commit-graph.con commit commit-graph: fix buffer read-overflow (d2b86fb)
   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#include "hashmap.h"
  17#include "replace-object.h"
  18#include "progress.h"
  19
  20#define GRAPH_SIGNATURE 0x43475048 /* "CGPH" */
  21#define GRAPH_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
  22#define GRAPH_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
  23#define GRAPH_CHUNKID_DATA 0x43444154 /* "CDAT" */
  24#define GRAPH_CHUNKID_LARGEEDGES 0x45444745 /* "EDGE" */
  25
  26#define GRAPH_DATA_WIDTH 36
  27
  28#define GRAPH_VERSION_1 0x1
  29#define GRAPH_VERSION GRAPH_VERSION_1
  30
  31#define GRAPH_OID_VERSION_SHA1 1
  32#define GRAPH_OID_LEN_SHA1 GIT_SHA1_RAWSZ
  33#define GRAPH_OID_VERSION GRAPH_OID_VERSION_SHA1
  34#define GRAPH_OID_LEN GRAPH_OID_LEN_SHA1
  35
  36#define GRAPH_OCTOPUS_EDGES_NEEDED 0x80000000
  37#define GRAPH_PARENT_MISSING 0x7fffffff
  38#define GRAPH_EDGE_LAST_MASK 0x7fffffff
  39#define GRAPH_PARENT_NONE 0x70000000
  40
  41#define GRAPH_LAST_EDGE 0x80000000
  42
  43#define GRAPH_HEADER_SIZE 8
  44#define GRAPH_FANOUT_SIZE (4 * 256)
  45#define GRAPH_CHUNKLOOKUP_WIDTH 12
  46#define GRAPH_MIN_SIZE (GRAPH_HEADER_SIZE + 4 * GRAPH_CHUNKLOOKUP_WIDTH \
  47                        + GRAPH_FANOUT_SIZE + GRAPH_OID_LEN)
  48
  49char *get_commit_graph_filename(const char *obj_dir)
  50{
  51        return xstrfmt("%s/info/commit-graph", obj_dir);
  52}
  53
  54static struct commit_graph *alloc_commit_graph(void)
  55{
  56        struct commit_graph *g = xcalloc(1, sizeof(*g));
  57        g->graph_fd = -1;
  58
  59        return g;
  60}
  61
  62extern int read_replace_refs;
  63
  64static int commit_graph_compatible(struct repository *r)
  65{
  66        if (!r->gitdir)
  67                return 0;
  68
  69        if (read_replace_refs) {
  70                prepare_replace_object(r);
  71                if (hashmap_get_size(&r->objects->replace_map->map))
  72                        return 0;
  73        }
  74
  75        prepare_commit_graft(r);
  76        if (r->parsed_objects && r->parsed_objects->grafts_nr)
  77                return 0;
  78        if (is_repository_shallow(r))
  79                return 0;
  80
  81        return 1;
  82}
  83
  84struct commit_graph *load_commit_graph_one(const char *graph_file)
  85{
  86        void *graph_map;
  87        size_t graph_size;
  88        struct stat st;
  89        struct commit_graph *ret;
  90        int fd = git_open(graph_file);
  91
  92        if (fd < 0)
  93                return NULL;
  94        if (fstat(fd, &st)) {
  95                close(fd);
  96                return NULL;
  97        }
  98        graph_size = xsize_t(st.st_size);
  99
 100        if (graph_size < GRAPH_MIN_SIZE) {
 101                close(fd);
 102                die(_("graph file %s is too small"), graph_file);
 103        }
 104        graph_map = xmmap(NULL, graph_size, PROT_READ, MAP_PRIVATE, fd, 0);
 105        ret = parse_commit_graph(graph_map, fd, graph_size);
 106
 107        if (!ret) {
 108                munmap(graph_map, graph_size);
 109                close(fd);
 110                exit(1);
 111        }
 112
 113        return ret;
 114}
 115
 116struct commit_graph *parse_commit_graph(void *graph_map, int fd,
 117                                        size_t graph_size)
 118{
 119        const unsigned char *data, *chunk_lookup;
 120        uint32_t i;
 121        struct commit_graph *graph;
 122        uint64_t last_chunk_offset;
 123        uint32_t last_chunk_id;
 124        uint32_t graph_signature;
 125        unsigned char graph_version, hash_version;
 126
 127        if (!graph_map)
 128                return NULL;
 129
 130        if (graph_size < GRAPH_MIN_SIZE)
 131                return NULL;
 132
 133        data = (const unsigned char *)graph_map;
 134
 135        graph_signature = get_be32(data);
 136        if (graph_signature != GRAPH_SIGNATURE) {
 137                error(_("graph signature %X does not match signature %X"),
 138                      graph_signature, GRAPH_SIGNATURE);
 139                return NULL;
 140        }
 141
 142        graph_version = *(unsigned char*)(data + 4);
 143        if (graph_version != GRAPH_VERSION) {
 144                error(_("graph version %X does not match version %X"),
 145                      graph_version, GRAPH_VERSION);
 146                return NULL;
 147        }
 148
 149        hash_version = *(unsigned char*)(data + 5);
 150        if (hash_version != GRAPH_OID_VERSION) {
 151                error(_("hash version %X does not match version %X"),
 152                      hash_version, GRAPH_OID_VERSION);
 153                return NULL;
 154        }
 155
 156        graph = alloc_commit_graph();
