contrib/libarchive/libarchive/archive_read.c

   1 /*-
   2  * Copyright (c) 2003-2011 Tim Kientzle
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
  15  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  16  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  17  * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
  18  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  19  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  20  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  21  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  22  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  23  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  24  */
  25
  26 /*
  27  * This file contains the "essential" portions of the read API, that
  28  * is, stuff that will probably always be used by any client that
  29  * actually needs to read an archive.  Optional pieces have been, as
  30  * far as possible, separated out into separate files to avoid
  31  * needlessly bloating statically-linked clients.
  32  */
  33
  34 #include "archive_platform.h"
  35 __FBSDID("$FreeBSD$");
  36
  37 #ifdef HAVE_ERRNO_H
  38 #include <errno.h>
  39 #endif
  40 #include <stdio.h>
  41 #ifdef HAVE_STDLIB_H
  42 #include <stdlib.h>
  43 #endif
  44 #ifdef HAVE_STRING_H
  45 #include <string.h>
  46 #endif
  47 #ifdef HAVE_UNISTD_H
  48 #include <unistd.h>
  49 #endif
  50
  51 #include "archive.h"
  52 #include "archive_entry.h"
  53 #include "archive_private.h"
  54 #include "archive_read_private.h"
  55
  56 #define minimum(a, b) (a < b ? a : b)
  57
  58 static int      choose_filters(struct archive_read *);
  59 static int      choose_format(struct archive_read *);
  60 static struct archive_vtable *archive_read_vtable(void);
  61 static int64_t  _archive_filter_bytes(struct archive *, int);
  62 static int      _archive_filter_code(struct archive *, int);
  63 static const char *_archive_filter_name(struct archive *, int);
  64 static int  _archive_filter_count(struct archive *);
  65 static int      _archive_read_close(struct archive *);
  66 static int      _archive_read_data_block(struct archive *,
  67                     const void **, size_t *, int64_t *);
  68 static int      _archive_read_free(struct archive *);
  69 static int      _archive_read_next_header(struct archive *,
  70                     struct archive_entry **);
  71 static int      _archive_read_next_header2(struct archive *,
  72                     struct archive_entry *);
  73 static int64_t  advance_file_pointer(struct archive_read_filter *, int64_t);
  74
  75 static struct archive_vtable *
  76 archive_read_vtable(void)
  77 {
  78         static struct archive_vtable av;
  79         static int inited = 0;
  80
  81         if (!inited) {
  82                 av.archive_filter_bytes = _archive_filter_bytes;
  83                 av.archive_filter_code = _archive_filter_code;
  84                 av.archive_filter_name = _archive_filter_name;
  85                 av.archive_filter_count = _archive_filter_count;
  86                 av.archive_read_data_block = _archive_read_data_block;
  87                 av.archive_read_next_header = _archive_read_next_header;
  88                 av.archive_read_next_header2 = _archive_read_next_header2;
  89                 av.archive_free = _archive_read_free;
  90                 av.archive_close = _archive_read_close;
  91                 inited = 1;
  92         }
  93         return (&av);
  94 }
  95
  96 /*
  97  * Allocate, initialize and return a struct archive object.
  98  */
  99 struct archive *
 100 archive_read_new(void)
 101 {
 102         struct archive_read *a;
 103
 104         a = (struct archive_read *)malloc(sizeof(*a));
 105         if (a == NULL)
 106                 return (NULL);
 107         memset(a, 0, sizeof(*a));
 108         a->archive.magic = ARCHIVE_READ_MAGIC;
 109
 110         a->archive.state = ARCHIVE_STATE_NEW;
 111         a->entry = archive_entry_new2(&a->archive);
 112         a->archive.vtable = archive_read_vtable();
 113
 114         return (&a->archive);
 115 }
 116
 117 /*
 118  * Record the do-not-extract-to file. This belongs in archive_read_extract.c.
 119  */
 120 void
 121 archive_read_extract_set_skip_file(struct archive *_a, int64_t d, int64_t i)
 122 {
 123         struct archive_read *a = (struct archive_read *)_a;
 124
 125         if (ARCHIVE_OK != __archive_check_magic(_a, ARCHIVE_READ_MAGIC,
 126                 ARCHIVE_STATE_ANY, "archive_read_extract_set_skip_file"))
 127                 return;
 128         a->skip_file_set = 1;
 129         a->skip_file_dev = d;
 130         a->skip_file_ino = i;
 131 }
 132
 133 /*
 134  * Open the archive
 135  */
 136 int
 137 archive_read_open(struct archive *a, void *client_data,
 138     archive_open_callback *client_opener, archive_read_callback *client_reader,
 139     archive_close_callback *client_closer)
 140 {
 141         /* Old archive_read_open() is just a thin shell around
 142          * archive_read_open1. */
 143         archive_read_set_open_callback(a, client_opener);
 144         archive_read_set_read_callback(a, client_reader);
 145         archive_read_set_close_callback(a, client_closer);
 146         archive_read_set_callback_data(a, client_data);
 147         return archive_read_open1(a);
 148 }
 149
 150
 151 int
 152 archive_read_open2(struct archive *a, void *client_data,
 153     archive_open_callback *client_opener,
 154     archive_read_callback *client_reader,
 155     archive_skip_callback *client_skipper,
 156     archive_close_callback *client_closer)
 157 {
 158         /* Old archive_read_open2() is just a thin shell around
 159          * archive_read_open1. */
 160         archive_read_set_callback_data(a, client_data);
 161         archive_read_set_open_callback(a, client_opener);
 162         archive_read_set_read_callback(a, client_reader);
 163         archive_read_set_skip_callback(a, client_skipper);
 164         archive_read_set_close_callback(a, client_closer);
 165         return archive_read_open1(a);
 166 }
 167
 168 static ssize_t
 169 client_read_proxy(struct archive_read_filter *self, const void **buff)
 170 {
 171         ssize_t r;
 172         r = (self->archive->client.reader)(&self->archive->archive,
 173             self->data, buff);
 174         return (r);
 175 }
 176
 177 static int64_t
 178 client_skip_proxy(struct archive_read_filter *self, int64_t request)
 179 {
 180         if (request < 0)
 181                 __archive_errx(1, "Negative skip requested.");
 182         if (request == 0)
 183                 return 0;
 184
 185         if (self->archive->client.skipper != NULL) {
 186                 /* Seek requests over 1GiB are broken down into
 187                  * multiple seeks.  This avoids overflows when the
 188                  * requests get passed through 32-bit arguments. */
 189                 int64_t skip_limit = (int64_t)1 << 30;
 190                 int64_t total = 0;
 191                 for (;;) {
 192                         int64_t get, ask = request;
 193                         if (ask > skip_limit)
 194                                 ask = skip_limit;
 195                         get = (self->archive->client.skipper)
 196                                 (&self->archive->archive, self->data, ask);
 197                         if (get == 0)
 198                                 return (total);
 199                         request -= get;
 200                         total += get;
 201                 }
 202         } else if (self->archive->client.seeker != NULL
 203                 && request > 64 * 1024) {
 204                 /* If the client provided a seeker but not a skipper,
 205                  * we can use the seeker to skip forward.
 206                  *
 207                  * Note: This isn't always a good idea.  The client
 208                  * skipper is allowed to skip by less than requested
 209                  * if it needs to maintain block alignment.  The
 210                  * seeker is not allowed to play such games, so using
 211                  * the seeker here may be a performance loss compared
 212                  * to just reading and discarding.  That's why we
 213                  * only do this for skips of over 64k.
