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: head/lib/libarchive/archive_write_set_format_pax.c 201162 2009-12-29 05:47:46Z kientzle $"); 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 46 struct sparse_block { 47 struct sparse_block *next; 48 int is_hole; 49 uint64_t offset; 50 uint64_t remaining; 51 }; 52 53 struct 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 66 static void add_pax_attr(struct archive_string *, const char *key, 67 const char *value); 68 static void add_pax_attr_int(struct archive_string *, 69 const char *key, int64_t value); 70 static void add_pax_attr_time(struct archive_string *, 71 const char *key, int64_t sec, 72 unsigned long nanos); 73 static ssize_t archive_write_pax_data(struct archive_write *, 74 const void *, size_t); 75 static int archive_write_pax_close(struct archive_write *); 76 static int archive_write_pax_free(struct archive_write *); 77 static int archive_write_pax_finish_entry(struct archive_write *); 78 static int archive_write_pax_header(struct archive_write *, 79 struct archive_entry *); 80 static int archive_write_pax_options(struct archive_write *, 81 const char *, const char *); 82 static char *base64_encode(const char *src, size_t len); 83 static char *build_gnu_sparse_name(char *dest, const char *src); 84 static char *build_pax_attribute_name(char *dest, const char *src); 85 static char *build_ustar_entry_name(char *dest, const char *src, 86 size_t src_length, const char *insert); 87 static char *format_int(char *dest, int64_t); 88 static int has_non_ASCII(const char *); 89 static void sparse_list_clear(struct pax *); 90 static int sparse_list_add(struct pax *, int64_t, int64_t); 91 static 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 */ 100 int 101 archive_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 */ 118 int 119 archive_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 *)malloc(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 memset(pax, 0, sizeof(*pax)); 137 a->format_data = pax; 138 a->format_name = "pax"; 139 a->format_options = archive_write_pax_options; 140 a->format_write_header = archive_write_pax_header; 141 a->format_write_data = archive_write_pax_data; 142 a->format_close = archive_write_pax_close; 143 a->format_free = archive_write_pax_free; 144 a->format_finish_entry = archive_write_pax_finish_entry; 145 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE; 146 a->archive.archive_format_name = "POSIX pax interchange"; 147 return (ARCHIVE_OK); 148 } 149 150 static int 151 archive_write_pax_options(struct archive_write *a, const char *key, 152 const char *val) 153 { 154 struct pax *pax = (struct pax *)a->format_data; 155 int ret = ARCHIVE_FAILED; 156 157 if (strcmp(key, "hdrcharset") == 0) { 158 /* 159 * The character-set we can use are defined in 160 * IEEE Std 1003.1-2001 161 */ 162 if (val == NULL || val[0] == 0) 163 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 164 "pax: hdrcharset option needs a character-set name"); 165 else if (strcmp(val, "BINARY") == 0 || 166 strcmp(val, "binary") == 0) { 167 /* 168 * Specify binary mode. We will not convert 169 * filenames, uname and gname to any charsets. 170 */ 171 pax->opt_binary = 1; 172 ret = ARCHIVE_OK; 173 } else if (strcmp(val, "UTF-8") == 0) { 174 /* 175 * Specify UTF-8 character-set to be used for 176 * filenames. This is almost the test that 177 * running platform supports the string conversion. 178 * Especially libarchive_test needs this trick for 179 * its test. 180 */ 181 pax->sconv_utf8 = archive_string_conversion_to_charset( 182 &(a->archive), "UTF-8", 0); 183 if (pax->sconv_utf8 == NULL) 184 ret = ARCHIVE_FATAL; 185 else 186 ret = ARCHIVE_OK; 187 } else 188 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 189 "pax: invalid charset name"); 190 return (ret); 191 } 192 193 /* Note: The "warn" return is just to inform the options 194 * supervisor that we didn't handle it. It will generate 195 * a suitable error if no one used this option. */ 196 return (ARCHIVE_WARN); 197 } 198 199 /* 200 * Note: This code assumes that 'nanos' has the same sign as 'sec', 201 * which implies that sec=-1, nanos=200000000 represents -1.2 seconds 202 * and not -0.8 seconds. This is a pretty pedantic point, as we're 203 * unlikely to encounter many real files created before Jan 1, 1970, 204 * much less ones with timestamps recorded to sub-second resolution. 205 */ 206 static void 207 add_pax_attr_time(struct archive_string *as, const char *key, 208 int64_t sec, unsigned long nanos) 209 { 210 int digit, i; 211 char *t; 212 /* 213 * Note that each byte contributes fewer than 3 base-10 214 * digits, so this will always be big enough. 215 */ 216 char tmp[1 + 3*sizeof(sec) + 1 + 3*sizeof(nanos)]; 217 218 tmp[sizeof(tmp) - 1] = 0; 219 t = tmp + sizeof(tmp) - 1; 220 221 /* Skip trailing zeros in the fractional part. */ 222 for (digit = 0, i = 10; i > 0 && digit == 0; i--) { 223 digit = nanos % 10; 224 nanos /= 10; 225 } 226 227 /* Only format the fraction if it's non-zero. */ 228 if (i > 0) { 229 while (i > 0) { 230 *--t = "0123456789"[digit]; 231 digit = nanos % 10; 232 nanos /= 10; 233 i--; 234 } 235 *--t = '.'; 236 } 237 t = format_int(t, sec); 238 239 add_pax_attr(as, key, t); 240 } 241 242 static char * 243 format_int(char *t, int64_t i) 244 { 245 uint64_t ui; 246 247 if (i < 0) 248 ui = (i == INT64_MIN) ? (uint64_t)(INT64_MAX) + 1 : (uint64_t)(-i); 249 else 250 ui = i; 251 252 do { 253 *--t = "0123456789"[ui % 10]; 254 } while (ui /= 10); 255 if (i < 0) 256 *--t = '-'; 257 return (t); 258 } 259 260 static void 261 add_pax_attr_int(struct archive_string *as, const char *key, int64_t value) 262 { 263 char tmp[1 + 3 * sizeof(value)]; 264 265 tmp[sizeof(tmp) - 1] = 0; 266 add_pax_attr(as, key, format_int(tmp + sizeof(tmp) - 1, value)); 267 } 268 269 /* 270 * Add a key/value attribute to the pax header. This function handles 271 * the length field and various other syntactic requirements. 272 */ 273 static void 274 add_pax_attr(struct archive_string *as, const char *key, const char *value) 275 { 276 int digits, i, len, next_ten; 277 char tmp[1 + 3 * sizeof(int)]; /* < 3 base-10 digits per byte */ 278 279 /*- 280 * PAX attributes have the following layout: 281 * <len> <space> <key> <=> <value> <nl> 282 */ 283 len = 1 + (int)strlen(key) + 1 + (int)strlen(value) + 1; 284 285 /* 286 * The <len> field includes the length of the <len> field, so 287 * computing the correct length is tricky. I start by 288 * counting the number of base-10 digits in 'len' and 289 * computing the next higher power of 10. 290 */ 291 next_ten = 1; 292 digits = 0; 293 i = len; 294 while (i > 0) { 295 i = i / 10; 296 digits++; 297 next_ten = next_ten * 10; 298 } 299 /* 300 * For example, if string without the length field is 99 301 * chars, then adding the 2 digit length "99" will force the 302 * total length past 100, requiring an extra digit. The next 303 * statement adjusts for this effect. 