fast-import.con commit fast-import: prevent producing bad delta (8fb3ad7)
   1/*
   2(See Documentation/git-fast-import.txt for maintained documentation.)
   3Format of STDIN stream:
   4
   5  stream ::= cmd*;
   6
   7  cmd ::= new_blob
   8        | new_commit
   9        | new_tag
  10        | reset_branch
  11        | checkpoint
  12        | progress
  13        ;
  14
  15  new_blob ::= 'blob' lf
  16    mark?
  17    file_content;
  18  file_content ::= data;
  19
  20  new_commit ::= 'commit' sp ref_str lf
  21    mark?
  22    ('author' (sp name)? sp '<' email '>' sp when lf)?
  23    'committer' (sp name)? sp '<' email '>' sp when lf
  24    commit_msg
  25    ('from' sp committish lf)?
  26    ('merge' sp committish lf)*
  27    (file_change | ls)*
  28    lf?;
  29  commit_msg ::= data;
  30
  31  ls ::= 'ls' sp '"' quoted(path) '"' lf;
  32
  33  file_change ::= file_clr
  34    | file_del
  35    | file_rnm
  36    | file_cpy
  37    | file_obm
  38    | file_inm;
  39  file_clr ::= 'deleteall' lf;
  40  file_del ::= 'D' sp path_str lf;
  41  file_rnm ::= 'R' sp path_str sp path_str lf;
  42  file_cpy ::= 'C' sp path_str sp path_str lf;
  43  file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
  44  file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
  45    data;
  46  note_obm ::= 'N' sp (hexsha1 | idnum) sp committish lf;
  47  note_inm ::= 'N' sp 'inline' sp committish lf
  48    data;
  49
  50  new_tag ::= 'tag' sp tag_str lf
  51    'from' sp committish lf
  52    ('tagger' (sp name)? sp '<' email '>' sp when lf)?
  53    tag_msg;
  54  tag_msg ::= data;
  55
  56  reset_branch ::= 'reset' sp ref_str lf
  57    ('from' sp committish lf)?
  58    lf?;
  59
  60  checkpoint ::= 'checkpoint' lf
  61    lf?;
  62
  63  progress ::= 'progress' sp not_lf* lf
  64    lf?;
  65
  66     # note: the first idnum in a stream should be 1 and subsequent
  67     # idnums should not have gaps between values as this will cause
  68     # the stream parser to reserve space for the gapped values.  An
  69     # idnum can be updated in the future to a new object by issuing
  70     # a new mark directive with the old idnum.
  71     #
  72  mark ::= 'mark' sp idnum lf;
  73  data ::= (delimited_data | exact_data)
  74    lf?;
  75
  76    # note: delim may be any string but must not contain lf.
  77    # data_line may contain any data but must not be exactly
  78    # delim.
  79  delimited_data ::= 'data' sp '<<' delim lf
  80    (data_line lf)*
  81    delim lf;
  82
  83     # note: declen indicates the length of binary_data in bytes.
  84     # declen does not include the lf preceding the binary data.
  85     #
  86  exact_data ::= 'data' sp declen lf
  87    binary_data;
  88
  89     # note: quoted strings are C-style quoting supporting \c for
  90     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  91     # is the signed byte value in octal.  Note that the only
  92     # characters which must actually be escaped to protect the
  93     # stream formatting is: \, " and LF.  Otherwise these values
  94     # are UTF8.
  95     #
  96  committish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
  97  ref_str     ::= ref;
  98  sha1exp_str ::= sha1exp;
  99  tag_str     ::= tag;
 100  path_str    ::= path    | '"' quoted(path)    '"' ;
 101  mode        ::= '100644' | '644'
 102                | '100755' | '755'
 103                | '120000'
 104                ;
 105
 106  declen ::= # unsigned 32 bit value, ascii base10 notation;
 107  bigint ::= # unsigned integer value, ascii base10 notation;
 108  binary_data ::= # file content, not interpreted;
 109
 110  when         ::= raw_when | rfc2822_when;
 111  raw_when     ::= ts sp tz;
 112  rfc2822_when ::= # Valid RFC 2822 date and time;
 113
 114  sp ::= # ASCII space character;
 115  lf ::= # ASCII newline (LF) character;
 116
 117     # note: a colon (':') must precede the numerical value assigned to
 118     # an idnum.  This is to distinguish it from a ref or tag name as
 119     # GIT does not permit ':' in ref or tag strings.
 120     #
 121  idnum   ::= ':' bigint;
 122  path    ::= # GIT style file path, e.g. "a/b/c";
 123  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
 124  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
 125  sha1exp ::= # Any valid GIT SHA1 expression;
 126  hexsha1 ::= # SHA1 in hexadecimal format;
 127
 128     # note: name and email are UTF8 strings, however name must not
 129     # contain '<' or lf and email must not contain any of the
 130     # following: '<', '>', lf.
 131     #
 132  name  ::= # valid GIT author/committer name;
 133  email ::= # valid GIT author/committer email;
 134  ts    ::= # time since the epoch in seconds, ascii base10 notation;
 135  tz    ::= # GIT style timezone;
 136
 137     # note: comments, ls and cat requests may appear anywhere
 138     # in the input, except within a data command.  Any form
 139     # of the data command always escapes the related input
 140     # from comment processing.
 141     #
 142     # In case it is not clear, the '#' that starts the comment
 143     # must be the first character on that line (an lf
 144     # preceded it).
 145     #
 146
 147  cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
 148  ls_tree  ::= 'ls' sp (hexsha1 | idnum) sp path_str lf;
 149
 150  comment ::= '#' not_lf* lf;
 151  not_lf  ::= # Any byte that is not ASCII newline (LF);
 152*/
 153
 154#include "builtin.h"
 155#include "cache.h"
 156#include "object.h"
 157#include "blob.h"
 158#include "tree.h"
 159#include "commit.h"
 160#include "delta.h"
 161#include "pack.h"
 162#include "refs.h"
 163#include "csum-file.h"
 164#include "quote.h"
 165#include "exec_cmd.h"
 166#include "dir.h"
 167
 168#define PACK_ID_BITS 16
 169#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 170#define DEPTH_BITS 13
 171#define MAX_DEPTH ((1<<DEPTH_BITS)-1)
 172
 173/*
 174 * We abuse the setuid bit on directories to mean "do not delta".
 175 */
 176#define NO_DELTA S_ISUID
 177
 178struct object_entry {
 179        struct pack_idx_entry idx;
 180        struct object_entry *next;
 181        uint32_t type : TYPE_BITS,
 182                pack_id : PACK_ID_BITS,
 183                depth : DEPTH_BITS;
 184};
 185
 186struct object_entry_pool {
 187        struct object_entry_pool *next_pool;
 188        struct object_entry *next_free;
 189        struct object_entry *end;
 190        struct object_entry entries[FLEX_ARRAY]; /* more */
 191};
 192
 193struct mark_set {
 194        union {
 195                struct object_entry *marked[1024];
 196                struct mark_set *sets[1024];
 197        } data;
 198        unsigned int shift;
 199};
 200
 201struct last_object {
 202        struct strbuf data;
 203        off_t offset;
 204        unsigned int depth;
 205        unsigned no_swap : 1;
 206};
 207
 208struct mem_pool {
 209        struct mem_pool *next_pool;
 210        char *next_free;
 211        char *end;
 212        uintmax_t space[FLEX_ARRAY]; /* more */
 213};
 214
 215struct atom_str {
 216        struct atom_str *next_atom;
 217        unsigned short str_len;
 218        char str_dat[FLEX_ARRAY]; /* more */
 219};
 220
 221struct tree_content;
 222struct tree_entry {
 223        struct tree_content *tree;
 224        struct atom_str *name;
 225        struct tree_entry_ms {
 226                uint16_t mode;
 227                unsigned char sha1[20];
 228        } versions[2];
 229};
 230
 231struct tree_content {
 232        unsigned int entry_capacity; /* must match avail_tree_content */
 233        unsigned int entry_count;
 234        unsigned int delta_depth;
 235        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 236};
 237
 238struct avail_tree_content {
 239        unsigned int entry_capacity; /* must match tree_content */
 240        struct avail_tree_content *next_avail;
 241};
 242
 243struct branch {
 244        struct branch *table_next_branch;
 245        struct branch *active_next_branch;
 246        const char *name;
 247        struct tree_entry branch_tree;
 248        uintmax_t last_commit;
 249        uintmax_t num_notes;
 250        unsigned active : 1;
 251        unsigned pack_id : PACK_ID_BITS;
 252        unsigned char sha1[20];
 253};
 254
 255struct tag {
 256        struct tag *next_tag;
 257        const char *name;
 258        unsigned int pack_id;
 259        unsigned char sha1[20];
 260};
 261
 262struct hash_list {
 263        struct hash_list *next;
 264        unsigned char sha1[20];
 265};
 266
 267typedef enum {
 268        WHENSPEC_RAW = 1,
 269        WHENSPEC_RFC2822,
 270        WHENSPEC_NOW
 271} whenspec_type;
 272
 273struct recent_command {
 274        struct recent_command *prev;
 275        struct recent_command *next;
 276        char *buf;
 277};
 278
 279/* Configured limits on output */
 280static unsigned long max_depth = 10;
 281static off_t max_packsize;
 282static int force_update;
 283static int pack_compression_level = Z_DEFAULT_COMPRESSION;
 284static int pack_compression_seen;
 285
 286/* Stats and misc. counters */
 287static uintmax_t alloc_count;
 288static uintmax_t marks_set_count;
 289static uintmax_t object_count_by_type[1 << TYPE_BITS];
 290static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 291static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 292static unsigned long object_count;
 293static unsigned long branch_count;
 294static unsigned long branch_load_count;
 295static int failure;
 296static FILE *pack_edges;
 297static unsigned int show_stats = 1;
 298static int global_argc;
 299static const char **global_argv;
 300
 301/* Memory pools */
 302static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 303static size_t total_allocd;
 304static struct mem_pool *mem_pool;
 305
 306/* Atom management */
 307static unsigned int atom_table_sz = 4451;
 308static unsigned int atom_cnt;
 309static struct atom_str **atom_table;
 310
 311/* The .pack file being generated */
 312static unsigned int pack_id;
 313static struct sha1file *pack_file;
 314static struct packed_git *pack_data;
 315static struct packed_git **all_packs;
 316static off_t pack_size;
 317
 318/* Table of objects we've written. */
 319static unsigned int object_entry_alloc = 5000;
 320static struct object_entry_pool *blocks;
 321static struct object_entry *object_table[1 << 16];
 322static struct mark_set *marks;
 323static const char *export_marks_file;
 324static const char *import_marks_file;
 325static int import_marks_file_from_stream;
 326static int import_marks_file_ignore_missing;
 327static int relative_marks_paths;
 328
 329/* Our last blob */
 330static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
 331
 332/* Tree management */
 333static unsigned int tree_entry_alloc = 1000;
 334static void *avail_tree_entry;
 335static unsigned int avail_tree_table_sz = 100;
 336static struct avail_tree_content **avail_tree_table;
 337static struct strbuf old_tree = STRBUF_INIT;
 338static struct strbuf new_tree = STRBUF_INIT;
 339
 340/* Branch data */
 341static unsigned long max_active_branches = 5;
 342static unsigned long cur_active_branches;
 343static unsigned long branch_table_sz = 1039;
 344static struct branch **branch_table;
 345static struct branch *active_branches;
 346
 347/* Tag data */
 348static struct tag *first_tag;
 349static struct tag *last_tag;
 350
 351/* Input stream parsing */
 352static whenspec_type whenspec = WHENSPEC_RAW;
 353static struct strbuf command_buf = STRBUF_INIT;
 354static int unread_command_buf;
 355static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 356static struct recent_command *cmd_tail = &cmd_hist;
 357static struct recent_command *rc_free;
 358static unsigned int cmd_save = 100;
 359static uintmax_t next_mark;
 360static struct strbuf new_data = STRBUF_INIT;
 361static int seen_data_command;
 362
 363/* Signal handling */
 364static volatile sig_atomic_t checkpoint_requested;
 365
 366/* Where to write output of cat-blob commands */
 367static int cat_blob_fd = STDOUT_FILENO;
 368
 369static void parse_argv(void);
 370static void parse_cat_blob(void);
 371static void parse_ls(struct branch *b);
 372
 373static void write_branch_report(FILE *rpt, struct branch *b)
 374{
 375        fprintf(rpt, "%s:\n", b->name);
 376
 377        fprintf(rpt, "  status      :");
 378        if (b->active)
 379                fputs(" active", rpt);
 380        if (b->branch_tree.tree)
 381                fputs(" loaded", rpt);
 382        if (is_null_sha1(b->branch_tree.versions[1].sha1))
 383                fputs(" dirty", rpt);
 384        fputc('\n', rpt);
 385
 386        fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
 387        fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
 388        fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
 389        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 390
 391        fputs("  last pack   : ", rpt);
 392        if (b->pack_id < MAX_PACK_ID)
 393                fprintf(rpt, "%u", b->pack_id);
 394        fputc('\n', rpt);
 395
 396        fputc('\n', rpt);
 397}
 398
 399static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
 400
 401static void write_crash_report(const char *err)
 402{
 403        char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
 404        FILE *rpt = fopen(loc, "w");
 405        struct branch *b;
 406        unsigned long lu;
 407        struct recent_command *rc;
 408
 409        if (!rpt) {
 410                error("can't write crash report %s: %s", loc, strerror(errno));
 411                return;
 412        }
 413
 414        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 415
 416        fprintf(rpt, "fast-import crash report:\n");
 417        fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
 418        fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
 419        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
 420        fputc('\n', rpt);
 421
 422        fputs("fatal: ", rpt);
 423        fputs(err, rpt);
 424        fputc('\n', rpt);
 425
 426        fputc('\n', rpt);
 427        fputs("Most Recent Commands Before Crash\n", rpt);
 428        fputs("---------------------------------\n", rpt);
 429        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 430                if (rc->next == &cmd_hist)
 431                        fputs("* ", rpt);
 432                else
 433                        fputs("  ", rpt);
 434                fputs(rc->buf, rpt);
 435                fputc('\n', rpt);
 436        }
 437
 438        fputc('\n', rpt);
 439        fputs("Active Branch LRU\n", rpt);
 440        fputs("-----------------\n", rpt);
 441        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 442                cur_active_branches,
 443                max_active_branches);
 444        fputc('\n', rpt);
 445        fputs("  pos  clock name\n", rpt);
