fast-import.con commit object-store: rename and expand packed_git's sha1 member (538b152)
   1#include "builtin.h"
   2#include "cache.h"
   3#include "repository.h"
   4#include "config.h"
   5#include "lockfile.h"
   6#include "object.h"
   7#include "blob.h"
   8#include "tree.h"
   9#include "commit.h"
  10#include "delta.h"
  11#include "pack.h"
  12#include "refs.h"
  13#include "csum-file.h"
  14#include "quote.h"
  15#include "dir.h"
  16#include "run-command.h"
  17#include "packfile.h"
  18#include "object-store.h"
  19#include "mem-pool.h"
  20#include "commit-reach.h"
  21
  22#define PACK_ID_BITS 16
  23#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
  24#define DEPTH_BITS 13
  25#define MAX_DEPTH ((1<<DEPTH_BITS)-1)
  26
  27/*
  28 * We abuse the setuid bit on directories to mean "do not delta".
  29 */
  30#define NO_DELTA S_ISUID
  31
  32struct object_entry {
  33        struct pack_idx_entry idx;
  34        struct object_entry *next;
  35        uint32_t type : TYPE_BITS,
  36                pack_id : PACK_ID_BITS,
  37                depth : DEPTH_BITS;
  38};
  39
  40struct object_entry_pool {
  41        struct object_entry_pool *next_pool;
  42        struct object_entry *next_free;
  43        struct object_entry *end;
  44        struct object_entry entries[FLEX_ARRAY]; /* more */
  45};
  46
  47struct mark_set {
  48        union {
  49                struct object_entry *marked[1024];
  50                struct mark_set *sets[1024];
  51        } data;
  52        unsigned int shift;
  53};
  54
  55struct last_object {
  56        struct strbuf data;
  57        off_t offset;
  58        unsigned int depth;
  59        unsigned no_swap : 1;
  60};
  61
  62struct atom_str {
  63        struct atom_str *next_atom;
  64        unsigned short str_len;
  65        char str_dat[FLEX_ARRAY]; /* more */
  66};
  67
  68struct tree_content;
  69struct tree_entry {
  70        struct tree_content *tree;
  71        struct atom_str *name;
  72        struct tree_entry_ms {
  73                uint16_t mode;
  74                struct object_id oid;
  75        } versions[2];
  76};
  77
  78struct tree_content {
  79        unsigned int entry_capacity; /* must match avail_tree_content */
  80        unsigned int entry_count;
  81        unsigned int delta_depth;
  82        struct tree_entry *entries[FLEX_ARRAY]; /* more */
  83};
  84
  85struct avail_tree_content {
  86        unsigned int entry_capacity; /* must match tree_content */
  87        struct avail_tree_content *next_avail;
  88};
  89
  90struct branch {
  91        struct branch *table_next_branch;
  92        struct branch *active_next_branch;
  93        const char *name;
  94        struct tree_entry branch_tree;
  95        uintmax_t last_commit;
  96        uintmax_t num_notes;
  97        unsigned active : 1;
  98        unsigned delete : 1;
  99        unsigned pack_id : PACK_ID_BITS;
 100        struct object_id oid;
 101};
 102
 103struct tag {
 104        struct tag *next_tag;
 105        const char *name;
 106        unsigned int pack_id;
 107        struct object_id oid;
 108};
 109
 110struct hash_list {
 111        struct hash_list *next;
 112        struct object_id oid;
 113};
 114
 115typedef enum {
 116        WHENSPEC_RAW = 1,
 117        WHENSPEC_RFC2822,
 118        WHENSPEC_NOW
 119} whenspec_type;
 120
 121struct recent_command {
 122        struct recent_command *prev;
 123        struct recent_command *next;
 124        char *buf;
 125};
 126
 127/* Configured limits on output */
 128static unsigned long max_depth = 50;
 129static off_t max_packsize;
 130static int unpack_limit = 100;
 131static int force_update;
 132
 133/* Stats and misc. counters */
 134static uintmax_t alloc_count;
 135static uintmax_t marks_set_count;
 136static uintmax_t object_count_by_type[1 << TYPE_BITS];
 137static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 138static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 139static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
 140static unsigned long object_count;
 141static unsigned long branch_count;
 142static unsigned long branch_load_count;
 143static int failure;
 144static FILE *pack_edges;
 145static unsigned int show_stats = 1;
 146static int global_argc;
 147static const char **global_argv;
 148
 149/* Memory pools */
 150static struct mem_pool fi_mem_pool =  {NULL, 2*1024*1024 -
 151                                       sizeof(struct mp_block), 0 };
 152
 153/* Atom management */
 154static unsigned int atom_table_sz = 4451;
 155static unsigned int atom_cnt;
 156static struct atom_str **atom_table;
 157
 158/* The .pack file being generated */
 159static struct pack_idx_option pack_idx_opts;
 160static unsigned int pack_id;
 161static struct hashfile *pack_file;
 162static struct packed_git *pack_data;
 163static struct packed_git **all_packs;
 164static off_t pack_size;
 165
 166/* Table of objects we've written. */
 167static unsigned int object_entry_alloc = 5000;
 168static struct object_entry_pool *blocks;
 169static struct object_entry *object_table[1 << 16];
 170static struct mark_set *marks;
 171static const char *export_marks_file;
 172static const char *import_marks_file;
 173static int import_marks_file_from_stream;
 174static int import_marks_file_ignore_missing;
 175static int import_marks_file_done;
 176static int relative_marks_paths;
 177
 178/* Our last blob */
 179static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
 180
 181/* Tree management */
 182static unsigned int tree_entry_alloc = 1000;
 183static void *avail_tree_entry;
 184static unsigned int avail_tree_table_sz = 100;
 185static struct avail_tree_content **avail_tree_table;
 186static size_t tree_entry_allocd;
 187static struct strbuf old_tree = STRBUF_INIT;
 188static struct strbuf new_tree = STRBUF_INIT;
 189
 190/* Branch data */
 191static unsigned long max_active_branches = 5;
 192static unsigned long cur_active_branches;
 193static unsigned long branch_table_sz = 1039;
 194static struct branch **branch_table;
 195static struct branch *active_branches;
 196
 197/* Tag data */
 198static struct tag *first_tag;
 199static struct tag *last_tag;
 200
 201/* Input stream parsing */
 202static whenspec_type whenspec = WHENSPEC_RAW;
 203static struct strbuf command_buf = STRBUF_INIT;
 204static int unread_command_buf;
 205static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 206static struct recent_command *cmd_tail = &cmd_hist;
 207static struct recent_command *rc_free;
 208static unsigned int cmd_save = 100;
 209static uintmax_t next_mark;
 210static struct strbuf new_data = STRBUF_INIT;
 211static int seen_data_command;
 212static int require_explicit_termination;
 213
 214/* Signal handling */
 215static volatile sig_atomic_t checkpoint_requested;
 216
 217/* Where to write output of cat-blob commands */
 218static int cat_blob_fd = STDOUT_FILENO;
 219
 220static void parse_argv(void);
 221static void parse_get_mark(const char *p);
 222static void parse_cat_blob(const char *p);
 223static void parse_ls(const char *p, struct branch *b);
 224
 225static void write_branch_report(FILE *rpt, struct branch *b)
 226{
 227        fprintf(rpt, "%s:\n", b->name);
 228
 229        fprintf(rpt, "  status      :");
 230        if (b->active)
 231                fputs(" active", rpt);
 232        if (b->branch_tree.tree)
 233                fputs(" loaded", rpt);
 234        if (is_null_oid(&b->branch_tree.versions[1].oid))
 235                fputs(" dirty", rpt);
 236        fputc('\n', rpt);
 237
 238        fprintf(rpt, "  tip commit  : %s\n", oid_to_hex(&b->oid));
 239        fprintf(rpt, "  old tree    : %s\n",
 240                oid_to_hex(&b->branch_tree.versions[0].oid));
 241        fprintf(rpt, "  cur tree    : %s\n",
 242                oid_to_hex(&b->branch_tree.versions[1].oid));
 243        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 244
 245        fputs("  last pack   : ", rpt);
 246        if (b->pack_id < MAX_PACK_ID)
 247                fprintf(rpt, "%u", b->pack_id);
 248        fputc('\n', rpt);
 249
 250        fputc('\n', rpt);
 251}
 252
 253static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
 254
 255static void write_crash_report(const char *err)
 256{
 257        char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
 258        FILE *rpt = fopen(loc, "w");
 259        struct branch *b;
 260        unsigned long lu;
 261        struct recent_command *rc;
 262
 263        if (!rpt) {
 264                error_errno("can't write crash report %s", loc);
 265                free(loc);
 266                return;
 267        }
 268
 269        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 270
 271        fprintf(rpt, "fast-import crash report:\n");
 272        fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
 273        fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
 274        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
 275        fputc('\n', rpt);
 276
 277        fputs("fatal: ", rpt);
 278        fputs(err, rpt);
 279        fputc('\n', rpt);
 280
 281        fputc('\n', rpt);
 282        fputs("Most Recent Commands Before Crash\n", rpt);
 283        fputs("---------------------------------\n", rpt);
 284        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 285                if (rc->next == &cmd_hist)
 286                        fputs("* ", rpt);
 287                else
 288                        fputs("  ", rpt);
 289                fputs(rc->buf, rpt);
 290                fputc('\n', rpt);
 291        }
 292
 293        fputc('\n', rpt);
 294        fputs("Active Branch LRU\n", rpt);
 295        fputs("-----------------\n", rpt);
 296        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 297                cur_active_branches,
 298                max_active_branches);
 299        fputc('\n', rpt);
 300        fputs("  pos  clock name\n", rpt);
 301        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 302        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 303                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 304                        ++lu, b->last_commit, b->name);
 305
 306        fputc('\n', rpt);
 307        fputs("Inactive Branches\n", rpt);
 308        fputs("-----------------\n", rpt);
 309        for (lu = 0; lu < branch_table_sz; lu++) {
 310                for (b = branch_table[lu]; b; b = b->table_next_branch)
 311                        write_branch_report(rpt, b);
 312        }
 313
 314        if (first_tag) {
 315                struct tag *tg;
 316                fputc('\n', rpt);
 317                fputs("Annotated Tags\n", rpt);
 318                fputs("--------------\n", rpt);
 319                for (tg = first_tag; tg; tg = tg->next_tag) {
 320                        fputs(oid_to_hex(&tg->oid), rpt);
 321                        fputc(' ', rpt);
 322                        fputs(tg->name, rpt);
 323                        fputc('\n', rpt);
 324                }
 325        }
 326
 327        fputc('\n', rpt);
 328        fputs("Marks\n", rpt);
 329        fputs("-----\n", rpt);
 330        if (export_marks_file)
 331                fprintf(rpt, "  exported to %s\n", export_marks_file);
 332        else
 333                dump_marks_helper(rpt, 0, marks);
 334
 335        fputc('\n', rpt);
 336        fputs("-------------------\n", rpt);
 337        fputs("END OF CRASH REPORT\n", rpt);
 338        fclose(rpt);
 339        free(loc);
 340}
 341
 342static void end_packfile(void);
 343static void unkeep_all_packs(void);
 344static void dump_marks(void);
 345
 346static NORETURN void die_nicely(const char *err, va_list params)
 347{
 348        static int zombie;
 349        char message[2 * PATH_MAX];
 350
 351        vsnprintf(message, sizeof(message), err, params);
 352        fputs("fatal: ", stderr);
 353        fputs(message, stderr);
 354        fputc('\n', stderr);
 355
 356        if (!zombie) {
 357                zombie = 1;
 358                write_crash_report(message);
 359                end_packfile();
 360                unkeep_all_packs();
 361                dump_marks();
 362        }
 363        exit(128);
 364}
 365
 366#ifndef SIGUSR1 /* Windows, for example */
 367
 368static void set_checkpoint_signal(void)
 369{
 370}
 371
 372#else
 373
 374static void checkpoint_signal(int signo)
 375{
 376        checkpoint_requested = 1;
 377}
 378
 379static void set_checkpoint_signal(void)
 380{
 381        struct sigaction sa;
 382
 383        memset(&sa, 0, sizeof(sa));
 384        sa.sa_handler = checkpoint_signal;
 385        sigemptyset(&sa.sa_mask);
 386        sa.sa_flags = SA_RESTART;
 387        sigaction(SIGUSR1, &sa, NULL);
 388}
 389
 390#endif
 391
 392static void alloc_objects(unsigned int cnt)
 393{
 394        struct object_entry_pool *b;
 395
 396        b = xmalloc(sizeof(struct object_entry_pool)
 397                + cnt * sizeof(struct object_entry));
 398        b->next_pool = blocks;
 399        b->next_free = b->entries;
 400        b->end = b->entries + cnt;
 401        blocks = b;
 402        alloc_count += cnt;
 403}
 404
 405static struct object_entry *new_object(struct object_id *oid)
 406{
 407        struct object_entry *e;
 408
 409        if (blocks->next_free == blocks->end)
 410                alloc_objects(object_entry_alloc);
 411
 412        e = blocks->next_free++;
 413        oidcpy(&e->idx.oid, oid);
 414        return e;
 415}
 416
 417static struct object_entry *find_object(struct object_id *oid)
 418{
 419        unsigned int h = oid->hash[0] << 8 | oid->hash[1];
 420        struct object_entry *e;
 421        for (e = object_table[h]; e; e = e->next)
 422                if (oideq(oid, &e->idx.oid))
 423                        return e;
 424        return NULL;
 425}
 426
 427static struct object_entry *insert_object(struct object_id *oid)
 428{
 429        unsigned int h = oid->hash[0] << 8 | oid->hash[1];
 430        struct object_entry *e = object_table[h];
 431
 432        while (e) {
 433                if (oideq(oid, &e->idx.oid))
 434                        return e;
 435                e = e->next;
 436        }
 437
 438        e = new_object(oid);
 439        e->next = object_table[h];
 440        e->idx.offset = 0;
 441        object_table[h] = e;
 442        return e;
 443}
 444
 445static void invalidate_pack_id(unsigned int id)
 446{
 447        unsigned int h;
 448        unsigned long lu;
 449        struct tag *t;
 450
 451        for (h = 0; h < ARRAY_SIZE(object_table); h++) {
 452                struct object_entry *e;
 453
 454                for (e = object_table[h]; e; e = e->next)
 455                        if (e->pack_id == id)
 456                                e->pack_id = MAX_PACK_ID;
 457        }
 458
 459        for (lu = 0; lu < branch_table_sz; lu++) {
 460                struct branch *b;
 461
 462                for (b = branch_table[lu]; b; b = b->table_next_branch)
 463                        if (b->pack_id == id)
 464                                b->pack_id = MAX_PACK_ID;
 465        }
 466
 467        for (t = first_tag; t; t = t->next_tag)
 468                if (t->pack_id == id)
 469                        t->pack_id = MAX_PACK_ID;
 470}
 471
 472static unsigned int hc_str(const char *s, size_t len)
 473{
 474        unsigned int r = 0;
 475        while (len-- > 0)
 476                r = r * 31 + *s++;
 477        return r;
 478}
 479
 480static char *pool_strdup(const char *s)
 481{
 482        size_t len = strlen(s) + 1;
 483        char *r = mem_pool_alloc(&fi_mem_pool, len);
 484        memcpy(r, s, len);
 485        return r;
 486}
 487
 488static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 489{
 490        struct mark_set *s = marks;
 491        while ((idnum >> s->shift) >= 1024) {
 492                s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
 493                s->shift = marks->shift + 10;
 494                s->data.sets[0] = marks;
 495                marks = s;
 496        }
 497        while (s->shift) {
 498                uintmax_t i = idnum >> s->shift;
 499                idnum -= i << s->shift;
 500                if (!s->data.sets[i]) {
 501                        s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
 502                        s->data.sets[i]->shift = s->shift - 10;
 503                }
 504                s = s->data.sets[i];
 505        }
 506        if (!s->data.marked[idnum])
 507                marks_set_count++;
 508        s->data.marked[idnum] = oe;
 509}
 510
 511static struct object_entry *find_mark(uintmax_t idnum)
 512{
 513        uintmax_t orig_idnum = idnum;
 514        struct mark_set *s = marks;
