fast-import.con commit Corrected buffer overflow during automatic checkpoint in fast-import. (5d6f3ef)
   1/*
   2Format of STDIN stream:
   3
   4  stream ::= cmd*;
   5
   6  cmd ::= new_blob
   7        | new_commit
   8        | new_tag
   9        | reset_branch
  10        ;
  11
  12  new_blob ::= 'blob' lf
  13        mark?
  14    file_content;
  15  file_content ::= data;
  16
  17  new_commit ::= 'commit' sp ref_str lf
  18    mark?
  19    ('author' sp name '<' email '>' ts tz lf)?
  20    'committer' sp name '<' email '>' ts tz lf
  21    commit_msg
  22    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  23    ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
  24    file_change*
  25    lf;
  26  commit_msg ::= data;
  27
  28  file_change ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf
  29                | 'D' sp path_str lf
  30                ;
  31  mode ::= '644' | '755';
  32
  33  new_tag ::= 'tag' sp tag_str lf
  34    'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
  35        'tagger' sp name '<' email '>' ts tz lf
  36    tag_msg;
  37  tag_msg ::= data;
  38
  39  reset_branch ::= 'reset' sp ref_str lf
  40    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  41    lf;
  42
  43  checkpoint ::= 'checkpoint' lf
  44    lf;
  45
  46     # note: the first idnum in a stream should be 1 and subsequent
  47     # idnums should not have gaps between values as this will cause
  48     # the stream parser to reserve space for the gapped values.  An
  49         # idnum can be updated in the future to a new object by issuing
  50     # a new mark directive with the old idnum.
  51         #
  52  mark ::= 'mark' sp idnum lf;
  53
  54     # note: declen indicates the length of binary_data in bytes.
  55     # declen does not include the lf preceeding or trailing the
  56     # binary data.
  57     #
  58  data ::= 'data' sp declen lf
  59    binary_data
  60        lf;
  61
  62     # note: quoted strings are C-style quoting supporting \c for
  63     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  64         # is the signed byte value in octal.  Note that the only
  65     # characters which must actually be escaped to protect the
  66     # stream formatting is: \, " and LF.  Otherwise these values
  67         # are UTF8.
  68     #
  69  ref_str     ::= ref     | '"' quoted(ref)     '"' ;
  70  sha1exp_str ::= sha1exp | '"' quoted(sha1exp) '"' ;
  71  tag_str     ::= tag     | '"' quoted(tag)     '"' ;
  72  path_str    ::= path    | '"' quoted(path)    '"' ;
  73
  74  declen ::= # unsigned 32 bit value, ascii base10 notation;
  75  binary_data ::= # file content, not interpreted;
  76
  77  sp ::= # ASCII space character;
  78  lf ::= # ASCII newline (LF) character;
  79
  80     # note: a colon (':') must precede the numerical value assigned to
  81         # an idnum.  This is to distinguish it from a ref or tag name as
  82     # GIT does not permit ':' in ref or tag strings.
  83         #
  84  idnum   ::= ':' declen;
  85  path    ::= # GIT style file path, e.g. "a/b/c";
  86  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
  87  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
  88  sha1exp ::= # Any valid GIT SHA1 expression;
  89  hexsha1 ::= # SHA1 in hexadecimal format;
  90
  91     # note: name and email are UTF8 strings, however name must not
  92         # contain '<' or lf and email must not contain any of the
  93     # following: '<', '>', lf.
  94         #
  95  name  ::= # valid GIT author/committer name;
  96  email ::= # valid GIT author/committer email;
  97  ts    ::= # time since the epoch in seconds, ascii base10 notation;
  98  tz    ::= # GIT style timezone;
  99*/
 100
 101#include "builtin.h"
 102#include "cache.h"
 103#include "object.h"
 104#include "blob.h"
 105#include "tree.h"
 106#include "delta.h"
 107#include "pack.h"
 108#include "refs.h"
 109#include "csum-file.h"
 110#include "strbuf.h"
 111#include "quote.h"
 112
 113struct object_entry
 114{
 115        struct object_entry *next;
 116        unsigned long offset;
 117        unsigned type : TYPE_BITS;
 118        unsigned pack_id : 16;
 119        unsigned char sha1[20];
 120};
 121
 122struct object_entry_pool
 123{
 124        struct object_entry_pool *next_pool;
 125        struct object_entry *next_free;
 126        struct object_entry *end;
 127        struct object_entry entries[FLEX_ARRAY]; /* more */
 128};
 129
 130struct mark_set
 131{
 132        int shift;
 133        union {
 134                struct object_entry *marked[1024];
 135                struct mark_set *sets[1024];
 136        } data;
 137};
 138
 139struct last_object
 140{
 141        void *data;
 142        unsigned long len;
 143        unsigned long offset;
 144        unsigned int depth;
 145        unsigned no_free:1;
 146};
 147
 148struct mem_pool
 149{
 150        struct mem_pool *next_pool;
 151        char *next_free;
 152        char *end;
 153        char space[FLEX_ARRAY]; /* more */
 154};
 155
 156struct atom_str
 157{
 158        struct atom_str *next_atom;
 159        int str_len;
 160        char str_dat[FLEX_ARRAY]; /* more */
 161};
 162
 163struct tree_content;
 164struct tree_entry
 165{
 166        struct tree_content *tree;
 167        struct atom_str* name;
 168        struct tree_entry_ms
 169        {
 170                unsigned int mode;
 171                unsigned char sha1[20];
 172        } versions[2];
 173};
 174
 175struct tree_content
 176{
 177        unsigned int entry_capacity; /* must match avail_tree_content */
 178        unsigned int entry_count;
 179        unsigned int delta_depth;
 180        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 181};
 182
 183struct avail_tree_content
 184{
 185        unsigned int entry_capacity; /* must match tree_content */
 186        struct avail_tree_content *next_avail;
 187};
 188
 189struct branch
 190{
 191        struct branch *table_next_branch;
 192        struct branch *active_next_branch;
 193        const char *name;
 194        unsigned long last_commit;
 195        struct tree_entry branch_tree;
 196        unsigned char sha1[20];
 197};
 198
 199struct tag
 200{
 201        struct tag *next_tag;
 202        const char *name;
 203        unsigned char sha1[20];
 204};
 205
 206struct dbuf
 207{
 208        void *buffer;
 209        size_t capacity;
 210};
 211
 212struct hash_list
 213{
 214        struct hash_list *next;
 215        unsigned char sha1[20];
 216};
 217
 218/* Stats and misc. counters */
 219static unsigned long max_depth = 10;
 220static unsigned long max_objects = -1;
 221static unsigned long max_packsize = -1;
 222static unsigned long alloc_count;
 223static unsigned long branch_count;
 224static unsigned long branch_load_count;
 225static unsigned long object_count;
 226static unsigned long marks_set_count;
 227static unsigned long object_count_by_type[1 << TYPE_BITS];
 228static unsigned long duplicate_count_by_type[1 << TYPE_BITS];
 229static unsigned long delta_count_by_type[1 << TYPE_BITS];
 230
 231/* Memory pools */
 232static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 233static size_t total_allocd;
 234static struct mem_pool *mem_pool;
 235
 236/* Atom management */
 237static unsigned int atom_table_sz = 4451;
 238static unsigned int atom_cnt;
 239static struct atom_str **atom_table;
 240
 241/* The .pack file being generated */
 242static const char *base_name;
 243static unsigned int pack_id;
 244static char *idx_name;
 245static struct packed_git *pack_data;
 246static struct packed_git **all_packs;
 247static int pack_fd;
 248static unsigned long pack_size;
 249static unsigned char pack_sha1[20];
 250
 251/* Table of objects we've written. */
 252static unsigned int object_entry_alloc = 5000;
 253static struct object_entry_pool *blocks;
 254static struct object_entry *object_table[1 << 16];
 255static struct mark_set *marks;
 256static const char* mark_file;
 257
 258/* Our last blob */
 259static struct last_object last_blob;
 260
 261/* Tree management */
 262static unsigned int tree_entry_alloc = 1000;
 263static void *avail_tree_entry;
 264static unsigned int avail_tree_table_sz = 100;
 265static struct avail_tree_content **avail_tree_table;
 266static struct dbuf old_tree;
 267static struct dbuf new_tree;
 268
 269/* Branch data */
 270static unsigned long max_active_branches = 5;
 271static unsigned long cur_active_branches;
 272static unsigned long branch_table_sz = 1039;