 214                  */
 215                 int64_t before = self->position;
 216                 int64_t after = (self->archive->client.seeker)
 217                     (&self->archive->archive, self->data, request, SEEK_CUR);
 218                 if (after != before + request)
 219                         return ARCHIVE_FATAL;
 220                 return after - before;
 221         }
 222         return 0;
 223 }
 224
 225 static int64_t
 226 client_seek_proxy(struct archive_read_filter *self, int64_t offset, int whence)
 227 {
 228         /* DO NOT use the skipper here!  If we transparently handled
 229          * forward seek here by using the skipper, that will break
 230          * other libarchive code that assumes a successful forward
 231          * seek means it can also seek backwards.
 232          */
 233         if (self->archive->client.seeker == NULL)
 234                 return (ARCHIVE_FAILED);
 235         return (self->archive->client.seeker)(&self->archive->archive,
 236             self->data, offset, whence);
 237 }
 238
 239 static int
 240 client_close_proxy(struct archive_read_filter *self)
 241 {
 242         int r = ARCHIVE_OK, r2;
 243         unsigned int i;
 244
 245         if (self->archive->client.closer == NULL)
 246                 return (r);
 247         for (i = 0; i < self->archive->client.nodes; i++)
 248         {
 249                 r2 = (self->archive->client.closer)
 250                         ((struct archive *)self->archive,
 251                                 self->archive->client.dataset[i].data);
 252                 if (r > r2)
 253                         r = r2;
 254         }
 255         return (r);
 256 }
 257
 258 static int
 259 client_open_proxy(struct archive_read_filter *self)
 260 {
 261   int r = ARCHIVE_OK;
 262         if (self->archive->client.opener != NULL)
 263                 r = (self->archive->client.opener)(
 264                     (struct archive *)self->archive, self->data);
 265         return (r);
 266 }
 267
 268 static int
 269 client_switch_proxy(struct archive_read_filter *self, unsigned int iindex)
 270 {
 271   int r1 = ARCHIVE_OK, r2 = ARCHIVE_OK;
 272         void *data2 = NULL;
 273
 274         /* Don't do anything if already in the specified data node */
 275         if (self->archive->client.cursor == iindex)
 276                 return (ARCHIVE_OK);
 277
 278         self->archive->client.cursor = iindex;
 279         data2 = self->archive->client.dataset[self->archive->client.cursor].data;
 280         if (self->archive->client.switcher != NULL)
 281         {
 282                 r1 = r2 = (self->archive->client.switcher)
 283                         ((struct archive *)self->archive, self->data, data2);
 284                 self->data = data2;
 285         }
 286         else
 287         {
 288                 /* Attempt to call close and open instead */
 289                 if (self->archive->client.closer != NULL)
 290                         r1 = (self->archive->client.closer)
 291                                 ((struct archive *)self->archive, self->data);
 292                 self->data = data2;
 293                 if (self->archive->client.opener != NULL)
 294                         r2 = (self->archive->client.opener)
 295                                 ((struct archive *)self->archive, self->data);
 296         }
 297         return (r1 < r2) ? r1 : r2;
 298 }
 299
 300 int
 301 archive_read_set_open_callback(struct archive *_a,
 302     archive_open_callback *client_opener)
 303 {
 304         struct archive_read *a = (struct archive_read *)_a;
 305         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
 306             "archive_read_set_open_callback");
 307         a->client.opener = client_opener;
 308         return ARCHIVE_OK;
 309 }
 310
 311 int
 312 archive_read_set_read_callback(struct archive *_a,
 313     archive_read_callback *client_reader)
 314 {
 315         struct archive_read *a = (struct archive_read *)_a;
 316         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
 317             "archive_read_set_read_callback");
 318         a->client.reader = client_reader;
 319         return ARCHIVE_OK;
 320 }
 321
 322 int
 323 archive_read_set_skip_callback(struct archive *_a,
 324     archive_skip_callback *client_skipper)
 325 {
 326         struct archive_read *a = (struct archive_read *)_a;
 327         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
 328             "archive_read_set_skip_callback");
 329         a->client.skipper = client_skipper;
 330         return ARCHIVE_OK;
 331 }
 332
 333 int
 334 archive_read_set_seek_callback(struct archive *_a,
 335     archive_seek_callback *client_seeker)
 336 {
 337         struct archive_read *a = (struct archive_read *)_a;
 338         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
 339             "archive_read_set_seek_callback");
 340         a->client.seeker = client_seeker;
 341         return ARCHIVE_OK;
 342 }
 343
 344 int
 345 archive_read_set_close_callback(struct archive *_a,
 346     archive_close_callback *client_closer)
 347 {
 348         struct archive_read *a = (struct archive_read *)_a;
 349         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
 350             "archive_read_set_close_callback");
 351         a->client.closer = client_closer;
 352         return ARCHIVE_OK;
 353 }
 354
 355 int
 356 archive_read_set_switch_callback(struct archive *_a,
 357     archive_switch_callback *client_switcher)
 358 {
 359         struct archive_read *a = (struct archive_read *)_a;
 360         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
 361             "archive_read_set_switch_callback");
 362         a->client.switcher = client_switcher;
 363         return ARCHIVE_OK;
 364 }
 365
 366 int
 367 archive_read_set_callback_data(struct archive *_a, void *client_data)
 368 {
 369         return archive_read_set_callback_data2(_a, client_data, 0);
 370 }
 371
 372 int
 373 archive_read_set_callback_data2(struct archive *_a, void *client_data,
 374     unsigned int iindex)
 375 {
 376         struct archive_read *a = (struct archive_read *)_a;
 377         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
 378             "archive_read_set_callback_data2");
 379
 380         if (a->client.nodes == 0)
 381         {
 382                 a->client.dataset = (struct archive_read_data_node *)
 383                     calloc(1, sizeof(*a->client.dataset));
 384                 if (a->client.dataset == NULL)
 385                 {
 386                         archive_set_error(&a->archive, ENOMEM,
 387                                 "No memory.");
 388                         return ARCHIVE_FATAL;
 389                 }
 390                 a->client.nodes = 1;
 391         }
 392
 393         if (iindex > a->client.nodes - 1)
 394         {
 395                 archive_set_error(&a->archive, EINVAL,
 396                         "Invalid index specified.");
 397                 return ARCHIVE_FATAL;
 398         }
 399         a->client.dataset[iindex].data = client_data;
 400         a->client.dataset[iindex].begin_position = -1;
 401         a->client.dataset[iindex].total_size = -1;
 402         return ARCHIVE_OK;
 403 }
 404
 405 int
 406 archive_read_add_callback_data(struct archive *_a, void *client_data,
 407     unsigned int iindex)
 408 {
 409         struct archive_read *a = (struct archive_read *)_a;
 410         void *p;
 411         unsigned int i;
 412
 413         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
 414             "archive_read_add_callback_data");
 415         if (iindex > a->client.nodes) {
 416                 archive_set_error(&a->archive, EINVAL,
 417                         "Invalid index specified.");
 418                 return ARCHIVE_FATAL;
 419         }
 420         p = realloc(a->client.dataset, sizeof(*a->client.dataset)
 421                 * (++(a->client.nodes)));
 422         if (p == NULL) {
 423                 archive_set_error(&a->archive, ENOMEM,
 424                         "No memory.");