 446        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 447        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 448                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 449                        ++lu, b->last_commit, b->name);
 450
 451        fputc('\n', rpt);
 452        fputs("Inactive Branches\n", rpt);
 453        fputs("-----------------\n", rpt);
 454        for (lu = 0; lu < branch_table_sz; lu++) {
 455                for (b = branch_table[lu]; b; b = b->table_next_branch)
 456                        write_branch_report(rpt, b);
 457        }
 458
 459        if (first_tag) {
 460                struct tag *tg;
 461                fputc('\n', rpt);
 462                fputs("Annotated Tags\n", rpt);
 463                fputs("--------------\n", rpt);
 464                for (tg = first_tag; tg; tg = tg->next_tag) {
 465                        fputs(sha1_to_hex(tg->sha1), rpt);
 466                        fputc(' ', rpt);
 467                        fputs(tg->name, rpt);
 468                        fputc('\n', rpt);
 469                }
 470        }
 471
 472        fputc('\n', rpt);
 473        fputs("Marks\n", rpt);
 474        fputs("-----\n", rpt);
 475        if (export_marks_file)
 476                fprintf(rpt, "  exported to %s\n", export_marks_file);
 477        else
 478                dump_marks_helper(rpt, 0, marks);
 479
 480        fputc('\n', rpt);
 481        fputs("-------------------\n", rpt);
 482        fputs("END OF CRASH REPORT\n", rpt);
 483        fclose(rpt);
 484}
 485
 486static void end_packfile(void);
 487static void unkeep_all_packs(void);
 488static void dump_marks(void);
 489
 490static NORETURN void die_nicely(const char *err, va_list params)
 491{
 492        static int zombie;
 493        char message[2 * PATH_MAX];
 494
 495        vsnprintf(message, sizeof(message), err, params);
 496        fputs("fatal: ", stderr);
 497        fputs(message, stderr);
 498        fputc('\n', stderr);
 499
 500        if (!zombie) {
 501                zombie = 1;
 502                write_crash_report(message);
 503                end_packfile();
 504                unkeep_all_packs();
 505                dump_marks();
 506        }
 507        exit(128);
 508}
 509
 510#ifndef SIGUSR1 /* Windows, for example */
 511
 512static void set_checkpoint_signal(void)
 513{
 514}
 515
 516#else
 517
 518static void checkpoint_signal(int signo)
 519{
 520        checkpoint_requested = 1;
 521}
 522
 523static void set_checkpoint_signal(void)
 524{
 525        struct sigaction sa;
 526
 527        memset(&sa, 0, sizeof(sa));
 528        sa.sa_handler = checkpoint_signal;
 529        sigemptyset(&sa.sa_mask);
 530        sa.sa_flags = SA_RESTART;
 531        sigaction(SIGUSR1, &sa, NULL);
 532}
 533
 534#endif
 535
 536static void alloc_objects(unsigned int cnt)
 537{
 538        struct object_entry_pool *b;
 539
 540        b = xmalloc(sizeof(struct object_entry_pool)
 541                + cnt * sizeof(struct object_entry));
 542        b->next_pool = blocks;
 543        b->next_free = b->entries;
 544        b->end = b->entries + cnt;
 545        blocks = b;
 546        alloc_count += cnt;
 547}
 548
 549static struct object_entry *new_object(unsigned char *sha1)
 550{
 551        struct object_entry *e;
 552
 553        if (blocks->next_free == blocks->end)
 554                alloc_objects(object_entry_alloc);
 555
 556        e = blocks->next_free++;
 557        hashcpy(e->idx.sha1, sha1);
 558        return e;
 559}
 560
 561static struct object_entry *find_object(unsigned char *sha1)
 562{
 563        unsigned int h = sha1[0] << 8 | sha1[1];
 564        struct object_entry *e;
 565        for (e = object_table[h]; e; e = e->next)
 566                if (!hashcmp(sha1, e->idx.sha1))
 567                        return e;
 568        return NULL;
 569}
 570
 571static struct object_entry *insert_object(unsigned char *sha1)
 572{
 573        unsigned int h = sha1[0] << 8 | sha1[1];
 574        struct object_entry *e = object_table[h];
 575
 576        while (e) {
 577                if (!hashcmp(sha1, e->idx.sha1))
 578                        return e;
 579                e = e->next;
 580        }
 581
 582        e = new_object(sha1);
 583        e->next = object_table[h];
 584        e->idx.offset = 0;
 585        object_table[h] = e;
 586        return e;
 587}
 588
 589static unsigned int hc_str(const char *s, size_t len)
 590{
 591        unsigned int r = 0;
 592        while (len-- > 0)
 593                r = r * 31 + *s++;
 594        return r;
 595}
 596
 597static void *pool_alloc(size_t len)
 598{
 599        struct mem_pool *p;
 600        void *r;
 601
 602        /* round up to a 'uintmax_t' alignment */
 603        if (len & (sizeof(uintmax_t) - 1))
 604                len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
 605
 606        for (p = mem_pool; p; p = p->next_pool)
 607                if ((p->end - p->next_free >= len))
 608                        break;
 609
 610        if (!p) {
 611                if (len >= (mem_pool_alloc/2)) {
 612                        total_allocd += len;
 613                        return xmalloc(len);
 614                }
 615                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 616                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 617                p->next_pool = mem_pool;
 618                p->next_free = (char *) p->space;
 619                p->end = p->next_free + mem_pool_alloc;
 620                mem_pool = p;
 621        }
 622
 623        r = p->next_free;
 624        p->next_free += len;
 625        return r;
 626}
 627
 628static void *pool_calloc(size_t count, size_t size)
 629{
 630        size_t len = count * size;
 631        void *r = pool_alloc(len);
 632        memset(r, 0, len);
 633        return r;
 634}
 635
 636static char *pool_strdup(const char *s)
 637{
 638        char *r = pool_alloc(strlen(s) + 1);
 639        strcpy(r, s);
 640        return r;
 641}
 642
 643static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 644{
 645        struct mark_set *s = marks;
 646        while ((idnum >> s->shift) >= 1024) {
 647                s = pool_calloc(1, sizeof(struct mark_set));
 648                s->shift = marks->shift + 10;
 649                s->data.sets[0] = marks;
 650                marks = s;
 651        }
 652        while (s->shift) {
 653                uintmax_t i = idnum >> s->shift;
 654                idnum -= i << s->shift;
 655                if (!s->data.sets[i]) {
 656                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 657                        s->data.sets[i]->shift = s->shift - 10;
 658                }
 659                s = s->data.sets[i];
 660        }
 661        if (!s->data.marked[idnum])
 662                marks_set_count++;
 663        s->data.marked[idnum] = oe;
 664}
 665
 666static struct object_entry *find_mark(uintmax_t idnum)
 667{
 668        uintmax_t orig_idnum = idnum;
 669        struct mark_set *s = marks;
 670        struct object_entry *oe = NULL;
 671        if ((idnum >> s->shift) < 1024) {
 672                while (s && s->shift) {
 673                        uintmax_t i = idnum >> s->shift;
 674                        idnum -= i << s->shift;
 675                        s = s->data.sets[i];
 676                }
 677                if (s)
 678                        oe = s->data.marked[idnum];
 679        }
 680        if (!oe)
 681                die("mark :%" PRIuMAX " not declared", orig_idnum);
 682        return oe;
 683}
 684
 685static struct atom_str *to_atom(const char *s, unsigned short len)
 686{
 687        unsigned int hc = hc_str(s, len) % atom_table_sz;
 688        struct atom_str *c;
 689
 690        for (c = atom_table[hc]; c; c = c->next_atom)
 691                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 692                        return c;
 693
 694        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 695        c->str_len = len;
 696        strncpy(c->str_dat, s, len);
 697        c->str_dat[len] = 0;
 698        c->next_atom = atom_table[hc];
 699        atom_table[hc] = c;
 700        atom_cnt++;
 701        return c;
 702}
 703
 704static struct branch *lookup_branch(const char *name)
 705{
 706        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 707        struct branch *b;
 708
 709        for (b = branch_table[hc]; b; b = b->table_next_branch)
 710                if (!strcmp(name, b->name))
 711                        return b;
 712        return NULL;
 713}
 714
 715static struct branch *new_branch(const char *name)
 716{
 717        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 718        struct branch *b = lookup_branch(name);
 719
 720        if (b)
 721                die("Invalid attempt to create duplicate branch: %s", name);
 722        switch (check_ref_format(name)) {
 723        case 0: break; /* its valid */
 724        case CHECK_REF_FORMAT_ONELEVEL:
 725                break; /* valid, but too few '/', allow anyway */
 726        default:
 727                die("Branch name doesn't conform to GIT standards: %s", name);
 728        }
 729
 730        b = pool_calloc(1, sizeof(struct branch));
 731        b->name = pool_strdup(name);
 732        b->table_next_branch = branch_table[hc];
 733        b->branch_tree.versions[0].mode = S_IFDIR;
 734        b->branch_tree.versions[1].mode = S_IFDIR;
 735        b->num_notes = 0;
 736        b->active = 0;
 737        b->pack_id = MAX_PACK_ID;
 738        branch_table[hc] = b;
 739        branch_count++;
 740        return b;
 741}
 742
 743static unsigned int hc_entries(unsigned int cnt)
 744{
 745        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 746        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 747}
 748
 749static struct tree_content *new_tree_content(unsigned int cnt)
 750{
 751        struct avail_tree_content *f, *l = NULL;
 752        struct tree_content *t;
 753        unsigned int hc = hc_entries(cnt);
 754
 755        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 756                if (f->entry_capacity >= cnt)
 757                        break;
 758
 759        if (f) {
 760                if (l)
 761                        l->next_avail = f->next_avail;
 762                else
 763                        avail_tree_table[hc] = f->next_avail;
 764        } else {
 765                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 766                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 767                f->entry_capacity = cnt;
 768        }
 769
 770        t = (struct tree_content*)f;
 771        t->entry_count = 0;
 772        t->delta_depth = 0;
 773        return t;
 774}
 775
 776static void release_tree_entry(struct tree_entry *e);
 777static void release_tree_content(struct tree_content *t)
 778{
 779        struct avail_tree_content *f = (struct avail_tree_content*)t;
 780        unsigned int hc = hc_entries(f->entry_capacity);
 781        f->next_avail = avail_tree_table[hc];
 782        avail_tree_table[hc] = f;
 783}
 784
 785static void release_tree_content_recursive(struct tree_content *t)
 786{
 787        unsigned int i;
 788        for (i = 0; i < t->entry_count; i++)
 789                release_tree_entry(t->entries[i]);
 790        release_tree_content(t);
 791}
 792
 793static struct tree_content *grow_tree_content(
 794        struct tree_content *t,
 795        int amt)
 796{
 797        struct tree_content *r = new_tree_content(t->entry_count + amt);
 798        r->entry_count = t->entry_count;
 799        r->delta_depth = t->delta_depth;
 800        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 801        release_tree_content(t);
 802        return r;
 803}
 804
 805static struct tree_entry *new_tree_entry(void)
 806{
 807        struct tree_entry *e;
 808
 809        if (!avail_tree_entry) {
 810                unsigned int n = tree_entry_alloc;
 811                total_allocd += n * sizeof(struct tree_entry);
 812                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 813                while (n-- > 1) {
 814                        *((void**)e) = e + 1;
 815                        e++;
 816                }
 817                *((void**)e) = NULL;
 818        }
 819
 820        e = avail_tree_entry;
 821        avail_tree_entry = *((void**)e);
 822        return e;
 823}
 824
 825static void release_tree_entry(struct tree_entry *e)
 826{
 827        if (e->tree)
 828                release_tree_content_recursive(e->tree);
 829        *((void**)e) = avail_tree_entry;
 830        avail_tree_entry = e;
 831}
 832
 833static struct tree_content *dup_tree_content(struct tree_content *s)
 834{
 835        struct tree_content *d;
 836        struct tree_entry *a, *b;
 837        unsigned int i;
 838
 839        if (!s)
 840                return NULL;
 841        d = new_tree_content(s->entry_count);
 842        for (i = 0; i < s->entry_count; i++) {
 843                a = s->entries[i];
 844                b = new_tree_entry();
 845                memcpy(b, a, sizeof(*a));
 846                if (a->tree && is_null_sha1(b->versions[1].sha1))
 847                        b->tree = dup_tree_content(a->tree);
 848                else
 849                        b->tree = NULL;
 850                d->entries[i] = b;
 851        }
 852        d->entry_count = s->entry_count;
 853        d->delta_depth = s->delta_depth;
 854
 855        return d;
 856}
 857
 858static void start_packfile(void)
 859{
 860        static char tmpfile[PATH_MAX];
 861        struct packed_git *p;
 862        struct pack_header hdr;
 863        int pack_fd;
 864
 865        pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
 866                              "pack/tmp_pack_XXXXXX");
 867        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 868        strcpy(p->pack_name, tmpfile);
 869        p->pack_fd = pack_fd;
 870        p->do_not_close = 1;
 871        pack_file = sha1fd(pack_fd, p->pack_name);
 872
 873        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 874        hdr.hdr_version = htonl(2);
 875        hdr.hdr_entries = 0;
 876        sha1write(pack_file, &hdr, sizeof(hdr));
 877
 878        pack_data = p;
 879        pack_size = sizeof(hdr);
 880        object_count = 0;
 881
 882        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 883        all_packs[pack_id] = p;
 884}
 885
 886static const char *create_index(void)
 887{
 888        const char *tmpfile;
 889        struct pack_idx_entry **idx, **c, **last;
 890        struct object_entry *e;
 891        struct object_entry_pool *o;
 892
 893        /* Build the table of object IDs. */
 894        idx = xmalloc(object_count * sizeof(*idx));
 895        c = idx;
 896        for (o = blocks; o; o = o->next_pool)
 897                for (e = o->next_free; e-- != o->entries;)
