commit-graph.con commit commit-graph: clean up leaked memory during write (f4dbdfc)
   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 "progress.h"
  17
  18#define GRAPH_SIGNATURE 0x43475048 /* "CGPH" */
  19#define GRAPH_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
  20#define GRAPH_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
  21#define GRAPH_CHUNKID_DATA 0x43444154 /* "CDAT" */
  22#define GRAPH_CHUNKID_LARGEEDGES 0x45444745 /* "EDGE" */
  23
  24#define GRAPH_DATA_WIDTH 36
  25
  26#define GRAPH_VERSION_1 0x1
  27#define GRAPH_VERSION GRAPH_VERSION_1
  28
  29#define GRAPH_OID_VERSION_SHA1 1
  30#define GRAPH_OID_LEN_SHA1 GIT_SHA1_RAWSZ
  31#define GRAPH_OID_VERSION GRAPH_OID_VERSION_SHA1
  32#define GRAPH_OID_LEN GRAPH_OID_LEN_SHA1
  33
  34#define GRAPH_OCTOPUS_EDGES_NEEDED 0x80000000
  35#define GRAPH_PARENT_MISSING 0x7fffffff
  36#define GRAPH_EDGE_LAST_MASK 0x7fffffff
  37#define GRAPH_PARENT_NONE 0x70000000
  38
  39#define GRAPH_LAST_EDGE 0x80000000
  40
  41#define GRAPH_HEADER_SIZE 8
  42#define GRAPH_FANOUT_SIZE (4 * 256)
  43#define GRAPH_CHUNKLOOKUP_WIDTH 12
  44#define GRAPH_MIN_SIZE (GRAPH_HEADER_SIZE + 4 * GRAPH_CHUNKLOOKUP_WIDTH \
  45                        + GRAPH_FANOUT_SIZE + GRAPH_OID_LEN)
  46
  47char *get_commit_graph_filename(const char *obj_dir)
  48{
  49        return xstrfmt("%s/info/commit-graph", obj_dir);
  50}
  51
  52static struct commit_graph *alloc_commit_graph(void)
  53{
  54        struct commit_graph *g = xcalloc(1, sizeof(*g));
  55        g->graph_fd = -1;
  56
  57        return g;
  58}
  59
  60struct commit_graph *load_commit_graph_one(const char *graph_file)
  61{
  62        void *graph_map;
  63        const unsigned char *data, *chunk_lookup;
  64        size_t graph_size;
  65        struct stat st;
  66        uint32_t i;
  67        struct commit_graph *graph;
  68        int fd = git_open(graph_file);
  69        uint64_t last_chunk_offset;
  70        uint32_t last_chunk_id;
  71        uint32_t graph_signature;
  72        unsigned char graph_version, hash_version;
  73
  74        if (fd < 0)
  75                return NULL;
  76        if (fstat(fd, &st)) {
  77                close(fd);
  78                return NULL;
  79        }
  80        graph_size = xsize_t(st.st_size);
  81
  82        if (graph_size < GRAPH_MIN_SIZE) {
  83                close(fd);
  84                die(_("graph file %s is too small"), graph_file);
  85        }
  86        graph_map = xmmap(NULL, graph_size, PROT_READ, MAP_PRIVATE, fd, 0);
  87        data = (const unsigned char *)graph_map;
  88
  89        graph_signature = get_be32(data);
  90        if (graph_signature != GRAPH_SIGNATURE) {
  91                error(_("graph signature %X does not match signature %X"),
  92                      graph_signature, GRAPH_SIGNATURE);
  93                goto cleanup_fail;
  94        }
  95
  96        graph_version = *(unsigned char*)(data + 4);
  97        if (graph_version != GRAPH_VERSION) {
  98                error(_("graph version %X does not match version %X"),
  99                      graph_version, GRAPH_VERSION);
 100                goto cleanup_fail;
 101        }
 102
 103        hash_version = *(unsigned char*)(data + 5);
 104        if (hash_version != GRAPH_OID_VERSION) {
 105                error(_("hash version %X does not match version %X"),
 106                      hash_version, GRAPH_OID_VERSION);
 107                goto cleanup_fail;
 108        }
 109
 110        graph = alloc_commit_graph();
 111
 112        graph->hash_len = GRAPH_OID_LEN;
 113        graph->num_chunks = *(unsigned char*)(data + 6);
 114        graph->graph_fd = fd;
 115        graph->data = graph_map;
 116        graph->data_len = graph_size;
 117
 118        last_chunk_id = 0;
 119        last_chunk_offset = 8;
 120        chunk_lookup = data + 8;
 121        for (i = 0; i < graph->num_chunks; i++) {
 122                uint32_t chunk_id = get_be32(chunk_lookup + 0);
 123                uint64_t chunk_offset = get_be64(chunk_lookup + 4);
 124                int chunk_repeated = 0;
 125
 126                chunk_lookup += GRAPH_CHUNKLOOKUP_WIDTH;
 127
 128                if (chunk_offset > graph_size - GIT_MAX_RAWSZ) {
 129                        error(_("improper chunk offset %08x%08x"), (uint32_t)(chunk_offset >> 32),
 130                              (uint32_t)chunk_offset);
 131                        goto cleanup_fail;
 132                }
 133
 134                switch (chunk_id) {
 135                case GRAPH_CHUNKID_OIDFANOUT:
 136                        if (graph->chunk_oid_fanout)
 137                                chunk_repeated = 1;
 138                        else
 139                                graph->chunk_oid_fanout = (uint32_t*)(data + chunk_offset);
 140                        break;
 141
