1/*- 2 * Copyright (c) 2003-2007 Tim Kientzle 3 * Copyright (c) 2011-2012 Michihiro NAKAJIMA 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 27#include "archive_platform.h" 28__FBSDID("$FreeBSD: releng/10.3/contrib/libarchive/libarchive/archive_read_support_format_tar.c 306941 2016-10-10 07:18:54Z delphij $"); 29 30#ifdef HAVE_ERRNO_H 31#include <errno.h> 32#endif 33#include <stddef.h> 34#ifdef HAVE_STDLIB_H 35#include <stdlib.h> 36#endif 37#ifdef HAVE_STRING_H 38#include <string.h> 39#endif 40 41#include "archive.h" 42#include "archive_acl_private.h" /* For ACL parsing routines. */ 43#include "archive_entry.h" 44#include "archive_entry_locale.h" 45#include "archive_private.h" 46#include "archive_read_private.h" 47 48#define tar_min(a,b) ((a) < (b) ? (a) : (b)) 49 50/* 51 * Layout of POSIX 'ustar' tar header. 52 */ 53struct archive_entry_header_ustar { 54 char name[100]; 55 char mode[8]; 56 char uid[8]; 57 char gid[8]; 58 char size[12]; 59 char mtime[12]; 60 char checksum[8]; 61 char typeflag[1]; 62 char linkname[100]; /* "old format" header ends here */ 63 char magic[6]; /* For POSIX: "ustar\0" */ 64 char version[2]; /* For POSIX: "00" */ 65 char uname[32]; 66 char gname[32]; 67 char rdevmajor[8]; 68 char rdevminor[8]; 69 char prefix[155]; 70}; 71 72/* 73 * Structure of GNU tar header 74 */ 75struct gnu_sparse { 76 char offset[12]; 77 char numbytes[12]; 78}; 79 80struct archive_entry_header_gnutar { 81 char name[100]; 82 char mode[8]; 83 char uid[8]; 84 char gid[8]; 85 char size[12]; 86 char mtime[12]; 87 char checksum[8]; 88 char typeflag[1]; 89 char linkname[100]; 90 char magic[8]; /* "ustar \0" (note blank/blank/null at end) */ 91 char uname[32]; 92 char gname[32]; 93 char rdevmajor[8]; 94 char rdevminor[8]; 95 char atime[12]; 96 char ctime[12]; 97 char offset[12]; 98 char longnames[4]; 99 char unused[1]; 100 struct gnu_sparse sparse[4]; 101 char isextended[1]; 102 char realsize[12]; 103 /* 104 * Old GNU format doesn't use POSIX 'prefix' field; they use 105 * the 'L' (longname) entry instead. 106 */ 107}; 108 109/* 110 * Data specific to this format. 111 */ 112struct sparse_block { 113 struct sparse_block *next; 114 int64_t offset; 115 int64_t remaining; 116 int hole; 117}; 118 119struct tar { 120 struct archive_string acl_text; 121 struct archive_string entry_pathname; 122 /* For "GNU.sparse.name" and other similar path extensions. */ 123 struct archive_string entry_pathname_override; 124 struct archive_string entry_linkpath; 125 struct archive_string entry_uname; 126 struct archive_string entry_gname; 127 struct archive_string longlink; 128 struct archive_string longname; 129 struct archive_string pax_header; 130 struct archive_string pax_global; 131 struct archive_string line; 132 int pax_hdrcharset_binary; 133 int header_recursion_depth; 134 int64_t entry_bytes_remaining; 135 int64_t entry_offset; 136 int64_t entry_padding; 137 int64_t entry_bytes_unconsumed; 138 int64_t realsize; 139 int sparse_allowed; 140 struct sparse_block *sparse_list; 141 struct sparse_block *sparse_last; 142 int64_t sparse_offset; 143 int64_t sparse_numbytes; 144 int sparse_gnu_major; 145 int sparse_gnu_minor; 146 char sparse_gnu_pending; 147 148 struct archive_string localname; 149 struct archive_string_conv *opt_sconv; 150 struct archive_string_conv *sconv; 151 struct archive_string_conv *sconv_acl; 152 struct archive_string_conv *sconv_default; 153 int init_default_conversion; 154 int compat_2x; 155}; 156 157static int archive_block_is_null(const char *p); 158static char *base64_decode(const char *, size_t, size_t *); 159static int gnu_add_sparse_entry(struct archive_read *, struct tar *, 160 int64_t offset, int64_t remaining); 161 162static void gnu_clear_sparse_list(struct tar *); 163static int gnu_sparse_old_read(struct archive_read *, struct tar *, 164 const struct archive_entry_header_gnutar *header, size_t *); 165static int gnu_sparse_old_parse(struct archive_read *, struct tar *, 166 const struct gnu_sparse *sparse, int length); 167static int gnu_sparse_01_parse(struct archive_read *, struct tar *, 168 const char *); 169static ssize_t gnu_sparse_10_read(struct archive_read *, struct tar *, 170 size_t *); 171static int header_Solaris_ACL(struct archive_read *, struct tar *, 172 struct archive_entry *, const void *, size_t *); 173static int header_common(struct archive_read *, struct tar *, 174 struct archive_entry *, const void *); 175static int header_old_tar(struct archive_read *, struct tar *, 176 struct archive_entry *, const void *); 177static int header_pax_extensions(struct archive_read *, struct tar *, 178 struct archive_entry *, const void *, size_t *); 179static int header_pax_global(struct archive_read *, struct tar *, 180 struct archive_entry *, const void *h, size_t *); 181static int header_longlink(struct archive_read *, struct tar *, 182 struct archive_entry *, const void *h, size_t *); 183static int header_longname(struct archive_read *, struct tar *, 184 struct archive_entry *, const void *h, size_t *); 185static int read_mac_metadata_blob(struct archive_read *, struct tar *, 186 struct archive_entry *, const void *h, size_t *); 187static int header_volume(struct archive_read *, struct tar *, 188 struct archive_entry *, const void *h, size_t *); 189static int header_ustar(struct archive_read *, struct tar *, 190 struct archive_entry *, const void *h); 191static int header_gnutar(struct archive_read *, struct tar *, 192 struct archive_entry *, const void *h, size_t *); 193static int archive_read_format_tar_bid(struct archive_read *, int); 194static int archive_read_format_tar_options(struct archive_read *, 195 const char *, const char *); 196static int archive_read_format_tar_cleanup(struct archive_read *); 197static int archive_read_format_tar_read_data(struct archive_read *a, 198 const void **buff, size_t *size, int64_t *offset); 199static int archive_read_format_tar_skip(struct archive_read *a); 200static int archive_read_format_tar_read_header(struct archive_read *, 201 struct archive_entry *); 202static int checksum(struct archive_read *, const void *); 203static int pax_attribute(struct archive_read *, struct tar *, 204 struct archive_entry *, char *key, char *value); 205static int pax_header(struct archive_read *, struct tar *, 206 struct archive_entry *, char *attr); 207static void pax_time(const char *, int64_t *sec, long *nanos); 208static ssize_t readline(struct archive_read *, struct tar *, const char **, 209 ssize_t limit, size_t *); 210static int read_body_to_string(struct archive_read *, struct tar *, 211 struct archive_string *, const void *h, size_t *); 212static int solaris_sparse_parse(struct archive_read *, struct tar *, 213 struct archive_entry *, const char *); 214static int64_t tar_atol(const char *, size_t); 215static int64_t tar_atol10(const char *, size_t); 216static int64_t tar_atol256(const char *, size_t); 217static int64_t tar_atol8(const char *, size_t); 218static int tar_read_header(struct archive_read *, struct tar *, 219 struct archive_entry *, size_t *); 220static int tohex(int c); 221static char *url_decode(const char *); 222static void tar_flush_unconsumed(struct archive_read *, size_t *); 223 224 225int 226archive_read_support_format_gnutar(struct archive *a) 227{ 228 archive_check_magic(a, ARCHIVE_READ_MAGIC, 229 ARCHIVE_STATE_NEW, "archive_read_support_format_gnutar"); 230 return (archive_read_support_format_tar(a)); 231} 232 233 234int 235archive_read_support_format_tar(struct archive *_a) 236{ 237 struct archive_read *a = (struct archive_read *)_a; 238 struct tar *tar; 239 int r; 240 241 archive_check_magic(_a, ARCHIVE_READ_MAGIC, 242 ARCHIVE_STATE_NEW, "archive_read_support_format_tar"); 243 244 tar = (struct tar *)calloc(1, sizeof(*tar)); 245 if (tar == NULL) { 246 archive_set_error(&a->archive, ENOMEM, 247 "Can't allocate tar data"); 248 return (ARCHIVE_FATAL); 249 } 250 251 r = __archive_read_register_format(a, tar, "tar", 252 archive_read_format_tar_bid, 253 archive_read_format_tar_options, 254 archive_read_format_tar_read_header, 255 archive_read_format_tar_read_data, 256 archive_read_format_tar_skip, 257 NULL, 258 archive_read_format_tar_cleanup); 259 260 if (r != ARCHIVE_OK) 261 free(tar); 262 return (ARCHIVE_OK); 263} 264 265static int 266archive_read_format_tar_cleanup(struct archive_read *a) 267{ 268 struct tar *tar; 269 270 tar = (struct tar *)(a->format->data); 271 gnu_clear_sparse_list(tar); 272 archive_string_free(&tar->acl_text); 273 archive_string_free(&tar->entry_pathname); 274 archive_string_free(&tar->entry_pathname_override); 275 archive_string_free(&tar->entry_linkpath); 276 archive_string_free(&tar->entry_uname); 277 archive_string_free(&tar->entry_gname); 278 archive_string_free(&tar->line); 279 archive_string_free(&tar->pax_global); 280 archive_string_free(&tar->pax_header); 281 archive_string_free(&tar->longname); 282 archive_string_free(&tar->longlink); 283 archive_string_free(&tar->localname); 284 free(tar); 285 (a->format->data) = NULL; 286 return (ARCHIVE_OK); 287} 288 289 290static int 291archive_read_format_tar_bid(struct archive_read *a, int best_bid) 292{ 293 int bid; 294 const char *h; 295 const struct archive_entry_header_ustar *header; 296 297 (void)best_bid; /* UNUSED */ 298 299 bid = 0; 300 301 /* Now let's look at the actual header and see if it matches. */ 302 h = __archive_read_ahead(a, 512, NULL); 303 if (h == NULL) 304 return (-1); 305 306 /* If it's an end-of-archive mark, we can handle it. */ 307 if (h[0] == 0 && archive_block_is_null(h)) { 308 /* 309 * Usually, I bid the number of bits verified, but 310 * in this case, 4096 seems excessive so I picked 10 as 311 * an arbitrary but reasonable-seeming value. 312 */ 313 return (10); 314 } 315 316 /* If it's not an end-of-archive mark, it must have a valid checksum.*/ 317 if (!checksum(a, h)) 318 return (0); 319 bid += 48; /* Checksum is usually 6 octal digits. */ 320 321 header = (const struct archive_entry_header_ustar *)h; 322 323 /* Recognize POSIX formats. */ 324 if ((memcmp(header->magic, "ustar\0", 6) == 0) 325 && (memcmp(header->version, "00", 2) == 0)) 326 bid += 56; 327 328 /* Recognize GNU tar format. */ 329 if ((memcmp(header->magic, "ustar ", 6) == 0) 330 && (memcmp(header->version, " \0", 2) == 0)) 331 bid += 56; 332 333 /* Type flag must be null, digit or A-Z, a-z. */ 334 if (header->typeflag[0] != 0 && 335 !( header->typeflag[0] >= '0' && header->typeflag[0] <= '9') && 336 !( header->typeflag[0] >= 'A' && header->typeflag[0] <= 'Z') && 337 !