archive_write_set_format_pax.c revision 311041
1/*- 2 * Copyright (c) 2003-2007 Tim Kientzle 3 * Copyright (c) 2010-2012 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/11/contrib/libarchive/libarchive/archive_write_set_format_pax.c 311041 2017-01-02 01:41:31Z mm $"); 29 30#ifdef HAVE_ERRNO_H 31#include <errno.h> 32#endif 33#ifdef HAVE_STDLIB_H 34#include <stdlib.h> 35#endif 36#ifdef HAVE_STRING_H 37#include <string.h> 38#endif 39 40#include "archive.h" 41#include "archive_entry.h" 42#include "archive_entry_locale.h" 43#include "archive_private.h" 44#include "archive_write_private.h" 45 46struct sparse_block { 47 struct sparse_block *next; 48 int is_hole; 49 uint64_t offset; 50 uint64_t remaining; 51}; 52 53struct pax { 54 uint64_t entry_bytes_remaining; 55 uint64_t entry_padding; 56 struct archive_string l_url_encoded_name; 57 struct archive_string pax_header; 58 struct archive_string sparse_map; 59 size_t sparse_map_padding; 60 struct sparse_block *sparse_list; 61 struct sparse_block *sparse_tail; 62 struct archive_string_conv *sconv_utf8; 63 int opt_binary; 64}; 65 66static void add_pax_attr(struct archive_string *, const char *key, 67 const char *value); 68static void add_pax_attr_int(struct archive_string *, 69 const char *key, int64_t value); 70static void add_pax_attr_time(struct archive_string *, 71 const char *key, int64_t sec, 72 unsigned long nanos); 73static ssize_t archive_write_pax_data(struct archive_write *, 74 const void *, size_t); 75static int archive_write_pax_close(struct archive_write *); 76static int archive_write_pax_free(struct archive_write *); 77static int archive_write_pax_finish_entry(struct archive_write *); 78static int archive_write_pax_header(struct archive_write *, 79 struct archive_entry *); 80static int archive_write_pax_options(struct archive_write *, 81 const char *, const char *); 82static char *base64_encode(const char *src, size_t len); 83static char *build_gnu_sparse_name(char *dest, const char *src); 84static char *build_pax_attribute_name(char *dest, const char *src); 85static char *build_ustar_entry_name(char *dest, const char *src, 86 size_t src_length, const char *insert); 87static char *format_int(char *dest, int64_t); 88static int has_non_ASCII(const char *); 89static void sparse_list_clear(struct pax *); 90static int sparse_list_add(struct pax *, int64_t, int64_t); 91static char *url_encode(const char *in); 92 93/* 94 * Set output format to 'restricted pax' format. 95 * 96 * This is the same as normal 'pax', but tries to suppress 97 * the pax header whenever possible. This is the default for 98 * bsdtar, for instance. 99 */ 100int 101archive_write_set_format_pax_restricted(struct archive *_a) 102{ 103 struct archive_write *a = (struct archive_write *)_a; 104 int r; 105 106 archive_check_magic(_a, ARCHIVE_WRITE_MAGIC, 107 ARCHIVE_STATE_NEW, "archive_write_set_format_pax_restricted"); 108 109 r = archive_write_set_format_pax(&a->archive); 110 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_RESTRICTED; 111 a->archive.archive_format_name = "restricted POSIX pax interchange"; 112 return (r); 113} 114 115/* 116 * Set output format to 'pax' format. 117 */ 118int 119archive_write_set_format_pax(struct archive *_a) 120{ 121 struct archive_write *a = (struct archive_write *)_a; 122 struct pax *pax; 123 124 archive_check_magic(_a, ARCHIVE_WRITE_MAGIC, 125 ARCHIVE_STATE_NEW, "archive_write_set_format_pax"); 126 127 if (a->format_free != NULL) 128 (a->format_free)(a); 129 130 pax = (struct pax *)calloc(1, sizeof(*pax)); 131 if (pax == NULL) { 132 archive_set_error(&a->archive, ENOMEM, 133 "Can't allocate pax data"); 134 return (ARCHIVE_FATAL); 135 } 136 a->format_data = pax; 137 a->format_name = "pax"; 138 a->format_options = archive_write_pax_options; 139 a->format_write_header = archive_write_pax_header; 140 a->format_write_data = archive_write_pax_data; 141 a->format_close = archive_write_pax_close; 142 a->format_free = archive_write_pax_free; 143 a->format_finish_entry = archive_write_pax_finish_entry; 144 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE; 145 a->archive.archive_format_name = "POSIX pax interchange"; 146 return (ARCHIVE_OK); 147} 148 149static int 150archive_write_pax_options(struct archive_write *a, const char *key, 151 const char *val) 152{ 153 struct pax *pax = (struct pax *)a->format_data; 154 int ret = ARCHIVE_FAILED; 155 156 if (strcmp(key, "hdrcharset") == 0) { 157 /* 158 * The character-set we can use are defined in 159 * IEEE Std 1003.1-2001 160 */ 161 if (val == NULL || val[0] == 0) 162 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 163 "pax: hdrcharset option needs a character-set name"); 164 else if (strcmp(val, "BINARY") == 0 || 165 strcmp(val, "binary") == 0) { 166 /* 167 * Specify binary mode. We will not convert 168 * filenames, uname and gname to any charsets. 169 */ 170 pax->opt_binary = 1; 171 ret = ARCHIVE_OK; 172 } else if (strcmp(val, "UTF-8") == 0) { 173 /* 174 * Specify UTF-8 character-set to be used for 175 * filenames. This is almost the test that 176 * running platform supports the string conversion. 177 * Especially libarchive_test needs this trick for 178 * its test. 179 */ 180 pax->sconv_utf8 = archive_string_conversion_to_charset( 181 &(a->archive), "UTF-8", 0); 182 if (pax->sconv_utf8 == NULL) 183 ret = ARCHIVE_FATAL; 184 else 185 ret = ARCHIVE_OK; 186 } else 187 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 188 "pax: invalid charset name"); 189 return (ret); 190 } 191 192 /* Note: The "warn" return is just to inform the options 193 * supervisor that we didn't handle it. It will generate 194 * a suitable error if no one used this option. */ 195 return (ARCHIVE_WARN); 196} 197 198/* 199 * Note: This code assumes that 'nanos' has the same sign as 'sec', 200 * which implies that sec=-1, nanos=200000000 represents -1.2 seconds 201 * and not -0.8 seconds. This is a pretty pedantic point, as we're 202 * unlikely to encounter many real files created before Jan 1, 1970, 203 * much less ones with timestamps recorded to sub-second resolution. 204 */ 205static void 206add_pax_attr_time(struct archive_string *as, const char *key, 207 int64_t sec, unsigned long nanos) 208{ 209 int digit, i; 210 char *t; 211 /* 212 * Note that each byte contributes fewer than 3 base-10 213 * digits, so this will always be big enough. 214 */ 215 char tmp[1 + 3*sizeof(sec) + 1 + 3*sizeof(nanos)]; 216 217 tmp[sizeof(tmp) - 1] = 0; 218 t = tmp + sizeof(tmp) - 1; 219 220 /* Skip trailing zeros in the fractional part. */ 221 for (digit = 0, i = 10; i > 0 && digit == 0; i--) { 222 digit = nanos % 10; 223 nanos /= 10; 224 } 225 226 /* Only format the fraction if it's non-zero. */ 227 if (i > 0) { 228 while (i > 0) { 229 *--t = "0123456789"[digit]; 230 digit = nanos % 10; 231 nanos /= 10; 232 i--; 233 } 234 *--t = '.'; 235 } 236 t = format_int(t, sec); 237 238 add_pax_attr(as, key, t); 239} 240 241static char * 242format_int(char *t, int64_t i) 243{ 244 uint64_t ui; 245 246 if (i < 0) 247 ui = (i == INT64_MIN) ? (uint64_t)(INT64_MAX) + 1 : (uint64_t)(-i); 248 else 249 ui = i; 250 251 do { 252 *--t = "0123456789"[ui % 10]; 253 } while (ui /= 10); 254 if (i < 0) 255 *--t = '-'; 256 return (t); 257} 258 259static void 260add_pax_attr_int(struct archive_string *as, const char *key, int64_t value) 261{ 262 char tmp[1 + 3 * sizeof(value)]; 263 264 tmp[sizeof(tmp) - 1] = 0; 265 add_pax_attr(as, key, format_int(tmp + sizeof(tmp) - 1, value)); 266} 267 268/* 269 * Add a key/value attribute to the pax header. This function handles 270 * the length field and various other syntactic requirements. 