 273static struct branch **branch_table;
 274static struct branch *active_branches;
 275
 276/* Tag data */
 277static struct tag *first_tag;
 278static struct tag *last_tag;
 279
 280/* Input stream parsing */
 281static struct strbuf command_buf;
 282static unsigned long next_mark;
 283static struct dbuf new_data;
 284static FILE* branch_log;
 285
 286
 287static void alloc_objects(unsigned int cnt)
 288{
 289        struct object_entry_pool *b;
 290
 291        b = xmalloc(sizeof(struct object_entry_pool)
 292                + cnt * sizeof(struct object_entry));
 293        b->next_pool = blocks;
 294        b->next_free = b->entries;
 295        b->end = b->entries + cnt;
 296        blocks = b;
 297        alloc_count += cnt;
 298}
 299
 300static struct object_entry* new_object(unsigned char *sha1)
 301{
 302        struct object_entry *e;
 303
 304        if (blocks->next_free == blocks->end)
 305                alloc_objects(object_entry_alloc);
 306
 307        e = blocks->next_free++;
 308        hashcpy(e->sha1, sha1);
 309        return e;
 310}
 311
 312static struct object_entry* find_object(unsigned char *sha1)
 313{
 314        unsigned int h = sha1[0] << 8 | sha1[1];
 315        struct object_entry *e;
 316        for (e = object_table[h]; e; e = e->next)
 317                if (!hashcmp(sha1, e->sha1))
 318                        return e;
 319        return NULL;
 320}
 321
 322static struct object_entry* insert_object(unsigned char *sha1)
 323{
 324        unsigned int h = sha1[0] << 8 | sha1[1];
 325        struct object_entry *e = object_table[h];
 326        struct object_entry *p = NULL;
 327
 328        while (e) {
 329                if (!hashcmp(sha1, e->sha1))
 330                        return e;
 331                p = e;
 332                e = e->next;
 333        }
 334
 335        e = new_object(sha1);
 336        e->next = NULL;
 337        e->offset = 0;
 338        if (p)
 339                p->next = e;
 340        else
 341                object_table[h] = e;
 342        return e;
 343}
 344
 345static unsigned int hc_str(const char *s, size_t len)
 346{
 347        unsigned int r = 0;
 348        while (len-- > 0)
 349                r = r * 31 + *s++;
 350        return r;
 351}
 352
 353static void* pool_alloc(size_t len)
 354{
 355        struct mem_pool *p;
 356        void *r;
 357
 358        for (p = mem_pool; p; p = p->next_pool)
 359                if ((p->end - p->next_free >= len))
 360                        break;
 361
 362        if (!p) {
 363                if (len >= (mem_pool_alloc/2)) {
 364                        total_allocd += len;
 365                        return xmalloc(len);
 366                }
 367                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 368                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 369                p->next_pool = mem_pool;
 370                p->next_free = p->space;
 371                p->end = p->next_free + mem_pool_alloc;
 372                mem_pool = p;
 373        }
 374
 375        r = p->next_free;
 376        /* round out to a pointer alignment */
 377        if (len & (sizeof(void*) - 1))
 378                len += sizeof(void*) - (len & (sizeof(void*) - 1));
 379        p->next_free += len;
 380        return r;
 381}
 382
 383static void* pool_calloc(size_t count, size_t size)
 384{
 385        size_t len = count * size;
 386        void *r = pool_alloc(len);
 387        memset(r, 0, len);
 388        return r;
 389}
 390
 391static char* pool_strdup(const char *s)
 392{
 393        char *r = pool_alloc(strlen(s) + 1);
 394        strcpy(r, s);
 395        return r;
 396}
 397
 398static void size_dbuf(struct dbuf *b, size_t maxlen)
 399{
 400        if (b->buffer) {
 401                if (b->capacity >= maxlen)
 402                        return;
 403                free(b->buffer);
 404        }
 405        b->capacity = ((maxlen / 1024) + 1) * 1024;
 406        b->buffer = xmalloc(b->capacity);
 407}
 408
 409static void insert_mark(unsigned long idnum, struct object_entry *oe)
 410{
 411        struct mark_set *s = marks;
 412        while ((idnum >> s->shift) >= 1024) {
 413                s = pool_calloc(1, sizeof(struct mark_set));
 414                s->shift = marks->shift + 10;
 415                s->data.sets[0] = marks;
 416                marks = s;
 417        }
 418        while (s->shift) {
 419                unsigned long i = idnum >> s->shift;
 420                idnum -= i << s->shift;
 421                if (!s->data.sets[i]) {
 422                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 423                        s->data.sets[i]->shift = s->shift - 10;
 424                }
 425                s = s->data.sets[i];
 426        }
 427        if (!s->data.marked[idnum])
 428                marks_set_count++;
 429        s->data.marked[idnum] = oe;
 430}
 431
 432static struct object_entry* find_mark(unsigned long idnum)
 433{
 434        unsigned long orig_idnum = idnum;
 435        struct mark_set *s = marks;
 436        struct object_entry *oe = NULL;
 437        if ((idnum >> s->shift) < 1024) {
 438                while (s && s->shift) {
 439                        unsigned long i = idnum >> s->shift;
 440                        idnum -= i << s->shift;
 441                        s = s->data.sets[i];
 442                }
 443                if (s)
 444                        oe = s->data.marked[idnum];
 445        }
 446        if (!oe)
 447                die("mark :%lu not declared", orig_idnum);
 448        return oe;
 449}
 450
 451static struct atom_str* to_atom(const char *s, size_t len)
 452{
 453        unsigned int hc = hc_str(s, len) % atom_table_sz;
 454        struct atom_str *c;
 455
 456        for (c = atom_table[hc]; c; c = c->next_atom)
 457                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 458                        return c;
 459
 460        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 461        c->str_len = len;
 462        strncpy(c->str_dat, s, len);
 463        c->str_dat[len] = 0;
 464        c->next_atom = atom_table[hc];
 465        atom_table[hc] = c;
 466        atom_cnt++;
 467        return c;
 468}
 469
 470static struct branch* lookup_branch(const char *name)
 471{
 472        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 473        struct branch *b;
 474
 475        for (b = branch_table[hc]; b; b = b->table_next_branch)
 476                if (!strcmp(name, b->name))
 477                        return b;
 478        return NULL;
 479}
 480
 481static struct branch* new_branch(const char *name)
 482{
 483        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 484        struct branch* b = lookup_branch(name);
 485
 486        if (b)
 487                die("Invalid attempt to create duplicate branch: %s", name);
 488        if (check_ref_format(name))
 489                die("Branch name doesn't conform to GIT standards: %s", name);
 490
 491        b = pool_calloc(1, sizeof(struct branch));
 492        b->name = pool_strdup(name);
 493        b->table_next_branch = branch_table[hc];
 494        b->branch_tree.versions[0].mode = S_IFDIR;
 495        b->branch_tree.versions[1].mode = S_IFDIR;
 496        branch_table[hc] = b;
 497        branch_count++;
 498        return b;
 499}
 500
 501static unsigned int hc_entries(unsigned int cnt)
 502{
 503        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 504        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 505}
 506
 507static struct tree_content* new_tree_content(unsigned int cnt)
 508{
 509        struct avail_tree_content *f, *l = NULL;
 510        struct tree_content *t;
 511        unsigned int hc = hc_entries(cnt);
 512
 513        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 514                if (f->entry_capacity >= cnt)
 515                        break;
 516
 517        if (f) {
 518                if (l)
 519                        l->next_avail = f->next_avail;
 520                else
 521                        avail_tree_table[hc] = f->next_avail;
 522        } else {
 523                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 524                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 525                f->entry_capacity = cnt;
 526        }
 527
 528        t = (struct tree_content*)f;
 529        t->entry_count = 0;
 530        t->delta_depth = 0;
 531        return t;
 532}
 533
 534static void release_tree_entry(struct tree_entry *e);