 515        struct object_entry *oe = NULL;
 516        if ((idnum >> s->shift) < 1024) {
 517                while (s && s->shift) {
 518                        uintmax_t i = idnum >> s->shift;
 519                        idnum -= i << s->shift;
 520                        s = s->data.sets[i];
 521                }
 522                if (s)
 523                        oe = s->data.marked[idnum];
 524        }
 525        if (!oe)
 526                die("mark :%" PRIuMAX " not declared", orig_idnum);
 527        return oe;
 528}
 529
 530static struct atom_str *to_atom(const char *s, unsigned short len)
 531{
 532        unsigned int hc = hc_str(s, len) % atom_table_sz;
 533        struct atom_str *c;
 534
 535        for (c = atom_table[hc]; c; c = c->next_atom)
 536                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 537                        return c;
 538
 539        c = mem_pool_alloc(&fi_mem_pool, sizeof(struct atom_str) + len + 1);
 540        c->str_len = len;
 541        memcpy(c->str_dat, s, len);
 542        c->str_dat[len] = 0;
 543        c->next_atom = atom_table[hc];
 544        atom_table[hc] = c;
 545        atom_cnt++;
 546        return c;
 547}
 548
 549static struct branch *lookup_branch(const char *name)
 550{
 551        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 552        struct branch *b;
 553
 554        for (b = branch_table[hc]; b; b = b->table_next_branch)
 555                if (!strcmp(name, b->name))
 556                        return b;
 557        return NULL;
 558}
 559
 560static struct branch *new_branch(const char *name)
 561{
 562        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 563        struct branch *b = lookup_branch(name);
 564
 565        if (b)
 566                die("Invalid attempt to create duplicate branch: %s", name);
 567        if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
 568                die("Branch name doesn't conform to GIT standards: %s", name);
 569
 570        b = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct branch));
 571        b->name = pool_strdup(name);
 572        b->table_next_branch = branch_table[hc];
 573        b->branch_tree.versions[0].mode = S_IFDIR;
 574        b->branch_tree.versions[1].mode = S_IFDIR;
 575        b->num_notes = 0;
 576        b->active = 0;
 577        b->pack_id = MAX_PACK_ID;
 578        branch_table[hc] = b;
 579        branch_count++;
 580        return b;
 581}
 582
 583static unsigned int hc_entries(unsigned int cnt)
 584{
 585        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 586        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 587}
 588
 589static struct tree_content *new_tree_content(unsigned int cnt)
 590{
 591        struct avail_tree_content *f, *l = NULL;
 592        struct tree_content *t;
 593        unsigned int hc = hc_entries(cnt);
 594
 595        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 596                if (f->entry_capacity >= cnt)
 597                        break;
 598
 599        if (f) {
 600                if (l)
 601                        l->next_avail = f->next_avail;
 602                else
 603                        avail_tree_table[hc] = f->next_avail;
 604        } else {
 605                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 606                f = mem_pool_alloc(&fi_mem_pool, sizeof(*t) + sizeof(t->entries[0]) * cnt);
 607                f->entry_capacity = cnt;
 608        }
 609
 610        t = (struct tree_content*)f;
 611        t->entry_count = 0;
 612        t->delta_depth = 0;
 613        return t;
 614}
 615
 616static void release_tree_entry(struct tree_entry *e);
 617static void release_tree_content(struct tree_content *t)
 618{
 619        struct avail_tree_content *f = (struct avail_tree_content*)t;
 620        unsigned int hc = hc_entries(f->entry_capacity);
 621        f->next_avail = avail_tree_table[hc];
 622        avail_tree_table[hc] = f;
 623}
 624
 625static void release_tree_content_recursive(struct tree_content *t)
 626{
 627        unsigned int i;
 628        for (i = 0; i < t->entry_count; i++)
 629                release_tree_entry(t->entries[i]);
 630        release_tree_content(t);
 631}
 632
 633static struct tree_content *grow_tree_content(
 634        struct tree_content *t,
 635        int amt)
 636{
 637        struct tree_content *r = new_tree_content(t->entry_count + amt);
 638        r->entry_count = t->entry_count;
 639        r->delta_depth = t->delta_depth;
 640        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 641        release_tree_content(t);
 642        return r;
 643}
 644
 645static struct tree_entry *new_tree_entry(void)
 646{
 647        struct tree_entry *e;
 648
 649        if (!avail_tree_entry) {
 650                unsigned int n = tree_entry_alloc;
 651                tree_entry_allocd += n * sizeof(struct tree_entry);
 652                ALLOC_ARRAY(e, n);
 653                avail_tree_entry = e;
 654                while (n-- > 1) {
 655                        *((void**)e) = e + 1;
 656                        e++;
 657                }
 658                *((void**)e) = NULL;
 659        }
 660
 661        e = avail_tree_entry;
 662        avail_tree_entry = *((void**)e);
 663        return e;
 664}
 665
 666static void release_tree_entry(struct tree_entry *e)
 667{
 668        if (e->tree)
 669                release_tree_content_recursive(e->tree);
 670        *((void**)e) = avail_tree_entry;
 671        avail_tree_entry = e;
 672}
 673
 674static struct tree_content *dup_tree_content(struct tree_content *s)
 675{
 676        struct tree_content *d;
 677        struct tree_entry *a, *b;
 678        unsigned int i;
 679
 680        if (!s)
 681                return NULL;
 682        d = new_tree_content(s->entry_count);
 683        for (i = 0; i < s->entry_count; i++) {
 684                a = s->entries[i];
 685                b = new_tree_entry();
 686                memcpy(b, a, sizeof(*a));
 687                if (a->tree && is_null_oid(&b->versions[1].oid))
 688                        b->tree = dup_tree_content(a->tree);
 689                else
 690                        b->tree = NULL;
 691                d->entries[i] = b;
 692        }
 693        d->entry_count = s->entry_count;
 694        d->delta_depth = s->delta_depth;
 695
 696        return d;
 697}
 698
 699static void start_packfile(void)
 700{
 701        struct strbuf tmp_file = STRBUF_INIT;
 702        struct packed_git *p;
 703        struct pack_header hdr;
 704        int pack_fd;
 705
 706        pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
 707        FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
 708        strbuf_release(&tmp_file);
 709
 710        p->pack_fd = pack_fd;
 711        p->do_not_close = 1;
 712        pack_file = hashfd(pack_fd, p->pack_name);
 713
 714        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 715        hdr.hdr_version = htonl(2);
 716        hdr.hdr_entries = 0;
 717        hashwrite(pack_file, &hdr, sizeof(hdr));
 718
 719        pack_data = p;
 720        pack_size = sizeof(hdr);
 721        object_count = 0;
 722
 723        REALLOC_ARRAY(all_packs, pack_id + 1);
 724        all_packs[pack_id] = p;
 725}
 726
 727static const char *create_index(void)
 728{
 729        const char *tmpfile;
 730        struct pack_idx_entry **idx, **c, **last;
 731        struct object_entry *e;
 732        struct object_entry_pool *o;
 733
 734        /* Build the table of object IDs. */
 735        ALLOC_ARRAY(idx, object_count);
 736        c = idx;
 737        for (o = blocks; o; o = o->next_pool)
 738                for (e = o->next_free; e-- != o->entries;)
 739                        if (pack_id == e->pack_id)
 740                                *c++ = &e->idx;
 741        last = idx + object_count;
 742        if (c != last)
 743                die("internal consistency error creating the index");
 744
 745        tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts,
 746                                 pack_data->hash);
 747        free(idx);
 748        return tmpfile;
 749}
 750
 751static char *keep_pack(const char *curr_index_name)
 752{
 753        static const char *keep_msg = "fast-import";
 754        struct strbuf name = STRBUF_INIT;
 755        int keep_fd;
 756
 757        odb_pack_name(&name, pack_data->hash, "keep");
 758        keep_fd = odb_pack_keep(name.buf);
 759        if (keep_fd < 0)
 760                die_errno("cannot create keep file");
 761        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 762        if (close(keep_fd))
 763                die_errno("failed to write keep file");
 764
 765        odb_pack_name(&name, pack_data->hash, "pack");
 766        if (finalize_object_file(pack_data->pack_name, name.buf))
 767                die("cannot store pack file");
 768
 769        odb_pack_name(&name, pack_data->hash, "idx");
 770        if (finalize_object_file(curr_index_name, name.buf))
 771                die("cannot store index file");
 772        free((void *)curr_index_name);
 773        return strbuf_detach(&name, NULL);
 774}
 775
 776static void unkeep_all_packs(void)
 777{
 778        struct strbuf name = STRBUF_INIT;
 779        int k;
 780
 781        for (k = 0; k < pack_id; k++) {
 782                struct packed_git *p = all_packs[k];
 783                odb_pack_name(&name, p->hash, "keep");
 784                unlink_or_warn(name.buf);
 785        }
 786        strbuf_release(&name);
 787}
 788
 789static int loosen_small_pack(const struct packed_git *p)
 790{
 791        struct child_process unpack = CHILD_PROCESS_INIT;
 792
 793        if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
 794                die_errno("Failed seeking to start of '%s'", p->pack_name);
 795
 796        unpack.in = p->pack_fd;
 797        unpack.git_cmd = 1;
 798        unpack.stdout_to_stderr = 1;
 799        argv_array_push(&unpack.args, "unpack-objects");
 800        if (!show_stats)
 801                argv_array_push(&unpack.args, "-q");
 802
 803        return run_command(&unpack);
 804}
 805
 806static void end_packfile(void)
 807{
 808        static int running;
 809
 810        if (running || !pack_data)
 811                return;
 812
 813        running = 1;
 814        clear_delta_base_cache();
 815        if (object_count) {
 816                struct packed_git *new_p;
 817                struct object_id cur_pack_oid;
 818                char *idx_name;
 819                int i;
 820                struct branch *b;
 821                struct tag *t;
 822
 823                close_pack_windows(pack_data);
 824                finalize_hashfile(pack_file, cur_pack_oid.hash, 0);
 825                fixup_pack_header_footer(pack_data->pack_fd, pack_data->hash,
 826                                         pack_data->pack_name, object_count,
 827                                         cur_pack_oid.hash, pack_size);
 828
 829                if (object_count <= unpack_limit) {
 830                        if (!loosen_small_pack(pack_data)) {
 831                                invalidate_pack_id(pack_id);
 832                                goto discard_pack;
 833                        }
 834                }
 835
 836                close(pack_data->pack_fd);
 837                idx_name = keep_pack(create_index());
 838
 839                /* Register the packfile with core git's machinery. */
 840                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 841                if (!new_p)
 842                        die("core git rejected index %s", idx_name);
 843                all_packs[pack_id] = new_p;
 844                install_packed_git(the_repository, new_p);
 845                free(idx_name);
 846
 847                /* Print the boundary */
 848                if (pack_edges) {
 849                        fprintf(pack_edges, "%s:", new_p->pack_name);
 850                        for (i = 0; i < branch_table_sz; i++) {
 851                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 852                                        if (b->pack_id == pack_id)
 853                                                fprintf(pack_edges, " %s",
 854                                                        oid_to_hex(&b->oid));
 855                                }
 856                        }
 857                        for (t = first_tag; t; t = t->next_tag) {
 858                                if (t->pack_id == pack_id)
 859                                        fprintf(pack_edges, " %s",
 860                                                oid_to_hex(&t->oid));
 861                        }
 862                        fputc('\n', pack_edges);
 863                        fflush(pack_edges);
 864                }
 865
 866                pack_id++;
 867        }
 868        else {
 869discard_pack:
 870                close(pack_data->pack_fd);
 871                unlink_or_warn(pack_data->pack_name);
 872        }
 873        FREE_AND_NULL(pack_data);
 874        running = 0;
 875
 876        /* We can't carry a delta across packfiles. */
 877        strbuf_release(&last_blob.data);
 878        last_blob.offset = 0;
 879        last_blob.depth = 0;
 880}
 881
 882static void cycle_packfile(void)
 883{
 884        end_packfile();
 885        start_packfile();
 886}
 887
 888static int store_object(
 889        enum object_type type,
 890        struct strbuf *dat,
 891        struct last_object *last,
 892        struct object_id *oidout,
 893        uintmax_t mark)
 894{
 895        void *out, *delta;
 896        struct object_entry *e;
 897        unsigned char hdr[96];
 898        struct object_id oid;
 899        unsigned long hdrlen, deltalen;
 900        git_hash_ctx c;
 901        git_zstream s;
 902
 903        hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
 904                           type_name(type), (unsigned long)dat->len) + 1;
 905        the_hash_algo->init_fn(&c);
 906        the_hash_algo->update_fn(&c, hdr, hdrlen);
 907        the_hash_algo->update_fn(&c, dat->buf, dat->len);
 908        the_hash_algo->final_fn(oid.hash, &c);
 909        if (oidout)
 910                oidcpy(oidout, &oid);
 911
 912        e = insert_object(&oid);
 913        if (mark)
 914                insert_mark(mark, e);
 915        if (e->idx.offset) {
 916                duplicate_count_by_type[type]++;
 917                return 1;
 918        } else if (find_sha1_pack(oid.hash,
 919                                  get_all_packs(the_repository))) {
 920                e->type = type;
 921                e->pack_id = MAX_PACK_ID;
 922                e->idx.offset = 1; /* just not zero! */
 923                duplicate_count_by_type[type]++;
 924                return 1;
 925        }
 926
 927        if (last && last->data.len && last->data.buf && last->depth < max_depth
 928                && dat->len > the_hash_algo->rawsz) {
 929
 930                delta_count_attempts_by_type[type]++;
 931                delta = diff_delta(last->data.buf, last->data.len,
 932                        dat->buf, dat->len,
 933                        &deltalen, dat->len - the_hash_algo->rawsz);
 934        } else
 935                delta = NULL;
 936
 937        git_deflate_init(&s, pack_compression_level);
 938        if (delta) {
 939                s.next_in = delta;
 940                s.avail_in = deltalen;
 941        } else {
 942                s.next_in = (void *)dat->buf;
 943                s.avail_in = dat->len;
 944        }
 945        s.avail_out = git_deflate_bound(&s, s.avail_in);
 946        s.next_out = out = xmalloc(s.avail_out);
 947        while (git_deflate(&s, Z_FINISH) == Z_OK)
 948                ; /* nothing */
 949        git_deflate_end(&s);
 950
 951        /* Determine if we should auto-checkpoint. */
 952        if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
 953                || (pack_size + 60 + s.total_out) < pack_size) {
 954
 955                /* This new object needs to *not* have the current pack_id. */
 956                e->pack_id = pack_id + 1;
 957                cycle_packfile();
 958
 959                /* We cannot carry a delta into the new pack. */
 960                if (delta) {
 961                        FREE_AND_NULL(delta);
 962
 963                        git_deflate_init(&s, pack_compression_level);
 964                        s.next_in = (void *)dat->buf;
 965                        s.avail_in = dat->len;
 966                        s.avail_out = git_deflate_bound(&s, s.avail_in);
 967                        s.next_out = out = xrealloc(out, s.avail_out);
 968                        while (git_deflate(&s, Z_FINISH) == Z_OK)
