commit-graph.con commit Merge branch 'jk/trailer-fixes' (fba9654)
   1#include "cache.h"
   2#include "config.h"
   3#include "dir.h"
   4#include "git-compat-util.h"
   5#include "lockfile.h"
   6#include "pack.h"
   7#include "packfile.h"
   8#include "commit.h"
   9#include "object.h"
  10#include "refs.h"
  11#include "revision.h"
  12#include "sha1-lookup.h"
  13#include "commit-graph.h"
  14#include "object-store.h"
  15#include "alloc.h"
  16
  17#define GRAPH_SIGNATURE 0x43475048 /* "CGPH" */
  18#define GRAPH_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
  19#define GRAPH_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
  20#define GRAPH_CHUNKID_DATA 0x43444154 /* "CDAT" */
  21#define GRAPH_CHUNKID_LARGEEDGES 0x45444745 /* "EDGE" */
  22
  23#define GRAPH_DATA_WIDTH 36
  24
  25#define GRAPH_VERSION_1 0x1
  26#define GRAPH_VERSION GRAPH_VERSION_1
  27
  28#define GRAPH_OID_VERSION_SHA1 1
  29#define GRAPH_OID_LEN_SHA1 GIT_SHA1_RAWSZ
  30#define GRAPH_OID_VERSION GRAPH_OID_VERSION_SHA1
  31#define GRAPH_OID_LEN GRAPH_OID_LEN_SHA1
  32
  33#define GRAPH_OCTOPUS_EDGES_NEEDED 0x80000000
  34#define GRAPH_PARENT_MISSING 0x7fffffff
  35#define GRAPH_EDGE_LAST_MASK 0x7fffffff
  36#define GRAPH_PARENT_NONE 0x70000000
  37
  38#define GRAPH_LAST_EDGE 0x80000000
  39
  40#define GRAPH_HEADER_SIZE 8
  41#define GRAPH_FANOUT_SIZE (4 * 256)
  42#define GRAPH_CHUNKLOOKUP_WIDTH 12
  43#define GRAPH_MIN_SIZE (GRAPH_HEADER_SIZE + 4 * GRAPH_CHUNKLOOKUP_WIDTH \
  44                        + GRAPH_FANOUT_SIZE + GRAPH_OID_LEN)
  45
  46char *get_commit_graph_filename(const char *obj_dir)
  47{
  48        return xstrfmt("%s/info/commit-graph", obj_dir);
  49}
  50
  51static struct commit_graph *alloc_commit_graph(void)
  52{
  53        struct commit_graph *g = xcalloc(1, sizeof(*g));
  54        g->graph_fd = -1;
  55
  56        return g;
  57}
  58
  59struct commit_graph *load_commit_graph_one(const char *graph_file)
  60{
  61        void *graph_map;
  62        const unsigned char *data, *chunk_lookup;
  63        size_t graph_size;
  64        struct stat st;
  65        uint32_t i;
  66        struct commit_graph *graph;
  67        int fd = git_open(graph_file);
  68        uint64_t last_chunk_offset;
  69        uint32_t last_chunk_id;
  70        uint32_t graph_signature;
  71        unsigned char graph_version, hash_version;
  72
  73        if (fd < 0)
  74                return NULL;
  75        if (fstat(fd, &st)) {
  76                close(fd);
  77                return NULL;
  78        }
  79        graph_size = xsize_t(st.st_size);
  80
  81        if (graph_size < GRAPH_MIN_SIZE) {
  82                close(fd);
  83                die(_("graph file %s is too small"), graph_file);
  84        }
  85        graph_map = xmmap(NULL, graph_size, PROT_READ, MAP_PRIVATE, fd, 0);
  86        data = (const unsigned char *)graph_map;
  87
  88        graph_signature = get_be32(data);
  89        if (graph_signature != GRAPH_SIGNATURE) {
  90                error(_("graph signature %X does not match signature %X"),
  91                      graph_signature, GRAPH_SIGNATURE);
  92                goto cleanup_fail;
  93        }
  94
  95        graph_version = *(unsigned char*)(data + 4);
  96        if (graph_version != GRAPH_VERSION) {
  97                error(_("graph version %X does not match version %X"),
  98                      graph_version, GRAPH_VERSION);
  99                goto cleanup_fail;
 100        }
 101
 102        hash_version = *(unsigned char*)(data + 5);
 103        if (hash_version != GRAPH_OID_VERSION) {
 104                error(_("hash version %X does not match version %X"),
 105                      hash_version, GRAPH_OID_VERSION);
 106                goto cleanup_fail;
 107        }
 108
 109        graph = alloc_commit_graph();
 110
 111        graph->hash_len = GRAPH_OID_LEN;
 112        graph->num_chunks = *(unsigned char*)(data + 6);
 113        graph->graph_fd = fd;
 114        graph->data = graph_map;
 115        graph->data_len = graph_size;
 116
 117        last_chunk_id = 0;
 118        last_chunk_offset = 8;