271 */ 272static void 273add_pax_attr(struct archive_string *as, const char *key, const char *value) 274{ 275 int digits, i, len, next_ten; 276 char tmp[1 + 3 * sizeof(int)]; /* < 3 base-10 digits per byte */ 277 278 /*- 279 * PAX attributes have the following layout: 280 * <len> <space> <key> <=> <value> <nl> 281 */ 282 len = 1 + (int)strlen(key) + 1 + (int)strlen(value) + 1; 283 284 /* 285 * The <len> field includes the length of the <len> field, so 286 * computing the correct length is tricky. I start by 287 * counting the number of base-10 digits in 'len' and 288 * computing the next higher power of 10. 289 */ 290 next_ten = 1; 291 digits = 0; 292 i = len; 293 while (i > 0) { 294 i = i / 10; 295 digits++; 296 next_ten = next_ten * 10; 297 } 298 /* 299 * For example, if string without the length field is 99 300 * chars, then adding the 2 digit length "99" will force the 301 * total length past 100, requiring an extra digit. The next 302 * statement adjusts for this effect. 303 */ 304 if (len + digits >= next_ten) 305 digits++; 306 307 /* Now, we have the right length so we can build the line. */ 308 tmp[sizeof(tmp) - 1] = 0; /* Null-terminate the work area. */ 309 archive_strcat(as, format_int(tmp + sizeof(tmp) - 1, len + digits)); 310 archive_strappend_char(as, ' '); 311 archive_strcat(as, key); 312 archive_strappend_char(as, '='); 313 archive_strcat(as, value); 314 archive_strappend_char(as, '\n'); 315} 316 317static int 318archive_write_pax_header_xattrs(struct archive_write *a, 319 struct pax *pax, struct archive_entry *entry) 320{ 321 struct archive_string s; 322 int i = archive_entry_xattr_reset(entry); 323 324 while (i--) { 325 const char *name; 326 const void *value; 327 char *encoded_value; 328 char *url_encoded_name = NULL, *encoded_name = NULL; 329 size_t size; 330 int r; 331 332 archive_entry_xattr_next(entry, &name, &value, &size); 333 url_encoded_name = url_encode(name); 334 if (url_encoded_name != NULL) { 335 /* Convert narrow-character to UTF-8. */ 336 r = archive_strcpy_l(&(pax->l_url_encoded_name), 337 url_encoded_name, pax->sconv_utf8); 338 free(url_encoded_name); /* Done with this. */ 339 if (r == 0) 340 encoded_name = pax->l_url_encoded_name.s; 341 else if (errno == ENOMEM) { 342 archive_set_error(&a->archive, ENOMEM, 343 "Can't allocate memory for Linkname"); 344 return (ARCHIVE_FATAL); 345 } 346 } 347 348 encoded_value = base64_encode((const char *)value, size); 349 350 if (encoded_name != NULL && encoded_value != NULL) { 351 archive_string_init(&s); 352 archive_strcpy(&s, "LIBARCHIVE.xattr."); 353 archive_strcat(&s, encoded_name); 354 add_pax_attr(&(pax->pax_header), s.s, encoded_value); 355 archive_string_free(&s); 356 } 357 free(encoded_value); 358 } 359 return (ARCHIVE_OK); 360} 361 362static int 363get_entry_hardlink(struct archive_write *a, struct archive_entry *entry, 364 const char **name, size_t *length, struct archive_string_conv *sc) 365{ 366 int r; 367 368 r = archive_entry_hardlink_l(entry, name, length, sc); 369 if (r != 0) { 370 if (errno == ENOMEM) { 371 archive_set_error(&a->archive, ENOMEM, 372 "Can't allocate memory for Linkname"); 373 return (ARCHIVE_FATAL); 374 } 375 return (ARCHIVE_WARN); 376 } 377 return (ARCHIVE_OK); 378} 379 380static int 381get_entry_pathname(struct archive_write *a, struct archive_entry *entry, 382 const char **name, size_t *length, struct archive_string_conv *sc) 383{ 384 int r; 385 386 r = archive_entry_pathname_l(entry, name, length, sc); 387 if (r != 0) { 388 if (errno == ENOMEM) { 389 archive_set_error(&a->archive, ENOMEM, 390 "Can't allocate memory for Pathname"); 391 return (ARCHIVE_FATAL); 392 } 393 return (ARCHIVE_WARN); 394 } 395 return (ARCHIVE_OK); 396} 397 398static int 399get_entry_uname(struct archive_write *a, struct archive_entry *entry, 400 const char **name, size_t *length, struct archive_string_conv *sc) 401{ 402 int r; 403 404 r = archive_entry_uname_l(entry, name, length, sc); 405 if (r != 0) { 406 if (errno == ENOMEM) { 407 archive_set_error(&a->archive, ENOMEM, 408 "Can't allocate memory for Uname"); 409 return (ARCHIVE_FATAL); 410 } 411 return (ARCHIVE_WARN); 412 } 413 return (ARCHIVE_OK); 414} 415 416static int 417get_entry_gname(struct archive_write *a, struct archive_entry *entry, 418 const char **name, size_t *length, struct archive_string_conv *sc) 419{ 420 int r; 421 422 r = archive_entry_gname_l(entry, name, length, sc); 423 if (r != 0) { 424 if (errno == ENOMEM) { 425 archive_set_error(&a->archive, ENOMEM, 426 "Can't allocate memory for Gname"); 427 return (ARCHIVE_FATAL); 428 } 429 return (ARCHIVE_WARN); 430 } 431 return (ARCHIVE_OK); 432} 433 434static int 435get_entry_symlink(struct archive_write *a, struct archive_entry *entry, 436 const char **name, size_t *length, struct archive_string_conv *sc) 437{ 438 int r; 439 440 r = archive_entry_symlink_l(entry, name, length, sc); 441 if (r != 0) { 442 if (errno == ENOMEM) { 443 archive_set_error(&a->archive, ENOMEM, 444 "Can't allocate memory for Linkname"); 445 return (ARCHIVE_FATAL); 446 } 447 return (ARCHIVE_WARN); 448 } 449 return (ARCHIVE_OK); 450} 451 452/* 453 * TODO: Consider adding 'comment' and 'charset' fields to 454 * archive_entry so that clients can specify them. Also, consider 455 * adding generic key/value tags so clients can add arbitrary 456 * key/value data. 457 * 458 * TODO: Break up this 700-line function!!!! Yowza! 459 */ 460static int 461archive_write_pax_header(struct archive_write *a, 462 struct archive_entry *entry_original) 463{ 464 struct archive_entry *entry_main; 465 const char *p; 466 const char *suffix; 467 int need_extension, r, ret; 468 int sparse_count; 469 uint64_t sparse_total, real_size; 470 struct pax *pax; 471 const char *hardlink; 472 const char *path = NULL, *linkpath = NULL; 473 const char *uname = NULL, *gname = NULL; 474 const void *mac_metadata; 475 size_t mac_metadata_size; 476 struct archive_string_conv *sconv; 477 size_t hardlink_length, path_length, linkpath_length; 478 size_t uname_length, gname_length; 479 480 char paxbuff[512]; 481 char ustarbuff[512]; 482 char ustar_entry_name[256]; 483 char pax_entry_name[256]; 484 char gnu_sparse_name[256]; 485 struct archive_string entry_name; 486 487 ret = ARCHIVE_OK; 488 need_extension = 0; 489 pax = (struct pax *)a->format_data; 490 491 /* Sanity check. */ 492 if (archive_entry_pathname(entry_original) == NULL) { 493 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 494 "Can't record entry in tar file without pathname"); 495 return (ARCHIVE_FAILED); 496 } 497 498 /* 499 * Choose a header encoding. 500 */ 501 if (pax->opt_binary) 502 sconv = NULL;/* Binary mode. */ 503 else { 504 /* Header encoding is UTF-8. */ 505 if (pax->sconv_utf8 == NULL) { 506 /* Initialize the string conversion object 507 * we must need */ 508 pax->sconv_utf8 = archive_string_conversion_to_charset( 509 &(a->archive), "UTF-8", 1); 510 if (pax->sconv_utf8 == NULL) 511 /* Couldn't allocate memory */ 512 return (ARCHIVE_FAILED); 513 } 514 sconv = pax->sconv_utf8; 515 } 516 517 r = get_entry_hardlink(a, entry_original, &hardlink, 518 &hardlink_length, sconv); 519 if (r == ARCHIVE_FATAL) 520 return (r); 521 else if (r != ARCHIVE_OK) { 522 r = get_entry_hardlink(a, entry_original, &hardlink, 523 &hardlink_length, NULL); 524 if (r == ARCHIVE_FATAL) 525 return (r); 526 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 527 "Can't translate linkname '%s' to %s", hardlink, 528 archive_string_conversion_charset_name(sconv)); 529 ret = ARCHIVE_WARN; 530 sconv = NULL;/* The header charset switches to binary mode. */ 531 } 532 533 /* Make sure this is a type of entry that we can handle here */ 534 if (hardlink == NULL) { 535 switch (archive_entry_filetype(entry_original)) { 536 case AE_IFBLK: 537 case AE_IFCHR: 538 case AE_IFIFO: 539 case AE_IFLNK: 540 case AE_IFREG: 541 break; 542 case AE_IFDIR: 543 { 544 /* 545 * Ensure a trailing '/'. Modify the original 546 * entry so the client sees the change. 