304 */ 305 if (len + digits >= next_ten) 306 digits++; 307 308 /* Now, we have the right length so we can build the line. */ 309 tmp[sizeof(tmp) - 1] = 0; /* Null-terminate the work area. */ 310 archive_strcat(as, format_int(tmp + sizeof(tmp) - 1, len + digits)); 311 archive_strappend_char(as, ' '); 312 archive_strcat(as, key); 313 archive_strappend_char(as, '='); 314 archive_strcat(as, value); 315 archive_strappend_char(as, '\n'); 316 } 317 318 static int 319 archive_write_pax_header_xattrs(struct archive_write *a, 320 struct pax *pax, struct archive_entry *entry) 321 { 322 struct archive_string s; 323 int i = archive_entry_xattr_reset(entry); 324 325 while (i--) { 326 const char *name; 327 const void *value; 328 char *encoded_value; 329 char *url_encoded_name = NULL, *encoded_name = NULL; 330 size_t size; 331 int r; 332 333 archive_entry_xattr_next(entry, &name, &value, &size); 334 url_encoded_name = url_encode(name); 335 if (url_encoded_name != NULL) { 336 /* Convert narrow-character to UTF-8. */ 337 r = archive_strcpy_l(&(pax->l_url_encoded_name), 338 url_encoded_name, pax->sconv_utf8); 339 free(url_encoded_name); /* Done with this. */ 340 if (r == 0) 341 encoded_name = pax->l_url_encoded_name.s; 342 else if (errno == ENOMEM) { 343 archive_set_error(&a->archive, ENOMEM, 344 "Can't allocate memory for Linkname"); 345 return (ARCHIVE_FATAL); 346 } 347 } 348 349 encoded_value = base64_encode((const char *)value, size); 350 351 if (encoded_name != NULL && encoded_value != NULL) { 352 archive_string_init(&s); 353 archive_strcpy(&s, "LIBARCHIVE.xattr."); 354 archive_strcat(&s, encoded_name); 355 add_pax_attr(&(pax->pax_header), s.s, encoded_value); 356 archive_string_free(&s); 357 } 358 free(encoded_value); 359 } 360 return (ARCHIVE_OK); 361 } 362 363 static int 364 get_entry_hardlink(struct archive_write *a, struct archive_entry *entry, 365 const char **name, size_t *length, struct archive_string_conv *sc) 366 { 367 int r; 368 369 r = archive_entry_hardlink_l(entry, name, length, sc); 370 if (r != 0) { 371 if (errno == ENOMEM) { 372 archive_set_error(&a->archive, ENOMEM, 373 "Can't allocate memory for Linkname"); 374 return (ARCHIVE_FATAL); 375 } 376 return (ARCHIVE_WARN); 377 } 378 return (ARCHIVE_OK); 379 } 380 381 static int 382 get_entry_pathname(struct archive_write *a, struct archive_entry *entry, 383 const char **name, size_t *length, struct archive_string_conv *sc) 384 { 385 int r; 386 387 r = archive_entry_pathname_l(entry, name, length, sc); 388 if (r != 0) { 389 if (errno == ENOMEM) { 390 archive_set_error(&a->archive, ENOMEM, 391 "Can't allocate memory for Pathname"); 392 return (ARCHIVE_FATAL); 393 } 394 return (ARCHIVE_WARN); 395 } 396 return (ARCHIVE_OK); 397 } 398 399 static int 400 get_entry_uname(struct archive_write *a, struct archive_entry *entry, 401 const char **name, size_t *length, struct archive_string_conv *sc) 402 { 403 int r; 404 405 r = archive_entry_uname_l(entry, name, length, sc); 406 if (r != 0) { 407 if (errno == ENOMEM) { 408 archive_set_error(&a->archive, ENOMEM, 409 "Can't allocate memory for Uname"); 410 return (ARCHIVE_FATAL); 411 } 412 return (ARCHIVE_WARN); 413 } 414 return (ARCHIVE_OK); 415 } 416 417 static int 418 get_entry_gname(struct archive_write *a, struct archive_entry *entry, 419 const char **name, size_t *length, struct archive_string_conv *sc) 420 { 421 int r; 422 423 r = archive_entry_gname_l(entry, name, length, sc); 424 if (r != 0) { 425 if (errno == ENOMEM) { 426 archive_set_error(&a->archive, ENOMEM, 427 "Can't allocate memory for Gname"); 428 return (ARCHIVE_FATAL); 429 } 430 return (ARCHIVE_WARN); 431 } 432 return (ARCHIVE_OK); 433 } 434 435 static int 436 get_entry_symlink(struct archive_write *a, struct archive_entry *entry, 437 const char **name, size_t *length, struct archive_string_conv *sc) 438 { 439 int r; 440 441 r = archive_entry_symlink_l(entry, name, length, sc); 442 if (r != 0) { 443 if (errno == ENOMEM) { 444 archive_set_error(&a->archive, ENOMEM, 445 "Can't allocate memory for Linkname"); 446 return (ARCHIVE_FATAL); 447 } 448 return (ARCHIVE_WARN); 449 } 450 return (ARCHIVE_OK); 451 } 452 453 /* 454 * TODO: Consider adding 'comment' and 'charset' fields to 455 * archive_entry so that clients can specify them. Also, consider 456 * adding generic key/value tags so clients can add arbitrary 457 * key/value data. 458 * 459 * TODO: Break up this 700-line function!!!! Yowza! 460 */ 461 static int 462 archive_write_pax_header(struct archive_write *a, 463 struct archive_entry *entry_original) 464 { 465 struct archive_entry *entry_main; 466 const char *p; 467 const char *suffix; 468 int need_extension, r, ret; 469 int sparse_count; 470 uint64_t sparse_total, real_size; 471 struct pax *pax; 472 const char *hardlink; 473 const char *path = NULL, *linkpath = NULL; 474 const char *uname = NULL, *gname = NULL; 475 const void *mac_metadata; 476 size_t mac_metadata_size; 477 struct archive_string_conv *sconv; 478 size_t hardlink_length, path_length, linkpath_length; 479 size_t uname_length, gname_length; 480 481 char paxbuff[512]; 482 char ustarbuff[512]; 483 char ustar_entry_name[256]; 484 char pax_entry_name[256]; 485 char gnu_sparse_name[256]; 486 struct archive_string entry_name; 487 488 ret = ARCHIVE_OK; 489 need_extension = 0; 490 pax = (struct pax *)a->format_data; 491 492 /* Sanity check. */ 493 if (archive_entry_pathname(entry_original) == NULL) { 494 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 495 "Can't record entry in tar file without pathname"); 496 return (ARCHIVE_FAILED); 497 } 498 499 /* 500 * Choose a header encoding. 501 */ 502 if (pax->opt_binary) 503 sconv = NULL;/* Binary mode. */ 504 else { 505 /* Header encoding is UTF-8. */ 506 if (pax->sconv_utf8 == NULL) { 507 /* Initialize the string conversion object 508 * we must need */ 509 pax->sconv_utf8 = archive_string_conversion_to_charset( 510 &(a->archive), "UTF-8", 1); 511 if (pax->sconv_utf8 == NULL) 512 /* Couldn't allocate memory */ 513 return (ARCHIVE_FAILED); 514 } 515 sconv = pax->sconv_utf8; 516 } 517 518 r = get_entry_hardlink(a, entry_original, &hardlink, 519 &hardlink_length, sconv); 520 if (r == ARCHIVE_FATAL) 521 return (r); 522 else if (r != ARCHIVE_OK) { 523 r = get_entry_hardlink(a, entry_original, &hardlink, 524 &hardlink_length, NULL); 525 if (r == ARCHIVE_FATAL) 526 return (r); 527 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 528 "Can't translate linkname '%s' to %s", hardlink, 529 archive_string_conversion_charset_name(sconv)); 530 ret = ARCHIVE_WARN; 531 sconv = NULL;/* The header charset switches to binary mode. */ 532 } 533 534 /* Make sure this is a type of entry that we can handle here */ 535 if (hardlink == NULL) { 536 switch (archive_entry_filetype(entry_original)) { 537 case AE_IFBLK: 538 case AE_IFCHR: 539 case AE_IFIFO: 540 case AE_IFLNK: 541 case AE_IFREG: 542 break; 543 case AE_IFDIR: 544 { 545 /* 546 * Ensure a trailing '/'. Modify the original 547 * entry so the client sees the change. 548 */ 549 #if defined(_WIN32) && !defined(__CYGWIN__) 550 const wchar_t *wp; 551 552 wp = archive_entry_pathname_w(entry_original); 553 if (wp != NULL && wp[wcslen(wp) -1] != L'/') { 554 struct archive_wstring ws; 555 556 archive_string_init(&ws); 557 path_length = wcslen(wp); 558 if (archive_wstring_ensure(&ws, 559 path_length + 2) == NULL) { 560 archive_set_error(&a->archive, ENOMEM, 561 "Can't allocate pax data"); 562 archive_wstring_free(&ws); 563 return(ARCHIVE_FATAL); 564 } 565 /* Should we keep '\' ? */ 566 if (wp[path_length -1] == L'\\') 567 path_length--; 568 archive_wstrncpy(&ws, wp, path_length); 569 archive_wstrappend_wchar(&ws, L'/'); 570 archive_entry_copy_pathname_w( 571 entry_original, ws.s); 572 archive_wstring_free(&ws); 573 p = NULL; 574 } else 575 #endif 576 p = archive_entry_pathname(entry_original); 577 /* 578 * On Windows, this is a backup operation just in 579 * case getting WCS failed. On POSIX, this is a 580 * normal operation. 