 969                                ; /* nothing */
 970                        git_deflate_end(&s);
 971                }
 972        }
 973
 974        e->type = type;
 975        e->pack_id = pack_id;
 976        e->idx.offset = pack_size;
 977        object_count++;
 978        object_count_by_type[type]++;
 979
 980        crc32_begin(pack_file);
 981
 982        if (delta) {
 983                off_t ofs = e->idx.offset - last->offset;
 984                unsigned pos = sizeof(hdr) - 1;
 985
 986                delta_count_by_type[type]++;
 987                e->depth = last->depth + 1;
 988
 989                hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
 990                                                      OBJ_OFS_DELTA, deltalen);
 991                hashwrite(pack_file, hdr, hdrlen);
 992                pack_size += hdrlen;
 993
 994                hdr[pos] = ofs & 127;
 995                while (ofs >>= 7)
 996                        hdr[--pos] = 128 | (--ofs & 127);
 997                hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
 998                pack_size += sizeof(hdr) - pos;
 999        } else {
1000                e->depth = 0;
1001                hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1002                                                      type, dat->len);
1003                hashwrite(pack_file, hdr, hdrlen);
1004                pack_size += hdrlen;
1005        }
1006
1007        hashwrite(pack_file, out, s.total_out);
1008        pack_size += s.total_out;
1009
1010        e->idx.crc32 = crc32_end(pack_file);
1011
1012        free(out);
1013        free(delta);
1014        if (last) {
1015                if (last->no_swap) {
1016                        last->data = *dat;
1017                } else {
1018                        strbuf_swap(&last->data, dat);
1019                }
1020                last->offset = e->idx.offset;
1021                last->depth = e->depth;
1022        }
1023        return 0;
1024}
1025
1026static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1027{
1028        if (hashfile_truncate(pack_file, checkpoint))
1029                die_errno("cannot truncate pack to skip duplicate");
1030        pack_size = checkpoint->offset;
1031}
1032
1033static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1034{
1035        size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1036        unsigned char *in_buf = xmalloc(in_sz);
1037        unsigned char *out_buf = xmalloc(out_sz);
1038        struct object_entry *e;
1039        struct object_id oid;
1040        unsigned long hdrlen;
1041        off_t offset;
1042        git_hash_ctx c;
1043        git_zstream s;
1044        struct hashfile_checkpoint checkpoint;
1045        int status = Z_OK;
1046
1047        /* Determine if we should auto-checkpoint. */
1048        if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1049                || (pack_size + 60 + len) < pack_size)
1050                cycle_packfile();
1051
1052        hashfile_checkpoint(pack_file, &checkpoint);
1053        offset = checkpoint.offset;
1054
1055        hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1056
1057        the_hash_algo->init_fn(&c);
1058        the_hash_algo->update_fn(&c, out_buf, hdrlen);
1059
1060        crc32_begin(pack_file);
1061
1062        git_deflate_init(&s, pack_compression_level);
1063
1064        hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1065
1066        s.next_out = out_buf + hdrlen;
1067        s.avail_out = out_sz - hdrlen;
1068
1069        while (status != Z_STREAM_END) {
1070                if (0 < len && !s.avail_in) {
1071                        size_t cnt = in_sz < len ? in_sz : (size_t)len;
1072                        size_t n = fread(in_buf, 1, cnt, stdin);
1073                        if (!n && feof(stdin))
1074                                die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1075
1076                        the_hash_algo->update_fn(&c, in_buf, n);
1077                        s.next_in = in_buf;
1078                        s.avail_in = n;
1079                        len -= n;
1080                }
1081
1082                status = git_deflate(&s, len ? 0 : Z_FINISH);
1083
1084                if (!s.avail_out || status == Z_STREAM_END) {
1085                        size_t n = s.next_out - out_buf;
1086                        hashwrite(pack_file, out_buf, n);
1087                        pack_size += n;
1088                        s.next_out = out_buf;
1089                        s.avail_out = out_sz;
1090                }
1091
1092                switch (status) {
1093                case Z_OK:
1094                case Z_BUF_ERROR:
1095                case Z_STREAM_END:
1096                        continue;
1097                default:
1098                        die("unexpected deflate failure: %d", status);
1099                }
1100        }
1101        git_deflate_end(&s);
1102        the_hash_algo->final_fn(oid.hash, &c);
1103
1104        if (oidout)
1105                oidcpy(oidout, &oid);
1106
1107        e = insert_object(&oid);
1108
1109        if (mark)
1110                insert_mark(mark, e);
1111
1112        if (e->idx.offset) {
1113                duplicate_count_by_type[OBJ_BLOB]++;
1114                truncate_pack(&checkpoint);
1115
1116        } else if (find_sha1_pack(oid.hash,
1117                                  get_all_packs(the_repository))) {
1118                e->type = OBJ_BLOB;
1119                e->pack_id = MAX_PACK_ID;
1120                e->idx.offset = 1; /* just not zero! */
1121                duplicate_count_by_type[OBJ_BLOB]++;
1122                truncate_pack(&checkpoint);
1123
1124        } else {
1125                e->depth = 0;
1126                e->type = OBJ_BLOB;
1127                e->pack_id = pack_id;
1128                e->idx.offset = offset;
1129                e->idx.crc32 = crc32_end(pack_file);
1130                object_count++;
1131                object_count_by_type[OBJ_BLOB]++;
1132        }
1133
1134        free(in_buf);
1135        free(out_buf);
1136}
1137
1138/* All calls must be guarded by find_object() or find_mark() to
1139 * ensure the 'struct object_entry' passed was written by this
1140 * process instance.  We unpack the entry by the offset, avoiding
1141 * the need for the corresponding .idx file.  This unpacking rule
1142 * works because we only use OBJ_REF_DELTA within the packfiles
1143 * created by fast-import.
1144 *
1145 * oe must not be NULL.  Such an oe usually comes from giving
1146 * an unknown SHA-1 to find_object() or an undefined mark to
1147 * find_mark().  Callers must test for this condition and use
1148 * the standard read_sha1_file() when it happens.
1149 *
1150 * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1151 * find_mark(), where the mark was reloaded from an existing marks
1152 * file and is referencing an object that this fast-import process
1153 * instance did not write out to a packfile.  Callers must test for
1154 * this condition and use read_sha1_file() instead.
1155 */
1156static void *gfi_unpack_entry(
1157        struct object_entry *oe,
1158        unsigned long *sizep)
1159{
1160        enum object_type type;
1161        struct packed_git *p = all_packs[oe->pack_id];
1162        if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1163                /* The object is stored in the packfile we are writing to
1164                 * and we have modified it since the last time we scanned
1165                 * back to read a previously written object.  If an old
1166                 * window covered [p->pack_size, p->pack_size + rawsz) its
1167                 * data is stale and is not valid.  Closing all windows
1168                 * and updating the packfile length ensures we can read
1169                 * the newly written data.
1170                 */
1171                close_pack_windows(p);
1172                hashflush(pack_file);
1173
1174                /* We have to offer rawsz bytes additional on the end of
1175                 * the packfile as the core unpacker code assumes the
1176                 * footer is present at the file end and must promise
1177                 * at least rawsz bytes within any window it maps.  But
1178                 * we don't actually create the footer here.
1179                 */
1180                p->pack_size = pack_size + the_hash_algo->rawsz;
1181        }
1182        return unpack_entry(the_repository, p, oe->idx.offset, &type, sizep);
1183}
1184
1185static const char *get_mode(const char *str, uint16_t *modep)
1186{
1187        unsigned char c;
1188        uint16_t mode = 0;
1189
1190        while ((c = *str++) != ' ') {
1191                if (c < '0' || c > '7')
1192                        return NULL;
1193                mode = (mode << 3) + (c - '0');
1194        }
1195        *modep = mode;
1196        return str;
1197}
1198
1199static void load_tree(struct tree_entry *root)
1200{
1201        struct object_id *oid = &root->versions[1].oid;
1202        struct object_entry *myoe;
1203        struct tree_content *t;
1204        unsigned long size;
1205        char *buf;
1206        const char *c;
1207
1208        root->tree = t = new_tree_content(8);
1209        if (is_null_oid(oid))
1210                return;
1211
1212        myoe = find_object(oid);
1213        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1214                if (myoe->type != OBJ_TREE)
1215                        die("Not a tree: %s", oid_to_hex(oid));
1216                t->delta_depth = myoe->depth;
1217                buf = gfi_unpack_entry(myoe, &size);
1218                if (!buf)
1219                        die("Can't load tree %s", oid_to_hex(oid));
1220        } else {
1221                enum object_type type;
1222                buf = read_object_file(oid, &type, &size);
1223                if (!buf || type != OBJ_TREE)
1224                        die("Can't load tree %s", oid_to_hex(oid));
1225        }
1226
1227        c = buf;
1228        while (c != (buf + size)) {
1229                struct tree_entry *e = new_tree_entry();
1230
1231                if (t->entry_count == t->entry_capacity)
1232                        root->tree = t = grow_tree_content(t, t->entry_count);
1233                t->entries[t->entry_count++] = e;
1234
1235                e->tree = NULL;
1236                c = get_mode(c, &e->versions[1].mode);
1237                if (!c)
1238                        die("Corrupt mode in %s", oid_to_hex(oid));
1239                e->versions[0].mode = e->versions[1].mode;
1240                e->name = to_atom(c, strlen(c));
1241                c += e->name->str_len + 1;
1242                hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1243                hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1244                c += GIT_SHA1_RAWSZ;
1245        }
1246        free(buf);
1247}
1248
1249static int tecmp0 (const void *_a, const void *_b)
1250{
1251        struct tree_entry *a = *((struct tree_entry**)_a);
1252        struct tree_entry *b = *((struct tree_entry**)_b);
1253        return base_name_compare(
1254                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1255                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1256}
1257
1258static int tecmp1 (const void *_a, const void *_b)
1259{
1260        struct tree_entry *a = *((struct tree_entry**)_a);
1261        struct tree_entry *b = *((struct tree_entry**)_b);
1262        return base_name_compare(
1263                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1264                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1265}
1266
1267static void mktree(struct tree_content *t, int v, struct strbuf *b)
1268{
1269        size_t maxlen = 0;
1270        unsigned int i;
1271
1272        if (!v)
1273                QSORT(t->entries, t->entry_count, tecmp0);
1274        else
1275                QSORT(t->entries, t->entry_count, tecmp1);
1276
1277        for (i = 0; i < t->entry_count; i++) {
1278                if (t->entries[i]->versions[v].mode)
1279                        maxlen += t->entries[i]->name->str_len + 34;
1280        }
1281
1282        strbuf_reset(b);
1283        strbuf_grow(b, maxlen);
1284        for (i = 0; i < t->entry_count; i++) {
1285                struct tree_entry *e = t->entries[i];
1286                if (!e->versions[v].mode)
1287                        continue;
1288                strbuf_addf(b, "%o %s%c",
1289                        (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1290                        e->name->str_dat, '\0');
1291                strbuf_add(b, e->versions[v].oid.hash, GIT_SHA1_RAWSZ);
1292        }
1293}
1294
1295static void store_tree(struct tree_entry *root)
1296{
1297        struct tree_content *t;
1298        unsigned int i, j, del;
1299        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1300        struct object_entry *le = NULL;
1301
1302        if (!is_null_oid(&root->versions[1].oid))
1303                return;
1304
1305        if (!root->tree)
1306                load_tree(root);
1307        t = root->tree;
1308
1309        for (i = 0; i < t->entry_count; i++) {
1310                if (t->entries[i]->tree)
1311                        store_tree(t->entries[i]);
1312        }
1313
1314        if (!(root->versions[0].mode & NO_DELTA))
1315                le = find_object(&root->versions[0].oid);
1316        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1317                mktree(t, 0, &old_tree);
1318                lo.data = old_tree;
1319                lo.offset = le->idx.offset;
1320                lo.depth = t->delta_depth;
1321        }
1322
1323        mktree(t, 1, &new_tree);
1324        store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1325
1326        t->delta_depth = lo.depth;
1327        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1328                struct tree_entry *e = t->entries[i];
1329                if (e->versions[1].mode) {
1330                        e->versions[0].mode = e->versions[1].mode;
1331                        oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1332                        t->entries[j++] = e;
1333                } else {
1334                        release_tree_entry(e);
1335                        del++;
1336                }
1337        }
1338        t->entry_count -= del;
1339}
1340
1341static void tree_content_replace(
1342        struct tree_entry *root,
1343        const struct object_id *oid,
1344        const uint16_t mode,
1345        struct tree_content *newtree)
1346{
1347        if (!S_ISDIR(mode))
1348                die("Root cannot be a non-directory");
1349        oidclr(&root->versions[0].oid);
1350        oidcpy(&root->versions[1].oid, oid);
1351        if (root->tree)
1352                release_tree_content_recursive(root->tree);
1353        root->tree = newtree;
1354}
1355
1356static int tree_content_set(
1357        struct tree_entry *root,
1358        const char *p,
1359        const struct object_id *oid,
1360        const uint16_t mode,
1361        struct tree_content *subtree)
1362{
1363        struct tree_content *t;
1364        const char *slash1;
1365        unsigned int i, n;
1366        struct tree_entry *e;
1367
1368        slash1 = strchrnul(p, '/');
1369        n = slash1 - p;
1370        if (!n)
1371                die("Empty path component found in input");
1372        if (!*slash1 && !S_ISDIR(mode) && subtree)
1373                die("Non-directories cannot have subtrees");
1374
1375        if (!root->tree)
1376                load_tree(root);
1377        t = root->tree;
1378        for (i = 0; i < t->entry_count; i++) {
1379                e = t->entries[i];
1380                if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1381                        if (!*slash1) {
1382                                if (!S_ISDIR(mode)
1383                                                && e->versions[1].mode == mode
1384                                                && oideq(&e->versions[1].oid, oid))
1385                                        return 0;
1386                                e->versions[1].mode = mode;
1387                                oidcpy(&e->versions[1].oid, oid);
1388                                if (e->tree)
1389                                        release_tree_content_recursive(e->tree);
1390                                e->tree = subtree;
1391
1392                                /*
1393                                 * We need to leave e->versions[0].sha1 alone
1394                                 * to avoid modifying the preimage tree used
1395                                 * when writing out the parent directory.