547 */ 548#if defined(_WIN32) && !defined(__CYGWIN__) 549 const wchar_t *wp; 550 551 wp = archive_entry_pathname_w(entry_original); 552 if (wp != NULL && wp[wcslen(wp) -1] != L'/') { 553 struct archive_wstring ws; 554 555 archive_string_init(&ws); 556 path_length = wcslen(wp); 557 if (archive_wstring_ensure(&ws, 558 path_length + 2) == NULL) { 559 archive_set_error(&a->archive, ENOMEM, 560 "Can't allocate pax data"); 561 archive_wstring_free(&ws); 562 return(ARCHIVE_FATAL); 563 } 564 /* Should we keep '\' ? */ 565 if (wp[path_length -1] == L'\\') 566 path_length--; 567 archive_wstrncpy(&ws, wp, path_length); 568 archive_wstrappend_wchar(&ws, L'/'); 569 archive_entry_copy_pathname_w( 570 entry_original, ws.s); 571 archive_wstring_free(&ws); 572 p = NULL; 573 } else 574#endif 575 p = archive_entry_pathname(entry_original); 576 /* 577 * On Windows, this is a backup operation just in 578 * case getting WCS failed. On POSIX, this is a 579 * normal operation. 580 */ 581 if (p != NULL && p[strlen(p) - 1] != '/') { 582 struct archive_string as; 583 584 archive_string_init(&as); 585 path_length = strlen(p); 586 if (archive_string_ensure(&as, 587 path_length + 2) == NULL) { 588 archive_set_error(&a->archive, ENOMEM, 589 "Can't allocate pax data"); 590 archive_string_free(&as); 591 return(ARCHIVE_FATAL); 592 } 593#if defined(_WIN32) && !defined(__CYGWIN__) 594 /* NOTE: This might break the pathname 595 * if the current code page is CP932 and 596 * the pathname includes a character '\' 597 * as a part of its multibyte pathname. */ 598 if (p[strlen(p) -1] == '\\') 599 path_length--; 600 else 601#endif 602 archive_strncpy(&as, p, path_length); 603 archive_strappend_char(&as, '/'); 604 archive_entry_copy_pathname( 605 entry_original, as.s); 606 archive_string_free(&as); 607 } 608 break; 609 } 610 case AE_IFSOCK: 611 archive_set_error(&a->archive, 612 ARCHIVE_ERRNO_FILE_FORMAT, 613 "tar format cannot archive socket"); 614 return (ARCHIVE_FAILED); 615 default: 616 archive_set_error(&a->archive, 617 ARCHIVE_ERRNO_FILE_FORMAT, 618 "tar format cannot archive this (type=0%lo)", 619 (unsigned long) 620 archive_entry_filetype(entry_original)); 621 return (ARCHIVE_FAILED); 622 } 623 } 624 625 /* 626 * If Mac OS metadata blob is here, recurse to write that 627 * as a separate entry. This is really a pretty poor design: 628 * In particular, it doubles the overhead for long filenames. 629 * TODO: Help Apple folks design something better and figure 630 * out how to transition from this legacy format. 631 * 632 * Note that this code is present on every platform; clients 633 * on non-Mac are unlikely to ever provide this data, but 634 * applications that copy entries from one archive to another 635 * should not lose data just because the local filesystem 636 * can't store it. 637 */ 638 mac_metadata = 639 archive_entry_mac_metadata(entry_original, &mac_metadata_size); 640 if (mac_metadata != NULL) { 641 const char *oname; 642 char *name, *bname; 643 size_t name_length; 644 struct archive_entry *extra = archive_entry_new2(&a->archive); 645 646 oname = archive_entry_pathname(entry_original); 647 name_length = strlen(oname); 648 name = malloc(name_length + 3); 649 if (name == NULL || extra == NULL) { 650 /* XXX error message */ 651 archive_entry_free(extra); 652 free(name); 653 return (ARCHIVE_FAILED); 654 } 655 strcpy(name, oname); 656 /* Find last '/'; strip trailing '/' characters */ 657 bname = strrchr(name, '/'); 658 while (bname != NULL && bname[1] == '\0') { 659 *bname = '\0'; 660 bname = strrchr(name, '/'); 661 } 662 if (bname == NULL) { 663 memmove(name + 2, name, name_length + 1); 664 memmove(name, "._", 2); 665 } else { 666 bname += 1; 667 memmove(bname + 2, bname, strlen(bname) + 1); 668 memmove(bname, "._", 2); 669 } 670 archive_entry_copy_pathname(extra, name); 671 free(name); 672 673 archive_entry_set_size(extra, mac_metadata_size); 674 archive_entry_set_filetype(extra, AE_IFREG); 675 archive_entry_set_perm(extra, 676 archive_entry_perm(entry_original)); 677 archive_entry_set_mtime(extra, 678 archive_entry_mtime(entry_original), 679 archive_entry_mtime_nsec(entry_original)); 680 archive_entry_set_gid(extra, 681 archive_entry_gid(entry_original)); 682 archive_entry_set_gname(extra, 683 archive_entry_gname(entry_original)); 684 archive_entry_set_uid(extra, 685 archive_entry_uid(entry_original)); 686 archive_entry_set_uname(extra, 687 archive_entry_uname(entry_original)); 688 689 /* Recurse to write the special copyfile entry. */ 690 r = archive_write_pax_header(a, extra); 691 archive_entry_free(extra); 692 if (r < ARCHIVE_WARN) 693 return (r); 694 if (r < ret) 695 ret = r; 696 r = (int)archive_write_pax_data(a, mac_metadata, 697 mac_metadata_size); 698 if (r < ARCHIVE_WARN) 699 return (r); 700 if (r < ret) 701 ret = r; 702 r = archive_write_pax_finish_entry(a); 703 if (r < ARCHIVE_WARN) 704 return (r); 705 if (r < ret) 706 ret = r; 707 } 708 709 /* Copy entry so we can modify it as needed. */ 710#if defined(_WIN32) && !defined(__CYGWIN__) 711 /* Make sure the path separators in pathname, hardlink and symlink 712 * are all slash '/', not the Windows path separator '\'. */ 713 entry_main = __la_win_entry_in_posix_pathseparator(entry_original); 714 if (entry_main == entry_original) 715 entry_main = archive_entry_clone(entry_original); 716#else 717 entry_main = archive_entry_clone(entry_original); 718#endif 719 if (entry_main == NULL) { 720 archive_set_error(&a->archive, ENOMEM, 721 "Can't allocate pax data"); 722 return(ARCHIVE_FATAL); 723 } 724 archive_string_empty(&(pax->pax_header)); /* Blank our work area. */ 725 archive_string_empty(&(pax->sparse_map)); 726 sparse_total = 0; 727 sparse_list_clear(pax); 728 729 if (hardlink == NULL && 730 archive_entry_filetype(entry_main) == AE_IFREG) 731 sparse_count = archive_entry_sparse_reset(entry_main); 732 else 733 sparse_count = 0; 734 if (sparse_count) { 735 int64_t offset, length, last_offset = 0; 736 /* Get the last entry of sparse block. */ 737 while (archive_entry_sparse_next( 738 entry_main, &offset, &length) == ARCHIVE_OK) 739 last_offset = offset + length; 740 741 /* If the last sparse block does not reach the end of file, 742 * We have to add a empty sparse block as the last entry to 743 * manage storing file data. */ 744 if (last_offset < archive_entry_size(entry_main)) 745 archive_entry_sparse_add_entry(entry_main, 746 archive_entry_size(entry_main), 0); 747 sparse_count = archive_entry_sparse_reset(entry_main); 748 } 749 750 /* 751 * First, check the name fields and see if any of them 752 * require binary coding. If any of them does, then all of 753 * them do. 754 */ 755 r = get_entry_pathname(a, entry_main, &path, &path_length, sconv); 756 if (r == ARCHIVE_FATAL) 757 return (r); 758 else if (r != ARCHIVE_OK) { 759 r = get_entry_pathname(a, entry_main, &path, 760 &path_length, NULL); 761 if (r == ARCHIVE_FATAL) 762 return (r); 763 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 764 "Can't translate pathname '%s' to %s", path, 765 archive_string_conversion_charset_name(sconv)); 