581 */ 582 if (p != NULL && p[strlen(p) - 1] != '/') { 583 struct archive_string as; 584 585 archive_string_init(&as); 586 path_length = strlen(p); 587 if (archive_string_ensure(&as, 588 path_length + 2) == NULL) { 589 archive_set_error(&a->archive, ENOMEM, 590 "Can't allocate pax data"); 591 archive_string_free(&as); 592 return(ARCHIVE_FATAL); 593 } 594 #if defined(_WIN32) && !defined(__CYGWIN__) 595 /* NOTE: This might break the pathname 596 * if the current code page is CP932 and 597 * the pathname includes a character '\' 598 * as a part of its multibyte pathname. */ 599 if (p[strlen(p) -1] == '\\') 600 path_length--; 601 else 602 #endif 603 archive_strncpy(&as, p, path_length); 604 archive_strappend_char(&as, '/'); 605 archive_entry_copy_pathname( 606 entry_original, as.s); 607 archive_string_free(&as); 608 } 609 break; 610 } 611 case AE_IFSOCK: 612 archive_set_error(&a->archive, 613 ARCHIVE_ERRNO_FILE_FORMAT, 614 "tar format cannot archive socket"); 615 return (ARCHIVE_FAILED); 616 default: 617 archive_set_error(&a->archive, 618 ARCHIVE_ERRNO_FILE_FORMAT, 619 "tar format cannot archive this (type=0%lo)", 620 (unsigned long) 621 archive_entry_filetype(entry_original)); 622 return (ARCHIVE_FAILED); 623 } 624 } 625 626 /* 627 * If Mac OS metadata blob is here, recurse to write that 628 * as a separate entry. This is really a pretty poor design: 629 * In particular, it doubles the overhead for long filenames. 630 * TODO: Help Apple folks design something better and figure 631 * out how to transition from this legacy format. 632 * 633 * Note that this code is present on every platform; clients 634 * on non-Mac are unlikely to ever provide this data, but 635 * applications that copy entries from one archive to another 636 * should not lose data just because the local filesystem 637 * can't store it. 638 */ 639 mac_metadata = 640 archive_entry_mac_metadata(entry_original, &mac_metadata_size); 641 if (mac_metadata != NULL) { 642 const char *oname; 643 char *name, *bname; 644 size_t name_length; 645 struct archive_entry *extra = archive_entry_new2(&a->archive); 646 647 oname = archive_entry_pathname(entry_original); 648 name_length = strlen(oname); 649 name = malloc(name_length + 3); 650 if (name == NULL || extra == NULL) { 651 /* XXX error message */ 652 archive_entry_free(extra); 653 free(name); 654 return (ARCHIVE_FAILED); 655 } 656 strcpy(name, oname); 657 /* Find last '/'; strip trailing '/' characters */ 658 bname = strrchr(name, '/'); 659 while (bname != NULL && bname[1] == '\0') { 660 *bname = '\0'; 661 bname = strrchr(name, '/'); 662 } 663 if (bname == NULL) { 664 memmove(name + 2, name, name_length + 1); 665 memmove(name, "._", 2); 666 } else { 667 bname += 1; 668 memmove(bname + 2, bname, strlen(bname) + 1); 669 memmove(bname, "._", 2); 670 } 671 archive_entry_copy_pathname(extra, name); 672 free(name); 673 674 archive_entry_set_size(extra, mac_metadata_size); 675 archive_entry_set_filetype(extra, AE_IFREG); 676 archive_entry_set_perm(extra, 677 archive_entry_perm(entry_original)); 678 archive_entry_set_mtime(extra, 679 archive_entry_mtime(entry_original), 680 archive_entry_mtime_nsec(entry_original)); 681 archive_entry_set_gid(extra, 682 archive_entry_gid(entry_original)); 683 archive_entry_set_gname(extra, 684 archive_entry_gname(entry_original)); 685 archive_entry_set_uid(extra, 686 archive_entry_uid(entry_original)); 687 archive_entry_set_uname(extra, 688 archive_entry_uname(entry_original)); 689 690 /* Recurse to write the special copyfile entry. */ 691 r = archive_write_pax_header(a, extra); 692 archive_entry_free(extra); 693 if (r < ARCHIVE_WARN) 694 return (r); 695 if (r < ret) 696 ret = r; 697 r = (int)archive_write_pax_data(a, mac_metadata, 698 mac_metadata_size); 699 if (r < ARCHIVE_WARN) 700 return (r); 701 if (r < ret) 702 ret = r; 703 r = archive_write_pax_finish_entry(a); 704 if (r < ARCHIVE_WARN) 705 return (r); 706 if (r < ret) 707 ret = r; 708 } 709 710 /* Copy entry so we can modify it as needed. */ 711 #if defined(_WIN32) && !defined(__CYGWIN__) 712 /* Make sure the path separators in pahtname, hardlink and symlink 713 * are all slash '/', not the Windows path separator '\'. */ 714 entry_main = __la_win_entry_in_posix_pathseparator(entry_original); 715 if (entry_main == entry_original) 716 entry_main = archive_entry_clone(entry_original); 717 #else 718 entry_main = archive_entry_clone(entry_original); 719 #endif 720 if (entry_main == NULL) { 721 archive_set_error(&a->archive, ENOMEM, 722 "Can't allocate pax data"); 723 return(ARCHIVE_FATAL); 724 } 725 archive_string_empty(&(pax->pax_header)); /* Blank our work area. */ 726 archive_string_empty(&(pax->sparse_map)); 727 sparse_total = 0; 728 sparse_list_clear(pax); 729 730 if (hardlink == NULL && 731 archive_entry_filetype(entry_main) == AE_IFREG) 732 sparse_count = archive_entry_sparse_reset(entry_main); 733 else 734 sparse_count = 0; 735 if (sparse_count) { 736 int64_t offset, length, last_offset = 0; 737 /* Get the last entry of sparse block. */ 738 while (archive_entry_sparse_next( 739 entry_main, &offset, &length) == ARCHIVE_OK) 740 last_offset = offset + length; 741 742 /* If the last sparse block does not reach the end of file, 743 * We have to add a empty sparse block as the last entry to 744 * manage storing file data. */ 745 if (last_offset < archive_entry_size(entry_main)) 746 archive_entry_sparse_add_entry(entry_main, 747 archive_entry_size(entry_main), 0); 748 sparse_count = archive_entry_sparse_reset(entry_main); 749 } 750 751 /* 752 * First, check the name fields and see if any of them 753 * require binary coding. If any of them does, then all of 754 * them do. 755 */ 756 r = get_entry_pathname(a, entry_main, &path, &path_length, sconv); 757 if (r == ARCHIVE_FATAL) 758 return (r); 759 else if (r != ARCHIVE_OK) { 760 r = get_entry_pathname(a, entry_main, &path, 761 &path_length, NULL); 762 if (r == ARCHIVE_FATAL) 763 return (r); 764 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 765 "Can't translate pathname '%s' to %s", path, 766 archive_string_conversion_charset_name(sconv)); 767 ret = ARCHIVE_WARN; 768 sconv = NULL;/* The header charset switches to binary mode. */ 769 } 770 r = get_entry_uname(a, entry_main, &uname, &uname_length, sconv); 771 if (r == ARCHIVE_FATAL) 772 return (r); 773 else if (r != ARCHIVE_OK) { 774 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL); 775 if (r == ARCHIVE_FATAL) 776 return (r); 777 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 778 "Can't translate uname '%s' to %s", uname, 779 archive_string_conversion_charset_name(sconv)); 780 ret = ARCHIVE_WARN; 781 sconv = NULL;/* The header charset switches to binary mode. */ 782 } 783 r = get_entry_gname(a, entry_main, &gname, &gname_length, sconv); 784 if (r == ARCHIVE_FATAL) 785 return (r); 786 else if (r != ARCHIVE_OK) { 787 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL); 788 if (r == ARCHIVE_FATAL) 789 return (r); 790 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 