archive_read_support_format_tar.c revision 286082
1212526Simp/*- 2212526Simp * Copyright (c) 2003-2007 Tim Kientzle 3252356Sdavide * 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: stable/10/contrib/libarchive/libarchive/archive_read_support_format_tar.c 286082 2015-07-30 17:04:15Z bdrewery $"); 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 struct sparse_block *sparse_list; 140 struct sparse_block *sparse_last; 141 int64_t sparse_offset; 142 int64_t sparse_numbytes; 143 int sparse_gnu_major; 144 int sparse_gnu_minor; 145 char sparse_gnu_pending; 146 147 struct archive_string localname; 148 struct archive_string_conv *opt_sconv; 149 struct archive_string_conv *sconv; 150 struct archive_string_conv *sconv_acl; 151 struct archive_string_conv *sconv_default; 152 int init_default_conversion; 153 int compat_2x; 154}; 155 156static int archive_block_is_null(const char *p); 157static char *base64_decode(const char *, size_t, size_t *); 158static int gnu_add_sparse_entry(struct archive_read *, struct tar *, 159 int64_t offset, int64_t remaining); 160 161static void gnu_clear_sparse_list(struct tar *); 162static int gnu_sparse_old_read(struct archive_read *, struct tar *, 163 const struct archive_entry_header_gnutar *header, size_t *); 164static int gnu_sparse_old_parse(struct archive_read *, struct tar *, 165 const struct gnu_sparse *sparse, int length); 166static int gnu_sparse_01_parse(struct archive_read *, struct tar *, 167 const char *); 168static ssize_t gnu_sparse_10_read(struct archive_read *, struct tar *, 169 size_t *); 170static int header_Solaris_ACL(struct archive_read *, struct tar *, 171 struct archive_entry *, const void *, size_t *); 172static int header_common(struct archive_read *, struct tar *, 173 struct archive_entry *, const void *); 174static int header_old_tar(struct archive_read *, struct tar *, 175 struct archive_entry *, const void *); 176static int header_pax_extensions(struct archive_read *, struct tar *, 177 struct archive_entry *, const void *, size_t *); 178static int header_pax_global(struct archive_read *, struct tar *, 179 struct archive_entry *, const void *h, size_t *); 180static int header_longlink(struct archive_read *, struct tar *, 181 struct archive_entry *, const void *h, size_t *); 182static int header_longname(struct archive_read *, struct tar *, 183 struct archive_entry *, const void *h, size_t *); 184static int read_mac_metadata_blob(struct archive_read *, struct tar *, 185 struct archive_entry *, const void *h, size_t *); 186static int header_volume(struct archive_read *, struct tar *, 187 struct archive_entry *, const void *h, size_t *); 188static int header_ustar(struct archive_read *, struct tar *, 189 struct archive_entry *, const void *h); 190static int header_gnutar(struct archive_read *, struct tar *, 191 struct archive_entry *, const void *h, size_t *); 192static int archive_read_format_tar_bid(struct archive_read *, int); 193static int archive_read_format_tar_options(struct archive_read *, 194 const char *, const char *); 195static int archive_read_format_tar_cleanup(struct archive_read *); 196static int archive_read_format_tar_read_data(struct archive_read *a, 197 const void **buff, size_t *size, int64_t *offset); 198static int archive_read_format_tar_skip(struct archive_read *a); 199static int archive_read_format_tar_read_header(struct archive_read *, 200 struct archive_entry *); 201static int checksum(struct archive_read *, const void *); 202static int pax_attribute(struct archive_read *, struct tar *, 203 struct archive_entry *, char *key, char *value); 204static int pax_header(struct archive_read *, struct tar *, 205 struct archive_entry *, char *attr); 206static void pax_time(const char *, int64_t *sec, long *nanos); 207static ssize_t readline(struct archive_read *, struct tar *, const char **, 208 ssize_t limit, size_t *); 209static int read_body_to_string(struct archive_read *, struct tar *, 210 struct archive_string *, const void *h, size_t *); 211static int solaris_sparse_parse(struct archive_read *, struct tar *, 212 struct archive_entry *, const char *); 213static int64_t tar_atol(const char *, size_t); 214static int64_t tar_atol10(const char *, size_t); 215static int64_t tar_atol256(const char *, size_t); 216static int64_t tar_atol8(const char *, size_t); 217static int tar_read_header(struct archive_read *, struct tar *, 218 struct archive_entry *, size_t *); 219static int tohex(int c); 220static char *url_decode(const char *); 221static void tar_flush_unconsumed(struct archive_read *, size_t *); 222 223 224int 225archive_read_support_format_gnutar(struct archive *a) 226{ 227 archive_check_magic(a, ARCHIVE_READ_MAGIC, 228 ARCHIVE_STATE_NEW, "archive_read_support_format_gnutar"); 229 return (archive_read_support_format_tar(a)); 230} 231 232 233int 234archive_read_support_format_tar(struct archive *_a) 235{ 236 struct archive_read *a = (struct archive_read *)_a; 237 struct tar *tar; 238 int r; 239 240 archive_check_magic(_a, ARCHIVE_READ_MAGIC, 241 ARCHIVE_STATE_NEW, "archive_read_support_format_tar"); 242 243 tar = (struct tar *)calloc(1, sizeof(*tar)); 244 if (tar == NULL) { 245 archive_set_error(&a->archive, ENOMEM, 246 "Can't allocate tar data"); 247 return (ARCHIVE_FATAL); 248 } 249 250 r = __archive_read_register_format(a, tar, "tar", 251 archive_read_format_tar_bid, 252 archive_read_format_tar_options, 253 archive_read_format_tar_read_header, 254 archive_read_format_tar_read_data, 255 archive_read_format_tar_skip, 256 NULL, 257 archive_read_format_tar_cleanup); 258 259 if (r != ARCHIVE_OK) 260 free(tar); 261 return (ARCHIVE_OK); 262} 263 264static int 265archive_read_format_tar_cleanup(struct archive_read *a) 266{ 267 struct tar *tar; 268 269 tar = (struct tar *)(a->format->data); 270 gnu_clear_sparse_list(tar); 271 archive_string_free(&tar->acl_text); 272 archive_string_free(&tar->entry_pathname); 273 archive_string_free(&tar->entry_pathname_override); 274 archive_string_free(&tar->entry_linkpath); 275 archive_string_free(&tar->entry_uname); 276 archive_string_free(&tar->entry_gname); 277 archive_string_free(&tar->line); 278 archive_string_free(&tar->pax_global); 279 archive_string_free(&tar->pax_header); 280 archive_string_free(&tar->longname); 281 archive_string_free(&tar->longlink); 282 archive_string_free(&tar->localname); 283 free(tar); 284 (a->format->data) = NULL; 285 return (ARCHIVE_OK); 286} 287 288 289static int 290archive_read_format_tar_bid(struct archive_read *a, int best_bid) 291{ 292 int bid; 293 const char *h; 294 const struct archive_entry_header_ustar *header; 295 296 (void)best_bid; /* UNUSED */ 297 298 bid = 0; 299 300 /* Now let's look at the actual header and see if it matches. */ 301 h = __archive_read_ahead(a, 512, NULL); 302 if (h == NULL) 303 return (-1); 304 305 /* If it's an end-of-archive mark, we can handle it. */ 306 if (h[0] == 0 && archive_block_is_null(h)) { 307 /* 308 * Usually, I bid the number of bits verified, but 309 * in this case, 4096 seems excessive so I picked 10 as 310 * an arbitrary but reasonable-seeming value. 311 */ 312 return (10); 313 } 314 315 /* If it's not an end-of-archive mark, it must have a valid checksum.*/ 316 if (!checksum(a, h)) 317 return (0); 318 bid += 48; /* Checksum is usually 6 octal digits. */ 319 320 header = (const struct archive_entry_header_ustar *)h; 321 322 /* Recognize POSIX formats. */ 323 if ((memcmp(header->magic, "ustar\0", 6) == 0) 324 && (memcmp(header->version, "00", 2) == 0)) 325 bid += 56; 326 327 /* Recognize GNU tar format. */ 328 if ((memcmp(header->magic, "ustar ", 6) == 0) 329 && (memcmp(header->version, " \0", 2) == 0)) 330 bid += 56; 331 332 /* Type flag must be null, digit or A-Z, a-z. */ 333 if (header->typeflag[0] != 0 && 334 !( header->typeflag[0] >= '0' && header->typeflag[0] <= '9') && 335 !( header->typeflag[0] >= 'A' && header->typeflag[0] <= 'Z') && 336 !( header->typeflag[0] >= 'a' && header->typeflag[0] <= 'z') ) 337 return (0); 338 bid += 2; /* 6 bits of variation in an 8-bit field leaves 2 bits. */ 339 340 /* Sanity check: Look at first byte of mode field. */ 341 switch (255 & (unsigned)header->mode[0]) { 342 case 0: case 255: 343 /* Base-256 value: No further verification possible! */ 344 break; 345 case ' ': /* Not recommended, but not illegal, either. */ 346 break; 347 case '0': case '1': case '2': case '3': 348 case '4': case '5': case '6': case '7': 349 /* Octal Value. */ 350 /* TODO: Check format of remainder of this field. */ 351 break; 352 default: 353 /* Not a valid mode; bail out here. */ 354 return (0); 355 } 356 /* TODO: Sanity test uid/gid/size/mtime/rdevmajor/rdevminor fields. */ 357 358 return (bid); 359} 360 361static int 362archive_read_format_tar_options(struct archive_read *a, 363 const char *key, const char *val) 364{ 365 struct tar *tar; 366 int ret = ARCHIVE_FAILED; 367 368 tar = (struct tar *)(a->format->data); 369 if (strcmp(key, "compat-2x") == 0) { 370 /* Handle UTF-8 filnames as libarchive 2.x */ 371 tar->compat_2x = (val != NULL)?1:0; 372 tar->init_default_conversion = tar->compat_2x; 373 return (ARCHIVE_OK); 374 } else if (strcmp(key, "hdrcharset") == 0) { 375 if (val == NULL || val[0] == 0) 376 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 377 "tar: hdrcharset option needs a character-set name"); 378 else { 379 tar->opt_sconv = 380 archive_string_conversion_from_charset( 381 &a->archive, val, 0); 382 if (tar->opt_sconv != NULL) 383 ret = ARCHIVE_OK; 384 else 385 ret = ARCHIVE_FATAL; 386 } 387 return (ret); 388 } 389 390 /* Note: The "warn" return is just to inform the options 391 * supervisor that we didn't handle it. It will generate 392 * a suitable error if no one used this option. */ 393 return (ARCHIVE_WARN); 394} 395 396/* utility function- this exists to centralize the logic of tracking 397 * how much unconsumed data we have floating around, and to consume 398 * anything outstanding since we're going to do read_aheads 399 */ 400static void 401tar_flush_unconsumed(struct archive_read *a, size_t *unconsumed) 402{ 403 if (*unconsumed) { 404/* 405 void *data = (void *)__archive_read_ahead(a, *unconsumed, NULL); 406 * this block of code is to poison claimed unconsumed space, ensuring 407 * things break if it is in use still. 