data, size) != 0)
 721                        die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
 722                free(has_data);
 723        }
 724
 725        if (strict) {
 726                read_lock();
 727                if (type == OBJ_BLOB) {
 728                        struct blob *blob = lookup_blob(sha1);
 729                        if (blob)
 730                                blob->object.flags |= FLAG_CHECKED;
 731                        else
 732                                die(_("invalid blob object %s"), sha1_to_hex(sha1));
 733                } else {
 734                        struct object *obj;
 735                        int eaten;
 736                        void *buf = (void *) data;
 737
 738                        if (!buf)
 739                                buf = new_data = get_data_from_pack(obj_entry);
 740
 741                        /*
 742                         * we do not need to free the memory here, as the
 743                         * buf is deleted by the caller.
 744                         */
 745                        obj = parse_object_buffer(sha1, type, size, buf, &eaten);
 746                        if (!obj)
 747                                die(_("invalid %s"), typename(type));
 748                        if (fsck_object(obj, 1, fsck_error_function))
 749                                die(_("Error in object"));
 750                        if (fsck_walk(obj, mark_link, NULL))
 751                                die(_("Not all child objects of %s are reachable"), sha1_to_hex(obj->sha1));
 752
 753                        if (obj->type == OBJ_TREE) {
 754                                struct tree *item = (struct tree *) obj;
 755                                item->buffer = NULL;
 756                        }
 757                        if (obj->type == OBJ_COMMIT) {
 758                                struct commit *commit = (struct commit *) obj;
 759                                commit->buffer = NULL;
 760                        }
 761                        obj->flags |= FLAG_CHECKED;
 762                }
 763                read_unlock();
 764        }
 765
 766        free(new_data);
 767}
 768
 769/*
 770 * This function is part of find_unresolved_deltas(). There are two
 771 * walkers going in the opposite ways.
 772 *
 773 * The first one in find_unresolved_deltas() traverses down from
 774 * parent node to children, deflating nodes along the way. However,
 775 * memory for deflated nodes is limited by delta_base_cache_limit, so
 776 * at some point parent node's deflated content may be freed.
 777 *
 778 * The second walker is this function, which goes from current node up
 779 * to top parent if necessary to deflate the node. In normal
 780 * situation, its parent node would be already deflated, so it just
 781 * needs to apply delta.
 782 *
 783 * In the worst case scenario, parent node is no longer deflated because
 784 * we're running out of delta_base_cache_limit; we need to re-deflate
 785 * parents, possibly up to the top base.
 786 *
 787 * All deflated objects here are subject to be freed if we exceed
 788 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
 789 * just need to make sure the last node is not freed.
 790 */
 791static void *get_base_data(struct base_data *c)
 792{
 793        if (!c->data) {
 794                struct object_entry *obj = c->obj;
 795                struct base_data **delta = NULL;
 796                int delta_nr = 0, delta_alloc = 0;
 797
 798                while (is_delta_type(c->obj->type) && !c->data) {
 799                        ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
 800                        delta[delta_nr++] = c;
 801                        c = c->base;
 802                }
 803                if (!delta_nr) {
 804                        c->data = get_data_from_pack(obj);
 805                        c->size = obj->size;
 806                        get_thread_data()->base_cache_used += c->size;
 807                        prune_base_data(c);
 808                }
 809                for (; delta_nr > 0; delta_nr--) {
 810                        void *base, *raw;
 811                        c = delta[delta_nr - 1];
 812                        obj = c->obj;
 813                        base = get_base_data(c->base);
 814                        raw = get_data_from_pack(obj);
 815                        c->data = patch_delta(
 816                                base, c->base->size,
 817                                raw, obj->size,
 818                                &c->size);
 819                        free(raw);
 820                        if (!c->data)
 821                                bad_object(obj->idx.offset, _("failed to apply delta"));
 822                        get_thread_data()->base_cache_used += c->size;
 823                        prune_base_data(c);
 824                }
 825                free(delta);
 826        }
 827        return c->data;
 828}
 829
 830static void resolve_delta(struct object_entry *delta_obj,
 831                          struct base_data *base, struct base_data *result)
 832{
 833        void *base_data, *delta_data;
 834
 835        delta_obj->real_type = base->obj->real_type;
 836        delta_obj->delta_depth = base->obj->delta_depth + 1;
 837        if (deepest_delta < delta_obj->delta_depth)
 838                deepest_delta = delta_obj->delta_depth;
