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) { 651 /* XXX error message */ 652 return (ARCHIVE_FAILED); 653 } 654 strcpy(name, oname); 655 /* Find last '/'; strip trailing '/' characters */ 656 bname = strrchr(name, '/'); 657 while (bname != NULL && bname[1] == '\0') { 658 *bname = '\0'; 659 bname = strrchr(name, '/'); 660 } 661 if (bname == NULL) { 662 memmove(name + 2, name, name_length + 1); 663 memmove(name, "._", 2); 664 } else { 665 bname += 1; 666 memmove(bname + 2, bname, strlen(bname) + 1); 667 memmove(bname, "._", 2); 668 } 669 archive_entry_copy_pathname(extra, name); 670 free(name); 671 672 archive_entry_set_size(extra, mac_metadata_size); 673 archive_entry_set_filetype(extra, AE_IFREG); 674 archive_entry_set_perm(extra, 675 archive_entry_perm(entry_original)); 676 archive_entry_set_mtime(extra, 677 archive_entry_mtime(entry_original), 678 archive_entry_mtime_nsec(entry_original)); 679 archive_entry_set_gid(extra, 680 archive_entry_gid(entry_original)); 681 archive_entry_set_gname(extra, 682 archive_entry_gname(entry_original)); 683 archive_entry_set_uid(extra, 684 archive_entry_uid(entry_original)); 685 archive_entry_set_uname(extra, 686 archive_entry_uname(entry_original)); 687 688 /* Recurse to write the special copyfile entry. */ 689 r = archive_write_pax_header(a, extra); 690 if (r < ARCHIVE_WARN) 691 return (r); 692 if (r < ret) 693 ret = r; 694 r = archive_write_pax_data(a, mac_metadata, mac_metadata_size); 695 if (r < ARCHIVE_WARN) 696 return (r); 697 if (r < ret) 698 ret = r; 699 r = archive_write_pax_finish_entry(a); 700 if (r < ARCHIVE_WARN) 701 return (r); 702 if (r < ret) 703 ret = r; 704 } 705 706 /* Copy entry so we can modify it as needed. */ 707 #if defined(_WIN32) && !defined(__CYGWIN__) 708 /* Make sure the path separators in pahtname, hardlink and symlink 709 * are all slash '/', not the Windows path separator '\'. */ 710 entry_main = __la_win_entry_in_posix_pathseparator(entry_original); 711 if (entry_main == entry_original) 712 entry_main = archive_entry_clone(entry_original); 713 #else 714 entry_main = archive_entry_clone(entry_original); 715 #endif 716 if (entry_main == NULL) { 717 archive_set_error(&a->archive, ENOMEM, 718 "Can't allocate pax data"); 719 return(ARCHIVE_FATAL); 720 } 721 archive_string_empty(&(pax->pax_header)); /* Blank our work area. */ 722 archive_string_empty(&(pax->sparse_map)); 723 sparse_total = 0; 724 sparse_list_clear(pax); 725 726 if (hardlink == NULL && 727 archive_entry_filetype(entry_main) == AE_IFREG) 728 sparse_count = archive_entry_sparse_reset(entry_main); 729 else 730 sparse_count = 0; 731 if (sparse_count) { 732 int64_t offset, length, last_offset = 0; 733 /* Get the last entry of sparse block. */ 734 while (archive_entry_sparse_next( 735 entry_main, &offset, &length) == ARCHIVE_OK) 736 last_offset = offset + length; 737 738 /* If the last sparse block does not reach the end of file, 739 * We have to add a empty sparse block as the last entry to 740 * manage storing file data. */ 741 if (last_offset < archive_entry_size(entry_main)) 742 archive_entry_sparse_add_entry(entry_main, 743 archive_entry_size(entry_main), 0); 744 sparse_count = archive_entry_sparse_reset(entry_main); 745 } 746 747 /* 748 * First, check the name fields and see if any of them 749 * require binary coding. If any of them does, then all of 750 * them do. 751 */ 752 r = get_entry_pathname(a, entry_main, &path, &path_length, sconv); 753 if (r == ARCHIVE_FATAL) 754 return (r); 755 else if (r != ARCHIVE_OK) { 756 r = get_entry_pathname(a, entry_main, &path, 757 &path_length, NULL); 758 if (r == ARCHIVE_FATAL) 759 return (r); 760 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 761 "Can't translate pathname '%s' to %s", path, 762 archive_string_conversion_charset_name(sconv)); 763 ret = ARCHIVE_WARN; 764 sconv = NULL;/* The header charset switches to binary mode. */ 765 } 766 r = get_entry_uname(a, entry_main, &uname, &uname_length, sconv); 767 if (r == ARCHIVE_FATAL) 768 return (r); 769 else if (r != ARCHIVE_OK) { 770 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL); 771 if (r == ARCHIVE_FATAL) 772 return (r); 773 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 774 "Can't translate uname '%s' to %s", uname, 775 archive_string_conversion_charset_name(sconv)); 776 ret = ARCHIVE_WARN; 777 sconv = NULL;/* The header charset switches to binary mode. */ 778 } 779 r = get_entry_gname(a, entry_main, &gname, &gname_length, sconv); 780 if (r == ARCHIVE_FATAL) 781 return (r); 782 else if (r != ARCHIVE_OK) { 783 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL); 784 if (r == ARCHIVE_FATAL) 785 return (r); 786 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 787 "Can't translate gname '%s' to %s", gname, 788 archive_string_conversion_charset_name(sconv)); 789 ret = ARCHIVE_WARN; 790 sconv = NULL;/* The header charset switches to binary mode. */ 791 } 792 linkpath = hardlink; 793 linkpath_length = hardlink_length; 794 if (linkpath == NULL) { 795 r = get_entry_symlink(a, entry_main, &linkpath, 796 &linkpath_length, sconv); 797 if (r == ARCHIVE_FATAL) 798 return (r); 799 else if (r != ARCHIVE_OK) { 800 r = get_entry_symlink(a, entry_main, &linkpath, 801 &linkpath_length, NULL); 802 if (r == ARCHIVE_FATAL) 803 return (r); 804 archive_set_error(&a->archive, 805 ARCHIVE_ERRNO_FILE_FORMAT, 806 "Can't translate linkname '%s' to %s", linkpath, 807 archive_string_conversion_charset_name(sconv)); 808 ret = ARCHIVE_WARN; 809 sconv = NULL; 810 } 811 } 812 813 /* If any string conversions failed, get all attributes 814 * in binary-mode. */ 815 if (sconv == NULL && !pax->opt_binary) { 816 if (hardlink != NULL) { 817 r = get_entry_hardlink(a, entry_main, &hardlink, 818 &hardlink_length, NULL); 819 if (r == ARCHIVE_FATAL) 820 return (r); 821 linkpath = hardlink; 822 linkpath_length = hardlink_length; 823 } 824 r = get_entry_pathname(a, entry_main, &path, 825 &path_length, NULL); 826 if (r == ARCHIVE_FATAL) 827 return (r); 828 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL); 829 if (r == ARCHIVE_FATAL) 830 return (r); 831 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL); 832 if (r == ARCHIVE_FATAL) 833 return (r); 834 } 835 836 /* Store the header encoding first, to be nice to readers. */ 837 if (sconv == NULL) 838 add_pax_attr(&(pax->pax_header), "hdrcharset", "BINARY"); 839 840 841 /* 842 * If name is too long, or has non-ASCII characters, add 843 * 'path' to pax extended attrs. (Note that an unconvertible 844 * name must have non-ASCII characters.) 845 */ 846 if (has_non_ASCII(path)) { 847 /* We have non-ASCII characters. */ 848 add_pax_attr(&(pax->pax_header), "path", path); 849 archive_entry_set_pathname(entry_main, 850 build_ustar_entry_name(ustar_entry_name, 851 path, path_length, NULL)); 852 need_extension = 1; 853 } else { 854 /* We have an all-ASCII path; we'd like to just store 855 * it in the ustar header if it will fit. Yes, this 856 * duplicates some of the logic in 857 * archive_write_set_format_ustar.c 858 */ 859 if (path_length <= 100) { 860 /* Fits in the old 100-char tar name field. */ 861 } else { 862 /* Find largest suffix that will fit. */ 863 /* Note: strlen() > 100, so strlen() - 100 - 1 >= 0 */ 864 suffix = strchr(path + path_length - 100 - 1, '/'); 865 /* Don't attempt an empty prefix. */ 866 if (suffix == path) 867 suffix = strchr(suffix + 1, '/'); 868 /* We can put it in the ustar header if it's 869 * all ASCII and it's either <= 100 characters 870 * or can be split at a '/' into a prefix <= 871 * 155 chars and a suffix <= 100 chars. (Note 872 * the strchr() above will return NULL exactly 873 * when the path can't be split.) 