 142                case GRAPH_CHUNKID_OIDLOOKUP:
 143                        if (graph->chunk_oid_lookup)
 144                                chunk_repeated = 1;
 145                        else
 146                                graph->chunk_oid_lookup = data + chunk_offset;
 147                        break;
 148
 149                case GRAPH_CHUNKID_DATA:
 150                        if (graph->chunk_commit_data)
 151                                chunk_repeated = 1;
 152                        else
 153                                graph->chunk_commit_data = data + chunk_offset;
 154                        break;
 155
 156                case GRAPH_CHUNKID_LARGEEDGES:
 157                        if (graph->chunk_large_edges)
 158                                chunk_repeated = 1;
 159                        else
 160                                graph->chunk_large_edges = data + chunk_offset;
 161                        break;
 162                }
 163
 164                if (chunk_repeated) {
 165                        error(_("chunk id %08x appears multiple times"), chunk_id);
 166                        goto cleanup_fail;
 167                }
 168
 169                if (last_chunk_id == GRAPH_CHUNKID_OIDLOOKUP)
 170                {
 171                        graph->num_commits = (chunk_offset - last_chunk_offset)
 172                                             / graph->hash_len;
 173                }
 174
 175                last_chunk_id = chunk_id;
 176                last_chunk_offset = chunk_offset;
 177        }
 178
 179        return graph;
 180
 181cleanup_fail:
 182        munmap(graph_map, graph_size);
 183        close(fd);
 184        exit(1);
 185}
 186
 187static void prepare_commit_graph_one(struct repository *r, const char *obj_dir)
 188{
 189        char *graph_name;
 190
 191        if (r->objects->commit_graph)
 192                return;
 193
 194        graph_name = get_commit_graph_filename(obj_dir);
 195        r->objects->commit_graph =
 196                load_commit_graph_one(graph_name);
 197
 198        FREE_AND_NULL(graph_name);
 199}
 200
 201/*
 202 * Return 1 if commit_graph is non-NULL, and 0 otherwise.
 203 *
 204 * On the first invocation, this function attemps to load the commit
 205 * graph if the_repository is configured to have one.
 206 */
 207static int prepare_commit_graph(struct repository *r)
 208{
 209        struct alternate_object_database *alt;
 210        char *obj_dir;
 211        int config_value;
 212
 213        if (r->objects->commit_graph_attempted)
 214                return !!r->objects->commit_graph;
 215        r->objects->commit_graph_attempted = 1;
 216
 217        if (repo_config_get_bool(r, "core.commitgraph", &config_value) ||
 218            !config_value)
 219                /*
 220                 * This repository is not configured to use commit graphs, so
 221                 * do not load one. (But report commit_graph_attempted anyway
 222                 * so that commit graph loading is not attempted again for this
 223                 * repository.)
 224                 */
 225                return 0;
 226
 227        obj_dir = r->objects->objectdir;
 228        prepare_commit_graph_one(r, obj_dir);
 229        prepare_alt_odb(r);
 230        for (alt = r->objects->alt_odb_list;
 231             !r->objects->commit_graph && alt;
 232             alt = alt->next)
 233                prepare_commit_graph_one(r, alt->path);
 234        return !!r->objects->commit_graph;
 235}
 236
 237static void close_commit_graph(void)
 238{
 239        free_commit_graph(the_repository->objects->commit_graph);
 240        the_repository->objects->commit_graph = NULL;
 241}
 242
 243static int bsearch_graph(struct commit_graph *g, struct object_id *oid, uint32_t *pos)
 244{
 245        return bsearch_hash(oid->hash, g->chunk_oid_fanout,
 246                            g->chunk_oid_lookup, g->hash_len, pos);
 247}
 248
 249static struct commit_list **insert_parent_or_die(struct commit_graph *g,
 250                                                 uint64_t pos,
 251                                                 struct commit_list **pptr)
 252{
 253        struct commit *c;
 254        struct object_id oid;
 255
 256        if (pos >= g->num_commits)
 257                die("invalid parent position %"PRIu64, pos);
 258
 259        hashcpy(oid.hash, g->chunk_oid_lookup + g->hash_len * pos);
 260        c = lookup_commit(the_repository, &oid);
 261        if (!c)
 262                die(_("could not find commit %s"), oid_to_hex(&oid));
 263        c->graph_pos = pos;
 264        return &commit_list_insert(c, pptr)->next;
 265}
 266
 267static void fill_commit_graph_info(struct commit *item, struct commit_graph *g, uint32_t pos)
 268{
 269        const unsigned char *commit_data = g->chunk_commit_data + GRAPH_DATA_WIDTH * pos;