408 * currently it WILL break things, so enable it only for debugging this issue 409 if (data) { 410 memset(data, 0xff, *unconsumed); 411 } 412*/ 413 __archive_read_consume(a, *unconsumed); 414 *unconsumed = 0; 415 } 416} 417 418/* 419 * The function invoked by archive_read_next_header(). This 420 * just sets up a few things and then calls the internal 421 * tar_read_header() function below. 422 */ 423static int 424archive_read_format_tar_read_header(struct archive_read *a, 425 struct archive_entry *entry) 426{ 427 /* 428 * When converting tar archives to cpio archives, it is 429 * essential that each distinct file have a distinct inode 430 * number. To simplify this, we keep a static count here to 431 * assign fake dev/inode numbers to each tar entry. Note that 432 * pax format archives may overwrite this with something more 433 * useful. 434 * 435 * Ideally, we would track every file read from the archive so 436 * that we could assign the same dev/ino pair to hardlinks, 437 * but the memory required to store a complete lookup table is 438 * probably not worthwhile just to support the relatively 439 * obscure tar->cpio conversion case. 440 */ 441 static int default_inode; 442 static int default_dev; 443 struct tar *tar; 444 const char *p; 445 int r; 446 size_t l, unconsumed = 0; 447 448 /* Assign default device/inode values. */ 449 archive_entry_set_dev(entry, 1 + default_dev); /* Don't use zero. */ 450 archive_entry_set_ino(entry, ++default_inode); /* Don't use zero. */ 451 /* Limit generated st_ino number to 16 bits. */ 452 if (default_inode >= 0xffff) { 453 ++default_dev; 454 default_inode = 0; 455 } 456 457 tar = (struct tar *)(a->format->data); 458 tar->entry_offset = 0; 459 gnu_clear_sparse_list(tar); 460 tar->realsize = -1; /* Mark this as "unset" */ 461 462 /* Setup default string conversion. */ 463 tar->sconv = tar->opt_sconv; 464 if (tar->sconv == NULL) { 465 if (!tar->init_default_conversion) { 466 tar->sconv_default = 467 archive_string_default_conversion_for_read(&(a->archive)); 468 tar->init_default_conversion = 1; 469 } 470 tar->sconv = tar->sconv_default; 471 } 472 473 r = tar_read_header(a, tar, entry, &unconsumed); 474 475 tar_flush_unconsumed(a, &unconsumed); 476 477 /* 478 * "non-sparse" files are really just sparse files with 479 * a single block. 480 */ 481 if (tar->sparse_list == NULL) { 482 if (gnu_add_sparse_entry(a, tar, 0, tar->entry_bytes_remaining) 483 != ARCHIVE_OK) 484 return (ARCHIVE_FATAL); 485 } else { 486 struct sparse_block *sb; 487 488 for (sb = tar->sparse_list; sb != NULL; sb = sb->next) { 489 if (!sb->hole) 490 archive_entry_sparse_add_entry(entry, 491 sb->offset, sb->remaining); 492 } 493 } 494 495 if (r == ARCHIVE_OK) { 496 /* 497 * "Regular" entry with trailing '/' is really 498 * directory: This is needed for certain old tar 499 * variants and even for some broken newer ones. 500 */ 501 const wchar_t *wp; 502 wp = archive_entry_pathname_w(entry); 503 if (wp != NULL) { 504 l = wcslen(wp); 505 if (archive_entry_filetype(entry) == AE_IFREG 506 && wp[l-1] == L'/') 507 archive_entry_set_filetype(entry, AE_IFDIR); 508 } else { 509 p = archive_entry_pathname(entry); 510 if (p == NULL) 511 return (ARCHIVE_FAILED); 512 l = strlen(p); 513 if (archive_entry_filetype(entry) == AE_IFREG 514 && p[l-1] == '/') 515 archive_entry_set_filetype(entry, AE_IFDIR); 516 } 517 } 518 return (r); 519} 520 521static int 522archive_read_format_tar_read_data(struct archive_read *a, 523 const void **buff, size_t *size, int64_t *offset) 524{ 525 ssize_t bytes_read; 526 struct tar *tar; 527 struct sparse_block *p; 528 529 tar = (struct tar *)(a->format->data); 530 531 for (;;) { 532 /* Remove exhausted entries from sparse list. */ 533 while (tar->sparse_list != NULL && 534 tar->sparse_list->remaining == 0) { 535 p = tar->sparse_list; 536 tar->sparse_list = p->next; 537 free(p); 538 } 539 540 if (tar->entry_bytes_unconsumed) { 541 __archive_read_consume(a, tar->entry_bytes_unconsumed); 542 tar->entry_bytes_unconsumed = 0; 543 } 544 545 /* If we're at end of file, return EOF. */ 546 if (tar->sparse_list == NULL || 547 tar->entry_bytes_remaining == 0) { 548 if (__archive_read_consume(a, tar->entry_padding) < 0) 549 return (ARCHIVE_FATAL); 550 tar->entry_padding = 0; 551 *buff = NULL; 552 *size = 0; 553 *offset = tar->realsize; 554 return (ARCHIVE_EOF); 555 } 556 557 *buff = __archive_read_ahead(a, 1, &bytes_read); 558 if (bytes_read < 0) 559 return (ARCHIVE_FATAL); 560 if (*buff == NULL) { 561 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 562 "Truncated tar archive"); 563 return (ARCHIVE_FATAL); 564 } 565 if (bytes_read > tar->entry_bytes_remaining) 566 bytes_read = (ssize_t)tar->entry_bytes_remaining; 567 /* Don't read more than is available in the 568 * current sparse block. */ 569 if (tar->sparse_list->remaining < bytes_read) 570 bytes_read = (ssize_t)tar->sparse_list->remaining; 571 *size = bytes_read; 572 *offset = tar->sparse_list->offset; 573 tar->sparse_list->remaining -= bytes_read; 574 tar->sparse_list->offset += bytes_read; 575 tar->entry_bytes_remaining -= bytes_read; 576 tar->entry_bytes_unconsumed = bytes_read; 577 578 if (!tar->sparse_list->hole) 579 return (ARCHIVE_OK); 580 /* Current is hole data and skip this. */ 581 } 582} 583 584static int 585archive_read_format_tar_skip(struct archive_read *a) 586{ 587 int64_t bytes_skipped; 588 int64_t request; 589 struct sparse_block *p; 590 struct tar* tar; 591 592 tar = (struct tar *)(a->format->data); 593 594 /* Do not consume the hole of a sparse file. */ 595 request = 0; 596 for (p = tar->sparse_list; p != NULL; p = p->next) { 597 if (!p->hole) 598 request += p->remaining; 599 } 600 if (request > tar->entry_bytes_remaining) 601 request = tar->entry_bytes_remaining; 602 request += tar->entry_padding + tar->entry_bytes_unconsumed; 603 604 bytes_skipped = __archive_read_consume(a, request); 605 if (bytes_skipped < 0) 606 return (ARCHIVE_FATAL); 607 608 tar->entry_bytes_remaining = 0; 609 tar->entry_bytes_unconsumed = 0; 610 tar->entry_padding = 0; 611 612 /* Free the sparse list. */ 613 gnu_clear_sparse_list(tar); 614 615 return (ARCHIVE_OK); 616} 617 618/* 619 * This function recursively interprets all of the headers associated 620 * with a single entry. 621 */ 622static int 623tar_read_header(struct archive_read *a, struct tar *tar, 624 struct archive_entry *entry, size_t *unconsumed) 625{ 626 ssize_t bytes; 627 int err; 628 const char *h; 629 const struct archive_entry_header_ustar *header; 630 const struct archive_entry_header_gnutar *gnuheader; 631 632 tar_flush_unconsumed(a, unconsumed); 633 634 /* Read 512-byte header record */ 635 h = __archive_read_ahead(a, 512, &bytes); 636 if (bytes < 0) 637 return ((int)bytes); 638 if (bytes == 0) { /* EOF at a block boundary. */ 639 /* Some writers do omit the block of nulls. <sigh> */ 640 return (ARCHIVE_EOF); 641 } 642 if (bytes < 512) { /* Short block at EOF; this is bad. */ 643 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 644 "Truncated tar archive"); 645 return (ARCHIVE_FATAL); 646 } 647 *unconsumed = 512; 648 649 /* Check for end-of-archive mark. */ 650 if (h[0] == 0 && archive_block_is_null(h)) { 651 /* Try to consume a second all-null record, as well. */ 652 tar_flush_unconsumed(a, unconsumed); 653 h = __archive_read_ahead(a, 512, NULL); 654 if (h != NULL) 655 __archive_read_consume(a, 512); 656 archive_clear_error(&a->archive); 657 if (a->archive.archive_format_name == NULL) { 658 a->archive.archive_format = ARCHIVE_FORMAT_TAR; 659 a->archive.archive_format_name = "tar"; 660 } 661 return (ARCHIVE_EOF); 662 } 663 664 /* 665 * Note: If the checksum fails and we return ARCHIVE_RETRY, 666 * then the client is likely to just retry. This is a very 667 * crude way to search for the next valid header! 668 * 669 * TODO: Improve this by implementing a real header scan. 670 */ 671 if (!checksum(a, h)) { 672 tar_flush_unconsumed(a, unconsumed); 673 archive_set_error(&a->archive, EINVAL, "Damaged tar archive"); 674 return (ARCHIVE_RETRY); /* Retryable: Invalid header */ 675 } 676 677 if (++tar->header_recursion_depth > 32) { 678 tar_flush_unconsumed(a, unconsumed); 679 archive_set_error(&a->archive, EINVAL, "Too many special headers"); 680 return (ARCHIVE_WARN); 681 } 682 683 /* Determine the format variant. */ 684 header = (const struct archive_entry_header_ustar *)h; 685 686 switch(header->typeflag[0]) { 687 case 'A': /* Solaris tar ACL */ 688 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE; 689 a->archive.archive_format_name = "Solaris tar"; 690 err = header_Solaris_ACL(a, tar, entry, h, unconsumed); 691 break; 692 case 'g': /* POSIX-standard 'g' header. */ 693 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE; 694 a->archive.archive_format_name = "POSIX pax interchange format"; 695 err = header_pax_global(a, tar, entry, h, unconsumed); 696 break; 697 case 'K': /* Long link name (GNU tar, others) */ 698 err = header_longlink(a, tar, entry, h, unconsumed); 699 break; 700 case 'L': /* Long filename (GNU tar, others) */ 701 err = header_longname(a, tar, entry, h, unconsumed); 702 break; 703 case 'V': /* GNU volume header */ 704 err = header_volume(a, tar, entry, h, unconsumed); 705 break; 706 case 'X': /* Used by SUN tar; same as 'x'. */ 707 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE; 708 a->archive.archive_format_name = 709 "POSIX pax interchange format (Sun variant)"; 710 err = header_pax_extensions(a, tar, entry, h, unconsumed); 711 break; 712 case 'x': /* POSIX-standard 'x' header. */ 713 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE; 714 a->archive.archive_format_name = "POSIX pax interchange format"; 715 err = header_pax_extensions(a, tar, entry, h, unconsumed); 716 break; 717 default: 718 gnuheader = (const struct archive_entry_header_gnutar *)h; 719 if (memcmp(gnuheader->magic, "ustar \0", 8) == 0) { 720 a->archive.archive_format = ARCHIVE_FORMAT_TAR_GNUTAR; 721 a->archive.archive_format_name = "GNU tar format"; 722 err = header_gnutar(a, tar, entry, h, unconsumed); 723 } else if (memcmp(header->magic, "ustar", 5) == 0) { 724 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) { 725 a->archive.archive_format = ARCHIVE_FORMAT_TAR_USTAR; 726 a->archive.archive_format_name = "POSIX ustar format"; 727 } 728 err = header_ustar(a, tar, entry, h); 729 } else { 730 a->archive.archive_format = ARCHIVE_FORMAT_TAR; 731 a->archive.archive_format_name = "tar (non-POSIX)"; 732 err = header_old_tar(a, tar, entry, h); 733 } 734 } 735 if (err == ARCHIVE_FATAL) 736 return (err); 737 738 tar_flush_unconsumed(a, unconsumed); 739 740 h = NULL; 741 header = NULL; 742 743 --tar->header_recursion_depth; 744 /* Yuck. Apple's design here ends up storing long pathname 745 * extensions for both the AppleDouble extension entry and the 746 * regular entry. 