 898                        if (pack_id == e->pack_id)
 899                                *c++ = &e->idx;
 900        last = idx + object_count;
 901        if (c != last)
 902                die("internal consistency error creating the index");
 903
 904        tmpfile = write_idx_file(NULL, idx, object_count, pack_data->sha1);
 905        free(idx);
 906        return tmpfile;
 907}
 908
 909static char *keep_pack(const char *curr_index_name)
 910{
 911        static char name[PATH_MAX];
 912        static const char *keep_msg = "fast-import";
 913        int keep_fd;
 914
 915        keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
 916        if (keep_fd < 0)
 917                die_errno("cannot create keep file");
 918        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 919        if (close(keep_fd))
 920                die_errno("failed to write keep file");
 921
 922        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 923                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 924        if (move_temp_to_file(pack_data->pack_name, name))
 925                die("cannot store pack file");
 926
 927        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 928                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 929        if (move_temp_to_file(curr_index_name, name))
 930                die("cannot store index file");
 931        free((void *)curr_index_name);
 932        return name;
 933}
 934
 935static void unkeep_all_packs(void)
 936{
 937        static char name[PATH_MAX];
 938        int k;
 939
 940        for (k = 0; k < pack_id; k++) {
 941                struct packed_git *p = all_packs[k];
 942                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 943                         get_object_directory(), sha1_to_hex(p->sha1));
 944                unlink_or_warn(name);
 945        }
 946}
 947
 948static void end_packfile(void)
 949{
 950        struct packed_git *old_p = pack_data, *new_p;
 951
 952        clear_delta_base_cache();
 953        if (object_count) {
 954                unsigned char cur_pack_sha1[20];
 955                char *idx_name;
 956                int i;
 957                struct branch *b;
 958                struct tag *t;
 959
 960                close_pack_windows(pack_data);
 961                sha1close(pack_file, cur_pack_sha1, 0);
 962                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 963                                    pack_data->pack_name, object_count,
 964                                    cur_pack_sha1, pack_size);
 965                close(pack_data->pack_fd);
 966                idx_name = keep_pack(create_index());
 967
 968                /* Register the packfile with core git's machinery. */
 969                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 970                if (!new_p)
 971                        die("core git rejected index %s", idx_name);
 972                all_packs[pack_id] = new_p;
 973                install_packed_git(new_p);
 974
 975                /* Print the boundary */
 976                if (pack_edges) {
 977                        fprintf(pack_edges, "%s:", new_p->pack_name);
 978                        for (i = 0; i < branch_table_sz; i++) {
 979                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 980                                        if (b->pack_id == pack_id)
 981                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 982                                }
 983                        }
 984                        for (t = first_tag; t; t = t->next_tag) {
 985                                if (t->pack_id == pack_id)
 986                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 987                        }
 988                        fputc('\n', pack_edges);
 989                        fflush(pack_edges);
 990                }
 991
 992                pack_id++;
 993        }
 994        else {
 995                close(old_p->pack_fd);
 996                unlink_or_warn(old_p->pack_name);
 997        }
 998        free(old_p);
 999
1000        /* We can't carry a delta across packfiles. */
1001        strbuf_release(&last_blob.data);
1002        last_blob.offset = 0;
1003        last_blob.depth = 0;
1004}
1005
1006static void cycle_packfile(void)
1007{
1008        end_packfile();
1009        start_packfile();
1010}
1011
1012static int store_object(
1013        enum object_type type,
1014        struct strbuf *dat,
1015        struct last_object *last,
1016        unsigned char *sha1out,
1017        uintmax_t mark)
1018{
1019        void *out, *delta;
1020        struct object_entry *e;
1021        unsigned char hdr[96];
1022        unsigned char sha1[20];
1023        unsigned long hdrlen, deltalen;
1024        git_SHA_CTX c;
1025        z_stream s;
1026
1027        hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1028                (unsigned long)dat->len) + 1;
1029        git_SHA1_Init(&c);
1030        git_SHA1_Update(&c, hdr, hdrlen);
1031        git_SHA1_Update(&c, dat->buf, dat->len);
1032        git_SHA1_Final(sha1, &c);
1033        if (sha1out)
1034                hashcpy(sha1out, sha1);
1035
1036        e = insert_object(sha1);
1037        if (mark)
1038                insert_mark(mark, e);
1039        if (e->idx.offset) {
1040                duplicate_count_by_type[type]++;
1041                return 1;
1042        } else if (find_sha1_pack(sha1, packed_git)) {
1043                e->type = type;
1044                e->pack_id = MAX_PACK_ID;
1045                e->idx.offset = 1; /* just not zero! */
1046                duplicate_count_by_type[type]++;
1047                return 1;
1048        }
1049
1050        if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1051                delta = diff_delta(last->data.buf, last->data.len,
1052                        dat->buf, dat->len,
1053                        &deltalen, dat->len - 20);
1054        } else
1055                delta = NULL;
1056
1057        memset(&s, 0, sizeof(s));
1058        deflateInit(&s, pack_compression_level);
1059        if (delta) {
1060                s.next_in = delta;
1061                s.avail_in = deltalen;
1062        } else {
1063                s.next_in = (void *)dat->buf;
1064                s.avail_in = dat->len;
1065        }
1066        s.avail_out = deflateBound(&s, s.avail_in);
1067        s.next_out = out = xmalloc(s.avail_out);
1068        while (deflate(&s, Z_FINISH) == Z_OK)
1069                /* nothing */;
1070        deflateEnd(&s);
1071
1072        /* Determine if we should auto-checkpoint. */
1073        if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1074                || (pack_size + 60 + s.total_out) < pack_size) {
1075
1076                /* This new object needs to *not* have the current pack_id. */
1077                e->pack_id = pack_id + 1;
1078                cycle_packfile();
1079
1080                /* We cannot carry a delta into the new pack. */
1081                if (delta) {
1082                        free(delta);
1083                        delta = NULL;
1084
1085                        memset(&s, 0, sizeof(s));
1086                        deflateInit(&s, pack_compression_level);
1087                        s.next_in = (void *)dat->buf;
1088                        s.avail_in = dat->len;
1089                        s.avail_out = deflateBound(&s, s.avail_in);
1090                        s.next_out = out = xrealloc(out, s.avail_out);
1091                        while (deflate(&s, Z_FINISH) == Z_OK)
1092                                /* nothing */;
1093                        deflateEnd(&s);
1094                }
1095        }
1096
1097        e->type = type;
1098        e->pack_id = pack_id;
1099        e->idx.offset = pack_size;
1100        object_count++;
1101        object_count_by_type[type]++;
1102
1103        crc32_begin(pack_file);
1104
1105        if (delta) {
1106                off_t ofs = e->idx.offset - last->offset;
1107                unsigned pos = sizeof(hdr) - 1;
1108
1109                delta_count_by_type[type]++;
1110                e->depth = last->depth + 1;
1111
1112                hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1113                sha1write(pack_file, hdr, hdrlen);
1114                pack_size += hdrlen;
1115
1116                hdr[pos] = ofs & 127;
1117                while (ofs >>= 7)
1118                        hdr[--pos] = 128 | (--ofs & 127);
1119                sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1120                pack_size += sizeof(hdr) - pos;
1121        } else {
1122                e->depth = 0;
1123                hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1124                sha1write(pack_file, hdr, hdrlen);
1125                pack_size += hdrlen;
1126        }
1127
1128        sha1write(pack_file, out, s.total_out);
1129        pack_size += s.total_out;
1130
1131        e->idx.crc32 = crc32_end(pack_file);
1132
1133        free(out);
1134        free(delta);
1135        if (last) {
1136                if (last->no_swap) {
1137                        last->data = *dat;
1138                } else {
1139                        strbuf_swap(&last->data, dat);
1140                }
1141                last->offset = e->idx.offset;
1142                last->depth = e->depth;
1143        }
1144        return 0;
1145}
1146
1147static void truncate_pack(off_t to, git_SHA_CTX *ctx)
1148{
1149        if (ftruncate(pack_data->pack_fd, to)
1150         || lseek(pack_data->pack_fd, to, SEEK_SET) != to)
1151                die_errno("cannot truncate pack to skip duplicate");
1152        pack_size = to;
1153
1154        /* yes this is a layering violation */
1155        pack_file->total = to;
1156        pack_file->offset = 0;
1157        pack_file->ctx = *ctx;
1158}
1159
1160static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1161{
1162        size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1163        unsigned char *in_buf = xmalloc(in_sz);
1164        unsigned char *out_buf = xmalloc(out_sz);
1165        struct object_entry *e;
1166        unsigned char sha1[20];
1167        unsigned long hdrlen;
1168        off_t offset;
1169        git_SHA_CTX c;
1170        git_SHA_CTX pack_file_ctx;
1171        z_stream s;
1172        int status = Z_OK;
1173
1174        /* Determine if we should auto-checkpoint. */
1175        if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1176                || (pack_size + 60 + len) < pack_size)
1177                cycle_packfile();
1178
1179        offset = pack_size;
1180
1181        /* preserve the pack_file SHA1 ctx in case we have to truncate later */
1182        sha1flush(pack_file);
1183        pack_file_ctx = pack_file->ctx;
1184
1185        hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1186        if (out_sz <= hdrlen)
1187                die("impossibly large object header");
1188
1189        git_SHA1_Init(&c);
1190        git_SHA1_Update(&c, out_buf, hdrlen);
1191
1192        crc32_begin(pack_file);
1193
1194        memset(&s, 0, sizeof(s));
1195        deflateInit(&s, pack_compression_level);
1196
1197        hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1198        if (out_sz <= hdrlen)
1199                die("impossibly large object header");
1200
1201        s.next_out = out_buf + hdrlen;
1202        s.avail_out = out_sz - hdrlen;
1203
1204        while (status != Z_STREAM_END) {
1205                if (0 < len && !s.avail_in) {
1206                        size_t cnt = in_sz < len ? in_sz : (size_t)len;
1207                        size_t n = fread(in_buf, 1, cnt, stdin);
1208                        if (!n && feof(stdin))
1209                                die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1210
1211                        git_SHA1_Update(&c, in_buf, n);
1212                        s.next_in = in_buf;
1213                        s.avail_in = n;
1214                        len -= n;
1215                }
1216
1217                status = deflate(&s, len ? 0 : Z_FINISH);
1218
1219                if (!s.avail_out || status == Z_STREAM_END) {
1220                        size_t n = s.next_out - out_buf;
1221                        sha1write(pack_file, out_buf, n);
1222                        pack_size += n;
1223                        s.next_out = out_buf;
1224                        s.avail_out = out_sz;
1225                }
1226
1227                switch (status) {
1228                case Z_OK:
1229                case Z_BUF_ERROR:
1230                case Z_STREAM_END:
1231                        continue;
1232                default:
1233                        die("unexpected deflate failure: %d", status);
1234                }
1235        }
1236        deflateEnd(&s);
1237        git_SHA1_Final(sha1, &c);
1238
1239        if (sha1out)
1240                hashcpy(sha1out, sha1);
1241
1242        e = insert_object(sha1);
1243
1244        if (mark)
1245                insert_mark(mark, e);
1246
1247        if (e->idx.offset) {
1248                duplicate_count_by_type[OBJ_BLOB]++;
1249                truncate_pack(offset, &pack_file_ctx);
1250
1251        } else if (find_sha1_pack(sha1, packed_git)) {
1252                e->type = OBJ_BLOB;
1253                e->pack_id = MAX_PACK_ID;
1254                e->idx.offset = 1; /* just not zero! */
1255                duplicate_count_by_type[OBJ_BLOB]++;
1256                truncate_pack(offset, &pack_file_ctx);
1257
1258        } else {
1259                e->depth = 0;
1260                e->type = OBJ_BLOB;
1261                e->pack_id = pack_id;
1262                e->idx.offset = offset;
1263                e->idx.crc32 = crc32_end(pack_file);
1264                object_count++;
1265                object_count_by_type[OBJ_BLOB]++;
1266        }
1267
1268        free(in_buf);
1269        free(out_buf);
1270}
1271
1272/* All calls must be guarded by find_object() or find_mark() to
1273 * ensure the 'struct object_entry' passed was written by this
1274 * process instance.  We unpack the entry by the offset, avoiding
1275 * the need for the corresponding .idx file.  This unpacking rule
1276 * works because we only use OBJ_REF_DELTA within the packfiles
1277 * created by fast-import.
1278 *
1279 * oe must not be NULL.  Such an oe usually comes from giving
1280 * an unknown SHA-1 to find_object() or an undefined mark to
1281 * find_mark().  Callers must test for this condition and use
1282 * the standard read_sha1_file() when it happens.
1283 *
1284 * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1285 * find_mark(), where the mark was reloaded from an existing marks
1286 * file and is referencing an object that this fast-import process
1287 * instance did not write out to a packfile.  Callers must test for
1288 * this condition and use read_sha1_file() instead.