 270        item->graph_pos = pos;
 271        item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
 272}
 273
 274static int fill_commit_in_graph(struct commit *item, struct commit_graph *g, uint32_t pos)
 275{
 276        uint32_t edge_value;
 277        uint32_t *parent_data_ptr;
 278        uint64_t date_low, date_high;
 279        struct commit_list **pptr;
 280        const unsigned char *commit_data = g->chunk_commit_data + (g->hash_len + 16) * pos;
 281
 282        item->object.parsed = 1;
 283        item->graph_pos = pos;
 284
 285        item->maybe_tree = NULL;
 286
 287        date_high = get_be32(commit_data + g->hash_len + 8) & 0x3;
 288        date_low = get_be32(commit_data + g->hash_len + 12);
 289        item->date = (timestamp_t)((date_high << 32) | date_low);
 290
 291        item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
 292
 293        pptr = &item->parents;
 294
 295        edge_value = get_be32(commit_data + g->hash_len);
 296        if (edge_value == GRAPH_PARENT_NONE)
 297                return 1;
 298        pptr = insert_parent_or_die(g, edge_value, pptr);
 299
 300        edge_value = get_be32(commit_data + g->hash_len + 4);
 301        if (edge_value == GRAPH_PARENT_NONE)
 302                return 1;
 303        if (!(edge_value & GRAPH_OCTOPUS_EDGES_NEEDED)) {
 304                pptr = insert_parent_or_die(g, edge_value, pptr);
 305                return 1;
 306        }
 307
 308        parent_data_ptr = (uint32_t*)(g->chunk_large_edges +
 309                          4 * (uint64_t)(edge_value & GRAPH_EDGE_LAST_MASK));
 310        do {
 311                edge_value = get_be32(parent_data_ptr);
 312                pptr = insert_parent_or_die(g,
 313                                            edge_value & GRAPH_EDGE_LAST_MASK,
 314                                            pptr);
 315                parent_data_ptr++;
 316        } while (!(edge_value & GRAPH_LAST_EDGE));
 317
 318        return 1;
 319}
 320
 321static int find_commit_in_graph(struct commit *item, struct commit_graph *g, uint32_t *pos)
 322{
 323        if (item->graph_pos != COMMIT_NOT_FROM_GRAPH) {
 324                *pos = item->graph_pos;
 325                return 1;
 326        } else {
 327                return bsearch_graph(g, &(item->object.oid), pos);
 328        }
 329}
 330
 331static int parse_commit_in_graph_one(struct commit_graph *g, struct commit *item)
 332{
 333        uint32_t pos;
 334
 335        if (item->object.parsed)
 336                return 1;
 337
 338        if (find_commit_in_graph(item, g, &pos))
 339                return fill_commit_in_graph(item, g, pos);
 340
 341        return 0;
 342}
 343
 344int parse_commit_in_graph(struct repository *r, struct commit *item)
 345{
 346        if (!prepare_commit_graph(r))
 347                return 0;
 348        return parse_commit_in_graph_one(r->objects->commit_graph, item);
 349}
 350
 351void load_commit_graph_info(struct repository *r, struct commit *item)
 352{
 353        uint32_t pos;
 354        if (!prepare_commit_graph(r))
 355                return;
 356        if (find_commit_in_graph(item, r->objects->commit_graph, &pos))
 357                fill_commit_graph_info(item, r->objects->commit_graph, pos);
 358}
 359
 360static struct tree *load_tree_for_commit(struct commit_graph *g, struct commit *c)
 361{
 362        struct object_id oid;
 363        const unsigned char *commit_data = g->chunk_commit_data +
 364                                           GRAPH_DATA_WIDTH * (c->graph_pos);
 365
 366        hashcpy(oid.hash, commit_data);
 367        c->maybe_tree = lookup_tree(the_repository, &oid);
 368
 369        return c->maybe_tree;
 370}
 371
 372static struct tree *get_commit_tree_in_graph_one(struct commit_graph *g,
 373                                                 const struct commit *c)
 374{
 375        if (c->maybe_tree)
 376                return c->maybe_tree;
 377        if (c->graph_pos == COMMIT_NOT_FROM_GRAPH)
 378                BUG("get_commit_tree_in_graph_one called from non-commit-graph commit");
 379
 380        return load_tree_for_commit(g, (struct commit *)c);
 381}
 382
 383struct tree *get_commit_tree_in_graph(struct repository *r, const struct commit *c)
 384{
 385        return get_commit_tree_in_graph_one(r->objects->commit_graph, c);
 386}
 387
 388static void write_graph_chunk_fanout(struct hashfile *f,
 389                                     struct commit **commits,
 390                                     int nr_commits)
 391{
 392        int i, count = 0;
 393        struct commit **list = commits;
 394
 395        /*
 396         * Write the first-level table (the list is sorted,
 397         * but we use a 256-entry lookup to be able to avoid
 398         * having to do eight extra binary search iterations).