747 */ 748 /* TODO: Should this be disabled on non-Mac platforms? */ 749 if ((err == ARCHIVE_WARN || err == ARCHIVE_OK) && 750 tar->header_recursion_depth == 0) { 751 int err2 = read_mac_metadata_blob(a, tar, entry, h, unconsumed); 752 if (err2 < err) 753 err = err2; 754 } 755 756 /* We return warnings or success as-is. Anything else is fatal. */ 757 if (err == ARCHIVE_WARN || err == ARCHIVE_OK) { 758 if (tar->sparse_gnu_pending) { 759 if (tar->sparse_gnu_major == 1 && 760 tar->sparse_gnu_minor == 0) { 761 ssize_t bytes_read; 762 763 tar->sparse_gnu_pending = 0; 764 /* Read initial sparse map. */ 765 bytes_read = gnu_sparse_10_read(a, tar, unconsumed); 766 tar->entry_bytes_remaining -= bytes_read; 767 if (bytes_read < 0) 768 return ((int)bytes_read); 769 } else { 770 archive_set_error(&a->archive, 771 ARCHIVE_ERRNO_MISC, 772 "Unrecognized GNU sparse file format"); 773 return (ARCHIVE_WARN); 774 } 775 tar->sparse_gnu_pending = 0; 776 } 777 return (err); 778 } 779 if (err == ARCHIVE_EOF) 780 /* EOF when recursively reading a header is bad. */ 781 archive_set_error(&a->archive, EINVAL, "Damaged tar archive"); 782 return (ARCHIVE_FATAL); 783} 784 785/* 786 * Return true if block checksum is correct. 787 */ 788static int 789checksum(struct archive_read *a, const void *h) 790{ 791 const unsigned char *bytes; 792 const struct archive_entry_header_ustar *header; 793 int check, i, sum; 794 795 (void)a; /* UNUSED */ 796 bytes = (const unsigned char *)h; 797 header = (const struct archive_entry_header_ustar *)h; 798 799 /* 800 * Test the checksum. Note that POSIX specifies _unsigned_ 801 * bytes for this calculation. 802 */ 803 sum = (int)tar_atol(header->checksum, sizeof(header->checksum)); 804 check = 0; 805 for (i = 0; i < 148; i++) 806 check += (unsigned char)bytes[i]; 807 for (; i < 156; i++) 808 check += 32; 809 for (; i < 512; i++) 810 check += (unsigned char)bytes[i]; 811 if (sum == check) 812 return (1); 813 814 /* 815 * Repeat test with _signed_ bytes, just in case this archive 816 * was created by an old BSD, Solaris, or HP-UX tar with a 817 * broken checksum calculation. 818 */ 819 check = 0; 820 for (i = 0; i < 148; i++) 821 check += (signed char)bytes[i]; 822 for (; i < 156; i++) 823 check += 32; 824 for (; i < 512; i++) 825 check += (signed char)bytes[i]; 826 if (sum == check) 827 return (1); 828 829 return (0); 830} 831 832/* 833 * Return true if this block contains only nulls. 834 */ 835static int 836archive_block_is_null(const char *p) 837{ 838 unsigned i; 839 840 for (i = 0; i < 512; i++) 841 if (*p++) 842 return (0); 843 return (1); 844} 845 846/* 847 * Interpret 'A' Solaris ACL header 848 */ 849static int 850header_Solaris_ACL(struct archive_read *a, struct tar *tar, 851 struct archive_entry *entry, const void *h, size_t *unconsumed) 852{ 853 const struct archive_entry_header_ustar *header; 854 size_t size; 855 int err; 856 int64_t type; 857 char *acl, *p; 858 859 /* 860 * read_body_to_string adds a NUL terminator, but we need a little 861 * more to make sure that we don't overrun acl_text later. 862 */ 863 header = (const struct archive_entry_header_ustar *)h; 864 size = (size_t)tar_atol(header->size, sizeof(header->size)); 865 err = read_body_to_string(a, tar, &(tar->acl_text), h, unconsumed); 866 if (err != ARCHIVE_OK) 867 return (err); 868 869 /* Recursively read next header */ 870 err = tar_read_header(a, tar, entry, unconsumed); 871 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN)) 872 return (err); 873 874 /* TODO: Examine the first characters to see if this 875 * is an AIX ACL descriptor. We'll likely never support 876 * them, but it would be polite to recognize and warn when 877 * we do see them. */ 878 879 /* Leading octal number indicates ACL type and number of entries. */ 880 p = acl = tar->acl_text.s; 881 type = 0; 882 while (*p != '\0' && p < acl + size) { 883 if (*p < '0' || *p > '7') { 884 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 885 "Malformed Solaris ACL attribute (invalid digit)"); 886 return(ARCHIVE_WARN); 887 } 888 type <<= 3; 889 type += *p - '0'; 890 if (type > 077777777) { 891 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 892 "Malformed Solaris ACL attribute (count too large)"); 893 return (ARCHIVE_WARN); 894 } 895 p++; 896 } 897 switch ((int)type & ~0777777) { 898 case 01000000: 899 /* POSIX.1e ACL */ 900 break; 901 case 03000000: 902 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 903 "Solaris NFSv4 ACLs not supported"); 904 return (ARCHIVE_WARN); 905 default: 906 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 907 "Malformed Solaris ACL attribute (unsupported type %o)", 908 (int)type); 909 return (ARCHIVE_WARN); 910 } 911 p++; 912 913 if (p >= acl + size) { 914 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 915 "Malformed Solaris ACL attribute (body overflow)"); 916 return(ARCHIVE_WARN); 917 } 918 919 /* ACL text is null-terminated; find the end. */ 920 size -= (p - acl); 921 acl = p; 922 923 while (*p != '\0' && p < acl + size) 924 p++; 925 926 if (tar->sconv_acl == NULL) { 927 tar->sconv_acl = archive_string_conversion_from_charset( 928 &(a->archive), "UTF-8", 1); 929 if (tar->sconv_acl == NULL) 930 return (ARCHIVE_FATAL); 931 } 932 archive_strncpy(&(tar->localname), acl, p - acl); 933 err = archive_acl_parse_l(archive_entry_acl(entry), 934 tar->localname.s, ARCHIVE_ENTRY_ACL_TYPE_ACCESS, tar->sconv_acl); 935 if (err != ARCHIVE_OK) { 936 if (errno == ENOMEM) { 937 archive_set_error(&a->archive, ENOMEM, 938 "Can't allocate memory for ACL"); 939 } else 940 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 941 "Malformed Solaris ACL attribute (unparsable)"); 942 } 943 return (err); 944} 945 946/* 947 * Interpret 'K' long linkname header. 948 */ 949static int 950header_longlink(struct archive_read *a, struct tar *tar, 951 struct archive_entry *entry, const void *h, size_t *unconsumed) 952{ 953 int err; 954 955 err = read_body_to_string(a, tar, &(tar->longlink), h, unconsumed); 956 if (err != ARCHIVE_OK) 957 return (err); 958 err = tar_read_header(a, tar, entry, unconsumed); 959 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN)) 960 return (err); 961 /* Set symlink if symlink already set, else hardlink. */ 962 archive_entry_copy_link(entry, tar->longlink.s); 963 return (ARCHIVE_OK); 964} 965 966static int 967set_conversion_failed_error(struct archive_read *a, 968 struct archive_string_conv *sconv, const char *name) 969{ 970 if (errno == ENOMEM) { 971 archive_set_error(&a->archive, ENOMEM, 972 "Can't allocate memory for %s", name); 973 return (ARCHIVE_FATAL); 974 } 975 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 976 "%s can't be converted from %s to current locale.", 977 name, archive_string_conversion_charset_name(sconv)); 978 return (ARCHIVE_WARN); 979} 980 981/* 982 * Interpret 'L' long filename header. 983 */ 984static int 985header_longname(struct archive_read *a, struct tar *tar, 986 struct archive_entry *entry, const void *h, size_t *unconsumed) 987{ 988 int err; 989 990 err = read_body_to_string(a, tar, &(tar->longname), h, unconsumed); 991 if (err != ARCHIVE_OK) 992 return (err); 993 /* Read and parse "real" header, then override name. */ 994 err = tar_read_header(a, tar, entry, unconsumed); 995 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN)) 996 return (err); 997 if (archive_entry_copy_pathname_l(entry, tar->longname.s, 998 archive_strlen(&(tar->longname)), tar->sconv) != 0) 999 err = set_conversion_failed_error(a, tar->sconv, "Pathname"); 1000 return (err); 1001} 1002 1003 1004/* 1005 * Interpret 'V' GNU tar volume header. 1006 */ 1007static int 1008header_volume(struct archive_read *a, struct tar *tar, 1009 struct archive_entry *entry, const void *h, size_t *unconsumed) 1010{ 1011 (void)h; 1012 1013 /* Just skip this and read the next header. */ 1014 return (tar_read_header(a, tar, entry, unconsumed)); 1015} 1016 1017/* 1018 * Read body of an archive entry into an archive_string object. 1019 */ 1020static int 1021read_body_to_string(struct archive_read *a, struct tar *tar, 1022 struct archive_string *as, const void *h, size_t *unconsumed) 1023{ 1024 int64_t size; 1025 const struct archive_entry_header_ustar *header; 1026 const void *src; 1027 1028 (void)tar; /* UNUSED */ 1029 header = (const struct archive_entry_header_ustar *)h; 1030 size = tar_atol(header->size, sizeof(header->size)); 1031 if ((size > 1048576) || (size < 0)) { 1032 archive_set_error(&a->archive, EINVAL, 1033 "Special header too large"); 1034 return (ARCHIVE_FATAL); 1035 } 1036 1037 /* Fail if we can't make our buffer big enough. */ 1038 if (archive_string_ensure(as, (size_t)size+1) == NULL) { 1039 archive_set_error(&a->archive, ENOMEM, 1040 "No memory"); 1041 return (ARCHIVE_FATAL); 1042 } 1043 1044 tar_flush_unconsumed(a, unconsumed); 1045 1046 /* Read the body into the string. */ 1047 *unconsumed = (size_t)((size + 511) & ~ 511); 1048 src = __archive_read_ahead(a, *unconsumed, NULL); 1049 if (src == NULL) { 1050 *unconsumed = 0; 1051 return (ARCHIVE_FATAL); 1052 } 1053 memcpy(as->s, src, (size_t)size); 1054 as->s[size] = '\0'; 1055 as->length = (size_t)size; 1056 return (ARCHIVE_OK); 1057} 1058 1059/* 1060 * Parse out common header elements. 1061 * 1062 * This would be the same as header_old_tar, except that the 1063 * filename is handled slightly differently for old and POSIX 1064 * entries (POSIX entries support a 'prefix'). This factoring 1065 * allows header_old_tar and header_ustar 1066 * to handle filenames differently, while still putting most of the 1067 * common parsing into one place. 