874 */ 875 if (suffix == NULL /* Suffix > 100 chars. */ 876 || suffix[1] == '\0' /* empty suffix */ 877 || suffix - path > 155) /* Prefix > 155 chars */ 878 { 879 add_pax_attr(&(pax->pax_header), "path", path); 880 archive_entry_set_pathname(entry_main, 881 build_ustar_entry_name(ustar_entry_name, 882 path, path_length, NULL)); 883 need_extension = 1; 884 } 885 } 886 } 887 888 if (linkpath != NULL) { 889 /* If link name is too long or has non-ASCII characters, add 890 * 'linkpath' to pax extended attrs. */ 891 if (linkpath_length > 100 || has_non_ASCII(linkpath)) { 892 add_pax_attr(&(pax->pax_header), "linkpath", linkpath); 893 if (linkpath_length > 100) { 894 if (hardlink != NULL) 895 archive_entry_set_hardlink(entry_main, 896 "././@LongHardLink"); 897 else 898 archive_entry_set_symlink(entry_main, 899 "././@LongSymLink"); 900 } 901 need_extension = 1; 902 } 903 } 904 /* Save a pathname since it will be renamed if `entry_main` has 905 * sparse blocks. */ 906 archive_string_init(&entry_name); 907 archive_strcpy(&entry_name, archive_entry_pathname(entry_main)); 908 909 /* If file size is too large, add 'size' to pax extended attrs. */ 910 if (archive_entry_size(entry_main) >= (((int64_t)1) << 33)) { 911 add_pax_attr_int(&(pax->pax_header), "size", 912 archive_entry_size(entry_main)); 913 need_extension = 1; 914 } 915 916 /* If numeric GID is too large, add 'gid' to pax extended attrs. */ 917 if ((unsigned int)archive_entry_gid(entry_main) >= (1 << 18)) { 918 add_pax_attr_int(&(pax->pax_header), "gid", 919 archive_entry_gid(entry_main)); 920 need_extension = 1; 921 } 922 923 /* If group name is too large or has non-ASCII characters, add 924 * 'gname' to pax extended attrs. */ 925 if (gname != NULL) { 926 if (gname_length > 31 || has_non_ASCII(gname)) { 927 add_pax_attr(&(pax->pax_header), "gname", gname); 928 need_extension = 1; 929 } 930 } 931 932 /* If numeric UID is too large, add 'uid' to pax extended attrs. */ 933 if ((unsigned int)archive_entry_uid(entry_main) >= (1 << 18)) { 934 add_pax_attr_int(&(pax->pax_header), "uid", 935 archive_entry_uid(entry_main)); 936 need_extension = 1; 937 } 938 939 /* Add 'uname' to pax extended attrs if necessary. */ 940 if (uname != NULL) { 941 if (uname_length > 31 || has_non_ASCII(uname)) { 942 add_pax_attr(&(pax->pax_header), "uname", uname); 943 need_extension = 1; 944 } 945 } 946 947 /* 948 * POSIX/SUSv3 doesn't provide a standard key for large device 949 * numbers. I use the same keys here that Joerg Schilling 950 * used for 'star.' (Which, somewhat confusingly, are called 951 * "devXXX" even though they code "rdev" values.) No doubt, 952 * other implementations use other keys. Note that there's no 953 * reason we can't write the same information into a number of 954 * different keys. 955 * 956 * Of course, this is only needed for block or char device entries. 957 */ 958 if (archive_entry_filetype(entry_main) == AE_IFBLK 959 || archive_entry_filetype(entry_main) == AE_IFCHR) { 960 /* 961 * If rdevmajor is too large, add 'SCHILY.devmajor' to 962 * extended attributes. 963 */ 964 int rdevmajor, rdevminor; 965 rdevmajor = archive_entry_rdevmajor(entry_main); 966 rdevminor = archive_entry_rdevminor(entry_main); 967 if (rdevmajor >= (1 << 18)) { 968 add_pax_attr_int(&(pax->pax_header), "SCHILY.devmajor", 969 rdevmajor); 970 /* 971 * Non-strict formatting below means we don't 972 * have to truncate here. Not truncating improves 973 * the chance that some more modern tar archivers 974 * (such as GNU tar 1.13) can restore the full 975 * value even if they don't understand the pax 976 * extended attributes. See my rant below about 977 * file size fields for additional details. 978 */ 979 /* archive_entry_set_rdevmajor(entry_main, 980 rdevmajor & ((1 << 18) - 1)); */ 981 need_extension = 1; 982 } 983 984 /* 985 * If devminor is too large, add 'SCHILY.devminor' to 986 * extended attributes. 