midx.con commit midx: read objects from multi-pack-index (3715a63)
   1#include "cache.h"
   2#include "config.h"
   3#include "csum-file.h"
   4#include "dir.h"
   5#include "lockfile.h"
   6#include "packfile.h"
   7#include "object-store.h"
   8#include "sha1-lookup.h"
   9#include "midx.h"
  10
  11#define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
  12#define MIDX_VERSION 1
  13#define MIDX_BYTE_FILE_VERSION 4
  14#define MIDX_BYTE_HASH_VERSION 5
  15#define MIDX_BYTE_NUM_CHUNKS 6
  16#define MIDX_BYTE_NUM_PACKS 8
  17#define MIDX_HASH_VERSION 1
  18#define MIDX_HEADER_SIZE 12
  19#define MIDX_HASH_LEN 20
  20#define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + MIDX_HASH_LEN)
  21
  22#define MIDX_MAX_CHUNKS 5
  23#define MIDX_CHUNK_ALIGNMENT 4
  24#define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
  25#define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
  26#define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
  27#define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
  28#define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
  29#define MIDX_CHUNKLOOKUP_WIDTH (sizeof(uint32_t) + sizeof(uint64_t))
  30#define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
  31#define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
  32#define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
  33#define MIDX_LARGE_OFFSET_NEEDED 0x80000000
  34
  35static char *get_midx_filename(const char *object_dir)
  36{
  37        return xstrfmt("%s/pack/multi-pack-index", object_dir);
  38}
  39
  40struct multi_pack_index *load_multi_pack_index(const char *object_dir)
  41{
  42        struct multi_pack_index *m = NULL;
  43        int fd;
  44        struct stat st;
  45        size_t midx_size;
  46        void *midx_map = NULL;
  47        uint32_t hash_version;
  48        char *midx_name = get_midx_filename(object_dir);
  49        uint32_t i;
  50        const char *cur_pack_name;
  51
  52        fd = git_open(midx_name);
  53
  54        if (fd < 0)
  55                goto cleanup_fail;
  56        if (fstat(fd, &st)) {
  57                error_errno(_("failed to read %s"), midx_name);
  58                goto cleanup_fail;
  59        }
  60
  61        midx_size = xsize_t(st.st_size);
  62
  63        if (midx_size < MIDX_MIN_SIZE) {
  64                error(_("multi-pack-index file %s is too small"), midx_name);
  65                goto cleanup_fail;
  66        }
  67
  68        FREE_AND_NULL(midx_name);
  69
  70        midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
  71
  72        FLEX_ALLOC_MEM(m, object_dir, object_dir, strlen(object_dir));
  73        m->fd = fd;
  74        m->data = midx_map;
  75        m->data_len = midx_size;
  76
  77        m->signature = get_be32(m->data);
  78        if (m->signature != MIDX_SIGNATURE) {
  79                error(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
  80                      m->signature, MIDX_SIGNATURE);
  81                goto cleanup_fail;
  82        }
  83
  84        m->version = m->data[MIDX_BYTE_FILE_VERSION];
  85        if (m->version != MIDX_VERSION) {
  86                error(_("multi-pack-index version %d not recognized"),
  87                      m->version);
  88                goto cleanup_fail;
  89        }
  90
  91        hash_version = m->data[MIDX_BYTE_HASH_VERSION];
  92        if (hash_version != MIDX_HASH_VERSION) {
  93                error(_("hash version %u does not match"), hash_version);
  94                goto cleanup_fail;
  95        }
  96        m->hash_len = MIDX_HASH_LEN;
  97
  98        m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
  99
 100        m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
 101
 102        for (i = 0; i < m->num_chunks; i++) {
 103                uint32_t chunk_id = get_be32(m->data + MIDX_HEADER_SIZE +
 104                                             MIDX_CHUNKLOOKUP_WIDTH * i);
 105                uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 +
 106                                                 MIDX_CHUNKLOOKUP_WIDTH * i);
 107
 108                switch (chunk_id) {
 109                        case MIDX_CHUNKID_PACKNAMES:
 110                                m->chunk_pack_names = m->data + chunk_offset;
 111                                break;
 112
 113                        case MIDX_CHUNKID_OIDFANOUT:
 114                                m->chunk_oid_fanout = (uint32_t *)(m->data + chunk_offset);
 115                                break;
 116
 117                        case MIDX_CHUNKID_OIDLOOKUP:
 118                                m->chunk_oid_lookup = m->data + chunk_offset;
 119                                break;
 120
 121                        case MIDX_CHUNKID_OBJECTOFFSETS:
 122                                m->chunk_object_offsets = m->data + chunk_offset;
 123                                break;
 124
 125                        case MIDX_CHUNKID_LARGEOFFSETS:
 126                                m->chunk_large_offsets = m->data + chunk_offset;
 127                                break;
 128
 129                        case 0:
 130                                die(_("terminating multi-pack-index chunk id appears earlier than expected"));
 131                                break;
 132
 133                        default:
 134                                /*
 135                                 * Do nothing on unrecognized chunks, allowing future
 136                                 * extensions to add optional chunks.