 399         */
 400        for (i = 0; i < 256; i++) {
 401                while (count < nr_commits) {
 402                        if ((*list)->object.oid.hash[0] != i)
 403                                break;
 404                        count++;
 405                        list++;
 406                }
 407
 408                hashwrite_be32(f, count);
 409        }
 410}
 411
 412static void write_graph_chunk_oids(struct hashfile *f, int hash_len,
 413                                   struct commit **commits, int nr_commits)
 414{
 415        struct commit **list = commits;
 416        int count;
 417        for (count = 0; count < nr_commits; count++, list++)
 418                hashwrite(f, (*list)->object.oid.hash, (int)hash_len);
 419}
 420
 421static const unsigned char *commit_to_sha1(size_t index, void *table)
 422{
 423        struct commit **commits = table;
 424        return commits[index]->object.oid.hash;
 425}
 426
 427static void write_graph_chunk_data(struct hashfile *f, int hash_len,
 428                                   struct commit **commits, int nr_commits)
 429{
 430        struct commit **list = commits;
 431        struct commit **last = commits + nr_commits;
 432        uint32_t num_extra_edges = 0;
 433
 434        while (list < last) {
 435                struct commit_list *parent;
 436                int edge_value;
 437                uint32_t packedDate[2];
 438
 439                parse_commit(*list);
 440                hashwrite(f, get_commit_tree_oid(*list)->hash, hash_len);
 441
 442                parent = (*list)->parents;
 443
 444                if (!parent)
 445                        edge_value = GRAPH_PARENT_NONE;
 446                else {
 447                        edge_value = sha1_pos(parent->item->object.oid.hash,
 448                                              commits,
 449                                              nr_commits,
 450                                              commit_to_sha1);
 451
 452                        if (edge_value < 0)
 453                                edge_value = GRAPH_PARENT_MISSING;
 454                }
 455
 456                hashwrite_be32(f, edge_value);
 457
 458                if (parent)
 459                        parent = parent->next;
 460
 461                if (!parent)
 462                        edge_value = GRAPH_PARENT_NONE;
 463                else if (parent->next)
 464                        edge_value = GRAPH_OCTOPUS_EDGES_NEEDED | num_extra_edges;
 465                else {
 466                        edge_value = sha1_pos(parent->item->object.oid.hash,
 467                                              commits,
 468                                              nr_commits,
 469                                              commit_to_sha1);
 470                        if (edge_value < 0)
 471                                edge_value = GRAPH_PARENT_MISSING;
 472                }
 473
 474                hashwrite_be32(f, edge_value);
 475
 476                if (edge_value & GRAPH_OCTOPUS_EDGES_NEEDED) {
 477                        do {
 478                                num_extra_edges++;
 479                                parent = parent->next;
 480                        } while (parent);
 481                }
 482
 483                if (sizeof((*list)->date) > 4)
 484                        packedDate[0] = htonl(((*list)->date >> 32) & 0x3);
 485                else
 486                        packedDate[0] = 0;
 487
 488                packedDate[0] |= htonl((*list)->generation << 2);
 489
 490                packedDate[1] = htonl((*list)->date);
 491                hashwrite(f, packedDate, 8);
 492
 493                list++;
 494        }
 495}
 496
 497static void write_graph_chunk_large_edges(struct hashfile *f,
 498                                          struct commit **commits,
 499                                          int nr_commits)
 500{
 501        struct commit **list = commits;
 502        struct commit **last = commits + nr_commits;
 503        struct commit_list *parent;
 504
 505        while (list < last) {
 506                int num_parents = 0;
 507                for (parent = (*list)->parents; num_parents < 3 && parent;
 508                     parent = parent->next)
 509                        num_parents++;
 510
 511                if (num_parents <= 2) {
 512                        list++;
 513                        continue;
 514                }
 515
 516                /* Since num_parents > 2, this initializer is safe. */
 517                for (parent = (*list)->parents->next; parent; parent = parent->next) {
 518                        int edge_value = sha1_pos(parent->item->object.oid.hash,
 519                                                  commits,
 520                                                  nr_commits,
 521                                                  commit_to_sha1);
 522
 523                        if (edge_value < 0)
 524                                edge_value = GRAPH_PARENT_MISSING;
 525                        else if (!parent->next)
 526                                edge_value |= GRAPH_LAST_EDGE;
 527
 528                        hashwrite_be32(f, edge_value);
 529                }
 530
 531                list++;
 532        }
 533}
 534
 535static int commit_compare(const void *_a, const void *_b)
 536{
 537        const struct object_id *a = (const struct object_id *)_a;
 538        const struct object_id *b = (const struct object_id *)_b;
 539        return oidcmp(a, b);
 540}
 541
 542struct packed_commit_list {
 543        struct commit **list;
 544        int nr;
 545        int alloc;
 546};
 547
 548struct packed_oid_list {
 549        struct object_id *list;
 550        int nr;
 551        int alloc;
 552        struct progress *progress;
 553        int progress_done;
 554};
 555
 556static int add_packed_commits(const struct object_id *oid,
 557                              struct packed_git *pack,
 558                              uint32_t pos,
 559                              void *data)
 560{
 561        struct packed_oid_list *list = (struct packed_oid_list*)data;
 562        enum object_type type;
 563        off_t offset = nth_packed_object_offset(pack, pos);
 564        struct object_info oi = OBJECT_INFO_INIT;
 565
 566        if (list->progress)
 567                display_progress(list->progress, ++list->progress_done);
 568
 569        oi.typep = &type;
 570        if (packed_object_info(the_repository, pack, offset, &oi) < 0)
 571                die(_("unable to get type of object %s"), oid_to_hex(oid));
 572
 573        if (type != OBJ_COMMIT)
 574                return 0;
 575
 576        ALLOC_GROW(list->list, list->nr + 1, list->alloc);
 577        oidcpy(&(list->list[list->nr]), oid);
 578        list->nr++;
 579
 580        return 0;
 581}
 582
 583static void add_missing_parents(struct packed_oid_list *oids, struct commit *commit)
 584{
 585        struct commit_list *parent;
 586        for (parent = commit->parents; parent; parent = parent->next) {
 587                if (!(parent->item->object.flags & UNINTERESTING)) {
 588                        ALLOC_GROW(oids->list, oids->nr + 1, oids->alloc);
 589                        oidcpy(&oids->list[oids->nr], &(parent->item->object.oid));
 590                        oids->nr++;
 591                        parent->item->object.flags |= UNINTERESTING;
 592                }
 593        }
 594}
 595
 596static void close_reachable(struct packed_oid_list *oids, int report_progress)
 597{
 598        int i;
 599        struct commit *commit;
 600        struct progress *progress = NULL;
 601        int j = 0;
 602
 603        if (report_progress)
 604                progress = start_delayed_progress(
 605                        _("Annotating commits in commit graph"), 0);
 606        for (i = 0; i < oids->nr; i++) {
 607                display_progress(progress, ++j);
 608                commit = lookup_commit(the_repository, &oids->list[i]);
 609                if (commit)
 610                        commit->object.flags |= UNINTERESTING;
 611        }
 612
 613        /*
 614         * As this loop runs, oids->nr may grow, but not more
 615         * than the number of missing commits in the reachable
 616         * closure.