 535static void release_tree_content(struct tree_content *t)
 536{
 537        struct avail_tree_content *f = (struct avail_tree_content*)t;
 538        unsigned int hc = hc_entries(f->entry_capacity);
 539        f->next_avail = avail_tree_table[hc];
 540        avail_tree_table[hc] = f;
 541}
 542
 543static void release_tree_content_recursive(struct tree_content *t)
 544{
 545        unsigned int i;
 546        for (i = 0; i < t->entry_count; i++)
 547                release_tree_entry(t->entries[i]);
 548        release_tree_content(t);
 549}
 550
 551static struct tree_content* grow_tree_content(
 552        struct tree_content *t,
 553        int amt)
 554{
 555        struct tree_content *r = new_tree_content(t->entry_count + amt);
 556        r->entry_count = t->entry_count;
 557        r->delta_depth = t->delta_depth;
 558        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 559        release_tree_content(t);
 560        return r;
 561}
 562
 563static struct tree_entry* new_tree_entry()
 564{
 565        struct tree_entry *e;
 566
 567        if (!avail_tree_entry) {
 568                unsigned int n = tree_entry_alloc;
 569                total_allocd += n * sizeof(struct tree_entry);
 570                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 571                while (n-- > 1) {
 572                        *((void**)e) = e + 1;
 573                        e++;
 574                }
 575                *((void**)e) = NULL;
 576        }
 577
 578        e = avail_tree_entry;
 579        avail_tree_entry = *((void**)e);
 580        return e;
 581}
 582
 583static void release_tree_entry(struct tree_entry *e)
 584{
 585        if (e->tree)
 586                release_tree_content_recursive(e->tree);
 587        *((void**)e) = avail_tree_entry;
 588        avail_tree_entry = e;
 589}
 590
 591static void yread(int fd, void *buffer, size_t length)
 592{
 593        ssize_t ret = 0;
 594        while (ret < length) {
 595                ssize_t size = xread(fd, (char *) buffer + ret, length - ret);
 596                if (!size)
 597                        die("Read from descriptor %i: end of stream", fd);
 598                if (size < 0)
 599                        die("Read from descriptor %i: %s", fd, strerror(errno));
 600                ret += size;
 601        }
 602}
 603
 604static void start_packfile()
 605{
 606        struct packed_git *p;
 607        struct pack_header hdr;
 608
 609        idx_name = xmalloc(strlen(base_name) + 11);
 610        p = xcalloc(1, sizeof(*p) + strlen(base_name) + 13);
 611        sprintf(p->pack_name, "%s%5.5i.pack", base_name, pack_id + 1);
 612        sprintf(idx_name, "%s%5.5i.idx", base_name, pack_id + 1);
 613
 614        pack_fd = open(p->pack_name, O_RDWR|O_CREAT|O_EXCL, 0666);
 615        if (pack_fd < 0)
 616                die("Can't create %s: %s", p->pack_name, strerror(errno));
 617        p->pack_fd = pack_fd;
 618
 619        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 620        hdr.hdr_version = htonl(2);
 621        hdr.hdr_entries = 0;
 622        write_or_die(pack_fd, &hdr, sizeof(hdr));
 623
 624        pack_data = p;
 625        pack_size = sizeof(hdr);
 626        object_count = 0;
 627
 628        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 629        all_packs[pack_id] = p;
 630}
 631
 632static void fixup_header_footer()
 633{
 634        SHA_CTX c;
 635        char hdr[8];
 636        unsigned long cnt;
 637        char *buf;
 638
 639        if (lseek(pack_fd, 0, SEEK_SET) != 0)
 640                die("Failed seeking to start: %s", strerror(errno));
 641
 642        SHA1_Init(&c);
 643        yread(pack_fd, hdr, 8);
 644        SHA1_Update(&c, hdr, 8);
 645
 646        cnt = htonl(object_count);
 647        SHA1_Update(&c, &cnt, 4);
 648        write_or_die(pack_fd, &cnt, 4);
 649
 650        buf = xmalloc(128 * 1024);
 651        for (;;) {
 652                size_t n = xread(pack_fd, buf, 128 * 1024);
 653                if (n <= 0)
 654                        break;
 655                SHA1_Update(&c, buf, n);
 656        }
 657        free(buf);
 658
 659        SHA1_Final(pack_sha1, &c);
 660        write_or_die(pack_fd, pack_sha1, sizeof(pack_sha1));
 661}
 662
 663static int oecmp (const void *a_, const void *b_)
 664{
 665        struct object_entry *a = *((struct object_entry**)a_);
 666        struct object_entry *b = *((struct object_entry**)b_);
 667        return hashcmp(a->sha1, b->sha1);
 668}
 669
 670static void write_index(const char *idx_name)
 671{
 672        struct sha1file *f;
 673        struct object_entry **idx, **c, **last, *e;
 674        struct object_entry_pool *o;
 675        unsigned int array[256];
 676        int i;
 677
 678        /* Build the sorted table of object IDs. */
 679        idx = xmalloc(object_count * sizeof(struct object_entry*));
 680        c = idx;
 681        for (o = blocks; o; o = o->next_pool)
 682                for (e = o->next_free; e-- != o->entries;)
 683                        if (pack_id == e->pack_id)
 684                                *c++ = e;
 685        last = idx + object_count;
 686        if (c != last)
 687                die("internal consistency error creating the index");
 688        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
 689
 690        /* Generate the fan-out array. */
 691        c = idx;
 692        for (i = 0; i < 256; i++) {
 693                struct object_entry **next = c;;
 694                while (next < last) {
 695                        if ((*next)->sha1[0] != i)
 696                                break;
 697                        next++;
 698                }
 699                array[i] = htonl(next - idx);
 700                c = next;
 701        }
 702
 703        f = sha1create("%s", idx_name);
 704        sha1write(f, array, 256 * sizeof(int));
 705        for (c = idx; c != last; c++) {
 706                unsigned int offset = htonl((*c)->offset);
 707                sha1write(f, &offset, 4);
 708                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
 709        }
 710        sha1write(f, pack_sha1, sizeof(pack_sha1));
 711        sha1close(f, NULL, 1);
 712        free(idx);
 713}
 714
 715static void end_packfile()
 716{
 717        struct packed_git *old_p = pack_data, *new_p;
 718
 719        if (object_count) {
 720                fixup_header_footer();
 721                write_index(idx_name);
 722                fprintf(stdout, "%s\n", old_p->pack_name);
 723                fflush(stdout);
 724
 725                /* Register the packfile with core git's machinary. */
 726                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 727                if (!new_p)
 728                        die("core git rejected index %s", idx_name);
 729                new_p->windows = old_p->windows;
 730                new_p->pack_fd = old_p->pack_fd;
 731                all_packs[pack_id++] = new_p;
 732                install_packed_git(new_p);
 733        }
 734        else {
 735                close(pack_fd);
 736                unlink(old_p->pack_name);
 737        }
 738        free(old_p);
 739        free(idx_name);
 740
 741        /* We can't carry a delta across packfiles. */
 742        free(last_blob.data);
 743        last_blob.data = NULL;
 744        last_blob.len = 0;
 745        last_blob.offset = 0;
 746        last_blob.depth = 0;
 747}
 748
 749static void checkpoint()
 750{
 751        end_packfile();
 752        start_packfile();
 753}
 754
 755static size_t encode_header(
 756        enum object_type type,
 757        size_t size,
 758        unsigned char *hdr)
 759{
 760        int n = 1;
 761        unsigned char c;
 762
 763        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 764                die("bad type %d", type);
 765
 766        c = (type << 4) | (size & 15);
 767        size >>= 4;
 768        while (size) {
 769                *hdr++ = c | 0x80;
 770                c = size & 0x7f;
 771                size >>= 7;
 772                n++;
 773        }
 774        *hdr = c;
 775        return n;
 776}
 777
 778static int store_object(
 779        enum object_type type,
 780        void *dat,
 781        size_t datlen,
 782        struct last_object *last,
 783        unsigned char *sha1out,
 784        unsigned long mark)
 785{
 786        void *out, *delta;
 787        struct object_entry *e;
 788        unsigned char hdr[96];
 789        unsigned char sha1[20];