 137                                 */
 138                                break;
 139                }
 140        }
 141
 142        if (!m->chunk_pack_names)
 143                die(_("multi-pack-index missing required pack-name chunk"));
 144        if (!m->chunk_oid_fanout)
 145                die(_("multi-pack-index missing required OID fanout chunk"));
 146        if (!m->chunk_oid_lookup)
 147                die(_("multi-pack-index missing required OID lookup chunk"));
 148        if (!m->chunk_object_offsets)
 149                die(_("multi-pack-index missing required object offsets chunk"));
 150
 151        m->num_objects = ntohl(m->chunk_oid_fanout[255]);
 152
 153        m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
 154        m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
 155
 156        cur_pack_name = (const char *)m->chunk_pack_names;
 157        for (i = 0; i < m->num_packs; i++) {
 158                m->pack_names[i] = cur_pack_name;
 159
 160                cur_pack_name += strlen(cur_pack_name) + 1;
 161
 162                if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0) {
 163                        error(_("multi-pack-index pack names out of order: '%s' before '%s'"),
 164                              m->pack_names[i - 1],
 165                              m->pack_names[i]);
 166                        goto cleanup_fail;
 167                }
 168        }
 169
 170        return m;
 171
 172cleanup_fail:
 173        free(m);
 174        free(midx_name);
 175        if (midx_map)
 176                munmap(midx_map, midx_size);
 177        if (0 <= fd)
 178                close(fd);
 179        return NULL;
 180}
 181
 182static int prepare_midx_pack(struct multi_pack_index *m, uint32_t pack_int_id)
 183{
 184        struct strbuf pack_name = STRBUF_INIT;
 185
 186        if (pack_int_id >= m->num_packs)
 187                BUG("bad pack-int-id");
 188
 189        if (m->packs[pack_int_id])
 190                return 0;
 191
 192        strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
 193                    m->pack_names[pack_int_id]);
 194
 195        m->packs[pack_int_id] = add_packed_git(pack_name.buf, pack_name.len, 1);
 196        strbuf_release(&pack_name);
 197        return !m->packs[pack_int_id];
 198}
 199
 200int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
 201{
 202        return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
 203                            MIDX_HASH_LEN, result);
 204}
 205
 206static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
 207{
 208        const unsigned char *offset_data;
 209        uint32_t offset32;
 210
 211        offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
 212        offset32 = get_be32(offset_data + sizeof(uint32_t));
 213
 214        if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
 215                if (sizeof(offset32) < sizeof(uint64_t))
 216                        die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
 217
 218                offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
 219                return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
 220        }
 221
 222        return offset32;
 223}
 224
 225static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
 226{
 227        return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
 228}
 229
 230static int nth_midxed_pack_entry(struct multi_pack_index *m, struct pack_entry *e, uint32_t pos)
 231{
 232        uint32_t pack_int_id;
 233        struct packed_git *p;
 234
 235        if (pos >= m->num_objects)
 236                return 0;
 237
 238        pack_int_id = nth_midxed_pack_int_id(m, pos);
 239
 240        if (prepare_midx_pack(m, pack_int_id))
 241                die(_("error preparing packfile from multi-pack-index"));
 242        p = m->packs[pack_int_id];
 243
 244        /*
 245        * We are about to tell the caller where they can locate the
 246        * requested object.  We better make sure the packfile is
 247        * still here and can be accessed before supplying that
 248        * answer, as it may have been deleted since the MIDX was
 249        * loaded!