1289 */
1290static void *gfi_unpack_entry(
1291        struct object_entry *oe,
1292        unsigned long *sizep)
1293{
1294        enum object_type type;
1295        struct packed_git *p = all_packs[oe->pack_id];
1296        if (p == pack_data && p->pack_size < (pack_size + 20)) {
1297                /* The object is stored in the packfile we are writing to
1298                 * and we have modified it since the last time we scanned
1299                 * back to read a previously written object.  If an old
1300                 * window covered [p->pack_size, p->pack_size + 20) its
1301                 * data is stale and is not valid.  Closing all windows
1302                 * and updating the packfile length ensures we can read
1303                 * the newly written data.
1304                 */
1305                close_pack_windows(p);
1306                sha1flush(pack_file);
1307
1308                /* We have to offer 20 bytes additional on the end of
1309                 * the packfile as the core unpacker code assumes the
1310                 * footer is present at the file end and must promise
1311                 * at least 20 bytes within any window it maps.  But
1312                 * we don't actually create the footer here.
1313                 */
1314                p->pack_size = pack_size + 20;
1315        }
1316        return unpack_entry(p, oe->idx.offset, &type, sizep);
1317}
1318
1319static const char *get_mode(const char *str, uint16_t *modep)
1320{
1321        unsigned char c;
1322        uint16_t mode = 0;
1323
1324        while ((c = *str++) != ' ') {
1325                if (c < '0' || c > '7')
1326                        return NULL;
1327                mode = (mode << 3) + (c - '0');
1328        }
1329        *modep = mode;
1330        return str;
1331}
1332
1333static void load_tree(struct tree_entry *root)
1334{
1335        unsigned char *sha1 = root->versions[1].sha1;
1336        struct object_entry *myoe;
1337        struct tree_content *t;
1338        unsigned long size;
1339        char *buf;
1340        const char *c;
1341
1342        root->tree = t = new_tree_content(8);
1343        if (is_null_sha1(sha1))
1344                return;
1345
1346        myoe = find_object(sha1);
1347        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1348                if (myoe->type != OBJ_TREE)
1349                        die("Not a tree: %s", sha1_to_hex(sha1));
1350                t->delta_depth = myoe->depth;
1351                buf = gfi_unpack_entry(myoe, &size);
1352                if (!buf)
1353                        die("Can't load tree %s", sha1_to_hex(sha1));
1354        } else {
1355                enum object_type type;
1356                buf = read_sha1_file(sha1, &type, &size);
1357                if (!buf || type != OBJ_TREE)
1358                        die("Can't load tree %s", sha1_to_hex(sha1));
1359        }
1360
1361        c = buf;
1362        while (c != (buf + size)) {
1363                struct tree_entry *e = new_tree_entry();
1364
1365                if (t->entry_count == t->entry_capacity)
1366                        root->tree = t = grow_tree_content(t, t->entry_count);
1367                t->entries[t->entry_count++] = e;
1368
1369                e->tree = NULL;
1370                c = get_mode(c, &e->versions[1].mode);
1371                if (!c)
1372                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1373                e->versions[0].mode = e->versions[1].mode;
1374                e->name = to_atom(c, strlen(c));
1375                c += e->name->str_len + 1;
1376                hashcpy(e->versions[0].sha1, (unsigned char *)c);
1377                hashcpy(e->versions[1].sha1, (unsigned char *)c);
1378                c += 20;
1379        }
1380        free(buf);
1381}
1382
1383static int tecmp0 (const void *_a, const void *_b)
1384{
1385        struct tree_entry *a = *((struct tree_entry**)_a);
1386        struct tree_entry *b = *((struct tree_entry**)_b);
1387        return base_name_compare(
1388                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1389                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1390}
1391
1392static int tecmp1 (const void *_a, const void *_b)
1393{
1394        struct tree_entry *a = *((struct tree_entry**)_a);
1395        struct tree_entry *b = *((struct tree_entry**)_b);
1396        return base_name_compare(
1397                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1398                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1399}
1400
1401static void mktree(struct tree_content *t, int v, struct strbuf *b)
1402{
1403        size_t maxlen = 0;
1404        unsigned int i;
1405
1406        if (!v)
1407                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1408        else
1409                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1410
1411        for (i = 0; i < t->entry_count; i++) {
1412                if (t->entries[i]->versions[v].mode)
1413                        maxlen += t->entries[i]->name->str_len + 34;
1414        }
1415
1416        strbuf_reset(b);
1417        strbuf_grow(b, maxlen);
1418        for (i = 0; i < t->entry_count; i++) {
1419                struct tree_entry *e = t->entries[i];
1420                if (!e->versions[v].mode)
1421                        continue;
1422                strbuf_addf(b, "%o %s%c",
1423                        (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1424                        e->name->str_dat, '\0');
1425                strbuf_add(b, e->versions[v].sha1, 20);
1426        }
1427}
1428
1429static void store_tree(struct tree_entry *root)
1430{
1431        struct tree_content *t = root->tree;
1432        unsigned int i, j, del;
1433        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1434        struct object_entry *le = NULL;
1435
1436        if (!is_null_sha1(root->versions[1].sha1))
1437                return;
1438
1439        for (i = 0; i < t->entry_count; i++) {
1440                if (t->entries[i]->tree)
1441                        store_tree(t->entries[i]);
1442        }
1443
1444        if (!(root->versions[0].mode & NO_DELTA))
1445                le = find_object(root->versions[0].sha1);
1446        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1447                mktree(t, 0, &old_tree);
1448                lo.data = old_tree;
1449                lo.offset = le->idx.offset;
1450                lo.depth = t->delta_depth;
1451        }
1452
1453        mktree(t, 1, &new_tree);
1454        store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1455
1456        t->delta_depth = lo.depth;
1457        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1458                struct tree_entry *e = t->entries[i];
1459                if (e->versions[1].mode) {
1460                        e->versions[0].mode = e->versions[1].mode;
1461                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1462                        t->entries[j++] = e;
1463                } else {
1464                        release_tree_entry(e);
1465                        del++;
1466                }
1467        }
1468        t->entry_count -= del;
1469}
1470
1471static void tree_content_replace(
1472        struct tree_entry *root,
1473        const unsigned char *sha1,
1474        const uint16_t mode,
1475        struct tree_content *newtree)
1476{
1477        if (!S_ISDIR(mode))
1478                die("Root cannot be a non-directory");
1479        hashclr(root->versions[0].sha1);
1480        hashcpy(root->versions[1].sha1, sha1);
1481        if (root->tree)
1482                release_tree_content_recursive(root->tree);
1483        root->tree = newtree;
1484}
1485
1486static int tree_content_set(
1487        struct tree_entry *root,
1488        const char *p,
1489        const unsigned char *sha1,
1490        const uint16_t mode,
1491        struct tree_content *subtree)
1492{
1493        struct tree_content *t;
1494        const char *slash1;
1495        unsigned int i, n;
1496        struct tree_entry *e;
1497
1498        slash1 = strchr(p, '/');
1499        if (slash1)
1500                n = slash1 - p;
1501        else
1502                n = strlen(p);
1503        if (!n)
1504                die("Empty path component found in input");
1505        if (!slash1 && !S_ISDIR(mode) && subtree)
1506                die("Non-directories cannot have subtrees");
1507
1508        if (!root->tree)
1509                load_tree(root);
1510        t = root->tree;
1511        for (i = 0; i < t->entry_count; i++) {
1512                e = t->entries[i];
1513                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1514                        if (!slash1) {
1515                                if (!S_ISDIR(mode)
1516                                                && e->versions[1].mode == mode
1517                                                && !hashcmp(e->versions[1].sha1, sha1))
1518                                        return 0;
1519                                e->versions[1].mode = mode;
1520                                hashcpy(e->versions[1].sha1, sha1);
1521                                if (e->tree)
1522                                        release_tree_content_recursive(e->tree);
1523                                e->tree = subtree;
1524
1525                                /*
1526                                 * We need to leave e->versions[0].sha1 alone
1527                                 * to avoid modifying the preimage tree used
1528                                 * when writing out the parent directory.
1529                                 * But after replacing the subdir with a
1530                                 * completely different one, it's not a good
1531                                 * delta base any more, and besides, we've
1532                                 * thrown away the tree entries needed to
1533                                 * make a delta against it.
1534                                 *
1535                                 * So let's just explicitly disable deltas
1536                                 * for the subtree.
1537                                 */
1538                                if (S_ISDIR(e->versions[0].mode))
1539                                        e->versions[0].mode |= NO_DELTA;
1540
1541                                hashclr(root->versions[1].sha1);
1542                                return 1;
1543                        }
1544                        if (!S_ISDIR(e->versions[1].mode)) {
1545                                e->tree = new_tree_content(8);
1546                                e->versions[1].mode = S_IFDIR;
1547                        }
1548                        if (!e->tree)
1549                                load_tree(e);
1550                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1551                                hashclr(root->versions[1].sha1);
1552                                return 1;
1553                        }
1554                        return 0;
1555                }
1556        }
1557
1558        if (t->entry_count == t->entry_capacity)
1559                root->tree = t = grow_tree_content(t, t->entry_count);
1560        e = new_tree_entry();
1561        e->name = to_atom(p, n);
1562        e->versions[0].mode = 0;
1563        hashclr(e->versions[0].sha1);
1564        t->entries[t->entry_count++] = e;
1565        if (slash1) {
1566                e->tree = new_tree_content(8);
1567                e->versions[1].mode = S_IFDIR;
1568                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1569        } else {
1570                e->tree = subtree;
1571                e->versions[1].mode = mode;
1572                hashcpy(e->versions[1].sha1, sha1);
1573        }
1574        hashclr(root->versions[1].sha1);
1575        return 1;
1576}
1577
1578static int tree_content_remove(
1579        struct tree_entry *root,
1580        const char *p,
1581        struct tree_entry *backup_leaf)
1582{
1583        struct tree_content *t;
1584        const char *slash1;
1585        unsigned int i, n;
1586        struct tree_entry *e;
1587
1588        slash1 = strchr(p, '/');
1589        if (slash1)
1590                n = slash1 - p;
1591        else
1592                n = strlen(p);
1593
1594        if (!root->tree)
1595                load_tree(root);
1596        t = root->tree;
1597        for (i = 0; i < t->entry_count; i++) {
1598                e = t->entries[i];
1599                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1600                        if (slash1 && !S_ISDIR(e->versions[1].mode))
1601                                /*
1602                                 * If p names a file in some subdirectory, and a
1603                                 * file or symlink matching the name of the
1604                                 * parent directory of p exists, then p cannot
1605                                 * exist and need not be deleted.
1606                                 */
1607                                return 1;
1608                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1609                                goto del_entry;
1610                        if (!e->tree)
1611                                load_tree(e);
1612                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1613                                for (n = 0; n < e->tree->entry_count; n++) {
1614                                        if (e->tree->entries[n]->versions[1].mode) {
1615                                                hashclr(root->versions[1].sha1);
1616                                                return 1;
1617                                        }
1618                                }
1619                                backup_leaf = NULL;
1620                                goto del_entry;
1621                        }
1622                        return 0;
1623                }
1624        }
1625        return 0;
1626
1627del_entry:
1628        if (backup_leaf)
1629                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1630        else if (e->tree)
1631                release_tree_content_recursive(e->tree);
1632        e->tree = NULL;
1633        e->versions[1].mode = 0;
1634        hashclr(e->versions[1].sha1);
1635        hashclr(root->versions[1].sha1);
1636        return 1;
1637}
1638
1639static int tree_content_get(
1640        struct tree_entry *root,
1641        const char *p,
1642        struct tree_entry *leaf)
1643{
1644        struct tree_content *t;
1645        const char *slash1;
1646        unsigned int i, n;
1647        struct tree_entry *e;
1648
1649        slash1 = strchr(p, '/');
1650        if (slash1)
1651                n = slash1 - p;
1652        else
1653                n = strlen(p);
1654
1655        if (!root->tree)
1656                load_tree(root);
1657        t = root->tree;
1658        for (i = 0; i < t->entry_count; i++) {
1659                e = t->entries[i];
1660                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1661                        if (!slash1) {
1662                                memcpy(leaf, e, sizeof(*leaf));
1663                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1664                                        leaf->tree = dup_tree_content(e->tree);
1665                                else
1666                                        leaf->tree = NULL;
1667                                return 1;
1668                        }
1669                        if (!S_ISDIR(e->versions[1].mode))
1670                                return 0;
1671                        if (!e->tree)
1672                                load_tree(e);
1673                        return tree_content_get(e, slash1 + 1, leaf);
1674                }
1675        }
1676        return 0;
1677}
1678
1679static int update_branch(struct branch *b)
1680{
1681        static const char *msg = "fast-import";
1682        struct ref_lock *lock;
1683        unsigned char old_sha1[20];
1684
1685        if (is_null_sha1(b->sha1))
1686                return 0;
1687        if (read_ref(b->name, old_sha1))
1688                hashclr(old_sha1);
1689        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1690        if (!lock)
1691                return error("Unable to lock %s", b->name);
1692        if (!force_update && !is_null_sha1(old_sha1)) {
1693                struct commit *old_cmit, *new_cmit;
1694
1695                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1696                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1697                if (!old_cmit || !new_cmit) {
1698                        unlock_ref(lock);
1699                        return error("Branch %s is missing commits.", b->name);
1700                }
1701
1702                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1703                        unlock_ref(lock);
1704                        warning("Not updating %s"
1705                                " (new tip %s does not contain %s)",
1706                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1707                        return -1;
1708                }
1709        }
1710        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1711                return error("Unable to update %s", b->name);
1712        return 0;
1713}
1714
1715static void dump_branches(void)
1716{
1717        unsigned int i;
1718        struct branch *b;
1719
1720        for (i = 0; i < branch_table_sz; i++) {
1721                for (b = branch_table[i]; b; b = b->table_next_branch)
1722                        failure |= update_branch(b);
1723        }
1724}
1725
1726static void dump_tags(void)
1727{
1728        static const char *msg = "fast-import";
1729        struct tag *t;
1730        struct ref_lock *lock;
1731        char ref_name[PATH_MAX];
1732
1733        for (t = first_tag; t; t = t->next_tag) {
1734                sprintf(ref_name, "tags/%s", t->name);
1735                lock = lock_ref_sha1(ref_name, NULL);
1736                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1737                        failure |= error("Unable to update %s", ref_name);
1738        }
1739}
1740
1741static void dump_marks_helper(FILE *f,
1742        uintmax_t base,
1743        struct mark_set *m)
1744{
1745        uintmax_t k;
1746        if (m->shift) {
1747                for (k = 0; k < 1024; k++) {
1748                        if (m->data.sets[k])
1749                                dump_marks_helper(f, base + (k << m->shift),
1750                                        m->data.sets[k]);
1751                }
1752        } else {
1753                for (k = 0; k < 1024; k++) {
1754                        if (m->data.marked[k])
1755                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1756                                        sha1_to_hex(m->data.marked[k]->idx.sha1));
1757                }
1758        }
1759}
1760
1761static void dump_marks(void)
1762{
1763        static struct lock_file mark_lock;
1764        int mark_fd;
1765        FILE *f;
1766
1767        if (!export_marks_file)
1768                return;
1769
1770        mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1771        if (mark_fd < 0) {
1772                failure |= error("Unable to write marks file %s: %s",
1773                        export_marks_file, strerror(errno));
1774                return;
1775        }
1776
1777        f = fdopen(mark_fd, "w");
1778        if (!f) {
1779                int saved_errno = errno;
1780                rollback_lock_file(&mark_lock);
1781                failure |= error("Unable to write marks file %s: %s",
1782                        export_marks_file, strerror(saved_errno));
1783                return;
1784        }
1785
1786        /*
1787         * Since the lock file was fdopen()'ed, it should not be close()'ed.