 839        delta_obj->base_object_no = base->obj - objects;
 840        delta_data = get_data_from_pack(delta_obj);
 841        base_data = get_base_data(base);
 842        result->obj = delta_obj;
 843        result->data = patch_delta(base_data, base->size,
 844                                   delta_data, delta_obj->size, &result->size);
 845        free(delta_data);
 846        if (!result->data)
 847                bad_object(delta_obj->idx.offset, _("failed to apply delta"));
 848        hash_sha1_file(result->data, result->size,
 849                       typename(delta_obj->real_type), delta_obj->idx.sha1);
 850        sha1_object(result->data, NULL, result->size, delta_obj->real_type,
 851                    delta_obj->idx.sha1);
 852        counter_lock();
 853        nr_resolved_deltas++;
 854        counter_unlock();
 855}
 856
 857static struct base_data *find_unresolved_deltas_1(struct base_data *base,
 858                                                  struct base_data *prev_base)
 859{
 860        if (base->ref_last == -1 && base->ofs_last == -1) {
 861                union delta_base base_spec;
 862
 863                hashcpy(base_spec.sha1, base->obj->idx.sha1);
 864                find_delta_children(&base_spec,
 865                                    &base->ref_first, &base->ref_last, OBJ_REF_DELTA);
 866
 867                memset(&base_spec, 0, sizeof(base_spec));
 868                base_spec.offset = base->obj->idx.offset;
 869                find_delta_children(&base_spec,
 870                                    &base->ofs_first, &base->ofs_last, OBJ_OFS_DELTA);
 871
 872                if (base->ref_last == -1 && base->ofs_last == -1) {
 873                        free(base->data);
 874                        return NULL;
 875                }
 876
 877                link_base_data(prev_base, base);
 878        }
 879
 880        if (base->ref_first <= base->ref_last) {
 881                struct object_entry *child = objects + deltas[base->ref_first].obj_no;
 882                struct base_data *result = alloc_base_data();
 883
 884                assert(child->real_type == OBJ_REF_DELTA);
 885                resolve_delta(child, base, result);
 886                if (base->ref_first == base->ref_last && base->ofs_last == -1)
 887                        free_base_data(base);
 888
 889                base->ref_first++;
 890                return result;
 891        }
 892
 893        if (base->ofs_first <= base->ofs_last) {
 894                struct object_entry *child = objects + deltas[base->ofs_first].obj_no;
 895                struct base_data *result = alloc_base_data();
 896
 897                assert(child->real_type == OBJ_OFS_DELTA);
 898                resolve_delta(child, base, result);
 899                if (base->ofs_first == base->ofs_last)
 900                        free_base_data(base);
 901
 902                base->ofs_first++;
 903                return result;
 904        }
 905
 906        unlink_base_data(base);
 907        return NULL;
 908}
 909
 910static void find_unresolved_deltas(struct base_data *base)
 911{
 912        struct base_data *new_base, *prev_base = NULL;
 913        for (;;) {
 914                new_base = find_unresolved_deltas_1(base, prev_base);
 915
 916                if (new_base) {
 917                        prev_base = base;
 918                        base = new_base;
 919                } else {
 920                        free(base);
 921                        base = prev_base;
 922                        if (!base)
 923                                return;
 924                        prev_base = base->base;
 925                }
 926        }
 927}
 928
 929static int compare_delta_entry(const void *a, const void *b)
 930{
 931        const struct delta_entry *delta_a = a;
 932        const struct delta_entry *delta_b = b;
 933
 934        /* group by type (ref vs ofs) and then by value (sha-1 or offset) */
 935        return compare_delta_bases(&delta_a->base, &delta_b->base,
 936                                   objects[delta_a->obj_no].type,
 937                                   objects[delta_b->obj_no].type);
 938}
 939
 940static void resolve_base(struct object_entry *obj)
 941{
 942        struct base_data *base_obj = alloc_base_data();
 943        base_obj->obj = obj;
 944        base_obj->data = NULL;
 945        find_unresolved_deltas(base_obj);
 946}
 947
 948#ifndef NO_PTHREADS
 949static void *threaded_second_pass(void *data)
 950{
 951        set_thread_data(data);
 952        for (;;) {
 953                int i;
 954                work_lock();
 955                display_progress(progress, nr_resolved_deltas);
 956                while (nr_dispatched < nr_objects &&
 957                       is_delta_type(objects[nr_dispatched].type))
 958                        nr_dispatched++;
 959                if (nr_dispatched >= nr_objects) {
 960                        work_unlock();
 961                        break;
 962                }
 963                i = nr_dispatched++;
 964                work_unlock();
 965
 966                resolve_base(&objects[i]);
 967        }
 968        return NULL;
 969}
 970#endif
 971
 972/*
 973 * First pass:
 974 * - find locations of all objects;
 975 * - calculate SHA1 of all non-delta objects;
 976 * - remember base (SHA1 or offset) for all deltas.