 250        */
 251        if (!is_pack_valid(p))
 252                return 0;
 253
 254        e->offset = nth_midxed_offset(m, pos);
 255        e->p = p;
 256
 257        return 1;
 258}
 259
 260int fill_midx_entry(const struct object_id *oid, struct pack_entry *e, struct multi_pack_index *m)
 261{
 262        uint32_t pos;
 263
 264        if (!bsearch_midx(oid, m, &pos))
 265                return 0;
 266
 267        return nth_midxed_pack_entry(m, e, pos);
 268}
 269
 270int prepare_multi_pack_index_one(struct repository *r, const char *object_dir)
 271{
 272        struct multi_pack_index *m = r->objects->multi_pack_index;
 273        struct multi_pack_index *m_search;
 274        int config_value;
 275
 276        if (repo_config_get_bool(r, "core.multipackindex", &config_value) ||
 277            !config_value)
 278                return 0;
 279
 280        for (m_search = m; m_search; m_search = m_search->next)
 281                if (!strcmp(object_dir, m_search->object_dir))
 282                        return 1;
 283
 284        r->objects->multi_pack_index = load_multi_pack_index(object_dir);
 285
 286        if (r->objects->multi_pack_index) {
 287                r->objects->multi_pack_index->next = m;
 288                return 1;
 289        }
 290
 291        return 0;
 292}
 293
 294static size_t write_midx_header(struct hashfile *f,
 295                                unsigned char num_chunks,
 296                                uint32_t num_packs)
 297{
 298        unsigned char byte_values[4];
 299
 300        hashwrite_be32(f, MIDX_SIGNATURE);
 301        byte_values[0] = MIDX_VERSION;
 302        byte_values[1] = MIDX_HASH_VERSION;
 303        byte_values[2] = num_chunks;
 304        byte_values[3] = 0; /* unused */
 305        hashwrite(f, byte_values, sizeof(byte_values));
 306        hashwrite_be32(f, num_packs);
 307
 308        return MIDX_HEADER_SIZE;
 309}
 310
 311struct pack_list {
 312        struct packed_git **list;
 313        char **names;
 314        uint32_t nr;
 315        uint32_t alloc_list;
 316        uint32_t alloc_names;
 317        size_t pack_name_concat_len;
 318};
 319
 320static void add_pack_to_midx(const char *full_path, size_t full_path_len,
 321                             const char *file_name, void *data)
 322{
 323        struct pack_list *packs = (struct pack_list *)data;
 324
 325        if (ends_with(file_name, ".idx")) {
 326                ALLOC_GROW(packs->list, packs->nr + 1, packs->alloc_list);
 327                ALLOC_GROW(packs->names, packs->nr + 1, packs->alloc_names);
 328
 329                packs->list[packs->nr] = add_packed_git(full_path,
 330                                                        full_path_len,
 331                                                        0);
 332
 333                if (!packs->list[packs->nr]) {
 334                        warning(_("failed to add packfile '%s'"),
 335                                full_path);
 336                        return;
 337                }
 338
 339                if (open_pack_index(packs->list[packs->nr])) {
 340                        warning(_("failed to open pack-index '%s'"),
 341                                full_path);
 342                        close_pack(packs->list[packs->nr]);
 343                        FREE_AND_NULL(packs->list[packs->nr]);
 344                        return;
 345                }
 346
 347                packs->names[packs->nr] = xstrdup(file_name);
 348                packs->pack_name_concat_len += strlen(file_name) + 1;
 349                packs->nr++;
 350        }
 351}
 352
 353struct pack_pair {
 354        uint32_t pack_int_id;
 355        char *pack_name;
 356};
 357
 358static int pack_pair_compare(const void *_a, const void *_b)
 359{
 360        struct pack_pair *a = (struct pack_pair *)_a;
 361        struct pack_pair *b = (struct pack_pair *)_b;
 362        return strcmp(a->pack_name, b->pack_name);
 363}
 364
 365static void sort_packs_by_name(char **pack_names, uint32_t nr_packs, uint32_t *perm)
 366{
 367        uint32_t i;
 368        struct pack_pair *pairs;
 369
 370        ALLOC_ARRAY(pairs, nr_packs);
 371
 372        for (i = 0; i < nr_packs; i++) {
 373                pairs[i].pack_int_id = i;
 374                pairs[i].pack_name = pack_names[i];
 375        }
 376