 119        chunk_lookup = data + 8;
 120        for (i = 0; i < graph->num_chunks; i++) {
 121                uint32_t chunk_id = get_be32(chunk_lookup + 0);
 122                uint64_t chunk_offset = get_be64(chunk_lookup + 4);
 123                int chunk_repeated = 0;
 124
 125                chunk_lookup += GRAPH_CHUNKLOOKUP_WIDTH;
 126
 127                if (chunk_offset > graph_size - GIT_MAX_RAWSZ) {
 128                        error(_("improper chunk offset %08x%08x"), (uint32_t)(chunk_offset >> 32),
 129                              (uint32_t)chunk_offset);
 130                        goto cleanup_fail;
 131                }
 132
 133                switch (chunk_id) {
 134                case GRAPH_CHUNKID_OIDFANOUT:
 135                        if (graph->chunk_oid_fanout)
 136                                chunk_repeated = 1;
 137                        else
 138                                graph->chunk_oid_fanout = (uint32_t*)(data + chunk_offset);
 139                        break;
 140
 141                case GRAPH_CHUNKID_OIDLOOKUP:
 142                        if (graph->chunk_oid_lookup)
 143                                chunk_repeated = 1;
 144                        else
 145                                graph->chunk_oid_lookup = data + chunk_offset;
 146                        break;
 147
 148                case GRAPH_CHUNKID_DATA:
 149                        if (graph->chunk_commit_data)
 150                                chunk_repeated = 1;
 151                        else
 152                                graph->chunk_commit_data = data + chunk_offset;
 153                        break;
 154
 155                case GRAPH_CHUNKID_LARGEEDGES:
 156                        if (graph->chunk_large_edges)
 157                                chunk_repeated = 1;
 158                        else
 159                                graph->chunk_large_edges = data + chunk_offset;
 160                        break;
 161                }
 162
 163                if (chunk_repeated) {
 164                        error(_("chunk id %08x appears multiple times"), chunk_id);
 165                        goto cleanup_fail;
 166                }
 167
 168                if (last_chunk_id == GRAPH_CHUNKID_OIDLOOKUP)
 169                {
 170                        graph->num_commits = (chunk_offset - last_chunk_offset)
 171                                             / graph->hash_len;
 172                }
 173
 174                last_chunk_id = chunk_id;
 175                last_chunk_offset = chunk_offset;
 176        }
 177
 178        return graph;
 179
 180cleanup_fail:
 181        munmap(graph_map, graph_size);
 182        close(fd);
 183        exit(1);
 184}
 185
 186static void prepare_commit_graph_one(struct repository *r, const char *obj_dir)
 187{
 188        char *graph_name;
 189
 190        if (r->objects->commit_graph)
 191                return;
 192
 193        graph_name = get_commit_graph_filename(obj_dir);
 194        r->objects->commit_graph =
 195                load_commit_graph_one(graph_name);
 196
 197        FREE_AND_NULL(graph_name);
 198}
 199
 200/*
 201 * Return 1 if commit_graph is non-NULL, and 0 otherwise.
 202 *
 203 * On the first invocation, this function attemps to load the commit
 204 * graph if the_repository is configured to have one.
 205 */
 206static int prepare_commit_graph(struct repository *r)
 207{
 208        struct alternate_object_database *alt;
 209        char *obj_dir;
 210        int config_value;
 211
 212        if (r->objects->commit_graph_attempted)
 213                return !!r->objects->commit_graph;
 214        r->objects->commit_graph_attempted = 1;
 215
 216        if (repo_config_get_bool(r, "core.commitgraph", &config_value) ||
 217            !config_value)
 218                /*
 219                 * This repository is not configured to use commit graphs, so
 220                 * do not load one. (But report commit_graph_attempted anyway
 221                 * so that commit graph loading is not attempted again for this
 222                 * repository.)