 157
 158        graph->hash_len = GRAPH_OID_LEN;
 159        graph->num_chunks = *(unsigned char*)(data + 6);
 160        graph->graph_fd = fd;
 161        graph->data = graph_map;
 162        graph->data_len = graph_size;
 163
 164        last_chunk_id = 0;
 165        last_chunk_offset = 8;
 166        chunk_lookup = data + 8;
 167        for (i = 0; i < graph->num_chunks; i++) {
 168                uint32_t chunk_id;
 169                uint64_t chunk_offset;
 170                int chunk_repeated = 0;
 171
 172                if (data + graph_size - chunk_lookup <
 173                    GRAPH_CHUNKLOOKUP_WIDTH) {
 174                        error(_("chunk lookup table entry missing; graph file may be incomplete"));
 175                        free(graph);
 176                        return NULL;
 177                }
 178
 179                chunk_id = get_be32(chunk_lookup + 0);
 180                chunk_offset = get_be64(chunk_lookup + 4);
 181
 182                chunk_lookup += GRAPH_CHUNKLOOKUP_WIDTH;
 183
 184                if (chunk_offset > graph_size - GIT_MAX_RAWSZ) {
 185                        error(_("improper chunk offset %08x%08x"), (uint32_t)(chunk_offset >> 32),
 186                              (uint32_t)chunk_offset);
 187                        free(graph);
 188                        return NULL;
 189                }
 190
 191                switch (chunk_id) {
 192                case GRAPH_CHUNKID_OIDFANOUT:
 193                        if (graph->chunk_oid_fanout)
 194                                chunk_repeated = 1;
 195                        else
 196                                graph->chunk_oid_fanout = (uint32_t*)(data + chunk_offset);
 197                        break;
 198
 199                case GRAPH_CHUNKID_OIDLOOKUP:
 200                        if (graph->chunk_oid_lookup)
 201                                chunk_repeated = 1;
 202                        else
 203                                graph->chunk_oid_lookup = data + chunk_offset;
 204                        break;
 205
 206                case GRAPH_CHUNKID_DATA:
 207                        if (graph->chunk_commit_data)
 208                                chunk_repeated = 1;
 209                        else
 210                                graph->chunk_commit_data = data + chunk_offset;
 211                        break;
 212
 213                case GRAPH_CHUNKID_LARGEEDGES:
 214                        if (graph->chunk_large_edges)
 215                                chunk_repeated = 1;
 216                        else
 217                                graph->chunk_large_edges = data + chunk_offset;
 218                        break;
 219                }
 220
 221                if (chunk_repeated) {
 222                        error(_("chunk id %08x appears multiple times"), chunk_id);
 223                        free(graph);
 224                        return NULL;
 225                }
 226
 227                if (last_chunk_id == GRAPH_CHUNKID_OIDLOOKUP)
 228                {
 229                        graph->num_commits = (chunk_offset - last_chunk_offset)
 230                                             / graph->hash_len;
 231                }
 232
 233                last_chunk_id = chunk_id;
 234                last_chunk_offset = chunk_offset;
 235        }
 236
 237        return graph;
 238}
 239
 240static void prepare_commit_graph_one(struct repository *r, const char *obj_dir)
 241{
 242        char *graph_name;
 243
 244        if (r->objects->commit_graph)
 245                return;
 246
 247        graph_name = get_commit_graph_filename(obj_dir);
 248        r->objects->commit_graph =
 249                load_commit_graph_one(graph_name);
 250
 251        FREE_AND_NULL(graph_name);
 252}
 253
 254/*
 255 * Return 1 if commit_graph is non-NULL, and 0 otherwise.
 256 *
 257 * On the first invocation, this function attemps to load the commit
 258 * graph if the_repository is configured to have one.
 259 */
 260static int prepare_commit_graph(struct repository *r)
 261{
 262        struct object_directory *odb;
 263        int config_value;
 264
 265        if (r->objects->commit_graph_attempted)
 266                return !!r->objects->commit_graph;
 267        r->objects->commit_graph_attempted = 1;
 268
 269        if (!git_env_bool(GIT_TEST_COMMIT_GRAPH, 0) &&
 270            (repo_config_get_bool(r, "core.commitgraph", &config_value) ||
 271            !config_value))
 272                /*
 273                 * This repository is not configured to use commit graphs, so
 274                 * do not load one. (But report commit_graph_attempted anyway
 275                 * so that commit graph loading is not attempted again for this
 276                 * repository.)