1788         * Assign -1 to the lock file descriptor so that commit_lock_file()
1789         * won't try to close() it.
1790         */
1791        mark_lock.fd = -1;
1792
1793        dump_marks_helper(f, 0, marks);
1794        if (ferror(f) || fclose(f)) {
1795                int saved_errno = errno;
1796                rollback_lock_file(&mark_lock);
1797                failure |= error("Unable to write marks file %s: %s",
1798                        export_marks_file, strerror(saved_errno));
1799                return;
1800        }
1801
1802        if (commit_lock_file(&mark_lock)) {
1803                int saved_errno = errno;
1804                rollback_lock_file(&mark_lock);
1805                failure |= error("Unable to commit marks file %s: %s",
1806                        export_marks_file, strerror(saved_errno));
1807                return;
1808        }
1809}
1810
1811static void read_marks(void)
1812{
1813        char line[512];
1814        FILE *f = fopen(import_marks_file, "r");
1815        if (f)
1816                ;
1817        else if (import_marks_file_ignore_missing && errno == ENOENT)
1818                return; /* Marks file does not exist */
1819        else
1820                die_errno("cannot read '%s'", import_marks_file);
1821        while (fgets(line, sizeof(line), f)) {
1822                uintmax_t mark;
1823                char *end;
1824                unsigned char sha1[20];
1825                struct object_entry *e;
1826
1827                end = strchr(line, '\n');
1828                if (line[0] != ':' || !end)
1829                        die("corrupt mark line: %s", line);
1830                *end = 0;
1831                mark = strtoumax(line + 1, &end, 10);
1832                if (!mark || end == line + 1
1833                        || *end != ' ' || get_sha1(end + 1, sha1))
1834                        die("corrupt mark line: %s", line);
1835                e = find_object(sha1);
1836                if (!e) {
1837                        enum object_type type = sha1_object_info(sha1, NULL);
1838                        if (type < 0)
1839                                die("object not found: %s", sha1_to_hex(sha1));
1840                        e = insert_object(sha1);
1841                        e->type = type;
1842                        e->pack_id = MAX_PACK_ID;
1843                        e->idx.offset = 1; /* just not zero! */
1844                }
1845                insert_mark(mark, e);
1846        }
1847        fclose(f);
1848}
1849
1850
1851static int read_next_command(void)
1852{
1853        static int stdin_eof = 0;
1854
1855        if (stdin_eof) {
1856                unread_command_buf = 0;
1857                return EOF;
1858        }
1859
1860        for (;;) {
1861                if (unread_command_buf) {
1862                        unread_command_buf = 0;
1863                } else {
1864                        struct recent_command *rc;
1865
1866                        strbuf_detach(&command_buf, NULL);
1867                        stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1868                        if (stdin_eof)
1869                                return EOF;
1870
1871                        if (!seen_data_command
1872                                && prefixcmp(command_buf.buf, "feature ")
1873                                && prefixcmp(command_buf.buf, "option ")) {
1874                                parse_argv();
1875                        }
1876
1877                        rc = rc_free;
1878                        if (rc)
1879                                rc_free = rc->next;
1880                        else {
1881                                rc = cmd_hist.next;
1882                                cmd_hist.next = rc->next;
1883                                cmd_hist.next->prev = &cmd_hist;
1884                                free(rc->buf);
1885                        }
1886
1887                        rc->buf = command_buf.buf;
1888                        rc->prev = cmd_tail;
1889                        rc->next = cmd_hist.prev;
1890                        rc->prev->next = rc;
1891                        cmd_tail = rc;
1892                }
1893                if (!prefixcmp(command_buf.buf, "cat-blob ")) {
1894                        parse_cat_blob();
1895                        continue;
1896                }
1897                if (command_buf.buf[0] == '#')
1898                        continue;
1899                return 0;
1900        }
1901}
1902
1903static void skip_optional_lf(void)
1904{
1905        int term_char = fgetc(stdin);
1906        if (term_char != '\n' && term_char != EOF)
1907                ungetc(term_char, stdin);
1908}
1909
1910static void parse_mark(void)
1911{
1912        if (!prefixcmp(command_buf.buf, "mark :")) {
1913                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1914                read_next_command();
1915        }
1916        else
1917                next_mark = 0;
1918}
1919
1920static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1921{
1922        strbuf_reset(sb);
1923
1924        if (prefixcmp(command_buf.buf, "data "))
1925                die("Expected 'data n' command, found: %s", command_buf.buf);
1926
1927        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1928                char *term = xstrdup(command_buf.buf + 5 + 2);
1929                size_t term_len = command_buf.len - 5 - 2;
1930
1931                strbuf_detach(&command_buf, NULL);
1932                for (;;) {
1933                        if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1934                                die("EOF in data (terminator '%s' not found)", term);
1935                        if (term_len == command_buf.len
1936                                && !strcmp(term, command_buf.buf))
1937                                break;
1938                        strbuf_addbuf(sb, &command_buf);
1939                        strbuf_addch(sb, '\n');
1940                }
1941                free(term);
1942        }
1943        else {
1944                uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1945                size_t n = 0, length = (size_t)len;
1946
1947                if (limit && limit < len) {
1948                        *len_res = len;
1949                        return 0;
1950                }
1951                if (length < len)
1952                        die("data is too large to use in this context");
1953
1954                while (n < length) {
1955                        size_t s = strbuf_fread(sb, length - n, stdin);
1956                        if (!s && feof(stdin))
1957                                die("EOF in data (%lu bytes remaining)",
1958                                        (unsigned long)(length - n));
1959                        n += s;
1960                }
1961        }
1962
1963        skip_optional_lf();
1964        return 1;
1965}
1966
1967static int validate_raw_date(const char *src, char *result, int maxlen)
1968{
1969        const char *orig_src = src;
1970        char *endp;
1971        unsigned long num;
1972
1973        errno = 0;
1974
1975        num = strtoul(src, &endp, 10);
1976        /* NEEDSWORK: perhaps check for reasonable values? */
1977        if (errno || endp == src || *endp != ' ')
1978                return -1;
1979
1980        src = endp + 1;
1981        if (*src != '-' && *src != '+')
1982                return -1;
1983
1984        num = strtoul(src + 1, &endp, 10);
1985        if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1986            1400 < num)
1987                return -1;
1988
1989        strcpy(result, orig_src);
1990        return 0;
1991}
1992
1993static char *parse_ident(const char *buf)
1994{
1995        const char *gt;
1996        size_t name_len;
1997        char *ident;
1998
1999        gt = strrchr(buf, '>');
2000        if (!gt)
2001                die("Missing > in ident string: %s", buf);
2002        gt++;
2003        if (*gt != ' ')
2004                die("Missing space after > in ident string: %s", buf);
2005        gt++;
2006        name_len = gt - buf;
2007        ident = xmalloc(name_len + 24);
2008        strncpy(ident, buf, name_len);
2009
2010        switch (whenspec) {
2011        case WHENSPEC_RAW:
2012                if (validate_raw_date(gt, ident + name_len, 24) < 0)
2013                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
2014                break;
2015        case WHENSPEC_RFC2822:
2016                if (parse_date(gt, ident + name_len, 24) < 0)
2017                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
2018                break;
2019        case WHENSPEC_NOW:
2020                if (strcmp("now", gt))
2021                        die("Date in ident must be 'now': %s", buf);
2022                datestamp(ident + name_len, 24);
2023                break;
2024        }
2025
2026        return ident;
2027}
2028
2029static void parse_and_store_blob(
2030        struct last_object *last,
2031        unsigned char *sha1out,
2032        uintmax_t mark)
2033{
2034        static struct strbuf buf = STRBUF_INIT;
2035        uintmax_t len;
2036
2037        if (parse_data(&buf, big_file_threshold, &len))
2038                store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2039        else {
2040                if (last) {
2041                        strbuf_release(&last->data);
2042                        last->offset = 0;
2043                        last->depth = 0;
2044                }
2045                stream_blob(len, sha1out, mark);
2046                skip_optional_lf();
2047        }
2048}
2049
2050static void parse_new_blob(void)
2051{
2052        read_next_command();
2053        parse_mark();
2054        parse_and_store_blob(&last_blob, NULL, next_mark);
2055}
2056
2057static void unload_one_branch(void)
2058{
2059        while (cur_active_branches
2060                && cur_active_branches >= max_active_branches) {
2061                uintmax_t min_commit = ULONG_MAX;
2062                struct branch *e, *l = NULL, *p = NULL;
2063
2064                for (e = active_branches; e; e = e->active_next_branch) {
2065                        if (e->last_commit < min_commit) {
2066                                p = l;
2067                                min_commit = e->last_commit;
2068                        }
2069                        l = e;
2070                }
2071
2072                if (p) {
2073                        e = p->active_next_branch;
2074                        p->active_next_branch = e->active_next_branch;
2075                } else {
2076                        e = active_branches;
2077                        active_branches = e->active_next_branch;
2078                }
2079                e->active = 0;
2080                e->active_next_branch = NULL;
2081                if (e->branch_tree.tree) {
2082                        release_tree_content_recursive(e->branch_tree.tree);
2083                        e->branch_tree.tree = NULL;
2084                }
2085                cur_active_branches--;
2086        }
2087}
2088
2089static void load_branch(struct branch *b)
2090{
2091        load_tree(&b->branch_tree);
2092        if (!b->active) {
2093                b->active = 1;
2094                b->active_next_branch = active_branches;
2095                active_branches = b;
2096                cur_active_branches++;
2097                branch_load_count++;
2098        }
2099}
2100
2101static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2102{
2103        unsigned char fanout = 0;
2104        while ((num_notes >>= 8))
2105                fanout++;
2106        return fanout;
2107}
2108
2109static void construct_path_with_fanout(const char *hex_sha1,
2110                unsigned char fanout, char *path)
2111{
2112        unsigned int i = 0, j = 0;
2113        if (fanout >= 20)
2114                die("Too large fanout (%u)", fanout);
2115        while (fanout) {
2116                path[i++] = hex_sha1[j++];
2117                path[i++] = hex_sha1[j++];
2118                path[i++] = '/';
2119                fanout--;
2120        }
2121        memcpy(path + i, hex_sha1 + j, 40 - j);
2122        path[i + 40 - j] = '\0';
2123}
2124
2125static uintmax_t do_change_note_fanout(
2126                struct tree_entry *orig_root, struct tree_entry *root,
2127                char *hex_sha1, unsigned int hex_sha1_len,
2128                char *fullpath, unsigned int fullpath_len,
2129                unsigned char fanout)
2130{
2131        struct tree_content *t = root->tree;
2132        struct tree_entry *e, leaf;
2133        unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2134        uintmax_t num_notes = 0;
2135        unsigned char sha1[20];
2136        char realpath[60];
2137
2138        for (i = 0; t && i < t->entry_count; i++) {
2139                e = t->entries[i];
2140                tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2141                tmp_fullpath_len = fullpath_len;
2142
2143                /*
2144                 * We're interested in EITHER existing note entries (entries
2145                 * with exactly 40 hex chars in path, not including directory
2146                 * separators), OR directory entries that may contain note
2147                 * entries (with < 40 hex chars in path).
2148                 * Also, each path component in a note entry must be a multiple
2149                 * of 2 chars.