( header->typeflag[0] >= 'a' && header->typeflag[0] <= 'z') ) 338 return (0); 339 bid += 2; /* 6 bits of variation in an 8-bit field leaves 2 bits. */ 340 341 /* Sanity check: Look at first byte of mode field. */ 342 switch (255 & (unsigned)header->mode[0]) { 343 case 0: case 255: 344 /* Base-256 value: No further verification possible! */ 345 break; 346 case ' ': /* Not recommended, but not illegal, either. */ 347 break; 348 case '0': case '1': case '2': case '3': 349 case '4': case '5': case '6': case '7': 350 /* Octal Value. */ 351 /* TODO: Check format of remainder of this field. */ 352 break; 353 default: 354 /* Not a valid mode; bail out here. */ 355 return (0); 356 } 357 /* TODO: Sanity test uid/gid/size/mtime/rdevmajor/rdevminor fields. */ 358 359 return (bid); 360} 361 362static int 363archive_read_format_tar_options(struct archive_read *a, 364 const char *key, const char *val) 365{ 366 struct tar *tar; 367 int ret = ARCHIVE_FAILED; 368 369 tar = (struct tar *)(a->format->data); 370 if (strcmp(key, "compat-2x") == 0) { 371 /* Handle UTF-8 filnames as libarchive 2.x */ 372 tar->compat_2x = (val != NULL)?1:0; 373 tar->init_default_conversion = tar->compat_2x; 374 return (ARCHIVE_OK); 375 } else if (strcmp(key, "hdrcharset") == 0) { 376 if (val == NULL || val[0] == 0) 377 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 378 "tar: hdrcharset option needs a character-set name"); 379 else { 380 tar->opt_sconv = 381 archive_string_conversion_from_charset( 382 &a->archive, val, 0); 383 if (tar->opt_sconv != NULL) 384 ret = ARCHIVE_OK; 385 else 386 ret = ARCHIVE_FATAL; 387 } 388 return (ret); 389 } 390 391 /* Note: The "warn" return is just to inform the options 392 * supervisor that we didn't handle it. It will generate 393 * a suitable error if no one used this option. */ 394 return (ARCHIVE_WARN); 395} 396 397/* utility function- this exists to centralize the logic of tracking 398 * how much unconsumed data we have floating around, and to consume 399 * anything outstanding since we're going to do read_aheads 400 */ 401static void 402tar_flush_unconsumed(struct archive_read *a, size_t *unconsumed) 403{ 404 if (*unconsumed) { 405/* 406 void *data = (void *)__archive_read_ahead(a, *unconsumed, NULL); 407 * this block of code is to poison claimed unconsumed space, ensuring 408 * things break if it is in use still. 409 * currently it WILL break things, so enable it only for debugging this issue 410 if (data) { 411 memset(data, 0xff, *unconsumed); 412 } 413*/ 414 __archive_read_consume(a, *unconsumed); 415 *unconsumed = 0; 416 } 417} 418 419/* 420 * The function invoked by archive_read_next_header(). This 421 * just sets up a few things and then calls the internal 422 * tar_read_header() function below. 423 */ 424static int 425archive_read_format_tar_read_header(struct archive_read *a, 426 struct archive_entry *entry) 427{ 428 /* 429 * When converting tar archives to cpio archives, it is 430 * essential that each distinct file have a distinct inode 431 * number. To simplify this, we keep a static count here to 432 * assign fake dev/inode numbers to each tar entry. Note that 433 * pax format archives may overwrite this with something more 434 * useful. 435 * 436 * Ideally, we would track every file read from the archive so 437 * that we could assign the same dev/ino pair to hardlinks, 438 * but the memory required to store a complete lookup table is 439 * probably not worthwhile just to support the relatively 440 * obscure tar->cpio conversion case. 441 */ 442 static int default_inode; 443 static int default_dev; 444 struct tar *tar; 445 const char *p; 446 int r; 447 size_t l, unconsumed = 0; 448 449 /* Assign default device/inode values. */ 450 archive_entry_set_dev(entry, 1 + default_dev); /* Don't use zero. */ 451 archive_entry_set_ino(entry, ++default_inode); /* Don't use zero. */ 452 /* Limit generated st_ino number to 16 bits. */ 453 if (default_inode >= 0xffff) { 454 ++default_dev; 455 default_inode = 0; 456 } 457 458 tar = (struct tar *)(a->format->data); 459 tar->entry_offset = 0; 460 gnu_clear_sparse_list(tar); 461 tar->realsize = -1; /* Mark this as "unset" */ 462 463 /* Setup default string conversion. */ 464 tar->sconv = tar->opt_sconv; 465 if (tar->sconv == NULL) { 466 if (!tar->init_default_conversion) { 467 tar->sconv_default = 468 archive_string_default_conversion_for_read(&(a->archive)); 469 tar->init_default_conversion = 1; 470 } 471 tar->sconv = tar->sconv_default; 472 } 473 474 r = tar_read_header(a, tar, entry, &unconsumed); 475 476 tar_flush_unconsumed(a, &unconsumed); 477 478 /* 479 * "non-sparse" files are really just sparse files with 480 * a single block. 481 */ 482 if (tar->sparse_list == NULL) { 483 if (gnu_add_sparse_entry(a, tar, 0, tar->entry_bytes_remaining) 484 != ARCHIVE_OK) 485 return (ARCHIVE_FATAL); 486 } else { 487 struct sparse_block *sb; 488 489 for (sb = tar->sparse_list; sb != NULL; sb = sb->next) { 490 if (!sb->hole) 491 archive_entry_sparse_add_entry(entry, 492 sb->offset, sb->remaining); 493 } 494 } 495 496 if (r == ARCHIVE_OK) { 497 /* 498 * "Regular" entry with trailing '/' is really 499 * directory: This is needed for certain old tar 500 * variants and even for some broken newer ones. 501 */ 502 const wchar_t *wp; 503 wp = archive_entry_pathname_w(entry); 504 if (wp != NULL) { 505 l = wcslen(wp); 506 if (archive_entry_filetype(entry) == AE_IFREG 507 && wp[l-1] == L'/') 508 archive_entry_set_filetype(entry, AE_IFDIR); 509 } else { 510 p = archive_entry_pathname(entry); 511 if (p == NULL) 512 return (ARCHIVE_FAILED); 513 l = strlen(p); 514 if (archive_entry_filetype(entry) == AE_IFREG 515 && p[l-1] == '/') 516 archive_entry_set_filetype(entry, AE_IFDIR); 517 } 518 } 519 return (r); 520} 521 522static int 523archive_read_format_tar_read_data(struct archive_read *a, 524 const void **buff, size_t *size, int64_t *offset) 525{ 526 ssize_t bytes_read; 527 struct tar *tar; 528 struct sparse_block *p; 529 530 tar = (struct tar *)(a->format->data); 531 532 for (;;) { 533 /* Remove exhausted entries from sparse list. */ 534 while (tar->sparse_list != NULL && 535 tar->sparse_list->remaining == 0) { 536 p = tar->sparse_list; 537 tar->sparse_list = p->next; 538 free(p); 539 } 540 541 if (tar->entry_bytes_unconsumed) { 542 __archive_read_consume(a, tar->entry_bytes_unconsumed); 543 tar->entry_bytes_unconsumed = 0; 544 } 545 546 /* If we're at end of file, return EOF. */ 547 if (tar->sparse_list == NULL || 548 tar->entry_bytes_remaining == 0) { 549 if (__archive_read_consume(a, tar->entry_padding) < 0) 550 return (ARCHIVE_FATAL); 551 tar->entry_padding = 0; 552 *buff = NULL; 553 *size = 0; 554 *offset = tar->realsize; 555 return (ARCHIVE_EOF); 556 } 557 558 *buff = __archive_read_ahead(a, 1, &bytes_read); 559 if (bytes_read < 0) 560 return (ARCHIVE_FATAL); 561 if (*buff == NULL) { 562 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 563 "Truncated tar archive"); 564 return (ARCHIVE_FATAL); 565 } 566 if (bytes_read > tar->entry_bytes_remaining) 567 bytes_read = (ssize_t)tar->entry_bytes_remaining; 568 /* Don't read more than is available in the 569 * current sparse block. */ 570 if (tar->sparse_list->remaining < bytes_read) 571 bytes_read = (ssize_t)tar->sparse_list->remaining; 572 *size = bytes_read; 573 *offset = tar->sparse_list->offset; 574 tar->sparse_list->remaining -= bytes_read; 575 tar->sparse_list->offset += bytes_read; 576 tar->entry_bytes_remaining -= bytes_read; 577 tar->entry_bytes_unconsumed = bytes_read; 578 579 if (!tar->sparse_list->hole) 580 return (ARCHIVE_OK); 581 /* Current is hole data and skip this. */ 582 } 583} 584 585static int 586archive_read_format_tar_skip(struct archive_read *a) 587{ 588 int64_t bytes_skipped; 589 int64_t request; 590 struct sparse_block *p; 591 struct tar* tar; 592 593 tar = (struct tar *)(a->format->data); 594 595 /* Do not consume the hole of a sparse file. */ 596 request = 0; 597 for (p = tar->sparse_list; p != NULL; p = p->next) { 598 if (!p->hole) 599 request += p->remaining; 600 } 601 if (request > tar->entry_bytes_remaining) 602 request = tar->entry_bytes_remaining; 603 request += tar->entry_padding + tar->entry_bytes_unconsumed; 604 605 bytes_skipped = __archive_read_consume(a, request); 606 if (bytes_skipped < 0) 607 return (ARCHIVE_FATAL); 608 609 tar->entry_bytes_remaining = 0; 610 tar->entry_bytes_unconsumed = 0; 611 tar->entry_padding = 0; 612 613 /* Free the sparse list. */ 614 gnu_clear_sparse_list(tar); 615 616 return (ARCHIVE_OK); 617} 618 619/* 620 * This function recursively interprets all of the headers associated 621 * with a single entry. 622 */ 623static int 624tar_read_header(struct archive_read *a, struct tar *tar, 625 struct archive_entry *entry, size_t *unconsumed) 626{ 627 ssize_t bytes; 628 int err; 629 const char *h; 630 const struct archive_entry_header_ustar *header; 631 const struct archive_entry_header_gnutar *gnuheader; 632 633 tar_flush_unconsumed(a, unconsumed); 634 635 /* Read 512-byte header record */ 636 h = __archive_read_ahead(a, 512, &bytes); 637 if (bytes < 0) 638 return ((int)bytes); 639 if (bytes == 0) { /* EOF at a block boundary. */ 640 /* Some writers do omit the block of nulls. <sigh> */ 641 return (ARCHIVE_EOF); 642 } 643 if (bytes < 512) { /* Short block at EOF; this is bad. */ 644 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 645 "Truncated tar archive"); 646 return (ARCHIVE_FATAL); 647 } 648 *unconsumed = 512; 649 650 /* Check for end-of-archive mark. */ 651 if (h[0] == 0 && archive_block_is_null(h)) { 652 /* Try to consume a second all-null record, as well. */ 653 tar_flush_unconsumed(a, unconsumed); 654 h = __archive_read_ahead(a, 512, NULL); 655 if (h != NULL) 656 __archive_read_consume(a, 512); 657 archive_clear_error(&a->archive); 658 if (a->archive.archive_format_name == NULL) { 659 a->archive.archive_format = ARCHIVE_FORMAT_TAR; 660 a->archive.archive_format_name = "tar"; 661 } 662 return (ARCHIVE_EOF); 663 } 664 665 /* 666 * Note: If the checksum fails and we return ARCHIVE_RETRY, 667 * then the client is likely to just retry. This is a very 668 * crude way to search for the next valid header! 