 277                 */
 278                return 0;
 279
 280        if (!commit_graph_compatible(r))
 281                return 0;
 282
 283        prepare_alt_odb(r);
 284        for (odb = r->objects->odb;
 285             !r->objects->commit_graph && odb;
 286             odb = odb->next)
 287                prepare_commit_graph_one(r, odb->path);
 288        return !!r->objects->commit_graph;
 289}
 290
 291int generation_numbers_enabled(struct repository *r)
 292{
 293        uint32_t first_generation;
 294        struct commit_graph *g;
 295        if (!prepare_commit_graph(r))
 296               return 0;
 297
 298        g = r->objects->commit_graph;
 299
 300        if (!g->num_commits)
 301                return 0;
 302
 303        first_generation = get_be32(g->chunk_commit_data +
 304                                    g->hash_len + 8) >> 2;
 305
 306        return !!first_generation;
 307}
 308
 309void close_commit_graph(struct repository *r)
 310{
 311        free_commit_graph(r->objects->commit_graph);
 312        r->objects->commit_graph = NULL;
 313}
 314
 315static int bsearch_graph(struct commit_graph *g, struct object_id *oid, uint32_t *pos)
 316{
 317        return bsearch_hash(oid->hash, g->chunk_oid_fanout,
 318                            g->chunk_oid_lookup, g->hash_len, pos);
 319}
 320
 321static struct commit_list **insert_parent_or_die(struct commit_graph *g,
 322                                                 uint64_t pos,
 323                                                 struct commit_list **pptr)
 324{
 325        struct commit *c;
 326        struct object_id oid;
 327
 328        if (pos >= g->num_commits)
 329                die("invalid parent position %"PRIu64, pos);
 330
 331        hashcpy(oid.hash, g->chunk_oid_lookup + g->hash_len * pos);
 332        c = lookup_commit(the_repository, &oid);
 333        if (!c)
 334                die(_("could not find commit %s"), oid_to_hex(&oid));
 335        c->graph_pos = pos;
 336        return &commit_list_insert(c, pptr)->next;
 337}
 338
 339static void fill_commit_graph_info(struct commit *item, struct commit_graph *g, uint32_t pos)
 340{
 341        const unsigned char *commit_data = g->chunk_commit_data + GRAPH_DATA_WIDTH * pos;
 342        item->graph_pos = pos;
 343        item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
 344}
 345
 346static int fill_commit_in_graph(struct commit *item, struct commit_graph *g, uint32_t pos)
 347{
 348        uint32_t edge_value;
 349        uint32_t *parent_data_ptr;
 350        uint64_t date_low, date_high;
 351        struct commit_list **pptr;
 352        const unsigned char *commit_data = g->chunk_commit_data + (g->hash_len + 16) * pos;
 353
 354        item->object.parsed = 1;
 355        item->graph_pos = pos;
 356
 357        item->maybe_tree = NULL;
 358
 359        date_high = get_be32(commit_data + g->hash_len + 8) & 0x3;
 360        date_low = get_be32(commit_data + g->hash_len + 12);
 361        item->date = (timestamp_t)((date_high << 32) | date_low);
 362
 363        item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
 364
 365        pptr = &item->parents;
 366
 367        edge_value = get_be32(commit_data + g->hash_len);
 368        if (edge_value == GRAPH_PARENT_NONE)
 369                return 1;
 370        pptr = insert_parent_or_die(g, edge_value, pptr);
 371
 372        edge_value = get_be32(commit_data + g->hash_len + 4);
 373        if (edge_value == GRAPH_PARENT_NONE)
 374                return 1;
 375        if (!(edge_value & GRAPH_OCTOPUS_EDGES_NEEDED)) {
 376                pptr = insert_parent_or_die(g, edge_value, pptr);
 377                return 1;
 378        }
 379
 380        parent_data_ptr = (uint32_t*)(g->chunk_large_edges +
 381                          4 * (uint64_t)(edge_value & GRAPH_EDGE_LAST_MASK));
 382        do {
 383                edge_value = get_be32(parent_data_ptr);
 384                pptr = insert_parent_or_die(g,
 385                                            edge_value & GRAPH_EDGE_LAST_MASK,
 386                                            pptr);
 387                parent_data_ptr++;
 388        } while (!(edge_value & GRAPH_LAST_EDGE));
 389
 390        return 1;
 391}
 392
 393static int find_commit_in_graph(struct commit *item, struct commit_graph *g, uint32_t *pos)
 394{
 395        if (item->graph_pos != COMMIT_NOT_FROM_GRAPH) {
 396                *pos = item->graph_pos;
 397                return 1;
 398        } else {
 399                return bsearch_graph(g, &(item->object.oid), pos);
 400        }
 401}
 402
 403static int parse_commit_in_graph_one(struct commit_graph *g, struct commit *item)
 404{
 405        uint32_t pos;
 406
 407        if (item->object.parsed)
 408                return 1;
 409
 410        if (find_commit_in_graph(item, g, &pos))
 411                return fill_commit_in_graph(item, g, pos);
 412
 413        return 0;
 414}
 415
 416int parse_commit_in_graph(struct repository *r, struct commit *item)
 417{
 418        if (!prepare_commit_graph(r))
 419                return 0;
 420        return parse_commit_in_graph_one(r->objects->commit_graph, item);
 421}
 422
 423void load_commit_graph_info(struct repository *r, struct commit *item)
 424{
 425        uint32_t pos;
 426        if (!prepare_commit_graph(r))
 427                return;
 428        if (find_commit_in_graph(item, r->objects->commit_graph, &pos))
 429                fill_commit_graph_info(item, r->objects->commit_graph, pos);
 430}
 431
 432static struct tree *load_tree_for_commit(struct commit_graph *g, struct commit *c)
 433{
 434        struct object_id oid;
 435        const unsigned char *commit_data = g->chunk_commit_data +
 436                                           GRAPH_DATA_WIDTH * (c->graph_pos);
 437
 438        hashcpy(oid.hash, commit_data);
 439        c->maybe_tree = lookup_tree(the_repository, &oid);
 440
 441        return c->maybe_tree;
 442}
 443
 444static struct tree *get_commit_tree_in_graph_one(struct commit_graph *g,
 445                                                 const struct commit *c)
 446{
 447        if (c->maybe_tree)
 448                return c->maybe_tree;
 449        if (c->graph_pos == COMMIT_NOT_FROM_GRAPH)
 450                BUG("get_commit_tree_in_graph_one called from non-commit-graph commit");
 451
 452        return load_tree_for_commit(g, (struct commit *)c);
 453}
 454
 455struct tree *get_commit_tree_in_graph(struct repository *r, const struct commit *c)
 456{
 457        return get_commit_tree_in_graph_one(r->objects->commit_graph, c);
 458}
 459
 460static void write_graph_chunk_fanout(struct hashfile *f,
 461                                     struct commit **commits,
 462                                     int nr_commits)
 463{
 464        int i, count = 0;
 465        struct commit **list = commits;
 466
 467        /*
 468         * Write the first-level table (the list is sorted,
 469         * but we use a 256-entry lookup to be able to avoid
 470         * having to do eight extra binary search iterations).