1396                                 * But after replacing the subdir with a
1397                                 * completely different one, it's not a good
1398                                 * delta base any more, and besides, we've
1399                                 * thrown away the tree entries needed to
1400                                 * make a delta against it.
1401                                 *
1402                                 * So let's just explicitly disable deltas
1403                                 * for the subtree.
1404                                 */
1405                                if (S_ISDIR(e->versions[0].mode))
1406                                        e->versions[0].mode |= NO_DELTA;
1407
1408                                oidclr(&root->versions[1].oid);
1409                                return 1;
1410                        }
1411                        if (!S_ISDIR(e->versions[1].mode)) {
1412                                e->tree = new_tree_content(8);
1413                                e->versions[1].mode = S_IFDIR;
1414                        }
1415                        if (!e->tree)
1416                                load_tree(e);
1417                        if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1418                                oidclr(&root->versions[1].oid);
1419                                return 1;
1420                        }
1421                        return 0;
1422                }
1423        }
1424
1425        if (t->entry_count == t->entry_capacity)
1426                root->tree = t = grow_tree_content(t, t->entry_count);
1427        e = new_tree_entry();
1428        e->name = to_atom(p, n);
1429        e->versions[0].mode = 0;
1430        oidclr(&e->versions[0].oid);
1431        t->entries[t->entry_count++] = e;
1432        if (*slash1) {
1433                e->tree = new_tree_content(8);
1434                e->versions[1].mode = S_IFDIR;
1435                tree_content_set(e, slash1 + 1, oid, mode, subtree);
1436        } else {
1437                e->tree = subtree;
1438                e->versions[1].mode = mode;
1439                oidcpy(&e->versions[1].oid, oid);
1440        }
1441        oidclr(&root->versions[1].oid);
1442        return 1;
1443}
1444
1445static int tree_content_remove(
1446        struct tree_entry *root,
1447        const char *p,
1448        struct tree_entry *backup_leaf,
1449        int allow_root)
1450{
1451        struct tree_content *t;
1452        const char *slash1;
1453        unsigned int i, n;
1454        struct tree_entry *e;
1455
1456        slash1 = strchrnul(p, '/');
1457        n = slash1 - p;
1458
1459        if (!root->tree)
1460                load_tree(root);
1461
1462        if (!*p && allow_root) {
1463                e = root;
1464                goto del_entry;
1465        }
1466
1467        t = root->tree;
1468        for (i = 0; i < t->entry_count; i++) {
1469                e = t->entries[i];
1470                if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1471                        if (*slash1 && !S_ISDIR(e->versions[1].mode))
1472                                /*
1473                                 * If p names a file in some subdirectory, and a
1474                                 * file or symlink matching the name of the
1475                                 * parent directory of p exists, then p cannot
1476                                 * exist and need not be deleted.
1477                                 */
1478                                return 1;
1479                        if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1480                                goto del_entry;
1481                        if (!e->tree)
1482                                load_tree(e);
1483                        if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1484                                for (n = 0; n < e->tree->entry_count; n++) {
1485                                        if (e->tree->entries[n]->versions[1].mode) {
1486                                                oidclr(&root->versions[1].oid);
1487                                                return 1;
1488                                        }
1489                                }
1490                                backup_leaf = NULL;
1491                                goto del_entry;
1492                        }
1493                        return 0;
1494                }
1495        }
1496        return 0;
1497
1498del_entry:
1499        if (backup_leaf)
1500                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1501        else if (e->tree)
1502                release_tree_content_recursive(e->tree);
1503        e->tree = NULL;
1504        e->versions[1].mode = 0;
1505        oidclr(&e->versions[1].oid);
1506        oidclr(&root->versions[1].oid);
1507        return 1;
1508}
1509
1510static int tree_content_get(
1511        struct tree_entry *root,
1512        const char *p,
1513        struct tree_entry *leaf,
1514        int allow_root)
1515{
1516        struct tree_content *t;
1517        const char *slash1;
1518        unsigned int i, n;
1519        struct tree_entry *e;
1520
1521        slash1 = strchrnul(p, '/');
1522        n = slash1 - p;
1523        if (!n && !allow_root)
1524                die("Empty path component found in input");
1525
1526        if (!root->tree)
1527                load_tree(root);
1528
1529        if (!n) {
1530                e = root;
1531                goto found_entry;
1532        }
1533
1534        t = root->tree;
1535        for (i = 0; i < t->entry_count; i++) {
1536                e = t->entries[i];
1537                if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1538                        if (!*slash1)
1539                                goto found_entry;
1540                        if (!S_ISDIR(e->versions[1].mode))
1541                                return 0;
1542                        if (!e->tree)
1543                                load_tree(e);
1544                        return tree_content_get(e, slash1 + 1, leaf, 0);
1545                }
1546        }
1547        return 0;
1548
1549found_entry:
1550        memcpy(leaf, e, sizeof(*leaf));
1551        if (e->tree && is_null_oid(&e->versions[1].oid))
1552                leaf->tree = dup_tree_content(e->tree);
1553        else
1554                leaf->tree = NULL;
1555        return 1;
1556}
1557
1558static int update_branch(struct branch *b)
1559{
1560        static const char *msg = "fast-import";
1561        struct ref_transaction *transaction;
1562        struct object_id old_oid;
1563        struct strbuf err = STRBUF_INIT;
1564
1565        if (is_null_oid(&b->oid)) {
1566                if (b->delete)
1567                        delete_ref(NULL, b->name, NULL, 0);
1568                return 0;
1569        }
1570        if (read_ref(b->name, &old_oid))
1571                oidclr(&old_oid);
1572        if (!force_update && !is_null_oid(&old_oid)) {
1573                struct commit *old_cmit, *new_cmit;
1574
1575                old_cmit = lookup_commit_reference_gently(the_repository,
1576                                                          &old_oid, 0);
1577                new_cmit = lookup_commit_reference_gently(the_repository,
1578                                                          &b->oid, 0);
1579                if (!old_cmit || !new_cmit)
1580                        return error("Branch %s is missing commits.", b->name);
1581
1582                if (!in_merge_bases(old_cmit, new_cmit)) {
1583                        warning("Not updating %s"
1584                                " (new tip %s does not contain %s)",
1585                                b->name, oid_to_hex(&b->oid),
1586                                oid_to_hex(&old_oid));
1587                        return -1;
1588                }
1589        }
1590        transaction = ref_transaction_begin(&err);
1591        if (!transaction ||
1592            ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1593                                   0, msg, &err) ||
1594            ref_transaction_commit(transaction, &err)) {
1595                ref_transaction_free(transaction);
1596                error("%s", err.buf);
1597                strbuf_release(&err);
1598                return -1;
1599        }
1600        ref_transaction_free(transaction);
1601        strbuf_release(&err);
1602        return 0;
1603}
1604
1605static void dump_branches(void)
1606{
1607        unsigned int i;
1608        struct branch *b;
1609
1610        for (i = 0; i < branch_table_sz; i++) {
1611                for (b = branch_table[i]; b; b = b->table_next_branch)
1612                        failure |= update_branch(b);
1613        }
1614}
1615
1616static void dump_tags(void)
1617{
1618        static const char *msg = "fast-import";
1619        struct tag *t;
1620        struct strbuf ref_name = STRBUF_INIT;
1621        struct strbuf err = STRBUF_INIT;
1622        struct ref_transaction *transaction;
1623
1624        transaction = ref_transaction_begin(&err);
1625        if (!transaction) {
1626                failure |= error("%s", err.buf);
1627                goto cleanup;
1628        }
1629        for (t = first_tag; t; t = t->next_tag) {
1630                strbuf_reset(&ref_name);
1631                strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1632
1633                if (ref_transaction_update(transaction, ref_name.buf,
1634                                           &t->oid, NULL, 0, msg, &err)) {
1635                        failure |= error("%s", err.buf);
1636                        goto cleanup;
1637                }
1638        }
1639        if (ref_transaction_commit(transaction, &err))
1640                failure |= error("%s", err.buf);
1641
1642 cleanup:
1643        ref_transaction_free(transaction);
1644        strbuf_release(&ref_name);
1645        strbuf_release(&err);
1646}
1647
1648static void dump_marks_helper(FILE *f,
1649        uintmax_t base,
1650        struct mark_set *m)
1651{
1652        uintmax_t k;
1653        if (m->shift) {
1654                for (k = 0; k < 1024; k++) {
1655                        if (m->data.sets[k])
1656                                dump_marks_helper(f, base + (k << m->shift),
1657                                        m->data.sets[k]);
1658                }
1659        } else {
1660                for (k = 0; k < 1024; k++) {
1661                        if (m->data.marked[k])
1662                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1663                                        oid_to_hex(&m->data.marked[k]->idx.oid));
1664                }
1665        }
1666}
1667
1668static void dump_marks(void)
1669{
1670        struct lock_file mark_lock = LOCK_INIT;
1671        FILE *f;
1672
1673        if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1674                return;
1675
1676        if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1677                failure |= error_errno("Unable to write marks file %s",
1678                                       export_marks_file);
1679                return;
1680        }
1681
1682        f = fdopen_lock_file(&mark_lock, "w");
1683        if (!f) {
1684                int saved_errno = errno;
1685                rollback_lock_file(&mark_lock);
1686                failure |= error("Unable to write marks file %s: %s",
1687                        export_marks_file, strerror(saved_errno));
1688                return;
1689        }
1690
1691        dump_marks_helper(f, 0, marks);
1692        if (commit_lock_file(&mark_lock)) {
1693                failure |= error_errno("Unable to write file %s",
1694                                       export_marks_file);
1695                return;
1696        }
1697}
1698
1699static void read_marks(void)
1700{
1701        char line[512];
1702        FILE *f = fopen(import_marks_file, "r");
1703        if (f)
1704                ;
1705        else if (import_marks_file_ignore_missing && errno == ENOENT)
1706                goto done; /* Marks file does not exist */
1707        else
1708                die_errno("cannot read '%s'", import_marks_file);
1709        while (fgets(line, sizeof(line), f)) {
1710                uintmax_t mark;
1711                char *end;
1712                struct object_id oid;
1713                struct object_entry *e;
1714
1715                end = strchr(line, '\n');
1716                if (line[0] != ':' || !end)
1717                        die("corrupt mark line: %s", line);
1718                *end = 0;
1719                mark = strtoumax(line + 1, &end, 10);
1720                if (!mark || end == line + 1
1721                        || *end != ' ' || get_oid_hex(end + 1, &oid))
1722                        die("corrupt mark line: %s", line);
1723                e = find_object(&oid);
1724                if (!e) {
1725                        enum object_type type = oid_object_info(the_repository,
1726                                                                &oid, NULL);
1727                        if (type < 0)
1728                                die("object not found: %s", oid_to_hex(&oid));
1729                        e = insert_object(&oid);
1730                        e->type = type;
1731                        e->pack_id = MAX_PACK_ID;
1732                        e->idx.offset = 1; /* just not zero! */
1733                }
1734                insert_mark(mark, e);
1735        }
1736        fclose(f);
1737done:
1738        import_marks_file_done = 1;
1739}
1740
1741
1742static int read_next_command(void)
1743{
1744        static int stdin_eof = 0;
1745
1746        if (stdin_eof) {
1747                unread_command_buf = 0;
1748                return EOF;
1749        }
1750
1751        for (;;) {
1752                const char *p;
1753
1754                if (unread_command_buf) {
1755                        unread_command_buf = 0;
1756                } else {
1757                        struct recent_command *rc;
1758
1759                        strbuf_detach(&command_buf, NULL);
1760                        stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1761                        if (stdin_eof)
1762                                return EOF;
1763
1764                        if (!seen_data_command
1765                                && !starts_with(command_buf.buf, "feature ")
1766                                && !starts_with(command_buf.buf, "option ")) {
1767                                parse_argv();
1768                        }
1769
1770                        rc = rc_free;
1771                        if (rc)
1772                                rc_free = rc->next;
1773                        else {
1774                                rc = cmd_hist.next;
1775                                cmd_hist.next = rc->next;