 790        unsigned long hdrlen, deltalen;
 791        SHA_CTX c;
 792        z_stream s;
 793
 794        hdrlen = sprintf((char*)hdr,"%s %lu",type_names[type],datlen) + 1;
 795        SHA1_Init(&c);
 796        SHA1_Update(&c, hdr, hdrlen);
 797        SHA1_Update(&c, dat, datlen);
 798        SHA1_Final(sha1, &c);
 799        if (sha1out)
 800                hashcpy(sha1out, sha1);
 801
 802        e = insert_object(sha1);
 803        if (mark)
 804                insert_mark(mark, e);
 805        if (e->offset) {
 806                duplicate_count_by_type[type]++;
 807                return 1;
 808        }
 809
 810        if (last && last->data && last->depth < max_depth) {
 811                delta = diff_delta(last->data, last->len,
 812                        dat, datlen,
 813                        &deltalen, 0);
 814                if (delta && deltalen >= datlen) {
 815                        free(delta);
 816                        delta = NULL;
 817                }
 818        } else
 819                delta = NULL;
 820
 821        memset(&s, 0, sizeof(s));
 822        deflateInit(&s, zlib_compression_level);
 823        if (delta) {
 824                s.next_in = delta;
 825                s.avail_in = deltalen;
 826        } else {
 827                s.next_in = dat;
 828                s.avail_in = datlen;
 829        }
 830        s.avail_out = deflateBound(&s, s.avail_in);
 831        s.next_out = out = xmalloc(s.avail_out);
 832        while (deflate(&s, Z_FINISH) == Z_OK)
 833                /* nothing */;
 834        deflateEnd(&s);
 835
 836        /* Determine if we should auto-checkpoint. */
 837        if ((object_count + 1) > max_objects
 838                || (object_count + 1) < object_count
 839                || (pack_size + 60 + s.total_out) > max_packsize
 840                || (pack_size + 60 + s.total_out) < pack_size) {
 841
 842                /* This new object needs to *not* have the current pack_id. */
 843                e->pack_id = pack_id + 1;
 844                checkpoint();
 845
 846                /* We cannot carry a delta into the new pack. */
 847                if (delta) {
 848                        free(delta);
 849                        delta = NULL;
 850
 851                        memset(&s, 0, sizeof(s));
 852                        deflateInit(&s, zlib_compression_level);
 853                        s.next_in = dat;
 854                        s.avail_in = datlen;
 855                        s.avail_out = deflateBound(&s, s.avail_in);
 856                        s.next_out = out = xrealloc(out, s.avail_out);
 857                        while (deflate(&s, Z_FINISH) == Z_OK)
 858                                /* nothing */;
 859                        deflateEnd(&s);
 860                }
 861        }
 862
 863        e->type = type;
 864        e->pack_id = pack_id;
 865        e->offset = pack_size;
 866        object_count++;
 867        object_count_by_type[type]++;
 868
 869        if (delta) {
 870                unsigned long ofs = e->offset - last->offset;
 871                unsigned pos = sizeof(hdr) - 1;
 872
 873                delta_count_by_type[type]++;
 874                last->depth++;
 875
 876                hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
 877                write_or_die(pack_fd, hdr, hdrlen);
 878                pack_size += hdrlen;
 879
 880                hdr[pos] = ofs & 127;
 881                while (ofs >>= 7)
 882                        hdr[--pos] = 128 | (--ofs & 127);
 883                write_or_die(pack_fd, hdr + pos, sizeof(hdr) - pos);
 884                pack_size += sizeof(hdr) - pos;
 885        } else {
 886                if (last)
 887                        last->depth = 0;
 888                hdrlen = encode_header(type, datlen, hdr);
 889                write_or_die(pack_fd, hdr, hdrlen);
 890                pack_size += hdrlen;
 891        }
 892
 893        write_or_die(pack_fd, out, s.total_out);
 894        pack_size += s.total_out;
 895
 896        free(out);
 897        if (delta)
 898                free(delta);
 899        if (last) {
 900                if (last->data && !last->no_free)
 901                        free(last->data);
 902                last->data = dat;
 903                last->offset = e->offset;
 904                last->len = datlen;
 905        }
 906        return 0;
 907}
 908
 909static void *gfi_unpack_entry(
 910        struct object_entry *oe,
 911        unsigned long *sizep)
 912{
 913        static char type[20];
 914        struct packed_git *p = all_packs[oe->pack_id];
 915        if (p == pack_data)
 916                p->pack_size = pack_size + 20;
 917        return unpack_entry(p, oe->offset, type, sizep);
 918}
 919
 920static const char *get_mode(const char *str, unsigned int *modep)
 921{
 922        unsigned char c;
 923        unsigned int mode = 0;
 924
 925        while ((c = *str++) != ' ') {
 926                if (c < '0' || c > '7')
 927                        return NULL;
 928                mode = (mode << 3) + (c - '0');
 929        }
 930        *modep = mode;
 931        return str;
 932}
 933
 934static void load_tree(struct tree_entry *root)
 935{
 936        unsigned char* sha1 = root->versions[1].sha1;
 937        struct object_entry *myoe;
 938        struct tree_content *t;
 939        unsigned long size;
 940        char *buf;
 941        const char *c;
 942
 943        root->tree = t = new_tree_content(8);
 944        if (is_null_sha1(sha1))
 945                return;
 946
 947        myoe = find_object(sha1);
 948        if (myoe) {
 949                if (myoe->type != OBJ_TREE)
 950                        die("Not a tree: %s", sha1_to_hex(sha1));
 951                t->delta_depth = 0;
 952                buf = gfi_unpack_entry(myoe, &size);
 953        } else {
 954                char type[20];
 955                buf = read_sha1_file(sha1, type, &size);
 956                if (!buf || strcmp(type, tree_type))
 957                        die("Can't load tree %s", sha1_to_hex(sha1));
 958        }
 959
 960        c = buf;
 961        while (c != (buf + size)) {
 962                struct tree_entry *e = new_tree_entry();
 963
 964                if (t->entry_count == t->entry_capacity)
 965                        root->tree = t = grow_tree_content(t, 8);
 966                t->entries[t->entry_count++] = e;
 967
 968                e->tree = NULL;
 969                c = get_mode(c, &e->versions[1].mode);
 970                if (!c)
 971                        die("Corrupt mode in %s", sha1_to_hex(sha1));
 972                e->versions[0].mode = e->versions[1].mode;
 973                e->name = to_atom(c, strlen(c));
 974                c += e->name->str_len + 1;
 975                hashcpy(e->versions[0].sha1, (unsigned char*)c);
 976                hashcpy(e->versions[1].sha1, (unsigned char*)c);
 977                c += 20;
 978        }
 979        free(buf);
 980}
 981
 982static int tecmp0 (const void *_a, const void *_b)
 983{
 984        struct tree_entry *a = *((struct tree_entry**)_a);
 985        struct tree_entry *b = *((struct tree_entry**)_b);
 986        return base_name_compare(
 987                a->name->str_dat, a->name->str_len, a->versions[0].mode,
 988                b->name->str_dat, b->name->str_len, b->versions[0].mode);
 989}
 990
 991static int tecmp1 (const void *_a, const void *_b)
 992{
 993        struct tree_entry *a = *((struct tree_entry**)_a);
 994        struct tree_entry *b = *((struct tree_entry**)_b);
 995        return base_name_compare(
 996                a->name->str_dat, a->name->str_len, a->versions[1].mode,
 997                b->name->str_dat, b->name->str_len, b->versions[1].mode);
 998}
 999
1000static void mktree(struct tree_content *t,
1001        int v,
1002        unsigned long *szp,
1003        struct dbuf *b)
1004{
1005        size_t maxlen = 0;
1006        unsigned int i;
1007        char *c;
1008
1009        if (!v)
1010                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1011        else
1012                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1013
1014        for (i = 0; i < t->entry_count; i++) {
1015                if (t->entries[i]->versions[v].mode)
1016                        maxlen += t->entries[i]->name->str_len + 34;
1017        }
1018
1019        size_dbuf(b, maxlen);
1020        c = b->buffer;
1021        for (i = 0; i < t->entry_count; i++) {