987 */ 988 if (rdevminor >= (1 << 18)) { 989 add_pax_attr_int(&(pax->pax_header), "SCHILY.devminor", 990 rdevminor); 991 /* Truncation is not necessary here, either. */ 992 /* archive_entry_set_rdevminor(entry_main, 993 rdevminor & ((1 << 18) - 1)); */ 994 need_extension = 1; 995 } 996 } 997 998 /* 999 * Technically, the mtime field in the ustar header can 1000 * support 33 bits, but many platforms use signed 32-bit time 1001 * values. The cutoff of 0x7fffffff here is a compromise. 1002 * Yes, this check is duplicated just below; this helps to 1003 * avoid writing an mtime attribute just to handle a 1004 * high-resolution timestamp in "restricted pax" mode. 1005 */ 1006 if (!need_extension && 1007 ((archive_entry_mtime(entry_main) < 0) 1008 || (archive_entry_mtime(entry_main) >= 0x7fffffff))) 1009 need_extension = 1; 1010 1011 /* I use a star-compatible file flag attribute. */ 1012 p = archive_entry_fflags_text(entry_main); 1013 if (!need_extension && p != NULL && *p != '\0') 1014 need_extension = 1; 1015 1016 /* If there are non-trivial ACL entries, we need an extension. */ 1017 if (!need_extension && archive_entry_acl_count(entry_original, 1018 ARCHIVE_ENTRY_ACL_TYPE_ACCESS) > 0) 1019 need_extension = 1; 1020 1021 /* If there are non-trivial ACL entries, we need an extension. */ 1022 if (!need_extension && archive_entry_acl_count(entry_original, 1023 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT) > 0) 1024 need_extension = 1; 1025 1026 /* If there are extended attributes, we need an extension */ 1027 if (!need_extension && archive_entry_xattr_count(entry_original) > 0) 1028 need_extension = 1; 1029 1030 /* If there are sparse info, we need an extension */ 1031 if (!need_extension && sparse_count > 0) 1032 need_extension = 1; 1033 1034 /* 1035 * The following items are handled differently in "pax 1036 * restricted" format. In particular, in "pax restricted" 1037 * format they won't be added unless need_extension is 1038 * already set (we're already generating an extended header, so 1039 * may as well include these). 1040 */ 1041 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_RESTRICTED || 1042 need_extension) { 1043 1044 if (archive_entry_mtime(entry_main) < 0 || 1045 archive_entry_mtime(entry_main) >= 0x7fffffff || 1046 archive_entry_mtime_nsec(entry_main) != 0) 1047 add_pax_attr_time(&(pax->pax_header), "mtime", 1048 archive_entry_mtime(entry_main), 1049 archive_entry_mtime_nsec(entry_main)); 1050 1051 if (archive_entry_ctime(entry_main) != 0 || 1052 archive_entry_ctime_nsec(entry_main) != 0) 1053 add_pax_attr_time(&(pax->pax_header), "ctime", 1054 archive_entry_ctime(entry_main), 1055 archive_entry_ctime_nsec(entry_main)); 1056 1057 if (archive_entry_atime(entry_main) != 0 || 1058 archive_entry_atime_nsec(entry_main) != 0) 1059 add_pax_attr_time(&(pax->pax_header), "atime", 1060 archive_entry_atime(entry_main), 1061 archive_entry_atime_nsec(entry_main)); 1062 1063 /* Store birth/creationtime only if it's earlier than mtime */ 1064 if (archive_entry_birthtime_is_set(entry_main) && 1065 archive_entry_birthtime(entry_main) 1066 < archive_entry_mtime(entry_main)) 1067 add_pax_attr_time(&(pax->pax_header), 1068 "LIBARCHIVE.creationtime", 1069 archive_entry_birthtime(entry_main), 1070 archive_entry_birthtime_nsec(entry_main)); 1071 1072 /* I use a star-compatible file flag attribute. */ 1073 p = archive_entry_fflags_text(entry_main); 1074 if (p != NULL && *p != '\0') 1075 add_pax_attr(&(pax->pax_header), "SCHILY.fflags", p); 1076 1077 /* I use star-compatible ACL attributes. */ 1078 r = archive_entry_acl_text_l(entry_original, 1079 ARCHIVE_ENTRY_ACL_TYPE_ACCESS | 1080 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID, 1081 &p, NULL, pax->sconv_utf8); 1082 if (r != 0) { 1083 if (errno == ENOMEM) { 1084 archive_set_error(&a->archive, ENOMEM, 1085 "Can't allocate memory for " 1086 "ACL.access"); 1087 return (ARCHIVE_FATAL); 1088 } 1089 archive_set_error(&a->archive, 1090 ARCHIVE_ERRNO_FILE_FORMAT, 1091 "Can't translate ACL.access to UTF-8"); 1092 ret = ARCHIVE_WARN; 1093 } else if (p != NULL && *p != '\0') { 1094 add_pax_attr(&(pax->pax_header), 1095 "SCHILY.acl.access", p); 1096 } 1097 r = archive_entry_acl_text_l(entry_original, 1098 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT | 1099 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID, 1100 &p, NULL, pax->sconv_utf8); 1101 if (r != 0) { 1102 if (errno == ENOMEM) { 1103 archive_set_error(&a->archive, ENOMEM, 1104 "Can't allocate memory for " 1105 "ACL.default"); 1106 return (ARCHIVE_FATAL); 1107 } 1108 archive_set_error(&a->archive, 1109 ARCHIVE_ERRNO_FILE_FORMAT, 1110 "Can't translate ACL.default to UTF-8"); 1111 ret = ARCHIVE_WARN; 1112 } else if (p != NULL && *p != '\0') { 1113 add_pax_attr(&(pax->pax_header), 1114 "SCHILY.acl.default", p); 1115 } 1116 1117 /* We use GNU-tar-compatible sparse attributes. */ 1118 if (sparse_count > 0) { 1119 int64_t soffset, slength; 1120 1121 add_pax_attr_int(&(pax->pax_header), 1122 "GNU.sparse.major", 1); 1123 add_pax_attr_int(&(pax->pax_header), 1124 "GNU.sparse.minor", 0); 1125 add_pax_attr(&(pax->pax_header), 1126 "GNU.sparse.name", entry_name.s); 1127 add_pax_attr_int(&(pax->pax_header), 1128 "GNU.sparse.realsize", 1129 archive_entry_size(entry_main)); 1130 1131 /* Rename the file name which will be used for 1132 * ustar header to a special name, which GNU 1133 * PAX Format 1.0 requires */ 1134 archive_entry_set_pathname(entry_main, 1135 build_gnu_sparse_name(gnu_sparse_name, 1136 entry_name.s)); 1137 1138 /* 1139 * - Make a sparse map, which will precede a file data. 1140 * - Get the total size of available data of sparse. 1141 */ 1142 archive_string_sprintf(&(pax->sparse_map), "%d\n", 1143 sparse_count); 1144 while (archive_entry_sparse_next(entry_main, 1145 &soffset, &slength) == ARCHIVE_OK) { 1146 archive_string_sprintf(&(pax->sparse_map), 1147 "%jd\n%jd\n", 1148 (intmax_t)soffset, 1149 (intmax_t)slength); 1150 sparse_total += slength; 1151 if (sparse_list_add(pax, soffset, slength) 1152 != ARCHIVE_OK) { 1153 archive_set_error(&a->archive, 1154 ENOMEM, 1155 "Can't allocate memory"); 1156 archive_entry_free(entry_main); 1157 archive_string_free(&entry_name); 1158 return (ARCHIVE_FATAL); 1159 } 1160 } 1161 } 1162 1163 /* Store extended attributes */ 1164 if (archive_write_pax_header_xattrs(a, pax, entry_original) 1165 == ARCHIVE_FATAL) { 1166 archive_entry_free(entry_main); 1167 archive_string_free(&entry_name); 1168 return (ARCHIVE_FATAL); 1169 } 1170 } 1171 1172 /* Only regular files have data. */ 1173 if (archive_entry_filetype(entry_main) != AE_IFREG) 1174 archive_entry_set_size(entry_main, 0); 1175 1176 /* 1177 * Pax-restricted does not store data for hardlinks, in order 1178 * to improve compatibility with ustar. 1179 */ 1180 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE && 1181 hardlink != NULL) 1182 archive_entry_set_size(entry_main, 0); 1183 1184 /* 1185 * XXX Full pax interchange format does permit a hardlink 1186 * entry to have data associated with it. I'm not supporting 1187 * that here because the client expects me to tell them whether 1188 * or not this format expects data for hardlinks. If I 1189 * don't check here, then every pax archive will end up with 1190 * duplicated data for hardlinks. Someday, there may be 1191 * need to select this behavior, in which case the following 1192 * will need to be revisited. XXX 1193 */ 1194 if (hardlink != NULL) 1195 archive_entry_set_size(entry_main, 0); 1196 1197 /* Save a real file size. */ 1198 real_size = archive_entry_size(entry_main); 1199 /* 1200 * Overwrite a file size by the total size of sparse blocks and 1201 * the size of sparse map info. That file size is the length of 1202 * the data, which we will exactly store into an archive file. 1203 */ 1204 if (archive_strlen(&(pax->sparse_map))) { 1205 size_t mapsize = archive_strlen(&(pax->sparse_map)); 1206 pax->sparse_map_padding = 0x1ff & (-(ssize_t)mapsize); 1207 archive_entry_set_size(entry_main, 1208 mapsize + pax->sparse_map_padding + sparse_total); 1209 } 1210 1211 /* Format 'ustar' header for main entry. 