669 * 670 * TODO: Improve this by implementing a real header scan. 671 */ 672 if (!checksum(a, h)) { 673 tar_flush_unconsumed(a, unconsumed); 674 archive_set_error(&a->archive, EINVAL, "Damaged tar archive"); 675 return (ARCHIVE_RETRY); /* Retryable: Invalid header */ 676 } 677 678 if (++tar->header_recursion_depth > 32) { 679 tar_flush_unconsumed(a, unconsumed); 680 archive_set_error(&a->archive, EINVAL, "Too many special headers"); 681 return (ARCHIVE_WARN); 682 } 683 684 /* Determine the format variant. */ 685 header = (const struct archive_entry_header_ustar *)h; 686 687 switch(header->typeflag[0]) { 688 case 'A': /* Solaris tar ACL */ 689 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE; 690 a->archive.archive_format_name = "Solaris tar"; 691 err = header_Solaris_ACL(a, tar, entry, h, unconsumed); 692 break; 693 case 'g': /* POSIX-standard 'g' header. */ 694 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE; 695 a->archive.archive_format_name = "POSIX pax interchange format"; 696 err = header_pax_global(a, tar, entry, h, unconsumed); 697 break; 698 case 'K': /* Long link name (GNU tar, others) */ 699 err = header_longlink(a, tar, entry, h, unconsumed); 700 break; 701 case 'L': /* Long filename (GNU tar, others) */ 702 err = header_longname(a, tar, entry, h, unconsumed); 703 break; 704 case 'V': /* GNU volume header */ 705 err = header_volume(a, tar, entry, h, unconsumed); 706 break; 707 case 'X': /* Used by SUN tar; same as 'x'. */ 708 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE; 709 a->archive.archive_format_name = 710 "POSIX pax interchange format (Sun variant)"; 711 err = header_pax_extensions(a, tar, entry, h, unconsumed); 712 break; 713 case 'x': /* POSIX-standard 'x' header. */ 714 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE; 715 a->archive.archive_format_name = "POSIX pax interchange format"; 716 err = header_pax_extensions(a, tar, entry, h, unconsumed); 717 break; 718 default: 719 gnuheader = (const struct archive_entry_header_gnutar *)h; 720 if (memcmp(gnuheader->magic, "ustar \0", 8) == 0) { 721 a->archive.archive_format = ARCHIVE_FORMAT_TAR_GNUTAR; 722 a->archive.archive_format_name = "GNU tar format"; 723 err = header_gnutar(a, tar, entry, h, unconsumed); 724 } else if (memcmp(header->magic, "ustar", 5) == 0) { 725 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) { 726 a->archive.archive_format = ARCHIVE_FORMAT_TAR_USTAR; 727 a->archive.archive_format_name = "POSIX ustar format"; 728 } 729 err = header_ustar(a, tar, entry, h); 730 } else { 731 a->archive.archive_format = ARCHIVE_FORMAT_TAR; 732 a->archive.archive_format_name = "tar (non-POSIX)"; 733 err = header_old_tar(a, tar, entry, h); 734 } 735 } 736 if (err == ARCHIVE_FATAL) 737 return (err); 738 739 tar_flush_unconsumed(a, unconsumed); 740 741 h = NULL; 742 header = NULL; 743 744 --tar->header_recursion_depth; 745 /* Yuck. Apple's design here ends up storing long pathname 746 * extensions for both the AppleDouble extension entry and the 747 * regular entry. 748 */ 749 /* TODO: Should this be disabled on non-Mac platforms? */ 750 if ((err == ARCHIVE_WARN || err == ARCHIVE_OK) && 751 tar->header_recursion_depth == 0) { 752 int err2 = read_mac_metadata_blob(a, tar, entry, h, unconsumed); 753 if (err2 < err) 754 err = err2; 755 } 756 757 /* We return warnings or success as-is. Anything else is fatal. */ 758 if (err == ARCHIVE_WARN || err == ARCHIVE_OK) { 759 if (tar->sparse_gnu_pending) { 760 if (tar->sparse_gnu_major == 1 && 761 tar->sparse_gnu_minor == 0) { 762 ssize_t bytes_read; 763 764 tar->sparse_gnu_pending = 0; 765 /* Read initial sparse map. */ 766 bytes_read = gnu_sparse_10_read(a, tar, unconsumed); 767 tar->entry_bytes_remaining -= bytes_read; 768 if (bytes_read < 0) 769 return ((int)bytes_read); 770 } else { 771 archive_set_error(&a->archive, 772 ARCHIVE_ERRNO_MISC, 773 "Unrecognized GNU sparse file format"); 774 return (ARCHIVE_WARN); 775 } 776 tar->sparse_gnu_pending = 0; 777 } 778 return (err); 779 } 780 if (err == ARCHIVE_EOF) 781 /* EOF when recursively reading a header is bad. */ 782 archive_set_error(&a->archive, EINVAL, "Damaged tar archive"); 783 return (ARCHIVE_FATAL); 784} 785 786/* 787 * Return true if block checksum is correct. 788 */ 789static int 790checksum(struct archive_read *a, const void *h) 791{ 792 const unsigned char *bytes; 793 const struct archive_entry_header_ustar *header; 794 int check, i, sum; 795 796 (void)a; /* UNUSED */ 797 bytes = (const unsigned char *)h; 798 header = (const struct archive_entry_header_ustar *)h; 799 800 /* 801 * Test the checksum. Note that POSIX specifies _unsigned_ 802 * bytes for this calculation. 803 */ 804 sum = (int)tar_atol(header->checksum, sizeof(header->checksum)); 805 check = 0; 806 for (i = 0; i < 148; i++) 807 check += (unsigned char)bytes[i]; 808 for (; i < 156; i++) 809 check += 32; 810 for (; i < 512; i++) 811 check += (unsigned char)bytes[i]; 812 if (sum == check) 813 return (1); 814 815 /* 816 * Repeat test with _signed_ bytes, just in case this archive 817 * was created by an old BSD, Solaris, or HP-UX tar with a 818 * broken checksum calculation. 819 */ 820 check = 0; 821 for (i = 0; i < 148; i++) 822 check += (signed char)bytes[i]; 823 for (; i < 156; i++) 824 check += 32; 825 for (; i < 512; i++) 826 check += (signed char)bytes[i]; 827 if (sum == check) 828 return (1); 829 830 return (0); 831} 832 833/* 834 * Return true if this block contains only nulls. 835 */ 836static int 837archive_block_is_null(const char *p) 838{ 839 unsigned i; 840 841 for (i = 0; i < 512; i++) 842 if (*p++) 843 return (0); 844 return (1); 845} 846 847/* 848 * Interpret 'A' Solaris ACL header 849 */ 850static int 851header_Solaris_ACL(struct archive_read *a, struct tar *tar, 852 struct archive_entry *entry, const void *h, size_t *unconsumed) 853{ 854 const struct archive_entry_header_ustar *header; 855 size_t size; 856 int err; 857 int64_t type; 858 char *acl, *p; 859 860 /* 861 * read_body_to_string adds a NUL terminator, but we need a little 862 * more to make sure that we don't overrun acl_text later. 863 */ 864 header = (const struct archive_entry_header_ustar *)h; 865 size = (size_t)tar_atol(header->size, sizeof(header->size)); 866 err = read_body_to_string(a, tar, &(tar->acl_text), h, unconsumed); 867 if (err != ARCHIVE_OK) 868 return (err); 869 870 /* Recursively read next header */ 871 err = tar_read_header(a, tar, entry, unconsumed); 872 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN)) 873 return (err); 874 875 /* TODO: Examine the first characters to see if this 876 * is an AIX ACL descriptor. We'll likely never support 877 * them, but it would be polite to recognize and warn when 878 * we do see them. */ 879 880 /* Leading octal number indicates ACL type and number of entries. */ 881 p = acl = tar->acl_text.s; 882 type = 0; 883 while (*p != '\0' && p < acl + size) { 884 if (*p < '0' || *p > '7') { 885 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 886 "Malformed Solaris ACL attribute (invalid digit)"); 887 return(ARCHIVE_WARN); 888 } 889 type <<= 3; 890 type += *p - '0'; 891 if (type > 077777777) { 892 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 893 "Malformed Solaris ACL attribute (count too large)"); 894 return (ARCHIVE_WARN); 895 } 896 p++; 897 } 898 switch ((int)type & ~0777777) { 899 case 01000000: 900 /* POSIX.1e ACL */ 901 break; 902 case 03000000: 903 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 904 "Solaris NFSv4 ACLs not supported"); 905 return (ARCHIVE_WARN); 906 default: 907 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 908 "Malformed Solaris ACL attribute (unsupported type %o)", 909 (int)type); 910 return (ARCHIVE_WARN); 911 } 912 p++; 913 914 if (p >= acl + size) { 915 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 916 "Malformed Solaris ACL attribute (body overflow)"); 917 return(ARCHIVE_WARN); 918 } 919 920 /* ACL text is null-terminated; find the end. */ 921 size -= (p - acl); 922 acl = p; 923 924 while (*p != '\0' && p < acl + size) 925 p++; 926 927 if (tar->sconv_acl == NULL) { 928 tar->sconv_acl = archive_string_conversion_from_charset( 929 &(a->archive), "UTF-8", 1); 930 if (tar->sconv_acl == NULL) 931 return (ARCHIVE_FATAL); 932 } 933 archive_strncpy(&(tar->localname), acl, p - acl); 934 err = archive_acl_parse_l(archive_entry_acl(entry), 935 tar->localname.s, ARCHIVE_ENTRY_ACL_TYPE_ACCESS, tar->sconv_acl); 936 if (err != ARCHIVE_OK) { 937 if (errno == ENOMEM) { 938 archive_set_error(&a->archive, ENOMEM, 939 "Can't allocate memory for ACL"); 940 } else 941 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 942 "Malformed Solaris ACL attribute (unparsable)"); 943 } 944 return (err); 945} 946 947/* 948 * Interpret 'K' long linkname header. 949 */ 950static int 951header_longlink(struct archive_read *a, struct tar *tar, 952 struct archive_entry *entry, const void *h, size_t *unconsumed) 953{ 954 int err; 955 956 err = read_body_to_string(a, tar, &(tar->longlink), h, unconsumed); 957 if (err != ARCHIVE_OK) 958 return (err); 959 err = tar_read_header(a, tar, entry, unconsumed); 960 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN)) 961 return (err); 962 /* Set symlink if symlink already set, else hardlink. */ 963 archive_entry_copy_link(entry, tar->longlink.s); 964 return (ARCHIVE_OK); 965} 966 967static int 968set_conversion_failed_error(struct archive_read *a, 969 struct archive_string_conv *sconv, const char *name) 970{ 971 if (errno == ENOMEM) { 972 archive_set_error(&a->archive, ENOMEM, 973 "Can't allocate memory for %s", name); 974 return (ARCHIVE_FATAL); 975 } 976 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 977 "%s can't be converted from %s to current locale.", 978 name, archive_string_conversion_charset_name(sconv)); 979 return (ARCHIVE_WARN); 980} 981 982/* 983 * Interpret 'L' long filename header. 984 */ 985static int 986header_longname(struct archive_read *a, struct tar *tar, 987 struct archive_entry *entry, const void *h, size_t *unconsumed) 988{ 989 int err; 990 991 err = read_body_to_string(a, tar, &(tar->longname), h, unconsumed); 992 if (err != ARCHIVE_OK) 993 return (err); 994 /* Read and parse "real" header, then override name. */ 995 err = tar_read_header(a, tar, entry, unconsumed); 996 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN)) 997 return (err); 998 if (archive_entry_copy_pathname_l(entry, tar->longname.s, 999 archive_strlen(&(tar->longname)), tar->sconv) != 0) 1000 err = set_conversion_failed_error(a, tar->sconv, "Pathname"); 1001 return (err); 1002} 1003 1004 1005/* 1006 * Interpret 'V' GNU tar volume header. 1007 */ 1008static int 1009header_volume(struct archive_read *a, struct tar *tar, 1010 struct archive_entry *entry, const void *h, size_t *unconsumed) 1011{ 1012 (void)h; 1013 1014 /* Just skip this and read the next header. */ 1015 return (tar_read_header(a, tar, entry, unconsumed)); 1016} 1017 1018/* 1019 * Read body of an archive entry into an archive_string object. 