 223                 */
 224                return 0;
 225
 226        obj_dir = r->objects->objectdir;
 227        prepare_commit_graph_one(r, obj_dir);
 228        prepare_alt_odb(r);
 229        for (alt = r->objects->alt_odb_list;
 230             !r->objects->commit_graph && alt;
 231             alt = alt->next)
 232                prepare_commit_graph_one(r, alt->path);
 233        return !!r->objects->commit_graph;
 234}
 235
 236int generation_numbers_enabled(struct repository *r)
 237{
 238        uint32_t first_generation;
 239        struct commit_graph *g;
 240        if (!prepare_commit_graph(r))
 241               return 0;
 242
 243        g = r->objects->commit_graph;
 244
 245        if (!g->num_commits)
 246                return 0;
 247
 248        first_generation = get_be32(g->chunk_commit_data +
 249                                    g->hash_len + 8) >> 2;
 250
 251        return !!first_generation;
 252}
 253
 254static void close_commit_graph(void)
 255{
 256        free_commit_graph(the_repository->objects->commit_graph);
 257        the_repository->objects->commit_graph = NULL;
 258}
 259
 260static int bsearch_graph(struct commit_graph *g, struct object_id *oid, uint32_t *pos)
 261{
 262        return bsearch_hash(oid->hash, g->chunk_oid_fanout,
 263                            g->chunk_oid_lookup, g->hash_len, pos);
 264}
 265
 266static struct commit_list **insert_parent_or_die(struct commit_graph *g,
 267                                                 uint64_t pos,
 268                                                 struct commit_list **pptr)
 269{
 270        struct commit *c;
 271        struct object_id oid;
 272
 273        if (pos >= g->num_commits)
 274                die("invalid parent position %"PRIu64, pos);
 275
 276        hashcpy(oid.hash, g->chunk_oid_lookup + g->hash_len * pos);
 277        c = lookup_commit(the_repository, &oid);
 278        if (!c)
 279                die(_("could not find commit %s"), oid_to_hex(&oid));
 280        c->graph_pos = pos;
 281        return &commit_list_insert(c, pptr)->next;
 282}
 283
 284static void fill_commit_graph_info(struct commit *item, struct commit_graph *g, uint32_t pos)
 285{
 286        const unsigned char *commit_data = g->chunk_commit_data + GRAPH_DATA_WIDTH * pos;
 287        item->graph_pos = pos;
 288        item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
 289}
 290
 291static int fill_commit_in_graph(struct commit *item, struct commit_graph *g, uint32_t pos)
 292{
 293        uint32_t edge_value;
 294        uint32_t *parent_data_ptr;
 295        uint64_t date_low, date_high;
 296        struct commit_list **pptr;
 297        const unsigned char *commit_data = g->chunk_commit_data + (g->hash_len + 16) * pos;
 298
 299        item->object.parsed = 1;
 300        item->graph_pos = pos;
 301
 302        item->maybe_tree = NULL;
 303
 304        date_high = get_be32(commit_data + g->hash_len + 8) & 0x3;
 305        date_low = get_be32(commit_data + g->hash_len + 12);
 306        item->date = (timestamp_t)((date_high << 32) | date_low);
 307
 308        item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
 309
 310        pptr = &item->parents;
 311
 312        edge_value = get_be32(commit_data + g->hash_len);
 313        if (edge_value == GRAPH_PARENT_NONE)
 314                return 1;
 315        pptr = insert_parent_or_die(g, edge_value, pptr);
 316
 317        edge_value = get_be32(commit_data + g->hash_len + 4);
 318        if (edge_value == GRAPH_PARENT_NONE)
 319                return 1;
 320        if (!(edge_value & GRAPH_OCTOPUS_EDGES_NEEDED)) {
 321                pptr = insert_parent_or_die(g, edge_value, pptr);
 322                return 1;
 323        }
 324
 325        parent_data_ptr = (uint32_t*)(g->chunk_large_edges +
 326                          4 * (uint64_t)(edge_value & GRAPH_EDGE_LAST_MASK));
 327        do {
 328                edge_value = get_be32(parent_data_ptr);
 329                pptr = insert_parent_or_die(g,
 330                                            edge_value & GRAPH_EDGE_LAST_MASK,
 331                                            pptr);
 332                parent_data_ptr++;
 333        } while (!(edge_value & GRAPH_LAST_EDGE));
 334
 335        return 1;
 336}
 337
 338static int find_commit_in_graph(struct commit *item, struct commit_graph *g, uint32_t *pos)
 339{
 340        if (item->graph_pos != COMMIT_NOT_FROM_GRAPH) {
 341                *pos = item->graph_pos;
 342                return 1;
 343        } else {
 344                return bsearch_graph(g, &(item->object.oid), pos);
 345        }
 346}
 347
 348static int parse_commit_in_graph_one(struct commit_graph *g, struct commit *item)
 349{
 350        uint32_t pos;
 351
 352        if (item->object.parsed)
 353                return 1;
 354
 355        if (find_commit_in_graph(item, g, &pos))
 356                return fill_commit_in_graph(item, g, pos);
 357
 358        return 0;
 359}
 360
 361int parse_commit_in_graph(struct repository *r, struct commit *item)
 362{
 363        if (!prepare_commit_graph(r))
 364                return 0;
 365        return parse_commit_in_graph_one(r->objects->commit_graph, item);
 366}
 367
 368void load_commit_graph_info(struct repository *r, struct commit *item)
 369{
 370        uint32_t pos;
 371        if (!prepare_commit_graph(r))
 372                return;
 373        if (find_commit_in_graph(item, r->objects->commit_graph, &pos))
 374                fill_commit_graph_info(item, r->objects->commit_graph, pos);
 375}
 376
 377static struct tree *load_tree_for_commit(struct commit_graph *g, struct commit *c)
 378{
 379        struct object_id oid;
 380        const unsigned char *commit_data = g->chunk_commit_data +
 381                                           GRAPH_DATA_WIDTH * (c->graph_pos);
 382
 383        hashcpy(oid.hash, commit_data);
 384        c->maybe_tree = lookup_tree(the_repository, &oid);
 385
 386        return c->maybe_tree;
 387}
 388
 389static struct tree *get_commit_tree_in_graph_one(struct commit_graph *g,
 390                                                 const struct commit *c)
 391{
 392        if (c->maybe_tree)
 393                return c->maybe_tree;
 394        if (c->graph_pos == COMMIT_NOT_FROM_GRAPH)
 395                BUG("get_commit_tree_in_graph_one called from non-commit-graph commit");
 396
 397        return load_tree_for_commit(g, (struct commit *)c);
 398}
 399
 400struct tree *get_commit_tree_in_graph(struct repository *r, const struct commit *c)
 401{
 402        return get_commit_tree_in_graph_one(r->objects->commit_graph, c);
 403}
 404
 405static void write_graph_chunk_fanout(struct hashfile *f,
 406                                     struct commit **commits,
 407                                     int nr_commits)
 408{
 409        int i, count = 0;
 410        struct commit **list = commits;
 411
 412        /*
 413         * Write the first-level table (the list is sorted,
 414         * but we use a 256-entry lookup to be able to avoid
 415         * having to do eight extra binary search iterations).