 425                 return ARCHIVE_FATAL;
 426         }
 427         a->client.dataset = (struct archive_read_data_node *)p;
 428         for (i = a->client.nodes - 1; i > iindex && i > 0; i--) {
 429                 a->client.dataset[i].data = a->client.dataset[i-1].data;
 430                 a->client.dataset[i].begin_position = -1;
 431                 a->client.dataset[i].total_size = -1;
 432         }
 433         a->client.dataset[iindex].data = client_data;
 434         a->client.dataset[iindex].begin_position = -1;
 435         a->client.dataset[iindex].total_size = -1;
 436         return ARCHIVE_OK;
 437 }
 438
 439 int
 440 archive_read_append_callback_data(struct archive *_a, void *client_data)
 441 {
 442         struct archive_read *a = (struct archive_read *)_a;
 443         return archive_read_add_callback_data(_a, client_data, a->client.nodes);
 444 }
 445
 446 int
 447 archive_read_prepend_callback_data(struct archive *_a, void *client_data)
 448 {
 449         return archive_read_add_callback_data(_a, client_data, 0);
 450 }
 451
 452 int
 453 archive_read_open1(struct archive *_a)
 454 {
 455         struct archive_read *a = (struct archive_read *)_a;
 456         struct archive_read_filter *filter, *tmp;
 457         int slot;
 458         int e = 0;
 459         unsigned int i;
 460
 461         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
 462             "archive_read_open");
 463         archive_clear_error(&a->archive);
 464
 465         if (a->client.reader == NULL) {
 466                 archive_set_error(&a->archive, EINVAL,
 467                     "No reader function provided to archive_read_open");
 468                 a->archive.state = ARCHIVE_STATE_FATAL;
 469                 return (ARCHIVE_FATAL);
 470         }
 471
 472         /* Open data source. */
 473         if (a->client.opener != NULL) {
 474                 e = (a->client.opener)(&a->archive, a->client.dataset[0].data);
 475                 if (e != 0) {
 476                         /* If the open failed, call the closer to clean up. */
 477                         if (a->client.closer) {
 478                                 for (i = 0; i < a->client.nodes; i++)
 479                                         (a->client.closer)(&a->archive,
 480                                             a->client.dataset[i].data);
 481                         }
 482                         return (e);
 483                 }
 484         }
 485
 486         filter = calloc(1, sizeof(*filter));
 487         if (filter == NULL)
 488                 return (ARCHIVE_FATAL);
 489         filter->bidder = NULL;
 490         filter->upstream = NULL;
 491         filter->archive = a;
 492         filter->data = a->client.dataset[0].data;
 493         filter->open = client_open_proxy;
 494         filter->read = client_read_proxy;
 495         filter->skip = client_skip_proxy;
 496         filter->seek = client_seek_proxy;
 497         filter->close = client_close_proxy;
 498         filter->sswitch = client_switch_proxy;
 499         filter->name = "none";
 500         filter->code = ARCHIVE_FILTER_NONE;
 501
 502         a->client.dataset[0].begin_position = 0;
 503         if (!a->filter || !a->bypass_filter_bidding)
 504         {
 505                 a->filter = filter;
 506                 /* Build out the input pipeline. */
 507                 e = choose_filters(a);
 508                 if (e < ARCHIVE_WARN) {
 509                         a->archive.state = ARCHIVE_STATE_FATAL;
 510                         return (ARCHIVE_FATAL);
 511                 }
 512         }
 513         else
 514         {
 515                 /* Need to add "NONE" type filter at the end of the filter chain */
 516                 tmp = a->filter;
 517                 while (tmp->upstream)
 518                         tmp = tmp->upstream;
 519                 tmp->upstream = filter;
 520         }
 521
 522         if (!a->format)
 523         {
 524                 slot = choose_format(a);
 525                 if (slot < 0) {
 526                         __archive_read_close_filters(a);
 527                         a->archive.state = ARCHIVE_STATE_FATAL;
 528                         return (ARCHIVE_FATAL);
 529                 }
 530                 a->format = &(a->formats[slot]);
 531         }
 532
 533         a->archive.state = ARCHIVE_STATE_HEADER;
 534
 535         /* Ensure libarchive starts from the first node in a multivolume set */
 536         client_switch_proxy(a->filter, 0);
 537         return (e);
 538 }
 539
 540 /*
 541  * Allow each registered stream transform to bid on whether
 542  * it wants to handle this stream.  Repeat until we've finished
 543  * building the pipeline.
 544  */
 545 static int
 546 choose_filters(struct archive_read *a)
 547 {
 548         int number_bidders, i, bid, best_bid, n;
 549         struct archive_read_filter_bidder *bidder, *best_bidder;
 550         struct archive_read_filter *filter;
 551         ssize_t avail;
 552         int r;
 553
 554         for (n = 0; n < 25; ++n) {
 555                 number_bidders = sizeof(a->bidders) / sizeof(a->bidders[0]);
 556
 557                 best_bid = 0;
 558                 best_bidder = NULL;
 559
 560                 bidder = a->bidders;
 561                 for (i = 0; i < number_bidders; i++, bidder++) {
 562                         if (bidder->bid != NULL) {
 563                                 bid = (bidder->bid)(bidder, a->filter);
 564                                 if (bid > best_bid) {
 565                                         best_bid = bid;
 566                                         best_bidder = bidder;
 567                                 }
 568                         }
 569                 }
 570
 571                 /* If no bidder, we're done. */
 572                 if (best_bidder == NULL) {
 573                         /* Verify the filter by asking it for some data. */
 574                         __archive_read_filter_ahead(a->filter, 1, &avail);
 575                         if (avail < 0) {
 576                                 __archive_read_close_filters(a);
 577                                 __archive_read_free_filters(a);
 578                                 return (ARCHIVE_FATAL);
 579                         }
 580                         a->archive.compression_name = a->filter->name;
 581                         a->archive.compression_code = a->filter->code;
 582                         return (ARCHIVE_OK);
 583                 }
 584
 585                 filter
 586                     = (struct archive_read_filter *)calloc(1, sizeof(*filter));
 587                 if (filter == NULL)
 588                         return (ARCHIVE_FATAL);
 589                 filter->bidder = best_bidder;
 590                 filter->archive = a;
 591                 filter->upstream = a->filter;
 592                 a->filter = filter;
 593                 r = (best_bidder->init)(a->filter);
 594                 if (r != ARCHIVE_OK) {
 595                         __archive_read_close_filters(a);
 596                         __archive_read_free_filters(a);
 597                         return (ARCHIVE_FATAL);
 598                 }
 599         }
 600         archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
 601             "Input requires too many filters for decoding");
 602         return (ARCHIVE_FATAL);
 603 }
 604
 605 /*
 606  * Read header of next entry.
 607  */
 608 static int
 609 _archive_read_next_header2(struct archive *_a, struct archive_entry *entry)
 610 {
 611         struct archive_read *a = (struct archive_read *)_a;
 612         int r1 = ARCHIVE_OK, r2;
 613
 614         archive_check_magic(_a, ARCHIVE_READ_MAGIC,
 615             ARCHIVE_STATE_HEADER | ARCHIVE_STATE_DATA,
 616             "archive_read_next_header");
 617
 618         archive_entry_clear(entry);
 619         archive_clear_error(&a->archive);
 620
 621         /*
 622          * If client didn't consume entire data, skip any remainder
 623          * (This is especially important for GNU incremental directories.)