1776                                cmd_hist.next->prev = &cmd_hist;
1777                                free(rc->buf);
1778                        }
1779
1780                        rc->buf = command_buf.buf;
1781                        rc->prev = cmd_tail;
1782                        rc->next = cmd_hist.prev;
1783                        rc->prev->next = rc;
1784                        cmd_tail = rc;
1785                }
1786                if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1787                        parse_get_mark(p);
1788                        continue;
1789                }
1790                if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1791                        parse_cat_blob(p);
1792                        continue;
1793                }
1794                if (command_buf.buf[0] == '#')
1795                        continue;
1796                return 0;
1797        }
1798}
1799
1800static void skip_optional_lf(void)
1801{
1802        int term_char = fgetc(stdin);
1803        if (term_char != '\n' && term_char != EOF)
1804                ungetc(term_char, stdin);
1805}
1806
1807static void parse_mark(void)
1808{
1809        const char *v;
1810        if (skip_prefix(command_buf.buf, "mark :", &v)) {
1811                next_mark = strtoumax(v, NULL, 10);
1812                read_next_command();
1813        }
1814        else
1815                next_mark = 0;
1816}
1817
1818static void parse_original_identifier(void)
1819{
1820        const char *v;
1821        if (skip_prefix(command_buf.buf, "original-oid ", &v))
1822                read_next_command();
1823}
1824
1825static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1826{
1827        const char *data;
1828        strbuf_reset(sb);
1829
1830        if (!skip_prefix(command_buf.buf, "data ", &data))
1831                die("Expected 'data n' command, found: %s", command_buf.buf);
1832
1833        if (skip_prefix(data, "<<", &data)) {
1834                char *term = xstrdup(data);
1835                size_t term_len = command_buf.len - (data - command_buf.buf);
1836
1837                strbuf_detach(&command_buf, NULL);
1838                for (;;) {
1839                        if (strbuf_getline_lf(&command_buf, stdin) == EOF)
1840                                die("EOF in data (terminator '%s' not found)", term);
1841                        if (term_len == command_buf.len
1842                                && !strcmp(term, command_buf.buf))
1843                                break;
1844                        strbuf_addbuf(sb, &command_buf);
1845                        strbuf_addch(sb, '\n');
1846                }
1847                free(term);
1848        }
1849        else {
1850                uintmax_t len = strtoumax(data, NULL, 10);
1851                size_t n = 0, length = (size_t)len;
1852
1853                if (limit && limit < len) {
1854                        *len_res = len;
1855                        return 0;
1856                }
1857                if (length < len)
1858                        die("data is too large to use in this context");
1859
1860                while (n < length) {
1861                        size_t s = strbuf_fread(sb, length - n, stdin);
1862                        if (!s && feof(stdin))
1863                                die("EOF in data (%lu bytes remaining)",
1864                                        (unsigned long)(length - n));
1865                        n += s;
1866                }
1867        }
1868
1869        skip_optional_lf();
1870        return 1;
1871}
1872
1873static int validate_raw_date(const char *src, struct strbuf *result)
1874{
1875        const char *orig_src = src;
1876        char *endp;
1877        unsigned long num;
1878
1879        errno = 0;
1880
1881        num = strtoul(src, &endp, 10);
1882        /* NEEDSWORK: perhaps check for reasonable values? */
1883        if (errno || endp == src || *endp != ' ')
1884                return -1;
1885
1886        src = endp + 1;
1887        if (*src != '-' && *src != '+')
1888                return -1;
1889
1890        num = strtoul(src + 1, &endp, 10);
1891        if (errno || endp == src + 1 || *endp || 1400 < num)
1892                return -1;
1893
1894        strbuf_addstr(result, orig_src);
1895        return 0;
1896}
1897
1898static char *parse_ident(const char *buf)
1899{
1900        const char *ltgt;
1901        size_t name_len;
1902        struct strbuf ident = STRBUF_INIT;
1903
1904        /* ensure there is a space delimiter even if there is no name */
1905        if (*buf == '<')
1906                --buf;
1907
1908        ltgt = buf + strcspn(buf, "<>");
1909        if (*ltgt != '<')
1910                die("Missing < in ident string: %s", buf);
1911        if (ltgt != buf && ltgt[-1] != ' ')
1912                die("Missing space before < in ident string: %s", buf);
1913        ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
1914        if (*ltgt != '>')
1915                die("Missing > in ident string: %s", buf);
1916        ltgt++;
1917        if (*ltgt != ' ')
1918                die("Missing space after > in ident string: %s", buf);
1919        ltgt++;
1920        name_len = ltgt - buf;
1921        strbuf_add(&ident, buf, name_len);
1922
1923        switch (whenspec) {
1924        case WHENSPEC_RAW:
1925                if (validate_raw_date(ltgt, &ident) < 0)
1926                        die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
1927                break;
1928        case WHENSPEC_RFC2822:
1929                if (parse_date(ltgt, &ident) < 0)
1930                        die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
1931                break;
1932        case WHENSPEC_NOW:
1933                if (strcmp("now", ltgt))
1934                        die("Date in ident must be 'now': %s", buf);
1935                datestamp(&ident);
1936                break;
1937        }
1938
1939        return strbuf_detach(&ident, NULL);
1940}
1941
1942static void parse_and_store_blob(
1943        struct last_object *last,
1944        struct object_id *oidout,
1945        uintmax_t mark)
1946{
1947        static struct strbuf buf = STRBUF_INIT;
1948        uintmax_t len;
1949
1950        if (parse_data(&buf, big_file_threshold, &len))
1951                store_object(OBJ_BLOB, &buf, last, oidout, mark);
1952        else {
1953                if (last) {
1954                        strbuf_release(&last->data);
1955                        last->offset = 0;
1956                        last->depth = 0;
1957                }
1958                stream_blob(len, oidout, mark);
1959                skip_optional_lf();
1960        }
1961}
1962
1963static void parse_new_blob(void)
1964{
1965        read_next_command();
1966        parse_mark();
1967        parse_original_identifier();
1968        parse_and_store_blob(&last_blob, NULL, next_mark);
1969}
1970
1971static void unload_one_branch(void)
1972{
1973        while (cur_active_branches
1974                && cur_active_branches >= max_active_branches) {
1975                uintmax_t min_commit = ULONG_MAX;
1976                struct branch *e, *l = NULL, *p = NULL;
1977
1978                for (e = active_branches; e; e = e->active_next_branch) {
1979                        if (e->last_commit < min_commit) {
1980                                p = l;
1981                                min_commit = e->last_commit;
1982                        }
1983                        l = e;
1984                }
1985
1986                if (p) {
1987                        e = p->active_next_branch;
1988                        p->active_next_branch = e->active_next_branch;
1989                } else {
1990                        e = active_branches;
1991                        active_branches = e->active_next_branch;
1992                }
1993                e->active = 0;
1994                e->active_next_branch = NULL;
1995                if (e->branch_tree.tree) {
1996                        release_tree_content_recursive(e->branch_tree.tree);
1997                        e->branch_tree.tree = NULL;
1998                }
1999                cur_active_branches--;
2000        }
2001}
2002
2003static void load_branch(struct branch *b)
2004{
2005        load_tree(&b->branch_tree);
2006        if (!b->active) {
2007                b->active = 1;
2008                b->active_next_branch = active_branches;
2009                active_branches = b;
2010                cur_active_branches++;
2011                branch_load_count++;
2012        }
2013}
2014
2015static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2016{
2017        unsigned char fanout = 0;
2018        while ((num_notes >>= 8))
2019                fanout++;
2020        return fanout;
2021}
2022
2023static void construct_path_with_fanout(const char *hex_sha1,
2024                unsigned char fanout, char *path)
2025{
2026        unsigned int i = 0, j = 0;
2027        if (fanout >= the_hash_algo->rawsz)
2028                die("Too large fanout (%u)", fanout);
2029        while (fanout) {
2030                path[i++] = hex_sha1[j++];
2031                path[i++] = hex_sha1[j++];
2032                path[i++] = '/';
2033                fanout--;
2034        }
2035        memcpy(path + i, hex_sha1 + j, the_hash_algo->hexsz - j);
2036        path[i + the_hash_algo->hexsz - j] = '\0';
2037}
2038
2039static uintmax_t do_change_note_fanout(
2040                struct tree_entry *orig_root, struct tree_entry *root,
2041                char *hex_oid, unsigned int hex_oid_len,
2042                char *fullpath, unsigned int fullpath_len,
2043                unsigned char fanout)
2044{
2045        struct tree_content *t;
2046        struct tree_entry *e, leaf;
2047        unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2048        uintmax_t num_notes = 0;
2049        struct object_id oid;
2050        char realpath[60];
2051
2052        if (!root->tree)
2053                load_tree(root);
2054        t = root->tree;
2055
2056        for (i = 0; t && i < t->entry_count; i++) {
2057                e = t->entries[i];
2058                tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2059                tmp_fullpath_len = fullpath_len;
2060
2061                /*
2062                 * We're interested in EITHER existing note entries (entries
2063                 * with exactly 40 hex chars in path, not including directory
2064                 * separators), OR directory entries that may contain note
2065                 * entries (with < 40 hex chars in path).
2066                 * Also, each path component in a note entry must be a multiple
2067                 * of 2 chars.
2068                 */
2069                if (!e->versions[1].mode ||
2070                    tmp_hex_oid_len > GIT_SHA1_HEXSZ ||
2071                    e->name->str_len % 2)
2072                        continue;
2073
2074                /* This _may_ be a note entry, or a subdir containing notes */
2075                memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2076                       e->name->str_len);
2077                if (tmp_fullpath_len)
2078                        fullpath[tmp_fullpath_len++] = '/';
2079                memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2080                       e->name->str_len);
2081                tmp_fullpath_len += e->name->str_len;
2082                fullpath[tmp_fullpath_len] = '\0';
2083
2084                if (tmp_hex_oid_len == GIT_SHA1_HEXSZ && !get_oid_hex(hex_oid, &oid)) {
2085                        /* This is a note entry */
2086                        if (fanout == 0xff) {
2087                                /* Counting mode, no rename */
2088                                num_notes++;
2089                                continue;
2090                        }
2091                        construct_path_with_fanout(hex_oid, fanout, realpath);
2092                        if (!strcmp(fullpath, realpath)) {
2093                                /* Note entry is in correct location */
2094                                num_notes++;
2095                                continue;
2096                        }
2097
2098                        /* Rename fullpath to realpath */
2099                        if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2100                                die("Failed to remove path %s", fullpath);
2101                        tree_content_set(orig_root, realpath,
2102                                &leaf.versions[1].oid,
2103                                leaf.versions[1].mode,
2104                                leaf.tree);
2105                } else if (S_ISDIR(e->versions[1].mode)) {
2106                        /* This is a subdir that may contain note entries */
2107                        num_notes += do_change_note_fanout(orig_root, e,
2108                                hex_oid, tmp_hex_oid_len,
2109                                fullpath, tmp_fullpath_len, fanout);
2110                }
2111
2112                /* The above may have reallocated the current tree_content */
2113                t = root->tree;
2114        }
2115        return num_notes;
2116}
2117
2118static uintmax_t change_note_fanout(struct tree_entry *root,
2119                unsigned char fanout)
2120{
2121        /*
2122         * The size of path is due to one slash between every two hex digits,
2123         * plus the terminating NUL.  Note that there is no slash at the end, so
2124         * the number of slashes is one less than half the number of hex
2125         * characters.
2126         */
2127        char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2128        return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2129}
2130
2131/*
2132 * Given a pointer into a string, parse a mark reference:
2133 *
2134 *   idnum ::= ':' bigint;
2135 *
2136 * Return the first character after the value in *endptr.
2137 *
2138 * Complain if the following character is not what is expected,
2139 * either a space or end of the string.
2140 */
2141static uintmax_t parse_mark_ref(const char *p, char **endptr)
2142{
2143        uintmax_t mark;
2144
2145        assert(*p == ':');
2146        p++;
2147        mark = strtoumax(p, endptr, 10);
2148        if (*endptr == p)
2149                die("No value after ':' in mark: %s", command_buf.buf);
2150        return mark;
2151}
2152
2153/*
2154 * Parse the mark reference, and complain if this is not the end of
2155 * the string.
2156 */
2157static uintmax_t parse_mark_ref_eol(const char *p)
2158{
2159        char *end;
2160        uintmax_t mark;
2161
2162        mark = parse_mark_ref(p, &end);
2163        if (*end != '\0')
2164                die("Garbage after mark: %s", command_buf.buf);
2165        return mark;
2166}
2167
2168/*
2169 * Parse the mark reference, demanding a trailing space.  Return a
2170 * pointer to the space.