 416         */
 417        for (i = 0; i < 256; i++) {
 418                while (count < nr_commits) {
 419                        if ((*list)->object.oid.hash[0] != i)
 420                                break;
 421                        count++;
 422                        list++;
 423                }
 424
 425                hashwrite_be32(f, count);
 426        }
 427}
 428
 429static void write_graph_chunk_oids(struct hashfile *f, int hash_len,
 430                                   struct commit **commits, int nr_commits)
 431{
 432        struct commit **list = commits;
 433        int count;
 434        for (count = 0; count < nr_commits; count++, list++)
 435                hashwrite(f, (*list)->object.oid.hash, (int)hash_len);
 436}
 437
 438static const unsigned char *commit_to_sha1(size_t index, void *table)
 439{
 440        struct commit **commits = table;
 441        return commits[index]->object.oid.hash;
 442}
 443
 444static void write_graph_chunk_data(struct hashfile *f, int hash_len,
 445                                   struct commit **commits, int nr_commits)
 446{
 447        struct commit **list = commits;
 448        struct commit **last = commits + nr_commits;
 449        uint32_t num_extra_edges = 0;
 450
 451        while (list < last) {
 452                struct commit_list *parent;
 453                int edge_value;
 454                uint32_t packedDate[2];
 455
 456                parse_commit(*list);
 457                hashwrite(f, get_commit_tree_oid(*list)->hash, hash_len);
 458
 459                parent = (*list)->parents;
 460
 461                if (!parent)
 462                        edge_value = GRAPH_PARENT_NONE;
 463                else {
 464                        edge_value = sha1_pos(parent->item->object.oid.hash,
 465                                              commits,
 466                                              nr_commits,
 467                                              commit_to_sha1);
 468
 469                        if (edge_value < 0)
 470                                edge_value = GRAPH_PARENT_MISSING;
 471                }
 472
 473                hashwrite_be32(f, edge_value);
 474
 475                if (parent)
 476                        parent = parent->next;
 477
 478                if (!parent)
 479                        edge_value = GRAPH_PARENT_NONE;
 480                else if (parent->next)
 481                        edge_value = GRAPH_OCTOPUS_EDGES_NEEDED | num_extra_edges;
 482                else {
 483                        edge_value = sha1_pos(parent->item->object.oid.hash,
 484                                              commits,
 485                                              nr_commits,
 486                                              commit_to_sha1);
 487                        if (edge_value < 0)
 488                                edge_value = GRAPH_PARENT_MISSING;
 489                }
 490
 491                hashwrite_be32(f, edge_value);
 492
 493                if (edge_value & GRAPH_OCTOPUS_EDGES_NEEDED) {
 494                        do {
 495                                num_extra_edges++;
 496                                parent = parent->next;
 497                        } while (parent);
 498                }
 499
 500                if (sizeof((*list)->date) > 4)
 501                        packedDate[0] = htonl(((*list)->date >> 32) & 0x3);
 502                else
 503                        packedDate[0] = 0;
 504
 505                packedDate[0] |= htonl((*list)->generation << 2);
 506
 507                packedDate[1] = htonl((*list)->date);
 508                hashwrite(f, packedDate, 8);
 509
 510                list++;
 511        }
 512}
 513
 514static void write_graph_chunk_large_edges(struct hashfile *f,
 515                                          struct commit **commits,
 516                                          int nr_commits)
 517{
 518        struct commit **list = commits;
 519        struct commit **last = commits + nr_commits;
 520        struct commit_list *parent;
 521
 522        while (list < last) {
 523                int num_parents = 0;
 524                for (parent = (*list)->parents; num_parents < 3 && parent;
 525                     parent = parent->next)
 526                        num_parents++;
 527
 528                if (num_parents <= 2) {
 529                        list++;
 530                        continue;
 531                }
 532
 533                /* Since num_parents > 2, this initializer is safe. */
 534                for (parent = (*list)->parents->next; parent; parent = parent->next) {
 535                        int edge_value = sha1_pos(parent->item->object.oid.hash,
 536                                                  commits,
 537                                                  nr_commits,
 538                                                  commit_to_sha1);
 539
 540                        if (edge_value < 0)
 541                                edge_value = GRAPH_PARENT_MISSING;
 542                        else if (!parent->next)
 543                                edge_value |= GRAPH_LAST_EDGE;
 544
 545                        hashwrite_be32(f, edge_value);
 546                }
 547
 548                list++;
 549        }
 550}
 551
 552static int commit_compare(const void *_a, const void *_b)
 553{
 554        const struct object_id *a = (const struct object_id *)_a;