 624          */
 625         if (a->archive.state == ARCHIVE_STATE_DATA) {
 626                 r1 = archive_read_data_skip(&a->archive);
 627                 if (r1 == ARCHIVE_EOF)
 628                         archive_set_error(&a->archive, EIO,
 629                             "Premature end-of-file.");
 630                 if (r1 == ARCHIVE_EOF || r1 == ARCHIVE_FATAL) {
 631                         a->archive.state = ARCHIVE_STATE_FATAL;
 632                         return (ARCHIVE_FATAL);
 633                 }
 634         }
 635
 636         /* Record start-of-header offset in uncompressed stream. */
 637         a->header_position = a->filter->position;
 638
 639         ++_a->file_count;
 640         r2 = (a->format->read_header)(a, entry);
 641
 642         /*
 643          * EOF and FATAL are persistent at this layer.  By
 644          * modifying the state, we guarantee that future calls to
 645          * read a header or read data will fail.
 646          */
 647         switch (r2) {
 648         case ARCHIVE_EOF:
 649                 a->archive.state = ARCHIVE_STATE_EOF;
 650                 --_a->file_count;/* Revert a file counter. */
 651                 break;
 652         case ARCHIVE_OK:
 653                 a->archive.state = ARCHIVE_STATE_DATA;
 654                 break;
 655         case ARCHIVE_WARN:
 656                 a->archive.state = ARCHIVE_STATE_DATA;
 657                 break;
 658         case ARCHIVE_RETRY:
 659                 break;
 660         case ARCHIVE_FATAL:
 661                 a->archive.state = ARCHIVE_STATE_FATAL;
 662                 break;
 663         }
 664
 665         a->read_data_output_offset = 0;
 666         a->read_data_remaining = 0;
 667         a->read_data_is_posix_read = 0;
 668         a->read_data_requested = 0;
 669         a->data_start_node = a->client.cursor;
 670         /* EOF always wins; otherwise return the worst error. */
 671         return (r2 < r1 || r2 == ARCHIVE_EOF) ? r2 : r1;
 672 }
 673
 674 int
 675 _archive_read_next_header(struct archive *_a, struct archive_entry **entryp)
 676 {
 677         int ret;
 678         struct archive_read *a = (struct archive_read *)_a;
 679         *entryp = NULL;
 680         ret = _archive_read_next_header2(_a, a->entry);
 681         *entryp = a->entry;
 682         return ret;
 683 }
 684
 685 /*
 686  * Allow each registered format to bid on whether it wants to handle
 687  * the next entry.  Return index of winning bidder.
 688  */
 689 static int
 690 choose_format(struct archive_read *a)
 691 {
 692         int slots;
 693         int i;
 694         int bid, best_bid;
 695         int best_bid_slot;
 696
 697         slots = sizeof(a->formats) / sizeof(a->formats[0]);
 698         best_bid = -1;
 699         best_bid_slot = -1;
 700
 701         /* Set up a->format for convenience of bidders. */
 702         a->format = &(a->formats[0]);
 703         for (i = 0; i < slots; i++, a->format++) {
 704                 if (a->format->bid) {
 705                         bid = (a->format->bid)(a, best_bid);
 706                         if (bid == ARCHIVE_FATAL)
 707                                 return (ARCHIVE_FATAL);
 708                         if (a->filter->position != 0)
 709                                 __archive_read_seek(a, 0, SEEK_SET);
 710                         if ((bid > best_bid) || (best_bid_slot < 0)) {
 711                                 best_bid = bid;
 712                                 best_bid_slot = i;
 713                         }
 714                 }
 715         }
 716
 717         /*
 718          * There were no bidders; this is a serious programmer error
 719          * and demands a quick and definitive abort.
 720          */
 721         if (best_bid_slot < 0) {
 722                 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
 723                     "No formats registered");
 724                 return (ARCHIVE_FATAL);
 725         }
 726
 727         /*
 728          * There were bidders, but no non-zero bids; this means we
 729          * can't support this stream.
 730          */
 731         if (best_bid < 1) {
 732                 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
 733                     "Unrecognized archive format");
 734                 return (ARCHIVE_FATAL);
 735         }
 736
 737         return (best_bid_slot);
 738 }
 739
 740 /*
 741  * Return the file offset (within the uncompressed data stream) where
 742  * the last header started.
 743  */
 744 int64_t
 745 archive_read_header_position(struct archive *_a)
 746 {
 747         struct archive_read *a = (struct archive_read *)_a;
 748         archive_check_magic(_a, ARCHIVE_READ_MAGIC,
 749             ARCHIVE_STATE_ANY, "archive_read_header_position");
 750         return (a->header_position);
 751 }
 752
 753 /*
 754  * Read data from an archive entry, using a read(2)-style interface.
 755  * This is a convenience routine that just calls
 756  * archive_read_data_block and copies the results into the client
 757  * buffer, filling any gaps with zero bytes.  Clients using this
 758  * API can be completely ignorant of sparse-file issues; sparse files
 759  * will simply be padded with nulls.
 760  *
 761  * DO NOT intermingle calls to this function and archive_read_data_block
 762  * to read a single entry body.
 763  */
 764 ssize_t
 765 archive_read_data(struct archive *_a, void *buff, size_t s)
 766 {
 767         struct archive_read *a = (struct archive_read *)_a;
 768         char    *dest;
 769         const void *read_buf;
 770         size_t   bytes_read;
 771         size_t   len;
 772         int      r;
 773
 774         bytes_read = 0;
 775         dest = (char *)buff;
 776
 777         while (s > 0) {
 778                 if (a->read_data_remaining == 0) {
 779                         read_buf = a->read_data_block;
 780                         a->read_data_is_posix_read = 1;
 781                         a->read_data_requested = s;
 782                         r = _archive_read_data_block(&a->archive, &read_buf,
 783                             &a->read_data_remaining, &a->read_data_offset);
 784                         a->read_data_block = read_buf;
 785                         if (r == ARCHIVE_EOF)
 786                                 return (bytes_read);
 787                         /*
 788                          * Error codes are all negative, so the status
 789                          * return here cannot be confused with a valid
 790                          * byte count.  (ARCHIVE_OK is zero.)
 791                          */
 792                         if (r < ARCHIVE_OK)
 793                                 return (r);
 794                 }
 795
 796                 if (a->read_data_offset < a->read_data_output_offset) {
 797                         archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
 798                             "Encountered out-of-order sparse blocks");
 799                         return (ARCHIVE_RETRY);
 800                 }
 801
 802                 /* Compute the amount of zero padding needed. */
 803                 if (a->read_data_output_offset + (int64_t)s <
 804                     a->read_data_offset) {
 805                         len = s;
 806                 } else if (a->read_data_output_offset <
 807                     a->read_data_offset) {
 808                         len = (size_t)(a->read_data_offset -
 809                             a->read_data_output_offset);
 810                 } else
 811                         len = 0;
 812
 813                 /* Add zeroes. */
 814                 memset(dest, 0, len);
 815                 s -= len;
 816                 a->read_data_output_offset += len;
 817                 dest += len;
 818                 bytes_read += len;
 819
 820                 /* Copy data if there is any space left. */
 821                 if (s > 0) {
 822                         len = a->read_data_remaining;
 823                         if (len > s)
 824                                 len = s;
 825                         memcpy(dest, a->read_data_block, len);
 826                         s -= len;
 827                         a->read_data_block += len;
 828                         a->read_data_remaining -= len;
 829                         a->read_data_output_offset += len;
 830                         a->read_data_offset += len;
 831                         dest += len;
 832                         bytes_read += len;
 833                 }
 834         }
 835         a->read_data_is_posix_read = 0;
 836         a->read_data_requested = 0;
 837         return (bytes_read);
 838 }
 839
 840 /*
 841  * Skip over all remaining data in this entry.