 377        QSORT(pairs, nr_packs, pack_pair_compare);
 378
 379        for (i = 0; i < nr_packs; i++) {
 380                pack_names[i] = pairs[i].pack_name;
 381                perm[pairs[i].pack_int_id] = i;
 382        }
 383
 384        free(pairs);
 385}
 386
 387struct pack_midx_entry {
 388        struct object_id oid;
 389        uint32_t pack_int_id;
 390        time_t pack_mtime;
 391        uint64_t offset;
 392};
 393
 394static int midx_oid_compare(const void *_a, const void *_b)
 395{
 396        const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
 397        const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
 398        int cmp = oidcmp(&a->oid, &b->oid);
 399
 400        if (cmp)
 401                return cmp;
 402
 403        if (a->pack_mtime > b->pack_mtime)
 404                return -1;
 405        else if (a->pack_mtime < b->pack_mtime)
 406                return 1;
 407
 408        return a->pack_int_id - b->pack_int_id;
 409}
 410
 411static void fill_pack_entry(uint32_t pack_int_id,
 412                            struct packed_git *p,
 413                            uint32_t cur_object,
 414                            struct pack_midx_entry *entry)
 415{
 416        if (!nth_packed_object_oid(&entry->oid, p, cur_object))
 417                die(_("failed to locate object %d in packfile"), cur_object);
 418
 419        entry->pack_int_id = pack_int_id;
 420        entry->pack_mtime = p->mtime;
 421
 422        entry->offset = nth_packed_object_offset(p, cur_object);
 423}
 424
 425/*
 426 * It is possible to artificially get into a state where there are many
 427 * duplicate copies of objects. That can create high memory pressure if
 428 * we are to create a list of all objects before de-duplication. To reduce
 429 * this memory pressure without a significant performance drop, automatically
 430 * group objects by the first byte of their object id. Use the IDX fanout
 431 * tables to group the data, copy to a local array, then sort.
 432 *
 433 * Copy only the de-duplicated entries (selected by most-recent modified time
 434 * of a packfile containing the object).
 435 */
 436static struct pack_midx_entry *get_sorted_entries(struct packed_git **p,
 437                                                  uint32_t *perm,
 438                                                  uint32_t nr_packs,
 439                                                  uint32_t *nr_objects)
 440{
 441        uint32_t cur_fanout, cur_pack, cur_object;
 442        uint32_t alloc_fanout, alloc_objects, total_objects = 0;
 443        struct pack_midx_entry *entries_by_fanout = NULL;
 444        struct pack_midx_entry *deduplicated_entries = NULL;
 445
 446        for (cur_pack = 0; cur_pack < nr_packs; cur_pack++)
 447                total_objects += p[cur_pack]->num_objects;
 448
 449        /*
 450         * As we de-duplicate by fanout value, we expect the fanout
 451         * slices to be evenly distributed, with some noise. Hence,
 452         * allocate slightly more than one 256th.
 453         */
 454        alloc_objects = alloc_fanout = total_objects > 3200 ? total_objects / 200 : 16;
 455
 456        ALLOC_ARRAY(entries_by_fanout, alloc_fanout);
 457        ALLOC_ARRAY(deduplicated_entries, alloc_objects);
 458        *nr_objects = 0;
 459
 460        for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
 461                uint32_t nr_fanout = 0;
 462
 463                for (cur_pack = 0; cur_pack < nr_packs; cur_pack++) {
 464                        uint32_t start = 0, end;
 465
 466                        if (cur_fanout)
 467                                start = get_pack_fanout(p[cur_pack], cur_fanout - 1);
 468                        end = get_pack_fanout(p[cur_pack], cur_fanout);
 469
 470                        for (cur_object = start; cur_object < end; cur_object++) {
 471                                ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
 472                                fill_pack_entry(perm[cur_pack], p[cur_pack], cur_object, &entries_by_fanout[nr_fanout]);
 473                                nr_fanout++;
 474                        }
 475                }
 476
 477                QSORT(entries_by_fanout, nr_fanout, midx_oid_compare);
 478
 479                /*
 480                 * The batch is now sorted by OID and then mtime (descending).