766 ret = ARCHIVE_WARN; 767 sconv = NULL;/* The header charset switches to binary mode. */ 768 } 769 r = get_entry_uname(a, entry_main, &uname, &uname_length, sconv); 770 if (r == ARCHIVE_FATAL) 771 return (r); 772 else if (r != ARCHIVE_OK) { 773 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL); 774 if (r == ARCHIVE_FATAL) 775 return (r); 776 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 777 "Can't translate uname '%s' to %s", uname, 778 archive_string_conversion_charset_name(sconv)); 779 ret = ARCHIVE_WARN; 780 sconv = NULL;/* The header charset switches to binary mode. */ 781 } 782 r = get_entry_gname(a, entry_main, &gname, &gname_length, sconv); 783 if (r == ARCHIVE_FATAL) 784 return (r); 785 else if (r != ARCHIVE_OK) { 786 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL); 787 if (r == ARCHIVE_FATAL) 788 return (r); 789 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 790 "Can't translate gname '%s' to %s", gname, 791 archive_string_conversion_charset_name(sconv)); 792 ret = ARCHIVE_WARN; 793 sconv = NULL;/* The header charset switches to binary mode. */ 794 } 795 linkpath = hardlink; 796 linkpath_length = hardlink_length; 797 if (linkpath == NULL) { 798 r = get_entry_symlink(a, entry_main, &linkpath, 799 &linkpath_length, sconv); 800 if (r == ARCHIVE_FATAL) 801 return (r); 802 else if (r != ARCHIVE_OK) { 803 r = get_entry_symlink(a, entry_main, &linkpath, 804 &linkpath_length, NULL); 805 if (r == ARCHIVE_FATAL) 806 return (r); 807 archive_set_error(&a->archive, 808 ARCHIVE_ERRNO_FILE_FORMAT, 809 "Can't translate linkname '%s' to %s", linkpath, 810 archive_string_conversion_charset_name(sconv)); 811 ret = ARCHIVE_WARN; 812 sconv = NULL; 813 } 814 } 815 816 /* If any string conversions failed, get all attributes 817 * in binary-mode. */ 818 if (sconv == NULL && !pax->opt_binary) { 819 if (hardlink != NULL) { 820 r = get_entry_hardlink(a, entry_main, &hardlink, 821 &hardlink_length, NULL); 822 if (r == ARCHIVE_FATAL) 823 return (r); 824 linkpath = hardlink; 825 linkpath_length = hardlink_length; 826 } 827 r = get_entry_pathname(a, entry_main, &path, 828 &path_length, NULL); 829 if (r == ARCHIVE_FATAL) 830 return (r); 831 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL); 832 if (r == ARCHIVE_FATAL) 833 return (r); 834 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL); 835 if (r == ARCHIVE_FATAL) 836 return (r); 837 } 838 839 /* Store the header encoding first, to be nice to readers. */ 840 if (sconv == NULL) 841 add_pax_attr(&(pax->pax_header), "hdrcharset", "BINARY"); 842 843 844 /* 845 * If name is too long, or has non-ASCII characters, add 846 * 'path' to pax extended attrs. (Note that an unconvertible 847 * name must have non-ASCII characters.) 848 */ 849 if (has_non_ASCII(path)) { 850 /* We have non-ASCII characters. */ 851 add_pax_attr(&(pax->pax_header), "path", path); 852 archive_entry_set_pathname(entry_main, 853 build_ustar_entry_name(ustar_entry_name, 854 path, path_length, NULL)); 855 need_extension = 1; 856 } else { 857 /* We have an all-ASCII path; we'd like to just store 858 * it in the ustar header if it will fit. Yes, this 859 * duplicates some of the logic in 860 * archive_write_set_format_ustar.c 861 */ 862 if (path_length <= 100) { 863 /* Fits in the old 100-char tar name field. */ 864 } else { 865 /* Find largest suffix that will fit. */ 866 /* Note: strlen() > 100, so strlen() - 100 - 1 >= 0 */ 867 suffix = strchr(path + path_length - 100 - 1, '/'); 868 /* Don't attempt an empty prefix. */ 869 if (suffix == path) 870 suffix = strchr(suffix + 1, '/'); 871 /* We can put it in the ustar header if it's 872 * all ASCII and it's either <= 100 characters 873 * or can be split at a '/' into a prefix <= 874 * 155 chars and a suffix <= 100 chars. (Note 875 * the strchr() above will return NULL exactly 876 * when the path can't be split.) 877 */ 878 if (suffix == NULL /* Suffix > 100 chars. */ 879 || suffix[1] == '\0' /* empty suffix */ 880 || suffix - path > 155) /* Prefix > 155 chars */ 881 { 882 add_pax_attr(&(pax->pax_header), "path", path); 883 archive_entry_set_pathname(entry_main, 884 build_ustar_entry_name(ustar_entry_name, 885 path, path_length, NULL)); 886 need_extension = 1; 887 } 888 } 889 } 890 891 if (linkpath != NULL) { 892 /* If link name is too long or has non-ASCII characters, add 893 * 'linkpath' to pax extended attrs. */ 894 if (linkpath_length > 100 || has_non_ASCII(linkpath)) { 895 add_pax_attr(&(pax->pax_header), "linkpath", linkpath); 896 if (linkpath_length > 100) { 897 if (hardlink != NULL) 898 archive_entry_set_hardlink(entry_main, 899 "././@LongHardLink"); 900 else 901 archive_entry_set_symlink(entry_main, 902 "././@LongSymLink"); 903 } 904 need_extension = 1; 905 } 906 } 907 /* Save a pathname since it will be renamed if `entry_main` has 908 * sparse blocks. */ 909 archive_string_init(&entry_name); 910 archive_strcpy(&entry_name, archive_entry_pathname(entry_main)); 911 912 /* If file size is too large, add 'size' to pax extended attrs. */ 913 if (archive_entry_size(entry_main) >= (((int64_t)1) << 33)) { 914 add_pax_attr_int(&(pax->pax_header), "size", 915 archive_entry_size(entry_main)); 916 need_extension = 1; 917 } 918 919 /* If numeric GID is too large, add 'gid' to pax extended attrs. */ 920 if ((unsigned int)archive_entry_gid(entry_main) >= (1 << 18)) { 921 add_pax_attr_int(&(pax->pax_header), "gid", 922 archive_entry_gid(entry_main)); 923 need_extension = 1; 924 } 925 926 /* If group name is too large or has non-ASCII characters, add 927 * 'gname' to pax extended attrs. */ 928 if (gname != NULL) { 929 if (gname_length > 31 || has_non_ASCII(gname)) { 930 add_pax_attr(&(pax->pax_header), "gname", gname); 931 need_extension = 1; 932 } 933 } 934 935 /* If numeric UID is too large, add 'uid' to pax extended attrs. */ 936 if ((unsigned int)archive_entry_uid(entry_main) >= (1 << 18)) { 937 add_pax_attr_int(&(pax->pax_header), "uid", 938 archive_entry_uid(entry_main)); 939 need_extension = 1; 940 } 941 942 /* Add 'uname' to pax extended attrs if necessary. */ 943 if (uname != NULL) { 944 if (uname_length > 31 || has_non_ASCII(uname)) { 945 add_pax_attr(&(pax->pax_header), "uname", uname); 946 need_extension = 1; 947 } 948 } 949 950 /* 951 * POSIX/SUSv3 doesn't provide a standard key for large device 952 * numbers. I use the same keys here that Joerg Schilling 953 * used for 'star.' (Which, somewhat confusingly, are called 954 * "devXXX" even though they code "rdev" values.) No doubt, 955 * other implementations use other keys. Note that there's no 956 * reason we can't write the same information into a number of 957 * different keys. 958 * 959 * Of course, this is only needed for block or char device entries. 960 */ 961 if (archive_entry_filetype(entry_main) == AE_IFBLK 962 || archive_entry_filetype(entry_main) == AE_IFCHR) { 963 /* 964 * If rdevmajor is too large, add 'SCHILY.devmajor' to 965 * extended attributes. 966 */ 967 int rdevmajor, rdevminor; 968 rdevmajor = archive_entry_rdevmajor(entry_main); 969 rdevminor = archive_entry_rdevminor(entry_main); 970 if (rdevmajor >= (1 << 18)) { 971 add_pax_attr_int(&(pax->pax_header), "SCHILY.devmajor", 972 rdevmajor); 973 /* 974 * Non-strict formatting below means we don't 975 * have to truncate here. Not truncating improves 976 * the chance that some more modern tar archivers 977 * (such as GNU tar 1.13) can restore the full 978 * value even if they don't understand the pax 979 * extended attributes. See my rant below about 980 * file size fields for additional details. 