 842  */
 843 int
 844 archive_read_data_skip(struct archive *_a)
 845 {
 846         struct archive_read *a = (struct archive_read *)_a;
 847         int r;
 848         const void *buff;
 849         size_t size;
 850         int64_t offset;
 851
 852         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_DATA,
 853             "archive_read_data_skip");
 854
 855         if (a->format->read_data_skip != NULL)
 856                 r = (a->format->read_data_skip)(a);
 857         else {
 858                 while ((r = archive_read_data_block(&a->archive,
 859                             &buff, &size, &offset))
 860                     == ARCHIVE_OK)
 861                         ;
 862         }
 863
 864         if (r == ARCHIVE_EOF)
 865                 r = ARCHIVE_OK;
 866
 867         a->archive.state = ARCHIVE_STATE_HEADER;
 868         return (r);
 869 }
 870
 871 int64_t
 872 archive_seek_data(struct archive *_a, int64_t offset, int whence)
 873 {
 874         struct archive_read *a = (struct archive_read *)_a;
 875         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_DATA,
 876             "archive_seek_data_block");
 877
 878         if (a->format->seek_data == NULL) {
 879                 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
 880                     "Internal error: "
 881                     "No format_seek_data_block function registered");
 882                 return (ARCHIVE_FATAL);
 883         }
 884
 885         return (a->format->seek_data)(a, offset, whence);
 886 }
 887
 888 /*
 889  * Read the next block of entry data from the archive.
 890  * This is a zero-copy interface; the client receives a pointer,
 891  * size, and file offset of the next available block of data.
 892  *
 893  * Returns ARCHIVE_OK if the operation is successful, ARCHIVE_EOF if
 894  * the end of entry is encountered.
 895  */
 896 static int
 897 _archive_read_data_block(struct archive *_a,
 898     const void **buff, size_t *size, int64_t *offset)
 899 {
 900         struct archive_read *a = (struct archive_read *)_a;
 901         archive_check_magic(_a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_DATA,
 902             "archive_read_data_block");
 903
 904         if (a->format->read_data == NULL) {
 905                 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
 906                     "Internal error: "
 907                     "No format_read_data_block function registered");
 908                 return (ARCHIVE_FATAL);
 909         }
 910
 911         return (a->format->read_data)(a, buff, size, offset);
 912 }
 913
 914 int
 915 __archive_read_close_filters(struct archive_read *a)
 916 {
 917         struct archive_read_filter *f = a->filter;
 918         int r = ARCHIVE_OK;
 919         /* Close each filter in the pipeline. */
 920         while (f != NULL) {
 921                 struct archive_read_filter *t = f->upstream;
 922                 if (!f->closed && f->close != NULL) {
 923                         int r1 = (f->close)(f);
 924                         f->closed = 1;
 925                         if (r1 < r)
 926                                 r = r1;
 927                 }
 928                 free(f->buffer);
 929                 f->buffer = NULL;
 930                 f = t;
 931         }
 932         return r;
 933 }
 934
 935 void
 936 __archive_read_free_filters(struct archive_read *a)
 937 {
 938         while (a->filter != NULL) {
 939                 struct archive_read_filter *t = a->filter->upstream;
 940                 free(a->filter);
 941                 a->filter = t;
 942         }
 943 }
 944
 945 /*
 946  * return the count of # of filters in use
 947  */
 948 static int
 949 _archive_filter_count(struct archive *_a)
 950 {
 951         struct archive_read *a = (struct archive_read *)_a;
 952         struct archive_read_filter *p = a->filter;
 953         int count = 0;
 954         while(p) {
 955                 count++;
 956                 p = p->upstream;
 957         }
 958         return count;
 959 }
 960
 961 /*
 962  * Close the file and all I/O.
 963  */
 964 static int
 965 _archive_read_close(struct archive *_a)
 966 {
 967         struct archive_read *a = (struct archive_read *)_a;
 968         int r = ARCHIVE_OK, r1 = ARCHIVE_OK;
 969
 970         archive_check_magic(&a->archive, ARCHIVE_READ_MAGIC,
 971             ARCHIVE_STATE_ANY | ARCHIVE_STATE_FATAL, "archive_read_close");
 972         if (a->archive.state == ARCHIVE_STATE_CLOSED)
 973                 return (ARCHIVE_OK);
 974         archive_clear_error(&a->archive);
 975         a->archive.state = ARCHIVE_STATE_CLOSED;
 976
 977         /* TODO: Clean up the formatters. */
 978
 979         /* Release the filter objects. */
 980         r1 = __archive_read_close_filters(a);
 981         if (r1 < r)
 982                 r = r1;
 983
 984         return (r);
 985 }
 986
 987 /*
 988  * Release memory and other resources.
 989  */
 990 static int
 991 _archive_read_free(struct archive *_a)
 992 {
 993         struct archive_read *a = (struct archive_read *)_a;
 994         int i, n;
 995         int slots;
 996         int r = ARCHIVE_OK;
 997
 998         if (_a == NULL)
 999                 return (ARCHIVE_OK);
1000         archive_check_magic(_a, ARCHIVE_READ_MAGIC,
1001             ARCHIVE_STATE_ANY | ARCHIVE_STATE_FATAL, "archive_read_free");
1002         if (a->archive.state != ARCHIVE_STATE_CLOSED
1003             && a->archive.state != ARCHIVE_STATE_FATAL)
1004                 r = archive_read_close(&a->archive);
1005
1006         /* Call cleanup functions registered by optional components. */
1007         if (a->cleanup_archive_extract != NULL)
1008                 r = (a->cleanup_archive_extract)(a);
1009
1010         /* Cleanup format-specific data. */
1011         slots = sizeof(a->formats) / sizeof(a->formats[0]);
1012         for (i = 0; i < slots; i++) {
1013                 a->format = &(a->formats[i]);
1014                 if (a->formats[i].cleanup)
1015                         (a->formats[i].cleanup)(a);
1016         }
1017
1018         /* Free the filters */
1019         __archive_read_free_filters(a);
1020
1021         /* Release the bidder objects. */
1022         n = sizeof(a->bidders)/sizeof(a->bidders[0]);
1023         for (i = 0; i < n; i++) {
1024                 if (a->bidders[i].free != NULL) {
1025                         int r1 = (a->bidders[i].free)(&a->bidders[i]);
1026                         if (r1 < r)
1027                                 r = r1;
1028                 }
1029         }
1030
1031         archive_string_free(&a->archive.error_string);
1032         if (a->entry)
1033                 archive_entry_free(a->entry);
1034         a->archive.magic = 0;
1035         __archive_clean(&a->archive);
1036         free(a->client.dataset);
1037         free(a);
1038         return (r);
1039 }
1040
1041 static struct archive_read_filter *
1042 get_filter(struct archive *_a, int n)
1043 {
1044         struct archive_read *a = (struct archive_read *)_a;
1045         struct archive_read_filter *f = a->filter;
1046         /* We use n == -1 for 'the last filter', which is always the
1047          * client proxy. */
1048         if (n == -1 && f != NULL) {
1049                 struct archive_read_filter *last = f;
1050                 f = f->upstream;
1051                 while (f != NULL) {
1052                         last = f;
1053                         f = f->upstream;
1054                 }
1055                 return (last);
1056         }
1057         if (n < 0)
1058                 return NULL;
1059         while (n > 0 && f != NULL) {
1060                 f = f->upstream;
1061                 --n;
1062         }
1063         return (f);
1064 }
1065
1066 static int
1067 _archive_filter_code(struct archive *_a, int n)
1068 {
1069         struct archive_read_filter *f = get_filter(_a, n);
1070         return f == NULL ? -1 : f->code;
1071 }
1072
1073 static const char *
1074 _archive_filter_name(struct archive *_a, int n)
1075 {
1076         struct archive_read_filter *f = get_filter(_a, n);
1077         return f == NULL ? NULL : f->name;
1078 }
1079
1080 static int64_t
1081 _archive_filter_bytes(struct archive *_a, int n)
1082 {
1083         struct archive_read_filter *f = get_filter(_a, n);
1084         return f == NULL ? -1 : f->position;
1085 }
1086
1087 /*
1088  * Used internally by read format handlers to register their bid and
1089  * initialization functions.