2171 */
2172static uintmax_t parse_mark_ref_space(const char **p)
2173{
2174        uintmax_t mark;
2175        char *end;
2176
2177        mark = parse_mark_ref(*p, &end);
2178        if (*end++ != ' ')
2179                die("Missing space after mark: %s", command_buf.buf);
2180        *p = end;
2181        return mark;
2182}
2183
2184static void file_change_m(const char *p, struct branch *b)
2185{
2186        static struct strbuf uq = STRBUF_INIT;
2187        const char *endp;
2188        struct object_entry *oe;
2189        struct object_id oid;
2190        uint16_t mode, inline_data = 0;
2191
2192        p = get_mode(p, &mode);
2193        if (!p)
2194                die("Corrupt mode: %s", command_buf.buf);
2195        switch (mode) {
2196        case 0644:
2197        case 0755:
2198                mode |= S_IFREG;
2199        case S_IFREG | 0644:
2200        case S_IFREG | 0755:
2201        case S_IFLNK:
2202        case S_IFDIR:
2203        case S_IFGITLINK:
2204                /* ok */
2205                break;
2206        default:
2207                die("Corrupt mode: %s", command_buf.buf);
2208        }
2209
2210        if (*p == ':') {
2211                oe = find_mark(parse_mark_ref_space(&p));
2212                oidcpy(&oid, &oe->idx.oid);
2213        } else if (skip_prefix(p, "inline ", &p)) {
2214                inline_data = 1;
2215                oe = NULL; /* not used with inline_data, but makes gcc happy */
2216        } else {
2217                if (parse_oid_hex(p, &oid, &p))
2218                        die("Invalid dataref: %s", command_buf.buf);
2219                oe = find_object(&oid);
2220                if (*p++ != ' ')
2221                        die("Missing space after SHA1: %s", command_buf.buf);
2222        }
2223
2224        strbuf_reset(&uq);
2225        if (!unquote_c_style(&uq, p, &endp)) {
2226                if (*endp)
2227                        die("Garbage after path in: %s", command_buf.buf);
2228                p = uq.buf;
2229        }
2230
2231        /* Git does not track empty, non-toplevel directories. */
2232        if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2233                tree_content_remove(&b->branch_tree, p, NULL, 0);
2234                return;
2235        }
2236
2237        if (S_ISGITLINK(mode)) {
2238                if (inline_data)
2239                        die("Git links cannot be specified 'inline': %s",
2240                                command_buf.buf);
2241                else if (oe) {
2242                        if (oe->type != OBJ_COMMIT)
2243                                die("Not a commit (actually a %s): %s",
2244                                        type_name(oe->type), command_buf.buf);
2245                }
2246                /*
2247                 * Accept the sha1 without checking; it expected to be in
2248                 * another repository.
2249                 */
2250        } else if (inline_data) {
2251                if (S_ISDIR(mode))
2252                        die("Directories cannot be specified 'inline': %s",
2253                                command_buf.buf);
2254                if (p != uq.buf) {
2255                        strbuf_addstr(&uq, p);
2256                        p = uq.buf;
2257                }
2258                read_next_command();
2259                parse_and_store_blob(&last_blob, &oid, 0);
2260        } else {
2261                enum object_type expected = S_ISDIR(mode) ?
2262                                                OBJ_TREE: OBJ_BLOB;
2263                enum object_type type = oe ? oe->type :
2264                                        oid_object_info(the_repository, &oid,
2265                                                        NULL);
2266                if (type < 0)
2267                        die("%s not found: %s",
2268                                        S_ISDIR(mode) ?  "Tree" : "Blob",
2269                                        command_buf.buf);
2270                if (type != expected)
2271                        die("Not a %s (actually a %s): %s",
2272                                type_name(expected), type_name(type),
2273                                command_buf.buf);
2274        }
2275
2276        if (!*p) {
2277                tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2278                return;
2279        }
2280        tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2281}
2282
2283static void file_change_d(const char *p, struct branch *b)
2284{
2285        static struct strbuf uq = STRBUF_INIT;
2286        const char *endp;
2287
2288        strbuf_reset(&uq);
2289        if (!unquote_c_style(&uq, p, &endp)) {
2290                if (*endp)
2291                        die("Garbage after path in: %s", command_buf.buf);
2292                p = uq.buf;
2293        }
2294        tree_content_remove(&b->branch_tree, p, NULL, 1);
2295}
2296
2297static void file_change_cr(const char *s, struct branch *b, int rename)
2298{
2299        const char *d;
2300        static struct strbuf s_uq = STRBUF_INIT;
2301        static struct strbuf d_uq = STRBUF_INIT;
2302        const char *endp;
2303        struct tree_entry leaf;
2304
2305        strbuf_reset(&s_uq);
2306        if (!unquote_c_style(&s_uq, s, &endp)) {
2307                if (*endp != ' ')
2308                        die("Missing space after source: %s", command_buf.buf);
2309        } else {
2310                endp = strchr(s, ' ');
2311                if (!endp)
2312                        die("Missing space after source: %s", command_buf.buf);
2313                strbuf_add(&s_uq, s, endp - s);
2314        }
2315        s = s_uq.buf;
2316
2317        endp++;
2318        if (!*endp)
2319                die("Missing dest: %s", command_buf.buf);
2320
2321        d = endp;
2322        strbuf_reset(&d_uq);
2323        if (!unquote_c_style(&d_uq, d, &endp)) {
2324                if (*endp)
2325                        die("Garbage after dest in: %s", command_buf.buf);
2326                d = d_uq.buf;
2327        }
2328
2329        memset(&leaf, 0, sizeof(leaf));
2330        if (rename)
2331                tree_content_remove(&b->branch_tree, s, &leaf, 1);
2332        else
2333                tree_content_get(&b->branch_tree, s, &leaf, 1);
2334        if (!leaf.versions[1].mode)
2335                die("Path %s not in branch", s);
2336        if (!*d) {      /* C "path/to/subdir" "" */
2337                tree_content_replace(&b->branch_tree,
2338                        &leaf.versions[1].oid,
2339                        leaf.versions[1].mode,
2340                        leaf.tree);
2341                return;
2342        }
2343        tree_content_set(&b->branch_tree, d,
2344                &leaf.versions[1].oid,
2345                leaf.versions[1].mode,
2346                leaf.tree);
2347}
2348
2349static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2350{
2351        static struct strbuf uq = STRBUF_INIT;
2352        struct object_entry *oe;
2353        struct branch *s;
2354        struct object_id oid, commit_oid;
2355        char path[60];
2356        uint16_t inline_data = 0;
2357        unsigned char new_fanout;
2358
2359        /*
2360         * When loading a branch, we don't traverse its tree to count the real
2361         * number of notes (too expensive to do this for all non-note refs).
2362         * This means that recently loaded notes refs might incorrectly have
2363         * b->num_notes == 0, and consequently, old_fanout might be wrong.
2364         *
2365         * Fix this by traversing the tree and counting the number of notes
2366         * when b->num_notes == 0. If the notes tree is truly empty, the
2367         * calculation should not take long.
2368         */
2369        if (b->num_notes == 0 && *old_fanout == 0) {
2370                /* Invoke change_note_fanout() in "counting mode". */
2371                b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2372                *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2373        }
2374
2375        /* Now parse the notemodify command. */
2376        /* <dataref> or 'inline' */
2377        if (*p == ':') {
2378                oe = find_mark(parse_mark_ref_space(&p));
2379                oidcpy(&oid, &oe->idx.oid);
2380        } else if (skip_prefix(p, "inline ", &p)) {
2381                inline_data = 1;
2382                oe = NULL; /* not used with inline_data, but makes gcc happy */
2383        } else {
2384                if (parse_oid_hex(p, &oid, &p))
2385                        die("Invalid dataref: %s", command_buf.buf);
2386                oe = find_object(&oid);
2387                if (*p++ != ' ')
2388                        die("Missing space after SHA1: %s", command_buf.buf);
2389        }
2390
2391        /* <commit-ish> */
2392        s = lookup_branch(p);
2393        if (s) {
2394                if (is_null_oid(&s->oid))
2395                        die("Can't add a note on empty branch.");
2396                oidcpy(&commit_oid, &s->oid);
2397        } else if (*p == ':') {
2398                uintmax_t commit_mark = parse_mark_ref_eol(p);
2399                struct object_entry *commit_oe = find_mark(commit_mark);
2400                if (commit_oe->type != OBJ_COMMIT)
2401                        die("Mark :%" PRIuMAX " not a commit", commit_mark);
2402                oidcpy(&commit_oid, &commit_oe->idx.oid);
2403        } else if (!get_oid(p, &commit_oid)) {
2404                unsigned long size;
2405                char *buf = read_object_with_reference(&commit_oid,
2406                                                       commit_type, &size,
2407                                                       &commit_oid);
2408                if (!buf || size < 46)
2409                        die("Not a valid commit: %s", p);
2410                free(buf);
2411        } else
2412                die("Invalid ref name or SHA1 expression: %s", p);
2413
2414        if (inline_data) {
2415                if (p != uq.buf) {
2416                        strbuf_addstr(&uq, p);
2417                        p = uq.buf;
2418                }
2419                read_next_command();
2420                parse_and_store_blob(&last_blob, &oid, 0);
2421        } else if (oe) {
2422                if (oe->type != OBJ_BLOB)
2423                        die("Not a blob (actually a %s): %s",
2424                                type_name(oe->type), command_buf.buf);
2425        } else if (!is_null_oid(&oid)) {
2426                enum object_type type = oid_object_info(the_repository, &oid,
2427                                                        NULL);
2428                if (type < 0)
2429                        die("Blob not found: %s", command_buf.buf);
2430                if (type != OBJ_BLOB)
2431                        die("Not a blob (actually a %s): %s",
2432                            type_name(type), command_buf.buf);
2433        }
2434
2435        construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2436        if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2437                b->num_notes--;
2438
2439        if (is_null_oid(&oid))
2440                return; /* nothing to insert */
2441
2442        b->num_notes++;
2443        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2444        construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2445        tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2446}
2447
2448static void file_change_deleteall(struct branch *b)
2449{
2450        release_tree_content_recursive(b->branch_tree.tree);
2451        oidclr(&b->branch_tree.versions[0].oid);
2452        oidclr(&b->branch_tree.versions[1].oid);
2453        load_tree(&b->branch_tree);
2454        b->num_notes = 0;
2455}
2456
2457static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2458{
2459        if (!buf || size < GIT_SHA1_HEXSZ + 6)
2460                die("Not a valid commit: %s", oid_to_hex(&b->oid));
2461        if (memcmp("tree ", buf, 5)
2462                || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2463                die("The commit %s is corrupt", oid_to_hex(&b->oid));
2464        oidcpy(&b->branch_tree.versions[0].oid,
2465               &b->branch_tree.versions[1].oid);
2466}
2467
2468static void parse_from_existing(struct branch *b)
2469{
2470        if (is_null_oid(&b->oid)) {
2471                oidclr(&b->branch_tree.versions[0].oid);
2472                oidclr(&b->branch_tree.versions[1].oid);
2473        } else {
2474                unsigned long size;
2475                char *buf;
2476
2477                buf = read_object_with_reference(&b->oid, commit_type, &size,
2478                                                 &b->oid);
2479                parse_from_commit(b, buf, size);
2480                free(buf);
2481        }
2482}
2483
2484static int parse_from(struct branch *b)
2485{
2486        const char *from;
2487        struct branch *s;
2488        struct object_id oid;
2489
2490        if (!skip_prefix(command_buf.buf, "from ", &from))
2491                return 0;
2492
2493        oidcpy(&oid, &b->branch_tree.versions[1].oid);
2494
2495        s = lookup_branch(from);
2496        if (b == s)
2497                die("Can't create a branch from itself: %s", b->name);
2498        else if (s) {
2499                struct object_id *t = &s->branch_tree.versions[1].oid;
2500                oidcpy(&b->oid, &s->oid);
2501                oidcpy(&b->branch_tree.versions[0].oid, t);
2502                oidcpy(&b->branch_tree.versions[1].oid, t);
2503        } else if (*from == ':') {
2504                uintmax_t idnum = parse_mark_ref_eol(from);
2505                struct object_entry *oe = find_mark(idnum);
2506                if (oe->type != OBJ_COMMIT)
2507                        die("Mark :%" PRIuMAX " not a commit", idnum);
2508                if (!oideq(&b->oid, &oe->idx.oid)) {
2509                        oidcpy(&b->oid, &oe->idx.oid);
2510                        if (oe->pack_id != MAX_PACK_ID) {
2511                                unsigned long size;
2512                                char *buf = gfi_unpack_entry(oe, &size);
2513                                parse_from_commit(b, buf, size);
2514                                free(buf);
2515                        } else
2516                                parse_from_existing(b);
2517                }
2518        } else if (!get_oid(from, &b->oid)) {
2519                parse_from_existing(b);
2520                if (is_null_oid(&b->oid))
2521                        b->delete = 1;
2522        }
2523        else
2524                die("Invalid ref name or SHA1 expression: %s", from);
2525
2526        if (b->branch_tree.tree && !oideq(&oid, &b->branch_tree.versions[1].oid)) {
2527                release_tree_content_recursive(b->branch_tree.tree);
2528                b->branch_tree.tree = NULL;
2529        }
2530
2531        read_next_command();
2532        return 1;
2533}
2534
2535static struct hash_list *parse_merge(unsigned int *count)
2536{
2537        struct hash_list *list = NULL, **tail = &list, *n;
2538        const char *from;
2539        struct branch *s;
2540
2541        *count = 0;
2542        while (skip_prefix(command_buf.buf, "merge ", &from)) {
2543                n = xmalloc(sizeof(*n));
2544                s = lookup_branch(from);
2545                if (s)
2546                        oidcpy(&n->oid, &s->oid);
2547                else if (*from == ':') {
2548                        uintmax_t idnum = parse_mark_ref_eol(from);
2549                        struct object_entry *oe = find_mark(idnum);
2550                        if (oe->type != OBJ_COMMIT)
2551                                die("Mark :%" PRIuMAX " not a commit", idnum);
2552                        oidcpy(&n->oid, &oe->idx.oid);
2553                } else if (!get_oid(from, &n->oid)) {
2554                        unsigned long size;
2555                        char *buf = read_object_with_reference(&n->oid,
2556                                                               commit_type,
2557                                                               &size, &n->oid);
2558                        if (!buf || size < 46)
2559                                die("Not a valid commit: %s", from);
2560                        free(buf);
2561                } else
2562                        die("Invalid ref name or SHA1 expression: %s", from);
2563
2564                n->next = NULL;
2565                *tail = n;
2566                tail = &n->next;