1022                struct tree_entry *e = t->entries[i];
1023                if (!e->versions[v].mode)
1024                        continue;
1025                c += sprintf(c, "%o", e->versions[v].mode);
1026                *c++ = ' ';
1027                strcpy(c, e->name->str_dat);
1028                c += e->name->str_len + 1;
1029                hashcpy((unsigned char*)c, e->versions[v].sha1);
1030                c += 20;
1031        }
1032        *szp = c - (char*)b->buffer;
1033}
1034
1035static void store_tree(struct tree_entry *root)
1036{
1037        struct tree_content *t = root->tree;
1038        unsigned int i, j, del;
1039        unsigned long new_len;
1040        struct last_object lo;
1041        struct object_entry *le;
1042
1043        if (!is_null_sha1(root->versions[1].sha1))
1044                return;
1045
1046        for (i = 0; i < t->entry_count; i++) {
1047                if (t->entries[i]->tree)
1048                        store_tree(t->entries[i]);
1049        }
1050
1051        le = find_object(root->versions[0].sha1);
1052        if (!S_ISDIR(root->versions[0].mode)
1053                || !le
1054                || le->pack_id != pack_id) {
1055                lo.data = NULL;
1056                lo.depth = 0;
1057        } else {
1058                mktree(t, 0, &lo.len, &old_tree);
1059                lo.data = old_tree.buffer;
1060                lo.offset = le->offset;
1061                lo.depth = t->delta_depth;
1062                lo.no_free = 1;
1063        }
1064
1065        mktree(t, 1, &new_len, &new_tree);
1066        store_object(OBJ_TREE, new_tree.buffer, new_len,
1067                &lo, root->versions[1].sha1, 0);
1068
1069        t->delta_depth = lo.depth;
1070        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1071                struct tree_entry *e = t->entries[i];
1072                if (e->versions[1].mode) {
1073                        e->versions[0].mode = e->versions[1].mode;
1074                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1075                        t->entries[j++] = e;
1076                } else {
1077                        release_tree_entry(e);
1078                        del++;
1079                }
1080        }
1081        t->entry_count -= del;
1082}
1083
1084static int tree_content_set(
1085        struct tree_entry *root,
1086        const char *p,
1087        const unsigned char *sha1,
1088        const unsigned int mode)
1089{
1090        struct tree_content *t = root->tree;
1091        const char *slash1;
1092        unsigned int i, n;
1093        struct tree_entry *e;
1094
1095        slash1 = strchr(p, '/');
1096        if (slash1)
1097                n = slash1 - p;
1098        else
1099                n = strlen(p);
1100
1101        for (i = 0; i < t->entry_count; i++) {
1102                e = t->entries[i];
1103                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1104                        if (!slash1) {
1105                                if (e->versions[1].mode == mode
1106                                                && !hashcmp(e->versions[1].sha1, sha1))
1107                                        return 0;
1108                                e->versions[1].mode = mode;
1109                                hashcpy(e->versions[1].sha1, sha1);
1110                                if (e->tree) {
1111                                        release_tree_content_recursive(e->tree);
1112                                        e->tree = NULL;
1113                                }
1114                                hashclr(root->versions[1].sha1);
1115                                return 1;
1116                        }
1117                        if (!S_ISDIR(e->versions[1].mode)) {
1118                                e->tree = new_tree_content(8);
1119                                e->versions[1].mode = S_IFDIR;
1120                        }
1121                        if (!e->tree)
1122                                load_tree(e);
1123                        if (tree_content_set(e, slash1 + 1, sha1, mode)) {
1124                                hashclr(root->versions[1].sha1);
1125                                return 1;
1126                        }
1127                        return 0;
1128                }
1129        }
1130
1131        if (t->entry_count == t->entry_capacity)
1132                root->tree = t = grow_tree_content(t, 8);
1133        e = new_tree_entry();
1134        e->name = to_atom(p, n);
1135        e->versions[0].mode = 0;
1136        hashclr(e->versions[0].sha1);
1137        t->entries[t->entry_count++] = e;
1138        if (slash1) {
1139                e->tree = new_tree_content(8);
1140                e->versions[1].mode = S_IFDIR;
1141                tree_content_set(e, slash1 + 1, sha1, mode);
1142        } else {
1143                e->tree = NULL;
1144                e->versions[1].mode = mode;
1145                hashcpy(e->versions[1].sha1, sha1);
1146        }
1147        hashclr(root->versions[1].sha1);
1148        return 1;
1149}
1150
1151static int tree_content_remove(struct tree_entry *root, const char *p)
1152{
1153        struct tree_content *t = root->tree;
1154        const char *slash1;
1155        unsigned int i, n;
1156        struct tree_entry *e;
1157
1158        slash1 = strchr(p, '/');
1159        if (slash1)
1160                n = slash1 - p;
1161        else
1162                n = strlen(p);
1163
1164        for (i = 0; i < t->entry_count; i++) {
1165                e = t->entries[i];
1166                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1167                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1168                                goto del_entry;
1169                        if (!e->tree)
1170                                load_tree(e);
1171                        if (tree_content_remove(e, slash1 + 1)) {
1172                                for (n = 0; n < e->tree->entry_count; n++) {
1173                                        if (e->tree->entries[n]->versions[1].mode) {
1174                                                hashclr(root->versions[1].sha1);
1175                                                return 1;
1176                                        }
1177                                }
1178                                goto del_entry;
1179                        }
1180                        return 0;
1181                }
1182        }
1183        return 0;
1184
1185del_entry:
1186        if (e->tree) {
1187                release_tree_content_recursive(e->tree);
1188                e->tree = NULL;
1189        }
1190        e->versions[1].mode = 0;
1191        hashclr(e->versions[1].sha1);
1192        hashclr(root->versions[1].sha1);
1193        return 1;
1194}
1195
1196static void dump_branches()
1197{
1198        static const char *msg = "fast-import";
1199        unsigned int i;
1200        struct branch *b;
1201        struct ref_lock *lock;
1202
1203        for (i = 0; i < branch_table_sz; i++) {
1204                for (b = branch_table[i]; b; b = b->table_next_branch) {
1205                        lock = lock_any_ref_for_update(b->name, NULL);
1206                        if (!lock || write_ref_sha1(lock, b->sha1, msg) < 0)
1207                                die("Can't write %s", b->name);
1208                }
1209        }
1210}
1211
1212static void dump_tags()
1213{
1214        static const char *msg = "fast-import";
1215        struct tag *t;
1216        struct ref_lock *lock;
1217        char path[PATH_MAX];
1218
1219        for (t = first_tag; t; t = t->next_tag) {
1220                sprintf(path, "refs/tags/%s", t->name);
1221                lock = lock_any_ref_for_update(path, NULL);
1222                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1223                        die("Can't write %s", path);
1224        }
1225}
1226
1227static void dump_marks_helper(FILE *f,
1228        unsigned long base,
1229        struct mark_set *m)
1230{
1231        int k;
1232        if (m->shift) {
1233                for (k = 0; k < 1024; k++) {
1234                        if (m->data.sets[k])
1235                                dump_marks_helper(f, (base + k) << m->shift,
1236                                        m->data.sets[k]);
1237                }
1238        } else {
1239                for (k = 0; k < 1024; k++) {
1240                        if (m->data.marked[k])
1241                                fprintf(f, ":%lu %s\n", base + k,