1068 */ 1069static int 1070header_common(struct archive_read *a, struct tar *tar, 1071 struct archive_entry *entry, const void *h) 1072{ 1073 const struct archive_entry_header_ustar *header; 1074 char tartype; 1075 int err = ARCHIVE_OK; 1076 1077 header = (const struct archive_entry_header_ustar *)h; 1078 if (header->linkname[0]) 1079 archive_strncpy(&(tar->entry_linkpath), 1080 header->linkname, sizeof(header->linkname)); 1081 else 1082 archive_string_empty(&(tar->entry_linkpath)); 1083 1084 /* Parse out the numeric fields (all are octal) */ 1085 archive_entry_set_mode(entry, 1086 (mode_t)tar_atol(header->mode, sizeof(header->mode))); 1087 archive_entry_set_uid(entry, tar_atol(header->uid, sizeof(header->uid))); 1088 archive_entry_set_gid(entry, tar_atol(header->gid, sizeof(header->gid))); 1089 tar->entry_bytes_remaining = tar_atol(header->size, sizeof(header->size)); 1090 if (tar->entry_bytes_remaining < 0) { 1091 tar->entry_bytes_remaining = 0; 1092 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1093 "Tar entry has negative size?"); 1094 err = ARCHIVE_WARN; 1095 } 1096 tar->realsize = tar->entry_bytes_remaining; 1097 archive_entry_set_size(entry, tar->entry_bytes_remaining); 1098 archive_entry_set_mtime(entry, tar_atol(header->mtime, sizeof(header->mtime)), 0); 1099 1100 /* Handle the tar type flag appropriately. */ 1101 tartype = header->typeflag[0]; 1102 1103 switch (tartype) { 1104 case '1': /* Hard link */ 1105 if (archive_entry_copy_hardlink_l(entry, tar->entry_linkpath.s, 1106 archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) { 1107 err = set_conversion_failed_error(a, tar->sconv, 1108 "Linkname"); 1109 if (err == ARCHIVE_FATAL) 1110 return (err); 1111 } 1112 /* 1113 * The following may seem odd, but: Technically, tar 1114 * does not store the file type for a "hard link" 1115 * entry, only the fact that it is a hard link. So, I 1116 * leave the type zero normally. But, pax interchange 1117 * format allows hard links to have data, which 1118 * implies that the underlying entry is a regular 1119 * file. 1120 */ 1121 if (archive_entry_size(entry) > 0) 1122 archive_entry_set_filetype(entry, AE_IFREG); 1123 1124 /* 1125 * A tricky point: Traditionally, tar readers have 1126 * ignored the size field when reading hardlink 1127 * entries, and some writers put non-zero sizes even 1128 * though the body is empty. POSIX blessed this 1129 * convention in the 1988 standard, but broke with 1130 * this tradition in 2001 by permitting hardlink 1131 * entries to store valid bodies in pax interchange 1132 * format, but not in ustar format. Since there is no 1133 * hard and fast way to distinguish pax interchange 1134 * from earlier archives (the 'x' and 'g' entries are 1135 * optional, after all), we need a heuristic. 1136 */ 1137 if (archive_entry_size(entry) == 0) { 1138 /* If the size is already zero, we're done. */ 1139 } else if (a->archive.archive_format 1140 == ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) { 1141 /* Definitely pax extended; must obey hardlink size. */ 1142 } else if (a->archive.archive_format == ARCHIVE_FORMAT_TAR 1143 || a->archive.archive_format == ARCHIVE_FORMAT_TAR_GNUTAR) 1144 { 1145 /* Old-style or GNU tar: we must ignore the size. */ 1146 archive_entry_set_size(entry, 0); 1147 tar->entry_bytes_remaining = 0; 1148 } else if (archive_read_format_tar_bid(a, 50) > 50) { 1149 /* 1150 * We don't know if it's pax: If the bid 1151 * function sees a valid ustar header 1152 * immediately following, then let's ignore 1153 * the hardlink size. 1154 */ 1155 archive_entry_set_size(entry, 0); 1156 tar->entry_bytes_remaining = 0; 1157 } 1158 /* 1159 * TODO: There are still two cases I'd like to handle: 1160 * = a ustar non-pax archive with a hardlink entry at 1161 * end-of-archive. (Look for block of nulls following?) 1162 * = a pax archive that has not seen any pax headers 1163 * and has an entry which is a hardlink entry storing 1164 * a body containing an uncompressed tar archive. 1165 * The first is worth addressing; I don't see any reliable 1166 * way to deal with the second possibility. 1167 */ 1168 break; 1169 case '2': /* Symlink */ 1170 archive_entry_set_filetype(entry, AE_IFLNK); 1171 archive_entry_set_size(entry, 0); 1172 tar->entry_bytes_remaining = 0; 1173 if (archive_entry_copy_symlink_l(entry, tar->entry_linkpath.s, 1174 archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) { 1175 err = set_conversion_failed_error(a, tar->sconv, 1176 "Linkname"); 1177 if (err == ARCHIVE_FATAL) 1178 return (err); 1179 } 1180 break; 1181 case '3': /* Character device */ 1182 archive_entry_set_filetype(entry, AE_IFCHR); 1183 archive_entry_set_size(entry, 0); 1184 tar->entry_bytes_remaining = 0; 1185 break; 1186 case '4': /* Block device */ 1187 archive_entry_set_filetype(entry, AE_IFBLK); 1188 archive_entry_set_size(entry, 0); 1189 tar->entry_bytes_remaining = 0; 1190 break; 1191 case '5': /* Dir */ 1192 archive_entry_set_filetype(entry, AE_IFDIR); 1193 archive_entry_set_size(entry, 0); 1194 tar->entry_bytes_remaining = 0; 1195 break; 1196 case '6': /* FIFO device */ 1197 archive_entry_set_filetype(entry, AE_IFIFO); 1198 archive_entry_set_size(entry, 0); 1199 tar->entry_bytes_remaining = 0; 1200 break; 1201 case 'D': /* GNU incremental directory type */ 1202 /* 1203 * No special handling is actually required here. 1204 * It might be nice someday to preprocess the file list and 1205 * provide it to the client, though. 1206 */ 1207 archive_entry_set_filetype(entry, AE_IFDIR); 1208 break; 1209 case 'M': /* GNU "Multi-volume" (remainder of file from last archive)*/ 1210 /* 1211 * As far as I can tell, this is just like a regular file 1212 * entry, except that the contents should be _appended_ to 1213 * the indicated file at the indicated offset. This may 1214 * require some API work to fully support. 1215 */ 1216 break; 1217 case 'N': /* Old GNU "long filename" entry. */ 1218 /* The body of this entry is a script for renaming 1219 * previously-extracted entries. Ugh. It will never 1220 * be supported by libarchive. */ 1221 archive_entry_set_filetype(entry, AE_IFREG); 1222 break; 1223 case 'S': /* GNU sparse files */ 1224 /* 1225 * Sparse files are really just regular files with 1226 * sparse information in the extended area. 1227 */ 1228 /* FALLTHROUGH */ 1229 default: /* Regular file and non-standard types */ 1230 /* 1231 * Per POSIX: non-recognized types should always be 1232 * treated as regular files. 1233 */ 1234 archive_entry_set_filetype(entry, AE_IFREG); 1235 break; 1236 } 1237 return (err); 1238} 1239 1240/* 1241 * Parse out header elements for "old-style" tar archives. 1242 */ 1243static int 1244header_old_tar(struct archive_read *a, struct tar *tar, 1245 struct archive_entry *entry, const void *h) 1246{ 1247 const struct archive_entry_header_ustar *header; 1248 int err = ARCHIVE_OK, err2; 1249 1250 /* Copy filename over (to ensure null termination). */ 1251 header = (const struct archive_entry_header_ustar *)h; 1252 if (archive_entry_copy_pathname_l(entry, 1253 header->name, sizeof(header->name), tar->sconv) != 0) { 1254 err = set_conversion_failed_error(a, tar->sconv, "Pathname"); 1255 if (err == ARCHIVE_FATAL) 1256 return (err); 1257 } 1258 1259 /* Grab rest of common fields */ 1260 err2 = header_common(a, tar, entry, h); 1261 if (err > err2) 1262 err = err2; 1263 1264 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining); 1265 return (err); 1266} 1267 1268/* 1269 * Read a Mac AppleDouble-encoded blob of file metadata, 1270 * if there is one. 1271 */ 1272static int 1273read_mac_metadata_blob(struct archive_read *a, struct tar *tar, 1274 struct archive_entry *entry, const void *h, size_t *unconsumed) 1275{ 1276 int64_t size; 1277 const void *data; 1278 const char *p, *name; 1279 const wchar_t *wp, *wname; 1280 1281 (void)h; /* UNUSED */ 1282 1283 wname = wp = archive_entry_pathname_w(entry); 1284 if (wp != NULL) { 1285 /* Find the last path element. */ 1286 for (; *wp != L'\0'; ++wp) { 1287 if (wp[0] == '/' && wp[1] != L'\0') 1288 wname = wp + 1; 1289 } 1290 /* 1291 * If last path element starts with "._", then 1292 * this is a Mac extension. 1293 */ 1294 if (wname[0] != L'.' || wname[1] != L'_' || wname[2] == L'\0') 1295 return ARCHIVE_OK; 1296 } else { 1297 /* Find the last path element. */ 1298 name = p = archive_entry_pathname(entry); 1299 if (p == NULL) 1300 return (ARCHIVE_FAILED); 1301 for (; *p != '\0'; ++p) { 1302 if (p[0] == '/' && p[1] != '\0') 1303 name = p + 1; 1304 } 1305 /* 1306 * If last path element starts with "._", then 1307 * this is a Mac extension. 1308 */ 1309 if (name[0] != '.' || name[1] != '_' || name[2] == '\0') 1310 return ARCHIVE_OK; 1311 } 1312 1313 /* Read the body as a Mac OS metadata blob. */ 1314 size = archive_entry_size(entry); 1315 1316 /* 1317 * TODO: Look beyond the body here to peek at the next header. 1318 * If it's a regular header (not an extension header) 1319 * that has the wrong name, just return the current 1320 * entry as-is, without consuming the body here. 1321 * That would reduce the risk of us mis-identifying 1322 * an ordinary file that just happened to have 1323 * a name starting with "._". 1324 * 1325 * Q: Is the above idea really possible? Even 1326 * when there are GNU or pax extension entries? 1327 */ 1328 data = __archive_read_ahead(a, (size_t)size, NULL); 1329 if (data == NULL) { 1330 *unconsumed = 0; 1331 return (ARCHIVE_FATAL); 1332 } 1333 archive_entry_copy_mac_metadata(entry, data, (size_t)size); 1334 *unconsumed = (size_t)((size + 511) & ~ 511); 1335 tar_flush_unconsumed(a, unconsumed); 1336 return (tar_read_header(a, tar, entry, unconsumed)); 1337} 1338 1339/* 1340 * Parse a file header for a pax extended archive entry. 