 555        const struct object_id *b = (const struct object_id *)_b;
 556        return oidcmp(a, b);
 557}
 558
 559struct packed_commit_list {
 560        struct commit **list;
 561        int nr;
 562        int alloc;
 563};
 564
 565struct packed_oid_list {
 566        struct object_id *list;
 567        int nr;
 568        int alloc;
 569};
 570
 571static int add_packed_commits(const struct object_id *oid,
 572                              struct packed_git *pack,
 573                              uint32_t pos,
 574                              void *data)
 575{
 576        struct packed_oid_list *list = (struct packed_oid_list*)data;
 577        enum object_type type;
 578        off_t offset = nth_packed_object_offset(pack, pos);
 579        struct object_info oi = OBJECT_INFO_INIT;
 580
 581        oi.typep = &type;
 582        if (packed_object_info(the_repository, pack, offset, &oi) < 0)
 583                die(_("unable to get type of object %s"), oid_to_hex(oid));
 584
 585        if (type != OBJ_COMMIT)
 586                return 0;
 587
 588        ALLOC_GROW(list->list, list->nr + 1, list->alloc);
 589        oidcpy(&(list->list[list->nr]), oid);
 590        list->nr++;
 591
 592        return 0;
 593}
 594
 595static void add_missing_parents(struct packed_oid_list *oids, struct commit *commit)
 596{
 597        struct commit_list *parent;
 598        for (parent = commit->parents; parent; parent = parent->next) {
 599                if (!(parent->item->object.flags & UNINTERESTING)) {
 600                        ALLOC_GROW(oids->list, oids->nr + 1, oids->alloc);
 601                        oidcpy(&oids->list[oids->nr], &(parent->item->object.oid));
 602                        oids->nr++;
 603                        parent->item->object.flags |= UNINTERESTING;
 604                }
 605        }
 606}
 607
 608static void close_reachable(struct packed_oid_list *oids)
 609{
 610        int i;
 611        struct commit *commit;
 612
 613        for (i = 0; i < oids->nr; i++) {
 614                commit = lookup_commit(the_repository, &oids->list[i]);
 615                if (commit)
 616                        commit->object.flags |= UNINTERESTING;
 617        }
 618
 619        /*
 620         * As this loop runs, oids->nr may grow, but not more
 621         * than the number of missing commits in the reachable
 622         * closure.
 623         */
 624        for (i = 0; i < oids->nr; i++) {
 625                commit = lookup_commit(the_repository, &oids->list[i]);
 626
 627                if (commit && !parse_commit(commit))
 628                        add_missing_parents(oids, commit);
 629        }
 630
 631        for (i = 0; i < oids->nr; i++) {
 632                commit = lookup_commit(the_repository, &oids->list[i]);
 633
 634                if (commit)
 635                        commit->object.flags &= ~UNINTERESTING;
 636        }
 637}
 638
 639static void compute_generation_numbers(struct packed_commit_list* commits)
 640{
 641        int i;
 642        struct commit_list *list = NULL;
 643
 644        for (i = 0; i < commits->nr; i++) {
 645                if (commits->list[i]->generation != GENERATION_NUMBER_INFINITY &&
 646                    commits->list[i]->generation != GENERATION_NUMBER_ZERO)
 647                        continue;
 648
 649                commit_list_insert(commits->list[i], &list);
 650                while (list) {
 651                        struct commit *current = list->item;
 652                        struct commit_list *parent;
 653                        int all_parents_computed = 1;
 654                        uint32_t max_generation = 0;
 655
 656                        for (parent = current->parents; parent; parent = parent->next) {
 657                                if (parent->item->generation == GENERATION_NUMBER_INFINITY ||
 658                                    parent->item->generation == GENERATION_NUMBER_ZERO) {
 659                                        all_parents_computed = 0;
 660                                        commit_list_insert(parent->item, &list);
 661                                        break;
 662                                } else if (parent->item->generation > max_generation) {
 663                                        max_generation = parent->item->generation;
 664                                }
 665                        }
 666
 667                        if (all_parents_computed) {
 668                                current->generation = max_generation + 1;
 669                                pop_commit(&list);
 670
 671                                if (current->generation > GENERATION_NUMBER_MAX)
 672                                        current->generation = GENERATION_NUMBER_MAX;
 673                        }
 674                }
 675        }
 676}
 677
 678static int add_ref_to_list(const char *refname,
 679                           const struct object_id *oid,
 680                           int flags, void *cb_data)
 681{
 682        struct string_list *list = (struct string_list *)cb_data;
 683
 684        string_list_append(list, oid_to_hex(oid));
 685        return 0;
 686}
 687
 688void write_commit_graph_reachable(const char *obj_dir, int append)
 689{
 690        struct string_list list;
 691
 692        string_list_init(&list, 1);
 693        for_each_ref(add_ref_to_list, &list);