2150                 */
2151                if (!e->versions[1].mode ||
2152                    tmp_hex_sha1_len > 40 ||
2153                    e->name->str_len % 2)
2154                        continue;
2155
2156                /* This _may_ be a note entry, or a subdir containing notes */
2157                memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2158                       e->name->str_len);
2159                if (tmp_fullpath_len)
2160                        fullpath[tmp_fullpath_len++] = '/';
2161                memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2162                       e->name->str_len);
2163                tmp_fullpath_len += e->name->str_len;
2164                fullpath[tmp_fullpath_len] = '\0';
2165
2166                if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2167                        /* This is a note entry */
2168                        construct_path_with_fanout(hex_sha1, fanout, realpath);
2169                        if (!strcmp(fullpath, realpath)) {
2170                                /* Note entry is in correct location */
2171                                num_notes++;
2172                                continue;
2173                        }
2174
2175                        /* Rename fullpath to realpath */
2176                        if (!tree_content_remove(orig_root, fullpath, &leaf))
2177                                die("Failed to remove path %s", fullpath);
2178                        tree_content_set(orig_root, realpath,
2179                                leaf.versions[1].sha1,
2180                                leaf.versions[1].mode,
2181                                leaf.tree);
2182                } else if (S_ISDIR(e->versions[1].mode)) {
2183                        /* This is a subdir that may contain note entries */
2184                        if (!e->tree)
2185                                load_tree(e);
2186                        num_notes += do_change_note_fanout(orig_root, e,
2187                                hex_sha1, tmp_hex_sha1_len,
2188                                fullpath, tmp_fullpath_len, fanout);
2189                }
2190
2191                /* The above may have reallocated the current tree_content */
2192                t = root->tree;
2193        }
2194        return num_notes;
2195}
2196
2197static uintmax_t change_note_fanout(struct tree_entry *root,
2198                unsigned char fanout)
2199{
2200        char hex_sha1[40], path[60];
2201        return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2202}
2203
2204static void file_change_m(struct branch *b)
2205{
2206        const char *p = command_buf.buf + 2;
2207        static struct strbuf uq = STRBUF_INIT;
2208        const char *endp;
2209        struct object_entry *oe = oe;
2210        unsigned char sha1[20];
2211        uint16_t mode, inline_data = 0;
2212
2213        p = get_mode(p, &mode);
2214        if (!p)
2215                die("Corrupt mode: %s", command_buf.buf);
2216        switch (mode) {
2217        case 0644:
2218        case 0755:
2219                mode |= S_IFREG;
2220        case S_IFREG | 0644:
2221        case S_IFREG | 0755:
2222        case S_IFLNK:
2223        case S_IFDIR:
2224        case S_IFGITLINK:
2225                /* ok */
2226                break;
2227        default:
2228                die("Corrupt mode: %s", command_buf.buf);
2229        }
2230
2231        if (*p == ':') {
2232                char *x;
2233                oe = find_mark(strtoumax(p + 1, &x, 10));
2234                hashcpy(sha1, oe->idx.sha1);
2235                p = x;
2236        } else if (!prefixcmp(p, "inline")) {
2237                inline_data = 1;
2238                p += 6;
2239        } else {
2240                if (get_sha1_hex(p, sha1))
2241                        die("Invalid SHA1: %s", command_buf.buf);
2242                oe = find_object(sha1);
2243                p += 40;
2244        }
2245        if (*p++ != ' ')
2246                die("Missing space after SHA1: %s", command_buf.buf);
2247
2248        strbuf_reset(&uq);
2249        if (!unquote_c_style(&uq, p, &endp)) {
2250                if (*endp)
2251                        die("Garbage after path in: %s", command_buf.buf);
2252                p = uq.buf;
2253        }
2254
2255        /* Git does not track empty, non-toplevel directories. */
2256        if (S_ISDIR(mode) && !memcmp(sha1, EMPTY_TREE_SHA1_BIN, 20) && *p) {
2257                tree_content_remove(&b->branch_tree, p, NULL);
2258                return;
2259        }
2260
2261        if (S_ISGITLINK(mode)) {
2262                if (inline_data)
2263                        die("Git links cannot be specified 'inline': %s",
2264                                command_buf.buf);
2265                else if (oe) {
2266                        if (oe->type != OBJ_COMMIT)
2267                                die("Not a commit (actually a %s): %s",
2268                                        typename(oe->type), command_buf.buf);
2269                }
2270                /*
2271                 * Accept the sha1 without checking; it expected to be in
2272                 * another repository.
2273                 */
2274        } else if (inline_data) {
2275                if (S_ISDIR(mode))
2276                        die("Directories cannot be specified 'inline': %s",
2277                                command_buf.buf);
2278                if (p != uq.buf) {
2279                        strbuf_addstr(&uq, p);
2280                        p = uq.buf;
2281                }
2282                read_next_command();
2283                parse_and_store_blob(&last_blob, sha1, 0);
2284        } else {
2285                enum object_type expected = S_ISDIR(mode) ?
2286                                                OBJ_TREE: OBJ_BLOB;
2287                enum object_type type = oe ? oe->type :
2288                                        sha1_object_info(sha1, NULL);
2289                if (type < 0)
2290                        die("%s not found: %s",
2291                                        S_ISDIR(mode) ?  "Tree" : "Blob",
2292                                        command_buf.buf);
2293                if (type != expected)
2294                        die("Not a %s (actually a %s): %s",
2295                                typename(expected), typename(type),
2296                                command_buf.buf);
2297        }
2298
2299        if (!*p) {
2300                tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2301                return;
2302        }
2303        tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2304}
2305
2306static void file_change_d(struct branch *b)
2307{
2308        const char *p = command_buf.buf + 2;
2309        static struct strbuf uq = STRBUF_INIT;
2310        const char *endp;
2311
2312        strbuf_reset(&uq);
2313        if (!unquote_c_style(&uq, p, &endp)) {
2314                if (*endp)
2315                        die("Garbage after path in: %s", command_buf.buf);
2316                p = uq.buf;
2317        }
2318        tree_content_remove(&b->branch_tree, p, NULL);
2319}
2320
2321static void file_change_cr(struct branch *b, int rename)
2322{
2323        const char *s, *d;
2324        static struct strbuf s_uq = STRBUF_INIT;
2325        static struct strbuf d_uq = STRBUF_INIT;
2326        const char *endp;
2327        struct tree_entry leaf;
2328
2329        s = command_buf.buf + 2;
2330        strbuf_reset(&s_uq);
2331        if (!unquote_c_style(&s_uq, s, &endp)) {
2332                if (*endp != ' ')
2333                        die("Missing space after source: %s", command_buf.buf);
2334        } else {
2335                endp = strchr(s, ' ');
2336                if (!endp)
2337                        die("Missing space after source: %s", command_buf.buf);
2338                strbuf_add(&s_uq, s, endp - s);
2339        }
2340        s = s_uq.buf;
2341
2342        endp++;
2343        if (!*endp)
2344                die("Missing dest: %s", command_buf.buf);
2345
2346        d = endp;
2347        strbuf_reset(&d_uq);
2348        if (!unquote_c_style(&d_uq, d, &endp)) {
2349                if (*endp)
2350                        die("Garbage after dest in: %s", command_buf.buf);
2351                d = d_uq.buf;
2352        }
2353
2354        memset(&leaf, 0, sizeof(leaf));
2355        if (rename)
2356                tree_content_remove(&b->branch_tree, s, &leaf);
2357        else
2358                tree_content_get(&b->branch_tree, s, &leaf);
2359        if (!leaf.versions[1].mode)
2360                die("Path %s not in branch", s);
2361        if (!*d) {      /* C "path/to/subdir" "" */
2362                tree_content_replace(&b->branch_tree,
2363                        leaf.versions[1].sha1,
2364                        leaf.versions[1].mode,
2365                        leaf.tree);
2366                return;
2367        }
2368        tree_content_set(&b->branch_tree, d,
2369                leaf.versions[1].sha1,
2370                leaf.versions[1].mode,
2371                leaf.tree);
2372}
2373
2374static void note_change_n(struct branch *b, unsigned char old_fanout)
2375{
2376        const char *p = command_buf.buf + 2;
2377        static struct strbuf uq = STRBUF_INIT;
2378        struct object_entry *oe = oe;
2379        struct branch *s;
2380        unsigned char sha1[20], commit_sha1[20];
2381        char path[60];
2382        uint16_t inline_data = 0;
2383        unsigned char new_fanout;
2384
2385        /* <dataref> or 'inline' */
2386        if (*p == ':') {
2387                char *x;
2388                oe = find_mark(strtoumax(p + 1, &x, 10));
2389                hashcpy(sha1, oe->idx.sha1);
2390                p = x;
2391        } else if (!prefixcmp(p, "inline")) {
2392                inline_data = 1;
2393                p += 6;
2394        } else {
2395                if (get_sha1_hex(p, sha1))
2396                        die("Invalid SHA1: %s", command_buf.buf);
2397                oe = find_object(sha1);
2398                p += 40;
2399        }
2400        if (*p++ != ' ')
2401                die("Missing space after SHA1: %s", command_buf.buf);
2402
2403        /* <committish> */
2404        s = lookup_branch(p);
2405        if (s) {
2406                hashcpy(commit_sha1, s->sha1);
2407        } else if (*p == ':') {
2408                uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2409                struct object_entry *commit_oe = find_mark(commit_mark);
2410                if (commit_oe->type != OBJ_COMMIT)
2411                        die("Mark :%" PRIuMAX " not a commit", commit_mark);
2412                hashcpy(commit_sha1, commit_oe->idx.sha1);
2413        } else if (!get_sha1(p, commit_sha1)) {
2414                unsigned long size;
2415                char *buf = read_object_with_reference(commit_sha1,
2416                        commit_type, &size, commit_sha1);
2417                if (!buf || size < 46)
2418                        die("Not a valid commit: %s", p);
2419                free(buf);
2420        } else
2421                die("Invalid ref name or SHA1 expression: %s", p);
2422
2423        if (inline_data) {
2424                if (p != uq.buf) {
2425                        strbuf_addstr(&uq, p);
2426                        p = uq.buf;
2427                }
2428                read_next_command();
2429                parse_and_store_blob(&last_blob, sha1, 0);
2430        } else if (oe) {
2431                if (oe->type != OBJ_BLOB)
2432                        die("Not a blob (actually a %s): %s",
2433                                typename(oe->type), command_buf.buf);
2434        } else if (!is_null_sha1(sha1)) {
2435                enum object_type type = sha1_object_info(sha1, NULL);
2436                if (type < 0)
2437                        die("Blob not found: %s", command_buf.buf);
2438                if (type != OBJ_BLOB)
2439                        die("Not a blob (actually a %s): %s",
2440                            typename(type), command_buf.buf);
2441        }
2442
2443        construct_path_with_fanout(sha1_to_hex(commit_sha1), old_fanout, path);
2444        if (tree_content_remove(&b->branch_tree, path, NULL))
2445                b->num_notes--;
2446
2447        if (is_null_sha1(sha1))
2448                return; /* nothing to insert */
2449
2450        b->num_notes++;
2451        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2452        construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2453        tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2454}
2455
2456static void file_change_deleteall(struct branch *b)
2457{
2458        release_tree_content_recursive(b->branch_tree.tree);
2459        hashclr(b->branch_tree.versions[0].sha1);
2460        hashclr(b->branch_tree.versions[1].sha1);
2461        load_tree(&b->branch_tree);
2462        b->num_notes = 0;
2463}
2464
2465static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2466{
2467        if (!buf || size < 46)
2468                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2469        if (memcmp("tree ", buf, 5)
2470                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2471                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2472        hashcpy(b->branch_tree.versions[0].sha1,
2473                b->branch_tree.versions[1].sha1);
2474}
2475
2476static void parse_from_existing(struct branch *b)
2477{
2478        if (is_null_sha1(b->sha1)) {
2479                hashclr(b->branch_tree.versions[0].sha1);
2480                hashclr(b->branch_tree.versions[1].sha1);
2481        } else {
2482                unsigned long size;
2483                char *buf;
2484
2485                buf = read_object_with_reference(b->sha1,
2486                        commit_type, &size, b->sha1);
2487                parse_from_commit(b, buf, size);
2488                free(buf);
2489        }
2490}
2491
2492static int parse_from(struct branch *b)
2493{
2494        const char *from;
2495        struct branch *s;
2496
2497        if (prefixcmp(command_buf.buf, "from "))
2498                return 0;
2499
2500        if (b->branch_tree.tree) {
2501                release_tree_content_recursive(b->branch_tree.tree);
2502                b->branch_tree.tree = NULL;
2503        }
2504
2505        from = strchr(command_buf.buf, ' ') + 1;
2506        s = lookup_branch(from);
2507        if (b == s)
2508                die("Can't create a branch from itself: %s", b->name);
2509        else if (s) {
2510                unsigned char *t = s->branch_tree.versions[1].sha1;
2511                hashcpy(b->sha1, s->sha1);
2512                hashcpy(b->branch_tree.versions[0].sha1, t);
2513                hashcpy(b->branch_tree.versions[1].sha1, t);
2514        } else if (*from == ':') {
2515                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2516                struct object_entry *oe = find_mark(idnum);
2517                if (oe->type != OBJ_COMMIT)
2518                        die("Mark :%" PRIuMAX " not a commit", idnum);
2519                hashcpy(b->sha1, oe->idx.sha1);
2520                if (oe->pack_id != MAX_PACK_ID) {
2521                        unsigned long size;
2522                        char *buf = gfi_unpack_entry(oe, &size);
2523                        parse_from_commit(b, buf, size);
2524                        free(buf);
2525                } else
2526                        parse_from_existing(b);
2527        } else if (!get_sha1(from, b->sha1))
2528                parse_from_existing(b);
2529        else
2530                die("Invalid ref name or SHA1 expression: %s", from);
2531
2532        read_next_command();
2533        return 1;
2534}
2535
2536static struct hash_list *parse_merge(unsigned int *count)
2537{
2538        struct hash_list *list = NULL, *n, *e = e;
2539        const char *from;
2540        struct branch *s;
2541
2542        *count = 0;
2543        while (!prefixcmp(command_buf.buf, "merge ")) {
2544                from = strchr(command_buf.buf, ' ') + 1;
2545                n = xmalloc(sizeof(*n));
2546                s = lookup_branch(from);
2547                if (s)
2548                        hashcpy(n->sha1, s->sha1);
2549                else if (*from == ':') {
2550                        uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2551                        struct object_entry *oe = find_mark(idnum);