 617         */
 618        for (i = 0; i < oids->nr; i++) {
 619                display_progress(progress, ++j);
 620                commit = lookup_commit(the_repository, &oids->list[i]);
 621
 622                if (commit && !parse_commit(commit))
 623                        add_missing_parents(oids, commit);
 624        }
 625
 626        for (i = 0; i < oids->nr; i++) {
 627                display_progress(progress, ++j);
 628                commit = lookup_commit(the_repository, &oids->list[i]);
 629
 630                if (commit)
 631                        commit->object.flags &= ~UNINTERESTING;
 632        }
 633        stop_progress(&progress);
 634}
 635
 636static void compute_generation_numbers(struct packed_commit_list* commits,
 637                                       int report_progress)
 638{
 639        int i;
 640        struct commit_list *list = NULL;
 641        struct progress *progress = NULL;
 642
 643        if (report_progress)
 644                progress = start_progress(
 645                        _("Computing commit graph generation numbers"),
 646                        commits->nr);
 647        for (i = 0; i < commits->nr; i++) {
 648                display_progress(progress, i + 1);
 649                if (commits->list[i]->generation != GENERATION_NUMBER_INFINITY &&
 650                    commits->list[i]->generation != GENERATION_NUMBER_ZERO)
 651                        continue;
 652
 653                commit_list_insert(commits->list[i], &list);
 654                while (list) {
 655                        struct commit *current = list->item;
 656                        struct commit_list *parent;
 657                        int all_parents_computed = 1;
 658                        uint32_t max_generation = 0;
 659
 660                        for (parent = current->parents; parent; parent = parent->next) {
 661                                if (parent->item->generation == GENERATION_NUMBER_INFINITY ||
 662                                    parent->item->generation == GENERATION_NUMBER_ZERO) {
 663                                        all_parents_computed = 0;
 664                                        commit_list_insert(parent->item, &list);
 665                                        break;
 666                                } else if (parent->item->generation > max_generation) {
 667                                        max_generation = parent->item->generation;
 668                                }
 669                        }
 670
 671                        if (all_parents_computed) {
 672                                current->generation = max_generation + 1;
 673                                pop_commit(&list);
 674
 675                                if (current->generation > GENERATION_NUMBER_MAX)
 676                                        current->generation = GENERATION_NUMBER_MAX;
 677                        }
 678                }
 679        }
 680        stop_progress(&progress);
 681}
 682
 683static int add_ref_to_list(const char *refname,
 684                           const struct object_id *oid,
 685                           int flags, void *cb_data)
 686{
 687        struct string_list *list = (struct string_list *)cb_data;
 688
 689        string_list_append(list, oid_to_hex(oid));
 690        return 0;
 691}
 692
 693void write_commit_graph_reachable(const char *obj_dir, int append,
 694                                  int report_progress)
 695{
 696        struct string_list list = STRING_LIST_INIT_DUP;
 697
 698        for_each_ref(add_ref_to_list, &list);
 699        write_commit_graph(obj_dir, NULL, &list, append, report_progress);
 700
 701        string_list_clear(&list, 0);
 702}
 703
 704void write_commit_graph(const char *obj_dir,
 705                        struct string_list *pack_indexes,
 706                        struct string_list *commit_hex,
 707                        int append, int report_progress)
 708{
 709        struct packed_oid_list oids;
 710        struct packed_commit_list commits;
 711        struct hashfile *f;
 712        uint32_t i, count_distinct = 0;
 713        char *graph_name;
 714        struct lock_file lk = LOCK_INIT;
 715        uint32_t chunk_ids[5];
 716        uint64_t chunk_offsets[5];
 717        int num_chunks;
 718        int num_extra_edges;
 719        struct commit_list *parent;
 720        struct progress *progress = NULL;
 721
 722        oids.nr = 0;
 723        oids.alloc = approximate_object_count() / 4;
 724        oids.progress = NULL;
 725        oids.progress_done = 0;
 726
 727        if (append) {
 728                prepare_commit_graph_one(the_repository, obj_dir);
 729                if (the_repository->objects->commit_graph)
 730                        oids.alloc += the_repository->objects->commit_graph->num_commits;
 731        }
 732
 733        if (oids.alloc < 1024)
 734                oids.alloc = 1024;
 735        ALLOC_ARRAY(oids.list, oids.alloc);
 736
 737        if (append && the_repository->objects->commit_graph) {
 738                struct commit_graph *commit_graph =
 739                        the_repository->objects->commit_graph;
 740                for (i = 0; i < commit_graph->num_commits; i++) {
 741                        const unsigned char *hash = commit_graph->chunk_oid_lookup +