 977 */
 978static void parse_pack_objects(unsigned char *sha1)
 979{
 980        int i, nr_delays = 0;
 981        struct delta_entry *delta = deltas;
 982        struct stat st;
 983
 984        if (verbose)
 985                progress = start_progress(
 986                                from_stdin ? _("Receiving objects") : _("Indexing objects"),
 987                                nr_objects);
 988        for (i = 0; i < nr_objects; i++) {
 989                struct object_entry *obj = &objects[i];
 990                void *data = unpack_raw_entry(obj, &delta->base, obj->idx.sha1);
 991                obj->real_type = obj->type;
 992                if (is_delta_type(obj->type)) {
 993                        nr_deltas++;
 994                        delta->obj_no = i;
 995                        delta++;
 996                } else if (!data) {
 997                        /* large blobs, check later */
 998                        obj->real_type = OBJ_BAD;
 999                        nr_delays++;
1000                } else
1001                        sha1_object(data, NULL, obj->size, obj->type, obj->idx.sha1);
1002                free(data);
1003                display_progress(progress, i+1);
1004        }
1005        objects[i].idx.offset = consumed_bytes;
1006        stop_progress(&progress);
1007
1008        /* Check pack integrity */
1009        flush();
1010        git_SHA1_Final(sha1, &input_ctx);
1011        if (hashcmp(fill(20), sha1))
1012                die(_("pack is corrupted (SHA1 mismatch)"));
1013        use(20);
1014
1015        /* If input_fd is a file, we should have reached its end now. */
1016        if (fstat(input_fd, &st))
1017                die_errno(_("cannot fstat packfile"));
1018        if (S_ISREG(st.st_mode) &&
1019                        lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
1020                die(_("pack has junk at the end"));
1021
1022        for (i = 0; i < nr_objects; i++) {
1023                struct object_entry *obj = &objects[i];
1024                if (obj->real_type != OBJ_BAD)
1025                        continue;
1026                obj->real_type = obj->type;
1027                sha1_object(NULL, obj, obj->size, obj->type, obj->idx.sha1);
1028                nr_delays--;
1029        }
1030        if (nr_delays)
1031                die(_("confusion beyond insanity in parse_pack_objects()"));
1032}
1033
1034/*
1035 * Second pass:
1036 * - for all non-delta objects, look if it is used as a base for
1037 *   deltas;
1038 * - if used as a base, uncompress the object and apply all deltas,
1039 *   recursively checking if the resulting object is used as a base
1040 *   for some more deltas.