1242                                        sha1_to_hex(m->data.marked[k]->sha1));
1243                }
1244        }
1245}
1246
1247static void dump_marks()
1248{
1249        if (mark_file)
1250        {
1251                FILE *f = fopen(mark_file, "w");
1252                dump_marks_helper(f, 0, marks);
1253                fclose(f);
1254        }
1255}
1256
1257static void read_next_command()
1258{
1259        read_line(&command_buf, stdin, '\n');
1260}
1261
1262static void cmd_mark()
1263{
1264        if (!strncmp("mark :", command_buf.buf, 6)) {
1265                next_mark = strtoul(command_buf.buf + 6, NULL, 10);
1266                read_next_command();
1267        }
1268        else
1269                next_mark = 0;
1270}
1271
1272static void* cmd_data (size_t *size)
1273{
1274        size_t n = 0;
1275        void *buffer;
1276        size_t length;
1277
1278        if (strncmp("data ", command_buf.buf, 5))
1279                die("Expected 'data n' command, found: %s", command_buf.buf);
1280
1281        length = strtoul(command_buf.buf + 5, NULL, 10);
1282        buffer = xmalloc(length);
1283
1284        while (n < length) {
1285                size_t s = fread((char*)buffer + n, 1, length - n, stdin);
1286                if (!s && feof(stdin))
1287                        die("EOF in data (%lu bytes remaining)", length - n);
1288                n += s;
1289        }
1290
1291        if (fgetc(stdin) != '\n')
1292                die("An lf did not trail the binary data as expected.");
1293
1294        *size = length;
1295        return buffer;
1296}
1297
1298static void cmd_new_blob()
1299{
1300        size_t l;
1301        void *d;
1302
1303        read_next_command();
1304        cmd_mark();
1305        d = cmd_data(&l);
1306
1307        if (store_object(OBJ_BLOB, d, l, &last_blob, NULL, next_mark))
1308                free(d);
1309}
1310
1311static void unload_one_branch()
1312{
1313        while (cur_active_branches
1314                && cur_active_branches >= max_active_branches) {
1315                unsigned long min_commit = ULONG_MAX;
1316                struct branch *e, *l = NULL, *p = NULL;
1317
1318                for (e = active_branches; e; e = e->active_next_branch) {
1319                        if (e->last_commit < min_commit) {
1320                                p = l;
1321                                min_commit = e->last_commit;
1322                        }
1323                        l = e;
1324                }
1325
1326                if (p) {
1327                        e = p->active_next_branch;
1328                        p->active_next_branch = e->active_next_branch;
1329                } else {
1330                        e = active_branches;
1331                        active_branches = e->active_next_branch;
1332                }
1333                e->active_next_branch = NULL;
1334                if (e->branch_tree.tree) {
1335                        release_tree_content_recursive(e->branch_tree.tree);
1336                        e->branch_tree.tree = NULL;
1337                }
1338                cur_active_branches--;
1339        }
1340}
1341
1342static void load_branch(struct branch *b)
1343{
1344        load_tree(&b->branch_tree);
1345        b->active_next_branch = active_branches;
1346        active_branches = b;
1347        cur_active_branches++;
1348        branch_load_count++;
1349}
1350
1351static void file_change_m(struct branch *b)
1352{
1353        const char *p = command_buf.buf + 2;
1354        char *p_uq;
1355        const char *endp;
1356        struct object_entry *oe;
1357        unsigned char sha1[20];
1358        unsigned int mode;
1359        char type[20];
1360
1361        p = get_mode(p, &mode);
1362        if (!p)
1363                die("Corrupt mode: %s", command_buf.buf);
1364        switch (mode) {
1365        case S_IFREG | 0644:
1366        case S_IFREG | 0755:
1367        case S_IFLNK:
1368        case 0644:
1369        case 0755:
1370                /* ok */
1371                break;
1372        default:
1373                die("Corrupt mode: %s", command_buf.buf);
1374        }
1375
1376        if (*p == ':') {
1377                char *x;
1378                oe = find_mark(strtoul(p + 1, &x, 10));
1379                hashcpy(sha1, oe->sha1);
1380                p = x;
1381        } else {
1382                if (get_sha1_hex(p, sha1))
1383                        die("Invalid SHA1: %s", command_buf.buf);
1384                oe = find_object(sha1);
1385                p += 40;
1386        }
1387        if (*p++ != ' ')
1388                die("Missing space after SHA1: %s", command_buf.buf);
1389
1390        p_uq = unquote_c_style(p, &endp);
1391        if (p_uq) {
1392                if (*endp)
1393                        die("Garbage after path in: %s", command_buf.buf);
1394                p = p_uq;
1395        }
1396
1397        if (oe) {
1398                if (oe->type != OBJ_BLOB)
1399                        die("Not a blob (actually a %s): %s",
1400                                command_buf.buf, type_names[oe->type]);
1401        } else {
1402                if (sha1_object_info(sha1, type, NULL))
1403                        die("Blob not found: %s", command_buf.buf);
1404                if (strcmp(blob_type, type))
1405                        die("Not a blob (actually a %s): %s",
1406                                command_buf.buf, type);
1407        }
1408
1409        tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode);
1410
1411        if (p_uq)
1412                free(p_uq);
1413}
1414
1415static void file_change_d(struct branch *b)
1416{
1417        const char *p = command_buf.buf + 2;
1418        char *p_uq;
1419        const char *endp;
1420
1421        p_uq = unquote_c_style(p, &endp);
1422        if (p_uq) {
1423                if (*endp)
1424                        die("Garbage after path in: %s", command_buf.buf);
1425                p = p_uq;
1426        }
1427        tree_content_remove(&b->branch_tree, p);
1428        if (p_uq)
1429                free(p_uq);
1430}
1431
1432static void cmd_from(struct branch *b)
1433{
1434        const char *from, *endp;
1435        char *str_uq;
1436        struct branch *s;
1437
1438        if (strncmp("from ", command_buf.buf, 5))
1439                return;
1440
1441        if (b->last_commit)
1442                die("Can't reinitailize branch %s", b->name);
1443
1444        from = strchr(command_buf.buf, ' ') + 1;
1445        str_uq = unquote_c_style(from, &endp);
1446        if (str_uq) {
1447                if (*endp)
1448                        die("Garbage after string in: %s", command_buf.buf);
1449                from = str_uq;
1450        }
1451
1452        s = lookup_branch(from);
1453        if (b == s)
1454                die("Can't create a branch from itself: %s", b->name);
1455        else if (s) {
1456                unsigned char *t = s->branch_tree.versions[1].sha1;
1457                hashcpy(b->sha1, s->sha1);
1458                hashcpy(b->branch_tree.versions[0].sha1, t);
1459                hashcpy(b->branch_tree.versions[1].sha1, t);
1460        } else if (*from == ':') {
1461                unsigned long idnum = strtoul(from + 1, NULL, 10);
1462                struct object_entry *oe = find_mark(idnum);
1463                unsigned long size;
1464                char *buf;
1465                if (oe->type != OBJ_COMMIT)
1466                        die("Mark :%lu not a commit", idnum);
1467                hashcpy(b->sha1, oe->sha1);
1468                buf = gfi_unpack_entry(oe, &size);
1469                if (!buf || size < 46)
1470                        die("Not a valid commit: %s", from);
1471                if (memcmp("tree ", buf, 5)
1472                        || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1473                        die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1474                free(buf);
1475                hashcpy(b->branch_tree.versions[0].sha1,
1476                        b->branch_tree.versions[1].sha1);
1477        } else if (!get_sha1(from, b->sha1)) {
1478                if (is_null_sha1(b->sha1)) {
1479                        hashclr(b->branch_tree.versions[0].sha1);
1480                        hashclr(b->branch_tree.versions[1].sha1);
1481                } else {
1482                        unsigned long size;
1483                        char *buf;
1484
1485                        buf = read_object_with_reference(b->sha1,
1486                                type_names[OBJ_COMMIT], &size, b->sha1);