2552                        if (oe->type != OBJ_COMMIT)
2553                                die("Mark :%" PRIuMAX " not a commit", idnum);
2554                        hashcpy(n->sha1, oe->idx.sha1);
2555                } else if (!get_sha1(from, n->sha1)) {
2556                        unsigned long size;
2557                        char *buf = read_object_with_reference(n->sha1,
2558                                commit_type, &size, n->sha1);
2559                        if (!buf || size < 46)
2560                                die("Not a valid commit: %s", from);
2561                        free(buf);
2562                } else
2563                        die("Invalid ref name or SHA1 expression: %s", from);
2564
2565                n->next = NULL;
2566                if (list)
2567                        e->next = n;
2568                else
2569                        list = n;
2570                e = n;
2571                (*count)++;
2572                read_next_command();
2573        }
2574        return list;
2575}
2576
2577static void parse_new_commit(void)
2578{
2579        static struct strbuf msg = STRBUF_INIT;
2580        struct branch *b;
2581        char *sp;
2582        char *author = NULL;
2583        char *committer = NULL;
2584        struct hash_list *merge_list = NULL;
2585        unsigned int merge_count;
2586        unsigned char prev_fanout, new_fanout;
2587
2588        /* Obtain the branch name from the rest of our command */
2589        sp = strchr(command_buf.buf, ' ') + 1;
2590        b = lookup_branch(sp);
2591        if (!b)
2592                b = new_branch(sp);
2593
2594        read_next_command();
2595        parse_mark();
2596        if (!prefixcmp(command_buf.buf, "author ")) {
2597                author = parse_ident(command_buf.buf + 7);
2598                read_next_command();
2599        }
2600        if (!prefixcmp(command_buf.buf, "committer ")) {
2601                committer = parse_ident(command_buf.buf + 10);
2602                read_next_command();
2603        }
2604        if (!committer)
2605                die("Expected committer but didn't get one");
2606        parse_data(&msg, 0, NULL);
2607        read_next_command();
2608        parse_from(b);
2609        merge_list = parse_merge(&merge_count);
2610
2611        /* ensure the branch is active/loaded */
2612        if (!b->branch_tree.tree || !max_active_branches) {
2613                unload_one_branch();
2614                load_branch(b);
2615        }
2616
2617        prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2618
2619        /* file_change* */
2620        while (command_buf.len > 0) {
2621                if (!prefixcmp(command_buf.buf, "M "))
2622                        file_change_m(b);
2623                else if (!prefixcmp(command_buf.buf, "D "))
2624                        file_change_d(b);
2625                else if (!prefixcmp(command_buf.buf, "R "))
2626                        file_change_cr(b, 1);
2627                else if (!prefixcmp(command_buf.buf, "C "))
2628                        file_change_cr(b, 0);
2629                else if (!prefixcmp(command_buf.buf, "N "))
2630                        note_change_n(b, prev_fanout);
2631                else if (!strcmp("deleteall", command_buf.buf))
2632                        file_change_deleteall(b);
2633                else if (!prefixcmp(command_buf.buf, "ls "))
2634                        parse_ls(b);
2635                else {
2636                        unread_command_buf = 1;
2637                        break;
2638                }
2639                if (read_next_command() == EOF)
2640                        break;
2641        }
2642
2643        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2644        if (new_fanout != prev_fanout)
2645                b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2646
2647        /* build the tree and the commit */
2648        store_tree(&b->branch_tree);
2649        hashcpy(b->branch_tree.versions[0].sha1,
2650                b->branch_tree.versions[1].sha1);
2651
2652        strbuf_reset(&new_data);
2653        strbuf_addf(&new_data, "tree %s\n",
2654                sha1_to_hex(b->branch_tree.versions[1].sha1));
2655        if (!is_null_sha1(b->sha1))
2656                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2657        while (merge_list) {
2658                struct hash_list *next = merge_list->next;
2659                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2660                free(merge_list);
2661                merge_list = next;
2662        }
2663        strbuf_addf(&new_data,
2664                "author %s\n"
2665                "committer %s\n"
2666                "\n",
2667                author ? author : committer, committer);
2668        strbuf_addbuf(&new_data, &msg);
2669        free(author);
2670        free(committer);
2671
2672        if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2673                b->pack_id = pack_id;
2674        b->last_commit = object_count_by_type[OBJ_COMMIT];
2675}
2676
2677static void parse_new_tag(void)
2678{
2679        static struct strbuf msg = STRBUF_INIT;
2680        char *sp;
2681        const char *from;
2682        char *tagger;
2683        struct branch *s;
2684        struct tag *t;
2685        uintmax_t from_mark = 0;
2686        unsigned char sha1[20];
2687        enum object_type type;
2688
2689        /* Obtain the new tag name from the rest of our command */
2690        sp = strchr(command_buf.buf, ' ') + 1;
2691        t = pool_alloc(sizeof(struct tag));
2692        t->next_tag = NULL;
2693        t->name = pool_strdup(sp);
2694        if (last_tag)
2695                last_tag->next_tag = t;
2696        else
2697                first_tag = t;
2698        last_tag = t;
2699        read_next_command();
2700
2701        /* from ... */
2702        if (prefixcmp(command_buf.buf, "from "))
2703                die("Expected from command, got %s", command_buf.buf);
2704        from = strchr(command_buf.buf, ' ') + 1;
2705        s = lookup_branch(from);
2706        if (s) {
2707                hashcpy(sha1, s->sha1);
2708                type = OBJ_COMMIT;
2709        } else if (*from == ':') {
2710                struct object_entry *oe;
2711                from_mark = strtoumax(from + 1, NULL, 10);
2712                oe = find_mark(from_mark);
2713                type = oe->type;
2714                hashcpy(sha1, oe->idx.sha1);
2715        } else if (!get_sha1(from, sha1)) {
2716                unsigned long size;
2717                char *buf;
2718
2719                buf = read_sha1_file(sha1, &type, &size);
2720                if (!buf || size < 46)
2721                        die("Not a valid commit: %s", from);
2722                free(buf);
2723        } else
2724                die("Invalid ref name or SHA1 expression: %s", from);
2725        read_next_command();
2726
2727        /* tagger ... */
2728        if (!prefixcmp(command_buf.buf, "tagger ")) {
2729                tagger = parse_ident(command_buf.buf + 7);
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                    sha1_to_hex(sha1), typename(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->sha1, 0))
2753                t->pack_id = MAX_PACK_ID;
2754        else
2755                t->pack_id = pack_id;
2756}
2757
2758static void parse_reset_branch(void)
2759{
2760        struct branch *b;
2761        char *sp;
2762
2763        /* Obtain the branch name from the rest of our command */
2764        sp = strchr(command_buf.buf, ' ') + 1;
2765        b = lookup_branch(sp);
2766        if (b) {
2767                hashclr(b->sha1);
2768                hashclr(b->branch_tree.versions[0].sha1);
2769                hashclr(b->branch_tree.versions[1].sha1);
2770                if (b->branch_tree.tree) {
2771                        release_tree_content_recursive(b->branch_tree.tree);
2772                        b->branch_tree.tree = NULL;
2773                }
2774        }
2775        else
2776                b = new_branch(sp);
2777        read_next_command();
2778        parse_from(b);
2779        if (command_buf.len > 0)
2780                unread_command_buf = 1;
2781}
2782
2783static void cat_blob_write(const char *buf, unsigned long size)
2784{
2785        if (write_in_full(cat_blob_fd, buf, size) != size)
2786                die_errno("Write to frontend failed");
2787}
2788
2789static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2790{
2791        struct strbuf line = STRBUF_INIT;
2792        unsigned long size;
2793        enum object_type type = 0;
2794        char *buf;
2795
2796        if (!oe || oe->pack_id == MAX_PACK_ID) {
2797                buf = read_sha1_file(sha1, &type, &size);
2798        } else {
2799                type = oe->type;
2800                buf = gfi_unpack_entry(oe, &size);
2801        }
2802
2803        /*
2804         * Output based on batch_one_object() from cat-file.c.
2805         */
2806        if (type <= 0) {
2807                strbuf_reset(&line);
2808                strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2809                cat_blob_write(line.buf, line.len);
2810                strbuf_release(&line);
2811                free(buf);
2812                return;
2813        }
2814        if (!buf)
2815                die("Can't read object %s", sha1_to_hex(sha1));
2816        if (type != OBJ_BLOB)
2817                die("Object %s is a %s but a blob was expected.",
2818                    sha1_to_hex(sha1), typename(type));
2819        strbuf_reset(&line);
2820        strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2821                                                typename(type), size);
2822        cat_blob_write(line.buf, line.len);
2823        strbuf_release(&line);
2824        cat_blob_write(buf, size);
2825        cat_blob_write("\n", 1);
2826        free(buf);
2827}
2828
2829static void parse_cat_blob(void)
2830{
2831        const char *p;
2832        struct object_entry *oe = oe;
2833        unsigned char sha1[20];
2834
2835        /* cat-blob SP <object> LF */
2836        p = command_buf.buf + strlen("cat-blob ");
2837        if (*p == ':') {
2838                char *x;
2839                oe = find_mark(strtoumax(p + 1, &x, 10));
2840                if (x == p + 1)
2841                        die("Invalid mark: %s", command_buf.buf);
2842                if (!oe)
2843                        die("Unknown mark: %s", command_buf.buf);
2844                if (*x)
2845                        die("Garbage after mark: %s", command_buf.buf);
2846                hashcpy(sha1, oe->idx.sha1);
2847        } else {
2848                if (get_sha1_hex(p, sha1))
2849                        die("Invalid SHA1: %s", command_buf.buf);
2850                if (p[40])
2851                        die("Garbage after SHA1: %s", command_buf.buf);
2852                oe = find_object(sha1);
2853        }
2854
2855        cat_blob(oe, sha1);
2856}
2857
2858static struct object_entry *dereference(struct object_entry *oe,
2859                                        unsigned char sha1[20])
2860{
2861        unsigned long size;
2862        char *buf = NULL;
2863        if (!oe) {
2864                enum object_type type = sha1_object_info(sha1, NULL);
2865                if (type < 0)
2866                        die("object not found: %s", sha1_to_hex(sha1));
2867                /* cache it! */
2868                oe = insert_object(sha1);
2869                oe->type = type;
2870                oe->pack_id = MAX_PACK_ID;
2871                oe->idx.offset = 1;
2872        }
2873        switch (oe->type) {
2874        case OBJ_TREE:  /* easy case. */
2875                return oe;
2876        case OBJ_COMMIT:
2877        case OBJ_TAG:
2878                break;
2879        default:
2880                die("Not a treeish: %s", command_buf.buf);
2881        }
2882
2883        if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
2884                buf = gfi_unpack_entry(oe, &size);
2885        } else {
2886                enum object_type unused;
2887                buf = read_sha1_file(sha1, &unused, &size);
2888        }
2889        if (!buf)
2890                die("Can't load object %s", sha1_to_hex(sha1));
2891
2892        /* Peel one layer. */
2893        switch (oe->type) {
2894        case OBJ_TAG:
2895                if (size < 40 + strlen("object ") ||
2896                    get_sha1_hex(buf + strlen("object "), sha1))
2897                        die("Invalid SHA1 in tag: %s", command_buf.buf);
2898                break;
2899        case OBJ_COMMIT:
2900                if (size < 40 + strlen("tree ") ||
2901                    get_sha1_hex(buf + strlen("tree "), sha1))
2902                        die("Invalid SHA1 in commit: %s", command_buf.buf);
2903        }
2904
2905        free(buf);
2906        return find_object(sha1);
2907}
2908
2909static struct object_entry *parse_treeish_dataref(const char **p)
2910{
2911        unsigned char sha1[20];
2912        struct object_entry *e;
2913
2914        if (**p == ':') {       /* <mark> */
2915                char *endptr;
2916                e = find_mark(strtoumax(*p + 1, &endptr, 10));
2917                if (endptr == *p + 1)
2918                        die("Invalid mark: %s", command_buf.buf);
2919                if (!e)
2920                        die("Unknown mark: %s", command_buf.buf);
2921                *p = endptr;
2922                hashcpy(sha1, e->idx.sha1);
2923        } else {        /* <sha1> */
2924                if (get_sha1_hex(*p, sha1))
2925                        die("Invalid SHA1: %s", command_buf.buf);
2926                e = find_object(sha1);
2927                *p += 40;
2928        }
2929
2930        while (!e || e->type != OBJ_TREE)
2931                e = dereference(e, sha1);
2932        return e;
2933}
2934
2935static void print_ls(int mode, const unsigned char *sha1, const char *path)
2936{
2937        static struct strbuf line = STRBUF_INIT;
2938
2939        /* See show_tree(). */
2940        const char *type =
2941                S_ISGITLINK(mode) ? commit_type :
2942                S_ISDIR(mode) ? tree_type :
2943                blob_type;
2944
2945        if (!mode) {
2946                /* missing SP path LF */
2947                strbuf_reset(&line);
2948                strbuf_addstr(&line, "missing ");
2949                quote_c_style(path, &line, NULL, 0);
2950                strbuf_addch(&line, '\n');
2951        } else {
2952                /* mode SP type SP object_name TAB path LF */
2953                strbuf_reset(&line);
2954                strbuf_addf(&line, "%06o %s %s\t",
2955                                mode & ~NO_DELTA, type, sha1_to_hex(sha1));
2956                quote_c_style(path, &line, NULL, 0);
2957                strbuf_addch(&line, '\n');
2958        }
2959        cat_blob_write(line.buf, line.len);
2960}
2961
2962static void parse_ls(struct branch *b)
2963{
2964        const char *p;
2965        struct tree_entry *root = NULL;
2966        struct tree_entry leaf = {NULL};
2967
2968        /* ls SP (<treeish> SP)? <path> */
2969        p = command_buf.buf + strlen("ls ");
2970        if (*p == '"') {
2971                if (!b)
2972                        die("Not in a commit: %s", command_buf.buf);
2973                root = &b->branch_tree;
2974        } else {
2975                struct object_entry *e = parse_treeish_dataref(&p);
2976                root = new_tree_entry();
2977                hashcpy(root->versions[1].sha1, e->idx.sha1);
2978                load_tree(root);
2979                if (*p++ != ' ')
2980                        die("Missing space after tree-ish: %s", command_buf.buf);
2981        }
2982        if (*p == '"') {
2983                static struct strbuf uq = STRBUF_INIT;
2984                const char *endp;
2985                strbuf_reset(&uq);
2986                if (unquote_c_style(&uq, p, &endp))
2987                        die("Invalid path: %s", command_buf.buf);
2988                if (*endp)
2989                        die("Garbage after path in: %s", command_buf.buf);
2990                p = uq.buf;
2991        }
2992        tree_content_get(root, p, &leaf);
2993        /*
2994         * A directory in preparation would have a sha1 of zero
2995         * until it is saved.  Save, for simplicity.