791 "Can't translate gname '%s' to %s", gname, 792 archive_string_conversion_charset_name(sconv)); 793 ret = ARCHIVE_WARN; 794 sconv = NULL;/* The header charset switches to binary mode. */ 795 } 796 linkpath = hardlink; 797 linkpath_length = hardlink_length; 798 if (linkpath == NULL) { 799 r = get_entry_symlink(a, entry_main, &linkpath, 800 &linkpath_length, sconv); 801 if (r == ARCHIVE_FATAL) 802 return (r); 803 else if (r != ARCHIVE_OK) { 804 r = get_entry_symlink(a, entry_main, &linkpath, 805 &linkpath_length, NULL); 806 if (r == ARCHIVE_FATAL) 807 return (r); 808 archive_set_error(&a->archive, 809 ARCHIVE_ERRNO_FILE_FORMAT, 810 "Can't translate linkname '%s' to %s", linkpath, 811 archive_string_conversion_charset_name(sconv)); 812 ret = ARCHIVE_WARN; 813 sconv = NULL; 814 } 815 } 816 817 /* If any string conversions failed, get all attributes 818 * in binary-mode. */ 819 if (sconv == NULL && !pax->opt_binary) { 820 if (hardlink != NULL) { 821 r = get_entry_hardlink(a, entry_main, &hardlink, 822 &hardlink_length, NULL); 823 if (r == ARCHIVE_FATAL) 824 return (r); 825 linkpath = hardlink; 826 linkpath_length = hardlink_length; 827 } 828 r = get_entry_pathname(a, entry_main, &path, 829 &path_length, NULL); 830 if (r == ARCHIVE_FATAL) 831 return (r); 832 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL); 833 if (r == ARCHIVE_FATAL) 834 return (r); 835 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL); 836 if (r == ARCHIVE_FATAL) 837 return (r); 838 } 839 840 /* Store the header encoding first, to be nice to readers. */ 841 if (sconv == NULL) 842 add_pax_attr(&(pax->pax_header), "hdrcharset", "BINARY"); 843 844 845 /* 846 * If name is too long, or has non-ASCII characters, add 847 * 'path' to pax extended attrs. (Note that an unconvertible 848 * name must have non-ASCII characters.) 849 */ 850 if (has_non_ASCII(path)) { 851 /* We have non-ASCII characters. */ 852 add_pax_attr(&(pax->pax_header), "path", path); 853 archive_entry_set_pathname(entry_main, 854 build_ustar_entry_name(ustar_entry_name, 855 path, path_length, NULL)); 856 need_extension = 1; 857 } else { 858 /* We have an all-ASCII path; we'd like to just store 859 * it in the ustar header if it will fit. Yes, this 860 * duplicates some of the logic in 861 * archive_write_set_format_ustar.c 862 */ 863 if (path_length <= 100) { 864 /* Fits in the old 100-char tar name field. */ 865 } else { 866 /* Find largest suffix that will fit. */ 867 /* Note: strlen() > 100, so strlen() - 100 - 1 >= 0 */ 868 suffix = strchr(path + path_length - 100 - 1, '/'); 869 /* Don't attempt an empty prefix. */ 870 if (suffix == path) 871 suffix = strchr(suffix + 1, '/'); 872 /* We can put it in the ustar header if it's 873 * all ASCII and it's either <= 100 characters 874 * or can be split at a '/' into a prefix <= 875 * 155 chars and a suffix <= 100 chars. (Note 876 * the strchr() above will return NULL exactly 877 * when the path can't be split.) 878 */ 879 if (suffix == NULL /* Suffix > 100 chars. */ 880 || suffix[1] == '\0' /* empty suffix */ 881 || suffix - path > 155) /* Prefix > 155 chars */ 882 { 883 add_pax_attr(&(pax->pax_header), "path", path); 884 archive_entry_set_pathname(entry_main, 885 build_ustar_entry_name(ustar_entry_name, 886 path, path_length, NULL)); 887 need_extension = 1; 888 } 889 } 890 } 891 892 if (linkpath != NULL) { 893 /* If link name is too long or has non-ASCII characters, add 894 * 'linkpath' to pax extended attrs. */ 895 if (linkpath_length > 100 || has_non_ASCII(linkpath)) { 896 add_pax_attr(&(pax->pax_header), "linkpath", linkpath); 897 if (linkpath_length > 100) { 898 if (hardlink != NULL) 899 archive_entry_set_hardlink(entry_main, 900 "././@LongHardLink"); 901 else 902 archive_entry_set_symlink(entry_main, 903 "././@LongSymLink"); 904 } 905 need_extension = 1; 906 } 907 } 908 /* Save a pathname since it will be renamed if `entry_main` has 909 * sparse blocks. */ 910 archive_string_init(&entry_name); 911 archive_strcpy(&entry_name, archive_entry_pathname(entry_main)); 912 913 /* If file size is too large, add 'size' to pax extended attrs. */ 914 if (archive_entry_size(entry_main) >= (((int64_t)1) << 33)) { 915 add_pax_attr_int(&(pax->pax_header), "size", 916 archive_entry_size(entry_main)); 917 need_extension = 1; 918 } 919 920 /* If numeric GID is too large, add 'gid' to pax extended attrs. */ 921 if ((unsigned int)archive_entry_gid(entry_main) >= (1 << 18)) { 922 add_pax_attr_int(&(pax->pax_header), "gid", 923 archive_entry_gid(entry_main)); 924 need_extension = 1; 925 } 926 927 /* If group name is too large or has non-ASCII characters, add 928 * 'gname' to pax extended attrs. */ 929 if (gname != NULL) { 930 if (gname_length > 31 || has_non_ASCII(gname)) { 931 add_pax_attr(&(pax->pax_header), "gname", gname); 932 need_extension = 1; 933 } 934 } 935 936 /* If numeric UID is too large, add 'uid' to pax extended attrs. */ 937 if ((unsigned int)archive_entry_uid(entry_main) >= (1 << 18)) { 938 add_pax_attr_int(&(pax->pax_header), "uid", 939 archive_entry_uid(entry_main)); 940 need_extension = 1; 941 } 942 943 /* Add 'uname' to pax extended attrs if necessary. */ 944 if (uname != NULL) { 945 if (uname_length > 31 || has_non_ASCII(uname)) { 946 add_pax_attr(&(pax->pax_header), "uname", uname); 947 need_extension = 1; 948 } 949 } 950 951 /* 952 * POSIX/SUSv3 doesn't provide a standard key for large device 953 * numbers. I use the same keys here that Joerg Schilling 954 * used for 'star.' (Which, somewhat confusingly, are called 955 * "devXXX" even though they code "rdev" values.) No doubt, 956 * other implementations use other keys. Note that there's no 957 * reason we can't write the same information into a number of 958 * different keys. 959 * 960 * Of course, this is only needed for block or char device entries. 961 */ 962 if (archive_entry_filetype(entry_main) == AE_IFBLK 963 || archive_entry_filetype(entry_main) == AE_IFCHR) { 964 /* 965 * If rdevmajor is too large, add 'SCHILY.devmajor' to 966 * extended attributes. 967 */ 968 int rdevmajor, rdevminor; 969 rdevmajor = archive_entry_rdevmajor(entry_main); 970 rdevminor = archive_entry_rdevminor(entry_main); 971 if (rdevmajor >= (1 << 18)) { 972 add_pax_attr_int(&(pax->pax_header), "SCHILY.devmajor", 973 rdevmajor); 974 /* 975 * Non-strict formatting below means we don't 976 * have to truncate here. Not truncating improves 977 * the chance that some more modern tar archivers 978 * (such as GNU tar 1.13) can restore the full 979 * value even if they don't understand the pax 980 * extended attributes. See my rant below about 981 * file size fields for additional details. 982 */ 983 /* archive_entry_set_rdevmajor(entry_main, 984 rdevmajor & ((1 << 18) - 1)); */ 985 need_extension = 1; 986 } 987 988 /* 989 * If devminor is too large, add 'SCHILY.devminor' to 990 * extended attributes. 991 */ 992 if (rdevminor >= (1 << 18)) { 993 add_pax_attr_int(&(pax->pax_header), "SCHILY.devminor", 994 rdevminor); 995 /* Truncation is not necessary here, either. */ 996 /* archive_entry_set_rdevminor(entry_main, 997 rdevminor & ((1 << 18) - 1)); */ 998 need_extension = 1; 999 } 1000 } 1001 1002 /* 1003 * Technically, the mtime field in the ustar header can 1004 * support 33 bits, but many platforms use signed 32-bit time 1005 * values. The cutoff of 0x7fffffff here is a compromise. 