 742                                commit_graph->hash_len * i;
 743                        hashcpy(oids.list[oids.nr++].hash, hash);
 744                }
 745        }
 746
 747        if (pack_indexes) {
 748                struct strbuf packname = STRBUF_INIT;
 749                int dirlen;
 750                strbuf_addf(&packname, "%s/pack/", obj_dir);
 751                dirlen = packname.len;
 752                if (report_progress) {
 753                        oids.progress = start_delayed_progress(
 754                                _("Finding commits for commit graph"), 0);
 755                        oids.progress_done = 0;
 756                }
 757                for (i = 0; i < pack_indexes->nr; i++) {
 758                        struct packed_git *p;
 759                        strbuf_setlen(&packname, dirlen);
 760                        strbuf_addstr(&packname, pack_indexes->items[i].string);
 761                        p = add_packed_git(packname.buf, packname.len, 1);
 762                        if (!p)
 763                                die(_("error adding pack %s"), packname.buf);
 764                        if (open_pack_index(p))
 765                                die(_("error opening index for %s"), packname.buf);
 766                        for_each_object_in_pack(p, add_packed_commits, &oids, 0);
 767                        close_pack(p);
 768                        free(p);
 769                }
 770                stop_progress(&oids.progress);
 771                strbuf_release(&packname);
 772        }
 773
 774        if (commit_hex) {
 775                if (report_progress)
 776                        progress = start_delayed_progress(
 777                                _("Finding commits for commit graph"),
 778                                commit_hex->nr);
 779                for (i = 0; i < commit_hex->nr; i++) {
 780                        const char *end;
 781                        struct object_id oid;
 782                        struct commit *result;
 783
 784                        display_progress(progress, i + 1);
 785                        if (commit_hex->items[i].string &&
 786                            parse_oid_hex(commit_hex->items[i].string, &oid, &end))
 787                                continue;
 788
 789                        result = lookup_commit_reference_gently(the_repository, &oid, 1);
 790
 791                        if (result) {
 792                                ALLOC_GROW(oids.list, oids.nr + 1, oids.alloc);
 793                                oidcpy(&oids.list[oids.nr], &(result->object.oid));
 794                                oids.nr++;
 795                        }
 796                }
 797                stop_progress(&progress);
 798        }
 799
 800        if (!pack_indexes && !commit_hex) {
 801                if (report_progress)
 802                        oids.progress = start_delayed_progress(
 803                                _("Finding commits for commit graph"), 0);
 804                for_each_packed_object(add_packed_commits, &oids, 0);
 805                stop_progress(&oids.progress);
 806        }
 807
 808        close_reachable(&oids, report_progress);
 809
 810        QSORT(oids.list, oids.nr, commit_compare);
 811
 812        count_distinct = 1;
 813        for (i = 1; i < oids.nr; i++) {
 814                if (oidcmp(&oids.list[i-1], &oids.list[i]))
 815                        count_distinct++;
 816        }
 817
 818        if (count_distinct >= GRAPH_PARENT_MISSING)
 819                die(_("the commit graph format cannot write %d commits"), count_distinct);
 820
 821        commits.nr = 0;
 822        commits.alloc = count_distinct;
 823        ALLOC_ARRAY(commits.list, commits.alloc);
 824
 825        num_extra_edges = 0;
 826        for (i = 0; i < oids.nr; i++) {
 827                int num_parents = 0;
 828                if (i > 0 && !oidcmp(&oids.list[i-1], &oids.list[i]))
 829                        continue;
 830
 831                commits.list[commits.nr] = lookup_commit(the_repository, &oids.list[i]);
 832                parse_commit(commits.list[commits.nr]);
 833
 834                for (parent = commits.list[commits.nr]->parents;
 835                     parent; parent = parent->next)
 836                        num_parents++;
 837
 838                if (num_parents > 2)
 839                        num_extra_edges += num_parents - 1;
 840
 841                commits.nr++;
 842        }
 843        num_chunks = num_extra_edges ? 4 : 3;
 844
 845        if (commits.nr >= GRAPH_PARENT_MISSING)
 846                die(_("too many commits to write graph"));
 847
 848        compute_generation_numbers(&commits, report_progress);
 849
 850        graph_name = get_commit_graph_filename(obj_dir);
 851        if (safe_create_leading_directories(graph_name)) {
 852                UNLEAK(graph_name);
 853                die_errno(_("unable to create leading directories of %s"),
 854                          graph_name);
 855        }
 856
 857        hold_lock_file_for_update(&lk, graph_name, LOCK_DIE_ON_ERROR);
 858        f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
 859
 860        hashwrite_be32(f, GRAPH_SIGNATURE);
 861
 862        hashwrite_u8(f, GRAPH_VERSION);
 863        hashwrite_u8(f, GRAPH_OID_VERSION);
 864        hashwrite_u8(f, num_chunks);