1041 */
1042static void resolve_deltas(void)
1043{
1044        int i;
1045
1046        if (!nr_deltas)
1047                return;
1048
1049        /* Sort deltas by base SHA1/offset for fast searching */
1050        qsort(deltas, nr_deltas, sizeof(struct delta_entry),
1051              compare_delta_entry);
1052
1053        if (verbose)
1054                progress = start_progress(_("Resolving deltas"), nr_deltas);
1055
1056#ifndef NO_PTHREADS
1057        nr_dispatched = 0;
1058        if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1059                init_thread();
1060                for (i = 0; i < nr_threads; i++) {
1061                        int ret = pthread_create(&thread_data[i].thread, NULL,
1062                                                 threaded_second_pass, thread_data + i);
1063                        if (ret)
1064                                die(_("unable to create thread: %s"),
1065                                    strerror(ret));
1066                }
1067                for (i = 0; i < nr_threads; i++)
1068                        pthread_join(thread_data[i].thread, NULL);
1069                cleanup_thread();
1070                return;
1071        }
1072#endif
1073
1074        for (i = 0; i < nr_objects; i++) {
1075                struct object_entry *obj = &objects[i];
1076
1077                if (is_delta_type(obj->type))
1078                        continue;
1079                resolve_base(obj);
1080                display_progress(progress, nr_resolved_deltas);
1081        }
1082}
1083
1084/*
1085 * Third pass:
1086 * - append objects to convert thin pack to full pack if required
1087 * - write the final 20-byte SHA-1
1088 */
1089static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved);
1090static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_sha1)
1091{
1092        if (nr_deltas == nr_resolved_deltas) {
1093                stop_progress(&progress);
1094                /* Flush remaining pack final 20-byte SHA1. */
1095                flush();
1096                return;
1097        }
1098
1099        if (fix_thin_pack) {
1100                struct sha1file *f;
1101                unsigned char read_sha1[20], tail_sha1[20];
1102                char msg[48];
1103                int nr_unresolved = nr_deltas - nr_resolved_deltas;
1104                int nr_objects_initial = nr_objects;
1105                if (nr_unresolved <= 0)
1106                        die(_("confusion beyond insanity"));
1107                objects = xrealloc(objects,
1108                                   (nr_objects + nr_unresolved + 1)
1109                                   * sizeof(*objects));
1110                f = sha1fd(output_fd, curr_pack);
1111                fix_unresolved_deltas(f, nr_unresolved);
1112                sprintf(msg, _("completed with %d local objects"),
1113                        nr_objects - nr_objects_initial);
1114                stop_progress_msg(&progress, msg);
1115                sha1close(f, tail_sha1, 0);
1116                hashcpy(read_sha1, pack_sha1);
1117                fixup_pack_header_footer(output_fd, pack_sha1,
1118                                         curr_pack, nr_objects,
1119                                         read_sha1, consumed_bytes-20);
1120                if (hashcmp(read_sha1, tail_sha1) != 0)
1121                        die(_("Unexpected tail checksum for %s "
1122                              "(disk corruption?)"), curr_pack);
1123        }
1124        if (nr_deltas != nr_resolved_deltas)
1125                die(Q_("pack has %d unresolved delta",
1126                       "pack has %d unresolved deltas",
1127                       nr_deltas - nr_resolved_deltas),
1128                    nr_deltas - nr_resolved_deltas);
1129}
1130
1131static int write_compressed(struct sha1file *f, void *in, unsigned int size)
1132{
1133        git_zstream stream;
1134        int status;
1135        unsigned char outbuf[4096];
1136
1137        memset(&stream, 0, sizeof(stream));
1138        git_deflate_init(&stream, zlib_compression_level);
1139        stream.next_in = in;
1140        stream.avail_in = size;
1141
1142        do {
1143                stream.next_out = outbuf;
1144                stream.avail_out = sizeof(outbuf);
1145                status = git_deflate(&stream, Z_FINISH);
1146                sha1write(f, outbuf, sizeof(outbuf) - stream.avail_out);
1147        } while (status == Z_OK);
1148
1149        if (status != Z_STREAM_END)
1150                die(_("unable to deflate appended object (%d)"), status);
1151        size = stream.total_out;
1152        git_deflate_end(&stream);
1153        return size;
1154}
1155
1156static struct object_entry *append_obj_to_pack(struct sha1file *f,
1157                               const unsigned char *sha1, void *buf,
1158                               unsigned long size, enum object_type type)
1159{
1160        struct object_entry *obj = &objects[nr_objects++];
1161        unsigned char header[10];
1162        unsigned long s = size;
1163        int n = 0;
1164        unsigned char c = (type << 4) | (s & 15);
1165        s >>= 4;
1166        while (s) {
1167                header[n++] = c | 0x80;
1168                c = s & 0x7f;
1169                s >>= 7;
1170        }
1171        header[n++] = c;
1172        crc32_begin(f);
1173        sha1write(f, header, n);
1174        obj[0].size = size;
1175        obj[0].hdr_size = n;
1176        obj[0].type = type;
1177        obj[0].real_type = type;
1178        obj[1].idx.offset = obj[0].idx.offset + n;
1179        obj[1].idx.offset += write_compressed(f, buf, size);
1180        obj[0].idx.crc32 = crc32_end(f);
1181        sha1flush(f);
1182        hashcpy(obj->idx.sha1, sha1);
1183        return obj;
1184}
1185
1186static int delta_pos_compare(const void *_a, const void *_b)
1187{
1188        struct delta_entry *a = *(struct delta_entry **)_a;
1189        struct delta_entry *b = *(struct delta_entry **)_b;
1190        return a->obj_no - b->obj_no;
1191}
1192
1193static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved)
1194{
1195        struct delta_entry **sorted_by_pos;
1196        int i, n = 0;
1197
1198        /*
1199         * Since many unresolved deltas may well be themselves base objects
1200         * for more unresolved deltas, we really want to include the
1201         * smallest number of base objects that would cover as much delta
1202         * as possible by picking the
1203         * trunc deltas first, allowing for other deltas to resolve without
1204         * additional base objects.  Since most base objects are to be found
1205         * before deltas depending on them, a good heuristic is to start
1206         * resolving deltas in the same order as their position in the pack.