 481                 * Take only the first duplicate.
 482                 */
 483                for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
 484                        if (cur_object && !oidcmp(&entries_by_fanout[cur_object - 1].oid,
 485                                                  &entries_by_fanout[cur_object].oid))
 486                                continue;
 487
 488                        ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
 489                        memcpy(&deduplicated_entries[*nr_objects],
 490                               &entries_by_fanout[cur_object],
 491                               sizeof(struct pack_midx_entry));
 492                        (*nr_objects)++;
 493                }
 494        }
 495
 496        free(entries_by_fanout);
 497        return deduplicated_entries;
 498}
 499
 500static size_t write_midx_pack_names(struct hashfile *f,
 501                                    char **pack_names,
 502                                    uint32_t num_packs)
 503{
 504        uint32_t i;
 505        unsigned char padding[MIDX_CHUNK_ALIGNMENT];
 506        size_t written = 0;
 507
 508        for (i = 0; i < num_packs; i++) {
 509                size_t writelen = strlen(pack_names[i]) + 1;
 510
 511                if (i && strcmp(pack_names[i], pack_names[i - 1]) <= 0)
 512                        BUG("incorrect pack-file order: %s before %s",
 513                            pack_names[i - 1],
 514                            pack_names[i]);
 515
 516                hashwrite(f, pack_names[i], writelen);
 517                written += writelen;
 518        }
 519
 520        /* add padding to be aligned */
 521        i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
 522        if (i < MIDX_CHUNK_ALIGNMENT) {
 523                memset(padding, 0, sizeof(padding));
 524                hashwrite(f, padding, i);
 525                written += i;
 526        }
 527
 528        return written;
 529}
 530
 531static size_t write_midx_oid_fanout(struct hashfile *f,
 532                                    struct pack_midx_entry *objects,
 533                                    uint32_t nr_objects)
 534{
 535        struct pack_midx_entry *list = objects;
 536        struct pack_midx_entry *last = objects + nr_objects;
 537        uint32_t count = 0;
 538        uint32_t i;
 539
 540        /*
 541        * Write the first-level table (the list is sorted,
 542        * but we use a 256-entry lookup to be able to avoid
 543        * having to do eight extra binary search iterations).
 544        */
 545        for (i = 0; i < 256; i++) {
 546                struct pack_midx_entry *next = list;
 547
 548                while (next < last && next->oid.hash[0] == i) {
 549                        count++;
 550                        next++;
 551                }
 552
 553                hashwrite_be32(f, count);
 554                list = next;
 555        }
 556
 557        return MIDX_CHUNK_FANOUT_SIZE;
 558}
 559
 560static size_t write_midx_oid_lookup(struct hashfile *f, unsigned char hash_len,
 561                                    struct pack_midx_entry *objects,
 562                                    uint32_t nr_objects)
 563{
 564        struct pack_midx_entry *list = objects;
 565        uint32_t i;
 566        size_t written = 0;
 567
 568        for (i = 0; i < nr_objects; i++) {
 569                struct pack_midx_entry *obj = list++;
 570
 571                if (i < nr_objects - 1) {
 572                        struct pack_midx_entry *next = list;
 573                        if (oidcmp(&obj->oid, &next->oid) >= 0)
 574                                BUG("OIDs not in order: %s >= %s",
 575                                    oid_to_hex(&obj->oid),
 576                                    oid_to_hex(&next->oid));
 577                }
 578
 579                hashwrite(f, obj->oid.hash, (int)hash_len);
 580                written += hash_len;
 581        }
 582
 583        return written;
 584}
 585
 586static size_t write_midx_object_offsets(struct hashfile *f, int large_offset_needed,
 587                                        struct pack_midx_entry *objects, uint32_t nr_objects)
 588{