 694        write_commit_graph(obj_dir, NULL, &list, append);
 695}
 696
 697void write_commit_graph(const char *obj_dir,
 698                        struct string_list *pack_indexes,
 699                        struct string_list *commit_hex,
 700                        int append)
 701{
 702        struct packed_oid_list oids;
 703        struct packed_commit_list commits;
 704        struct hashfile *f;
 705        uint32_t i, count_distinct = 0;
 706        char *graph_name;
 707        struct lock_file lk = LOCK_INIT;
 708        uint32_t chunk_ids[5];
 709        uint64_t chunk_offsets[5];
 710        int num_chunks;
 711        int num_extra_edges;
 712        struct commit_list *parent;
 713
 714        oids.nr = 0;
 715        oids.alloc = approximate_object_count() / 4;
 716
 717        if (append) {
 718                prepare_commit_graph_one(the_repository, obj_dir);
 719                if (the_repository->objects->commit_graph)
 720                        oids.alloc += the_repository->objects->commit_graph->num_commits;
 721        }
 722
 723        if (oids.alloc < 1024)
 724                oids.alloc = 1024;
 725        ALLOC_ARRAY(oids.list, oids.alloc);
 726
 727        if (append && the_repository->objects->commit_graph) {
 728                struct commit_graph *commit_graph =
 729                        the_repository->objects->commit_graph;
 730                for (i = 0; i < commit_graph->num_commits; i++) {
 731                        const unsigned char *hash = commit_graph->chunk_oid_lookup +
 732                                commit_graph->hash_len * i;
 733                        hashcpy(oids.list[oids.nr++].hash, hash);
 734                }
 735        }
 736
 737        if (pack_indexes) {
 738                struct strbuf packname = STRBUF_INIT;
 739                int dirlen;
 740                strbuf_addf(&packname, "%s/pack/", obj_dir);
 741                dirlen = packname.len;
 742                for (i = 0; i < pack_indexes->nr; i++) {
 743                        struct packed_git *p;
 744                        strbuf_setlen(&packname, dirlen);
 745                        strbuf_addstr(&packname, pack_indexes->items[i].string);
 746                        p = add_packed_git(packname.buf, packname.len, 1);
 747                        if (!p)
 748                                die(_("error adding pack %s"), packname.buf);
 749                        if (open_pack_index(p))
 750                                die(_("error opening index for %s"), packname.buf);
 751                        for_each_object_in_pack(p, add_packed_commits, &oids, 0);
 752                        close_pack(p);
 753                }
 754                strbuf_release(&packname);
 755        }
 756
 757        if (commit_hex) {
 758                for (i = 0; i < commit_hex->nr; i++) {
 759                        const char *end;
 760                        struct object_id oid;
 761                        struct commit *result;
 762
 763                        if (commit_hex->items[i].string &&
 764                            parse_oid_hex(commit_hex->items[i].string, &oid, &end))
 765                                continue;
 766
 767                        result = lookup_commit_reference_gently(the_repository, &oid, 1);
 768
 769                        if (result) {
 770                                ALLOC_GROW(oids.list, oids.nr + 1, oids.alloc);
 771                                oidcpy(&oids.list[oids.nr], &(result->object.oid));
 772                                oids.nr++;
 773                        }
 774                }
 775        }
 776
 777        if (!pack_indexes && !commit_hex)
 778                for_each_packed_object(add_packed_commits, &oids, 0);
 779
 780        close_reachable(&oids);
 781
 782        QSORT(oids.list, oids.nr, commit_compare);
 783
 784        count_distinct = 1;
 785        for (i = 1; i < oids.nr; i++) {
 786                if (oidcmp(&oids.list[i-1], &oids.list[i]))
 787                        count_distinct++;
 788        }
 789
 790        if (count_distinct >= GRAPH_PARENT_MISSING)
 791                die(_("the commit graph format cannot write %d commits"), count_distinct);
 792
 793        commits.nr = 0;
 794        commits.alloc = count_distinct;
 795        ALLOC_ARRAY(commits.list, commits.alloc);
 796
 797        num_extra_edges = 0;
 798        for (i = 0; i < oids.nr; i++) {
 799                int num_parents = 0;
 800                if (i > 0 && !oidcmp(&oids.list[i-1], &oids.list[i]))
 801                        continue;
 802
 803                commits.list[commits.nr] = lookup_commit(the_repository, &oids.list[i]);
 804                parse_commit(commits.list[commits.nr]);
 805
 806                for (parent = commits.list[commits.nr]->parents;
 807                     parent; parent = parent->next)
 808                        num_parents++;
 809
 810                if (num_parents > 2)
 811                        num_extra_edges += num_parents - 1;
 812
 813                commits.nr++;
 814        }
 815        num_chunks = num_extra_edges ? 4 : 3;
 816
 817        if (commits.nr >= GRAPH_PARENT_MISSING)
 818                die(_("too many commits to write graph"));
 819
 820        compute_generation_numbers(&commits);
 821
 822        graph_name = get_commit_graph_filename(obj_dir);
 823        if (safe_create_leading_directories(graph_name))