 865        hashwrite_u8(f, 0); /* unused padding byte */
 866
 867        chunk_ids[0] = GRAPH_CHUNKID_OIDFANOUT;
 868        chunk_ids[1] = GRAPH_CHUNKID_OIDLOOKUP;
 869        chunk_ids[2] = GRAPH_CHUNKID_DATA;
 870        if (num_extra_edges)
 871                chunk_ids[3] = GRAPH_CHUNKID_LARGEEDGES;
 872        else
 873                chunk_ids[3] = 0;
 874        chunk_ids[4] = 0;
 875
 876        chunk_offsets[0] = 8 + (num_chunks + 1) * GRAPH_CHUNKLOOKUP_WIDTH;
 877        chunk_offsets[1] = chunk_offsets[0] + GRAPH_FANOUT_SIZE;
 878        chunk_offsets[2] = chunk_offsets[1] + GRAPH_OID_LEN * commits.nr;
 879        chunk_offsets[3] = chunk_offsets[2] + (GRAPH_OID_LEN + 16) * commits.nr;
 880        chunk_offsets[4] = chunk_offsets[3] + 4 * num_extra_edges;
 881
 882        for (i = 0; i <= num_chunks; i++) {
 883                uint32_t chunk_write[3];
 884
 885                chunk_write[0] = htonl(chunk_ids[i]);
 886                chunk_write[1] = htonl(chunk_offsets[i] >> 32);
 887                chunk_write[2] = htonl(chunk_offsets[i] & 0xffffffff);
 888                hashwrite(f, chunk_write, 12);
 889        }
 890
 891        write_graph_chunk_fanout(f, commits.list, commits.nr);
 892        write_graph_chunk_oids(f, GRAPH_OID_LEN, commits.list, commits.nr);
 893        write_graph_chunk_data(f, GRAPH_OID_LEN, commits.list, commits.nr);
 894        write_graph_chunk_large_edges(f, commits.list, commits.nr);
 895
 896        close_commit_graph();
 897        finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_FSYNC);
 898        commit_lock_file(&lk);
 899
 900        free(graph_name);
 901        free(commits.list);
 902        free(oids.list);
 903}
 904
 905#define VERIFY_COMMIT_GRAPH_ERROR_HASH 2
 906static int verify_commit_graph_error;
 907
 908static void graph_report(const char *fmt, ...)
 909{
 910        va_list ap;
 911
 912        verify_commit_graph_error = 1;
 913        va_start(ap, fmt);
 914        vfprintf(stderr, fmt, ap);
 915        fprintf(stderr, "\n");
 916        va_end(ap);
 917}
 918
 919#define GENERATION_ZERO_EXISTS 1
 920#define GENERATION_NUMBER_EXISTS 2
 921
 922int verify_commit_graph(struct repository *r, struct commit_graph *g)
 923{
 924        uint32_t i, cur_fanout_pos = 0;
 925        struct object_id prev_oid, cur_oid, checksum;
 926        int generation_zero = 0;
 927        struct hashfile *f;
 928        int devnull;
 929        struct progress *progress = NULL;
 930
 931        if (!g) {
 932                graph_report("no commit-graph file loaded");
 933                return 1;
 934        }
 935
 936        verify_commit_graph_error = 0;
 937
 938        if (!g->chunk_oid_fanout)
 939                graph_report("commit-graph is missing the OID Fanout chunk");
 940        if (!g->chunk_oid_lookup)
 941                graph_report("commit-graph is missing the OID Lookup chunk");
 942        if (!g->chunk_commit_data)
 943                graph_report("commit-graph is missing the Commit Data chunk");
 944
 945        if (verify_commit_graph_error)
 946                return verify_commit_graph_error;
 947
 948        devnull = open("/dev/null", O_WRONLY);
 949        f = hashfd(devnull, NULL);
 950        hashwrite(f, g->data, g->data_len - g->hash_len);
 951        finalize_hashfile(f, checksum.hash, CSUM_CLOSE);
 952        if (hashcmp(checksum.hash, g->data + g->data_len - g->hash_len)) {
 953                graph_report(_("the commit-graph file has incorrect checksum and is likely corrupt"));
 954                verify_commit_graph_error = VERIFY_COMMIT_GRAPH_ERROR_HASH;
 955        }
 956
 957        for (i = 0; i < g->num_commits; i++) {
 958                struct commit *graph_commit;
 959
 960                hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
 961
 962                if (i && oidcmp(&prev_oid, &cur_oid) >= 0)
 963                        graph_report("commit-graph has incorrect OID order: %s then %s",
 964                                     oid_to_hex(&prev_oid),
 965                                     oid_to_hex(&cur_oid));
 966
 967                oidcpy(&prev_oid, &cur_oid);
 968
 969                while (cur_oid.hash[0] > cur_fanout_pos) {
 970                        uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
 971
 972                        if (i != fanout_value)
 973                                graph_report("commit-graph has incorrect fanout value: fanout[%d] = %u != %u",
 974                                             cur_fanout_pos, fanout_value, i);
 975                        cur_fanout_pos++;
 976                }
 977
 978                graph_commit = lookup_commit(r, &cur_oid);
 979                if (!parse_commit_in_graph_one(g, graph_commit))
 980                        graph_report("failed to parse %s from commit-graph",
 981                                     oid_to_hex(&cur_oid));
 982        }
 983
 984        while (cur_fanout_pos < 256) {
 985                uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
 986
 987                if (g->num_commits != fanout_value)
 988                        graph_report("commit-graph has incorrect fanout value: fanout[%d] = %u != %u",
 989                                     cur_fanout_pos, fanout_value, i);
 990
 991                cur_fanout_pos++;
 992        }
 993