1020 */ 1021static int 1022read_body_to_string(struct archive_read *a, struct tar *tar, 1023 struct archive_string *as, const void *h, size_t *unconsumed) 1024{ 1025 int64_t size; 1026 const struct archive_entry_header_ustar *header; 1027 const void *src; 1028 1029 (void)tar; /* UNUSED */ 1030 header = (const struct archive_entry_header_ustar *)h; 1031 size = tar_atol(header->size, sizeof(header->size)); 1032 if ((size > 1048576) || (size < 0)) { 1033 archive_set_error(&a->archive, EINVAL, 1034 "Special header too large"); 1035 return (ARCHIVE_FATAL); 1036 } 1037 1038 /* Fail if we can't make our buffer big enough. */ 1039 if (archive_string_ensure(as, (size_t)size+1) == NULL) { 1040 archive_set_error(&a->archive, ENOMEM, 1041 "No memory"); 1042 return (ARCHIVE_FATAL); 1043 } 1044 1045 tar_flush_unconsumed(a, unconsumed); 1046 1047 /* Read the body into the string. */ 1048 *unconsumed = (size_t)((size + 511) & ~ 511); 1049 src = __archive_read_ahead(a, *unconsumed, NULL); 1050 if (src == NULL) { 1051 *unconsumed = 0; 1052 return (ARCHIVE_FATAL); 1053 } 1054 memcpy(as->s, src, (size_t)size); 1055 as->s[size] = '\0'; 1056 as->length = (size_t)size; 1057 return (ARCHIVE_OK); 1058} 1059 1060/* 1061 * Parse out common header elements. 1062 * 1063 * This would be the same as header_old_tar, except that the 1064 * filename is handled slightly differently for old and POSIX 1065 * entries (POSIX entries support a 'prefix'). This factoring 1066 * allows header_old_tar and header_ustar 1067 * to handle filenames differently, while still putting most of the 1068 * common parsing into one place. 1069 */ 1070static int 1071header_common(struct archive_read *a, struct tar *tar, 1072 struct archive_entry *entry, const void *h) 1073{ 1074 const struct archive_entry_header_ustar *header; 1075 char tartype; 1076 int err = ARCHIVE_OK; 1077 1078 header = (const struct archive_entry_header_ustar *)h; 1079 if (header->linkname[0]) 1080 archive_strncpy(&(tar->entry_linkpath), 1081 header->linkname, sizeof(header->linkname)); 1082 else 1083 archive_string_empty(&(tar->entry_linkpath)); 1084 1085 /* Parse out the numeric fields (all are octal) */ 1086 archive_entry_set_mode(entry, 1087 (mode_t)tar_atol(header->mode, sizeof(header->mode))); 1088 archive_entry_set_uid(entry, tar_atol(header->uid, sizeof(header->uid))); 1089 archive_entry_set_gid(entry, tar_atol(header->gid, sizeof(header->gid))); 1090 tar->entry_bytes_remaining = tar_atol(header->size, sizeof(header->size)); 1091 if (tar->entry_bytes_remaining < 0) { 1092 tar->entry_bytes_remaining = 0; 1093 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1094 "Tar entry has negative size?"); 1095 err = ARCHIVE_WARN; 1096 } 1097 tar->realsize = tar->entry_bytes_remaining; 1098 archive_entry_set_size(entry, tar->entry_bytes_remaining); 1099 archive_entry_set_mtime(entry, tar_atol(header->mtime, sizeof(header->mtime)), 0); 1100 1101 /* Handle the tar type flag appropriately. */ 1102 tartype = header->typeflag[0]; 1103 1104 switch (tartype) { 1105 case '1': /* Hard link */ 1106 if (archive_entry_copy_hardlink_l(entry, tar->entry_linkpath.s, 1107 archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) { 1108 err = set_conversion_failed_error(a, tar->sconv, 1109 "Linkname"); 1110 if (err == ARCHIVE_FATAL) 1111 return (err); 1112 } 1113 /* 1114 * The following may seem odd, but: Technically, tar 1115 * does not store the file type for a "hard link" 1116 * entry, only the fact that it is a hard link. So, I 1117 * leave the type zero normally. But, pax interchange 1118 * format allows hard links to have data, which 1119 * implies that the underlying entry is a regular 1120 * file. 1121 */ 1122 if (archive_entry_size(entry) > 0) 1123 archive_entry_set_filetype(entry, AE_IFREG); 1124 1125 /* 1126 * A tricky point: Traditionally, tar readers have 1127 * ignored the size field when reading hardlink 1128 * entries, and some writers put non-zero sizes even 1129 * though the body is empty. POSIX blessed this 1130 * convention in the 1988 standard, but broke with 1131 * this tradition in 2001 by permitting hardlink 1132 * entries to store valid bodies in pax interchange 1133 * format, but not in ustar format. Since there is no 1134 * hard and fast way to distinguish pax interchange 1135 * from earlier archives (the 'x' and 'g' entries are 1136 * optional, after all), we need a heuristic. 1137 */ 1138 if (archive_entry_size(entry) == 0) { 1139 /* If the size is already zero, we're done. */ 1140 } else if (a->archive.archive_format 1141 == ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) { 1142 /* Definitely pax extended; must obey hardlink size. */ 1143 } else if (a->archive.archive_format == ARCHIVE_FORMAT_TAR 1144 || a->archive.archive_format == ARCHIVE_FORMAT_TAR_GNUTAR) 1145 { 1146 /* Old-style or GNU tar: we must ignore the size. */ 1147 archive_entry_set_size(entry, 0); 1148 tar->entry_bytes_remaining = 0; 1149 } else if (archive_read_format_tar_bid(a, 50) > 50) { 1150 /* 1151 * We don't know if it's pax: If the bid 1152 * function sees a valid ustar header 1153 * immediately following, then let's ignore 1154 * the hardlink size. 1155 */ 1156 archive_entry_set_size(entry, 0); 1157 tar->entry_bytes_remaining = 0; 1158 } 1159 /* 1160 * TODO: There are still two cases I'd like to handle: 1161 * = a ustar non-pax archive with a hardlink entry at 1162 * end-of-archive. (Look for block of nulls following?) 1163 * = a pax archive that has not seen any pax headers 1164 * and has an entry which is a hardlink entry storing 1165 * a body containing an uncompressed tar archive. 1166 * The first is worth addressing; I don't see any reliable 1167 * way to deal with the second possibility. 1168 */ 1169 break; 1170 case '2': /* Symlink */ 1171 archive_entry_set_filetype(entry, AE_IFLNK); 1172 archive_entry_set_size(entry, 0); 1173 tar->entry_bytes_remaining = 0; 1174 if (archive_entry_copy_symlink_l(entry, tar->entry_linkpath.s, 1175 archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) { 1176 err = set_conversion_failed_error(a, tar->sconv, 1177 "Linkname"); 1178 if (err == ARCHIVE_FATAL) 1179 return (err); 1180 } 1181 break; 1182 case '3': /* Character device */ 1183 archive_entry_set_filetype(entry, AE_IFCHR); 1184 archive_entry_set_size(entry, 0); 1185 tar->entry_bytes_remaining = 0; 1186 break; 1187 case '4': /* Block device */ 1188 archive_entry_set_filetype(entry, AE_IFBLK); 1189 archive_entry_set_size(entry, 0); 1190 tar->entry_bytes_remaining = 0; 1191 break; 1192 case '5': /* Dir */ 1193 archive_entry_set_filetype(entry, AE_IFDIR); 1194 archive_entry_set_size(entry, 0); 1195 tar->entry_bytes_remaining = 0; 1196 break; 1197 case '6': /* FIFO device */ 1198 archive_entry_set_filetype(entry, AE_IFIFO); 1199 archive_entry_set_size(entry, 0); 1200 tar->entry_bytes_remaining = 0; 1201 break; 1202 case 'D': /* GNU incremental directory type */ 1203 /* 1204 * No special handling is actually required here. 1205 * It might be nice someday to preprocess the file list and 1206 * provide it to the client, though. 1207 */ 1208 archive_entry_set_filetype(entry, AE_IFDIR); 1209 break; 1210 case 'M': /* GNU "Multi-volume" (remainder of file from last archive)*/ 1211 /* 1212 * As far as I can tell, this is just like a regular file 1213 * entry, except that the contents should be _appended_ to 1214 * the indicated file at the indicated offset. This may 1215 * require some API work to fully support. 1216 */ 1217 break; 1218 case 'N': /* Old GNU "long filename" entry. */ 1219 /* The body of this entry is a script for renaming 1220 * previously-extracted entries. Ugh. It will never 1221 * be supported by libarchive. */ 1222 archive_entry_set_filetype(entry, AE_IFREG); 1223 break; 1224 case 'S': /* GNU sparse files */ 1225 /* 1226 * Sparse files are really just regular files with 1227 * sparse information in the extended area. 1228 */ 1229 /* FALLTHROUGH */ 1230 case '0': 1231 /* 1232 * Enable sparse file "read" support only for regular 1233 * files and explicit GNU sparse files. However, we 1234 * don't allow non-standard file types to be sparse. 1235 */ 1236 tar->sparse_allowed = 1; 1237 /* FALLTHROUGH */ 1238 default: /* Regular file and non-standard types */ 1239 /* 1240 * Per POSIX: non-recognized types should always be 1241 * treated as regular files. 1242 */ 1243 archive_entry_set_filetype(entry, AE_IFREG); 1244 break; 1245 } 1246 return (err); 1247} 1248 1249/* 1250 * Parse out header elements for "old-style" tar archives. 1251 */ 1252static int 1253header_old_tar(struct archive_read *a, struct tar *tar, 1254 struct archive_entry *entry, const void *h) 1255{ 1256 const struct archive_entry_header_ustar *header; 1257 int err = ARCHIVE_OK, err2; 1258 1259 /* Copy filename over (to ensure null termination). */ 1260 header = (const struct archive_entry_header_ustar *)h; 1261 if (archive_entry_copy_pathname_l(entry, 1262 header->name, sizeof(header->name), tar->sconv) != 0) { 1263 err = set_conversion_failed_error(a, tar->sconv, "Pathname"); 1264 if (err == ARCHIVE_FATAL) 1265 return (err); 1266 } 1267 1268 /* Grab rest of common fields */ 1269 err2 = header_common(a, tar, entry, h); 1270 if (err > err2) 1271 err = err2; 1272 1273 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining); 1274 return (err); 1275} 1276 1277/* 1278 * Read a Mac AppleDouble-encoded blob of file metadata, 1279 * if there is one. 1280 */ 1281static int 1282read_mac_metadata_blob(struct archive_read *a, struct tar *tar, 1283 struct archive_entry *entry, const void *h, size_t *unconsumed) 1284{ 1285 int64_t size; 1286 const void *data; 1287 const char *p, *name; 1288 const wchar_t *wp, *wname; 1289 1290 (void)h; /* UNUSED */ 1291 1292 wname = wp = archive_entry_pathname_w(entry); 1293 if (wp != NULL) { 1294 /* Find the last path element. */ 1295 for (; *wp != L'\0'; ++wp) { 1296 if (wp[0] == '/' && wp[1] != L'\0') 1297 wname = wp + 1; 1298 } 1299 /* 1300 * If last path element starts with "._", then 1301 * this is a Mac extension. 1302 */ 1303 if (wname[0] != L'.' || wname[1] != L'_' || wname[2] == L'\0') 1304 return ARCHIVE_OK; 1305 } else { 1306 /* Find the last path element. */ 1307 name = p = archive_entry_pathname(entry); 1308 if (p == NULL) 1309 return (ARCHIVE_FAILED); 1310 for (; *p != '\0'; ++p) { 1311 if (p[0] == '/' && p[1] != '\0') 1312 name = p + 1; 1313 } 1314 /* 1315 * If last path element starts with "._", then 1316 * this is a Mac extension. 1317 */ 1318 if (name[0] != '.' || name[1] != '_' || name[2] == '\0') 1319 return ARCHIVE_OK; 1320 } 1321 1322 /* Read the body as a Mac OS metadata blob. */ 1323 size = archive_entry_size(entry); 1324 1325 /* 1326 * TODO: Look beyond the body here to peek at the next header. 