1341 */ 1342static int 1343header_pax_global(struct archive_read *a, struct tar *tar, 1344 struct archive_entry *entry, const void *h, size_t *unconsumed) 1345{ 1346 int err; 1347 1348 err = read_body_to_string(a, tar, &(tar->pax_global), h, unconsumed); 1349 if (err != ARCHIVE_OK) 1350 return (err); 1351 err = tar_read_header(a, tar, entry, unconsumed); 1352 return (err); 1353} 1354 1355static int 1356header_pax_extensions(struct archive_read *a, struct tar *tar, 1357 struct archive_entry *entry, const void *h, size_t *unconsumed) 1358{ 1359 int err, err2; 1360 1361 err = read_body_to_string(a, tar, &(tar->pax_header), h, unconsumed); 1362 if (err != ARCHIVE_OK) 1363 return (err); 1364 1365 /* Parse the next header. */ 1366 err = tar_read_header(a, tar, entry, unconsumed); 1367 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN)) 1368 return (err); 1369 1370 /* 1371 * TODO: Parse global/default options into 'entry' struct here 1372 * before handling file-specific options. 1373 * 1374 * This design (parse standard header, then overwrite with pax 1375 * extended attribute data) usually works well, but isn't ideal; 1376 * it would be better to parse the pax extended attributes first 1377 * and then skip any fields in the standard header that were 1378 * defined in the pax header. 1379 */ 1380 err2 = pax_header(a, tar, entry, tar->pax_header.s); 1381 err = err_combine(err, err2); 1382 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining); 1383 return (err); 1384} 1385 1386 1387/* 1388 * Parse a file header for a Posix "ustar" archive entry. This also 1389 * handles "pax" or "extended ustar" entries. 1390 */ 1391static int 1392header_ustar(struct archive_read *a, struct tar *tar, 1393 struct archive_entry *entry, const void *h) 1394{ 1395 const struct archive_entry_header_ustar *header; 1396 struct archive_string *as; 1397 int err = ARCHIVE_OK, r; 1398 1399 header = (const struct archive_entry_header_ustar *)h; 1400 1401 /* Copy name into an internal buffer to ensure null-termination. */ 1402 as = &(tar->entry_pathname); 1403 if (header->prefix[0]) { 1404 archive_strncpy(as, header->prefix, sizeof(header->prefix)); 1405 if (as->s[archive_strlen(as) - 1] != '/') 1406 archive_strappend_char(as, '/'); 1407 archive_strncat(as, header->name, sizeof(header->name)); 1408 } else { 1409 archive_strncpy(as, header->name, sizeof(header->name)); 1410 } 1411 if (archive_entry_copy_pathname_l(entry, as->s, archive_strlen(as), 1412 tar->sconv) != 0) { 1413 err = set_conversion_failed_error(a, tar->sconv, "Pathname"); 1414 if (err == ARCHIVE_FATAL) 1415 return (err); 1416 } 1417 1418 /* Handle rest of common fields. */ 1419 r = header_common(a, tar, entry, h); 1420 if (r == ARCHIVE_FATAL) 1421 return (r); 1422 if (r < err) 1423 err = r; 1424 1425 /* Handle POSIX ustar fields. */ 1426 if (archive_entry_copy_uname_l(entry, 1427 header->uname, sizeof(header->uname), tar->sconv) != 0) { 1428 err = set_conversion_failed_error(a, tar->sconv, "Uname"); 1429 if (err == ARCHIVE_FATAL) 1430 return (err); 1431 } 1432 1433 if (archive_entry_copy_gname_l(entry, 1434 header->gname, sizeof(header->gname), tar->sconv) != 0) { 1435 err = set_conversion_failed_error(a, tar->sconv, "Gname"); 1436 if (err == ARCHIVE_FATAL) 1437 return (err); 1438 } 1439 1440 /* Parse out device numbers only for char and block specials. */ 1441 if (header->typeflag[0] == '3' || header->typeflag[0] == '4') { 1442 archive_entry_set_rdevmajor(entry, (dev_t) 1443 tar_atol(header->rdevmajor, sizeof(header->rdevmajor))); 1444 archive_entry_set_rdevminor(entry, (dev_t) 1445 tar_atol(header->rdevminor, sizeof(header->rdevminor))); 1446 } 1447 1448 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining); 1449 1450 return (err); 1451} 1452 1453 1454/* 1455 * Parse the pax extended attributes record. 1456 * 1457 * Returns non-zero if there's an error in the data. 1458 */ 1459static int 1460pax_header(struct archive_read *a, struct tar *tar, 1461 struct archive_entry *entry, char *attr) 1462{ 1463 size_t attr_length, l, line_length; 1464 char *p; 1465 char *key, *value; 1466 struct archive_string *as; 1467 struct archive_string_conv *sconv; 1468 int err, err2; 1469 1470 attr_length = strlen(attr); 1471 tar->pax_hdrcharset_binary = 0; 1472 archive_string_empty(&(tar->entry_gname)); 1473 archive_string_empty(&(tar->entry_linkpath)); 1474 archive_string_empty(&(tar->entry_pathname)); 1475 archive_string_empty(&(tar->entry_pathname_override)); 1476 archive_string_empty(&(tar->entry_uname)); 1477 err = ARCHIVE_OK; 1478 while (attr_length > 0) { 1479 /* Parse decimal length field at start of line. */ 1480 line_length = 0; 1481 l = attr_length; 1482 p = attr; /* Record start of line. */ 1483 while (l>0) { 1484 if (*p == ' ') { 1485 p++; 1486 l--; 1487 break; 1488 } 1489 if (*p < '0' || *p > '9') { 1490 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1491 "Ignoring malformed pax extended attributes"); 1492 return (ARCHIVE_WARN); 1493 } 1494 line_length *= 10; 1495 line_length += *p - '0'; 1496 if (line_length > 999999) { 1497 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1498 "Rejecting pax extended attribute > 1MB"); 1499 return (ARCHIVE_WARN); 1500 } 1501 p++; 1502 l--; 1503 } 1504 1505 /* 1506 * Parsed length must be no bigger than available data, 1507 * at least 1, and the last character of the line must 1508 * be '\n'. 1509 */ 1510 if (line_length > attr_length 1511 || line_length < 1 1512 || attr[line_length - 1] != '\n') 1513 { 1514 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1515 "Ignoring malformed pax extended attribute"); 1516 return (ARCHIVE_WARN); 1517 } 1518 1519 /* Null-terminate the line. */ 1520 attr[line_length - 1] = '\0'; 1521 1522 /* Find end of key and null terminate it. */ 1523 key = p; 1524 if (key[0] == '=') 1525 return (-1); 1526 while (*p && *p != '=') 1527 ++p; 1528 if (*p == '\0') { 1529 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1530 "Invalid pax extended attributes"); 1531 return (ARCHIVE_WARN); 1532 } 1533 *p = '\0'; 1534 1535 /* Identify null-terminated 'value' portion. */ 1536 value = p + 1; 1537 1538 /* Identify this attribute and set it in the entry. */ 1539 err2 = pax_attribute(a, tar, entry, key, value); 1540 if (err2 == ARCHIVE_FATAL) 1541 return (err2); 1542 err = err_combine(err, err2); 1543 1544 /* Skip to next line */ 1545 attr += line_length; 1546 attr_length -= line_length; 1547 } 1548 1549 /* 1550 * PAX format uses UTF-8 as default charset for its metadata 1551 * unless hdrcharset=BINARY is present in its header. 1552 * We apply the charset specified by the hdrcharset option only 1553 * when the hdrcharset attribute(in PAX header) is BINARY because 1554 * we respect the charset described in PAX header and BINARY also 1555 * means that metadata(filename,uname and gname) character-set 1556 * is unknown. 1557 */ 1558 if (tar->pax_hdrcharset_binary) 1559 sconv = tar->opt_sconv; 1560 else { 1561 sconv = archive_string_conversion_from_charset( 1562 &(a->archive), "UTF-8", 1); 1563 if (sconv == NULL) 1564 return (ARCHIVE_FATAL); 1565 if (tar->compat_2x) 1566 archive_string_conversion_set_opt(sconv, 1567 SCONV_SET_OPT_UTF8_LIBARCHIVE2X); 1568 } 1569 1570 if (archive_strlen(&(tar->entry_gname)) > 0) { 1571 if (archive_entry_copy_gname_l(entry, tar->entry_gname.s, 1572 archive_strlen(&(tar->entry_gname)), sconv) != 0) { 1573 err = set_conversion_failed_error(a, sconv, "Gname"); 1574 if (err == ARCHIVE_FATAL) 1575 return (err); 1576 /* Use a converted an original name. */ 1577 archive_entry_copy_gname(entry, tar->entry_gname.s); 1578 } 1579 } 1580 if (archive_strlen(&(tar->entry_linkpath)) > 0) { 1581 if (archive_entry_copy_link_l(entry, tar->entry_linkpath.s, 1582 archive_strlen(&(tar->entry_linkpath)), sconv) != 0) { 1583 err = set_conversion_failed_error(a, sconv, "Linkname"); 1584 if (err == ARCHIVE_FATAL) 1585 return (err); 1586 /* Use a converted an original name. */ 1587 archive_entry_copy_link(entry, tar->entry_linkpath.s); 1588 } 1589 } 1590 /* 1591 * Some extensions (such as the GNU sparse file extensions) 1592 * deliberately store a synthetic name under the regular 'path' 1593 * attribute and the real file name under a different attribute. 1594 * Since we're supposed to not care about the order, we 1595 * have no choice but to store all of the various filenames 1596 * we find and figure it all out afterwards. This is the 1597 * figuring out part. 1598 */ 1599 as = NULL; 1600 if (archive_strlen(&(tar->entry_pathname_override)) > 0) 1601 as = &(tar->entry_pathname_override); 1602 else if (archive_strlen(&(tar->entry_pathname)) > 0) 1603 as = &(tar->entry_pathname); 1604 if (as != NULL) { 1605 if (archive_entry_copy_pathname_l(entry, as->s, 1606 archive_strlen(as), sconv) != 0) { 1607 err = set_conversion_failed_error(a, sconv, "Pathname"); 1608 if (err == ARCHIVE_FATAL) 1609 return (err); 1610 /* Use a converted an original name. */ 1611 archive_entry_copy_pathname(entry, as->s); 1612 } 1613 } 1614 if (archive_strlen(&(tar->entry_uname)) > 0) { 1615 if (archive_entry_copy_uname_l(entry, tar->entry_uname.s, 1616 archive_strlen(&(tar->entry_uname)), sconv) != 0) { 1617 err = set_conversion_failed_error(a, sconv, "Uname"); 1618 if (err == ARCHIVE_FATAL) 1619 return (err); 1620 /* Use a converted an original name. */ 1621 archive_entry_copy_uname(entry, tar->entry_uname.s); 1622 } 1623 } 1624 return (err); 1625} 1626 1627static int 1628pax_attribute_xattr(struct archive_entry *entry, 1629 char *name, char *value) 1630{ 1631 char *name_decoded; 1632 void *value_decoded; 1633 size_t value_len; 1634 1635 if (strlen(name) < 18 || (memcmp(name, "LIBARCHIVE.xattr.", 17)) != 0) 1636 return 3; 1637 1638 name += 17; 1639 1640 /* URL-decode name */ 1641 name_decoded = url_decode(name); 1642 if (name_decoded == NULL) 1643 return 2; 1644 1645 /* Base-64 decode value */ 1646 value_decoded = base64_decode(value, strlen(value), &value_len); 1647 if (value_decoded == NULL) { 1648 free(name_decoded); 1649 return 1; 1650 } 1651 1652 archive_entry_xattr_add_entry(entry, name_decoded, 1653 value_decoded, value_len); 1654 1655 free(name_decoded); 1656 free(value_decoded); 1657 return 0; 1658} 1659 1660/* 1661 * Parse a single key=value attribute. key/value pointers are 1662 * assumed to point into reasonably long-lived storage. 1663 * 1664 * Note that POSIX reserves all-lowercase keywords. Vendor-specific 1665 * extensions should always have keywords of the form "VENDOR.attribute" 1666 * In particular, it's quite feasible to support many different 1667 * vendor extensions here. I'm using "LIBARCHIVE" for extensions 1668 * unique to this library. 