 994        if (verify_commit_graph_error & ~VERIFY_COMMIT_GRAPH_ERROR_HASH)
 995                return verify_commit_graph_error;
 996
 997        progress = start_progress(_("Verifying commits in commit graph"),
 998                                  g->num_commits);
 999        for (i = 0; i < g->num_commits; i++) {
1000                struct commit *graph_commit, *odb_commit;
1001                struct commit_list *graph_parents, *odb_parents;
1002                uint32_t max_generation = 0;
1003
1004                display_progress(progress, i + 1);
1005                hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
1006
1007                graph_commit = lookup_commit(r, &cur_oid);
1008                odb_commit = (struct commit *)create_object(r, cur_oid.hash, alloc_commit_node(r));
1009                if (parse_commit_internal(odb_commit, 0, 0)) {
1010                        graph_report("failed to parse %s from object database",
1011                                     oid_to_hex(&cur_oid));
1012                        continue;
1013                }
1014
1015                if (oidcmp(&get_commit_tree_in_graph_one(g, graph_commit)->object.oid,
1016                           get_commit_tree_oid(odb_commit)))
1017                        graph_report("root tree OID for commit %s in commit-graph is %s != %s",
1018                                     oid_to_hex(&cur_oid),
1019                                     oid_to_hex(get_commit_tree_oid(graph_commit)),
1020                                     oid_to_hex(get_commit_tree_oid(odb_commit)));
1021
1022                graph_parents = graph_commit->parents;
1023                odb_parents = odb_commit->parents;
1024
1025                while (graph_parents) {
1026                        if (odb_parents == NULL) {
1027                                graph_report("commit-graph parent list for commit %s is too long",
1028                                             oid_to_hex(&cur_oid));
1029                                break;
1030                        }
1031
1032                        if (oidcmp(&graph_parents->item->object.oid, &odb_parents->item->object.oid))
1033                                graph_report("commit-graph parent for %s is %s != %s",
1034                                             oid_to_hex(&cur_oid),
1035                                             oid_to_hex(&graph_parents->item->object.oid),
1036                                             oid_to_hex(&odb_parents->item->object.oid));
1037
1038                        if (graph_parents->item->generation > max_generation)
1039                                max_generation = graph_parents->item->generation;
1040
1041                        graph_parents = graph_parents->next;
1042                        odb_parents = odb_parents->next;
1043                }
1044
1045                if (odb_parents != NULL)
1046                        graph_report("commit-graph parent list for commit %s terminates early",
1047                                     oid_to_hex(&cur_oid));
1048
1049                if (!graph_commit->generation) {
1050                        if (generation_zero == GENERATION_NUMBER_EXISTS)
1051                                graph_report("commit-graph has generation number zero for commit %s, but non-zero elsewhere",
1052                                             oid_to_hex(&cur_oid));
1053                        generation_zero = GENERATION_ZERO_EXISTS;
1054                } else if (generation_zero == GENERATION_ZERO_EXISTS)
1055                        graph_report("commit-graph has non-zero generation number for commit %s, but zero elsewhere",
1056                                     oid_to_hex(&cur_oid));
1057
1058                if (generation_zero == GENERATION_ZERO_EXISTS)
1059                        continue;
1060
1061                /*
1062                 * If one of our parents has generation GENERATION_NUMBER_MAX, then
1063                 * our generation is also GENERATION_NUMBER_MAX. Decrement to avoid
1064                 * extra logic in the following condition.
1065                 */
1066                if (max_generation == GENERATION_NUMBER_MAX)
1067                        max_generation--;
1068
1069                if (graph_commit->generation != max_generation + 1)
1070                        graph_report("commit-graph generation for commit %s is %u != %u",
1071                                     oid_to_hex(&cur_oid),
1072                                     graph_commit->generation,
1073                                     max_generation + 1);
1074
1075                if (graph_commit->date != odb_commit->date)
1076                        graph_report("commit date for commit %s in commit-graph is %"PRItime" != %"PRItime,
1077                                     oid_to_hex(&cur_oid),
1078                                     graph_commit->date,
1079                                     odb_commit->date);
1080        }
1081        stop_progress(&progress);
1082
1083        return verify_commit_graph_error;
1084}
1085
1086void free_commit_graph(struct commit_graph *g)
1087{
1088        if (!g)
1089                return;
1090        if (g->graph_fd >= 0) {
1091                munmap((void *)g->data, g->data_len);
1092                g->data = NULL;
1093                close(g->graph_fd);
1094        }
1095        free(g);
1096}