 471         */
 472        for (i = 0; i < 256; i++) {
 473                while (count < nr_commits) {
 474                        if ((*list)->object.oid.hash[0] != i)
 475                                break;
 476                        count++;
 477                        list++;
 478                }
 479
 480                hashwrite_be32(f, count);
 481        }
 482}
 483
 484static void write_graph_chunk_oids(struct hashfile *f, int hash_len,
 485                                   struct commit **commits, int nr_commits)
 486{
 487        struct commit **list = commits;
 488        int count;
 489        for (count = 0; count < nr_commits; count++, list++)
 490                hashwrite(f, (*list)->object.oid.hash, (int)hash_len);
 491}
 492
 493static const unsigned char *commit_to_sha1(size_t index, void *table)
 494{
 495        struct commit **commits = table;
 496        return commits[index]->object.oid.hash;
 497}
 498
 499static void write_graph_chunk_data(struct hashfile *f, int hash_len,
 500                                   struct commit **commits, int nr_commits)
 501{
 502        struct commit **list = commits;
 503        struct commit **last = commits + nr_commits;
 504        uint32_t num_extra_edges = 0;
 505
 506        while (list < last) {
 507                struct commit_list *parent;
 508                int edge_value;
 509                uint32_t packedDate[2];
 510
 511                parse_commit(*list);
 512                hashwrite(f, get_commit_tree_oid(*list)->hash, hash_len);
 513
 514                parent = (*list)->parents;
 515
 516                if (!parent)
 517                        edge_value = GRAPH_PARENT_NONE;
 518                else {
 519                        edge_value = sha1_pos(parent->item->object.oid.hash,
 520                                              commits,
 521                                              nr_commits,
 522                                              commit_to_sha1);
 523
 524                        if (edge_value < 0)
 525                                edge_value = GRAPH_PARENT_MISSING;
 526                }
 527
 528                hashwrite_be32(f, edge_value);
 529
 530                if (parent)
 531                        parent = parent->next;
 532
 533                if (!parent)
 534                        edge_value = GRAPH_PARENT_NONE;
 535                else if (parent->next)
 536                        edge_value = GRAPH_OCTOPUS_EDGES_NEEDED | num_extra_edges;
 537                else {
 538                        edge_value = sha1_pos(parent->item->object.oid.hash,
 539                                              commits,
 540                                              nr_commits,
 541                                              commit_to_sha1);
 542                        if (edge_value < 0)
 543                                edge_value = GRAPH_PARENT_MISSING;
 544                }
 545
 546                hashwrite_be32(f, edge_value);
 547
 548                if (edge_value & GRAPH_OCTOPUS_EDGES_NEEDED) {
 549                        do {
 550                                num_extra_edges++;
 551                                parent = parent->next;
 552                        } while (parent);
 553                }
 554
 555                if (sizeof((*list)->date) > 4)
 556                        packedDate[0] = htonl(((*list)->date >> 32) & 0x3);
 557                else
 558                        packedDate[0] = 0;
 559
 560                packedDate[0] |= htonl((*list)->generation << 2);
 561
 562                packedDate[1] = htonl((*list)->date);
 563                hashwrite(f, packedDate, 8);
 564
 565                list++;
 566        }
 567}
 568
 569static void write_graph_chunk_large_edges(struct hashfile *f,
 570                                          struct commit **commits,
 571                                          int nr_commits)
 572{
 573        struct commit **list = commits;
 574        struct commit **last = commits + nr_commits;
 575        struct commit_list *parent;
 576
 577        while (list < last) {
 578                int num_parents = 0;
 579                for (parent = (*list)->parents; num_parents < 3 && parent;
 580                     parent = parent->next)
 581                        num_parents++;
 582
 583                if (num_parents <= 2) {
 584                        list++;
 585                        continue;
 586                }
 587
 588                /* Since num_parents > 2, this initializer is safe. */
 589                for (parent = (*list)->parents->next; parent; parent = parent->next) {
 590                        int edge_value = sha1_pos(parent->item->object.oid.hash,
 591                                                  commits,
 592                                                  nr_commits,
 593                                                  commit_to_sha1);
 594
 595                        if (edge_value < 0)
 596                                edge_value = GRAPH_PARENT_MISSING;
 597                        else if (!parent->next)
 598                                edge_value |= GRAPH_LAST_EDGE;
 599
 600                        hashwrite_be32(f, edge_value);
 601                }
 602
 603                list++;
 604        }
 605}
 606
 607static int commit_compare(const void *_a, const void *_b)
 608{
 609        const struct object_id *a = (const struct object_id *)_a;
 610        const struct object_id *b = (const struct object_id *)_b;
 611        return oidcmp(a, b);
 612}
 613
 614struct packed_commit_list {
 615        struct commit **list;
 616        int nr;
 617        int alloc;
 618};
 619
 620struct packed_oid_list {
 621        struct object_id *list;
 622        int nr;
 623        int alloc;
 624        struct progress *progress;
 625        int progress_done;
 626};
 627
 628static int add_packed_commits(const struct object_id *oid,
 629                              struct packed_git *pack,
 630                              uint32_t pos,