1327 * If it's a regular header (not an extension header) 1328 * that has the wrong name, just return the current 1329 * entry as-is, without consuming the body here. 1330 * That would reduce the risk of us mis-identifying 1331 * an ordinary file that just happened to have 1332 * a name starting with "._". 1333 * 1334 * Q: Is the above idea really possible? Even 1335 * when there are GNU or pax extension entries? 1336 */ 1337 data = __archive_read_ahead(a, (size_t)size, NULL); 1338 if (data == NULL) { 1339 *unconsumed = 0; 1340 return (ARCHIVE_FATAL); 1341 } 1342 archive_entry_copy_mac_metadata(entry, data, (size_t)size); 1343 *unconsumed = (size_t)((size + 511) & ~ 511); 1344 tar_flush_unconsumed(a, unconsumed); 1345 return (tar_read_header(a, tar, entry, unconsumed)); 1346} 1347 1348/* 1349 * Parse a file header for a pax extended archive entry. 1350 */ 1351static int 1352header_pax_global(struct archive_read *a, struct tar *tar, 1353 struct archive_entry *entry, const void *h, size_t *unconsumed) 1354{ 1355 int err; 1356 1357 err = read_body_to_string(a, tar, &(tar->pax_global), h, unconsumed); 1358 if (err != ARCHIVE_OK) 1359 return (err); 1360 err = tar_read_header(a, tar, entry, unconsumed); 1361 return (err); 1362} 1363 1364static int 1365header_pax_extensions(struct archive_read *a, struct tar *tar, 1366 struct archive_entry *entry, const void *h, size_t *unconsumed) 1367{ 1368 int err, err2; 1369 1370 err = read_body_to_string(a, tar, &(tar->pax_header), h, unconsumed); 1371 if (err != ARCHIVE_OK) 1372 return (err); 1373 1374 /* Parse the next header. */ 1375 err = tar_read_header(a, tar, entry, unconsumed); 1376 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN)) 1377 return (err); 1378 1379 /* 1380 * TODO: Parse global/default options into 'entry' struct here 1381 * before handling file-specific options. 1382 * 1383 * This design (parse standard header, then overwrite with pax 1384 * extended attribute data) usually works well, but isn't ideal; 1385 * it would be better to parse the pax extended attributes first 1386 * and then skip any fields in the standard header that were 1387 * defined in the pax header. 1388 */ 1389 err2 = pax_header(a, tar, entry, tar->pax_header.s); 1390 err = err_combine(err, err2); 1391 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining); 1392 return (err); 1393} 1394 1395 1396/* 1397 * Parse a file header for a Posix "ustar" archive entry. This also 1398 * handles "pax" or "extended ustar" entries. 1399 */ 1400static int 1401header_ustar(struct archive_read *a, struct tar *tar, 1402 struct archive_entry *entry, const void *h) 1403{ 1404 const struct archive_entry_header_ustar *header; 1405 struct archive_string *as; 1406 int err = ARCHIVE_OK, r; 1407 1408 header = (const struct archive_entry_header_ustar *)h; 1409 1410 /* Copy name into an internal buffer to ensure null-termination. */ 1411 as = &(tar->entry_pathname); 1412 if (header->prefix[0]) { 1413 archive_strncpy(as, header->prefix, sizeof(header->prefix)); 1414 if (as->s[archive_strlen(as) - 1] != '/') 1415 archive_strappend_char(as, '/'); 1416 archive_strncat(as, header->name, sizeof(header->name)); 1417 } else { 1418 archive_strncpy(as, header->name, sizeof(header->name)); 1419 } 1420 if (archive_entry_copy_pathname_l(entry, as->s, archive_strlen(as), 1421 tar->sconv) != 0) { 1422 err = set_conversion_failed_error(a, tar->sconv, "Pathname"); 1423 if (err == ARCHIVE_FATAL) 1424 return (err); 1425 } 1426 1427 /* Handle rest of common fields. */ 1428 r = header_common(a, tar, entry, h); 1429 if (r == ARCHIVE_FATAL) 1430 return (r); 1431 if (r < err) 1432 err = r; 1433 1434 /* Handle POSIX ustar fields. */ 1435 if (archive_entry_copy_uname_l(entry, 1436 header->uname, sizeof(header->uname), tar->sconv) != 0) { 1437 err = set_conversion_failed_error(a, tar->sconv, "Uname"); 1438 if (err == ARCHIVE_FATAL) 1439 return (err); 1440 } 1441 1442 if (archive_entry_copy_gname_l(entry, 1443 header->gname, sizeof(header->gname), tar->sconv) != 0) { 1444 err = set_conversion_failed_error(a, tar->sconv, "Gname"); 1445 if (err == ARCHIVE_FATAL) 1446 return (err); 1447 } 1448 1449 /* Parse out device numbers only for char and block specials. */ 1450 if (header->typeflag[0] == '3' || header->typeflag[0] == '4') { 1451 archive_entry_set_rdevmajor(entry, (dev_t) 1452 tar_atol(header->rdevmajor, sizeof(header->rdevmajor))); 1453 archive_entry_set_rdevminor(entry, (dev_t) 1454 tar_atol(header->rdevminor, sizeof(header->rdevminor))); 1455 } 1456 1457 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining); 1458 1459 return (err); 1460} 1461 1462 1463/* 1464 * Parse the pax extended attributes record. 1465 * 1466 * Returns non-zero if there's an error in the data. 1467 */ 1468static int 1469pax_header(struct archive_read *a, struct tar *tar, 1470 struct archive_entry *entry, char *attr) 1471{ 1472 size_t attr_length, l, line_length; 1473 char *p; 1474 char *key, *value; 1475 struct archive_string *as; 1476 struct archive_string_conv *sconv; 1477 int err, err2; 1478 1479 attr_length = strlen(attr); 1480 tar->pax_hdrcharset_binary = 0; 1481 archive_string_empty(&(tar->entry_gname)); 1482 archive_string_empty(&(tar->entry_linkpath)); 1483 archive_string_empty(&(tar->entry_pathname)); 1484 archive_string_empty(&(tar->entry_pathname_override)); 1485 archive_string_empty(&(tar->entry_uname)); 1486 err = ARCHIVE_OK; 1487 while (attr_length > 0) { 1488 /* Parse decimal length field at start of line. */ 1489 line_length = 0; 1490 l = attr_length; 1491 p = attr; /* Record start of line. */ 1492 while (l>0) { 1493 if (*p == ' ') { 1494 p++; 1495 l--; 1496 break; 1497 } 1498 if (*p < '0' || *p > '9') { 1499 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1500 "Ignoring malformed pax extended attributes"); 1501 return (ARCHIVE_WARN); 1502 } 1503 line_length *= 10; 1504 line_length += *p - '0'; 1505 if (line_length > 999999) { 1506 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1507 "Rejecting pax extended attribute > 1MB"); 1508 return (ARCHIVE_WARN); 1509 } 1510 p++; 1511 l--; 1512 } 1513 1514 /* 1515 * Parsed length must be no bigger than available data, 1516 * at least 1, and the last character of the line must 1517 * be '\n'. 1518 */ 1519 if (line_length > attr_length 1520 || line_length < 1 1521 || attr[line_length - 1] != '\n') 1522 { 1523 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1524 "Ignoring malformed pax extended attribute"); 1525 return (ARCHIVE_WARN); 1526 } 1527 1528 /* Null-terminate the line. */ 1529 attr[line_length - 1] = '\0'; 1530 1531 /* Find end of key and null terminate it. */ 1532 key = p; 1533 if (key[0] == '=') 1534 return (-1); 1535 while (*p && *p != '=') 1536 ++p; 1537 if (*p == '\0') { 1538 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1539 "Invalid pax extended attributes"); 1540 return (ARCHIVE_WARN); 1541 } 1542 *p = '\0'; 1543 1544 /* Identify null-terminated 'value' portion. */ 1545 value = p + 1; 1546 1547 /* Identify this attribute and set it in the entry. */ 1548 err2 = pax_attribute(a, tar, entry, key, value); 1549 if (err2 == ARCHIVE_FATAL) 1550 return (err2); 1551 err = err_combine(err, err2); 1552 1553 /* Skip to next line */ 1554 attr += line_length; 1555 attr_length -= line_length; 1556 } 1557 1558 /* 1559 * PAX format uses UTF-8 as default charset for its metadata 1560 * unless hdrcharset=BINARY is present in its header. 1561 * We apply the charset specified by the hdrcharset option only 1562 * when the hdrcharset attribute(in PAX header) is BINARY because 1563 * we respect the charset described in PAX header and BINARY also 1564 * means that metadata(filename,uname and gname) character-set 1565 * is unknown. 1566 */ 1567 if (tar->pax_hdrcharset_binary) 1568 sconv = tar->opt_sconv; 1569 else { 1570 sconv = archive_string_conversion_from_charset( 1571 &(a->archive), "UTF-8", 1); 1572 if (sconv == NULL) 1573 return (ARCHIVE_FATAL); 1574 if (tar->compat_2x) 1575 archive_string_conversion_set_opt(sconv, 1576 SCONV_SET_OPT_UTF8_LIBARCHIVE2X); 1577 } 1578 1579 if (archive_strlen(&(tar->entry_gname)) > 0) { 1580 if (archive_entry_copy_gname_l(entry, tar->entry_gname.s, 1581 archive_strlen(&(tar->entry_gname)), sconv) != 0) { 1582 err = set_conversion_failed_error(a, sconv, "Gname"); 1583 if (err == ARCHIVE_FATAL) 1584 return (err); 1585 /* Use a converted an original name. */ 1586 archive_entry_copy_gname(entry, tar->entry_gname.s); 1587 } 1588 } 1589 if (archive_strlen(&(tar->entry_linkpath)) > 0) { 1590 if (archive_entry_copy_link_l(entry, tar->entry_linkpath.s, 1591 archive_strlen(&(tar->entry_linkpath)), sconv) != 0) { 1592 err = set_conversion_failed_error(a, sconv, "Linkname"); 1593 if (err == ARCHIVE_FATAL) 1594 return (err); 1595 /* Use a converted an original name. */ 1596 archive_entry_copy_link(entry, tar->entry_linkpath.s); 1597 } 1598 } 1599 /* 1600 * Some extensions (such as the GNU sparse file extensions) 1601 * deliberately store a synthetic name under the regular 'path' 1602 * attribute and the real file name under a different attribute. 1603 * Since we're supposed to not care about the order, we 1604 * have no choice but to store all of the various filenames 1605 * we find and figure it all out afterwards. This is the 1606 * figuring out part. 1607 */ 1608 as = NULL; 1609 if (archive_strlen(&(tar->entry_pathname_override)) > 0) 1610 as = &(tar->entry_pathname_override); 1611 else if (archive_strlen(&(tar->entry_pathname)) > 0) 1612 as = &(tar->entry_pathname); 1613 if (as != NULL) { 1614 if (archive_entry_copy_pathname_l(entry, as->s, 1615 archive_strlen(as), sconv) != 0) { 1616 err = set_conversion_failed_error(a, sconv, "Pathname"); 1617 if (err == ARCHIVE_FATAL) 1618 return (err); 1619 /* Use a converted an original name. */ 1620 archive_entry_copy_pathname(entry, as->s); 1621 } 1622 } 1623 if (archive_strlen(&(tar->entry_uname)) > 0) { 1624 if (archive_entry_copy_uname_l(entry, tar->entry_uname.s, 1625 archive_strlen(&(tar->entry_uname)), sconv) != 0) { 1626 err = set_conversion_failed_error(a, sconv, "Uname"); 1627 if (err == ARCHIVE_FATAL) 1628 return (err); 1629 /* Use a converted an original name. */ 1630 archive_entry_copy_uname(entry, tar->entry_uname.s); 1631 } 1632 } 1633 return (err); 1634} 1635 1636static int 1637pax_attribute_xattr(struct archive_entry *entry, 1638 char *name, char *value) 1639{ 1640 char *name_decoded; 1641 void *value_decoded; 1642 size_t value_len; 1643 1644 if (strlen(name) < 18 || (memcmp(name, "LIBARCHIVE.xattr.", 17)) != 0) 1645 return 3; 1646 1647 name += 17; 1648 1649 /* URL-decode name */ 1650 name_decoded = url_decode(name); 1651 if (name_decoded == NULL) 1652 return 2; 1653 1654 /* Base-64 decode value */ 1655 value_decoded = base64_decode(value, strlen(value), &value_len); 1656 if (value_decoded == NULL) { 1657 free(name_decoded); 1658 return 1; 1659 } 1660 1661 archive_entry_xattr_add_entry(entry, name_decoded, 1662 value_decoded, value_len); 1663 1664 free(name_decoded); 1665 free(value_decoded); 1666 return 0; 1667} 1668 1669/* 1670 * Parse a single key=value attribute. key/value pointers are 1671 * assumed to point into reasonably long-lived storage. 