1006 * Yes, this check is duplicated just below; this helps to 1007 * avoid writing an mtime attribute just to handle a 1008 * high-resolution timestamp in "restricted pax" mode. 1009 */ 1010 if (!need_extension && 1011 ((archive_entry_mtime(entry_main) < 0) 1012 || (archive_entry_mtime(entry_main) >= 0x7fffffff))) 1013 need_extension = 1; 1014 1015 /* I use a star-compatible file flag attribute. */ 1016 p = archive_entry_fflags_text(entry_main); 1017 if (!need_extension && p != NULL && *p != '\0') 1018 need_extension = 1; 1019 1020 /* If there are non-trivial ACL entries, we need an extension. */ 1021 if (!need_extension && archive_entry_acl_count(entry_original, 1022 ARCHIVE_ENTRY_ACL_TYPE_ACCESS) > 0) 1023 need_extension = 1; 1024 1025 /* If there are non-trivial ACL entries, we need an extension. */ 1026 if (!need_extension && archive_entry_acl_count(entry_original, 1027 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT) > 0) 1028 need_extension = 1; 1029 1030 /* If there are extended attributes, we need an extension */ 1031 if (!need_extension && archive_entry_xattr_count(entry_original) > 0) 1032 need_extension = 1; 1033 1034 /* If there are sparse info, we need an extension */ 1035 if (!need_extension && sparse_count > 0) 1036 need_extension = 1; 1037 1038 /* 1039 * The following items are handled differently in "pax 1040 * restricted" format. In particular, in "pax restricted" 1041 * format they won't be added unless need_extension is 1042 * already set (we're already generating an extended header, so 1043 * may as well include these). 1044 */ 1045 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_RESTRICTED || 1046 need_extension) { 1047 1048 if (archive_entry_mtime(entry_main) < 0 || 1049 archive_entry_mtime(entry_main) >= 0x7fffffff || 1050 archive_entry_mtime_nsec(entry_main) != 0) 1051 add_pax_attr_time(&(pax->pax_header), "mtime", 1052 archive_entry_mtime(entry_main), 1053 archive_entry_mtime_nsec(entry_main)); 1054 1055 if (archive_entry_ctime(entry_main) != 0 || 1056 archive_entry_ctime_nsec(entry_main) != 0) 1057 add_pax_attr_time(&(pax->pax_header), "ctime", 1058 archive_entry_ctime(entry_main), 1059 archive_entry_ctime_nsec(entry_main)); 1060 1061 if (archive_entry_atime(entry_main) != 0 || 1062 archive_entry_atime_nsec(entry_main) != 0) 1063 add_pax_attr_time(&(pax->pax_header), "atime", 1064 archive_entry_atime(entry_main), 1065 archive_entry_atime_nsec(entry_main)); 1066 1067 /* Store birth/creationtime only if it's earlier than mtime */ 1068 if (archive_entry_birthtime_is_set(entry_main) && 1069 archive_entry_birthtime(entry_main) 1070 < archive_entry_mtime(entry_main)) 1071 add_pax_attr_time(&(pax->pax_header), 1072 "LIBARCHIVE.creationtime", 1073 archive_entry_birthtime(entry_main), 1074 archive_entry_birthtime_nsec(entry_main)); 1075 1076 /* I use a star-compatible file flag attribute. */ 1077 p = archive_entry_fflags_text(entry_main); 1078 if (p != NULL && *p != '\0') 1079 add_pax_attr(&(pax->pax_header), "SCHILY.fflags", p); 1080 1081 /* I use star-compatible ACL attributes. */ 1082 r = archive_entry_acl_text_l(entry_original, 1083 ARCHIVE_ENTRY_ACL_TYPE_ACCESS | 1084 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID, 1085 &p, NULL, pax->sconv_utf8); 1086 if (r != 0) { 1087 if (errno == ENOMEM) { 1088 archive_set_error(&a->archive, ENOMEM, 1089 "Can't allocate memory for " 1090 "ACL.access"); 1091 return (ARCHIVE_FATAL); 1092 } 1093 archive_set_error(&a->archive, 1094 ARCHIVE_ERRNO_FILE_FORMAT, 1095 "Can't translate ACL.access to UTF-8"); 1096 ret = ARCHIVE_WARN; 1097 } else if (p != NULL && *p != '\0') { 1098 add_pax_attr(&(pax->pax_header), 1099 "SCHILY.acl.access", p); 1100 } 1101 r = archive_entry_acl_text_l(entry_original, 1102 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT | 1103 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID, 1104 &p, NULL, pax->sconv_utf8); 1105 if (r != 0) { 1106 if (errno == ENOMEM) { 1107 archive_set_error(&a->archive, ENOMEM, 1108 "Can't allocate memory for " 1109 "ACL.default"); 1110 return (ARCHIVE_FATAL); 1111 } 1112 archive_set_error(&a->archive, 1113 ARCHIVE_ERRNO_FILE_FORMAT, 1114 "Can't translate ACL.default to UTF-8"); 1115 ret = ARCHIVE_WARN; 1116 } else if (p != NULL && *p != '\0') { 1117 add_pax_attr(&(pax->pax_header), 1118 "SCHILY.acl.default", p); 1119 } 1120 1121 /* We use GNU-tar-compatible sparse attributes. */ 1122 if (sparse_count > 0) { 1123 int64_t soffset, slength; 1124 1125 add_pax_attr_int(&(pax->pax_header), 1126 "GNU.sparse.major", 1); 1127 add_pax_attr_int(&(pax->pax_header), 1128 "GNU.sparse.minor", 0); 1129 add_pax_attr(&(pax->pax_header), 1130 "GNU.sparse.name", entry_name.s); 1131 add_pax_attr_int(&(pax->pax_header), 1132 "GNU.sparse.realsize", 1133 archive_entry_size(entry_main)); 1134 1135 /* Rename the file name which will be used for 1136 * ustar header to a special name, which GNU 1137 * PAX Format 1.0 requires */ 1138 archive_entry_set_pathname(entry_main, 1139 build_gnu_sparse_name(gnu_sparse_name, 1140 entry_name.s)); 1141 1142 /* 1143 * - Make a sparse map, which will precede a file data. 1144 * - Get the total size of available data of sparse. 1145 */ 1146 archive_string_sprintf(&(pax->sparse_map), "%d\n", 1147 sparse_count); 1148 while (archive_entry_sparse_next(entry_main, 1149 &soffset, &slength) == ARCHIVE_OK) { 1150 archive_string_sprintf(&(pax->sparse_map), 1151 "%jd\n%jd\n", 1152 (intmax_t)soffset, 1153 (intmax_t)slength); 1154 sparse_total += slength; 1155 if (sparse_list_add(pax, soffset, slength) 1156 != ARCHIVE_OK) { 1157 archive_set_error(&a->archive, 1158 ENOMEM, 1159 "Can't allocate memory"); 1160 archive_entry_free(entry_main); 1161 archive_string_free(&entry_name); 1162 return (ARCHIVE_FATAL); 1163 } 1164 } 1165 } 1166 1167 /* Store extended attributes */ 1168 if (archive_write_pax_header_xattrs(a, pax, entry_original) 1169 == ARCHIVE_FATAL) { 1170 archive_entry_free(entry_main); 1171 archive_string_free(&entry_name); 1172 return (ARCHIVE_FATAL); 1173 } 1174 } 1175 1176 /* Only regular files have data. */ 1177 if (archive_entry_filetype(entry_main) != AE_IFREG) 1178 archive_entry_set_size(entry_main, 0); 1179 1180 /* 1181 * Pax-restricted does not store data for hardlinks, in order 1182 * to improve compatibility with ustar. 1183 */ 1184 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE && 1185 hardlink != NULL) 1186 archive_entry_set_size(entry_main, 0); 1187 1188 /* 1189 * XXX Full pax interchange format does permit a hardlink 1190 * entry to have data associated with it. I'm not supporting 1191 * that here because the client expects me to tell them whether 1192 * or not this format expects data for hardlinks. If I 1193 * don't check here, then every pax archive will end up with 1194 * duplicated data for hardlinks. Someday, there may be 1195 * need to select this behavior, in which case the following 1196 * will need to be revisited. XXX 1197 */ 1198 if (hardlink != NULL) 1199 archive_entry_set_size(entry_main, 0); 1200 1201 /* Save a real file size. */ 1202 real_size = archive_entry_size(entry_main); 1203 /* 1204 * Overwrite a file size by the total size of sparse blocks and 1205 * the size of sparse map info. That file size is the length of 1206 * the data, which we will exactly store into an archive file. 1207 */ 1208 if (archive_strlen(&(pax->sparse_map))) { 1209 size_t mapsize = archive_strlen(&(pax->sparse_map)); 1210 pax->sparse_map_padding = 0x1ff & (-(ssize_t)mapsize); 1211 archive_entry_set_size(entry_main, 1212 mapsize + pax->sparse_map_padding + sparse_total); 1213 } 1214 1215 /* Format 'ustar' header for main entry. 