1669 * 1670 * Investigate other vendor-specific extensions and see if 1671 * any of them look useful. 1672 */ 1673static int 1674pax_attribute(struct archive_read *a, struct tar *tar, 1675 struct archive_entry *entry, char *key, char *value) 1676{ 1677 int64_t s; 1678 long n; 1679 int err = ARCHIVE_OK, r; 1680 1681#ifndef __FreeBSD__ 1682 if (value == NULL) 1683 value = ""; /* Disable compiler warning; do not pass 1684 * NULL pointer to strlen(). */ 1685#endif 1686 switch (key[0]) { 1687 case 'G': 1688 /* GNU "0.0" sparse pax format. */ 1689 if (strcmp(key, "GNU.sparse.numblocks") == 0) { 1690 tar->sparse_offset = -1; 1691 tar->sparse_numbytes = -1; 1692 tar->sparse_gnu_major = 0; 1693 tar->sparse_gnu_minor = 0; 1694 } 1695 if (strcmp(key, "GNU.sparse.offset") == 0) { 1696 tar->sparse_offset = tar_atol10(value, strlen(value)); 1697 if (tar->sparse_numbytes != -1) { 1698 if (gnu_add_sparse_entry(a, tar, 1699 tar->sparse_offset, tar->sparse_numbytes) 1700 != ARCHIVE_OK) 1701 return (ARCHIVE_FATAL); 1702 tar->sparse_offset = -1; 1703 tar->sparse_numbytes = -1; 1704 } 1705 } 1706 if (strcmp(key, "GNU.sparse.numbytes") == 0) { 1707 tar->sparse_numbytes = tar_atol10(value, strlen(value)); 1708 if (tar->sparse_numbytes != -1) { 1709 if (gnu_add_sparse_entry(a, tar, 1710 tar->sparse_offset, tar->sparse_numbytes) 1711 != ARCHIVE_OK) 1712 return (ARCHIVE_FATAL); 1713 tar->sparse_offset = -1; 1714 tar->sparse_numbytes = -1; 1715 } 1716 } 1717 if (strcmp(key, "GNU.sparse.size") == 0) { 1718 tar->realsize = tar_atol10(value, strlen(value)); 1719 archive_entry_set_size(entry, tar->realsize); 1720 } 1721 1722 /* GNU "0.1" sparse pax format. */ 1723 if (strcmp(key, "GNU.sparse.map") == 0) { 1724 tar->sparse_gnu_major = 0; 1725 tar->sparse_gnu_minor = 1; 1726 if (gnu_sparse_01_parse(a, tar, value) != ARCHIVE_OK) 1727 return (ARCHIVE_WARN); 1728 } 1729 1730 /* GNU "1.0" sparse pax format */ 1731 if (strcmp(key, "GNU.sparse.major") == 0) { 1732 tar->sparse_gnu_major = (int)tar_atol10(value, strlen(value)); 1733 tar->sparse_gnu_pending = 1; 1734 } 1735 if (strcmp(key, "GNU.sparse.minor") == 0) { 1736 tar->sparse_gnu_minor = (int)tar_atol10(value, strlen(value)); 1737 tar->sparse_gnu_pending = 1; 1738 } 1739 if (strcmp(key, "GNU.sparse.name") == 0) { 1740 /* 1741 * The real filename; when storing sparse 1742 * files, GNU tar puts a synthesized name into 1743 * the regular 'path' attribute in an attempt 1744 * to limit confusion. ;-) 1745 */ 1746 archive_strcpy(&(tar->entry_pathname_override), value); 1747 } 1748 if (strcmp(key, "GNU.sparse.realsize") == 0) { 1749 tar->realsize = tar_atol10(value, strlen(value)); 1750 archive_entry_set_size(entry, tar->realsize); 1751 } 1752 break; 1753 case 'L': 1754 /* Our extensions */ 1755/* TODO: Handle arbitrary extended attributes... */ 1756/* 1757 if (strcmp(key, "LIBARCHIVE.xxxxxxx") == 0) 1758 archive_entry_set_xxxxxx(entry, value); 1759*/ 1760 if (strcmp(key, "LIBARCHIVE.creationtime") == 0) { 1761 pax_time(value, &s, &n); 1762 archive_entry_set_birthtime(entry, s, n); 1763 } 1764 if (memcmp(key, "LIBARCHIVE.xattr.", 17) == 0) 1765 pax_attribute_xattr(entry, key, value); 1766 break; 1767 case 'S': 1768 /* We support some keys used by the "star" archiver */ 1769 if (strcmp(key, "SCHILY.acl.access") == 0) { 1770 if (tar->sconv_acl == NULL) { 1771 tar->sconv_acl = 1772 archive_string_conversion_from_charset( 1773 &(a->archive), "UTF-8", 1); 1774 if (tar->sconv_acl == NULL) 1775 return (ARCHIVE_FATAL); 1776 } 1777 1778 r = archive_acl_parse_l(archive_entry_acl(entry), 1779 value, ARCHIVE_ENTRY_ACL_TYPE_ACCESS, 1780 tar->sconv_acl); 1781 if (r != ARCHIVE_OK) { 1782 err = r; 1783 if (err == ARCHIVE_FATAL) { 1784 archive_set_error(&a->archive, ENOMEM, 1785 "Can't allocate memory for " 1786 "SCHILY.acl.access"); 1787 return (err); 1788 } 1789 archive_set_error(&a->archive, 1790 ARCHIVE_ERRNO_MISC, 1791 "Parse error: SCHILY.acl.access"); 1792 } 1793 } else if (strcmp(key, "SCHILY.acl.default") == 0) { 1794 if (tar->sconv_acl == NULL) { 1795 tar->sconv_acl = 1796 archive_string_conversion_from_charset( 1797 &(a->archive), "UTF-8", 1); 1798 if (tar->sconv_acl == NULL) 1799 return (ARCHIVE_FATAL); 1800 } 1801 1802 r = archive_acl_parse_l(archive_entry_acl(entry), 1803 value, ARCHIVE_ENTRY_ACL_TYPE_DEFAULT, 1804 tar->sconv_acl); 1805 if (r != ARCHIVE_OK) { 1806 err = r; 1807 if (err == ARCHIVE_FATAL) { 1808 archive_set_error(&a->archive, ENOMEM, 1809 "Can't allocate memory for " 1810 "SCHILY.acl.default"); 1811 return (err); 1812 } 1813 archive_set_error(&a->archive, 1814 ARCHIVE_ERRNO_MISC, 1815 "Parse error: SCHILY.acl.default"); 1816 } 1817 } else if (strcmp(key, "SCHILY.devmajor") == 0) { 1818 archive_entry_set_rdevmajor(entry, 1819 (dev_t)tar_atol10(value, strlen(value))); 1820 } else if (strcmp(key, "SCHILY.devminor") == 0) { 1821 archive_entry_set_rdevminor(entry, 1822 (dev_t)tar_atol10(value, strlen(value))); 1823 } else if (strcmp(key, "SCHILY.fflags") == 0) { 1824 archive_entry_copy_fflags_text(entry, value); 1825 } else if (strcmp(key, "SCHILY.dev") == 0) { 1826 archive_entry_set_dev(entry, 1827 (dev_t)tar_atol10(value, strlen(value))); 1828 } else if (strcmp(key, "SCHILY.ino") == 0) { 1829 archive_entry_set_ino(entry, 1830 tar_atol10(value, strlen(value))); 1831 } else if (strcmp(key, "SCHILY.nlink") == 0) { 1832 archive_entry_set_nlink(entry, (unsigned) 1833 tar_atol10(value, strlen(value))); 1834 } else if (strcmp(key, "SCHILY.realsize") == 0) { 1835 tar->realsize = tar_atol10(value, strlen(value)); 1836 archive_entry_set_size(entry, tar->realsize); 1837 } else if (strcmp(key, "SUN.holesdata") == 0) { 1838 /* A Solaris extension for sparse. */ 1839 r = solaris_sparse_parse(a, tar, entry, value); 1840 if (r < err) { 1841 if (r == ARCHIVE_FATAL) 1842 return (r); 1843 err = r; 1844 archive_set_error(&a->archive, 1845 ARCHIVE_ERRNO_MISC, 1846 "Parse error: SUN.holesdata"); 1847 } 1848 } 1849 break; 1850 case 'a': 1851 if (strcmp(key, "atime") == 0) { 1852 pax_time(value, &s, &n); 1853 archive_entry_set_atime(entry, s, n); 1854 } 1855 break; 1856 case 'c': 1857 if (strcmp(key, "ctime") == 0) { 1858 pax_time(value, &s, &n); 1859 archive_entry_set_ctime(entry, s, n); 1860 } else if (strcmp(key, "charset") == 0) { 1861 /* TODO: Publish charset information in entry. */ 1862 } else if (strcmp(key, "comment") == 0) { 1863 /* TODO: Publish comment in entry. */ 1864 } 1865 break; 1866 case 'g': 1867 if (strcmp(key, "gid") == 0) { 1868 archive_entry_set_gid(entry, 1869 tar_atol10(value, strlen(value))); 1870 } else if (strcmp(key, "gname") == 0) { 1871 archive_strcpy(&(tar->entry_gname), value); 1872 } 1873 break; 1874 case 'h': 1875 if (strcmp(key, "hdrcharset") == 0) { 1876 if (strcmp(value, "BINARY") == 0) 1877 /* Binary mode. */ 1878 tar->pax_hdrcharset_binary = 1; 1879 else if (strcmp(value, "ISO-IR 10646 2000 UTF-8") == 0) 1880 tar->pax_hdrcharset_binary = 0; 1881 } 1882 break; 1883 case 'l': 1884 /* pax interchange doesn't distinguish hardlink vs. symlink. */ 1885 if (strcmp(key, "linkpath") == 0) { 1886 archive_strcpy(&(tar->entry_linkpath), value); 1887 } 1888 break; 1889 case 'm': 1890 if (strcmp(key, "mtime") == 0) { 1891 pax_time(value, &s, &n); 1892 archive_entry_set_mtime(entry, s, n); 1893 } 1894 break; 1895 case 'p': 1896 if (strcmp(key, "path") == 0) { 1897 archive_strcpy(&(tar->entry_pathname), value); 1898 } 1899 break; 1900 case 'r': 1901 /* POSIX has reserved 'realtime.*' */ 1902 break; 1903 case 's': 1904 /* POSIX has reserved 'security.*' */ 1905 /* Someday: if (strcmp(key, "security.acl") == 0) { ... } */ 1906 if (strcmp(key, "size") == 0) { 1907 /* "size" is the size of the data in the entry. */ 1908 tar->entry_bytes_remaining 1909 = tar_atol10(value, strlen(value)); 1910 /* 1911 * But, "size" is not necessarily the size of 1912 * the file on disk; if this is a sparse file, 1913 * the disk size may have already been set from 1914 * GNU.sparse.realsize or GNU.sparse.size or 1915 * an old GNU header field or SCHILY.realsize 1916 * or .... 1917 */ 1918 if (tar->realsize < 0) { 1919 archive_entry_set_size(entry, 1920 tar->entry_bytes_remaining); 1921 tar->realsize 1922 = tar->entry_bytes_remaining; 1923 } 1924 } 1925 break; 1926 case 'u': 1927 if (strcmp(key, "uid") == 0) { 1928 archive_entry_set_uid(entry, 1929 tar_atol10(value, strlen(value))); 1930 } else if (strcmp(key, "uname") == 0) { 1931 archive_strcpy(&(tar->entry_uname), value); 1932 } 1933 break; 1934 } 1935 return (err); 1936} 1937 1938 1939 1940/* 1941 * parse a decimal time value, which may include a fractional portion 1942 */ 1943static void 1944pax_time(const char *p, int64_t *ps, long *pn) 1945{ 1946 char digit; 1947 int64_t s; 1948 unsigned long l; 1949 int sign; 1950 int64_t limit, last_digit_limit; 1951 1952 limit = INT64_MAX / 10; 1953 last_digit_limit = INT64_MAX % 10; 1954 1955 s = 0; 1956 sign = 1; 1957 if (*p == '-') { 1958 sign = -1; 1959 p++; 1960 } 1961 while (*p >= '0' && *p <= '9') { 1962 digit = *p - '0'; 1963 if (s > limit || 1964 (s == limit && digit > last_digit_limit)) { 1965 s = INT64_MAX; 1966 break; 1967 } 1968 s = (s * 10) + digit; 1969 ++p; 1970 } 1971 1972 *ps = s * sign; 1973 1974 /* Calculate nanoseconds. */ 1975 *pn = 0; 1976 1977 if (*p != '.') 1978 return; 1979 1980 l = 100000000UL; 1981 do { 1982 ++p; 1983 if (*p >= '0' && *p <= '9') 1984 *pn += (*p - '0') * l; 1985 else 1986 break; 1987 } while (l /= 10); 1988} 1989 1990/* 1991 * Parse GNU tar header 1992 */ 1993static int 1994header_gnutar(struct archive_read *a, struct tar *tar, 1995 struct archive_entry *entry, const void *h, size_t *unconsumed) 1996{ 1997 const struct archive_entry_header_gnutar *header; 1998 int64_t t; 1999 int err = ARCHIVE_OK; 2000 2001 /* 2002 * GNU header is like POSIX ustar, except 'prefix' is 2003 * replaced with some other fields. This also means the 2004 * filename is stored as in old-style archives. 