streaming.con commit Merge branch 'jk/check-corrupt-objects-carefully' (a6fed65)
   1/*
   2 * Copyright (c) 2011, Google Inc.
   3 */
   4#include "cache.h"
   5#include "streaming.h"
   6
   7enum input_source {
   8        stream_error = -1,
   9        incore = 0,
  10        loose = 1,
  11        pack_non_delta = 2
  12};
  13
  14typedef int (*open_istream_fn)(struct git_istream *,
  15                               struct object_info *,
  16                               const unsigned char *,
  17                               enum object_type *);
  18typedef int (*close_istream_fn)(struct git_istream *);
  19typedef ssize_t (*read_istream_fn)(struct git_istream *, char *, size_t);
  20
  21struct stream_vtbl {
  22        close_istream_fn close;
  23        read_istream_fn read;
  24};
  25
  26#define open_method_decl(name) \
  27        int open_istream_ ##name \
  28        (struct git_istream *st, struct object_info *oi, \
  29         const unsigned char *sha1, \
  30         enum object_type *type)
  31
  32#define close_method_decl(name) \
  33        int close_istream_ ##name \
  34        (struct git_istream *st)
  35
  36#define read_method_decl(name) \
  37        ssize_t read_istream_ ##name \
  38        (struct git_istream *st, char *buf, size_t sz)
  39
  40/* forward declaration */
  41static open_method_decl(incore);
  42static open_method_decl(loose);
  43static open_method_decl(pack_non_delta);
  44static struct git_istream *attach_stream_filter(struct git_istream *st,
  45                                                struct stream_filter *filter);
  46
  47
  48static open_istream_fn open_istream_tbl[] = {
  49        open_istream_incore,
  50        open_istream_loose,
  51        open_istream_pack_non_delta,
  52};
  53
  54#define FILTER_BUFFER (1024*16)
  55
  56struct filtered_istream {
  57        struct git_istream *upstream;
  58        struct stream_filter *filter;
  59        char ibuf[FILTER_BUFFER];
  60        char obuf[FILTER_BUFFER];
  61        int i_end, i_ptr;
  62        int o_end, o_ptr;
  63        int input_finished;
  64};
  65
  66struct git_istream {
  67        const struct stream_vtbl *vtbl;
  68        unsigned long size; /* inflated size of full object */
  69        git_zstream z;
  70        enum { z_unused, z_used, z_done, z_error } z_state;
  71
  72        union {
  73                struct {
  74                        char *buf; /* from read_object() */
  75                        unsigned long read_ptr;
  76                } incore;
  77
  78                struct {
  79                        void *mapped;
  80                        unsigned long mapsize;
  81                        char hdr[32];
  82                        int hdr_avail;
  83                        int hdr_used;
  84                } loose;
  85
  86                struct {
  87                        struct packed_git *pack;
  88                        off_t pos;
  89                } in_pack;
  90
  91                struct filtered_istream filtered;
  92        } u;
  93};
  94
  95int close_istream(struct git_istream *st)
  96{
  97        int r = st->vtbl->close(st);
  98        free(st);
  99        return r;
 100}
 101
 102ssize_t read_istream(struct git_istream *st, void *buf, size_t sz)
 103{
 104        return st->vtbl->read(st, buf, sz);
 105}
 106
 107static enum input_source istream_source(const unsigned char *sha1,
 108                                        enum object_type *type,
 109                                        struct object_info *oi)
 110{
 111        unsigned long size;
 112        int status;
 113
 114        oi->sizep = &size;
 115        status = sha1_object_info_extended(sha1, oi);
 116        if (status < 0)
 117                return stream_error;
 118        *type = status;
 119
 120        switch (oi->whence) {
 121        case OI_LOOSE:
 122                return loose;
 123        case OI_PACKED:
 124                if (!oi->u.packed.is_delta && big_file_threshold < size)
 125                        return pack_non_delta;
 126                /* fallthru */
 127        default:
 128                return incore;
 129        }
 130}
 131
 132struct git_istream *open_istream(const unsigned char *sha1,
 133                                 enum object_type *type,
 134                                 unsigned long *size,
 135                                 struct stream_filter *filter)
 136{
 137        struct git_istream *st;
 138        struct object_info oi;
 139        const unsigned char *real = lookup_replace_object(sha1);
 140        enum input_source src = istream_source(real, type, &oi);
 141
 142        if (src < 0)
 143                return NULL;