1216 * 1217 * The trouble with file size: If the reader can't understand 1218 * the file size, they may not be able to locate the next 1219 * entry and the rest of the archive is toast. Pax-compliant 1220 * readers are supposed to ignore the file size in the main 1221 * header, so the question becomes how to maximize portability 1222 * for readers that don't support pax attribute extensions. 1223 * For maximum compatibility, I permit numeric extensions in 1224 * the main header so that the file size stored will always be 1225 * correct, even if it's in a format that only some 1226 * implementations understand. The technique used here is: 1227 * 1228 * a) If possible, follow the standard exactly. This handles 1229 * files up to 8 gigabytes minus 1. 1230 * 1231 * b) If that fails, try octal but omit the field terminator. 1232 * That handles files up to 64 gigabytes minus 1. 1233 * 1234 * c) Otherwise, use base-256 extensions. That handles files 1235 * up to 2^63 in this implementation, with the potential to 1236 * go up to 2^94. That should hold us for a while. ;-) 1237 * 1238 * The non-strict formatter uses similar logic for other 1239 * numeric fields, though they're less critical. 1240 */ 1241 if (__archive_write_format_header_ustar(a, ustarbuff, entry_main, -1, 0, 1242 NULL) == ARCHIVE_FATAL) 1243 return (ARCHIVE_FATAL); 1244 1245 /* If we built any extended attributes, write that entry first. */ 1246 if (archive_strlen(&(pax->pax_header)) > 0) { 1247 struct archive_entry *pax_attr_entry; 1248 time_t s; 1249 int64_t uid, gid; 1250 int mode; 1251 1252 pax_attr_entry = archive_entry_new2(&a->archive); 1253 p = entry_name.s; 1254 archive_entry_set_pathname(pax_attr_entry, 1255 build_pax_attribute_name(pax_entry_name, p)); 1256 archive_entry_set_size(pax_attr_entry, 1257 archive_strlen(&(pax->pax_header))); 1258 /* Copy uid/gid (but clip to ustar limits). */ 1259 uid = archive_entry_uid(entry_main); 1260 if (uid >= 1 << 18) 1261 uid = (1 << 18) - 1; 1262 archive_entry_set_uid(pax_attr_entry, uid); 1263 gid = archive_entry_gid(entry_main); 1264 if (gid >= 1 << 18) 1265 gid = (1 << 18) - 1; 1266 archive_entry_set_gid(pax_attr_entry, gid); 1267 /* Copy mode over (but not setuid/setgid bits) */ 1268 mode = archive_entry_mode(entry_main); 1269 #ifdef S_ISUID 1270 mode &= ~S_ISUID; 1271 #endif 1272 #ifdef S_ISGID 1273 mode &= ~S_ISGID; 1274 #endif 1275 #ifdef S_ISVTX 1276 mode &= ~S_ISVTX; 1277 #endif 1278 archive_entry_set_mode(pax_attr_entry, mode); 1279 1280 /* Copy uname/gname. */ 1281 archive_entry_set_uname(pax_attr_entry, 1282 archive_entry_uname(entry_main)); 1283 archive_entry_set_gname(pax_attr_entry, 1284 archive_entry_gname(entry_main)); 1285 1286 /* Copy mtime, but clip to ustar limits. */ 1287 s = archive_entry_mtime(entry_main); 1288 if (s < 0) { s = 0; } 1289 if (s >= 0x7fffffff) { s = 0x7fffffff; } 1290 archive_entry_set_mtime(pax_attr_entry, s, 0); 1291 1292 /* Standard ustar doesn't support atime. */ 1293 archive_entry_set_atime(pax_attr_entry, 0, 0); 1294 1295 /* Standard ustar doesn't support ctime. */ 1296 archive_entry_set_ctime(pax_attr_entry, 0, 0); 1297 1298 r = __archive_write_format_header_ustar(a, paxbuff, 1299 pax_attr_entry, 'x', 1, NULL); 1300 1301 archive_entry_free(pax_attr_entry); 1302 1303 /* Note that the 'x' header shouldn't ever fail to format */ 1304 if (r < ARCHIVE_WARN) { 1305 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1306 "archive_write_pax_header: " 1307 "'x' header failed?! This can't happen.\n"); 1308 return (ARCHIVE_FATAL); 1309 } else if (r < ret) 1310 ret = r; 1311 r = __archive_write_output(a, paxbuff, 512); 1312 if (r != ARCHIVE_OK) { 1313 sparse_list_clear(pax); 1314 pax->entry_bytes_remaining = 0; 1315 pax->entry_padding = 0; 1316 return (ARCHIVE_FATAL); 1317 } 1318 1319 pax->entry_bytes_remaining = archive_strlen(&(pax->pax_header)); 1320 pax->entry_padding = 1321 0x1ff & (-(int64_t)pax->entry_bytes_remaining); 1322 1323 r = __archive_write_output(a, pax->pax_header.s, 1324 archive_strlen(&(pax->pax_header))); 1325 if (r != ARCHIVE_OK) { 1326 /* If a write fails, we're pretty much toast. */ 1327 return (ARCHIVE_FATAL); 1328 } 1329 /* Pad out the end of the entry. */ 1330 r = __archive_write_nulls(a, (size_t)pax->entry_padding); 1331 if (r != ARCHIVE_OK) { 1332 /* If a write fails, we're pretty much toast. */ 1333 return (ARCHIVE_FATAL); 1334 } 1335 pax->entry_bytes_remaining = pax->entry_padding = 0; 1336 } 1337 1338 /* Write the header for main entry. */ 1339 r = __archive_write_output(a, ustarbuff, 512); 1340 if (r != ARCHIVE_OK) 1341 return (r); 1342 1343 /* 1344 * Inform the client of the on-disk size we're using, so 1345 * they can avoid unnecessarily writing a body for something 1346 * that we're just going to ignore. 1347 */ 1348 archive_entry_set_size(entry_original, real_size); 1349 if (pax->sparse_list == NULL && real_size > 0) { 1350 /* This is not a sparse file but we handle its data as 1351 * a sparse block. */ 1352 sparse_list_add(pax, 0, real_size); 1353 sparse_total = real_size; 1354 } 1355 pax->entry_padding = 0x1ff & (-(int64_t)sparse_total); 1356 archive_entry_free(entry_main); 1357 archive_string_free(&entry_name); 1358 1359 return (ret); 1360 } 1361 1362 /* 1363 * We need a valid name for the regular 'ustar' entry. This routine 1364 * tries to hack something more-or-less reasonable. 1365 * 1366 * The approach here tries to preserve leading dir names. We do so by 1367 * working with four sections: 1368 * 1) "prefix" directory names, 1369 * 2) "suffix" directory names, 1370 * 3) inserted dir name (optional), 1371 * 4) filename. 1372 * 1373 * These sections must satisfy the following requirements: 1374 * * Parts 1 & 2 together form an initial portion of the dir name. 1375 * * Part 3 is specified by the caller. (It should not contain a leading 1376 * or trailing '/'.) 1377 * * Part 4 forms an initial portion of the base filename. 1378 * * The filename must be <= 99 chars to fit the ustar 'name' field. 1379 * * Parts 2, 3, 4 together must be <= 99 chars to fit the ustar 'name' fld. 1380 * * Part 1 must be <= 155 chars to fit the ustar 'prefix' field. 1381 * * If the original name ends in a '/', the new name must also end in a '/' 1382 * * Trailing '/.' sequences may be stripped. 1383 * 1384 * Note: Recall that the ustar format does not store the '/' separating 1385 * parts 1 & 2, but does store the '/' separating parts 2 & 3. 