1672 * 1673 * Note that POSIX reserves all-lowercase keywords. Vendor-specific 1674 * extensions should always have keywords of the form "VENDOR.attribute" 1675 * In particular, it's quite feasible to support many different 1676 * vendor extensions here. I'm using "LIBARCHIVE" for extensions 1677 * unique to this library. 1678 * 1679 * Investigate other vendor-specific extensions and see if 1680 * any of them look useful. 1681 */ 1682static int 1683pax_attribute(struct archive_read *a, struct tar *tar, 1684 struct archive_entry *entry, char *key, char *value) 1685{ 1686 int64_t s; 1687 long n; 1688 int err = ARCHIVE_OK, r; 1689 1690#ifndef __FreeBSD__ 1691 if (value == NULL) 1692 value = ""; /* Disable compiler warning; do not pass 1693 * NULL pointer to strlen(). */ 1694#endif 1695 switch (key[0]) { 1696 case 'G': 1697 /* Reject GNU.sparse.* headers on non-regular files. */ 1698 if (strncmp(key, "GNU.sparse", 10) == 0 && 1699 !tar->sparse_allowed) { 1700 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1701 "Non-regular file cannot be sparse"); 1702 return (ARCHIVE_FATAL); 1703 } 1704 1705 /* GNU "0.0" sparse pax format. */ 1706 if (strcmp(key, "GNU.sparse.numblocks") == 0) { 1707 tar->sparse_offset = -1; 1708 tar->sparse_numbytes = -1; 1709 tar->sparse_gnu_major = 0; 1710 tar->sparse_gnu_minor = 0; 1711 } 1712 if (strcmp(key, "GNU.sparse.offset") == 0) { 1713 tar->sparse_offset = tar_atol10(value, strlen(value)); 1714 if (tar->sparse_numbytes != -1) { 1715 if (gnu_add_sparse_entry(a, tar, 1716 tar->sparse_offset, tar->sparse_numbytes) 1717 != ARCHIVE_OK) 1718 return (ARCHIVE_FATAL); 1719 tar->sparse_offset = -1; 1720 tar->sparse_numbytes = -1; 1721 } 1722 } 1723 if (strcmp(key, "GNU.sparse.numbytes") == 0) { 1724 tar->sparse_numbytes = tar_atol10(value, strlen(value)); 1725 if (tar->sparse_numbytes != -1) { 1726 if (gnu_add_sparse_entry(a, tar, 1727 tar->sparse_offset, tar->sparse_numbytes) 1728 != ARCHIVE_OK) 1729 return (ARCHIVE_FATAL); 1730 tar->sparse_offset = -1; 1731 tar->sparse_numbytes = -1; 1732 } 1733 } 1734 if (strcmp(key, "GNU.sparse.size") == 0) { 1735 tar->realsize = tar_atol10(value, strlen(value)); 1736 archive_entry_set_size(entry, tar->realsize); 1737 } 1738 1739 /* GNU "0.1" sparse pax format. */ 1740 if (strcmp(key, "GNU.sparse.map") == 0) { 1741 tar->sparse_gnu_major = 0; 1742 tar->sparse_gnu_minor = 1; 1743 if (gnu_sparse_01_parse(a, tar, value) != ARCHIVE_OK) 1744 return (ARCHIVE_WARN); 1745 } 1746 1747 /* GNU "1.0" sparse pax format */ 1748 if (strcmp(key, "GNU.sparse.major") == 0) { 1749 tar->sparse_gnu_major = (int)tar_atol10(value, strlen(value)); 1750 tar->sparse_gnu_pending = 1; 1751 } 1752 if (strcmp(key, "GNU.sparse.minor") == 0) { 1753 tar->sparse_gnu_minor = (int)tar_atol10(value, strlen(value)); 1754 tar->sparse_gnu_pending = 1; 1755 } 1756 if (strcmp(key, "GNU.sparse.name") == 0) { 1757 /* 1758 * The real filename; when storing sparse 1759 * files, GNU tar puts a synthesized name into 1760 * the regular 'path' attribute in an attempt 1761 * to limit confusion. ;-) 1762 */ 1763 archive_strcpy(&(tar->entry_pathname_override), value); 1764 } 1765 if (strcmp(key, "GNU.sparse.realsize") == 0) { 1766 tar->realsize = tar_atol10(value, strlen(value)); 1767 archive_entry_set_size(entry, tar->realsize); 1768 } 1769 break; 1770 case 'L': 1771 /* Our extensions */ 1772/* TODO: Handle arbitrary extended attributes... */ 1773/* 1774 if (strcmp(key, "LIBARCHIVE.xxxxxxx") == 0) 1775 archive_entry_set_xxxxxx(entry, value); 1776*/ 1777 if (strcmp(key, "LIBARCHIVE.creationtime") == 0) { 1778 pax_time(value, &s, &n); 1779 archive_entry_set_birthtime(entry, s, n); 1780 } 1781 if (memcmp(key, "LIBARCHIVE.xattr.", 17) == 0) 1782 pax_attribute_xattr(entry, key, value); 1783 break; 1784 case 'S': 1785 /* We support some keys used by the "star" archiver */ 1786 if (strcmp(key, "SCHILY.acl.access") == 0) { 1787 if (tar->sconv_acl == NULL) { 1788 tar->sconv_acl = 1789 archive_string_conversion_from_charset( 1790 &(a->archive), "UTF-8", 1); 1791 if (tar->sconv_acl == NULL) 1792 return (ARCHIVE_FATAL); 1793 } 1794 1795 r = archive_acl_parse_l(archive_entry_acl(entry), 1796 value, ARCHIVE_ENTRY_ACL_TYPE_ACCESS, 1797 tar->sconv_acl); 1798 if (r != ARCHIVE_OK) { 1799 err = r; 1800 if (err == ARCHIVE_FATAL) { 1801 archive_set_error(&a->archive, ENOMEM, 1802 "Can't allocate memory for " 1803 "SCHILY.acl.access"); 1804 return (err); 1805 } 1806 archive_set_error(&a->archive, 1807 ARCHIVE_ERRNO_MISC, 1808 "Parse error: SCHILY.acl.access"); 1809 } 1810 } else if (strcmp(key, "SCHILY.acl.default") == 0) { 1811 if (tar->sconv_acl == NULL) { 1812 tar->sconv_acl = 1813 archive_string_conversion_from_charset( 1814 &(a->archive), "UTF-8", 1); 1815 if (tar->sconv_acl == NULL) 1816 return (ARCHIVE_FATAL); 1817 } 1818 1819 r = archive_acl_parse_l(archive_entry_acl(entry), 1820 value, ARCHIVE_ENTRY_ACL_TYPE_DEFAULT, 1821 tar->sconv_acl); 1822 if (r != ARCHIVE_OK) { 1823 err = r; 1824 if (err == ARCHIVE_FATAL) { 1825 archive_set_error(&a->archive, ENOMEM, 1826 "Can't allocate memory for " 1827 "SCHILY.acl.default"); 1828 return (err); 1829 } 1830 archive_set_error(&a->archive, 1831 ARCHIVE_ERRNO_MISC, 1832 "Parse error: SCHILY.acl.default"); 1833 } 1834 } else if (strcmp(key, "SCHILY.devmajor") == 0) { 1835 archive_entry_set_rdevmajor(entry, 1836 (dev_t)tar_atol10(value, strlen(value))); 1837 } else if (strcmp(key, "SCHILY.devminor") == 0) { 1838 archive_entry_set_rdevminor(entry, 1839 (dev_t)tar_atol10(value, strlen(value))); 1840 } else if (strcmp(key, "SCHILY.fflags") == 0) { 1841 archive_entry_copy_fflags_text(entry, value); 1842 } else if (strcmp(key, "SCHILY.dev") == 0) { 1843 archive_entry_set_dev(entry, 1844 (dev_t)tar_atol10(value, strlen(value))); 1845 } else if (strcmp(key, "SCHILY.ino") == 0) { 1846 archive_entry_set_ino(entry, 1847 tar_atol10(value, strlen(value))); 1848 } else if (strcmp(key, "SCHILY.nlink") == 0) { 1849 archive_entry_set_nlink(entry, (unsigned) 1850 tar_atol10(value, strlen(value))); 1851 } else if (strcmp(key, "SCHILY.realsize") == 0) { 1852 tar->realsize = tar_atol10(value, strlen(value)); 1853 archive_entry_set_size(entry, tar->realsize); 1854 } else if (strcmp(key, "SUN.holesdata") == 0) { 1855 /* A Solaris extension for sparse. */ 1856 r = solaris_sparse_parse(a, tar, entry, value); 1857 if (r < err) { 1858 if (r == ARCHIVE_FATAL) 1859 return (r); 1860 err = r; 1861 archive_set_error(&a->archive, 1862 ARCHIVE_ERRNO_MISC, 1863 "Parse error: SUN.holesdata"); 1864 } 1865 } 1866 break; 1867 case 'a': 1868 if (strcmp(key, "atime") == 0) { 1869 pax_time(value, &s, &n); 1870 archive_entry_set_atime(entry, s, n); 1871 } 1872 break; 1873 case 'c': 1874 if (strcmp(key, "ctime") == 0) { 1875 pax_time(value, &s, &n); 1876 archive_entry_set_ctime(entry, s, n); 1877 } else if (strcmp(key, "charset") == 0) { 1878 /* TODO: Publish charset information in entry. */ 1879 } else if (strcmp(key, "comment") == 0) { 1880 /* TODO: Publish comment in entry. */ 1881 } 1882 break; 1883 case 'g': 1884 if (strcmp(key, "gid") == 0) { 1885 archive_entry_set_gid(entry, 1886 tar_atol10(value, strlen(value))); 1887 } else if (strcmp(key, "gname") == 0) { 1888 archive_strcpy(&(tar->entry_gname), value); 1889 } 1890 break; 1891 case 'h': 1892 if (strcmp(key, "hdrcharset") == 0) { 1893 if (strcmp(value, "BINARY") == 0) 1894 /* Binary mode. */ 1895 tar->pax_hdrcharset_binary = 1; 1896 else if (strcmp(value, "ISO-IR 10646 2000 UTF-8") == 0) 1897 tar->pax_hdrcharset_binary = 0; 1898 } 1899 break; 1900 case 'l': 1901 /* pax interchange doesn't distinguish hardlink vs. symlink. */ 1902 if (strcmp(key, "linkpath") == 0) { 1903 archive_strcpy(&(tar->entry_linkpath), value); 1904 } 1905 break; 1906 case 'm': 1907 if (strcmp(key, "mtime") == 0) { 1908 pax_time(value, &s, &n); 1909 archive_entry_set_mtime(entry, s, n); 1910 } 1911 break; 1912 case 'p': 1913 if (strcmp(key, "path") == 0) { 1914 archive_strcpy(&(tar->entry_pathname), value); 1915 } 1916 break; 1917 case 'r': 1918 /* POSIX has reserved 'realtime.*' */ 1919 break; 1920 case 's': 1921 /* POSIX has reserved 'security.*' */ 1922 /* Someday: if (strcmp(key, "security.acl") == 0) { ... } */ 1923 if (strcmp(key, "size") == 0) { 1924 /* "size" is the size of the data in the entry. */ 1925 tar->entry_bytes_remaining 1926 = tar_atol10(value, strlen(value)); 1927 /* 1928 * But, "size" is not necessarily the size of 1929 * the file on disk; if this is a sparse file, 1930 * the disk size may have already been set from 1931 * GNU.sparse.realsize or GNU.sparse.size or 1932 * an old GNU header field or SCHILY.realsize 1933 * or .... 1934 */ 1935 if (tar->realsize < 0) { 1936 archive_entry_set_size(entry, 1937 tar->entry_bytes_remaining); 1938 tar->realsize 1939 = tar->entry_bytes_remaining; 1940 } 1941 } 1942 break; 1943 case 'u': 1944 if (strcmp(key, "uid") == 0) { 1945 archive_entry_set_uid(entry, 1946 tar_atol10(value, strlen(value))); 1947 } else if (strcmp(key, "uname") == 0) { 1948 archive_strcpy(&(tar->entry_uname), value); 1949 } 1950 break; 1951 } 1952 return (err); 1953} 1954 1955 1956 1957/* 1958 * parse a decimal time value, which may include a fractional portion 1959 */ 1960static void 1961pax_time(const char *p, int64_t *ps, long *pn) 1962{ 1963 char digit; 1964 int64_t s; 1965 unsigned long l; 1966 int sign; 1967 int64_t limit, last_digit_limit; 1968 1969 limit = INT64_MAX / 10; 1970 last_digit_limit = INT64_MAX % 10; 1971 1972 s = 0; 1973 sign = 1; 1974 if (*p == '-') { 1975 sign = -1; 1976 p++; 1977 } 1978 while (*p >= '0' && *p <= '9') { 1979 digit = *p - '0'; 1980 if (s > limit || 1981 (s == limit && digit > last_digit_limit)) { 1982 s = INT64_MAX; 1983 break; 1984 } 1985 s = (s * 10) + digit; 1986 ++p; 1987 } 1988 1989 *ps = s * sign; 1990 1991 /* Calculate nanoseconds. */ 1992 *pn = 0; 1993 1994 if (*p != '.') 