 144
 145        st = xmalloc(sizeof(*st));
 146        if (open_istream_tbl[src](st, &oi, real, type)) {
 147                if (open_istream_incore(st, &oi, real, type)) {
 148                        free(st);
 149                        return NULL;
 150                }
 151        }
 152        if (st && filter) {
 153                /* Add "&& !is_null_stream_filter(filter)" for performance */
 154                struct git_istream *nst = attach_stream_filter(st, filter);
 155                if (!nst)
 156                        close_istream(st);
 157                st = nst;
 158        }
 159
 160        *size = st->size;
 161        return st;
 162}
 163
 164
 165/*****************************************************************
 166 *
 167 * Common helpers
 168 *
 169 *****************************************************************/
 170
 171static void close_deflated_stream(struct git_istream *st)
 172{
 173        if (st->z_state == z_used)
 174                git_inflate_end(&st->z);
 175}
 176
 177
 178/*****************************************************************
 179 *
 180 * Filtered stream
 181 *
 182 *****************************************************************/
 183
 184static close_method_decl(filtered)
 185{
 186        free_stream_filter(st->u.filtered.filter);
 187        return close_istream(st->u.filtered.upstream);
 188}
 189
 190static read_method_decl(filtered)
 191{
 192        struct filtered_istream *fs = &(st->u.filtered);
 193        size_t filled = 0;
 194
 195        while (sz) {
 196                /* do we already have filtered output? */
 197                if (fs->o_ptr < fs->o_end) {
 198                        size_t to_move = fs->o_end - fs->o_ptr;
 199                        if (sz < to_move)
 200                                to_move = sz;
 201                        memcpy(buf + filled, fs->obuf + fs->o_ptr, to_move);
 202                        fs->o_ptr += to_move;
 203                        sz -= to_move;
 204                        filled += to_move;
 205                        continue;
 206                }
 207                fs->o_end = fs->o_ptr = 0;
 208
 209                /* do we have anything to feed the filter with? */
 210                if (fs->i_ptr < fs->i_end) {
 211                        size_t to_feed = fs->i_end - fs->i_ptr;
 212                        size_t to_receive = FILTER_BUFFER;
 213                        if (stream_filter(fs->filter,
 214                                          fs->ibuf + fs->i_ptr, &to_feed,
 215                                          fs->obuf, &to_receive))
 216                                return -1;
 217                        fs->i_ptr = fs->i_end - to_feed;
 218                        fs->o_end = FILTER_BUFFER - to_receive;
 219                        continue;
 220                }
 221
 222                /* tell the filter to drain upon no more input */
 223                if (fs->input_finished) {
 224                        size_t to_receive = FILTER_BUFFER;
 225                        if (stream_filter(fs->filter,
 226                                          NULL, NULL,
 227                                          fs->obuf, &to_receive))
 228                                return -1;
 229                        fs->o_end = FILTER_BUFFER - to_receive;
 230                        if (!fs->o_end)
 231                                break;
 232                        continue;
 233                }
 234                fs->i_end = fs->i_ptr = 0;
 235
 236                /* refill the input from the upstream */
 237                if (!fs->input_finished) {
 238                        fs->i_end = read_istream(fs->upstream, fs->ibuf, FILTER_BUFFER);
 239                        if (fs->i_end < 0)
 240                                return -1;
 241                        if (fs->i_end)
 242                                continue;
 243                }
 244                fs->input_finished = 1;
 245        }
 246        return filled;
 247}
 248
 249static struct stream_vtbl filtered_vtbl = {
 250        close_istream_filtered,
 251        read_istream_filtered,
 252};
 253
 254static struct git_istream *attach_stream_filter(struct git_istream *st,
 255                                                struct stream_filter *filter)
 256{
 257        struct git_istream *ifs = xmalloc(sizeof(*ifs));
 258        struct filtered_istream *fs = &(ifs->u.filtered);
 259
 260        ifs->vtbl = &filtered_vtbl;
 261        fs->upstream = st;
 262        fs->filter = filter;
 263        fs->i_end = fs->i_ptr = 0;
 264        fs->o_end = fs->o_ptr = 0;
 265        fs->input_finished = 0;
 266        ifs->size = -1; /* unknown */