1386 */ 1387 static char * 1388 build_ustar_entry_name(char *dest, const char *src, size_t src_length, 1389 const char *insert) 1390 { 1391 const char *prefix, *prefix_end; 1392 const char *suffix, *suffix_end; 1393 const char *filename, *filename_end; 1394 char *p; 1395 int need_slash = 0; /* Was there a trailing slash? */ 1396 size_t suffix_length = 99; 1397 size_t insert_length; 1398 1399 /* Length of additional dir element to be added. */ 1400 if (insert == NULL) 1401 insert_length = 0; 1402 else 1403 /* +2 here allows for '/' before and after the insert. */ 1404 insert_length = strlen(insert) + 2; 1405 1406 /* Step 0: Quick bailout in a common case. */ 1407 if (src_length < 100 && insert == NULL) { 1408 strncpy(dest, src, src_length); 1409 dest[src_length] = '\0'; 1410 return (dest); 1411 } 1412 1413 /* Step 1: Locate filename and enforce the length restriction. */ 1414 filename_end = src + src_length; 1415 /* Remove trailing '/' chars and '/.' pairs. */ 1416 for (;;) { 1417 if (filename_end > src && filename_end[-1] == '/') { 1418 filename_end --; 1419 need_slash = 1; /* Remember to restore trailing '/'. */ 1420 continue; 1421 } 1422 if (filename_end > src + 1 && filename_end[-1] == '.' 1423 && filename_end[-2] == '/') { 1424 filename_end -= 2; 1425 need_slash = 1; /* "foo/." will become "foo/" */ 1426 continue; 1427 } 1428 break; 1429 } 1430 if (need_slash) 1431 suffix_length--; 1432 /* Find start of filename. */ 1433 filename = filename_end - 1; 1434 while ((filename > src) && (*filename != '/')) 1435 filename --; 1436 if ((*filename == '/') && (filename < filename_end - 1)) 1437 filename ++; 1438 /* Adjust filename_end so that filename + insert fits in 99 chars. */ 1439 suffix_length -= insert_length; 1440 if (filename_end > filename + suffix_length) 1441 filename_end = filename + suffix_length; 1442 /* Calculate max size for "suffix" section (#3 above). */ 1443 suffix_length -= filename_end - filename; 1444 1445 /* Step 2: Locate the "prefix" section of the dirname, including 1446 * trailing '/'. */ 1447 prefix = src; 1448 prefix_end = prefix + 155; 1449 if (prefix_end > filename) 1450 prefix_end = filename; 1451 while (prefix_end > prefix && *prefix_end != '/') 1452 prefix_end--; 1453 if ((prefix_end < filename) && (*prefix_end == '/')) 1454 prefix_end++; 1455 1456 /* Step 3: Locate the "suffix" section of the dirname, 1457 * including trailing '/'. */ 1458 suffix = prefix_end; 1459 suffix_end = suffix + suffix_length; /* Enforce limit. */ 1460 if (suffix_end > filename) 1461 suffix_end = filename; 1462 if (suffix_end < suffix) 1463 suffix_end = suffix; 1464 while (suffix_end > suffix && *suffix_end != '/') 1465 suffix_end--; 1466 if ((suffix_end < filename) && (*suffix_end == '/')) 1467 suffix_end++; 1468 1469 /* Step 4: Build the new name. */ 1470 /* The OpenBSD strlcpy function is safer, but less portable. */ 1471 /* Rather than maintain two versions, just use the strncpy version. */ 1472 p = dest; 1473 if (prefix_end > prefix) { 1474 strncpy(p, prefix, prefix_end - prefix); 1475 p += prefix_end - prefix; 1476 } 1477 if (suffix_end > suffix) { 1478 strncpy(p, suffix, suffix_end - suffix); 1479 p += suffix_end - suffix; 1480 } 1481 if (insert != NULL) { 1482 /* Note: assume insert does not have leading or trailing '/' */ 1483 strcpy(p, insert); 1484 p += strlen(insert); 1485 *p++ = '/'; 1486 } 1487 strncpy(p, filename, filename_end - filename); 1488 p += filename_end - filename; 1489 if (need_slash) 1490 *p++ = '/'; 1491 *p = '\0'; 1492 1493 return (dest); 1494 } 1495 1496 /* 1497 * The ustar header for the pax extended attributes must have a 1498 * reasonable name: SUSv3 requires 'dirname'/PaxHeader.'pid'/'filename' 1499 * where 'pid' is the PID of the archiving process. Unfortunately, 1500 * that makes testing a pain since the output varies for each run, 1501 * so I'm sticking with the simpler 'dirname'/PaxHeader/'filename' 1502 * for now. (Someday, I'll make this settable. Then I can use the 1503 * SUS recommendation as default and test harnesses can override it 1504 * to get predictable results.) 1505 * 1506 * Joerg Schilling has argued that this is unnecessary because, in 1507 * practice, if the pax extended attributes get extracted as regular 1508 * files, no one is going to bother reading those attributes to 1509 * manually restore them. Based on this, 'star' uses 1510 * /tmp/PaxHeader/'basename' as the ustar header name. This is a 1511 * tempting argument, in part because it's simpler than the SUSv3 1512 * recommendation, but I'm not entirely convinced. I'm also 1513 * uncomfortable with the fact that "/tmp" is a Unix-ism. 1514 * 1515 * The following routine leverages build_ustar_entry_name() above and 1516 * so is simpler than you might think. It just needs to provide the 1517 * additional path element and handle a few pathological cases). 1518 */ 1519 static char * 1520 build_pax_attribute_name(char *dest, const char *src) 1521 { 1522 char buff[64]; 1523 const char *p; 1524 1525 /* Handle the null filename case. */ 1526 if (src == NULL || *src == '\0') { 1527 strcpy(dest, "PaxHeader/blank"); 1528 return (dest); 1529 } 1530 1531 /* Prune final '/' and other unwanted final elements. */ 1532 p = src + strlen(src); 1533 for (;;) { 1534 /* Ends in "/", remove the '/' */ 1535 if (p > src && p[-1] == '/') { 1536 --p; 1537 continue; 1538 } 1539 /* Ends in "/.", remove the '.' */ 1540 if (p > src + 1 && p[-1] == '.' 1541 && p[-2] == '/') { 1542 --p; 1543 continue; 1544 } 1545 break; 1546 } 1547 1548 /* Pathological case: After above, there was nothing left. 1549 * This includes "/." "/./." "/.//./." etc. */ 1550 if (p == src) { 1551 strcpy(dest, "/PaxHeader/rootdir"); 1552 return (dest); 1553 } 1554 1555 /* Convert unadorned "." into a suitable filename. */ 1556 if (*src == '.' && p == src + 1) { 1557 strcpy(dest, "PaxHeader/currentdir"); 1558 return (dest); 1559 } 1560 1561 /* 1562 * TODO: Push this string into the 'pax' structure to avoid 1563 * recomputing it every time. That will also open the door 1564 * to having clients override it. 1565 */ 1566 #if HAVE_GETPID && 0 /* Disable this for now; see above comment. */ 1567 sprintf(buff, "PaxHeader.%d", getpid()); 1568 #else 1569 /* If the platform can't fetch the pid, don't include it. */ 1570 strcpy(buff, "PaxHeader"); 1571 #endif 1572 /* General case: build a ustar-compatible name adding 1573 * "/PaxHeader/". */ 1574 build_ustar_entry_name(dest, src, p - src, buff); 1575 1576 return (dest); 1577 } 1578 1579 /* 1580 * GNU PAX Format 1.0 requires the special name, which pattern is: 1581 * <dir>/GNUSparseFile.<pid>/<original file name> 1582 * 1583 * This function is used for only Sparse file, a file type of which 1584 * is regular file. 1585 */ 1586 static char * 1587 build_gnu_sparse_name(char *dest, const char *src) 1588 { 1589 char buff[64]; 1590 const char *p; 1591 1592 /* Handle the null filename case. */ 1593 if (src == NULL || *src == '\0') { 1594 strcpy(dest, "GNUSparseFile/blank"); 1595 return (dest); 1596 } 1597 1598 /* Prune final '/' and other unwanted final elements. */ 1599 p = src + strlen(src); 1600 for (;;) { 1601 /* Ends in "/", remove the '/' */ 1602 if (p > src && p[-1] == '/') { 1603 --p; 1604 continue; 1605 } 1606 /* Ends in "/.", remove the '.' */ 1607 if (p > src + 1 && p[-1] == '.' 1608 && p[-2] == '/') { 1609 --p; 1610 continue; 1611 } 1612 break; 1613 } 1614 1615 #if HAVE_GETPID && 0 /* Disable this as pax attribute name. */ 1616 sprintf(buff, "GNUSparseFile.