981 */ 982 /* archive_entry_set_rdevmajor(entry_main, 983 rdevmajor & ((1 << 18) - 1)); */ 984 need_extension = 1; 985 } 986 987 /* 988 * If devminor is too large, add 'SCHILY.devminor' to 989 * extended attributes. 990 */ 991 if (rdevminor >= (1 << 18)) { 992 add_pax_attr_int(&(pax->pax_header), "SCHILY.devminor", 993 rdevminor); 994 /* Truncation is not necessary here, either. */ 995 /* archive_entry_set_rdevminor(entry_main, 996 rdevminor & ((1 << 18) - 1)); */ 997 need_extension = 1; 998 } 999 } 1000 1001 /* 1002 * Technically, the mtime field in the ustar header can 1003 * support 33 bits, but many platforms use signed 32-bit time 1004 * values. The cutoff of 0x7fffffff here is a compromise. 1005 * Yes, this check is duplicated just below; this helps to 1006 * avoid writing an mtime attribute just to handle a 1007 * high-resolution timestamp in "restricted pax" mode. 1008 */ 1009 if (!need_extension && 1010 ((archive_entry_mtime(entry_main) < 0) 1011 || (archive_entry_mtime(entry_main) >= 0x7fffffff))) 1012 need_extension = 1; 1013 1014 /* I use a star-compatible file flag attribute. */ 1015 p = archive_entry_fflags_text(entry_main); 1016 if (!need_extension && p != NULL && *p != '\0') 1017 need_extension = 1; 1018 1019 /* If there are non-trivial ACL entries, we need an extension. */ 1020 if (!need_extension && archive_entry_acl_count(entry_original, 1021 ARCHIVE_ENTRY_ACL_TYPE_ACCESS) > 0) 1022 need_extension = 1; 1023 1024 /* If there are non-trivial ACL entries, we need an extension. */ 1025 if (!need_extension && archive_entry_acl_count(entry_original, 1026 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT) > 0) 1027 need_extension = 1; 1028 1029 /* If there are extended attributes, we need an extension */ 1030 if (!need_extension && archive_entry_xattr_count(entry_original) > 0) 1031 need_extension = 1; 1032 1033 /* If there are sparse info, we need an extension */ 1034 if (!need_extension && sparse_count > 0) 1035 need_extension = 1; 1036 1037 /* 1038 * Libarchive used to include these in extended headers for 1039 * restricted pax format, but that confused people who 1040 * expected ustar-like time semantics. So now we only include 1041 * them in full pax format. 1042 */ 1043 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_RESTRICTED) { 1044 if (archive_entry_ctime(entry_main) != 0 || 1045 archive_entry_ctime_nsec(entry_main) != 0) 1046 add_pax_attr_time(&(pax->pax_header), "ctime", 1047 archive_entry_ctime(entry_main), 1048 archive_entry_ctime_nsec(entry_main)); 1049 1050 if (archive_entry_atime(entry_main) != 0 || 1051 archive_entry_atime_nsec(entry_main) != 0) 1052 add_pax_attr_time(&(pax->pax_header), "atime", 1053 archive_entry_atime(entry_main), 1054 archive_entry_atime_nsec(entry_main)); 1055 1056 /* Store birth/creationtime only if it's earlier than mtime */ 1057 if (archive_entry_birthtime_is_set(entry_main) && 1058 archive_entry_birthtime(entry_main) 1059 < archive_entry_mtime(entry_main)) 1060 add_pax_attr_time(&(pax->pax_header), 1061 "LIBARCHIVE.creationtime", 1062 archive_entry_birthtime(entry_main), 1063 archive_entry_birthtime_nsec(entry_main)); 1064 } 1065 1066 /* 1067 * The following items are handled differently in "pax 1068 * restricted" format. In particular, in "pax restricted" 1069 * format they won't be added unless need_extension is 1070 * already set (we're already generating an extended header, so 1071 * may as well include these). 1072 */ 1073 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_RESTRICTED || 1074 need_extension) { 1075 if (archive_entry_mtime(entry_main) < 0 || 1076 archive_entry_mtime(entry_main) >= 0x7fffffff || 1077 archive_entry_mtime_nsec(entry_main) != 0) 1078 add_pax_attr_time(&(pax->pax_header), "mtime", 1079 archive_entry_mtime(entry_main), 1080 archive_entry_mtime_nsec(entry_main)); 1081 1082 /* I use a star-compatible file flag attribute. */ 1083 p = archive_entry_fflags_text(entry_main); 1084 if (p != NULL && *p != '\0') 1085 add_pax_attr(&(pax->pax_header), "SCHILY.fflags", p); 1086 1087 /* I use star-compatible ACL attributes. */ 1088 r = archive_entry_acl_text_l(entry_original, 1089 ARCHIVE_ENTRY_ACL_TYPE_ACCESS | 1090 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID, 1091 &p, NULL, pax->sconv_utf8); 1092 if (r != 0) { 1093 if (errno == ENOMEM) { 1094 archive_set_error(&a->archive, ENOMEM, 1095 "Can't allocate memory for " 1096 "ACL.access"); 1097 return (ARCHIVE_FATAL); 1098 } 1099 archive_set_error(&a->archive, 1100 ARCHIVE_ERRNO_FILE_FORMAT, 1101 "Can't translate ACL.access to UTF-8"); 1102 ret = ARCHIVE_WARN; 1103 } else if (p != NULL && *p != '\0') { 1104 add_pax_attr(&(pax->pax_header), 1105 "SCHILY.acl.access", p); 1106 } 1107 r = archive_entry_acl_text_l(entry_original, 1108 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT | 1109 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID, 1110 &p, NULL, pax->sconv_utf8); 1111 if (r != 0) { 1112 if (errno == ENOMEM) { 1113 archive_set_error(&a->archive, ENOMEM, 1114 "Can't allocate memory for " 1115 "ACL.default"); 1116 return (ARCHIVE_FATAL); 1117 } 1118 archive_set_error(&a->archive, 1119 ARCHIVE_ERRNO_FILE_FORMAT, 1120 "Can't translate ACL.default to UTF-8"); 1121 ret = ARCHIVE_WARN; 1122 } else if (p != NULL && *p != '\0') { 1123 add_pax_attr(&(pax->pax_header), 1124 "SCHILY.acl.default", p); 1125 } 1126 1127 /* We use GNU-tar-compatible sparse attributes. */ 1128 if (sparse_count > 0) { 1129 int64_t soffset, slength; 1130 1131 add_pax_attr_int(&(pax->pax_header), 1132 "GNU.sparse.major", 1); 1133 add_pax_attr_int(&(pax->pax_header), 1134 "GNU.sparse.minor", 0); 1135 add_pax_attr(&(pax->pax_header), 1136 "GNU.sparse.name", entry_name.s); 1137 add_pax_attr_int(&(pax->pax_header), 1138 "GNU.sparse.realsize", 1139 archive_entry_size(entry_main)); 1140 1141 /* Rename the file name which will be used for 1142 * ustar header to a special name, which GNU 1143 * PAX Format 1.0 requires */ 1144 archive_entry_set_pathname(entry_main, 1145 build_gnu_sparse_name(gnu_sparse_name, 1146 entry_name.s)); 1147 1148 /* 1149 * - Make a sparse map, which will precede a file data. 1150 * - Get the total size of available data of sparse. 1151 */ 1152 archive_string_sprintf(&(pax->sparse_map), "%d\n", 1153 sparse_count); 1154 while (archive_entry_sparse_next(entry_main, 1155 &soffset, &slength) == ARCHIVE_OK) { 1156 archive_string_sprintf(&(pax->sparse_map), 1157 "%jd\n%jd\n", 1158 (intmax_t)soffset, 1159 (intmax_t)slength); 1160 sparse_total += slength; 1161 if (sparse_list_add(pax, soffset, slength) 1162 != ARCHIVE_OK) { 1163 archive_set_error(&a->archive, 1164 ENOMEM, 1165 "Can't allocate memory"); 1166 archive_entry_free(entry_main); 1167 archive_string_free(&entry_name); 1168 return (ARCHIVE_FATAL); 1169 } 1170 } 1171 } 1172 1173 /* Store extended attributes */ 1174 if (archive_write_pax_header_xattrs(a, pax, entry_original) 1175 == ARCHIVE_FATAL) { 1176 archive_entry_free(entry_main); 1177 archive_string_free(&entry_name); 1178 return (ARCHIVE_FATAL); 1179 } 1180 } 1181 1182 /* Only regular files have data. */ 1183 if (archive_entry_filetype(entry_main) != AE_IFREG) 1184 archive_entry_set_size(entry_main, 0); 1185 1186 /* 1187 * Pax-restricted does not store data for hardlinks, in order 1188 * to improve compatibility with ustar. 