 589        struct pack_midx_entry *list = objects;
 590        uint32_t i, nr_large_offset = 0;
 591        size_t written = 0;
 592
 593        for (i = 0; i < nr_objects; i++) {
 594                struct pack_midx_entry *obj = list++;
 595
 596                hashwrite_be32(f, obj->pack_int_id);
 597
 598                if (large_offset_needed && obj->offset >> 31)
 599                        hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
 600                else if (!large_offset_needed && obj->offset >> 32)
 601                        BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
 602                            oid_to_hex(&obj->oid),
 603                            obj->offset);
 604                else
 605                        hashwrite_be32(f, (uint32_t)obj->offset);
 606
 607                written += MIDX_CHUNK_OFFSET_WIDTH;
 608        }
 609
 610        return written;
 611}
 612
 613static size_t write_midx_large_offsets(struct hashfile *f, uint32_t nr_large_offset,
 614                                       struct pack_midx_entry *objects, uint32_t nr_objects)
 615{
 616        struct pack_midx_entry *list = objects;
 617        size_t written = 0;
 618
 619        while (nr_large_offset) {
 620                struct pack_midx_entry *obj = list++;
 621                uint64_t offset = obj->offset;
 622
 623                if (!(offset >> 31))
 624                        continue;
 625
 626                hashwrite_be32(f, offset >> 32);
 627                hashwrite_be32(f, offset & 0xffffffffUL);
 628                written += 2 * sizeof(uint32_t);
 629
 630                nr_large_offset--;
 631        }
 632
 633        return written;
 634}
 635
 636int write_midx_file(const char *object_dir)
 637{
 638        unsigned char cur_chunk, num_chunks = 0;
 639        char *midx_name;
 640        uint32_t i;
 641        struct hashfile *f = NULL;
 642        struct lock_file lk;
 643        struct pack_list packs;
 644        uint32_t *pack_perm = NULL;
 645        uint64_t written = 0;
 646        uint32_t chunk_ids[MIDX_MAX_CHUNKS + 1];
 647        uint64_t chunk_offsets[MIDX_MAX_CHUNKS + 1];
 648        uint32_t nr_entries, num_large_offsets = 0;
 649        struct pack_midx_entry *entries = NULL;
 650        int large_offsets_needed = 0;
 651
 652        midx_name = get_midx_filename(object_dir);
 653        if (safe_create_leading_directories(midx_name)) {
 654                UNLEAK(midx_name);
 655                die_errno(_("unable to create leading directories of %s"),
 656                          midx_name);
 657        }
 658
 659        packs.nr = 0;
 660        packs.alloc_list = 16;
 661        packs.alloc_names = 16;
 662        packs.list = NULL;
 663        packs.pack_name_concat_len = 0;
 664        ALLOC_ARRAY(packs.list, packs.alloc_list);
 665        ALLOC_ARRAY(packs.names, packs.alloc_names);
 666
 667        for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
 668
 669        if (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
 670                packs.pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
 671                                              (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
 672
 673        ALLOC_ARRAY(pack_perm, packs.nr);
 674        sort_packs_by_name(packs.names, packs.nr, pack_perm);
 675
 676        entries = get_sorted_entries(packs.list, pack_perm, packs.nr, &nr_entries);
 677        for (i = 0; i < nr_entries; i++) {
 678                if (entries[i].offset > 0x7fffffff)
 679                        num_large_offsets++;
 680                if (entries[i].offset > 0xffffffff)
 681                        large_offsets_needed = 1;
 682        }
 683
 684        hold_lock_file_for_update(&lk, midx_name, LOCK_DIE_ON_ERROR);
 685        f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
 686        FREE_AND_NULL(midx_name);
 687
 688        cur_chunk = 0;
 689        num_chunks = large_offsets_needed ? 5 : 4;
 690
 691        written = write_midx_header(f, num_chunks, packs.nr);
 692
 693        chunk_ids[cur_chunk] = MIDX_CHUNKID_PACKNAMES;
 694        chunk_offsets[cur_chunk] = written + (num_chunks + 1) * MIDX_CHUNKLOOKUP_WIDTH;
 695