 824                die_errno(_("unable to create leading directories of %s"),
 825                          graph_name);
 826
 827        hold_lock_file_for_update(&lk, graph_name, LOCK_DIE_ON_ERROR);
 828        f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
 829
 830        hashwrite_be32(f, GRAPH_SIGNATURE);
 831
 832        hashwrite_u8(f, GRAPH_VERSION);
 833        hashwrite_u8(f, GRAPH_OID_VERSION);
 834        hashwrite_u8(f, num_chunks);
 835        hashwrite_u8(f, 0); /* unused padding byte */
 836
 837        chunk_ids[0] = GRAPH_CHUNKID_OIDFANOUT;
 838        chunk_ids[1] = GRAPH_CHUNKID_OIDLOOKUP;
 839        chunk_ids[2] = GRAPH_CHUNKID_DATA;
 840        if (num_extra_edges)
 841                chunk_ids[3] = GRAPH_CHUNKID_LARGEEDGES;
 842        else
 843                chunk_ids[3] = 0;
 844        chunk_ids[4] = 0;
 845
 846        chunk_offsets[0] = 8 + (num_chunks + 1) * GRAPH_CHUNKLOOKUP_WIDTH;
 847        chunk_offsets[1] = chunk_offsets[0] + GRAPH_FANOUT_SIZE;
 848        chunk_offsets[2] = chunk_offsets[1] + GRAPH_OID_LEN * commits.nr;
 849        chunk_offsets[3] = chunk_offsets[2] + (GRAPH_OID_LEN + 16) * commits.nr;
 850        chunk_offsets[4] = chunk_offsets[3] + 4 * num_extra_edges;
 851
 852        for (i = 0; i <= num_chunks; i++) {
 853                uint32_t chunk_write[3];
 854
 855                chunk_write[0] = htonl(chunk_ids[i]);
 856                chunk_write[1] = htonl(chunk_offsets[i] >> 32);
 857                chunk_write[2] = htonl(chunk_offsets[i] & 0xffffffff);
 858                hashwrite(f, chunk_write, 12);
 859        }
 860
 861        write_graph_chunk_fanout(f, commits.list, commits.nr);
 862        write_graph_chunk_oids(f, GRAPH_OID_LEN, commits.list, commits.nr);
 863        write_graph_chunk_data(f, GRAPH_OID_LEN, commits.list, commits.nr);
 864        write_graph_chunk_large_edges(f, commits.list, commits.nr);
 865
 866        close_commit_graph();
 867        finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_FSYNC);
 868        commit_lock_file(&lk);
 869
 870        free(oids.list);
 871        oids.alloc = 0;
 872        oids.nr = 0;
 873}
 874
 875#define VERIFY_COMMIT_GRAPH_ERROR_HASH 2
 876static int verify_commit_graph_error;
 877
 878static void graph_report(const char *fmt, ...)
 879{
 880        va_list ap;
 881
 882        verify_commit_graph_error = 1;
 883        va_start(ap, fmt);
 884        vfprintf(stderr, fmt, ap);
 885        fprintf(stderr, "\n");
 886        va_end(ap);
 887}
 888
 889#define GENERATION_ZERO_EXISTS 1
 890#define GENERATION_NUMBER_EXISTS 2
 891
 892int verify_commit_graph(struct repository *r, struct commit_graph *g)
 893{
 894        uint32_t i, cur_fanout_pos = 0;
 895        struct object_id prev_oid, cur_oid, checksum;
 896        int generation_zero = 0;
 897        struct hashfile *f;
 898        int devnull;
 899
 900        if (!g) {
 901                graph_report("no commit-graph file loaded");
 902                return 1;
 903        }
 904
 905        verify_commit_graph_error = 0;
 906
 907        if (!g->chunk_oid_fanout)
 908                graph_report("commit-graph is missing the OID Fanout chunk");
 909        if (!g->chunk_oid_lookup)
 910                graph_report("commit-graph is missing the OID Lookup chunk");
 911        if (!g->chunk_commit_data)
 912                graph_report("commit-graph is missing the Commit Data chunk");
 913
 914        if (verify_commit_graph_error)
 915                return verify_commit_graph_error;
 916
 917        devnull = open("/dev/null", O_WRONLY);
 918        f = hashfd(devnull, NULL);
 919        hashwrite(f, g->data, g->data_len - g->hash_len);
 920        finalize_hashfile(f, checksum.hash, CSUM_CLOSE);
 921        if (hashcmp(checksum.hash, g->data + g->data_len - g->hash_len)) {
 922                graph_report(_("the commit-graph file has incorrect checksum and is likely corrupt"));
 923                verify_commit_graph_error = VERIFY_COMMIT_GRAPH_ERROR_HASH;
 924        }
 925
 926        for (i = 0; i < g->num_commits; i++) {
 927                struct commit *graph_commit;
 928
 929                hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
 930
 931                if (i && oidcmp(&prev_oid, &cur_oid) >= 0)
 932                        graph_report("commit-graph has incorrect OID order: %s then %s",
 933                                     oid_to_hex(&prev_oid),
 934                                     oid_to_hex(&cur_oid));
 935
 936                oidcpy(&prev_oid, &cur_oid);
 937
 938                while (cur_oid.hash[0] > cur_fanout_pos) {
 939                        uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
 940
 941                        if (i != fanout_value)
 942                                graph_report("commit-graph has incorrect fanout value: fanout[%d] = %u != %u",
 943                                             cur_fanout_pos, fanout_value, i);
 944                        cur_fanout_pos++;
 945                }
 946
 947                graph_commit = lookup_commit(r, &cur_oid);
 948                if (!parse_commit_in_graph_one(g, graph_commit))
 949                        graph_report("failed to parse %s from commit-graph",
 950                                     oid_to_hex(&cur_oid));
 951        }
 952
 953        while (cur_fanout_pos < 256) {
 954                uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
 955