1995 return; 1996 1997 l = 100000000UL; 1998 do { 1999 ++p; 2000 if (*p >= '0' && *p <= '9') 2001 *pn += (*p - '0') * l; 2002 else 2003 break; 2004 } while (l /= 10); 2005} 2006 2007/* 2008 * Parse GNU tar header 2009 */ 2010static int 2011header_gnutar(struct archive_read *a, struct tar *tar, 2012 struct archive_entry *entry, const void *h, size_t *unconsumed) 2013{ 2014 const struct archive_entry_header_gnutar *header; 2015 int64_t t; 2016 int err = ARCHIVE_OK; 2017 2018 /* 2019 * GNU header is like POSIX ustar, except 'prefix' is 2020 * replaced with some other fields. This also means the 2021 * filename is stored as in old-style archives. 2022 */ 2023 2024 /* Grab fields common to all tar variants. */ 2025 err = header_common(a, tar, entry, h); 2026 if (err == ARCHIVE_FATAL) 2027 return (err); 2028 2029 /* Copy filename over (to ensure null termination). */ 2030 header = (const struct archive_entry_header_gnutar *)h; 2031 if (archive_entry_copy_pathname_l(entry, 2032 header->name, sizeof(header->name), tar->sconv) != 0) { 2033 err = set_conversion_failed_error(a, tar->sconv, "Pathname"); 2034 if (err == ARCHIVE_FATAL) 2035 return (err); 2036 } 2037 2038 /* Fields common to ustar and GNU */ 2039 /* XXX Can the following be factored out since it's common 2040 * to ustar and gnu tar? Is it okay to move it down into 2041 * header_common, perhaps? */ 2042 if (archive_entry_copy_uname_l(entry, 2043 header->uname, sizeof(header->uname), tar->sconv) != 0) { 2044 err = set_conversion_failed_error(a, tar->sconv, "Uname"); 2045 if (err == ARCHIVE_FATAL) 2046 return (err); 2047 } 2048 2049 if (archive_entry_copy_gname_l(entry, 2050 header->gname, sizeof(header->gname), tar->sconv) != 0) { 2051 err = set_conversion_failed_error(a, tar->sconv, "Gname"); 2052 if (err == ARCHIVE_FATAL) 2053 return (err); 2054 } 2055 2056 /* Parse out device numbers only for char and block specials */ 2057 if (header->typeflag[0] == '3' || header->typeflag[0] == '4') { 2058 archive_entry_set_rdevmajor(entry, (dev_t) 2059 tar_atol(header->rdevmajor, sizeof(header->rdevmajor))); 2060 archive_entry_set_rdevminor(entry, (dev_t) 2061 tar_atol(header->rdevminor, sizeof(header->rdevminor))); 2062 } else 2063 archive_entry_set_rdev(entry, 0); 2064 2065 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining); 2066 2067 /* Grab GNU-specific fields. */ 2068 t = tar_atol(header->atime, sizeof(header->atime)); 2069 if (t > 0) 2070 archive_entry_set_atime(entry, t, 0); 2071 t = tar_atol(header->ctime, sizeof(header->ctime)); 2072 if (t > 0) 2073 archive_entry_set_ctime(entry, t, 0); 2074 2075 if (header->realsize[0] != 0) { 2076 tar->realsize 2077 = tar_atol(header->realsize, sizeof(header->realsize)); 2078 archive_entry_set_size(entry, tar->realsize); 2079 } 2080 2081 if (header->sparse[0].offset[0] != 0) { 2082 if (gnu_sparse_old_read(a, tar, header, unconsumed) 2083 != ARCHIVE_OK) 2084 return (ARCHIVE_FATAL); 2085 } else { 2086 if (header->isextended[0] != 0) { 2087 /* XXX WTF? XXX */ 2088 } 2089 } 2090 2091 return (err); 2092} 2093 2094static int 2095gnu_add_sparse_entry(struct archive_read *a, struct tar *tar, 2096 int64_t offset, int64_t remaining) 2097{ 2098 struct sparse_block *p; 2099 2100 p = (struct sparse_block *)malloc(sizeof(*p)); 2101 if (p == NULL) { 2102 archive_set_error(&a->archive, ENOMEM, "Out of memory"); 2103 return (ARCHIVE_FATAL); 2104 } 2105 memset(p, 0, sizeof(*p)); 2106 if (tar->sparse_last != NULL) 2107 tar->sparse_last->next = p; 2108 else 2109 tar->sparse_list = p; 2110 tar->sparse_last = p; 2111 p->offset = offset; 2112 p->remaining = remaining; 2113 return (ARCHIVE_OK); 2114} 2115 2116static void 2117gnu_clear_sparse_list(struct tar *tar) 2118{ 2119 struct sparse_block *p; 2120 2121 while (tar->sparse_list != NULL) { 2122 p = tar->sparse_list; 2123 tar->sparse_list = p->next; 2124 free(p); 2125 } 2126 tar->sparse_last = NULL; 2127} 2128 2129/* 2130 * GNU tar old-format sparse data. 2131 * 2132 * GNU old-format sparse data is stored in a fixed-field 2133 * format. Offset/size values are 11-byte octal fields (same 2134 * format as 'size' field in ustart header). These are 2135 * stored in the header, allocating subsequent header blocks 2136 * as needed. Extending the header in this way is a pretty 2137 * severe POSIX violation; this design has earned GNU tar a 2138 * lot of criticism. 2139 */ 2140 2141static int 2142gnu_sparse_old_read(struct archive_read *a, struct tar *tar, 2143 const struct archive_entry_header_gnutar *header, size_t *unconsumed) 2144{ 2145 ssize_t bytes_read; 2146 const void *data; 2147 struct extended { 2148 struct gnu_sparse sparse[21]; 2149 char isextended[1]; 2150 char padding[7]; 2151 }; 2152 const struct extended *ext; 2153 2154 if (gnu_sparse_old_parse(a, tar, header->sparse, 4) != ARCHIVE_OK) 2155 return (ARCHIVE_FATAL); 2156 if (header->isextended[0] == 0) 2157 return (ARCHIVE_OK); 2158 2159 do { 2160 tar_flush_unconsumed(a, unconsumed); 2161 data = __archive_read_ahead(a, 512, &bytes_read); 2162 if (bytes_read < 0) 2163 return (ARCHIVE_FATAL); 2164 if (bytes_read < 512) { 2165 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 2166 "Truncated tar archive " 2167 "detected while reading sparse file data"); 2168 return (ARCHIVE_FATAL); 2169 } 2170 *unconsumed = 512; 2171 ext = (const struct extended *)data; 2172 if (gnu_sparse_old_parse(a, tar, ext->sparse, 21) != ARCHIVE_OK) 2173 return (ARCHIVE_FATAL); 2174 } while (ext->isextended[0] != 0); 2175 if (tar->sparse_list != NULL) 2176 tar->entry_offset = tar->sparse_list->offset; 2177 return (ARCHIVE_OK); 2178} 2179 2180static int 2181gnu_sparse_old_parse(struct archive_read *a, struct tar *tar, 2182 const struct gnu_sparse *sparse, int length) 2183{ 2184 while (length > 0 && sparse->offset[0] != 0) { 2185 if (gnu_add_sparse_entry(a, tar, 2186 tar_atol(sparse->offset, sizeof(sparse->offset)), 2187 tar_atol(sparse->numbytes, sizeof(sparse->numbytes))) 2188 != ARCHIVE_OK) 2189 return (ARCHIVE_FATAL); 2190 sparse++; 2191 length--; 2192 } 2193 return (ARCHIVE_OK); 2194} 2195 2196/* 2197 * GNU tar sparse format 0.0 2198 * 2199 * Beginning with GNU tar 1.15, sparse files are stored using 2200 * information in the pax extended header. The GNU tar maintainers 2201 * have gone through a number of variations in the process of working 2202 * out this scheme; fortunately, they're all numbered. 2203 * 2204 * Sparse format 0.0 uses attribute GNU.sparse.numblocks to store the 2205 * number of blocks, and GNU.sparse.offset/GNU.sparse.numbytes to 2206 * store offset/size for each block. The repeated instances of these 2207 * latter fields violate the pax specification (which frowns on 2208 * duplicate keys), so this format was quickly replaced. 2209 */ 2210 2211/* 2212 * GNU tar sparse format 0.1 2213 * 2214 * This version replaced the offset/numbytes attributes with 2215 * a single "map" attribute that stored a list of integers. This 2216 * format had two problems: First, the "map" attribute could be very 2217 * long, which caused problems for some implementations. More 2218 * importantly, the sparse data was lost when extracted by archivers 2219 * that didn't recognize this extension. 2220 */ 2221 2222static int 2223gnu_sparse_01_parse(struct archive_read *a, struct tar *tar, const char *p) 2224{ 2225 const char *e; 2226 int64_t offset = -1, size = -1; 2227 2228 for (;;) { 2229 e = p; 2230 while (*e != '\0' && *e != ',') { 2231 if (*e < '0' || *e > '9') 2232 return (ARCHIVE_WARN); 2233 e++; 2234 } 2235 if (offset < 0) { 2236 offset = tar_atol10(p, e - p); 2237 if (offset < 0) 2238 return (ARCHIVE_WARN); 2239 } else { 2240 size = tar_atol10(p, e - p); 2241 if (size < 0) 2242 return (ARCHIVE_WARN); 2243 if (gnu_add_sparse_entry(a, tar, offset, size) 2244 != ARCHIVE_OK) 2245 return (ARCHIVE_FATAL); 2246 offset = -1; 2247 } 2248 if (*e == '\0') 2249 return (ARCHIVE_OK); 2250 p = e + 1; 2251 } 2252} 2253 2254/* 2255 * GNU tar sparse format 1.0 2256 * 2257 * The idea: The offset/size data is stored as a series of base-10 2258 * ASCII numbers prepended to the file data, so that dearchivers that 2259 * don't support this format will extract the block map along with the 2260 * data and a separate post-process can restore the sparseness. 2261 * 2262 * Unfortunately, GNU tar 1.16 had a bug that added unnecessary 2263 * padding to the body of the file when using this format. GNU tar 2264 * 1.17 corrected this bug without bumping the version number, so 2265 * it's not possible to support both variants. This code supports 2266 * the later variant at the expense of not supporting the former. 2267 * 2268 * This variant also replaced GNU.sparse.size with GNU.sparse.realsize 2269 * and introduced the GNU.sparse.major/GNU.sparse.minor attributes. 2270 */ 2271 2272/* 2273 * Read the next line from the input, and parse it as a decimal 2274 * integer followed by '\n'. Returns positive integer value or 2275 * negative on error. 2276 */ 2277static int64_t 2278gnu_sparse_10_atol(struct archive_read *a, struct tar *tar, 2279 int64_t *remaining, size_t *unconsumed) 2280{ 2281 int64_t l, limit, last_digit_limit; 2282 const char *p; 2283 ssize_t bytes_read; 2284 int base, digit; 2285 2286 base = 10; 2287 limit = INT64_MAX / base; 2288 last_digit_limit = INT64_MAX % base; 2289 2290 /* 2291 * Skip any lines starting with '#'; GNU tar specs 2292 * don't require this, but they should. 