%d", getpid()); 1617 #else 1618 /* If the platform can't fetch the pid, don't include it. */ 1619 strcpy(buff, "GNUSparseFile"); 1620 #endif 1621 /* General case: build a ustar-compatible name adding 1622 * "/GNUSparseFile/". */ 1623 build_ustar_entry_name(dest, src, p - src, buff); 1624 1625 return (dest); 1626 } 1627 1628 /* Write two null blocks for the end of archive */ 1629 static int 1630 archive_write_pax_close(struct archive_write *a) 1631 { 1632 return (__archive_write_nulls(a, 512 * 2)); 1633 } 1634 1635 static int 1636 archive_write_pax_free(struct archive_write *a) 1637 { 1638 struct pax *pax; 1639 1640 pax = (struct pax *)a->format_data; 1641 if (pax == NULL) 1642 return (ARCHIVE_OK); 1643 1644 archive_string_free(&pax->pax_header); 1645 archive_string_free(&pax->sparse_map); 1646 archive_string_free(&pax->l_url_encoded_name); 1647 sparse_list_clear(pax); 1648 free(pax); 1649 a->format_data = NULL; 1650 return (ARCHIVE_OK); 1651 } 1652 1653 static int 1654 archive_write_pax_finish_entry(struct archive_write *a) 1655 { 1656 struct pax *pax; 1657 uint64_t remaining; 1658 int ret; 1659 1660 pax = (struct pax *)a->format_data; 1661 remaining = pax->entry_bytes_remaining; 1662 if (remaining == 0) { 1663 while (pax->sparse_list) { 1664 struct sparse_block *sb; 1665 if (!pax->sparse_list->is_hole) 1666 remaining += pax->sparse_list->remaining; 1667 sb = pax->sparse_list->next; 1668 free(pax->sparse_list); 1669 pax->sparse_list = sb; 1670 } 1671 } 1672 ret = __archive_write_nulls(a, (size_t)(remaining + pax->entry_padding)); 1673 pax->entry_bytes_remaining = pax->entry_padding = 0; 1674 return (ret); 1675 } 1676 1677 static ssize_t 1678 archive_write_pax_data(struct archive_write *a, const void *buff, size_t s) 1679 { 1680 struct pax *pax; 1681 size_t ws; 1682 size_t total; 1683 int ret; 1684 1685 pax = (struct pax *)a->format_data; 1686 1687 /* 1688 * According to GNU PAX format 1.0, write a sparse map 1689 * before the body. 1690 */ 1691 if (archive_strlen(&(pax->sparse_map))) { 1692 ret = __archive_write_output(a, pax->sparse_map.s, 1693 archive_strlen(&(pax->sparse_map))); 1694 if (ret != ARCHIVE_OK) 1695 return (ret); 1696 ret = __archive_write_nulls(a, pax->sparse_map_padding); 1697 if (ret != ARCHIVE_OK) 1698 return (ret); 1699 archive_string_empty(&(pax->sparse_map)); 1700 } 1701 1702 total = 0; 1703 while (total < s) { 1704 const unsigned char *p; 1705 1706 while (pax->sparse_list != NULL && 1707 pax->sparse_list->remaining == 0) { 1708 struct sparse_block *sb = pax->sparse_list->next; 1709 free(pax->sparse_list); 1710 pax->sparse_list = sb; 1711 } 1712 1713 if (pax->sparse_list == NULL) 1714 return (total); 1715 1716 p = ((const unsigned char *)buff) + total; 1717 ws = s - total; 1718 if (ws > pax->sparse_list->remaining) 1719 ws = (size_t)pax->sparse_list->remaining; 1720 1721 if (pax->sparse_list->is_hole) { 1722 /* Current block is hole thus we do not write 1723 * the body. */ 1724 pax->sparse_list->remaining -= ws; 1725 total += ws; 1726 continue; 1727 } 1728 1729 ret = __archive_write_output(a, p, ws); 1730 pax->sparse_list->remaining -= ws; 1731 total += ws; 1732 if (ret != ARCHIVE_OK) 1733 return (ret); 1734 } 1735 return (total); 1736 } 1737 1738 static int 1739 has_non_ASCII(const char *_p) 1740 { 1741 const unsigned char *p = (const unsigned char *)_p; 1742 1743 if (p == NULL) 1744 return (1); 1745 while (*p != '\0' && *p < 128) 1746 p++; 1747 return (*p != '\0'); 1748 } 1749 1750 /* 1751 * Used by extended attribute support; encodes the name 1752 * so that there will be no '=' characters in the result. 1753 */ 1754 static char * 1755 url_encode(const char *in) 1756 { 1757 const char *s; 1758 char *d; 1759 int out_len = 0; 1760 char *out; 1761 1762 for (s = in; *s != '\0'; s++) { 1763 if (*s < 33 || *s > 126 || *s == '%' || *s == '=') 1764 out_len += 3; 1765 else 1766 out_len++; 1767 } 1768 1769 out = (char *)malloc(out_len + 1); 1770 if (out == NULL) 1771 return (NULL); 1772 1773 for (s = in, d = out; *s != '\0'; s++) { 1774 /* encode any non-printable ASCII character or '%' or '=' */ 1775 if (*s < 33 || *s > 126 || *s == '%' || *s == '=') { 1776 /* URL encoding is '%' followed by two hex digits */ 1777 *d++ = '%'; 1778 *d++ = "0123456789ABCDEF"[0x0f & (*s >> 4)]; 1779 *d++ = "0123456789ABCDEF"[0x0f & *s]; 1780 } else { 1781 *d++ = *s; 1782 } 1783 } 1784 *d = '\0'; 1785 return (out); 1786 } 1787 1788 /* 1789 * Encode a sequence of bytes into a C string using base-64 encoding. 1790 * 1791 * Returns a null-terminated C string allocated with malloc(); caller 1792 * is responsible for freeing the result. 1793 */ 1794 static char * 1795 base64_encode(const char *s, size_t len) 1796 { 1797 static const char digits[64] = 1798 { 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O', 1799 'P','Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d', 1800 'e','f','g','h','i','j','k','l','m','n','o','p','q','r','s', 1801 't','u','v','w','x','y','z','0','1','2','3','4','5','6','7', 1802 '8','9','+','/' }; 1803 int v; 1804 char *d, *out; 1805 1806 /* 3 bytes becomes 4 chars, but round up and allow for trailing NUL */ 1807 out = (char *)malloc((len * 4 + 2) / 3 + 1); 1808 if (out == NULL) 1809 return (NULL); 1810 d = out; 1811 1812 /* Convert each group of 3 bytes into 4 characters. */ 1813 while (len >= 3) { 1814 v = (((int)s[0] << 16) & 0xff0000) 1815 | (((int)s[1] << 8) & 0xff00) 1816 | (((int)s[2]) & 0x00ff); 1817 s += 3; 1818 len -= 3; 1819 *d++ = digits[(v >> 18) & 0x3f]; 1820 *d++ = digits[(v >> 12) & 0x3f]; 1821 *d++ = digits[(v >> 6) & 0x3f]; 1822 *d++ = digits[(v) & 0x3f]; 1823 } 1824 /* Handle final group of 1 byte (2 chars) or 2 bytes (3 chars). */ 1825 switch (len) { 1826 case 0: break; 1827 case 1: 1828 v = (((int)s[0] << 16) & 0xff0000); 1829 *d++ = digits[(v >> 18) & 0x3f]; 1830 *d++ = digits[(v >> 12) & 0x3f]; 1831 break; 1832 case 2: 1833 v = (((int)s[0] << 16) & 0xff0000) 1834 | (((int)s[1] << 8) & 0xff00); 1835 *d++ = digits[(v >> 18) & 0x3f]; 1836 *d++ = digits[(v >> 12) & 0x3f]; 1837 *d++ = digits[(v >> 6) & 0x3f]; 1838 break; 1839 } 1840 /* Add trailing NUL character so output is a valid C string. */ 1841 *d = '\0'; 1842 return (out); 1843 } 1844 1845 static void 1846 sparse_list_clear(struct pax *pax) 1847 { 1848 while (pax->sparse_list != NULL) { 1849 struct sparse_block *sb = pax->sparse_list; 1850 pax->sparse_list = sb->next; 1851 free(sb); 1852 } 1853 pax->sparse_tail = NULL; 1854 } 1855 1856 static int 1857 _sparse_list_add_block(struct pax *pax, int64_t offset, int64_t length, 1858 int is_hole) 1859 { 1860 struct sparse_block *sb; 1861 1862 sb = (struct sparse_block *)malloc(sizeof(*sb)); 1863 if (sb == NULL) 1864 return (ARCHIVE_FATAL); 1865 sb->next = NULL; 1866 sb->is_hole = is_hole; 1867 sb->offset = offset; 1868 sb->remaining = length; 1869 if (pax->sparse_list == NULL || pax->sparse_tail == NULL) 1870 pax->sparse_list = pax->sparse_tail = sb; 1871 else { 1872 pax->sparse_tail->next = sb; 1873 pax->sparse_tail = sb; 1874 } 1875 return (ARCHIVE_OK); 1876 } 1877 1878 static int 1879 sparse_list_add(struct pax *pax, int64_t offset, int64_t length) 1880 { 1881 int64_t last_offset; 1882 int r; 1883 1884 if (pax->sparse_tail == NULL) 1885 last_offset = 0; 1886 else { 1887 last_offset = pax->sparse_tail->offset + 1888 pax->sparse_tail->remaining; 1889 } 1890 if (last_offset < offset) { 1891 /* Add a hole block. */ 1892 r = _sparse_list_add_block(pax, last_offset, 1893 offset - last_offset, 1); 1894 if (r != ARCHIVE_OK) 1895 return (r); 1896 } 1897 /* Add data block. */ 1898 return (_sparse_list_add_block(pax, offset, length, 0)); 1899 } 1900 1901