 956                if (g->num_commits != fanout_value)
 957                        graph_report("commit-graph has incorrect fanout value: fanout[%d] = %u != %u",
 958                                     cur_fanout_pos, fanout_value, i);
 959
 960                cur_fanout_pos++;
 961        }
 962
 963        if (verify_commit_graph_error & ~VERIFY_COMMIT_GRAPH_ERROR_HASH)
 964                return verify_commit_graph_error;
 965
 966        for (i = 0; i < g->num_commits; i++) {
 967                struct commit *graph_commit, *odb_commit;
 968                struct commit_list *graph_parents, *odb_parents;
 969                uint32_t max_generation = 0;
 970
 971                hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
 972
 973                graph_commit = lookup_commit(r, &cur_oid);
 974                odb_commit = (struct commit *)create_object(r, cur_oid.hash, alloc_commit_node(r));
 975                if (parse_commit_internal(odb_commit, 0, 0)) {
 976                        graph_report("failed to parse %s from object database",
 977                                     oid_to_hex(&cur_oid));
 978                        continue;
 979                }
 980
 981                if (oidcmp(&get_commit_tree_in_graph_one(g, graph_commit)->object.oid,
 982                           get_commit_tree_oid(odb_commit)))
 983                        graph_report("root tree OID for commit %s in commit-graph is %s != %s",
 984                                     oid_to_hex(&cur_oid),
 985                                     oid_to_hex(get_commit_tree_oid(graph_commit)),
 986                                     oid_to_hex(get_commit_tree_oid(odb_commit)));
 987
 988                graph_parents = graph_commit->parents;
 989                odb_parents = odb_commit->parents;
 990
 991                while (graph_parents) {
 992                        if (odb_parents == NULL) {
 993                                graph_report("commit-graph parent list for commit %s is too long",
 994                                             oid_to_hex(&cur_oid));
 995                                break;
 996                        }
 997
 998                        if (oidcmp(&graph_parents->item->object.oid, &odb_parents->item->object.oid))
 999                                graph_report("commit-graph parent for %s is %s != %s",
1000                                             oid_to_hex(&cur_oid),
1001                                             oid_to_hex(&graph_parents->item->object.oid),
1002                                             oid_to_hex(&odb_parents->item->object.oid));
1003
1004                        if (graph_parents->item->generation > max_generation)
1005                                max_generation = graph_parents->item->generation;
1006
1007                        graph_parents = graph_parents->next;
1008                        odb_parents = odb_parents->next;
1009                }
1010
1011                if (odb_parents != NULL)
1012                        graph_report("commit-graph parent list for commit %s terminates early",
1013                                     oid_to_hex(&cur_oid));
1014
1015                if (!graph_commit->generation) {
1016                        if (generation_zero == GENERATION_NUMBER_EXISTS)
1017                                graph_report("commit-graph has generation number zero for commit %s, but non-zero elsewhere",
1018                                             oid_to_hex(&cur_oid));
1019                        generation_zero = GENERATION_ZERO_EXISTS;
1020                } else if (generation_zero == GENERATION_ZERO_EXISTS)
1021                        graph_report("commit-graph has non-zero generation number for commit %s, but zero elsewhere",
1022                                     oid_to_hex(&cur_oid));
1023
1024                if (generation_zero == GENERATION_ZERO_EXISTS)
1025                        continue;
1026
1027                /*
1028                 * If one of our parents has generation GENERATION_NUMBER_MAX, then
1029                 * our generation is also GENERATION_NUMBER_MAX. Decrement to avoid
1030                 * extra logic in the following condition.
1031                 */
1032                if (max_generation == GENERATION_NUMBER_MAX)
1033                        max_generation--;
1034
1035                if (graph_commit->generation != max_generation + 1)
1036                        graph_report("commit-graph generation for commit %s is %u != %u",
1037                                     oid_to_hex(&cur_oid),
1038                                     graph_commit->generation,
1039                                     max_generation + 1);
1040
1041                if (graph_commit->date != odb_commit->date)
1042                        graph_report("commit date for commit %s in commit-graph is %"PRItime" != %"PRItime,
1043                                     oid_to_hex(&cur_oid),
1044                                     graph_commit->date,
1045                                     odb_commit->date);
1046        }
1047
1048        return verify_commit_graph_error;
1049}
1050
1051void free_commit_graph(struct commit_graph *g)
1052{
1053        if (!g)
1054                return;
1055        if (g->graph_fd >= 0) {
1056                munmap((void *)g->data, g->data_len);
1057                g->data = NULL;
1058                close(g->graph_fd);
1059        }
1060        free(g);
1061}