1207         */
1208        sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
1209        for (i = 0; i < nr_deltas; i++) {
1210                if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
1211                        continue;
1212                sorted_by_pos[n++] = &deltas[i];
1213        }
1214        qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
1215
1216        for (i = 0; i < n; i++) {
1217                struct delta_entry *d = sorted_by_pos[i];
1218                enum object_type type;
1219                struct base_data *base_obj = alloc_base_data();
1220
1221                if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1222                        continue;
1223                base_obj->data = read_sha1_file(d->base.sha1, &type, &base_obj->size);
1224                if (!base_obj->data)
1225                        continue;
1226
1227                if (check_sha1_signature(d->base.sha1, base_obj->data,
1228                                base_obj->size, typename(type)))
1229                        die(_("local object %s is corrupt"), sha1_to_hex(d->base.sha1));
1230                base_obj->obj = append_obj_to_pack(f, d->base.sha1,
1231                                        base_obj->data, base_obj->size, type);
1232                find_unresolved_deltas(base_obj);
1233                display_progress(progress, nr_resolved_deltas);
1234        }
1235        free(sorted_by_pos);
1236}
1237
1238static void final(const char *final_pack_name, const char *curr_pack_name,
1239                  const char *final_index_name, const char *curr_index_name,
1240                  const char *keep_name, const char *keep_msg,
1241                  unsigned char *sha1)
1242{
1243        const char *report = "pack";
1244        char name[PATH_MAX];
1245        int err;
1246
1247        if (!from_stdin) {
1248                close(input_fd);
1249        } else {
1250                fsync_or_die(output_fd, curr_pack_name);
1251                err = close(output_fd);
1252                if (err)
1253                        die_errno(_("error while closing pack file"));
1254        }
1255
1256        if (keep_msg) {
1257                int keep_fd, keep_msg_len = strlen(keep_msg);
1258
1259                if (!keep_name)
1260                        keep_fd = odb_pack_keep(name, sizeof(name), sha1);
1261                else
1262                        keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
1263
1264                if (keep_fd < 0) {
1265                        if (errno != EEXIST)
1266                                die_errno(_("cannot write keep file '%s'"),
1267                                          keep_name);
1268                } else {
1269                        if (keep_msg_len > 0) {
1270                                write_or_die(keep_fd, keep_msg, keep_msg_len);
1271                                write_or_die(keep_fd, "\n", 1);
1272                        }
1273                        if (close(keep_fd) != 0)
1274                                die_errno(_("cannot close written keep file '%s'"),
1275                                    keep_name);
1276                        report = "keep";
1277                }
1278        }
1279
1280        if (final_pack_name != curr_pack_name) {
1281                if (!final_pack_name) {
1282                        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
1283                                 get_object_directory(), sha1_to_hex(sha1));
1284                        final_pack_name = name;
1285                }
1286                if (move_temp_to_file(curr_pack_name, final_pack_name))
1287                        die(_("cannot store pack file"));
1288        } else if (from_stdin)
1289                chmod(final_pack_name, 0444);
1290
1291        if (final_index_name != curr_index_name) {
1292                if (!final_index_name) {
1293                        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
1294                                 get_object_directory(), sha1_to_hex(sha1));
1295                        final_index_name = name;
1296                }
1297                if (move_temp_to_file(curr_index_name, final_index_name))
1298                        die(_("cannot store index file"));
1299        } else
1300                chmod(final_index_name, 0444);
1301
1302        if (!from_stdin) {
1303                printf("%s\n", sha1_to_hex(sha1));
1304        } else {
1305                char buf[48];
1306                int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
1307                                   report, sha1_to_hex(sha1));
1308                write_or_die(1, buf, len);
1309
1310                /*
1311                 * Let's just mimic git-unpack-objects here and write
1312                 * the last part of the input buffer to stdout.