2005 */ 2006 2007 /* Grab fields common to all tar variants. */ 2008 err = header_common(a, tar, entry, h); 2009 if (err == ARCHIVE_FATAL) 2010 return (err); 2011 2012 /* Copy filename over (to ensure null termination). */ 2013 header = (const struct archive_entry_header_gnutar *)h; 2014 if (archive_entry_copy_pathname_l(entry, 2015 header->name, sizeof(header->name), tar->sconv) != 0) { 2016 err = set_conversion_failed_error(a, tar->sconv, "Pathname"); 2017 if (err == ARCHIVE_FATAL) 2018 return (err); 2019 } 2020 2021 /* Fields common to ustar and GNU */ 2022 /* XXX Can the following be factored out since it's common 2023 * to ustar and gnu tar? Is it okay to move it down into 2024 * header_common, perhaps? */ 2025 if (archive_entry_copy_uname_l(entry, 2026 header->uname, sizeof(header->uname), tar->sconv) != 0) { 2027 err = set_conversion_failed_error(a, tar->sconv, "Uname"); 2028 if (err == ARCHIVE_FATAL) 2029 return (err); 2030 } 2031 2032 if (archive_entry_copy_gname_l(entry, 2033 header->gname, sizeof(header->gname), tar->sconv) != 0) { 2034 err = set_conversion_failed_error(a, tar->sconv, "Gname"); 2035 if (err == ARCHIVE_FATAL) 2036 return (err); 2037 } 2038 2039 /* Parse out device numbers only for char and block specials */ 2040 if (header->typeflag[0] == '3' || header->typeflag[0] == '4') { 2041 archive_entry_set_rdevmajor(entry, (dev_t) 2042 tar_atol(header->rdevmajor, sizeof(header->rdevmajor))); 2043 archive_entry_set_rdevminor(entry, (dev_t) 2044 tar_atol(header->rdevminor, sizeof(header->rdevminor))); 2045 } else 2046 archive_entry_set_rdev(entry, 0); 2047 2048 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining); 2049 2050 /* Grab GNU-specific fields. */ 2051 t = tar_atol(header->atime, sizeof(header->atime)); 2052 if (t > 0) 2053 archive_entry_set_atime(entry, t, 0); 2054 t = tar_atol(header->ctime, sizeof(header->ctime)); 2055 if (t > 0) 2056 archive_entry_set_ctime(entry, t, 0); 2057 2058 if (header->realsize[0] != 0) { 2059 tar->realsize 2060 = tar_atol(header->realsize, sizeof(header->realsize)); 2061 archive_entry_set_size(entry, tar->realsize); 2062 } 2063 2064 if (header->sparse[0].offset[0] != 0) { 2065 if (gnu_sparse_old_read(a, tar, header, unconsumed) 2066 != ARCHIVE_OK) 2067 return (ARCHIVE_FATAL); 2068 } else { 2069 if (header->isextended[0] != 0) { 2070 /* XXX WTF? XXX */ 2071 } 2072 } 2073 2074 return (err); 2075} 2076 2077static int 2078gnu_add_sparse_entry(struct archive_read *a, struct tar *tar, 2079 int64_t offset, int64_t remaining) 2080{ 2081 struct sparse_block *p; 2082 2083 p = (struct sparse_block *)malloc(sizeof(*p)); 2084 if (p == NULL) { 2085 archive_set_error(&a->archive, ENOMEM, "Out of memory"); 2086 return (ARCHIVE_FATAL); 2087 } 2088 memset(p, 0, sizeof(*p)); 2089 if (tar->sparse_last != NULL) 2090 tar->sparse_last->next = p; 2091 else 2092 tar->sparse_list = p; 2093 tar->sparse_last = p; 2094 p->offset = offset; 2095 p->remaining = remaining; 2096 return (ARCHIVE_OK); 2097} 2098 2099static void 2100gnu_clear_sparse_list(struct tar *tar) 2101{ 2102 struct sparse_block *p; 2103 2104 while (tar->sparse_list != NULL) { 2105 p = tar->sparse_list; 2106 tar->sparse_list = p->next; 2107 free(p); 2108 } 2109 tar->sparse_last = NULL; 2110} 2111 2112/* 2113 * GNU tar old-format sparse data. 2114 * 2115 * GNU old-format sparse data is stored in a fixed-field 2116 * format. Offset/size values are 11-byte octal fields (same 2117 * format as 'size' field in ustart header). These are 2118 * stored in the header, allocating subsequent header blocks 2119 * as needed. Extending the header in this way is a pretty 2120 * severe POSIX violation; this design has earned GNU tar a 2121 * lot of criticism. 2122 */ 2123 2124static int 2125gnu_sparse_old_read(struct archive_read *a, struct tar *tar, 2126 const struct archive_entry_header_gnutar *header, size_t *unconsumed) 2127{ 2128 ssize_t bytes_read; 2129 const void *data; 2130 struct extended { 2131 struct gnu_sparse sparse[21]; 2132 char isextended[1]; 2133 char padding[7]; 2134 }; 2135 const struct extended *ext; 2136 2137 if (gnu_sparse_old_parse(a, tar, header->sparse, 4) != ARCHIVE_OK) 2138 return (ARCHIVE_FATAL); 2139 if (header->isextended[0] == 0) 2140 return (ARCHIVE_OK); 2141 2142 do { 2143 tar_flush_unconsumed(a, unconsumed); 2144 data = __archive_read_ahead(a, 512, &bytes_read); 2145 if (bytes_read < 0) 2146 return (ARCHIVE_FATAL); 2147 if (bytes_read < 512) { 2148 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 2149 "Truncated tar archive " 2150 "detected while reading sparse file data"); 2151 return (ARCHIVE_FATAL); 2152 } 2153 *unconsumed = 512; 2154 ext = (const struct extended *)data; 2155 if (gnu_sparse_old_parse(a, tar, ext->sparse, 21) != ARCHIVE_OK) 2156 return (ARCHIVE_FATAL); 2157 } while (ext->isextended[0] != 0); 2158 if (tar->sparse_list != NULL) 2159 tar->entry_offset = tar->sparse_list->offset; 2160 return (ARCHIVE_OK); 2161} 2162 2163static int 2164gnu_sparse_old_parse(struct archive_read *a, struct tar *tar, 2165 const struct gnu_sparse *sparse, int length) 2166{ 2167 while (length > 0 && sparse->offset[0] != 0) { 2168 if (gnu_add_sparse_entry(a, tar, 2169 tar_atol(sparse->offset, sizeof(sparse->offset)), 2170 tar_atol(sparse->numbytes, sizeof(sparse->numbytes))) 2171 != ARCHIVE_OK) 2172 return (ARCHIVE_FATAL); 2173 sparse++; 2174 length--; 2175 } 2176 return (ARCHIVE_OK); 2177} 2178 2179/* 2180 * GNU tar sparse format 0.0 2181 * 2182 * Beginning with GNU tar 1.15, sparse files are stored using 2183 * information in the pax extended header. The GNU tar maintainers 2184 * have gone through a number of variations in the process of working 2185 * out this scheme; fortunately, they're all numbered. 2186 * 2187 * Sparse format 0.0 uses attribute GNU.sparse.numblocks to store the 2188 * number of blocks, and GNU.sparse.offset/GNU.sparse.numbytes to 2189 * store offset/size for each block. The repeated instances of these 2190 * latter fields violate the pax specification (which frowns on 2191 * duplicate keys), so this format was quickly replaced. 2192 */ 2193 2194/* 2195 * GNU tar sparse format 0.1 2196 * 2197 * This version replaced the offset/numbytes attributes with 2198 * a single "map" attribute that stored a list of integers. This 2199 * format had two problems: First, the "map" attribute could be very 2200 * long, which caused problems for some implementations. More 2201 * importantly, the sparse data was lost when extracted by archivers 2202 * that didn't recognize this extension. 2203 */ 2204 2205static int 2206gnu_sparse_01_parse(struct archive_read *a, struct tar *tar, const char *p) 2207{ 2208 const char *e; 2209 int64_t offset = -1, size = -1; 2210 2211 for (;;) { 2212 e = p; 2213 while (*e != '\0' && *e != ',') { 2214 if (*e < '0' || *e > '9') 2215 return (ARCHIVE_WARN); 2216 e++; 2217 } 2218 if (offset < 0) { 2219 offset = tar_atol10(p, e - p); 2220 if (offset < 0) 2221 return (ARCHIVE_WARN); 2222 } else { 2223 size = tar_atol10(p, e - p); 2224 if (size < 0) 2225 return (ARCHIVE_WARN); 2226 if (gnu_add_sparse_entry(a, tar, offset, size) 2227 != ARCHIVE_OK) 2228 return (ARCHIVE_FATAL); 2229 offset = -1; 2230 } 2231 if (*e == '\0') 2232 return (ARCHIVE_OK); 2233 p = e + 1; 2234 } 2235} 2236 2237/* 2238 * GNU tar sparse format 1.0 2239 * 2240 * The idea: The offset/size data is stored as a series of base-10 2241 * ASCII numbers prepended to the file data, so that dearchivers that 2242 * don't support this format will extract the block map along with the 2243 * data and a separate post-process can restore the sparseness. 2244 * 2245 * Unfortunately, GNU tar 1.16 had a bug that added unnecessary 2246 * padding to the body of the file when using this format. GNU tar 2247 * 1.17 corrected this bug without bumping the version number, so 2248 * it's not possible to support both variants. This code supports 2249 * the later variant at the expense of not supporting the former. 2250 * 2251 * This variant also replaced GNU.sparse.size with GNU.sparse.realsize 2252 * and introduced the GNU.sparse.major/GNU.sparse.minor attributes. 2253 */ 2254 2255/* 2256 * Read the next line from the input, and parse it as a decimal 2257 * integer followed by '\n'. Returns positive integer value or 2258 * negative on error. 2259 */ 2260static int64_t 2261gnu_sparse_10_atol(struct archive_read *a, struct tar *tar, 2262 int64_t *remaining, size_t *unconsumed) 2263{ 2264 int64_t l, limit, last_digit_limit; 2265 const char *p; 2266 ssize_t bytes_read; 2267 int base, digit; 2268 2269 base = 10; 2270 limit = INT64_MAX / base; 2271 last_digit_limit = INT64_MAX % base; 2272 2273 /* 2274 * Skip any lines starting with '#'; GNU tar specs 2275 * don't require this, but they should. 2276 */ 2277 do { 2278 bytes_read = readline(a, tar, &p, 2279 (ssize_t)tar_min(*remaining, 100), unconsumed); 2280 if (bytes_read <= 0) 2281 return (ARCHIVE_FATAL); 2282 *remaining -= bytes_read; 2283 } while (p[0] == '#'); 2284 2285 l = 0; 2286 while (bytes_read > 0) { 2287 if (*p == '\n') 2288 return (l); 2289 if (*p < '0' || *p >= '0' + base) 2290 return (ARCHIVE_WARN); 2291 digit = *p - '0'; 2292 if (l > limit || (l == limit && digit > last_digit_limit)) 2293 l = INT64_MAX; /* Truncate on overflow. */ 2294 else 2295 l = (l * base) + digit; 2296 p++; 2297 bytes_read--; 2298 } 2299 /* TODO: Error message. */ 2300 return (ARCHIVE_WARN); 2301} 2302 2303/* 2304 * Returns length (in bytes) of the sparse data description 2305 * that was read. 2306 */ 2307static ssize_t 2308gnu_sparse_10_read(struct archive_read *a, struct tar *tar, size_t *unconsumed) 2309{ 2310 ssize_t bytes_read; 2311 int entries; 2312 int64_t offset, size, to_skip, remaining; 2313 2314 /* Clear out the existing sparse list. */ 2315 gnu_clear_sparse_list(tar); 2316 2317 remaining = tar->entry_bytes_remaining; 2318 2319 /* Parse entries. */ 2320 entries = (int)gnu_sparse_10_atol(a, tar, &remaining, unconsumed); 2321 if (entries < 0) 2322 return (ARCHIVE_FATAL); 2323 /* Parse the individual entries. */ 2324 while (entries-- > 0) { 2325 /* Parse offset/size */ 2326 offset = gnu_sparse_10_atol(a, tar, &remaining, unconsumed); 2327 if (offset < 0) 2328 return (ARCHIVE_FATAL); 2329 size = gnu_sparse_10_atol(a, tar, &remaining, unconsumed); 2330 if (size < 0) 2331 return (ARCHIVE_FATAL); 2332 /* Add a new sparse entry. */ 2333 if (gnu_add_sparse_entry(a, tar, offset, size) != ARCHIVE_OK) 2334 return (ARCHIVE_FATAL); 2335 } 2336 /* Skip rest of block... */ 2337 tar_flush_unconsumed(a, unconsumed); 2338 bytes_read = (ssize_t)(tar->entry_bytes_remaining - remaining); 2339 to_skip = 0x1ff & -bytes_read; 2340 if (to_skip != __archive_read_consume(a, to_skip)) 2341 return (ARCHIVE_FATAL); 2342 return ((ssize_t)(bytes_read + to_skip)); 2343} 2344 2345/* 2346 * Solaris pax extension for a sparse file. This is recorded with the 2347 * data and hole pairs. The way recording sparse information by Solaris' 2348 * pax simply indicates where data and sparse are, so the stored contents 2349 * consist of both data and hole. 2350 */ 2351static int 2352solaris_sparse_parse(struct archive_read *a, struct tar *tar, 2353 struct archive_entry *entry, const char *p) 2354{ 2355 const char *e; 2356 int64_t start, end; 2357 int hole = 1; 2358 2359 (void)entry; /* UNUSED */ 2360 2361 end = 0; 2362 if (*p == ' ') 2363 p++; 2364 else 2365 return (ARCHIVE_WARN); 2366 for (;;) { 2367 e = p; 2368 while (*e != '\0' && *e != ' ') { 2369 if (*e < '0' || *e > '9') 2370 return (ARCHIVE_WARN); 2371 e++; 2372 } 2373 start = end; 2374 end = tar_atol10(p, e - p); 2375 if (end < 0) 2376 return (ARCHIVE_WARN); 2377 if (start < end) { 2378 if (gnu_add_sparse_entry(a, tar, start, 2379 end - start) != ARCHIVE_OK) 2380 return (ARCHIVE_FATAL); 2381 tar->sparse_last->hole = hole; 2382 } 2383 if (*e == '\0') 2384 return (ARCHIVE_OK); 2385 p = e + 1; 2386 hole = hole == 0; 2387 } 2388} 2389 2390/*- 2391 * Convert text->integer. 