 631                              void *data)
 632{
 633        struct packed_oid_list *list = (struct packed_oid_list*)data;
 634        enum object_type type;
 635        off_t offset = nth_packed_object_offset(pack, pos);
 636        struct object_info oi = OBJECT_INFO_INIT;
 637
 638        if (list->progress)
 639                display_progress(list->progress, ++list->progress_done);
 640
 641        oi.typep = &type;
 642        if (packed_object_info(the_repository, pack, offset, &oi) < 0)
 643                die(_("unable to get type of object %s"), oid_to_hex(oid));
 644
 645        if (type != OBJ_COMMIT)
 646                return 0;
 647
 648        ALLOC_GROW(list->list, list->nr + 1, list->alloc);
 649        oidcpy(&(list->list[list->nr]), oid);
 650        list->nr++;
 651
 652        return 0;
 653}
 654
 655static void add_missing_parents(struct packed_oid_list *oids, struct commit *commit)
 656{
 657        struct commit_list *parent;
 658        for (parent = commit->parents; parent; parent = parent->next) {
 659                if (!(parent->item->object.flags & UNINTERESTING)) {
 660                        ALLOC_GROW(oids->list, oids->nr + 1, oids->alloc);
 661                        oidcpy(&oids->list[oids->nr], &(parent->item->object.oid));
 662                        oids->nr++;
 663                        parent->item->object.flags |= UNINTERESTING;
 664                }
 665        }
 666}
 667
 668static void close_reachable(struct packed_oid_list *oids, int report_progress)
 669{
 670        int i, j;
 671        struct commit *commit;
 672        struct progress *progress = NULL;
 673
 674        if (report_progress)
 675                progress = start_delayed_progress(
 676                        _("Loading known commits in commit graph"), j = 0);
 677        for (i = 0; i < oids->nr; i++) {
 678                display_progress(progress, ++j);
 679                commit = lookup_commit(the_repository, &oids->list[i]);
 680                if (commit)
 681                        commit->object.flags |= UNINTERESTING;
 682        }
 683        stop_progress(&progress);
 684
 685        /*
 686         * As this loop runs, oids->nr may grow, but not more
 687         * than the number of missing commits in the reachable
 688         * closure.
 689         */
 690        if (report_progress)
 691                progress = start_delayed_progress(
 692                        _("Expanding reachable commits in commit graph"), j = 0);
 693        for (i = 0; i < oids->nr; i++) {
 694                display_progress(progress, ++j);
 695                commit = lookup_commit(the_repository, &oids->list[i]);
 696
 697                if (commit && !parse_commit(commit))
 698                        add_missing_parents(oids, commit);
 699        }
 700        stop_progress(&progress);
 701
 702        if (report_progress)
 703                progress = start_delayed_progress(
 704                        _("Clearing commit marks in commit graph"), j = 0);
 705        for (i = 0; i < oids->nr; i++) {
 706                display_progress(progress, ++j);
 707                commit = lookup_commit(the_repository, &oids->list[i]);
 708
 709                if (commit)
 710                        commit->object.flags &= ~UNINTERESTING;
 711        }
 712        stop_progress(&progress);
 713}
 714
 715static void compute_generation_numbers(struct packed_commit_list* commits,
 716                                       int report_progress)
 717{
 718        int i;
 719        struct commit_list *list = NULL;
 720        struct progress *progress = NULL;
 721
 722        if (report_progress)
 723                progress = start_progress(
 724                        _("Computing commit graph generation numbers"),
 725                        commits->nr);
 726        for (i = 0; i < commits->nr; i++) {
 727                display_progress(progress, i + 1);
 728                if (commits->list[i]->generation != GENERATION_NUMBER_INFINITY &&
 729                    commits->list[i]->generation != GENERATION_NUMBER_ZERO)
 730                        continue;
 731
 732                commit_list_insert(commits->list[i], &list);
 733                while (list) {
 734                        struct commit *current = list->item;
 735                        struct commit_list *parent;
 736                        int all_parents_computed = 1;
 737                        uint32_t max_generation = 0;
 738
 739                        for (parent = current->parents; parent; parent = parent->next) {
 740                                if (parent->item->generation == GENERATION_NUMBER_INFINITY ||
 741                                    parent->item->generation == GENERATION_NUMBER_ZERO) {
 742                                        all_parents_computed = 0;
 743                                        commit_list_insert(parent->item, &list);
 744                                        break;
 745                                } else if (parent->item->generation > max_generation) {
 746                                        max_generation = parent->item->generation;
 747                                }
 748                        }
 749
 750                        if (all_parents_computed) {
 751                                current->generation = max_generation + 1;
 752                                pop_commit(&list);
 753
 754                                if (current->generation > GENERATION_NUMBER_MAX)
 755                                        current->generation = GENERATION_NUMBER_MAX;
 756                        }
 757                }
 758        }
 759        stop_progress(&progress);
 760}
 761
 762static int add_ref_to_list(const char *refname,
 763                           const struct object_id *oid,
 764                           int flags, void *cb_data)
 765{
 766        struct string_list *list = (struct string_list *)cb_data;
 767
 768        string_list_append(list, oid_to_hex(oid));
 769        return 0;
 770}
 771