2567
2568                (*count)++;
2569                read_next_command();
2570        }
2571        return list;
2572}
2573
2574static void parse_new_commit(const char *arg)
2575{
2576        static struct strbuf msg = STRBUF_INIT;
2577        struct branch *b;
2578        char *author = NULL;
2579        char *committer = NULL;
2580        struct hash_list *merge_list = NULL;
2581        unsigned int merge_count;
2582        unsigned char prev_fanout, new_fanout;
2583        const char *v;
2584
2585        b = lookup_branch(arg);
2586        if (!b)
2587                b = new_branch(arg);
2588
2589        read_next_command();
2590        parse_mark();
2591        parse_original_identifier();
2592        if (skip_prefix(command_buf.buf, "author ", &v)) {
2593                author = parse_ident(v);
2594                read_next_command();
2595        }
2596        if (skip_prefix(command_buf.buf, "committer ", &v)) {
2597                committer = parse_ident(v);
2598                read_next_command();
2599        }
2600        if (!committer)
2601                die("Expected committer but didn't get one");
2602        parse_data(&msg, 0, NULL);
2603        read_next_command();
2604        parse_from(b);
2605        merge_list = parse_merge(&merge_count);
2606
2607        /* ensure the branch is active/loaded */
2608        if (!b->branch_tree.tree || !max_active_branches) {
2609                unload_one_branch();
2610                load_branch(b);
2611        }
2612
2613        prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2614
2615        /* file_change* */
2616        while (command_buf.len > 0) {
2617                if (skip_prefix(command_buf.buf, "M ", &v))
2618                        file_change_m(v, b);
2619                else if (skip_prefix(command_buf.buf, "D ", &v))
2620                        file_change_d(v, b);
2621                else if (skip_prefix(command_buf.buf, "R ", &v))
2622                        file_change_cr(v, b, 1);
2623                else if (skip_prefix(command_buf.buf, "C ", &v))
2624                        file_change_cr(v, b, 0);
2625                else if (skip_prefix(command_buf.buf, "N ", &v))
2626                        note_change_n(v, b, &prev_fanout);
2627                else if (!strcmp("deleteall", command_buf.buf))
2628                        file_change_deleteall(b);
2629                else if (skip_prefix(command_buf.buf, "ls ", &v))
2630                        parse_ls(v, b);
2631                else {
2632                        unread_command_buf = 1;
2633                        break;
2634                }
2635                if (read_next_command() == EOF)
2636                        break;
2637        }
2638
2639        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2640        if (new_fanout != prev_fanout)
2641                b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2642
2643        /* build the tree and the commit */
2644        store_tree(&b->branch_tree);
2645        oidcpy(&b->branch_tree.versions[0].oid,
2646               &b->branch_tree.versions[1].oid);
2647
2648        strbuf_reset(&new_data);
2649        strbuf_addf(&new_data, "tree %s\n",
2650                oid_to_hex(&b->branch_tree.versions[1].oid));
2651        if (!is_null_oid(&b->oid))
2652                strbuf_addf(&new_data, "parent %s\n",
2653                            oid_to_hex(&b->oid));
2654        while (merge_list) {
2655                struct hash_list *next = merge_list->next;
2656                strbuf_addf(&new_data, "parent %s\n",
2657                            oid_to_hex(&merge_list->oid));
2658                free(merge_list);
2659                merge_list = next;
2660        }
2661        strbuf_addf(&new_data,
2662                "author %s\n"
2663                "committer %s\n"
2664                "\n",
2665                author ? author : committer, committer);
2666        strbuf_addbuf(&new_data, &msg);
2667        free(author);
2668        free(committer);
2669
2670        if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2671                b->pack_id = pack_id;
2672        b->last_commit = object_count_by_type[OBJ_COMMIT];
2673}
2674
2675static void parse_new_tag(const char *arg)
2676{
2677        static struct strbuf msg = STRBUF_INIT;
2678        const char *from;
2679        char *tagger;
2680        struct branch *s;
2681        struct tag *t;
2682        uintmax_t from_mark = 0;
2683        struct object_id oid;
2684        enum object_type type;
2685        const char *v;
2686
2687        t = mem_pool_alloc(&fi_mem_pool, sizeof(struct tag));
2688        memset(t, 0, sizeof(struct tag));
2689        t->name = pool_strdup(arg);
2690        if (last_tag)
2691                last_tag->next_tag = t;
2692        else
2693                first_tag = t;
2694        last_tag = t;
2695        read_next_command();
2696
2697        /* from ... */
2698        if (!skip_prefix(command_buf.buf, "from ", &from))
2699                die("Expected from command, got %s", command_buf.buf);
2700        s = lookup_branch(from);
2701        if (s) {
2702                if (is_null_oid(&s->oid))
2703                        die("Can't tag an empty branch.");
2704                oidcpy(&oid, &s->oid);
2705                type = OBJ_COMMIT;
2706        } else if (*from == ':') {
2707                struct object_entry *oe;
2708                from_mark = parse_mark_ref_eol(from);
2709                oe = find_mark(from_mark);
2710                type = oe->type;
2711                oidcpy(&oid, &oe->idx.oid);
2712        } else if (!get_oid(from, &oid)) {
2713                struct object_entry *oe = find_object(&oid);
2714                if (!oe) {
2715                        type = oid_object_info(the_repository, &oid, NULL);
2716                        if (type < 0)
2717                                die("Not a valid object: %s", from);
2718                } else
2719                        type = oe->type;
2720        } else
2721                die("Invalid ref name or SHA1 expression: %s", from);
2722        read_next_command();
2723
2724        /* original-oid ... */
2725        parse_original_identifier();
2726
2727        /* tagger ... */
2728        if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2729                tagger = parse_ident(v);
2730                read_next_command();
2731        } else
2732                tagger = NULL;
2733
2734        /* tag payload/message */
2735        parse_data(&msg, 0, NULL);
2736
2737        /* build the tag object */
2738        strbuf_reset(&new_data);
2739
2740        strbuf_addf(&new_data,
2741                    "object %s\n"
2742                    "type %s\n"
2743                    "tag %s\n",
2744                    oid_to_hex(&oid), type_name(type), t->name);
2745        if (tagger)
2746                strbuf_addf(&new_data,
2747                            "tagger %s\n", tagger);
2748        strbuf_addch(&new_data, '\n');
2749        strbuf_addbuf(&new_data, &msg);
2750        free(tagger);
2751
2752        if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, 0))
2753                t->pack_id = MAX_PACK_ID;
2754        else
2755                t->pack_id = pack_id;
2756}
2757
2758static void parse_reset_branch(const char *arg)
2759{
2760        struct branch *b;
2761
2762        b = lookup_branch(arg);
2763        if (b) {
2764                oidclr(&b->oid);
2765                oidclr(&b->branch_tree.versions[0].oid);
2766                oidclr(&b->branch_tree.versions[1].oid);
2767                if (b->branch_tree.tree) {
2768                        release_tree_content_recursive(b->branch_tree.tree);
2769                        b->branch_tree.tree = NULL;
2770                }
2771        }
2772        else
2773                b = new_branch(arg);
2774        read_next_command();
2775        parse_from(b);
2776        if (command_buf.len > 0)
2777                unread_command_buf = 1;
2778}
2779
2780static void cat_blob_write(const char *buf, unsigned long size)
2781{
2782        if (write_in_full(cat_blob_fd, buf, size) < 0)
2783                die_errno("Write to frontend failed");
2784}
2785
2786static void cat_blob(struct object_entry *oe, struct object_id *oid)
2787{
2788        struct strbuf line = STRBUF_INIT;
2789        unsigned long size;
2790        enum object_type type = 0;
2791        char *buf;
2792
2793        if (!oe || oe->pack_id == MAX_PACK_ID) {
2794                buf = read_object_file(oid, &type, &size);
2795        } else {
2796                type = oe->type;
2797                buf = gfi_unpack_entry(oe, &size);
2798        }
2799
2800        /*
2801         * Output based on batch_one_object() from cat-file.c.
2802         */
2803        if (type <= 0) {
2804                strbuf_reset(&line);
2805                strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2806                cat_blob_write(line.buf, line.len);
2807                strbuf_release(&line);
2808                free(buf);
2809                return;
2810        }
2811        if (!buf)
2812                die("Can't read object %s", oid_to_hex(oid));
2813        if (type != OBJ_BLOB)
2814                die("Object %s is a %s but a blob was expected.",
2815                    oid_to_hex(oid), type_name(type));
2816        strbuf_reset(&line);
2817        strbuf_addf(&line, "%s %s %"PRIuMAX"\n", oid_to_hex(oid),
2818                    type_name(type), (uintmax_t)size);
2819        cat_blob_write(line.buf, line.len);
2820        strbuf_release(&line);
2821        cat_blob_write(buf, size);
2822        cat_blob_write("\n", 1);
2823        if (oe && oe->pack_id == pack_id) {
2824                last_blob.offset = oe->idx.offset;
2825                strbuf_attach(&last_blob.data, buf, size, size);
2826                last_blob.depth = oe->depth;
2827        } else
2828                free(buf);
2829}
2830
2831static void parse_get_mark(const char *p)
2832{
2833        struct object_entry *oe;
2834        char output[GIT_MAX_HEXSZ + 2];
2835
2836        /* get-mark SP <object> LF */
2837        if (*p != ':')
2838                die("Not a mark: %s", p);
2839
2840        oe = find_mark(parse_mark_ref_eol(p));
2841        if (!oe)
2842                die("Unknown mark: %s", command_buf.buf);
2843
2844        xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
2845        cat_blob_write(output, GIT_SHA1_HEXSZ + 1);
2846}
2847
2848static void parse_cat_blob(const char *p)
2849{
2850        struct object_entry *oe;
2851        struct object_id oid;
2852
2853        /* cat-blob SP <object> LF */
2854        if (*p == ':') {
2855                oe = find_mark(parse_mark_ref_eol(p));
2856                if (!oe)
2857                        die("Unknown mark: %s", command_buf.buf);
2858                oidcpy(&oid, &oe->idx.oid);
2859        } else {
2860                if (parse_oid_hex(p, &oid, &p))
2861                        die("Invalid dataref: %s", command_buf.buf);
2862                if (*p)
2863                        die("Garbage after SHA1: %s", command_buf.buf);
2864                oe = find_object(&oid);
2865        }
2866
2867        cat_blob(oe, &oid);
2868}
2869
2870static struct object_entry *dereference(struct object_entry *oe,
2871                                        struct object_id *oid)
2872{
2873        unsigned long size;
2874        char *buf = NULL;
2875        if (!oe) {
2876                enum object_type type = oid_object_info(the_repository, oid,
2877                                                        NULL);
2878                if (type < 0)
2879                        die("object not found: %s", oid_to_hex(oid));
2880                /* cache it! */
2881                oe = insert_object(oid);
2882                oe->type = type;
2883                oe->pack_id = MAX_PACK_ID;
2884                oe->idx.offset = 1;
2885        }
2886        switch (oe->type) {
2887        case OBJ_TREE:  /* easy case. */
2888                return oe;
2889        case OBJ_COMMIT:
2890        case OBJ_TAG:
2891                break;
2892        default:
2893                die("Not a tree-ish: %s", command_buf.buf);
2894        }
2895
2896        if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
2897                buf = gfi_unpack_entry(oe, &size);
2898        } else {
2899                enum object_type unused;
2900                buf = read_object_file(oid, &unused, &size);
2901        }
2902        if (!buf)
2903                die("Can't load object %s", oid_to_hex(oid));
2904
2905        /* Peel one layer. */
2906        switch (oe->type) {
2907        case OBJ_TAG:
2908                if (size < GIT_SHA1_HEXSZ + strlen("object ") ||
2909                    get_oid_hex(buf + strlen("object "), oid))
2910                        die("Invalid SHA1 in tag: %s", command_buf.buf);
2911                break;
2912        case OBJ_COMMIT:
2913                if (size < GIT_SHA1_HEXSZ + strlen("tree ") ||
2914                    get_oid_hex(buf + strlen("tree "), oid))
2915                        die("Invalid SHA1 in commit: %s", command_buf.buf);
2916        }
2917
2918        free(buf);
2919        return find_object(oid);
2920}
2921
2922static struct object_entry *parse_treeish_dataref(const char **p)
2923{
2924        struct object_id oid;
2925        struct object_entry *e;
2926
2927        if (**p == ':') {       /* <mark> */
2928                e = find_mark(parse_mark_ref_space(p));
2929                if (!e)
2930                        die("Unknown mark: %s", command_buf.buf);
2931                oidcpy(&oid, &e->idx.oid);
2932        } else {        /* <sha1> */
2933                if (parse_oid_hex(*p, &oid, p))
2934                        die("Invalid dataref: %s", command_buf.buf);
2935                e = find_object(&oid);
2936                if (*(*p)++ != ' ')
2937                        die("Missing space after tree-ish: %s", command_buf.buf);
2938        }
2939
2940        while (!e || e->type != OBJ_TREE)
2941                e = dereference(e, &oid);
2942        return e;
2943}
2944
2945static void print_ls(int mode, const unsigned char *sha1, const char *path)
2946{
2947        static struct strbuf line = STRBUF_INIT;
2948
2949        /* See show_tree(). */
2950        const char *type =
2951                S_ISGITLINK(mode) ? commit_type :
2952                S_ISDIR(mode) ? tree_type :
2953                blob_type;
2954
2955        if (!mode) {
2956                /* missing SP path LF */
2957                strbuf_reset(&line);
2958                strbuf_addstr(&line, "missing ");
2959                quote_c_style(path, &line, NULL, 0);
2960                strbuf_addch(&line, '\n');
2961        } else {
2962                /* mode SP type SP object_name TAB path LF */
2963                strbuf_reset(&line);
2964                strbuf_addf(&line, "%06o %s %s\t",
2965                                mode & ~NO_DELTA, type, sha1_to_hex(sha1));
2966                quote_c_style(path, &line, NULL, 0);
2967                strbuf_addch(&line, '\n');
2968        }
2969        cat_blob_write(line.buf, line.len);
2970}
2971
2972static void parse_ls(const char *p, struct branch *b)
2973{
2974        struct tree_entry *root = NULL;
2975        struct tree_entry leaf = {NULL};
2976
2977        /* ls SP (<tree-ish> SP)? <path> */
2978        if (*p == '"') {
2979                if (!b)
2980                        die("Not in a commit: %s", command_buf.buf);
2981                root = &b->branch_tree;
2982        } else {
2983                struct object_entry *e = parse_treeish_dataref(&p);
2984                root = new_tree_entry();
2985                oidcpy(&root->versions[1].oid, &e->idx.oid);
2986                if (!is_null_oid(&root->versions[1].oid))
2987                        root->versions[1].mode = S_IFDIR;
2988                load_tree(root);
2989        }
2990        if (*p == '"') {
2991                static struct strbuf uq = STRBUF_INIT;
2992                const char *endp;
2993                strbuf_reset(&uq);
2994                if (unquote_c_style(&uq, p, &endp))
2995                        die("Invalid path: %s", command_buf.buf);
2996                if (*endp)
2997                        die("Garbage after path in: %s", command_buf.buf);
2998                p = uq.buf;
2999        }
3000        tree_content_get(root, p, &leaf, 1);
3001        /*
3002         * A directory in preparation would have a sha1 of zero
3003         * until it is saved.  Save, for simplicity.