1487                        if (!buf || size < 46)
1488                                die("Not a valid commit: %s", from);
1489                        if (memcmp("tree ", buf, 5)
1490                                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1491                                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1492                        free(buf);
1493                        hashcpy(b->branch_tree.versions[0].sha1,
1494                                b->branch_tree.versions[1].sha1);
1495                }
1496        } else
1497                die("Invalid ref name or SHA1 expression: %s", from);
1498
1499        read_next_command();
1500}
1501
1502static struct hash_list* cmd_merge(unsigned int *count)
1503{
1504        struct hash_list *list = NULL, *n, *e;
1505        const char *from, *endp;
1506        char *str_uq;
1507        struct branch *s;
1508
1509        *count = 0;
1510        while (!strncmp("merge ", command_buf.buf, 6)) {
1511                from = strchr(command_buf.buf, ' ') + 1;
1512                str_uq = unquote_c_style(from, &endp);
1513                if (str_uq) {
1514                        if (*endp)
1515                                die("Garbage after string in: %s", command_buf.buf);
1516                        from = str_uq;
1517                }
1518
1519                n = xmalloc(sizeof(*n));
1520                s = lookup_branch(from);
1521                if (s)
1522                        hashcpy(n->sha1, s->sha1);
1523                else if (*from == ':') {
1524                        unsigned long idnum = strtoul(from + 1, NULL, 10);
1525                        struct object_entry *oe = find_mark(idnum);
1526                        if (oe->type != OBJ_COMMIT)
1527                                die("Mark :%lu not a commit", idnum);
1528                        hashcpy(n->sha1, oe->sha1);
1529                } else if (get_sha1(from, n->sha1))
1530                        die("Invalid ref name or SHA1 expression: %s", from);
1531
1532                n->next = NULL;
1533                if (list)
1534                        e->next = n;
1535                else
1536                        list = n;
1537                e = n;
1538                *count++;
1539                read_next_command();
1540        }
1541        return list;
1542}
1543
1544static void cmd_new_commit()
1545{
1546        struct branch *b;
1547        void *msg;
1548        size_t msglen;
1549        char *str_uq;
1550        const char *endp;
1551        char *sp;
1552        char *author = NULL;
1553        char *committer = NULL;
1554        struct hash_list *merge_list = NULL;
1555        unsigned int merge_count;
1556
1557        /* Obtain the branch name from the rest of our command */
1558        sp = strchr(command_buf.buf, ' ') + 1;
1559        str_uq = unquote_c_style(sp, &endp);
1560        if (str_uq) {
1561                if (*endp)
1562                        die("Garbage after ref in: %s", command_buf.buf);
1563                sp = str_uq;
1564        }
1565        b = lookup_branch(sp);
1566        if (!b)
1567                b = new_branch(sp);
1568        if (str_uq)
1569                free(str_uq);
1570
1571        read_next_command();
1572        cmd_mark();
1573        if (!strncmp("author ", command_buf.buf, 7)) {
1574                author = strdup(command_buf.buf);
1575                read_next_command();
1576        }
1577        if (!strncmp("committer ", command_buf.buf, 10)) {
1578                committer = strdup(command_buf.buf);
1579                read_next_command();
1580        }
1581        if (!committer)
1582                die("Expected committer but didn't get one");
1583        msg = cmd_data(&msglen);
1584        read_next_command();
1585        cmd_from(b);
1586        merge_list = cmd_merge(&merge_count);
1587
1588        /* ensure the branch is active/loaded */
1589        if (!b->branch_tree.tree || !max_active_branches) {
1590                unload_one_branch();
1591                load_branch(b);
1592        }
1593
1594        /* file_change* */
1595        for (;;) {
1596                if (1 == command_buf.len)
1597                        break;
1598                else if (!strncmp("M ", command_buf.buf, 2))
1599                        file_change_m(b);
1600                else if (!strncmp("D ", command_buf.buf, 2))
1601                        file_change_d(b);
1602                else
1603                        die("Unsupported file_change: %s", command_buf.buf);
1604                read_next_command();
1605        }
1606
1607        /* build the tree and the commit */
1608        store_tree(&b->branch_tree);
1609        hashcpy(b->branch_tree.versions[0].sha1,
1610                b->branch_tree.versions[1].sha1);
1611        size_dbuf(&new_data, 97 + msglen
1612                + merge_count * 49
1613                + (author
1614                        ? strlen(author) + strlen(committer)
1615                        : 2 * strlen(committer)));
1616        sp = new_data.buffer;
1617        sp += sprintf(sp, "tree %s\n",
1618                sha1_to_hex(b->branch_tree.versions[1].sha1));
1619        if (!is_null_sha1(b->sha1))
1620                sp += sprintf(sp, "parent %s\n", sha1_to_hex(b->sha1));
1621        while (merge_list) {
1622                struct hash_list *next = merge_list->next;
1623                sp += sprintf(sp, "parent %s\n", sha1_to_hex(merge_list->sha1));
1624                free(merge_list);
1625                merge_list = next;
1626        }
1627        if (author)
1628                sp += sprintf(sp, "%s\n", author);
1629        else
1630                sp += sprintf(sp, "author %s\n", committer + 10);
1631        sp += sprintf(sp, "%s\n\n", committer);
1632        memcpy(sp, msg, msglen);
1633        sp += msglen;
1634        if (author)
1635                free(author);
1636        free(committer);
1637        free(msg);
1638
1639        store_object(OBJ_COMMIT,
1640                new_data.buffer, sp - (char*)new_data.buffer,
1641                NULL, b->sha1, next_mark);
1642        b->last_commit = object_count_by_type[OBJ_COMMIT];
1643
1644        if (branch_log) {
1645                int need_dq = quote_c_style(b->name, NULL, NULL, 0);
1646                fprintf(branch_log, "commit ");
1647                if (need_dq) {
1648                        fputc('"', branch_log);
1649                        quote_c_style(b->name, NULL, branch_log, 0);
1650                        fputc('"', branch_log);
1651                } else
1652                        fprintf(branch_log, "%s", b->name);
1653                fprintf(branch_log," :%lu %s\n",next_mark,sha1_to_hex(b->sha1));
1654        }
1655}
1656
1657static void cmd_new_tag()
1658{
1659        char *str_uq;
1660        const char *endp;
1661        char *sp;
1662        const char *from;
1663        char *tagger;
1664        struct branch *s;
1665        void *msg;
1666        size_t msglen;
1667        struct tag *t;
1668        unsigned long from_mark = 0;
1669        unsigned char sha1[20];
1670
1671        /* Obtain the new tag name from the rest of our command */
1672        sp = strchr(command_buf.buf, ' ') + 1;
1673        str_uq = unquote_c_style(sp, &endp);
1674        if (str_uq) {
1675                if (*endp)
1676                        die("Garbage after tag name in: %s", command_buf.buf);
1677                sp = str_uq;
1678        }
1679        t = pool_alloc(sizeof(struct tag));
1680        t->next_tag = NULL;
1681        t->name = pool_strdup(sp);
1682        if (last_tag)
1683                last_tag->next_tag = t;
1684        else
1685                first_tag = t;
1686        last_tag = t;
1687        if (str_uq)
1688                free(str_uq);
1689        read_next_command();
1690
1691        /* from ... */
1692        if (strncmp("from ", command_buf.buf, 5))
1693                die("Expected from command, got %s", command_buf.buf);
1694
1695        from = strchr(command_buf.buf, ' ') + 1;
1696        str_uq = unquote_c_style(from, &endp);
1697        if (str_uq) {
1698                if (*endp)
1699                        die("Garbage after string in: %s", command_buf.buf);
1700                from = str_uq;
1701        }
1702
1703        s = lookup_branch(from);
1704        if (s) {
1705                hashcpy(sha1, s->sha1);
1706        } else if (*from == ':') {
1707                from_mark = strtoul(from + 1, NULL, 10);