1212 * 1213 * The trouble with file size: If the reader can't understand 1214 * the file size, they may not be able to locate the next 1215 * entry and the rest of the archive is toast. Pax-compliant 1216 * readers are supposed to ignore the file size in the main 1217 * header, so the question becomes how to maximize portability 1218 * for readers that don't support pax attribute extensions. 1219 * For maximum compatibility, I permit numeric extensions in 1220 * the main header so that the file size stored will always be 1221 * correct, even if it's in a format that only some 1222 * implementations understand. The technique used here is: 1223 * 1224 * a) If possible, follow the standard exactly. This handles 1225 * files up to 8 gigabytes minus 1. 1226 * 1227 * b) If that fails, try octal but omit the field terminator. 1228 * That handles files up to 64 gigabytes minus 1. 1229 * 1230 * c) Otherwise, use base-256 extensions. That handles files 1231 * up to 2^63 in this implementation, with the potential to 1232 * go up to 2^94. That should hold us for a while. ;-) 1233 * 1234 * The non-strict formatter uses similar logic for other 1235 * numeric fields, though they're less critical. 1236 */ 1237 if (__archive_write_format_header_ustar(a, ustarbuff, entry_main, -1, 0, 1238 NULL) == ARCHIVE_FATAL) 1239 return (ARCHIVE_FATAL); 1240 1241 /* If we built any extended attributes, write that entry first. */ 1242 if (archive_strlen(&(pax->pax_header)) > 0) { 1243 struct archive_entry *pax_attr_entry; 1244 time_t s; 1245 int64_t uid, gid; 1246 int mode; 1247 1248 pax_attr_entry = archive_entry_new2(&a->archive); 1249 p = entry_name.s; 1250 archive_entry_set_pathname(pax_attr_entry, 1251 build_pax_attribute_name(pax_entry_name, p)); 1252 archive_entry_set_size(pax_attr_entry, 1253 archive_strlen(&(pax->pax_header))); 1254 /* Copy uid/gid (but clip to ustar limits). */ 1255 uid = archive_entry_uid(entry_main); 1256 if (uid >= 1 << 18) 1257 uid = (1 << 18) - 1; 1258 archive_entry_set_uid(pax_attr_entry, uid); 1259 gid = archive_entry_gid(entry_main); 1260 if (gid >= 1 << 18) 1261 gid = (1 << 18) - 1; 1262 archive_entry_set_gid(pax_attr_entry, gid); 1263 /* Copy mode over (but not setuid/setgid bits) */ 1264 mode = archive_entry_mode(entry_main); 1265 #ifdef S_ISUID 1266 mode &= ~S_ISUID; 1267 #endif 1268 #ifdef S_ISGID 1269 mode &= ~S_ISGID; 1270 #endif 1271 #ifdef S_ISVTX 1272 mode &= ~S_ISVTX; 1273 #endif 1274 archive_entry_set_mode(pax_attr_entry, mode); 1275 1276 /* Copy uname/gname. */ 1277 archive_entry_set_uname(pax_attr_entry, 1278 archive_entry_uname(entry_main)); 1279 archive_entry_set_gname(pax_attr_entry, 1280 archive_entry_gname(entry_main)); 1281 1282 /* Copy mtime, but clip to ustar limits. */ 1283 s = archive_entry_mtime(entry_main); 1284 if (s < 0) { s = 0; } 1285 if (s >= 0x7fffffff) { s = 0x7fffffff; } 1286 archive_entry_set_mtime(pax_attr_entry, s, 0); 1287 1288 /* Standard ustar doesn't support atime. */ 1289 archive_entry_set_atime(pax_attr_entry, 0, 0); 1290 1291 /* Standard ustar doesn't support ctime. */ 1292 archive_entry_set_ctime(pax_attr_entry, 0, 0); 1293 1294 r = __archive_write_format_header_ustar(a, paxbuff, 1295 pax_attr_entry, 'x', 1, NULL); 1296 1297 archive_entry_free(pax_attr_entry); 1298 1299 /* Note that the 'x' header shouldn't ever fail to format */ 1300 if (r < ARCHIVE_WARN) { 1301 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1302 "archive_write_pax_header: " 1303 "'x' header failed?! This can't happen.