1313                 */
1314                while (input_len) {
1315                        err = xwrite(1, input_buffer + input_offset, input_len);
1316                        if (err <= 0)
1317                                break;
1318                        input_len -= err;
1319                        input_offset += err;
1320                }
1321        }
1322}
1323
1324static int git_index_pack_config(const char *k, const char *v, void *cb)
1325{
1326        struct pack_idx_option *opts = cb;
1327
1328        if (!strcmp(k, "pack.indexversion")) {
1329                opts->version = git_config_int(k, v);
1330                if (opts->version > 2)
1331                        die(_("bad pack.indexversion=%"PRIu32), opts->version);
1332                return 0;
1333        }
1334        if (!strcmp(k, "pack.threads")) {
1335                nr_threads = git_config_int(k, v);
1336                if (nr_threads < 0)
1337                        die(_("invalid number of threads specified (%d)"),
1338                            nr_threads);
1339#ifdef NO_PTHREADS
1340                if (nr_threads != 1)
1341                        warning(_("no threads support, ignoring %s"), k);
1342                nr_threads = 1;
1343#endif
1344                return 0;
1345        }
1346        return git_default_config(k, v, cb);
1347}
1348
1349static int cmp_uint32(const void *a_, const void *b_)
1350{
1351        uint32_t a = *((uint32_t *)a_);
1352        uint32_t b = *((uint32_t *)b_);
1353
1354        return (a < b) ? -1 : (a != b);
1355}
1356
1357static void read_v2_anomalous_offsets(struct packed_git *p,
1358                                      struct pack_idx_option *opts)
1359{
1360        const uint32_t *idx1, *idx2;
1361        uint32_t i;
1362
1363        /* The address of the 4-byte offset table */
1364        idx1 = (((const uint32_t *)p->index_data)
1365                + 2 /* 8-byte header */
1366                + 256 /* fan out */
1367                + 5 * p->num_objects /* 20-byte SHA-1 table */
1368                + p->num_objects /* CRC32 table */
1369                );
1370
1371        /* The address of the 8-byte offset table */
1372        idx2 = idx1 + p->num_objects;
1373
1374        for (i = 0; i < p->num_objects; i++) {
1375                uint32_t off = ntohl(idx1[i]);
1376                if (!(off & 0x80000000))
1377                        continue;
1378                off = off & 0x7fffffff;
1379                if (idx2[off * 2])
1380                        continue;
1381                /*
1382                 * The real offset is ntohl(idx2[off * 2]) in high 4
1383                 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1384                 * octets.  But idx2[off * 2] is Zero!!!
1385                 */
1386                ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1387                opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1388        }
1389
1390        if (1 < opts->anomaly_nr)
1391                qsort(opts->anomaly, opts->anomaly_nr, sizeof(uint32_t), cmp_uint32);
1392}
1393
1394static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1395{
1396        struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1397
1398        if (!p)
1399                die(_("Cannot open existing pack file '%s'"), pack_name);
1400        if (open_pack_index(p))
1401                die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1402
1403        /* Read the attributes from the existing idx file */
1404        opts->version = p->index_version;
1405
1406        if (opts->version == 2)
1407                read_v2_anomalous_offsets(p, opts);
1408
1409        /*
1410         * Get rid of the idx file as we do not need it anymore.
1411         * NEEDSWORK: extract this bit from free_pack_by_name() in
1412         * sha1_file.c, perhaps?  It shouldn't matter very much as we
1413         * know we haven't installed this pack (hence we never have
1414         * read anything from it).