1090  */
1091 int
1092 __archive_read_register_format(struct archive_read *a,
1093     void *format_data,
1094     const char *name,
1095     int (*bid)(struct archive_read *, int),
1096     int (*options)(struct archive_read *, const char *, const char *),
1097     int (*read_header)(struct archive_read *, struct archive_entry *),
1098     int (*read_data)(struct archive_read *, const void **, size_t *, int64_t *),
1099     int (*read_data_skip)(struct archive_read *),
1100     int64_t (*seek_data)(struct archive_read *, int64_t, int),
1101     int (*cleanup)(struct archive_read *))
1102 {
1103         int i, number_slots;
1104
1105         archive_check_magic(&a->archive,
1106             ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
1107             "__archive_read_register_format");
1108
1109         number_slots = sizeof(a->formats) / sizeof(a->formats[0]);
1110
1111         for (i = 0; i < number_slots; i++) {
1112                 if (a->formats[i].bid == bid)
1113                         return (ARCHIVE_WARN); /* We've already installed */
1114                 if (a->formats[i].bid == NULL) {
1115                         a->formats[i].bid = bid;
1116                         a->formats[i].options = options;
1117                         a->formats[i].read_header = read_header;
1118                         a->formats[i].read_data = read_data;
1119                         a->formats[i].read_data_skip = read_data_skip;
1120                         a->formats[i].seek_data = seek_data;
1121                         a->formats[i].cleanup = cleanup;
1122                         a->formats[i].data = format_data;
1123                         a->formats[i].name = name;
1124                         return (ARCHIVE_OK);
1125                 }
1126         }
1127
1128         archive_set_error(&a->archive, ENOMEM,
1129             "Not enough slots for format registration");
1130         return (ARCHIVE_FATAL);
1131 }
1132
1133 /*
1134  * Used internally by decompression routines to register their bid and
1135  * initialization functions.
1136  */
1137 int
1138 __archive_read_get_bidder(struct archive_read *a,
1139     struct archive_read_filter_bidder **bidder)
1140 {
1141         int i, number_slots;
1142
1143         number_slots = sizeof(a->bidders) / sizeof(a->bidders[0]);
1144
1145         for (i = 0; i < number_slots; i++) {
1146                 if (a->bidders[i].bid == NULL) {
1147                         memset(a->bidders + i, 0, sizeof(a->bidders[0]));
1148                         *bidder = (a->bidders + i);
1149                         return (ARCHIVE_OK);
1150                 }
1151         }
1152
1153         archive_set_error(&a->archive, ENOMEM,
1154             "Not enough slots for filter registration");
1155         return (ARCHIVE_FATAL);
1156 }
1157
1158 /*
1159  * The next section implements the peek/consume internal I/O
1160  * system used by archive readers.  This system allows simple
1161  * read-ahead for consumers while preserving zero-copy operation
1162  * most of the time.
1163  *
1164  * The two key operations:
1165  *  * The read-ahead function returns a pointer to a block of data
1166  *    that satisfies a minimum request.
1167  *  * The consume function advances the file pointer.
1168  *
1169  * In the ideal case, filters generate blocks of data
1170  * and __archive_read_ahead() just returns pointers directly into
1171  * those blocks.  Then __archive_read_consume() just bumps those
1172  * pointers.  Only if your request would span blocks does the I/O
1173  * layer use a copy buffer to provide you with a contiguous block of
1174  * data.
1175  *
1176  * A couple of useful idioms:
1177  *  * "I just want some data."  Ask for 1 byte and pay attention to
1178  *    the "number of bytes available" from __archive_read_ahead().
1179  *    Consume whatever you actually use.
1180  *  * "I want to output a large block of data."  As above, ask for 1 byte,
1181  *    emit all that's available (up to whatever limit you have), consume
1182  *    it all, then repeat until you're done.  This effectively means that
1183  *    you're passing along the blocks that came from your provider.
1184  *  * "I want to peek ahead by a large amount."  Ask for 4k or so, then
1185  *    double and repeat until you get an error or have enough.  Note
1186  *    that the I/O layer will likely end up expanding its copy buffer
1187  *    to fit your request, so use this technique cautiously.  This
1188  *    technique is used, for example, by some of the format tasting
1189  *    code that has uncertain look-ahead needs.
1190  */
1191
1192 /*
1193  * Looks ahead in the input stream:
1194  *  * If 'avail' pointer is provided, that returns number of bytes available
1195  *    in the current buffer, which may be much larger than requested.
1196  *  * If end-of-file, *avail gets set to zero.
1197  *  * If error, *avail gets error code.
1198  *  * If request can be met, returns pointer to data.
1199  *  * If minimum request cannot be met, returns NULL.
1200  *
1201  * Note: If you just want "some data", ask for 1 byte and pay attention
1202  * to *avail, which will have the actual amount available.  If you
1203  * know exactly how many bytes you need, just ask for that and treat
1204  * a NULL return as an error.
1205  *
1206  * Important:  This does NOT move the file pointer.  See
1207  * __archive_read_consume() below.
1208  */
1209 const void *
1210 __archive_read_ahead(struct archive_read *a, size_t min, ssize_t *avail)
1211 {
1212         return (__archive_read_filter_ahead(a->filter, min, avail));
1213 }
1214
1215 const void *
1216 __archive_read_filter_ahead(struct archive_read_filter *filter,
1217     size_t min, ssize_t *avail)
1218 {
1219         ssize_t bytes_read;
1220         size_t tocopy;
1221
1222         if (filter->fatal) {
1223                 if (avail)
1224                         *avail = ARCHIVE_FATAL;
1225                 return (NULL);
1226         }
1227
1228         /*
1229          * Keep pulling more data until we can satisfy the request.
1230          */
1231         for (;;) {
1232
1233                 /*
1234                  * If we can satisfy from the copy buffer (and the
1235                  * copy buffer isn't empty), we're done.  In particular,
1236                  * note that min == 0 is a perfectly well-defined
1237                  * request.
1238                  */
1239                 if (filter->avail >= min && filter->avail > 0) {
1240                         if (avail != NULL)
1241                                 *avail = filter->avail;
1242                         return (filter->next);
1243                 }
1244
1245                 /*
1246                  * We can satisfy directly from client buffer if everything
1247                  * currently in the copy buffer is still in the client buffer.