2392 * 2393 * Traditional tar formats (including POSIX) specify base-8 for 2394 * all of the standard numeric fields. This is a significant limitation 2395 * in practice: 2396 * = file size is limited to 8GB 2397 * = rdevmajor and rdevminor are limited to 21 bits 2398 * = uid/gid are limited to 21 bits 2399 * 2400 * There are two workarounds for this: 2401 * = pax extended headers, which use variable-length string fields 2402 * = GNU tar and STAR both allow either base-8 or base-256 in 2403 * most fields. The high bit is set to indicate base-256. 2404 * 2405 * On read, this implementation supports both extensions. 2406 */ 2407static int64_t 2408tar_atol(const char *p, size_t char_cnt) 2409{ 2410 /* 2411 * Technically, GNU tar considers a field to be in base-256 2412 * only if the first byte is 0xff or 0x80. 2413 */ 2414 if (*p & 0x80) 2415 return (tar_atol256(p, char_cnt)); 2416 return (tar_atol8(p, char_cnt)); 2417} 2418 2419/* 2420 * Note that this implementation does not (and should not!) obey 2421 * locale settings; you cannot simply substitute strtol here, since 2422 * it does obey locale. 2423 */ 2424static int64_t 2425tar_atol_base_n(const char *p, size_t char_cnt, int base) 2426{ 2427 int64_t l, limit, last_digit_limit; 2428 int digit, sign; 2429 2430 limit = INT64_MAX / base; 2431 last_digit_limit = INT64_MAX % base; 2432 2433 /* the pointer will not be dereferenced if char_cnt is zero 2434 * due to the way the && operator is evaulated. 2435 */ 2436 while (char_cnt != 0 && (*p == ' ' || *p == '\t')) { 2437 p++; 2438 char_cnt--; 2439 } 2440 2441 sign = 1; 2442 if (char_cnt != 0 && *p == '-') { 2443 sign = -1; 2444 p++; 2445 char_cnt--; 2446 } 2447 2448 l = 0; 2449 if (char_cnt != 0) { 2450 digit = *p - '0'; 2451 while (digit >= 0 && digit < base && char_cnt != 0) { 2452 if (l>limit || (l == limit && digit > last_digit_limit)) { 2453 l = INT64_MAX; /* Truncate on overflow. */ 2454 break; 2455 } 2456 l = (l * base) + digit; 2457 digit = *++p - '0'; 2458 char_cnt--; 2459 } 2460 } 2461 return (sign < 0) ? -l : l; 2462} 2463 2464static int64_t 2465tar_atol8(const char *p, size_t char_cnt) 2466{ 2467 return tar_atol_base_n(p, char_cnt, 8); 2468} 2469 2470static int64_t 2471tar_atol10(const char *p, size_t char_cnt) 2472{ 2473 return tar_atol_base_n(p, char_cnt, 10); 2474} 2475 2476/* 2477 * Parse a base-256 integer. This is just a straight signed binary 2478 * value in big-endian order, except that the high-order bit is 2479 * ignored. 2480 */ 2481static int64_t 2482tar_atol256(const char *_p, size_t char_cnt) 2483{ 2484 int64_t l, upper_limit, lower_limit; 2485 const unsigned char *p = (const unsigned char *)_p; 2486 2487 upper_limit = INT64_MAX / 256; 2488 lower_limit = INT64_MIN / 256; 2489 2490 /* Pad with 1 or 0 bits, depending on sign. */ 2491 if ((0x40 & *p) == 0x40) 2492 l = (int64_t)-1; 2493 else 2494 l = 0; 2495 l = (l << 6) | (0x3f & *p++); 2496 while (--char_cnt > 0) { 2497 if (l > upper_limit) { 2498 l = INT64_MAX; /* Truncate on overflow */ 2499 break; 2500 } else if (l < lower_limit) { 2501 l = INT64_MIN; 2502 break; 2503 } 2504 l = (l << 8) | (0xff & (int64_t)*p++); 2505 } 2506 return (l); 2507} 2508 2509/* 2510 * Returns length of line (including trailing newline) 2511 * or negative on error. 'start' argument is updated to 2512 * point to first character of line. This avoids copying 2513 * when possible. 2514 */ 2515static ssize_t 2516readline(struct archive_read *a, struct tar *tar, const char **start, 2517 ssize_t limit, size_t *unconsumed) 2518{ 2519 ssize_t bytes_read; 2520 ssize_t total_size = 0; 2521 const void *t; 2522 const char *s; 2523 void *p; 2524 2525 tar_flush_unconsumed(a, unconsumed); 2526 2527 t = __archive_read_ahead(a, 1, &bytes_read); 2528 if (bytes_read <= 0) 2529 return (ARCHIVE_FATAL); 2530 s = t; /* Start of line? */ 2531 p = memchr(t, '\n', bytes_read); 2532 /* If we found '\n' in the read buffer, return pointer to that. */ 2533 if (p != NULL) { 2534 bytes_read = 1 + ((const char *)p) - s; 2535 if (bytes_read > limit) { 2536 archive_set_error(&a->archive, 2537 ARCHIVE_ERRNO_FILE_FORMAT, 2538 "Line too long"); 2539 return (ARCHIVE_FATAL); 2540 } 2541 *unconsumed = bytes_read; 2542 *start = s; 2543 return (bytes_read); 2544 } 2545 *unconsumed = bytes_read; 2546 /* Otherwise, we need to accumulate in a line buffer. */ 2547 for (;;) { 2548 if (total_size + bytes_read > limit) { 2549 archive_set_error(&a->archive, 2550 ARCHIVE_ERRNO_FILE_FORMAT, 2551 "Line too long"); 2552 return (ARCHIVE_FATAL); 2553 } 2554 if (archive_string_ensure(&tar->line, total_size + bytes_read) == NULL) { 2555 archive_set_error(&a->archive, ENOMEM, 2556 "Can't allocate working buffer"); 2557 return (ARCHIVE_FATAL); 2558 } 2559 memcpy(tar->line.s + total_size, t, bytes_read); 2560 tar_flush_unconsumed(a, unconsumed); 2561 total_size += bytes_read; 2562 /* If we found '\n', clean up and return. */ 2563 if (p != NULL) { 2564 *start = tar->line.s; 2565 return (total_size); 2566 } 2567 /* Read some more. */ 2568 t = __archive_read_ahead(a, 1, &bytes_read); 2569 if (bytes_read <= 0) 2570 return (ARCHIVE_FATAL); 2571 s = t; /* Start of line? */ 2572 p = memchr(t, '\n', bytes_read); 2573 /* If we found '\n', trim the read. */ 2574 if (p != NULL) { 2575 bytes_read = 1 + ((const char *)p) - s; 2576 } 2577 *unconsumed = bytes_read; 2578 } 2579} 2580 2581/* 2582 * base64_decode - Base64 decode 2583 * 2584 * This accepts most variations of base-64 encoding, including: 2585 * * with or without line breaks 2586 * * with or without the final group padded with '=' or '_' characters 2587 * (The most economical Base-64 variant does not pad the last group and 2588 * omits line breaks; RFC1341 used for MIME requires both.) 2589 */ 2590static char * 2591base64_decode(const char *s, size_t len, size_t *out_len) 2592{ 2593 static const unsigned char digits[64] = { 2594 'A','B','C','D','E','F','G','H','I','J','K','L','M','N', 2595 'O','P','Q','R','S','T','U','V','W','X','Y','Z','a','b', 2596 'c','d','e','f','g','h','i','j','k','l','m','n','o','p', 2597 'q','r','s','t','u','v','w','x','y','z','0','1','2','3', 2598 '4','5','6','7','8','9','+','/' }; 2599 static unsigned char decode_table[128]; 2600 char *out, *d; 2601 const unsigned char *src = (const unsigned char *)s; 2602 2603 /* If the decode table is not yet initialized, prepare it. */ 2604 if (decode_table[digits[1]] != 1) { 2605 unsigned i; 2606 memset(decode_table, 0xff, sizeof(decode_table)); 2607 for (i = 0; i < sizeof(digits); i++) 2608 decode_table[digits[i]] = i; 2609 } 2610 2611 /* Allocate enough space to hold the entire output. */ 2612 /* Note that we may not use all of this... */ 2613 out = (char *)malloc(len - len / 4 + 1); 2614 if (out == NULL) { 2615 *out_len = 0; 2616 return (NULL); 2617 } 2618 d = out; 2619 2620 while (len > 0) { 2621 /* Collect the next group of (up to) four characters. */ 2622 int v = 0; 2623 int group_size = 0; 2624 while (group_size < 4 && len > 0) { 2625 /* '=' or '_' padding indicates final group. */ 2626 if (*src == '=' || *src == '_') { 2627 len = 0; 2628 break; 2629 } 2630 /* Skip illegal characters (including line breaks) */ 2631 if (*src > 127 || *src < 32 2632 || decode_table[*src] == 0xff) { 2633 len--; 2634 src++; 2635 continue; 2636 } 2637 v <<= 6; 2638 v |= decode_table[*src++]; 2639 len --; 2640 group_size++; 2641 } 2642 /* Align a short group properly. */ 2643 v <<= 6 * (4 - group_size); 2644 /* Unpack the group we just collected. */ 2645 switch (group_size) { 2646 case 4: d[2] = v & 0xff; 2647 /* FALLTHROUGH */ 2648 case 3: d[1] = (v >> 8) & 0xff; 2649 /* FALLTHROUGH */ 2650 case 2: d[0] = (v >> 16) & 0xff; 2651 break; 2652 case 1: /* this is invalid! */ 2653 break; 2654 } 2655 d += group_size * 3 / 4; 2656 } 2657 2658 *out_len = d - out; 2659 return (out); 2660} 2661 2662static char * 2663url_decode(const char *in) 2664{ 2665 char *out, *d; 2666 const char *s; 2667 2668 out = (char *)malloc(strlen(in) + 1); 2669 if (out == NULL) 2670 return (NULL); 2671 for (s = in, d = out; *s != '\0'; ) { 2672 if (s[0] == '%' && s[1] != '\0' && s[2] != '\0') { 2673 /* Try to convert % escape */ 2674 int digit1 = tohex(s[1]); 2675 int digit2 = tohex(s[2]); 2676 if (digit1 >= 0 && digit2 >= 0) { 2677 /* Looks good, consume three chars */ 2678 s += 3; 2679 /* Convert output */ 2680 *d++ = ((digit1 << 4) | digit2); 2681 continue; 2682 } 2683 /* Else fall through and treat '%' as normal char */ 2684 } 2685 *d++ = *s++; 2686 } 2687 *d = '\0'; 2688 return (out); 2689} 2690 2691static int 2692tohex(int c) 2693{ 2694 if (c >= '0' && c <= '9') 2695 return (c - '0'); 2696 else if (c >= 'A' && c <= 'F') 2697 return (c - 'A' + 10); 2698 else if (c >= 'a' && c <= 'f') 2699 return (c - 'a' + 10); 2700 else 2701 return (-1); 2702} 2703