1189 */ 1190 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE && 1191 hardlink != NULL) 1192 archive_entry_set_size(entry_main, 0); 1193 1194 /* 1195 * XXX Full pax interchange format does permit a hardlink 1196 * entry to have data associated with it. I'm not supporting 1197 * that here because the client expects me to tell them whether 1198 * or not this format expects data for hardlinks. If I 1199 * don't check here, then every pax archive will end up with 1200 * duplicated data for hardlinks. Someday, there may be 1201 * need to select this behavior, in which case the following 1202 * will need to be revisited. XXX 1203 */ 1204 if (hardlink != NULL) 1205 archive_entry_set_size(entry_main, 0); 1206 1207 /* Save a real file size. */ 1208 real_size = archive_entry_size(entry_main); 1209 /* 1210 * Overwrite a file size by the total size of sparse blocks and 1211 * the size of sparse map info. That file size is the length of 1212 * the data, which we will exactly store into an archive file. 1213 */ 1214 if (archive_strlen(&(pax->sparse_map))) { 1215 size_t mapsize = archive_strlen(&(pax->sparse_map)); 1216 pax->sparse_map_padding = 0x1ff & (-(ssize_t)mapsize); 1217 archive_entry_set_size(entry_main, 1218 mapsize + pax->sparse_map_padding + sparse_total); 1219 } 1220 1221 /* Format 'ustar' header for main entry. 1222 * 1223 * The trouble with file size: If the reader can't understand 1224 * the file size, they may not be able to locate the next 1225 * entry and the rest of the archive is toast. Pax-compliant 1226 * readers are supposed to ignore the file size in the main 1227 * header, so the question becomes how to maximize portability 1228 * for readers that don't support pax attribute extensions. 1229 * For maximum compatibility, I permit numeric extensions in 1230 * the main header so that the file size stored will always be 1231 * correct, even if it's in a format that only some 1232 * implementations understand. The technique used here is: 1233 * 1234 * a) If possible, follow the standard exactly. This handles 1235 * files up to 8 gigabytes minus 1. 1236 * 1237 * b) If that fails, try octal but omit the field terminator. 1238 * That handles files up to 64 gigabytes minus 1. 1239 * 1240 * c) Otherwise, use base-256 extensions. That handles files 1241 * up to 2^63 in this implementation, with the potential to 1242 * go up to 2^94. That should hold us for a while. ;-) 1243 * 1244 * The non-strict formatter uses similar logic for other 1245 * numeric fields, though they're less critical. 1246 */ 1247 if (__archive_write_format_header_ustar(a, ustarbuff, entry_main, -1, 0, 1248 NULL) == ARCHIVE_FATAL) 1249 return (ARCHIVE_FATAL); 1250 1251 /* If we built any extended attributes, write that entry first. */ 1252 if (archive_strlen(&(pax->pax_header)) > 0) { 1253 struct archive_entry *pax_attr_entry; 1254 time_t s; 1255 int64_t uid, gid; 1256 int mode; 1257 1258 pax_attr_entry = archive_entry_new2(&a->archive); 1259 p = entry_name.s; 1260 archive_entry_set_pathname(pax_attr_entry, 1261 build_pax_attribute_name(pax_entry_name, p)); 1262 archive_entry_set_size(pax_attr_entry, 1263 archive_strlen(&(pax->pax_header))); 1264 /* Copy uid/gid (but clip to ustar limits). */ 1265 uid = archive_entry_uid(entry_main); 1266 if (uid >= 1 << 18) 1267 uid = (1 << 18) - 1; 1268 archive_entry_set_uid(pax_attr_entry, uid); 1269 gid = archive_entry_gid(entry_main); 1270 if (gid >= 1 << 18) 1271 gid = (1 << 18) - 1; 1272 archive_entry_set_gid(pax_attr_entry, gid); 1273 /* Copy mode over (but not setuid/setgid bits) */ 1274 mode = archive_entry_mode(entry_main); 1275#ifdef S_ISUID 1276 mode &= ~S_ISUID; 1277#endif 1278#ifdef S_ISGID 1279 mode &= ~S_ISGID; 1280#endif 1281#ifdef S_ISVTX 1282 mode &= ~S_ISVTX; 1283#endif 1284 archive_entry_set_mode(pax_attr_entry, mode); 1285 1286 /* Copy uname/gname. */ 1287 archive_entry_set_uname(pax_attr_entry, 1288 archive_entry_uname(entry_main)); 1289 archive_entry_set_gname(pax_attr_entry, 1290 archive_entry_gname(entry_main)); 1291 1292 /* Copy mtime, but clip to ustar limits. */ 1293 s = archive_entry_mtime(entry_main); 1294 if (s < 0) { s = 0; } 1295 if (s >= 0x7fffffff) { s = 0x7fffffff; } 1296 archive_entry_set_mtime(pax_attr_entry, s, 0); 1297 1298 /* Standard ustar doesn't support atime. */ 1299 archive_entry_set_atime(pax_attr_entry, 0, 0); 1300 1301 /* Standard ustar doesn't support ctime. */ 1302 archive_entry_set_ctime(pax_attr_entry, 0, 0); 1303 1304 r = __archive_write_format_header_ustar(a, paxbuff, 1305 pax_attr_entry, 'x', 1, NULL); 1306 1307 archive_entry_free(pax_attr_entry); 1308 1309 /* Note that the 'x' header shouldn't ever fail to format */ 1310 if (r < ARCHIVE_WARN) { 1311 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1312 "archive_write_pax_header: " 1313 "'x' header failed?! This can't happen.\n"); 1314 return (ARCHIVE_FATAL); 1315 } else if (r < ret) 1316 ret = r; 1317 r = __archive_write_output(a, paxbuff, 512); 1318 if (r != ARCHIVE_OK) { 1319 sparse_list_clear(pax); 1320 pax->entry_bytes_remaining = 0; 1321 pax->entry_padding = 0; 1322 return (ARCHIVE_FATAL); 1323 } 1324 1325 pax->entry_bytes_remaining = archive_strlen(&(pax->pax_header)); 1326 pax->entry_padding = 1327 0x1ff & (-(int64_t)pax->entry_bytes_remaining); 1328 1329 r = __archive_write_output(a, pax->pax_header.s, 1330 archive_strlen(&(pax->pax_header))); 1331 if (r != ARCHIVE_OK) { 1332 /* If a write fails, we're pretty much toast. */ 1333 return (ARCHIVE_FATAL); 1334 } 1335 /* Pad out the end of the entry. */ 1336 r = __archive_write_nulls(a, (size_t)pax->entry_padding); 1337 if (r != ARCHIVE_OK) { 1338 /* If a write fails, we're pretty much toast. */ 1339 return (ARCHIVE_FATAL); 1340 } 1341 pax->entry_bytes_remaining = pax->entry_padding = 0; 1342 } 1343 1344 /* Write the header for main entry. */ 1345 r = __archive_write_output(a, ustarbuff, 512); 1346 if (r != ARCHIVE_OK) 1347 return (r); 1348 1349 /* 1350 * Inform the client of the on-disk size we're using, so 1351 * they can avoid unnecessarily writing a body for something 1352 * that we're just going to ignore. 1353 */ 1354 archive_entry_set_size(entry_original, real_size); 1355 if (pax->sparse_list == NULL && real_size > 0) { 1356 /* This is not a sparse file but we handle its data as 1357 * a sparse block. */ 1358 sparse_list_add(pax, 0, real_size); 1359 sparse_total = real_size; 1360 } 1361 pax->entry_padding = 0x1ff & (-(int64_t)sparse_total); 1362 archive_entry_free(entry_main); 1363 archive_string_free(&entry_name); 1364 1365 return (ret); 1366} 1367 1368/* 1369 * We need a valid name for the regular 'ustar' entry. This routine 1370 * tries to hack something more-or-less reasonable. 1371 * 1372 * The approach here tries to preserve leading dir names. We do so by 1373 * working with four sections: 1374 * 1) "prefix" directory names, 1375 * 2) "suffix" directory names, 1376 * 3) inserted dir name (optional), 1377 * 4) filename. 1378 * 1379 * These sections must satisfy the following requirements: 1380 * * Parts 1 & 2 together form an initial portion of the dir name. 1381 * * Part 3 is specified by the caller. (It should not contain a leading 1382 * or trailing '/'.) 1383 * * Part 4 forms an initial portion of the base filename. 1384 * * The filename must be <= 99 chars to fit the ustar 'name' field. 1385 * * Parts 2, 3, 4 together must be <= 99 chars to fit the ustar 'name' fld. 1386 * * Part 1 must be <= 155 chars to fit the ustar 'prefix' field. 1387 * * If the original name ends in a '/', the new name must also end in a '/' 1388 * * Trailing '/.' sequences may be stripped. 1389 * 1390 * Note: Recall that the ustar format does not store the '/' separating 1391 * parts 1 & 2, but does store the '/' separating parts 2 & 3. 