 772void write_commit_graph_reachable(const char *obj_dir, int append,
 773                                  int report_progress)
 774{
 775        struct string_list list = STRING_LIST_INIT_DUP;
 776
 777        for_each_ref(add_ref_to_list, &list);
 778        write_commit_graph(obj_dir, NULL, &list, append, report_progress);
 779
 780        string_list_clear(&list, 0);
 781}
 782
 783void write_commit_graph(const char *obj_dir,
 784                        struct string_list *pack_indexes,
 785                        struct string_list *commit_hex,
 786                        int append, int report_progress)
 787{
 788        struct packed_oid_list oids;
 789        struct packed_commit_list commits;
 790        struct hashfile *f;
 791        uint32_t i, count_distinct = 0;
 792        char *graph_name;
 793        struct lock_file lk = LOCK_INIT;
 794        uint32_t chunk_ids[5];
 795        uint64_t chunk_offsets[5];
 796        int num_chunks;
 797        int num_extra_edges;
 798        struct commit_list *parent;
 799        struct progress *progress = NULL;
 800
 801        if (!commit_graph_compatible(the_repository))
 802                return;
 803
 804        oids.nr = 0;
 805        oids.alloc = approximate_object_count() / 32;
 806        oids.progress = NULL;
 807        oids.progress_done = 0;
 808
 809        if (append) {
 810                prepare_commit_graph_one(the_repository, obj_dir);
 811                if (the_repository->objects->commit_graph)
 812                        oids.alloc += the_repository->objects->commit_graph->num_commits;
 813        }
 814
 815        if (oids.alloc < 1024)
 816                oids.alloc = 1024;
 817        ALLOC_ARRAY(oids.list, oids.alloc);
 818
 819        if (append && the_repository->objects->commit_graph) {
 820                struct commit_graph *commit_graph =
 821                        the_repository->objects->commit_graph;
 822                for (i = 0; i < commit_graph->num_commits; i++) {
 823                        const unsigned char *hash = commit_graph->chunk_oid_lookup +
 824                                commit_graph->hash_len * i;
 825                        hashcpy(oids.list[oids.nr++].hash, hash);
 826                }
 827        }
 828
 829        if (pack_indexes) {
 830                struct strbuf packname = STRBUF_INIT;
 831                int dirlen;
 832                strbuf_addf(&packname, "%s/pack/", obj_dir);
 833                dirlen = packname.len;
 834                if (report_progress) {
 835                        oids.progress = start_delayed_progress(
 836                                _("Finding commits for commit graph"), 0);
 837                        oids.progress_done = 0;
 838                }
 839                for (i = 0; i < pack_indexes->nr; i++) {
 840                        struct packed_git *p;
 841                        strbuf_setlen(&packname, dirlen);
 842                        strbuf_addstr(&packname, pack_indexes->items[i].string);
 843                        p = add_packed_git(packname.buf, packname.len, 1);
 844                        if (!p)
 845                                die(_("error adding pack %s"), packname.buf);
 846                        if (open_pack_index(p))
 847                                die(_("error opening index for %s"), packname.buf);
 848                        for_each_object_in_pack(p, add_packed_commits, &oids, 0);
 849                        close_pack(p);
 850                        free(p);
 851                }
 852                stop_progress(&oids.progress);
 853                strbuf_release(&packname);
 854        }
 855
 856        if (commit_hex) {
 857                if (report_progress)
 858                        progress = start_delayed_progress(
 859                                _("Finding commits for commit graph"),
 860                                commit_hex->nr);
 861                for (i = 0; i < commit_hex->nr; i++) {
 862                        const char *end;
 863                        struct object_id oid;
 864                        struct commit *result;
 865
 866                        display_progress(progress, i + 1);
 867                        if (commit_hex->items[i].string &&
 868                            parse_oid_hex(commit_hex->items[i].string, &oid, &end))
 869                                continue;
 870
 871                        result = lookup_commit_reference_gently(the_repository, &oid, 1);
 872
 873                        if (result) {
 874                                ALLOC_GROW(oids.list, oids.nr + 1, oids.alloc);
 875                                oidcpy(&oids.list[oids.nr], &(result->object.oid));
 876                                oids.nr++;
 877                        }
 878                }
 879                stop_progress(&progress);
 880        }
 881
 882        if (!pack_indexes && !commit_hex) {
 883                if (report_progress)
 884                        oids.progress = start_delayed_progress(
 885                                _("Finding commits for commit graph"), 0);
 886                for_each_packed_object(add_packed_commits, &oids, 0);
 887                stop_progress(&oids.progress);
 888        }
 889
 890        close_reachable(&oids, report_progress);
 891
 892        QSORT(oids.list, oids.nr, commit_compare);
 893
 894        count_distinct = 1;
 895        for (i = 1; i < oids.nr; i++) {
 896                if (!oideq(&oids.list[i - 1], &oids.list[i]))
 897                        count_distinct++;
 898        }
 899
 900        if (count_distinct >= GRAPH_PARENT_MISSING)
 901                die(_("the commit graph format cannot write %d commits"), count_distinct);
 902
 903        commits.nr = 0;
 904        commits.alloc = count_distinct;
 905        ALLOC_ARRAY(commits.list, commits.alloc);
 906
 907        num_extra_edges = 0;
 908        for (i = 0; i < oids.nr; i++) {
 909                int num_parents = 0;
 910                if (i > 0 && oideq(&oids.list[i - 1], &oids.list[i]))
 911                        continue;