1415         */
1416        close_pack_index(p);
1417        free(p);
1418}
1419
1420static void show_pack_info(int stat_only)
1421{
1422        int i, baseobjects = nr_objects - nr_deltas;
1423        unsigned long *chain_histogram = NULL;
1424
1425        if (deepest_delta)
1426                chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1427
1428        for (i = 0; i < nr_objects; i++) {
1429                struct object_entry *obj = &objects[i];
1430
1431                if (is_delta_type(obj->type))
1432                        chain_histogram[obj->delta_depth - 1]++;
1433                if (stat_only)
1434                        continue;
1435                printf("%s %-6s %lu %lu %"PRIuMAX,
1436                       sha1_to_hex(obj->idx.sha1),
1437                       typename(obj->real_type), obj->size,
1438                       (unsigned long)(obj[1].idx.offset - obj->idx.offset),
1439                       (uintmax_t)obj->idx.offset);
1440                if (is_delta_type(obj->type)) {
1441                        struct object_entry *bobj = &objects[obj->base_object_no];
1442                        printf(" %u %s", obj->delta_depth, sha1_to_hex(bobj->idx.sha1));
1443                }
1444                putchar('\n');
1445        }
1446
1447        if (baseobjects)
1448                printf_ln(Q_("non delta: %d object",
1449                             "non delta: %d objects",
1450                             baseobjects),
1451                          baseobjects);
1452        for (i = 0; i < deepest_delta; i++) {
1453                if (!chain_histogram[i])
1454                        continue;
1455                printf_ln(Q_("chain length = %d: %lu object",
1456                             "chain length = %d: %lu objects",
1457                             chain_histogram[i]),
1458                          i + 1,
1459                          chain_histogram[i]);
1460        }
1461}
1462
1463int cmd_index_pack(int argc, const char **argv, const char *prefix)
1464{
1465        int i, fix_thin_pack = 0, verify = 0, stat_only = 0, stat = 0;
1466        const char *curr_pack, *curr_index;
1467        const char *index_name = NULL, *pack_name = NULL;
1468        const char *keep_name = NULL, *keep_msg = NULL;
1469        char *index_name_buf = NULL, *keep_name_buf = NULL;
1470        struct pack_idx_entry **idx_objects;
1471        struct pack_idx_option opts;
1472        unsigned char pack_sha1[20];
1473
1474        if (argc == 2 && !strcmp(argv[1], "-h"))
1475                usage(index_pack_usage);
1476
1477        read_replace_refs = 0;
1478
1479        reset_pack_idx_option(&opts);
1480        git_config(git_index_pack_config, &opts);
1481        if (prefix && chdir(prefix))
1482                die(_("Cannot come back to cwd"));
1483
1484        for (i = 1; i < argc; i++) {
1485                const char *arg = argv[i];
1486
1487                if (*arg == '-') {
1488                        if (!strcmp(arg, "--stdin")) {
1489                                from_stdin = 1;
1490                        } else if (!strcmp(arg, "--fix-thin")) {
1491                                fix_thin_pack = 1;
1492                        } else if (!strcmp(arg, "--strict")) {
1493                                strict = 1;
1494                        } else if (!strcmp(arg, "--verify")) {
1495                                verify = 1;
1496                        } else if (!strcmp(arg, "--verify-stat")) {
1497                                verify = 1;
1498                                stat = 1;
1499                        } else if (!strcmp(arg, "--verify-stat-only")) {
1500                                verify = 1;
1501                                stat = 1;
1502                                stat_only = 1;
1503                        } else if (!strcmp(arg, "--keep")) {
1504                                keep_msg = "";
1505                        } else if (!prefixcmp(arg, "--keep=")) {
1506                                keep_msg = arg + 7;
1507                        } else if (!prefixcmp(arg, "--threads=")) {
1508                                char *end;
1509                                nr_threads = strtoul(arg+10, &end, 0);
1510                                if (!arg[10] || *end || nr_threads < 0)
1511                                        usage(index_pack_usage);
1512#ifdef NO_PTHREADS
1513                                if (nr_threads != 1)
1514                                        warning(_("no threads support, "
1515                                                  "ignoring %s"), arg);
1516                                nr_threads = 1;
1517#endif
1518                        } else if (!prefixcmp(arg, "--pack_header=")) {
1519                                struct pack_header *hdr;
1520                                char *c;
1521
1522                                hdr = (struct pack_header *)input_buffer;
1523                                hdr->hdr_signature = htonl(PACK_SIGNATURE);