1392 */ 1393static char * 1394build_ustar_entry_name(char *dest, const char *src, size_t src_length, 1395 const char *insert) 1396{ 1397 const char *prefix, *prefix_end; 1398 const char *suffix, *suffix_end; 1399 const char *filename, *filename_end; 1400 char *p; 1401 int need_slash = 0; /* Was there a trailing slash? */ 1402 size_t suffix_length = 99; 1403 size_t insert_length; 1404 1405 /* Length of additional dir element to be added. */ 1406 if (insert == NULL) 1407 insert_length = 0; 1408 else 1409 /* +2 here allows for '/' before and after the insert. */ 1410 insert_length = strlen(insert) + 2; 1411 1412 /* Step 0: Quick bailout in a common case. */ 1413 if (src_length < 100 && insert == NULL) { 1414 strncpy(dest, src, src_length); 1415 dest[src_length] = '\0'; 1416 return (dest); 1417 } 1418 1419 /* Step 1: Locate filename and enforce the length restriction. */ 1420 filename_end = src + src_length; 1421 /* Remove trailing '/' chars and '/.' pairs. */ 1422 for (;;) { 1423 if (filename_end > src && filename_end[-1] == '/') { 1424 filename_end --; 1425 need_slash = 1; /* Remember to restore trailing '/'. */ 1426 continue; 1427 } 1428 if (filename_end > src + 1 && filename_end[-1] == '.' 1429 && filename_end[-2] == '/') { 1430 filename_end -= 2; 1431 need_slash = 1; /* "foo/." will become "foo/" */ 1432 continue; 1433 } 1434 break; 1435 } 1436 if (need_slash) 1437 suffix_length--; 1438 /* Find start of filename. */ 1439 filename = filename_end - 1; 1440 while ((filename > src) && (*filename != '/')) 1441 filename --; 1442 if ((*filename == '/') && (filename < filename_end - 1)) 1443 filename ++; 1444 /* Adjust filename_end so that filename + insert fits in 99 chars. */ 1445 suffix_length -= insert_length; 1446 if (filename_end > filename + suffix_length) 1447 filename_end = filename + suffix_length; 1448 /* Calculate max size for "suffix" section (#3 above). */ 1449 suffix_length -= filename_end - filename; 1450 1451 /* Step 2: Locate the "prefix" section of the dirname, including 1452 * trailing '/'. */ 1453 prefix = src; 1454 prefix_end = prefix + 155; 1455 if (prefix_end > filename) 1456 prefix_end = filename; 1457 while (prefix_end > prefix && *prefix_end != '/') 1458 prefix_end--; 1459 if ((prefix_end < filename) && (*prefix_end == '/')) 1460 prefix_end++; 1461 1462 /* Step 3: Locate the "suffix" section of the dirname, 1463 * including trailing '/'. */ 1464 suffix = prefix_end; 1465 suffix_end = suffix + suffix_length; /* Enforce limit. */ 1466 if (suffix_end > filename) 1467 suffix_end = filename; 1468 if (suffix_end < suffix) 1469 suffix_end = suffix; 1470 while (suffix_end > suffix && *suffix_end != '/') 1471 suffix_end--; 1472 if ((suffix_end < filename) && (*suffix_end == '/')) 1473 suffix_end++; 1474 1475 /* Step 4: Build the new name. */ 1476 /* The OpenBSD strlcpy function is safer, but less portable. */ 1477 /* Rather than maintain two versions, just use the strncpy version. */ 1478 p = dest; 1479 if (prefix_end > prefix) { 1480 strncpy(p, prefix, prefix_end - prefix); 1481 p += prefix_end - prefix; 1482 } 1483 if (suffix_end > suffix) { 1484 strncpy(p, suffix, suffix_end - suffix); 1485 p += suffix_end - suffix; 1486 } 1487 if (insert != NULL) { 1488 /* Note: assume insert does not have leading or trailing '/' */ 1489 strcpy(p, insert); 1490 p += strlen(insert); 1491 *p++ = '/'; 1492 } 1493 strncpy(p, filename, filename_end - filename); 1494 p += filename_end - filename; 1495 if (need_slash) 1496 *p++ = '/'; 1497 *p = '\0'; 1498 1499 return (dest); 1500} 1501 1502/* 1503 * The ustar header for the pax extended attributes must have a 1504 * reasonable name: SUSv3 requires 'dirname'/PaxHeader.'pid'/'filename' 1505 * where 'pid' is the PID of the archiving process. Unfortunately, 1506 * that makes testing a pain since the output varies for each run, 1507 * so I'm sticking with the simpler 'dirname'/PaxHeader/'filename' 1508 * for now. (Someday, I'll make this settable. Then I can use the 1509 * SUS recommendation as default and test harnesses can override it 1510 * to get predictable results.) 1511 * 1512 * Joerg Schilling has argued that this is unnecessary because, in 1513 * practice, if the pax extended attributes get extracted as regular 1514 * files, no one is going to bother reading those attributes to 1515 * manually restore them. Based on this, 'star' uses 1516 * /tmp/PaxHeader/'basename' as the ustar header name. This is a 1517 * tempting argument, in part because it's simpler than the SUSv3 1518 * recommendation, but I'm not entirely convinced. I'm also 1519 * uncomfortable with the fact that "/tmp" is a Unix-ism. 1520 * 1521 * The following routine leverages build_ustar_entry_name() above and 1522 * so is simpler than you might think. It just needs to provide the 1523 * additional path element and handle a few pathological cases). 1524 */ 1525static char * 1526build_pax_attribute_name(char *dest, const char *src) 1527{ 1528 char buff[64]; 1529 const char *p; 1530 1531 /* Handle the null filename case. */ 1532 if (src == NULL || *src == '\0') { 1533 strcpy(dest, "PaxHeader/blank"); 1534 return (dest); 1535 } 1536 1537 /* Prune final '/' and other unwanted final elements. */ 1538 p = src + strlen(src); 1539 for (;;) { 1540 /* Ends in "/", remove the '/' */ 1541 if (p > src && p[-1] == '/') { 1542 --p; 1543 continue; 1544 } 1545 /* Ends in "/.", remove the '.' */ 1546 if (p > src + 1 && p[-1] == '.' 1547 && p[-2] == '/') { 1548 --p; 1549 continue; 1550 } 1551 break; 1552 } 1553 1554 /* Pathological case: After above, there was nothing left. 1555 * This includes "/." "/./." "/.//./." etc. */ 1556 if (p == src) { 1557 strcpy(dest, "/PaxHeader/rootdir"); 1558 return (dest); 1559 } 1560 1561 /* Convert unadorned "." into a suitable filename. */ 1562 if (*src == '.' && p == src + 1) { 1563 strcpy(dest, "PaxHeader/currentdir"); 1564 return (dest); 1565 } 1566 1567 /* 1568 * TODO: Push this string into the 'pax' structure to avoid 1569 * recomputing it every time. That will also open the door 1570 * to having clients override it. 1571 */ 1572#if HAVE_GETPID && 0 /* Disable this for now; see above comment. */ 1573 sprintf(buff, "PaxHeader.%d", getpid()); 1574#else 1575 /* If the platform can't fetch the pid, don't include it. */ 1576 strcpy(buff, "PaxHeader"); 1577#endif 1578 /* General case: build a ustar-compatible name adding 1579 * "/PaxHeader/". */ 1580 build_ustar_entry_name(dest, src, p - src, buff); 1581 1582 return (dest); 1583} 1584 1585/* 1586 * GNU PAX Format 1.0 requires the special name, which pattern is: 1587 * <dir>/GNUSparseFile.<pid>/<original file name> 1588 * 1589 * This function is used for only Sparse file, a file type of which 1590 * is regular file. 1591 */ 1592static char * 1593build_gnu_sparse_name(char *dest, const char *src) 1594{ 1595 char buff[64]; 1596 const char *p; 1597 1598 /* Handle the null filename case. */ 1599 if (src == NULL || *src == '\0') { 1600 strcpy(dest, "GNUSparseFile/blank"); 1601 return (dest); 1602 } 1603 1604 /* Prune final '/' and other unwanted final elements. */ 1605 p = src + strlen(src); 1606 for (;;) { 1607 /* Ends in "/", remove the '/' */ 1608 if (p > src && p[-1] == '/') { 1609 --p; 1610 continue; 1611 } 1612 /* Ends in "/.", remove the '.' */ 1613 if (p > src + 1 && p[-1] == '.' 1614 && p[-2] == '/') { 1615 --p; 1616 continue; 1617 } 1618 break; 1619 } 1620 1621#if HAVE_GETPID && 0 /* Disable this as pax attribute name. */ 1622 sprintf(buff, "GNUSparseFile.