 912
 913                commits.list[commits.nr] = lookup_commit(the_repository, &oids.list[i]);
 914                parse_commit(commits.list[commits.nr]);
 915
 916                for (parent = commits.list[commits.nr]->parents;
 917                     parent; parent = parent->next)
 918                        num_parents++;
 919
 920                if (num_parents > 2)
 921                        num_extra_edges += num_parents - 1;
 922
 923                commits.nr++;
 924        }
 925        num_chunks = num_extra_edges ? 4 : 3;
 926
 927        if (commits.nr >= GRAPH_PARENT_MISSING)
 928                die(_("too many commits to write graph"));
 929
 930        compute_generation_numbers(&commits, report_progress);
 931
 932        graph_name = get_commit_graph_filename(obj_dir);
 933        if (safe_create_leading_directories(graph_name)) {
 934                UNLEAK(graph_name);
 935                die_errno(_("unable to create leading directories of %s"),
 936                          graph_name);
 937        }
 938
 939        hold_lock_file_for_update(&lk, graph_name, LOCK_DIE_ON_ERROR);
 940        f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
 941
 942        hashwrite_be32(f, GRAPH_SIGNATURE);
 943
 944        hashwrite_u8(f, GRAPH_VERSION);
 945        hashwrite_u8(f, GRAPH_OID_VERSION);
 946        hashwrite_u8(f, num_chunks);
 947        hashwrite_u8(f, 0); /* unused padding byte */
 948
 949        chunk_ids[0] = GRAPH_CHUNKID_OIDFANOUT;
 950        chunk_ids[1] = GRAPH_CHUNKID_OIDLOOKUP;
 951        chunk_ids[2] = GRAPH_CHUNKID_DATA;
 952        if (num_extra_edges)
 953                chunk_ids[3] = GRAPH_CHUNKID_LARGEEDGES;
 954        else
 955                chunk_ids[3] = 0;
 956        chunk_ids[4] = 0;
 957
 958        chunk_offsets[0] = 8 + (num_chunks + 1) * GRAPH_CHUNKLOOKUP_WIDTH;
 959        chunk_offsets[1] = chunk_offsets[0] + GRAPH_FANOUT_SIZE;
 960        chunk_offsets[2] = chunk_offsets[1] + GRAPH_OID_LEN * commits.nr;
 961        chunk_offsets[3] = chunk_offsets[2] + (GRAPH_OID_LEN + 16) * commits.nr;
 962        chunk_offsets[4] = chunk_offsets[3] + 4 * num_extra_edges;
 963
 964        for (i = 0; i <= num_chunks; i++) {
 965                uint32_t chunk_write[3];
 966
 967                chunk_write[0] = htonl(chunk_ids[i]);
 968                chunk_write[1] = htonl(chunk_offsets[i] >> 32);
 969                chunk_write[2] = htonl(chunk_offsets[i] & 0xffffffff);
 970                hashwrite(f, chunk_write, 12);
 971        }
 972
 973        write_graph_chunk_fanout(f, commits.list, commits.nr);
 974        write_graph_chunk_oids(f, GRAPH_OID_LEN, commits.list, commits.nr);
 975        write_graph_chunk_data(f, GRAPH_OID_LEN, commits.list, commits.nr);
 976        write_graph_chunk_large_edges(f, commits.list, commits.nr);
 977
 978        close_commit_graph(the_repository);
 979        finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_FSYNC);
 980        commit_lock_file(&lk);
 981
 982        free(graph_name);
 983        free(commits.list);
 984        free(oids.list);
 985}
 986
 987#define VERIFY_COMMIT_GRAPH_ERROR_HASH 2
 988static int verify_commit_graph_error;
 989
 990static void graph_report(const char *fmt, ...)
 991{
 992        va_list ap;
 993
 994        verify_commit_graph_error = 1;
 995        va_start(ap, fmt);
 996        vfprintf(stderr, fmt, ap);
 997        fprintf(stderr, "\n");
 998        va_end(ap);
 999}
1000
1001#define GENERATION_ZERO_EXISTS 1
1002#define GENERATION_NUMBER_EXISTS 2
1003
1004int verify_commit_graph(struct repository *r, struct commit_graph *g)
1005{
1006        uint32_t i, cur_fanout_pos = 0;
1007        struct object_id prev_oid, cur_oid, checksum;
1008        int generation_zero = 0;
1009        struct hashfile *f;
1010        int devnull;
1011        struct progress *progress = NULL;
1012
1013        if (!g) {
1014                graph_report("no commit-graph file loaded");
1015                return 1;
1016        }
1017
1018        verify_commit_graph_error = 0;
1019
1020        if (!g->chunk_oid_fanout)
1021                graph_report("commit-graph is missing the OID Fanout chunk");
1022        if (!g->chunk_oid_lookup)
1023                graph_report("commit-graph is missing the OID Lookup chunk");
1024        if (!g->chunk_commit_data)
1025                graph_report("commit-graph is missing the Commit Data chunk");
1026
1027        if (verify_commit_graph_error)
1028                return verify_commit_graph_error;
1029
1030        devnull = open("/dev/null", O_WRONLY);
1031        f = hashfd(devnull, NULL);
1032        hashwrite(f, g->data, g->data_len - g->hash_len);
1033        finalize_hashfile(f, checksum.hash, CSUM_CLOSE);
1034        if (!hasheq(checksum.hash, g->data + g->data_len - g->hash_len)) {
1035                graph_report(_("the commit-graph file has incorrect checksum and is likely corrupt"));
1036                verify_commit_graph_error = VERIFY_COMMIT_GRAPH_ERROR_HASH;
1037        }
1038
1039        for (i = 0; i < g->num_commits; i++) {
1040                struct commit *graph_commit;
1041
1042                hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
1043
1044                if (i && oidcmp(&prev_oid, &cur_oid) >= 0)
1045                        graph_report("commit-graph has incorrect OID order: %s then %s",
1046                                     oid_to_hex(&prev_oid),
1047                                     oid_to_hex(&cur_oid));
1048
1049                oidcpy(&prev_oid, &cur_oid);
1050
1051                while (cur_oid.hash[0] > cur_fanout_pos) {
1052                        uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
1053
1054                        if (i != fanout_value)
1055                                graph_report("commit-graph has incorrect fanout value: fanout[%d] = %u != %u",
1056                                             cur_fanout_pos, fanout_value, i);
1057                        cur_fanout_pos++;
1058                }
1059
1060                graph_commit = lookup_commit(r, &cur_oid);