3004         */
3005        if (S_ISDIR(leaf.versions[1].mode))
3006                store_tree(&leaf);
3007
3008        print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3009        if (leaf.tree)
3010                release_tree_content_recursive(leaf.tree);
3011        if (!b || root != &b->branch_tree)
3012                release_tree_entry(root);
3013}
3014
3015static void checkpoint(void)
3016{
3017        checkpoint_requested = 0;
3018        if (object_count) {
3019                cycle_packfile();
3020        }
3021        dump_branches();
3022        dump_tags();
3023        dump_marks();
3024}
3025
3026static void parse_checkpoint(void)
3027{
3028        checkpoint_requested = 1;
3029        skip_optional_lf();
3030}
3031
3032static void parse_progress(void)
3033{
3034        fwrite(command_buf.buf, 1, command_buf.len, stdout);
3035        fputc('\n', stdout);
3036        fflush(stdout);
3037        skip_optional_lf();
3038}
3039
3040static char* make_fast_import_path(const char *path)
3041{
3042        if (!relative_marks_paths || is_absolute_path(path))
3043                return xstrdup(path);
3044        return git_pathdup("info/fast-import/%s", path);
3045}
3046
3047static void option_import_marks(const char *marks,
3048                                        int from_stream, int ignore_missing)
3049{
3050        if (import_marks_file) {
3051                if (from_stream)
3052                        die("Only one import-marks command allowed per stream");
3053
3054                /* read previous mark file */
3055                if(!import_marks_file_from_stream)
3056                        read_marks();
3057        }
3058
3059        import_marks_file = make_fast_import_path(marks);
3060        safe_create_leading_directories_const(import_marks_file);
3061        import_marks_file_from_stream = from_stream;
3062        import_marks_file_ignore_missing = ignore_missing;
3063}
3064
3065static void option_date_format(const char *fmt)
3066{
3067        if (!strcmp(fmt, "raw"))
3068                whenspec = WHENSPEC_RAW;
3069        else if (!strcmp(fmt, "rfc2822"))
3070                whenspec = WHENSPEC_RFC2822;
3071        else if (!strcmp(fmt, "now"))
3072                whenspec = WHENSPEC_NOW;
3073        else
3074                die("unknown --date-format argument %s", fmt);
3075}
3076
3077static unsigned long ulong_arg(const char *option, const char *arg)
3078{
3079        char *endptr;
3080        unsigned long rv = strtoul(arg, &endptr, 0);
3081        if (strchr(arg, '-') || endptr == arg || *endptr)
3082                die("%s: argument must be a non-negative integer", option);
3083        return rv;
3084}
3085
3086static void option_depth(const char *depth)
3087{
3088        max_depth = ulong_arg("--depth", depth);
3089        if (max_depth > MAX_DEPTH)
3090                die("--depth cannot exceed %u", MAX_DEPTH);
3091}
3092
3093static void option_active_branches(const char *branches)
3094{
3095        max_active_branches = ulong_arg("--active-branches", branches);
3096}
3097
3098static void option_export_marks(const char *marks)
3099{
3100        export_marks_file = make_fast_import_path(marks);
3101        safe_create_leading_directories_const(export_marks_file);
3102}
3103
3104static void option_cat_blob_fd(const char *fd)
3105{
3106        unsigned long n = ulong_arg("--cat-blob-fd", fd);
3107        if (n > (unsigned long) INT_MAX)
3108                die("--cat-blob-fd cannot exceed %d", INT_MAX);
3109        cat_blob_fd = (int) n;
3110}
3111
3112static void option_export_pack_edges(const char *edges)
3113{
3114        if (pack_edges)
3115                fclose(pack_edges);
3116        pack_edges = xfopen(edges, "a");
3117}
3118
3119static int parse_one_option(const char *option)
3120{
3121        if (skip_prefix(option, "max-pack-size=", &option)) {
3122                unsigned long v;
3123                if (!git_parse_ulong(option, &v))
3124                        return 0;
3125                if (v < 8192) {
3126                        warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3127                        v *= 1024 * 1024;
3128                } else if (v < 1024 * 1024) {
3129                        warning("minimum max-pack-size is 1 MiB");
3130                        v = 1024 * 1024;
3131                }
3132                max_packsize = v;
3133        } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3134                unsigned long v;
3135                if (!git_parse_ulong(option, &v))
3136                        return 0;
3137                big_file_threshold = v;
3138        } else if (skip_prefix(option, "depth=", &option)) {
3139                option_depth(option);
3140        } else if (skip_prefix(option, "active-branches=", &option)) {
3141                option_active_branches(option);
3142        } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3143                option_export_pack_edges(option);
3144        } else if (starts_with(option, "quiet")) {
3145                show_stats = 0;
3146        } else if (starts_with(option, "stats")) {
3147                show_stats = 1;
3148        } else {
3149                return 0;
3150        }
3151
3152        return 1;
3153}
3154
3155static int parse_one_feature(const char *feature, int from_stream)
3156{
3157        const char *arg;
3158
3159        if (skip_prefix(feature, "date-format=", &arg)) {
3160                option_date_format(arg);
3161        } else if (skip_prefix(feature, "import-marks=", &arg)) {
3162                option_import_marks(arg, from_stream, 0);
3163        } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3164                option_import_marks(arg, from_stream, 1);
3165        } else if (skip_prefix(feature, "export-marks=", &arg)) {
3166                option_export_marks(arg);
3167        } else if (!strcmp(feature, "get-mark")) {
3168                ; /* Don't die - this feature is supported */
3169        } else if (!strcmp(feature, "cat-blob")) {
3170                ; /* Don't die - this feature is supported */
3171        } else if (!strcmp(feature, "relative-marks")) {
3172                relative_marks_paths = 1;
3173        } else if (!strcmp(feature, "no-relative-marks")) {
3174                relative_marks_paths = 0;
3175        } else if (!strcmp(feature, "done")) {
3176                require_explicit_termination = 1;
3177        } else if (!strcmp(feature, "force")) {
3178                force_update = 1;
3179        } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3180                ; /* do nothing; we have the feature */
3181        } else {
3182                return 0;
3183        }
3184
3185        return 1;
3186}
3187
3188static void parse_feature(const char *feature)
3189{
3190        if (seen_data_command)
3191                die("Got feature command '%s' after data command", feature);
3192
3193        if (parse_one_feature(feature, 1))
3194                return;
3195
3196        die("This version of fast-import does not support feature %s.", feature);
3197}
3198
3199static void parse_option(const char *option)
3200{
3201        if (seen_data_command)
3202                die("Got option command '%s' after data command", option);
3203
3204        if (parse_one_option(option))
3205                return;
3206
3207        die("This version of fast-import does not support option: %s", option);
3208}
3209
3210static void git_pack_config(void)
3211{
3212        int indexversion_value;
3213        int limit;
3214        unsigned long packsizelimit_value;
3215
3216        if (!git_config_get_ulong("pack.depth", &max_depth)) {
3217                if (max_depth > MAX_DEPTH)
3218                        max_depth = MAX_DEPTH;
3219        }
3220        if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3221                pack_idx_opts.version = indexversion_value;
3222                if (pack_idx_opts.version > 2)
3223                        git_die_config("pack.indexversion",
3224                                        "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3225        }
3226        if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3227                max_packsize = packsizelimit_value;
3228
3229        if (!git_config_get_int("fastimport.unpacklimit", &limit))
3230                unpack_limit = limit;
3231        else if (!git_config_get_int("transfer.unpacklimit", &limit))
3232                unpack_limit = limit;
3233
3234        git_config(git_default_config, NULL);
3235}
3236
3237static const char fast_import_usage[] =
3238"git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3239
3240static void parse_argv(void)
3241{
3242        unsigned int i;
3243
3244        for (i = 1; i < global_argc; i++) {
3245                const char *a = global_argv[i];
3246
3247                if (*a != '-' || !strcmp(a, "--"))
3248                        break;
3249
3250                if (!skip_prefix(a, "--", &a))
3251                        die("unknown option %s", a);
3252
3253                if (parse_one_option(a))
3254                        continue;
3255
3256                if (parse_one_feature(a, 0))
3257                        continue;
3258
3259                if (skip_prefix(a, "cat-blob-fd=", &a)) {
3260                        option_cat_blob_fd(a);
3261                        continue;
3262                }
3263
3264                die("unknown option --%s", a);
3265        }
3266        if (i != global_argc)
3267                usage(fast_import_usage);
3268
3269        seen_data_command = 1;
3270        if (import_marks_file)
3271                read_marks();
3272}
3273
3274int cmd_main(int argc, const char **argv)
3275{
3276        unsigned int i;
3277
3278        if (argc == 2 && !strcmp(argv[1], "-h"))
3279                usage(fast_import_usage);
3280
3281        setup_git_directory();
3282        reset_pack_idx_option(&pack_idx_opts);
3283        git_pack_config();
3284
3285        alloc_objects(object_entry_alloc);
3286        strbuf_init(&command_buf, 0);
3287        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3288        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3289        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3290        marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3291
3292        global_argc = argc;
3293        global_argv = argv;
3294
3295        rc_free = mem_pool_alloc(&fi_mem_pool, cmd_save * sizeof(*rc_free));
3296        for (i = 0; i < (cmd_save - 1); i++)
3297                rc_free[i].next = &rc_free[i + 1];
3298        rc_free[cmd_save - 1].next = NULL;
3299
3300        start_packfile();
3301        set_die_routine(die_nicely);
3302        set_checkpoint_signal();
3303        while (read_next_command() != EOF) {
3304                const char *v;
3305                if (!strcmp("blob", command_buf.buf))
3306                        parse_new_blob();
3307                else if (skip_prefix(command_buf.buf, "ls ", &v))
3308                        parse_ls(v, NULL);
3309                else if (skip_prefix(command_buf.buf, "commit ", &v))
3310                        parse_new_commit(v);
3311                else if (skip_prefix(command_buf.buf, "tag ", &v))
3312                        parse_new_tag(v);
3313                else if (skip_prefix(command_buf.buf, "reset ", &v))
3314                        parse_reset_branch(v);
3315                else if (!strcmp("checkpoint", command_buf.buf))
3316                        parse_checkpoint();
3317                else if (!strcmp("done", command_buf.buf))
3318                        break;
3319                else if (starts_with(command_buf.buf, "progress "))
3320                        parse_progress();
3321                else if (skip_prefix(command_buf.buf, "feature ", &v))
3322                        parse_feature(v);
3323                else if (skip_prefix(command_buf.buf, "option git ", &v))
3324                        parse_option(v);
3325                else if (starts_with(command_buf.buf, "option "))
3326                        /* ignore non-git options*/;
3327                else
3328                        die("Unsupported command: %s", command_buf.buf);
3329
3330                if (checkpoint_requested)
3331                        checkpoint();
3332        }
3333
3334        /* argv hasn't been parsed yet, do so */
3335        if (!seen_data_command)
3336                parse_argv();
3337
3338        if (require_explicit_termination && feof(stdin))
3339                die("stream ends early");
3340
3341        end_packfile();
3342
3343        dump_branches();
3344        dump_tags();
3345        unkeep_all_packs();
3346        dump_marks();
3347
3348        if (pack_edges)
3349                fclose(pack_edges);
3350
3351        if (show_stats) {
3352                uintmax_t total_count = 0, duplicate_count = 0;
3353                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3354                        total_count += object_count_by_type[i];
3355                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3356                        duplicate_count += duplicate_count_by_type[i];
3357
3358                fprintf(stderr, "%s statistics:\n", argv[0]);
3359                fprintf(stderr, "---------------------------------------------------------------------\n");
3360                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3361                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3362                fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB], delta_count_attempts_by_type[OBJ_BLOB]);
3363                fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE], delta_count_attempts_by_type[OBJ_TREE]);
3364                fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT], delta_count_attempts_by_type[OBJ_COMMIT]);
3365                fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG], delta_count_attempts_by_type[OBJ_TAG]);
3366                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3367                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3368                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3369                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (tree_entry_allocd + fi_mem_pool.pool_alloc + alloc_count*sizeof(struct object_entry))/1024);
3370                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)((tree_entry_allocd + fi_mem_pool.pool_alloc) /1024));
3371                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3372                fprintf(stderr, "---------------------------------------------------------------------\n");
3373                pack_report();
3374                fprintf(stderr, "---------------------------------------------------------------------\n");
3375                fprintf(stderr, "\n");
3376        }
3377
3378        return failure ? 1 : 0;
3379}