\n"); 1304 return (ARCHIVE_FATAL); 1305 } else if (r < ret) 1306 ret = r; 1307 r = __archive_write_output(a, paxbuff, 512); 1308 if (r != ARCHIVE_OK) { 1309 sparse_list_clear(pax); 1310 pax->entry_bytes_remaining = 0; 1311 pax->entry_padding = 0; 1312 return (ARCHIVE_FATAL); 1313 } 1314 1315 pax->entry_bytes_remaining = archive_strlen(&(pax->pax_header)); 1316 pax->entry_padding = 1317 0x1ff & (-(int64_t)pax->entry_bytes_remaining); 1318 1319 r = __archive_write_output(a, pax->pax_header.s, 1320 archive_strlen(&(pax->pax_header))); 1321 if (r != ARCHIVE_OK) { 1322 /* If a write fails, we're pretty much toast. */ 1323 return (ARCHIVE_FATAL); 1324 } 1325 /* Pad out the end of the entry. */ 1326 r = __archive_write_nulls(a, (size_t)pax->entry_padding); 1327 if (r != ARCHIVE_OK) { 1328 /* If a write fails, we're pretty much toast. */ 1329 return (ARCHIVE_FATAL); 1330 } 1331 pax->entry_bytes_remaining = pax->entry_padding = 0; 1332 } 1333 1334 /* Write the header for main entry. */ 1335 r = __archive_write_output(a, ustarbuff, 512); 1336 if (r != ARCHIVE_OK) 1337 return (r); 1338 1339 /* 1340 * Inform the client of the on-disk size we're using, so 1341 * they can avoid unnecessarily writing a body for something 1342 * that we're just going to ignore. 1343 */ 1344 archive_entry_set_size(entry_original, real_size); 1345 if (pax->sparse_list == NULL && real_size > 0) { 1346 /* This is not a sparse file but we handle its data as 1347 * a sparse block. */ 1348 sparse_list_add(pax, 0, real_size); 1349 sparse_total = real_size; 1350 } 1351 pax->entry_padding = 0x1ff & (-(int64_t)sparse_total); 1352 archive_entry_free(entry_main); 1353 archive_string_free(&entry_name); 1354 1355 return (ret); 1356 } 1357 1358 /* 1359 * We need a valid name for the regular 'ustar' entry. This routine 1360 * tries to hack something more-or-less reasonable. 1361 * 1362 * The approach here tries to preserve leading dir names. We do so by 1363 * working with four sections: 1364 * 1) "prefix" directory names, 1365 * 2) "suffix" directory names, 1366 * 3) inserted dir name (optional), 1367 * 4) filename. 1368 * 1369 * These sections must satisfy the following requirements: 1370 * * Parts 1 & 2 together form an initial portion of the dir name. 1371 * * Part 3 is specified by the caller. (It should not contain a leading 1372 * or trailing '/'.) 1373 * * Part 4 forms an initial portion of the base filename. 1374 * * The filename must be <= 99 chars to fit the ustar 'name' field. 1375 * * Parts 2, 3, 4 together must be <= 99 chars to fit the ustar 'name' fld. 1376 * * Part 1 must be <= 155 chars to fit the ustar 'prefix' field. 1377 * * If the original name ends in a '/', the new name must also end in a '/' 1378 * * Trailing '/.' sequences may be stripped. 1379 * 1380 * Note: Recall that the ustar format does not store the '/' separating 1381 * parts 1 & 2, but does store the '/' separating parts 2 & 3. 1382 */ 1383 static char * 1384 build_ustar_entry_name(char *dest, const char *src, size_t src_length, 1385 const char *insert) 1386 { 1387 const char *prefix, *prefix_end; 1388 const char *suffix, *suffix_end; 1389 const char *filename, *filename_end; 1390 char *p; 1391 int need_slash = 0; /* Was there a trailing slash? */ 1392 size_t suffix_length = 99; 1393 size_t insert_length; 1394 1395 /* Length of additional dir element to be added. */ 1396 if (insert == NULL) 1397 insert_length = 0; 1398 else 1399 /* +2 here allows for '/' before and after the insert. */ 1400 insert_length = strlen(insert) + 2; 1401 1402 /* Step 0: Quick bailout in a common case. */ 1403 if (src_length < 100 && insert == NULL) { 1404 strncpy(dest, src, src_length); 1405 dest[src_length] = '\0'; 1406 return (dest); 1407 } 1408 1409 /* Step 1: Locate filename and enforce the length restriction. */ 1410 filename_end = src + src_length; 1411 /* Remove trailing '/' chars and '/.' pairs. */ 1412 for (;;) { 1413 if (filename_end > src && filename_end[-1] == '/') { 1414 filename_end --; 1415 need_slash = 1; /* Remember to restore trailing '/'. */ 1416 continue; 1417 } 1418 if (filename_end > src + 1 && filename_end[-1] == '.' 