 696        cur_chunk++;
 697        chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDFANOUT;
 698        chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + packs.pack_name_concat_len;
 699
 700        cur_chunk++;
 701        chunk_ids[cur_chunk] = MIDX_CHUNKID_OIDLOOKUP;
 702        chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + MIDX_CHUNK_FANOUT_SIZE;
 703
 704        cur_chunk++;
 705        chunk_ids[cur_chunk] = MIDX_CHUNKID_OBJECTOFFSETS;
 706        chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_HASH_LEN;
 707
 708        cur_chunk++;
 709        chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] + nr_entries * MIDX_CHUNK_OFFSET_WIDTH;
 710        if (large_offsets_needed) {
 711                chunk_ids[cur_chunk] = MIDX_CHUNKID_LARGEOFFSETS;
 712
 713                cur_chunk++;
 714                chunk_offsets[cur_chunk] = chunk_offsets[cur_chunk - 1] +
 715                                           num_large_offsets * MIDX_CHUNK_LARGE_OFFSET_WIDTH;
 716        }
 717
 718        chunk_ids[cur_chunk] = 0;
 719
 720        for (i = 0; i <= num_chunks; i++) {
 721                if (i && chunk_offsets[i] < chunk_offsets[i - 1])
 722                        BUG("incorrect chunk offsets: %"PRIu64" before %"PRIu64,
 723                            chunk_offsets[i - 1],
 724                            chunk_offsets[i]);
 725
 726                if (chunk_offsets[i] % MIDX_CHUNK_ALIGNMENT)
 727                        BUG("chunk offset %"PRIu64" is not properly aligned",
 728                            chunk_offsets[i]);
 729
 730                hashwrite_be32(f, chunk_ids[i]);
 731                hashwrite_be32(f, chunk_offsets[i] >> 32);
 732                hashwrite_be32(f, chunk_offsets[i]);
 733
 734                written += MIDX_CHUNKLOOKUP_WIDTH;
 735        }
 736
 737        for (i = 0; i < num_chunks; i++) {
 738                if (written != chunk_offsets[i])
 739                        BUG("incorrect chunk offset (%"PRIu64" != %"PRIu64") for chunk id %"PRIx32,
 740                            chunk_offsets[i],
 741                            written,
 742                            chunk_ids[i]);
 743
 744                switch (chunk_ids[i]) {
 745                        case MIDX_CHUNKID_PACKNAMES:
 746                                written += write_midx_pack_names(f, packs.names, packs.nr);
 747                                break;
 748
 749                        case MIDX_CHUNKID_OIDFANOUT:
 750                                written += write_midx_oid_fanout(f, entries, nr_entries);
 751                                break;
 752
 753                        case MIDX_CHUNKID_OIDLOOKUP:
 754                                written += write_midx_oid_lookup(f, MIDX_HASH_LEN, entries, nr_entries);
 755                                break;
 756
 757                        case MIDX_CHUNKID_OBJECTOFFSETS:
 758                                written += write_midx_object_offsets(f, large_offsets_needed, entries, nr_entries);
 759                                break;
 760
 761                        case MIDX_CHUNKID_LARGEOFFSETS:
 762                                written += write_midx_large_offsets(f, num_large_offsets, entries, nr_entries);
 763                                break;
 764
 765                        default:
 766                                BUG("trying to write unknown chunk id %"PRIx32,
 767                                    chunk_ids[i]);
 768                }
 769        }
 770
 771        if (written != chunk_offsets[num_chunks])
 772                BUG("incorrect final offset %"PRIu64" != %"PRIu64,
 773                    written,
 774                    chunk_offsets[num_chunks]);
 775
 776        finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
 777        commit_lock_file(&lk);
 778
 779        for (i = 0; i < packs.nr; i++) {
 780                if (packs.list[i]) {
 781                        close_pack(packs.list[i]);
 782                        free(packs.list[i]);
 783                }
 784                free(packs.names[i]);
 785        }
 786
 787        free(packs.list);
 788        free(packs.names);
 789        free(entries);
 790        return 0;
 791}