%d", getpid()); 1623#else 1624 /* If the platform can't fetch the pid, don't include it. */ 1625 strcpy(buff, "GNUSparseFile"); 1626#endif 1627 /* General case: build a ustar-compatible name adding 1628 * "/GNUSparseFile/". */ 1629 build_ustar_entry_name(dest, src, p - src, buff); 1630 1631 return (dest); 1632} 1633 1634/* Write two null blocks for the end of archive */ 1635static int 1636archive_write_pax_close(struct archive_write *a) 1637{ 1638 return (__archive_write_nulls(a, 512 * 2)); 1639} 1640 1641static int 1642archive_write_pax_free(struct archive_write *a) 1643{ 1644 struct pax *pax; 1645 1646 pax = (struct pax *)a->format_data; 1647 if (pax == NULL) 1648 return (ARCHIVE_OK); 1649 1650 archive_string_free(&pax->pax_header); 1651 archive_string_free(&pax->sparse_map); 1652 archive_string_free(&pax->l_url_encoded_name); 1653 sparse_list_clear(pax); 1654 free(pax); 1655 a->format_data = NULL; 1656 return (ARCHIVE_OK); 1657} 1658 1659static int 1660archive_write_pax_finish_entry(struct archive_write *a) 1661{ 1662 struct pax *pax; 1663 uint64_t remaining; 1664 int ret; 1665 1666 pax = (struct pax *)a->format_data; 1667 remaining = pax->entry_bytes_remaining; 1668 if (remaining == 0) { 1669 while (pax->sparse_list) { 1670 struct sparse_block *sb; 1671 if (!pax->sparse_list->is_hole) 1672 remaining += pax->sparse_list->remaining; 1673 sb = pax->sparse_list->next; 1674 free(pax->sparse_list); 1675 pax->sparse_list = sb; 1676 } 1677 } 1678 ret = __archive_write_nulls(a, (size_t)(remaining + pax->entry_padding)); 1679 pax->entry_bytes_remaining = pax->entry_padding = 0; 1680 return (ret); 1681} 1682 1683static ssize_t 1684archive_write_pax_data(struct archive_write *a, const void *buff, size_t s) 1685{ 1686 struct pax *pax; 1687 size_t ws; 1688 size_t total; 1689 int ret; 1690 1691 pax = (struct pax *)a->format_data; 1692 1693 /* 1694 * According to GNU PAX format 1.0, write a sparse map 1695 * before the body. 1696 */ 1697 if (archive_strlen(&(pax->sparse_map))) { 1698 ret = __archive_write_output(a, pax->sparse_map.s, 1699 archive_strlen(&(pax->sparse_map))); 1700 if (ret != ARCHIVE_OK) 1701 return (ret); 1702 ret = __archive_write_nulls(a, pax->sparse_map_padding); 1703 if (ret != ARCHIVE_OK) 1704 return (ret); 1705 archive_string_empty(&(pax->sparse_map)); 1706 } 1707 1708 total = 0; 1709 while (total < s) { 1710 const unsigned char *p; 1711 1712 while (pax->sparse_list != NULL && 1713 pax->sparse_list->remaining == 0) { 1714 struct sparse_block *sb = pax->sparse_list->next; 1715 free(pax->sparse_list); 1716 pax->sparse_list = sb; 1717 } 1718 1719 if (pax->sparse_list == NULL) 1720 return (total); 1721 1722 p = ((const unsigned char *)buff) + total; 1723 ws = s - total; 1724 if (ws > pax->sparse_list->remaining) 1725 ws = (size_t)pax->sparse_list->remaining; 1726 1727 if (pax->sparse_list->is_hole) { 1728 /* Current block is hole thus we do not write 1729 * the body. */ 1730 pax->sparse_list->remaining -= ws; 1731 total += ws; 1732 continue; 1733 } 1734 1735 ret = __archive_write_output(a, p, ws); 1736 pax->sparse_list->remaining -= ws; 1737 total += ws; 1738 if (ret != ARCHIVE_OK) 1739 return (ret); 1740 } 1741 return (total); 1742} 1743 1744static int 1745has_non_ASCII(const char *_p) 1746{ 1747 const unsigned char *p = (const unsigned char *)_p; 1748 1749 if (p == NULL) 1750 return (1); 1751 while (*p != '\0' && *p < 128) 1752 p++; 1753 return (*p != '\0'); 1754} 1755 1756/* 1757 * Used by extended attribute support; encodes the name 1758 * so that there will be no '=' characters in the result. 1759 */ 1760static char * 1761url_encode(const char *in) 1762{ 1763 const char *s; 1764 char *d; 1765 int out_len = 0; 1766 char *out; 1767 1768 for (s = in; *s != '\0'; s++) { 1769 if (*s < 33 || *s > 126 || *s == '%' || *s == '=') 1770 out_len += 3; 1771 else 1772 out_len++; 1773 } 1774 1775 out = (char *)malloc(out_len + 1); 1776 if (out == NULL) 1777 return (NULL); 1778 1779 for (s = in, d = out; *s != '\0'; s++) { 1780 /* encode any non-printable ASCII character or '%' or '=' */ 1781 if (*s < 33 || *s > 126 || *s == '%' || *s == '=') { 1782 /* URL encoding is '%' followed by two hex digits */ 1783 *d++ = '%'; 1784 *d++ = "0123456789ABCDEF"[0x0f & (*s >> 4)]; 1785 *d++ = "0123456789ABCDEF"[0x0f & *s]; 1786 } else { 1787 *d++ = *s; 1788 } 1789 } 1790 *d = '\0'; 1791 return (out); 1792} 1793 1794/* 1795 * Encode a sequence of bytes into a C string using base-64 encoding. 1796 * 1797 * Returns a null-terminated C string allocated with malloc(); caller 1798 * is responsible for freeing the result. 1799 */ 1800static char * 1801base64_encode(const char *s, size_t len) 1802{ 1803 static const char digits[64] = 1804 { 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O', 1805 'P','Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d', 1806 'e','f','g','h','i','j','k','l','m','n','o','p','q','r','s', 1807 't','u','v','w','x','y','z','0','1','2','3','4','5','6','7', 1808 '8','9','+','/' }; 1809 int v; 1810 char *d, *out; 1811 1812 /* 3 bytes becomes 4 chars, but round up and allow for trailing NUL */ 1813 out = (char *)malloc((len * 4 + 2) / 3 + 1); 1814 if (out == NULL) 1815 return (NULL); 1816 d = out; 1817 1818 /* Convert each group of 3 bytes into 4 characters. */ 1819 while (len >= 3) { 1820 v = (((int)s[0] << 16) & 0xff0000) 1821 | (((int)s[1] << 8) & 0xff00) 1822 | (((int)s[2]) & 0x00ff); 1823 s += 3; 1824 len -= 3; 1825 *d++ = digits[(v >> 18) & 0x3f]; 1826 *d++ = digits[(v >> 12) & 0x3f]; 1827 *d++ = digits[(v >> 6) & 0x3f]; 1828 *d++ = digits[(v) & 0x3f]; 1829 } 1830 /* Handle final group of 1 byte (2 chars) or 2 bytes (3 chars). */ 1831 switch (len) { 1832 case 0: break; 1833 case 1: 1834 v = (((int)s[0] << 16) & 0xff0000); 1835 *d++ = digits[(v >> 18) & 0x3f]; 1836 *d++ = digits[(v >> 12) & 0x3f]; 1837 break; 1838 case 2: 1839 v = (((int)s[0] << 16) & 0xff0000) 1840 | (((int)s[1] << 8) & 0xff00); 1841 *d++ = digits[(v >> 18) & 0x3f]; 1842 *d++ = digits[(v >> 12) & 0x3f]; 1843 *d++ = digits[(v >> 6) & 0x3f]; 1844 break; 1845 } 1846 /* Add trailing NUL character so output is a valid C string. */ 1847 *d = '\0'; 1848 return (out); 1849} 1850 1851static void 1852sparse_list_clear(struct pax *pax) 1853{ 1854 while (pax->sparse_list != NULL) { 1855 struct sparse_block *sb = pax->sparse_list; 1856 pax->sparse_list = sb->next; 1857 free(sb); 1858 } 1859 pax->sparse_tail = NULL; 1860} 1861 1862static int 1863_sparse_list_add_block(struct pax *pax, int64_t offset, int64_t length, 1864 int is_hole) 1865{ 1866 struct sparse_block *sb; 1867 1868 sb = (struct sparse_block *)malloc(sizeof(*sb)); 1869 if (sb == NULL) 1870 return (ARCHIVE_FATAL); 1871 sb->next = NULL; 1872 sb->is_hole = is_hole; 1873 sb->offset = offset; 1874 sb->remaining = length; 1875 if (pax->sparse_list == NULL || pax->sparse_tail == NULL) 1876 pax->sparse_list = pax->sparse_tail = sb; 1877 else { 1878 pax->sparse_tail->next = sb; 1879 pax->sparse_tail = sb; 1880 } 1881 return (ARCHIVE_OK); 1882} 1883 1884static int 1885sparse_list_add(struct pax *pax, int64_t offset, int64_t length) 1886{ 1887 int64_t last_offset; 1888 int r; 1889 1890 if (pax->sparse_tail == NULL) 1891 last_offset = 0; 1892 else { 1893 last_offset = pax->sparse_tail->offset + 1894 pax->sparse_tail->remaining; 1895 } 1896 if (last_offset < offset) { 1897 /* Add a hole block. */ 1898 r = _sparse_list_add_block(pax, last_offset, 1899 offset - last_offset, 1); 1900 if (r != ARCHIVE_OK) 1901 return (r); 1902 } 1903 /* Add data block. */ 1904 return (_sparse_list_add_block(pax, offset, length, 0)); 1905} 1906 1907