1524                                hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1525                                if (*c != ',')
1526                                        die(_("bad %s"), arg);
1527                                hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1528                                if (*c)
1529                                        die(_("bad %s"), arg);
1530                                input_len = sizeof(*hdr);
1531                        } else if (!strcmp(arg, "-v")) {
1532                                verbose = 1;
1533                        } else if (!strcmp(arg, "-o")) {
1534                                if (index_name || (i+1) >= argc)
1535                                        usage(index_pack_usage);
1536                                index_name = argv[++i];
1537                        } else if (!prefixcmp(arg, "--index-version=")) {
1538                                char *c;
1539                                opts.version = strtoul(arg + 16, &c, 10);
1540                                if (opts.version > 2)
1541                                        die(_("bad %s"), arg);
1542                                if (*c == ',')
1543                                        opts.off32_limit = strtoul(c+1, &c, 0);
1544                                if (*c || opts.off32_limit & 0x80000000)
1545                                        die(_("bad %s"), arg);
1546                        } else
1547                                usage(index_pack_usage);
1548                        continue;
1549                }
1550
1551                if (pack_name)
1552                        usage(index_pack_usage);
1553                pack_name = arg;
1554        }
1555
1556        if (!pack_name && !from_stdin)
1557                usage(index_pack_usage);
1558        if (fix_thin_pack && !from_stdin)
1559                die(_("--fix-thin cannot be used without --stdin"));
1560        if (!index_name && pack_name) {
1561                int len = strlen(pack_name);
1562                if (!has_extension(pack_name, ".pack"))
1563                        die(_("packfile name '%s' does not end with '.pack'"),
1564                            pack_name);
1565                index_name_buf = xmalloc(len);
1566                memcpy(index_name_buf, pack_name, len - 5);
1567                strcpy(index_name_buf + len - 5, ".idx");
1568                index_name = index_name_buf;
1569        }
1570        if (keep_msg && !keep_name && pack_name) {
1571                int len = strlen(pack_name);
1572                if (!has_extension(pack_name, ".pack"))
1573                        die(_("packfile name '%s' does not end with '.pack'"),
1574                            pack_name);
1575                keep_name_buf = xmalloc(len);
1576                memcpy(keep_name_buf, pack_name, len - 5);
1577                strcpy(keep_name_buf + len - 5, ".keep");
1578                keep_name = keep_name_buf;
1579        }
1580        if (verify) {
1581                if (!index_name)
1582                        die(_("--verify with no packfile name given"));
1583                read_idx_option(&opts, index_name);
1584                opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1585        }
1586        if (strict)
1587                opts.flags |= WRITE_IDX_STRICT;
1588
1589#ifndef NO_PTHREADS
1590        if (!nr_threads) {
1591                nr_threads = online_cpus();
1592                /* An experiment showed that more threads does not mean faster */
1593                if (nr_threads > 3)
1594                        nr_threads = 3;
1595        }
1596#endif
1597
1598        curr_pack = open_pack_file(pack_name);
1599        parse_pack_header();
1600        objects = xcalloc(nr_objects + 1, sizeof(struct object_entry));
1601        deltas = xcalloc(nr_objects, sizeof(struct delta_entry));
1602        parse_pack_objects(pack_sha1);
1603        resolve_deltas();
1604        conclude_pack(fix_thin_pack, curr_pack, pack_sha1);
1605        free(deltas);
1606        if (strict)
1607                check_objects();
1608
1609        if (stat)
1610                show_pack_info(stat_only);
1611
1612        idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *));
1613        for (i = 0; i < nr_objects; i++)
1614                idx_objects[i] = &objects[i].idx;
1615        curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_sha1);
1616        free(idx_objects);
1617
1618        if (!verify)
1619                final(pack_name, curr_pack,
1620                      index_name, curr_index,
1621                      keep_name, keep_msg,
1622                      pack_sha1);
1623        else
1624                close(input_fd);
1625        free(objects);
1626        free(index_name_buf);
1627        free(keep_name_buf);
1628        if (pack_name == NULL)
1629                free((void *) curr_pack);
1630        if (index_name == NULL)
1631                free((void *) curr_index);
1632
1633        return 0;
1634}