1708                struct object_entry *oe = find_mark(from_mark);
1709                if (oe->type != OBJ_COMMIT)
1710                        die("Mark :%lu not a commit", from_mark);
1711                hashcpy(sha1, oe->sha1);
1712        } else if (!get_sha1(from, sha1)) {
1713                unsigned long size;
1714                char *buf;
1715
1716                buf = read_object_with_reference(sha1,
1717                        type_names[OBJ_COMMIT], &size, sha1);
1718                if (!buf || size < 46)
1719                        die("Not a valid commit: %s", from);
1720                free(buf);
1721        } else
1722                die("Invalid ref name or SHA1 expression: %s", from);
1723
1724        if (str_uq)
1725                free(str_uq);
1726        read_next_command();
1727
1728        /* tagger ... */
1729        if (strncmp("tagger ", command_buf.buf, 7))
1730                die("Expected tagger command, got %s", command_buf.buf);
1731        tagger = strdup(command_buf.buf);
1732
1733        /* tag payload/message */
1734        read_next_command();
1735        msg = cmd_data(&msglen);
1736
1737        /* build the tag object */
1738        size_dbuf(&new_data, 67+strlen(t->name)+strlen(tagger)+msglen);
1739        sp = new_data.buffer;
1740        sp += sprintf(sp, "object %s\n", sha1_to_hex(sha1));
1741        sp += sprintf(sp, "type %s\n", type_names[OBJ_COMMIT]);
1742        sp += sprintf(sp, "tag %s\n", t->name);
1743        sp += sprintf(sp, "%s\n\n", tagger);
1744        memcpy(sp, msg, msglen);
1745        sp += msglen;
1746        free(tagger);
1747        free(msg);
1748
1749        store_object(OBJ_TAG, new_data.buffer, sp - (char*)new_data.buffer,
1750                NULL, t->sha1, 0);
1751
1752        if (branch_log) {
1753                int need_dq = quote_c_style(t->name, NULL, NULL, 0);
1754                fprintf(branch_log, "tag ");
1755                if (need_dq) {
1756                        fputc('"', branch_log);
1757                        quote_c_style(t->name, NULL, branch_log, 0);
1758                        fputc('"', branch_log);
1759                } else
1760                        fprintf(branch_log, "%s", t->name);
1761                fprintf(branch_log," :%lu %s\n",from_mark,sha1_to_hex(t->sha1));
1762        }
1763}
1764
1765static void cmd_reset_branch()
1766{
1767        struct branch *b;
1768        char *str_uq;
1769        const char *endp;
1770        char *sp;
1771
1772        /* Obtain the branch name from the rest of our command */
1773        sp = strchr(command_buf.buf, ' ') + 1;
1774        str_uq = unquote_c_style(sp, &endp);
1775        if (str_uq) {
1776                if (*endp)
1777                        die("Garbage after ref in: %s", command_buf.buf);
1778                sp = str_uq;
1779        }
1780        b = lookup_branch(sp);
1781        if (b) {
1782                b->last_commit = 0;
1783                if (b->branch_tree.tree) {
1784                        release_tree_content_recursive(b->branch_tree.tree);
1785                        b->branch_tree.tree = NULL;
1786                }
1787        }
1788        else
1789                b = new_branch(sp);
1790        if (str_uq)
1791                free(str_uq);
1792        read_next_command();
1793        cmd_from(b);
1794}
1795
1796static void cmd_checkpoint()
1797{
1798        if (object_count)
1799                checkpoint();
1800        read_next_command();
1801}
1802
1803static const char fast_import_usage[] =
1804"git-fast-import [--objects=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file] [--branch-log=log] temp.pack";
1805
1806int main(int argc, const char **argv)
1807{
1808        int i;
1809        unsigned long est_obj_cnt = object_entry_alloc;
1810        unsigned long duplicate_count;
1811
1812        setup_ident();
1813        git_config(git_default_config);
1814
1815        for (i = 1; i < argc; i++) {
1816                const char *a = argv[i];
1817
1818                if (*a != '-' || !strcmp(a, "--"))
1819                        break;
1820                else if (!strncmp(a, "--objects=", 10))
1821                        est_obj_cnt = strtoul(a + 10, NULL, 0);
1822                else if (!strncmp(a, "--max-objects-per-pack=", 23))
1823                        max_objects = strtoul(a + 23, NULL, 0);
1824                else if (!strncmp(a, "--max-pack-size=", 16))
1825                        max_packsize = strtoul(a + 16, NULL, 0) * 1024 * 1024;
1826                else if (!strncmp(a, "--depth=", 8))
1827                        max_depth = strtoul(a + 8, NULL, 0);
1828                else if (!strncmp(a, "--active-branches=", 18))
1829                        max_active_branches = strtoul(a + 18, NULL, 0);
1830                else if (!strncmp(a, "--export-marks=", 15))
1831                        mark_file = a + 15;
1832                else if (!strncmp(a, "--branch-log=", 13)) {
1833                        branch_log = fopen(a + 13, "w");
1834                        if (!branch_log)
1835                                die("Can't create %s: %s", a + 13, strerror(errno));
1836                }
1837                else
1838                        die("unknown option %s", a);
1839        }
1840        if ((i+1) != argc)
1841                usage(fast_import_usage);
1842        base_name = argv[i];
1843
1844        alloc_objects(est_obj_cnt);
1845        strbuf_init(&command_buf);
1846
1847        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
1848        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
1849        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
1850        marks = pool_calloc(1, sizeof(struct mark_set));
1851
1852        start_packfile();
1853        for (;;) {
1854                read_next_command();
1855                if (command_buf.eof)
1856                        break;
1857                else if (!strcmp("blob", command_buf.buf))
1858                        cmd_new_blob();
1859                else if (!strncmp("commit ", command_buf.buf, 7))
1860                        cmd_new_commit();
1861                else if (!strncmp("tag ", command_buf.buf, 4))
1862                        cmd_new_tag();
1863                else if (!strncmp("reset ", command_buf.buf, 6))
1864                        cmd_reset_branch();
1865                else if (!strcmp("checkpoint", command_buf.buf))
1866                        cmd_checkpoint();
1867                else
1868                        die("Unsupported command: %s", command_buf.buf);
1869        }
1870        end_packfile();
1871
1872        dump_branches();
1873        dump_tags();
1874        dump_marks();
1875        if (branch_log)
1876                fclose(branch_log);
1877
1878        for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
1879                duplicate_count += duplicate_count_by_type[i];
1880
1881        fprintf(stderr, "%s statistics:\n", argv[0]);
1882        fprintf(stderr, "---------------------------------------------------------------------\n");
1883        fprintf(stderr, "Alloc'd objects: %10lu (%10lu overflow  )\n", alloc_count, alloc_count - est_obj_cnt);
1884        fprintf(stderr, "Total objects:   %10lu (%10lu duplicates                  )\n", object_count, duplicate_count);
1885        fprintf(stderr, "      blobs  :   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
1886        fprintf(stderr, "      trees  :   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
1887        fprintf(stderr, "      commits:   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
1888        fprintf(stderr, "      tags   :   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
1889        fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
1890        fprintf(stderr, "      marks:     %10u (%10lu unique    )\n", (1 << marks->shift) * 1024, marks_set_count);
1891        fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
1892        fprintf(stderr, "Memory total:    %10lu KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
1893        fprintf(stderr, "       pools:    %10lu KiB\n", total_allocd/1024);
1894        fprintf(stderr, "     objects:    %10lu KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
1895        fprintf(stderr, "---------------------------------------------------------------------\n");
1896        pack_report();
1897        fprintf(stderr, "---------------------------------------------------------------------\n");
1898        fprintf(stderr, "\n");
1899
1900        return 0;
1901}