 267        return ifs;
 268}
 269
 270/*****************************************************************
 271 *
 272 * Loose object stream
 273 *
 274 *****************************************************************/
 275
 276static read_method_decl(loose)
 277{
 278        size_t total_read = 0;
 279
 280        switch (st->z_state) {
 281        case z_done:
 282                return 0;
 283        case z_error:
 284                return -1;
 285        default:
 286                break;
 287        }
 288
 289        if (st->u.loose.hdr_used < st->u.loose.hdr_avail) {
 290                size_t to_copy = st->u.loose.hdr_avail - st->u.loose.hdr_used;
 291                if (sz < to_copy)
 292                        to_copy = sz;
 293                memcpy(buf, st->u.loose.hdr + st->u.loose.hdr_used, to_copy);
 294                st->u.loose.hdr_used += to_copy;
 295                total_read += to_copy;
 296        }
 297
 298        while (total_read < sz) {
 299                int status;
 300
 301                st->z.next_out = (unsigned char *)buf + total_read;
 302                st->z.avail_out = sz - total_read;
 303                status = git_inflate(&st->z, Z_FINISH);
 304
 305                total_read = st->z.next_out - (unsigned char *)buf;
 306
 307                if (status == Z_STREAM_END) {
 308                        git_inflate_end(&st->z);
 309                        st->z_state = z_done;
 310                        break;
 311                }
 312                if (status != Z_OK && (status != Z_BUF_ERROR || total_read < sz)) {
 313                        git_inflate_end(&st->z);
 314                        st->z_state = z_error;
 315                        return -1;
 316                }
 317        }
 318        return total_read;
 319}
 320
 321static close_method_decl(loose)
 322{
 323        close_deflated_stream(st);
 324        munmap(st->u.loose.mapped, st->u.loose.mapsize);
 325        return 0;
 326}
 327
 328static struct stream_vtbl loose_vtbl = {
 329        close_istream_loose,
 330        read_istream_loose,
 331};
 332
 333static open_method_decl(loose)
 334{
 335        st->u.loose.mapped = map_sha1_file(sha1, &st->u.loose.mapsize);
 336        if (!st->u.loose.mapped)
 337                return -1;
 338        if (unpack_sha1_header(&st->z,
 339                               st->u.loose.mapped,
 340                               st->u.loose.mapsize,
 341                               st->u.loose.hdr,
 342                               sizeof(st->u.loose.hdr)) < 0) {
 343                git_inflate_end(&st->z);
 344                munmap(st->u.loose.mapped, st->u.loose.mapsize);
 345                return -1;
 346        }
 347
 348        parse_sha1_header(st->u.loose.hdr, &st->size);
 349        st->u.loose.hdr_used = strlen(st->u.loose.hdr) + 1;
 350        st->u.loose.hdr_avail = st->z.total_out;
 351        st->z_state = z_used;
 352
 353        st->vtbl = &loose_vtbl;
 354        return 0;
 355}
 356
 357
 358/*****************************************************************
 359 *
 360 * Non-delta packed object stream
 361 *
 362 *****************************************************************/
 363
 364static read_method_decl(pack_non_delta)
 365{
 366        size_t total_read = 0;
 367
 368        switch (st->z_state) {
 369        case z_unused:
 370                memset(&st->z, 0, sizeof(st->z));
 371                git_inflate_init(&st->z);
 372                st->z_state = z_used;
 373                break;
 374        case z_done:
 375                return 0;
 376        case z_error:
 377                return -1;
 378        case z_used:
 379                break;
 380        }
 381
 382        while (total_read < sz) {
 383                int status;
 384                struct pack_window *window = NULL;
 385                unsigned char *mapped;
 386
 387                mapped = use_pack(st->u.in_pack.pack, &window,
 388                                  st->u.in_pack.pos, &st->z.avail_in);
 389
 390                st->z.next_out = (unsigned char *)buf + total_read;
 391                st->z.avail_out = sz - total_read;
 392                st->z.next_in = mapped;
 393                status = git_inflate(&st->z, Z_FINISH);
 394
 395                st->u.in_pack.pos += st->z.next_in - mapped;
 396                total_read = st->z.next_out - (unsigned char *)buf;
 397                unuse_pack(&window);
 398
 399                if (status == Z_STREAM_END) {
 400                        git_inflate_end(&st->z);
 401                        st->z_state = z_done;
 402                        break;
 403                }
 404                if (status != Z_OK && status != Z_BUF_ERROR) {