1248                  */
1249                 if (filter->client_total >= filter->client_avail + filter->avail
1250                     && filter->client_avail + filter->avail >= min) {
1251                         /* "Roll back" to client buffer. */
1252                         filter->client_avail += filter->avail;
1253                         filter->client_next -= filter->avail;
1254                         /* Copy buffer is now empty. */
1255                         filter->avail = 0;
1256                         filter->next = filter->buffer;
1257                         /* Return data from client buffer. */
1258                         if (avail != NULL)
1259                                 *avail = filter->client_avail;
1260                         return (filter->client_next);
1261                 }
1262
1263                 /* Move data forward in copy buffer if necessary. */
1264                 if (filter->next > filter->buffer &&
1265                     filter->next + min > filter->buffer + filter->buffer_size) {
1266                         if (filter->avail > 0)
1267                                 memmove(filter->buffer, filter->next,
1268                                     filter->avail);
1269                         filter->next = filter->buffer;
1270                 }
1271
1272                 /* If we've used up the client data, get more. */
1273                 if (filter->client_avail <= 0) {
1274                         if (filter->end_of_file) {
1275                                 if (avail != NULL)
1276                                         *avail = 0;
1277                                 return (NULL);
1278                         }
1279                         bytes_read = (filter->read)(filter,
1280                             &filter->client_buff);
1281                         if (bytes_read < 0) {           /* Read error. */
1282                                 filter->client_total = filter->client_avail = 0;
1283                                 filter->client_next =
1284                                     filter->client_buff = NULL;
1285                                 filter->fatal = 1;
1286                                 if (avail != NULL)
1287                                         *avail = ARCHIVE_FATAL;
1288                                 return (NULL);
1289                         }
1290                         if (bytes_read == 0) {
1291                                 /* Check for another client object first */
1292                                 if (filter->archive->client.cursor !=
1293                                       filter->archive->client.nodes - 1) {
1294                                         if (client_switch_proxy(filter,
1295                                             filter->archive->client.cursor + 1)
1296                                             == ARCHIVE_OK)
1297                                                 continue;
1298                                 }
1299                                 /* Premature end-of-file. */
1300                                 filter->client_total = filter->client_avail = 0;
1301                                 filter->client_next =
1302                                     filter->client_buff = NULL;
1303                                 filter->end_of_file = 1;
1304                                 /* Return whatever we do have. */
1305                                 if (avail != NULL)
1306                                         *avail = filter->avail;
1307                                 return (NULL);
1308                         }
1309                         filter->client_total = bytes_read;
1310                         filter->client_avail = filter->client_total;
1311                         filter->client_next = filter->client_buff;
1312                 } else {
1313                         /*
1314                          * We can't satisfy the request from the copy
1315                          * buffer or the existing client data, so we
1316                          * need to copy more client data over to the
1317                          * copy buffer.
1318                          */
1319
1320                         /* Ensure the buffer is big enough. */
1321                         if (min > filter->buffer_size) {
1322                                 size_t s, t;
1323                                 char *p;
1324
1325                                 /* Double the buffer; watch for overflow. */
1326                                 s = t = filter->buffer_size;
1327                                 if (s == 0)
1328                                         s = min;
1329                                 while (s < min) {
1330                                         t *= 2;
1331                                         if (t <= s) { /* Integer overflow! */
1332                                                 archive_set_error(
1333                                                     &filter->archive->archive,
1334                                                     ENOMEM,
1335                                                     "Unable to allocate copy"
1336                                                     " buffer");
1337                                                 filter->fatal = 1;
1338                                                 if (avail != NULL)
1339                                                         *avail = ARCHIVE_FATAL;
1340                                                 return (NULL);
1341                                         }
1342                                         s = t;
1343                                 }
1344                                 /* Now s >= min, so allocate a new buffer. */
1345                                 p = (char *)malloc(s);
1346                                 if (p == NULL) {
1347                                         archive_set_error(
1348                                                 &filter->archive->archive,
1349                                                 ENOMEM,
1350                                             "Unable to allocate copy buffer");
1351                                         filter->fatal = 1;
1352                                         if (avail != NULL)
1353                                                 *avail = ARCHIVE_FATAL;
1354                                         return (NULL);
1355                                 }
1356                                 /* Move data into newly-enlarged buffer. */
1357                                 if (filter->avail > 0)
1358                                         memmove(p, filter->next, filter->avail);
1359                                 free(filter->buffer);
1360                                 filter->next = filter->buffer = p;
1361                                 filter->buffer_size = s;
1362                         }
1363
1364                         /* We can add client data to copy buffer. */
1365                         /* First estimate: copy to fill rest of buffer. */
1366                         tocopy = (filter->buffer + filter->buffer_size)
1367                             - (filter->next + filter->avail);
1368                         /* Don't waste time buffering more than we need to. */
1369                         if (tocopy + filter->avail > min)
1370                                 tocopy = min - filter->avail;
1371                         /* Don't copy more than is available. */
1372                         if (tocopy > filter->client_avail)
1373                                 tocopy = filter->client_avail;
1374
1375                         memcpy(filter->next + filter->avail,
1376                             filter->client_next, tocopy);
1377                         /* Remove this data from client buffer. */
1378                         filter->client_next += tocopy;
1379                         filter->client_avail -= tocopy;
1380                         /* add it to copy buffer. */
1381                         filter->avail += tocopy;
1382                 }
1383         }
1384 }
1385
1386 /*
1387  * Move the file pointer forward.
1388  */
1389 int64_t
1390 __archive_read_consume(struct archive_read *a, int64_t request)
1391 {
1392         return (__archive_read_filter_consume(a->filter, request));
1393 }
1394
1395 int64_t
1396 __archive_read_filter_consume(struct archive_read_filter * filter,
1397     int64_t request)
1398 {
1399         int64_t skipped;
1400
1401         if (request < 0)
1402                 return ARCHIVE_FATAL;
1403         if (request == 0)
1404                 return 0;
1405
1406         skipped = advance_file_pointer(filter, request);
1407         if (skipped == request)
1408                 return (skipped);
1409         /* We hit EOF before we satisfied the skip request. */
1410         if (skipped < 0)  /* Map error code to 0 for error message below. */
1411                 skipped = 0;
1412         archive_set_error(&filter->archive->archive,
1413             ARCHIVE_ERRNO_MISC,
1414             "Truncated input file (needed %jd bytes, only %jd available)",
1415             (intmax_t)request, (intmax_t)skipped);
1416         return (ARCHIVE_FATAL);
1417 }
1418
1419 /*
1420  * Advance the file pointer by the amount requested.
1421  * Returns the amount actually advanced, which may be less than the
1422  * request if EOF is encountered first.
1423  * Returns a negative value if there's an I/O error.