2293 */ 2294 do { 2295 bytes_read = readline(a, tar, &p, 2296 (ssize_t)tar_min(*remaining, 100), unconsumed); 2297 if (bytes_read <= 0) 2298 return (ARCHIVE_FATAL); 2299 *remaining -= bytes_read; 2300 } while (p[0] == '#'); 2301 2302 l = 0; 2303 while (bytes_read > 0) { 2304 if (*p == '\n') 2305 return (l); 2306 if (*p < '0' || *p >= '0' + base) 2307 return (ARCHIVE_WARN); 2308 digit = *p - '0'; 2309 if (l > limit || (l == limit && digit > last_digit_limit)) 2310 l = INT64_MAX; /* Truncate on overflow. */ 2311 else 2312 l = (l * base) + digit; 2313 p++; 2314 bytes_read--; 2315 } 2316 /* TODO: Error message. */ 2317 return (ARCHIVE_WARN); 2318} 2319 2320/* 2321 * Returns length (in bytes) of the sparse data description 2322 * that was read. 2323 */ 2324static ssize_t 2325gnu_sparse_10_read(struct archive_read *a, struct tar *tar, size_t *unconsumed) 2326{ 2327 ssize_t bytes_read; 2328 int entries; 2329 int64_t offset, size, to_skip, remaining; 2330 2331 /* Clear out the existing sparse list. */ 2332 gnu_clear_sparse_list(tar); 2333 2334 remaining = tar->entry_bytes_remaining; 2335 2336 /* Parse entries. */ 2337 entries = (int)gnu_sparse_10_atol(a, tar, &remaining, unconsumed); 2338 if (entries < 0) 2339 return (ARCHIVE_FATAL); 2340 /* Parse the individual entries. */ 2341 while (entries-- > 0) { 2342 /* Parse offset/size */ 2343 offset = gnu_sparse_10_atol(a, tar, &remaining, unconsumed); 2344 if (offset < 0) 2345 return (ARCHIVE_FATAL); 2346 size = gnu_sparse_10_atol(a, tar, &remaining, unconsumed); 2347 if (size < 0) 2348 return (ARCHIVE_FATAL); 2349 /* Add a new sparse entry. */ 2350 if (gnu_add_sparse_entry(a, tar, offset, size) != ARCHIVE_OK) 2351 return (ARCHIVE_FATAL); 2352 } 2353 /* Skip rest of block... */ 2354 tar_flush_unconsumed(a, unconsumed); 2355 bytes_read = (ssize_t)(tar->entry_bytes_remaining - remaining); 2356 to_skip = 0x1ff & -bytes_read; 2357 if (to_skip != __archive_read_consume(a, to_skip)) 2358 return (ARCHIVE_FATAL); 2359 return ((ssize_t)(bytes_read + to_skip)); 2360} 2361 2362/* 2363 * Solaris pax extension for a sparse file. This is recorded with the 2364 * data and hole pairs. The way recording sparse information by Solaris' 2365 * pax simply indicates where data and sparse are, so the stored contents 2366 * consist of both data and hole. 2367 */ 2368static int 2369solaris_sparse_parse(struct archive_read *a, struct tar *tar, 2370 struct archive_entry *entry, const char *p) 2371{ 2372 const char *e; 2373 int64_t start, end; 2374 int hole = 1; 2375 2376 (void)entry; /* UNUSED */ 2377 2378 end = 0; 2379 if (*p == ' ') 2380 p++; 2381 else 2382 return (ARCHIVE_WARN); 2383 for (;;) { 2384 e = p; 2385 while (*e != '\0' && *e != ' ') { 2386 if (*e < '0' || *e > '9') 2387 return (ARCHIVE_WARN); 2388 e++; 2389 } 2390 start = end; 2391 end = tar_atol10(p, e - p); 2392 if (end < 0) 2393 return (ARCHIVE_WARN); 2394 if (start < end) { 2395 if (gnu_add_sparse_entry(a, tar, start, 2396 end - start) != ARCHIVE_OK) 2397 return (ARCHIVE_FATAL); 2398 tar->sparse_last->hole = hole; 2399 } 2400 if (*e == '\0') 2401 return (ARCHIVE_OK); 2402 p = e + 1; 2403 hole = hole == 0; 2404 } 2405} 2406 2407/*- 2408 * Convert text->integer. 2409 * 2410 * Traditional tar formats (including POSIX) specify base-8 for 2411 * all of the standard numeric fields. This is a significant limitation 2412 * in practice: 2413 * = file size is limited to 8GB 2414 * = rdevmajor and rdevminor are limited to 21 bits 2415 * = uid/gid are limited to 21 bits 2416 * 2417 * There are two workarounds for this: 2418 * = pax extended headers, which use variable-length string fields 2419 * = GNU tar and STAR both allow either base-8 or base-256 in 2420 * most fields. The high bit is set to indicate base-256. 2421 * 2422 * On read, this implementation supports both extensions. 2423 */ 2424static int64_t 2425tar_atol(const char *p, size_t char_cnt) 2426{ 2427 /* 2428 * Technically, GNU tar considers a field to be in base-256 2429 * only if the first byte is 0xff or 0x80. 2430 */ 2431 if (*p & 0x80) 2432 return (tar_atol256(p, char_cnt)); 2433 return (tar_atol8(p, char_cnt)); 2434} 2435 2436/* 2437 * Note that this implementation does not (and should not!) obey 2438 * locale settings; you cannot simply substitute strtol here, since 2439 * it does obey locale. 2440 */ 2441static int64_t 2442tar_atol_base_n(const char *p, size_t char_cnt, int base) 2443{ 2444 int64_t l, limit, last_digit_limit; 2445 int digit, sign; 2446 2447 limit = INT64_MAX / base; 2448 last_digit_limit = INT64_MAX % base; 2449 2450 /* the pointer will not be dereferenced if char_cnt is zero 2451 * due to the way the && operator is evaulated. 2452 */ 2453 while (char_cnt != 0 && (*p == ' ' || *p == '\t')) { 2454 p++; 2455 char_cnt--; 2456 } 2457 2458 sign = 1; 2459 if (char_cnt != 0 && *p == '-') { 2460 sign = -1; 2461 p++; 2462 char_cnt--; 2463 } 2464 2465 l = 0; 2466 if (char_cnt != 0) { 2467 digit = *p - '0'; 2468 while (digit >= 0 && digit < base && char_cnt != 0) { 2469 if (l>limit || (l == limit && digit > last_digit_limit)) { 2470 l = INT64_MAX; /* Truncate on overflow. */ 2471 break; 2472 } 2473 l = (l * base) + digit; 2474 digit = *++p - '0'; 2475 char_cnt--; 2476 } 2477 } 2478 return (sign < 0) ? -l : l; 2479} 2480 2481static int64_t 2482tar_atol8(const char *p, size_t char_cnt) 2483{ 2484 return tar_atol_base_n(p, char_cnt, 8); 2485} 2486 2487static int64_t 2488tar_atol10(const char *p, size_t char_cnt) 2489{ 2490 return tar_atol_base_n(p, char_cnt, 10); 2491} 2492 2493/* 2494 * Parse a base-256 integer. This is just a straight signed binary 2495 * value in big-endian order, except that the high-order bit is 2496 * ignored. 2497 */ 2498static int64_t 2499tar_atol256(const char *_p, size_t char_cnt) 2500{ 2501 int64_t l, upper_limit, lower_limit; 2502 const unsigned char *p = (const unsigned char *)_p; 2503 2504 upper_limit = INT64_MAX / 256; 2505 lower_limit = INT64_MIN / 256; 2506 2507 /* Pad with 1 or 0 bits, depending on sign. */ 2508 if ((0x40 & *p) == 0x40) 2509 l = (int64_t)-1; 2510 else 2511 l = 0; 2512 l = (l << 6) | (0x3f & *p++); 2513 while (--char_cnt > 0) { 2514 if (l > upper_limit) { 2515 l = INT64_MAX; /* Truncate on overflow */ 2516 break; 2517 } else if (l < lower_limit) { 2518 l = INT64_MIN; 2519 break; 2520 } 2521 l = (l << 8) | (0xff & (int64_t)*p++); 2522 } 2523 return (l); 2524} 2525 2526/* 2527 * Returns length of line (including trailing newline) 2528 * or negative on error. 'start' argument is updated to 2529 * point to first character of line. This avoids copying 2530 * when possible. 2531 */ 2532static ssize_t 2533readline(struct archive_read *a, struct tar *tar, const char **start, 2534 ssize_t limit, size_t *unconsumed) 2535{ 2536 ssize_t bytes_read; 2537 ssize_t total_size = 0; 2538 const void *t; 2539 const char *s; 2540 void *p; 2541 2542 tar_flush_unconsumed(a, unconsumed); 2543 2544 t = __archive_read_ahead(a, 1, &bytes_read); 2545 if (bytes_read <= 0) 2546 return (ARCHIVE_FATAL); 2547 s = t; /* Start of line? */ 2548 p = memchr(t, '\n', bytes_read); 2549 /* If we found '\n' in the read buffer, return pointer to that. */ 2550 if (p != NULL) { 2551 bytes_read = 1 + ((const char *)p) - s; 2552 if (bytes_read > limit) { 2553 archive_set_error(&a->archive, 2554 ARCHIVE_ERRNO_FILE_FORMAT, 2555 "Line too long"); 2556 return (ARCHIVE_FATAL); 2557 } 2558 *unconsumed = bytes_read; 2559 *start = s; 2560 return (bytes_read); 2561 } 2562 *unconsumed = bytes_read; 2563 /* Otherwise, we need to accumulate in a line buffer. */ 2564 for (;;) { 2565 if (total_size + bytes_read > limit) { 2566 archive_set_error(&a->archive, 2567 ARCHIVE_ERRNO_FILE_FORMAT, 2568 "Line too long"); 2569 return (ARCHIVE_FATAL); 2570 } 2571 if (archive_string_ensure(&tar->line, total_size + bytes_read) == NULL) { 2572 archive_set_error(&a->archive, ENOMEM, 2573 "Can't allocate working buffer"); 2574 return (ARCHIVE_FATAL); 2575 } 2576 memcpy(tar->line.s + total_size, t, bytes_read); 2577 tar_flush_unconsumed(a, unconsumed); 2578 total_size += bytes_read; 2579 /* If we found '\n', clean up and return. */ 2580 if (p != NULL) { 2581 *start = tar->line.s; 2582 return (total_size); 2583 } 2584 /* Read some more. */ 2585 t = __archive_read_ahead(a, 1, &bytes_read); 2586 if (bytes_read <= 0) 2587 return (ARCHIVE_FATAL); 2588 s = t; /* Start of line? */ 2589 p = memchr(t, '\n', bytes_read); 2590 /* If we found '\n', trim the read. */ 2591 if (p != NULL) { 2592 bytes_read = 1 + ((const char *)p) - s; 2593 } 2594 *unconsumed = bytes_read; 2595 } 2596} 2597 2598/* 2599 * base64_decode - Base64 decode 2600 * 2601 * This accepts most variations of base-64 encoding, including: 2602 * * with or without line breaks 2603 * * with or without the final group padded with '=' or '_' characters 2604 * (The most economical Base-64 variant does not pad the last group and 2605 * omits line breaks; RFC1341 used for MIME requires both.) 2606 */ 2607static char * 2608base64_decode(const char *s, size_t len, size_t *out_len) 2609{ 2610 static const unsigned char digits[64] = { 2611 'A','B','C','D','E','F','G','H','I','J','K','L','M','N', 2612 'O','P','Q','R','S','T','U','V','W','X','Y','Z','a','b', 2613 'c','d','e','f','g','h','i','j','k','l','m','n','o','p', 2614 'q','r','s','t','u','v','w','x','y','z','0','1','2','3', 2615 '4','5','6','7','8','9','+','/' }; 2616 static unsigned char decode_table[128]; 2617 char *out, *d; 2618 const unsigned char *src = (const unsigned char *)s; 2619 2620 /* If the decode table is not yet initialized, prepare it. */ 2621 if (decode_table[digits[1]] != 1) { 2622 unsigned i; 2623 memset(decode_table, 0xff, sizeof(decode_table)); 2624 for (i = 0; i < sizeof(digits); i++) 2625 decode_table[digits[i]] = i; 2626 } 2627 2628 /* Allocate enough space to hold the entire output. */ 2629 /* Note that we may not use all of this... */ 2630 out = (char *)malloc(len - len / 4 + 1); 2631 if (out == NULL) { 2632 *out_len = 0; 2633 return (NULL); 2634 } 2635 d = out; 2636 2637 while (len > 0) { 2638 /* Collect the next group of (up to) four characters. */ 2639 int v = 0; 2640 int group_size = 0; 2641 while (group_size < 4 && len > 0) { 2642 /* '=' or '_' padding indicates final group. */ 2643 if (*src == '=' || *src == '_') { 2644 len = 0; 2645 break; 2646 } 2647 /* Skip illegal characters (including line breaks) */ 2648 if (*src > 127 || *src < 32 2649 || decode_table[*src] == 0xff) { 2650 len--; 2651 src++; 2652 continue; 2653 } 2654 v <<= 6; 2655 v |= decode_table[*src++]; 2656 len --; 2657 group_size++; 2658 } 2659 /* Align a short group properly. */ 2660 v <<= 6 * (4 - group_size); 2661 /* Unpack the group we just collected. */ 2662 switch (group_size) { 2663 case 4: d[2] = v & 0xff; 2664 /* FALLTHROUGH */ 2665 case 3: d[1] = (v >> 8) & 0xff; 2666 /* FALLTHROUGH */ 2667 case 2: d[0] = (v >> 16) & 0xff; 2668 break; 2669 case 1: /* this is invalid! */ 2670 break; 2671 } 2672 d += group_size * 3 / 4; 2673 } 2674 2675 *out_len = d - out; 2676 return (out); 2677} 2678 2679static char * 2680url_decode(const char *in) 2681{ 2682 char *out, *d; 2683 const char *s; 2684 2685 out = (char *)malloc(strlen(in) + 1); 2686 if (out == NULL) 2687 return (NULL); 2688 for (s = in, d = out; *s != '\0'; ) { 2689 if (s[0] == '%' && s[1] != '\0' && s[2] != '\0') { 2690 /* Try to convert % escape */ 2691 int digit1 = tohex(s[1]); 2692 int digit2 = tohex(s[2]); 2693 if (digit1 >= 0 && digit2 >= 0) { 2694 /* Looks good, consume three chars */ 2695 s += 3; 2696 /* Convert output */ 2697 *d++ = ((digit1 << 4) | digit2); 2698 continue; 2699 } 2700 /* Else fall through and treat '%' as normal char */ 2701 } 2702 *d++ = *s++; 2703 } 2704 *d = '\0'; 2705 return (out); 2706} 2707 2708static int 2709tohex(int c) 2710{ 2711 if (c >= '0' && c <= '9') 2712 return (c - '0'); 2713 else if (c >= 'A' && c <= 'F') 2714 return (c - 'A' + 10); 2715 else if (c >= 'a' && c <= 'f') 2716 return (c - 'a' + 10); 2717 else 2718 return (-1); 2719} 2720