2996         */
2997        if (S_ISDIR(leaf.versions[1].mode))
2998                store_tree(&leaf);
2999
3000        print_ls(leaf.versions[1].mode, leaf.versions[1].sha1, p);
3001        if (!b || root != &b->branch_tree)
3002                release_tree_entry(root);
3003}
3004
3005static void checkpoint(void)
3006{
3007        checkpoint_requested = 0;
3008        if (object_count) {
3009                cycle_packfile();
3010                dump_branches();
3011                dump_tags();
3012                dump_marks();
3013        }
3014}
3015
3016static void parse_checkpoint(void)
3017{
3018        checkpoint_requested = 1;
3019        skip_optional_lf();
3020}
3021
3022static void parse_progress(void)
3023{
3024        fwrite(command_buf.buf, 1, command_buf.len, stdout);
3025        fputc('\n', stdout);
3026        fflush(stdout);
3027        skip_optional_lf();
3028}
3029
3030static char* make_fast_import_path(const char *path)
3031{
3032        struct strbuf abs_path = STRBUF_INIT;
3033
3034        if (!relative_marks_paths || is_absolute_path(path))
3035                return xstrdup(path);
3036        strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
3037        return strbuf_detach(&abs_path, NULL);
3038}
3039
3040static void option_import_marks(const char *marks,
3041                                        int from_stream, int ignore_missing)
3042{
3043        if (import_marks_file) {
3044                if (from_stream)
3045                        die("Only one import-marks command allowed per stream");
3046
3047                /* read previous mark file */
3048                if(!import_marks_file_from_stream)
3049                        read_marks();
3050        }
3051
3052        import_marks_file = make_fast_import_path(marks);
3053        safe_create_leading_directories_const(import_marks_file);
3054        import_marks_file_from_stream = from_stream;
3055        import_marks_file_ignore_missing = ignore_missing;
3056}
3057
3058static void option_date_format(const char *fmt)
3059{
3060        if (!strcmp(fmt, "raw"))
3061                whenspec = WHENSPEC_RAW;
3062        else if (!strcmp(fmt, "rfc2822"))
3063                whenspec = WHENSPEC_RFC2822;
3064        else if (!strcmp(fmt, "now"))
3065                whenspec = WHENSPEC_NOW;
3066        else
3067                die("unknown --date-format argument %s", fmt);
3068}
3069
3070static unsigned long ulong_arg(const char *option, const char *arg)
3071{
3072        char *endptr;
3073        unsigned long rv = strtoul(arg, &endptr, 0);
3074        if (strchr(arg, '-') || endptr == arg || *endptr)
3075                die("%s: argument must be a non-negative integer", option);
3076        return rv;
3077}
3078
3079static void option_depth(const char *depth)
3080{
3081        max_depth = ulong_arg("--depth", depth);
3082        if (max_depth > MAX_DEPTH)
3083                die("--depth cannot exceed %u", MAX_DEPTH);
3084}
3085
3086static void option_active_branches(const char *branches)
3087{
3088        max_active_branches = ulong_arg("--active-branches", branches);
3089}
3090
3091static void option_export_marks(const char *marks)
3092{
3093        export_marks_file = make_fast_import_path(marks);
3094        safe_create_leading_directories_const(export_marks_file);
3095}
3096
3097static void option_cat_blob_fd(const char *fd)
3098{
3099        unsigned long n = ulong_arg("--cat-blob-fd", fd);
3100        if (n > (unsigned long) INT_MAX)
3101                die("--cat-blob-fd cannot exceed %d", INT_MAX);
3102        cat_blob_fd = (int) n;
3103}
3104
3105static void option_export_pack_edges(const char *edges)
3106{
3107        if (pack_edges)
3108                fclose(pack_edges);
3109        pack_edges = fopen(edges, "a");
3110        if (!pack_edges)
3111                die_errno("Cannot open '%s'", edges);
3112}
3113
3114static int parse_one_option(const char *option)
3115{
3116        if (!prefixcmp(option, "max-pack-size=")) {
3117                unsigned long v;
3118                if (!git_parse_ulong(option + 14, &v))
3119                        return 0;
3120                if (v < 8192) {
3121                        warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3122                        v *= 1024 * 1024;
3123                } else if (v < 1024 * 1024) {
3124                        warning("minimum max-pack-size is 1 MiB");
3125                        v = 1024 * 1024;
3126                }
3127                max_packsize = v;
3128        } else if (!prefixcmp(option, "big-file-threshold=")) {
3129                unsigned long v;
3130                if (!git_parse_ulong(option + 19, &v))
3131                        return 0;
3132                big_file_threshold = v;
3133        } else if (!prefixcmp(option, "depth=")) {
3134                option_depth(option + 6);
3135        } else if (!prefixcmp(option, "active-branches=")) {
3136                option_active_branches(option + 16);
3137        } else if (!prefixcmp(option, "export-pack-edges=")) {
3138                option_export_pack_edges(option + 18);
3139        } else if (!prefixcmp(option, "quiet")) {
3140                show_stats = 0;
3141        } else if (!prefixcmp(option, "stats")) {
3142                show_stats = 1;
3143        } else {
3144                return 0;
3145        }
3146
3147        return 1;
3148}
3149
3150static int parse_one_feature(const char *feature, int from_stream)
3151{
3152        if (!prefixcmp(feature, "date-format=")) {
3153                option_date_format(feature + 12);
3154        } else if (!prefixcmp(feature, "import-marks=")) {
3155                option_import_marks(feature + 13, from_stream, 0);
3156        } else if (!prefixcmp(feature, "import-marks-if-exists=")) {
3157                option_import_marks(feature + strlen("import-marks-if-exists="),
3158                                        from_stream, 1);
3159        } else if (!prefixcmp(feature, "export-marks=")) {
3160                option_export_marks(feature + 13);
3161        } else if (!strcmp(feature, "cat-blob")) {
3162                ; /* Don't die - this feature is supported */
3163        } else if (!strcmp(feature, "relative-marks")) {
3164                relative_marks_paths = 1;
3165        } else if (!strcmp(feature, "no-relative-marks")) {
3166                relative_marks_paths = 0;
3167        } else if (!strcmp(feature, "force")) {
3168                force_update = 1;
3169        } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3170                ; /* do nothing; we have the feature */
3171        } else {
3172                return 0;
3173        }
3174
3175        return 1;
3176}
3177
3178static void parse_feature(void)
3179{
3180        char *feature = command_buf.buf + 8;
3181
3182        if (seen_data_command)
3183                die("Got feature command '%s' after data command", feature);
3184
3185        if (parse_one_feature(feature, 1))
3186                return;
3187
3188        die("This version of fast-import does not support feature %s.", feature);
3189}
3190
3191static void parse_option(void)
3192{
3193        char *option = command_buf.buf + 11;
3194
3195        if (seen_data_command)
3196                die("Got option command '%s' after data command", option);
3197
3198        if (parse_one_option(option))
3199                return;
3200
3201        die("This version of fast-import does not support option: %s", option);
3202}
3203
3204static int git_pack_config(const char *k, const char *v, void *cb)
3205{
3206        if (!strcmp(k, "pack.depth")) {
3207                max_depth = git_config_int(k, v);
3208                if (max_depth > MAX_DEPTH)
3209                        max_depth = MAX_DEPTH;
3210                return 0;
3211        }
3212        if (!strcmp(k, "pack.compression")) {
3213                int level = git_config_int(k, v);
3214                if (level == -1)
3215                        level = Z_DEFAULT_COMPRESSION;
3216                else if (level < 0 || level > Z_BEST_COMPRESSION)
3217                        die("bad pack compression level %d", level);
3218                pack_compression_level = level;
3219                pack_compression_seen = 1;
3220                return 0;
3221        }
3222        if (!strcmp(k, "pack.indexversion")) {
3223                pack_idx_default_version = git_config_int(k, v);
3224                if (pack_idx_default_version > 2)
3225                        die("bad pack.indexversion=%"PRIu32,
3226                            pack_idx_default_version);
3227                return 0;
3228        }
3229        if (!strcmp(k, "pack.packsizelimit")) {
3230                max_packsize = git_config_ulong(k, v);
3231                return 0;
3232        }
3233        return git_default_config(k, v, cb);
3234}
3235
3236static const char fast_import_usage[] =
3237"git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3238
3239static void parse_argv(void)
3240{
3241        unsigned int i;
3242
3243        for (i = 1; i < global_argc; i++) {
3244                const char *a = global_argv[i];
3245
3246                if (*a != '-' || !strcmp(a, "--"))
3247                        break;
3248
3249                if (parse_one_option(a + 2))
3250                        continue;
3251
3252                if (parse_one_feature(a + 2, 0))
3253                        continue;
3254
3255                if (!prefixcmp(a + 2, "cat-blob-fd=")) {
3256                        option_cat_blob_fd(a + 2 + strlen("cat-blob-fd="));
3257                        continue;
3258                }
3259
3260                die("unknown option %s", a);
3261        }
3262        if (i != global_argc)
3263                usage(fast_import_usage);
3264
3265        seen_data_command = 1;
3266        if (import_marks_file)
3267                read_marks();
3268}
3269
3270int main(int argc, const char **argv)
3271{
3272        unsigned int i;
3273
3274        git_extract_argv0_path(argv[0]);
3275
3276        if (argc == 2 && !strcmp(argv[1], "-h"))
3277                usage(fast_import_usage);
3278
3279        setup_git_directory();
3280        git_config(git_pack_config, NULL);
3281        if (!pack_compression_seen && core_compression_seen)
3282                pack_compression_level = core_compression_level;
3283
3284        alloc_objects(object_entry_alloc);
3285        strbuf_init(&command_buf, 0);
3286        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3287        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3288        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3289        marks = pool_calloc(1, sizeof(struct mark_set));
3290
3291        global_argc = argc;
3292        global_argv = argv;
3293
3294        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3295        for (i = 0; i < (cmd_save - 1); i++)
3296                rc_free[i].next = &rc_free[i + 1];
3297        rc_free[cmd_save - 1].next = NULL;
3298
3299        prepare_packed_git();
3300        start_packfile();
3301        set_die_routine(die_nicely);
3302        set_checkpoint_signal();
3303        while (read_next_command() != EOF) {
3304                if (!strcmp("blob", command_buf.buf))
3305                        parse_new_blob();
3306                else if (!prefixcmp(command_buf.buf, "ls "))
3307                        parse_ls(NULL);
3308                else if (!prefixcmp(command_buf.buf, "commit "))
3309                        parse_new_commit();
3310                else if (!prefixcmp(command_buf.buf, "tag "))
3311                        parse_new_tag();
3312                else if (!prefixcmp(command_buf.buf, "reset "))
3313                        parse_reset_branch();
3314                else if (!strcmp("checkpoint", command_buf.buf))
3315                        parse_checkpoint();
3316                else if (!prefixcmp(command_buf.buf, "progress "))
3317                        parse_progress();
3318                else if (!prefixcmp(command_buf.buf, "feature "))
3319                        parse_feature();
3320                else if (!prefixcmp(command_buf.buf, "option git "))
3321                        parse_option();
3322                else if (!prefixcmp(command_buf.buf, "option "))
3323                        /* ignore non-git options*/;
3324                else
3325                        die("Unsupported command: %s", command_buf.buf);
3326
3327                if (checkpoint_requested)
3328                        checkpoint();
3329        }
3330
3331        /* argv hasn't been parsed yet, do so */
3332        if (!seen_data_command)
3333                parse_argv();
3334
3335        end_packfile();
3336
3337        dump_branches();
3338        dump_tags();
3339        unkeep_all_packs();
3340        dump_marks();
3341
3342        if (pack_edges)
3343                fclose(pack_edges);
3344
3345        if (show_stats) {
3346                uintmax_t total_count = 0, duplicate_count = 0;
3347                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3348                        total_count += object_count_by_type[i];
3349                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3350                        duplicate_count += duplicate_count_by_type[i];
3351
3352                fprintf(stderr, "%s statistics:\n", argv[0]);
3353                fprintf(stderr, "---------------------------------------------------------------------\n");
3354                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3355                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3356                fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
3357                fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
3358                fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
3359                fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
3360                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3361                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3362                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3363                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3364                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3365                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3366                fprintf(stderr, "---------------------------------------------------------------------\n");
3367                pack_report();
3368                fprintf(stderr, "---------------------------------------------------------------------\n");
3369                fprintf(stderr, "\n");
3370        }
3371
3372        return failure ? 1 : 0;
3373}