 405                        git_inflate_end(&st->z);
 406                        st->z_state = z_error;
 407                        return -1;
 408                }
 409        }
 410        return total_read;
 411}
 412
 413static close_method_decl(pack_non_delta)
 414{
 415        close_deflated_stream(st);
 416        return 0;
 417}
 418
 419static struct stream_vtbl pack_non_delta_vtbl = {
 420        close_istream_pack_non_delta,
 421        read_istream_pack_non_delta,
 422};
 423
 424static open_method_decl(pack_non_delta)
 425{
 426        struct pack_window *window;
 427        enum object_type in_pack_type;
 428
 429        st->u.in_pack.pack = oi->u.packed.pack;
 430        st->u.in_pack.pos = oi->u.packed.offset;
 431        window = NULL;
 432
 433        in_pack_type = unpack_object_header(st->u.in_pack.pack,
 434                                            &window,
 435                                            &st->u.in_pack.pos,
 436                                            &st->size);
 437        unuse_pack(&window);
 438        switch (in_pack_type) {
 439        default:
 440                return -1; /* we do not do deltas for now */
 441        case OBJ_COMMIT:
 442        case OBJ_TREE:
 443        case OBJ_BLOB:
 444        case OBJ_TAG:
 445                break;
 446        }
 447        st->z_state = z_unused;
 448        st->vtbl = &pack_non_delta_vtbl;
 449        return 0;
 450}
 451
 452
 453/*****************************************************************
 454 *
 455 * In-core stream
 456 *
 457 *****************************************************************/
 458
 459static close_method_decl(incore)
 460{
 461        free(st->u.incore.buf);
 462        return 0;
 463}
 464
 465static read_method_decl(incore)
 466{
 467        size_t read_size = sz;
 468        size_t remainder = st->size - st->u.incore.read_ptr;
 469
 470        if (remainder <= read_size)
 471                read_size = remainder;
 472        if (read_size) {
 473                memcpy(buf, st->u.incore.buf + st->u.incore.read_ptr, read_size);
 474                st->u.incore.read_ptr += read_size;
 475        }
 476        return read_size;
 477}
 478
 479static struct stream_vtbl incore_vtbl = {
 480        close_istream_incore,
 481        read_istream_incore,
 482};
 483
 484static open_method_decl(incore)
 485{
 486        st->u.incore.buf = read_sha1_file_extended(sha1, type, &st->size, 0);
 487        st->u.incore.read_ptr = 0;
 488        st->vtbl = &incore_vtbl;
 489
 490        return st->u.incore.buf ? 0 : -1;
 491}
 492
 493
 494/****************************************************************
 495 * Users of streaming interface
 496 ****************************************************************/
 497
 498int stream_blob_to_fd(int fd, unsigned const char *sha1, struct stream_filter *filter,
 499                      int can_seek)
 500{
 501        struct git_istream *st;
 502        enum object_type type;
 503        unsigned long sz;
 504        ssize_t kept = 0;
 505        int result = -1;
 506
 507        st = open_istream(sha1, &type, &sz, filter);
 508        if (!st)
 509                return result;
 510        if (type != OBJ_BLOB)
 511                goto close_and_exit;
 512        for (;;) {
 513                char buf[1024 * 16];
 514                ssize_t wrote, holeto;
 515                ssize_t readlen = read_istream(st, buf, sizeof(buf));
 516
 517                if (readlen < 0)
 518                        goto close_and_exit;
 519                if (!readlen)
 520                        break;
 521                if (can_seek && sizeof(buf) == readlen) {
 522                        for (holeto = 0; holeto < readlen; holeto++)
 523                                if (buf[holeto])
 524                                        break;
 525                        if (readlen == holeto) {
 526                                kept += holeto;
 527                                continue;
 528                        }
 529                }
 530
 531                if (kept && lseek(fd, kept, SEEK_CUR) == (off_t) -1)
 532                        goto close_and_exit;
 533                else
 534                        kept = 0;
 535                wrote = write_in_full(fd, buf, readlen);
 536
 537                if (wrote != readlen)
 538                        goto close_and_exit;
 539        }
 540        if (kept && (lseek(fd, kept - 1, SEEK_CUR) == (off_t) -1 ||
 541                     write(fd, "", 1) != 1))
 542                goto close_and_exit;
 543        result = 0;
 544
 545 close_and_exit:
 546        close_istream(st);
 547        return result;
 548}