1424  */
1425 static int64_t
1426 advance_file_pointer(struct archive_read_filter *filter, int64_t request)
1427 {
1428         int64_t bytes_skipped, total_bytes_skipped = 0;
1429         ssize_t bytes_read;
1430         size_t min;
1431
1432         if (filter->fatal)
1433                 return (-1);
1434
1435         /* Use up the copy buffer first. */
1436         if (filter->avail > 0) {
1437                 min = (size_t)minimum(request, (int64_t)filter->avail);
1438                 filter->next += min;
1439                 filter->avail -= min;
1440                 request -= min;
1441                 filter->position += min;
1442                 total_bytes_skipped += min;
1443         }
1444
1445         /* Then use up the client buffer. */
1446         if (filter->client_avail > 0) {
1447                 min = (size_t)minimum(request, (int64_t)filter->client_avail);
1448                 filter->client_next += min;
1449                 filter->client_avail -= min;
1450                 request -= min;
1451                 filter->position += min;
1452                 total_bytes_skipped += min;
1453         }
1454         if (request == 0)
1455                 return (total_bytes_skipped);
1456
1457         /* If there's an optimized skip function, use it. */
1458         if (filter->skip != NULL) {
1459                 bytes_skipped = (filter->skip)(filter, request);
1460                 if (bytes_skipped < 0) {        /* error */
1461                         filter->fatal = 1;
1462                         return (bytes_skipped);
1463                 }
1464                 filter->position += bytes_skipped;
1465                 total_bytes_skipped += bytes_skipped;
1466                 request -= bytes_skipped;
1467                 if (request == 0)
1468                         return (total_bytes_skipped);
1469         }
1470
1471         /* Use ordinary reads as necessary to complete the request. */
1472         for (;;) {
1473                 bytes_read = (filter->read)(filter, &filter->client_buff);
1474                 if (bytes_read < 0) {
1475                         filter->client_buff = NULL;
1476                         filter->fatal = 1;
1477                         return (bytes_read);
1478                 }
1479
1480                 if (bytes_read == 0) {
1481                         if (filter->archive->client.cursor !=
1482                               filter->archive->client.nodes - 1) {
1483                                 if (client_switch_proxy(filter,
1484                                     filter->archive->client.cursor + 1)
1485                                     == ARCHIVE_OK)
1486                                         continue;
1487                         }
1488                         filter->client_buff = NULL;
1489                         filter->end_of_file = 1;
1490                         return (total_bytes_skipped);
1491                 }
1492
1493                 if (bytes_read >= request) {
1494                         filter->client_next =
1495                             ((const char *)filter->client_buff) + request;
1496                         filter->client_avail = (size_t)(bytes_read - request);
1497                         filter->client_total = bytes_read;
1498                         total_bytes_skipped += request;
1499                         filter->position += request;
1500                         return (total_bytes_skipped);
1501                 }
1502
1503                 filter->position += bytes_read;
1504                 total_bytes_skipped += bytes_read;
1505                 request -= bytes_read;
1506         }
1507 }
1508
1509 /**
1510  * Returns ARCHIVE_FAILED if seeking isn't supported.
1511  */
1512 int64_t
1513 __archive_read_seek(struct archive_read *a, int64_t offset, int whence)
1514 {
1515         return __archive_read_filter_seek(a->filter, offset, whence);
1516 }
1517
1518 int64_t
1519 __archive_read_filter_seek(struct archive_read_filter *filter, int64_t offset,
1520     int whence)
1521 {
1522         struct archive_read_client *client;
1523         int64_t r;
1524         unsigned int cursor;
1525
1526         if (filter->closed || filter->fatal)
1527                 return (ARCHIVE_FATAL);
1528         if (filter->seek == NULL)
1529                 return (ARCHIVE_FAILED);
1530
1531         client = &(filter->archive->client);
1532         switch (whence) {
1533         case SEEK_CUR:
1534                 /* Adjust the offset and use SEEK_SET instead */
1535                 offset += filter->position;
1536         case SEEK_SET:
1537                 cursor = 0;
1538                 while (1)
1539                 {
1540                         if (client->dataset[cursor].begin_position < 0 ||
1541                             client->dataset[cursor].total_size < 0 ||
1542                             client->dataset[cursor].begin_position +
1543                               client->dataset[cursor].total_size - 1 > offset ||
1544                             cursor + 1 >= client->nodes)
1545                                 break;
1546                         r = client->dataset[cursor].begin_position +
1547                                 client->dataset[cursor].total_size;
1548                         client->dataset[++cursor].begin_position = r;
1549                 }
1550                 while (1) {
1551                         r = client_switch_proxy(filter, cursor);
1552                         if (r != ARCHIVE_OK)
1553                                 return r;
1554                         if ((r = client_seek_proxy(filter, 0, SEEK_END)) < 0)
1555                                 return r;
1556                         client->dataset[cursor].total_size = r;
1557                         if (client->dataset[cursor].begin_position +
1558                             client->dataset[cursor].total_size - 1 > offset ||
1559                             cursor + 1 >= client->nodes)
1560                                 break;
1561                         r = client->dataset[cursor].begin_position +
1562                                 client->dataset[cursor].total_size;
1563                         client->dataset[++cursor].begin_position = r;
1564                 }
1565                 offset -= client->dataset[cursor].begin_position;
1566                 if (offset < 0)
1567                         offset = 0;
1568                 else if (offset > client->dataset[cursor].total_size - 1)
1569                         offset = client->dataset[cursor].total_size - 1;
1570                 if ((r = client_seek_proxy(filter, offset, SEEK_SET)) < 0)
1571                         return r;
1572                 break;
1573
1574         case SEEK_END:
1575                 cursor = 0;
1576                 while (1) {
1577                         if (client->dataset[cursor].begin_position < 0 ||
1578                             client->dataset[cursor].total_size < 0 ||
1579                             cursor + 1 >= client->nodes)
1580                                 break;
1581                         r = client->dataset[cursor].begin_position +
1582                                 client->dataset[cursor].total_size;
1583                         client->dataset[++cursor].begin_position = r;
1584                 }
1585                 while (1) {
1586                         r = client_switch_proxy(filter, cursor);
1587                         if (r != ARCHIVE_OK)
1588                                 return r;
1589                         if ((r = client_seek_proxy(filter, 0, SEEK_END)) < 0)
1590                                 return r;
1591                         client->dataset[cursor].total_size = r;
1592                         r = client->dataset[cursor].begin_position +
1593                                 client->dataset[cursor].total_size;
1594                         if (cursor + 1 >= client->nodes)
1595                                 break;
1596                         client->dataset[++cursor].begin_position = r;
1597                 }
1598                 while (1) {
1599                         if (r + offset >=
1600                             client->dataset[cursor].begin_position)
1601                                 break;
1602                         offset += client->dataset[cursor].total_size;
1603                         if (cursor == 0)
1604                                 break;
1605                         cursor--;
1606                         r = client->dataset[cursor].begin_position +
1607                                 client->dataset[cursor].total_size;
1608                 }
1609                 offset = (r + offset) - client->dataset[cursor].begin_position;
1610                 if ((r = client_switch_proxy(filter, cursor)) != ARCHIVE_OK)
1611                         return r;
1612                 r = client_seek_proxy(filter, offset, SEEK_SET);
1613                 if (r < ARCHIVE_OK)
1614                         return r;
1615                 break;
1616
1617         default:
1618                 return (ARCHIVE_FATAL);
1619         }
1620         r += client->dataset[cursor].begin_position;
1621
1622         if (r >= 0) {
1623                 /*
1624                  * Ouch.  Clearing the buffer like this hurts, especially
1625                  * at bid time.  A lot of our efficiency at bid time comes
1626                  * from having bidders reuse the data we've already read.
1627                  *
1628                  * TODO: If the seek request is in data we already
1629                  * have, then don't call the seek callback.
1630                  *
1631                  * TODO: Zip seeks to end-of-file at bid time.  If
1632                  * other formats also start doing this, we may need to
1633                  * find a way for clients to fudge the seek offset to
1634                  * a block boundary.
1635                  *
1636                  * Hmmm... If whence was SEEK_END, we know the file
1637                  * size is (r - offset).  Can we use that to simplify
1638                  * the TODO items above?
1639                  */
1640                 filter->avail = filter->client_avail = 0;
1641                 filter->next = filter->buffer;
1642                 filter->position = r;
1643                 filter->end_of_file = 0;
1644         }
1645         return r;
1646 }