1061                if (!parse_commit_in_graph_one(g, graph_commit))
1062                        graph_report("failed to parse %s from commit-graph",
1063                                     oid_to_hex(&cur_oid));
1064        }
1065
1066        while (cur_fanout_pos < 256) {
1067                uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
1068
1069                if (g->num_commits != fanout_value)
1070                        graph_report("commit-graph has incorrect fanout value: fanout[%d] = %u != %u",
1071                                     cur_fanout_pos, fanout_value, i);
1072
1073                cur_fanout_pos++;
1074        }
1075
1076        if (verify_commit_graph_error & ~VERIFY_COMMIT_GRAPH_ERROR_HASH)
1077                return verify_commit_graph_error;
1078
1079        progress = start_progress(_("Verifying commits in commit graph"),
1080                                  g->num_commits);
1081        for (i = 0; i < g->num_commits; i++) {
1082                struct commit *graph_commit, *odb_commit;
1083                struct commit_list *graph_parents, *odb_parents;
1084                uint32_t max_generation = 0;
1085
1086                display_progress(progress, i + 1);
1087                hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
1088
1089                graph_commit = lookup_commit(r, &cur_oid);
1090                odb_commit = (struct commit *)create_object(r, cur_oid.hash, alloc_commit_node(r));
1091                if (parse_commit_internal(odb_commit, 0, 0)) {
1092                        graph_report("failed to parse %s from object database",
1093                                     oid_to_hex(&cur_oid));
1094                        continue;
1095                }
1096
1097                if (!oideq(&get_commit_tree_in_graph_one(g, graph_commit)->object.oid,
1098                           get_commit_tree_oid(odb_commit)))
1099                        graph_report("root tree OID for commit %s in commit-graph is %s != %s",
1100                                     oid_to_hex(&cur_oid),
1101                                     oid_to_hex(get_commit_tree_oid(graph_commit)),
1102                                     oid_to_hex(get_commit_tree_oid(odb_commit)));
1103
1104                graph_parents = graph_commit->parents;
1105                odb_parents = odb_commit->parents;
1106
1107                while (graph_parents) {
1108                        if (odb_parents == NULL) {
1109                                graph_report("commit-graph parent list for commit %s is too long",
1110                                             oid_to_hex(&cur_oid));
1111                                break;
1112                        }
1113
1114                        if (!oideq(&graph_parents->item->object.oid, &odb_parents->item->object.oid))
1115                                graph_report("commit-graph parent for %s is %s != %s",
1116                                             oid_to_hex(&cur_oid),
1117                                             oid_to_hex(&graph_parents->item->object.oid),
1118                                             oid_to_hex(&odb_parents->item->object.oid));
1119
1120                        if (graph_parents->item->generation > max_generation)
1121                                max_generation = graph_parents->item->generation;
1122
1123                        graph_parents = graph_parents->next;
1124                        odb_parents = odb_parents->next;
1125                }
1126
1127                if (odb_parents != NULL)
1128                        graph_report("commit-graph parent list for commit %s terminates early",
1129                                     oid_to_hex(&cur_oid));
1130
1131                if (!graph_commit->generation) {
1132                        if (generation_zero == GENERATION_NUMBER_EXISTS)
1133                                graph_report("commit-graph has generation number zero for commit %s, but non-zero elsewhere",
1134                                             oid_to_hex(&cur_oid));
1135                        generation_zero = GENERATION_ZERO_EXISTS;
1136                } else if (generation_zero == GENERATION_ZERO_EXISTS)
1137                        graph_report("commit-graph has non-zero generation number for commit %s, but zero elsewhere",
1138                                     oid_to_hex(&cur_oid));
1139
1140                if (generation_zero == GENERATION_ZERO_EXISTS)
1141                        continue;
1142
1143                /*
1144                 * If one of our parents has generation GENERATION_NUMBER_MAX, then
1145                 * our generation is also GENERATION_NUMBER_MAX. Decrement to avoid
1146                 * extra logic in the following condition.
1147                 */
1148                if (max_generation == GENERATION_NUMBER_MAX)
1149                        max_generation--;
1150
1151                if (graph_commit->generation != max_generation + 1)
1152                        graph_report("commit-graph generation for commit %s is %u != %u",
1153                                     oid_to_hex(&cur_oid),
1154                                     graph_commit->generation,
1155                                     max_generation + 1);
1156
1157                if (graph_commit->date != odb_commit->date)
1158                        graph_report("commit date for commit %s in commit-graph is %"PRItime" != %"PRItime,
1159                                     oid_to_hex(&cur_oid),
1160                                     graph_commit->date,
1161                                     odb_commit->date);
1162        }
1163        stop_progress(&progress);
1164
1165        return verify_commit_graph_error;
1166}
1167
1168void free_commit_graph(struct commit_graph *g)
1169{
1170        if (!g)
1171                return;
1172        if (g->graph_fd >= 0) {
1173                munmap((void *)g->data, g->data_len);
1174                g->data = NULL;
1175                close(g->graph_fd);
1176        }
1177        free(g);
1178}