1419 && filename_end[-2] == '/') { 1420 filename_end -= 2; 1421 need_slash = 1; /* "foo/." will become "foo/" */ 1422 continue; 1423 } 1424 break; 1425 } 1426 if (need_slash) 1427 suffix_length--; 1428 /* Find start of filename. */ 1429 filename = filename_end - 1; 1430 while ((filename > src) && (*filename != '/')) 1431 filename --; 1432 if ((*filename == '/') && (filename < filename_end - 1)) 1433 filename ++; 1434 /* Adjust filename_end so that filename + insert fits in 99 chars. */ 1435 suffix_length -= insert_length; 1436 if (filename_end > filename + suffix_length) 1437 filename_end = filename + suffix_length; 1438 /* Calculate max size for "suffix" section (#3 above). */ 1439 suffix_length -= filename_end - filename; 1440 1441 /* Step 2: Locate the "prefix" section of the dirname, including 1442 * trailing '/'. */ 1443 prefix = src; 1444 prefix_end = prefix + 155; 1445 if (prefix_end > filename) 1446 prefix_end = filename; 1447 while (prefix_end > prefix && *prefix_end != '/') 1448 prefix_end--; 1449 if ((prefix_end < filename) && (*prefix_end == '/')) 1450 prefix_end++; 1451 1452 /* Step 3: Locate the "suffix" section of the dirname, 1453 * including trailing '/'. */ 1454 suffix = prefix_end; 1455 suffix_end = suffix + suffix_length; /* Enforce limit. */ 1456 if (suffix_end > filename) 1457 suffix_end = filename; 1458 if (suffix_end < suffix) 1459 suffix_end = suffix; 1460 while (suffix_end > suffix && *suffix_end != '/') 1461 suffix_end--; 1462 if ((suffix_end < filename) && (*suffix_end == '/')) 1463 suffix_end++; 1464 1465 /* Step 4: Build the new name. */ 1466 /* The OpenBSD strlcpy function is safer, but less portable. */ 1467 /* Rather than maintain two versions, just use the strncpy version. */ 1468 p = dest; 1469 if (prefix_end > prefix) { 1470 strncpy(p, prefix, prefix_end - prefix); 1471 p += prefix_end - prefix; 1472 } 1473 if (suffix_end > suffix) { 1474 strncpy(p, suffix, suffix_end - suffix); 1475 p += suffix_end - suffix; 1476 } 1477 if (insert != NULL) { 1478 /* Note: assume insert does not have leading or trailing '/' */ 1479 strcpy(p, insert); 1480 p += strlen(insert); 1481 *p++ = '/'; 1482 } 1483 strncpy(p, filename, filename_end - filename); 1484 p += filename_end - filename; 1485 if (need_slash) 1486 *p++ = '/'; 1487 *p = '\0'; 1488 1489 return (dest); 1490 } 1491 1492 /* 1493 * The ustar header for the pax extended attributes must have a 1494 * reasonable name: SUSv3 requires 'dirname'/PaxHeader.'pid'/'filename' 1495 * where 'pid' is the PID of the archiving process. Unfortunately, 1496 * that makes testing a pain since the output varies for each run, 1497 * so I'm sticking with the simpler 'dirname'/PaxHeader/'filename' 1498 * for now. (Someday, I'll make this settable. Then I can use the 1499 * SUS recommendation as default and test harnesses can override it 1500 * to get predictable results.) 1501 * 1502 * Joerg Schilling has argued that this is unnecessary because, in 1503 * practice, if the pax extended attributes get extracted as regular 1504 * files, no one is going to bother reading those attributes to 1505 * manually restore them. Based on this, 'star' uses 1506 * /tmp/PaxHeader/'basename' as the ustar header name. This is a 1507 * tempting argument, in part because it's simpler than the SUSv3 1508 * recommendation, but I'm not entirely convinced. I'm also 1509 * uncomfortable with the fact that "/tmp" is a Unix-ism. 1510 * 1511 * The following routine leverages build_ustar_entry_name() above and 1512 * so is simpler than you might think. It just needs to provide the 1513 * additional path element and handle a few pathological cases). 1514 */ 1515 static char * 1516 build_pax_attribute_name(char *dest, const char *src) 1517 { 1518 char buff[64]; 1519 const char *p; 1520 1521 /* Handle the null filename case. */ 1522 if (src == NULL || *src == '\0') { 1523 strcpy(dest, "PaxHeader/blank"); 1524 return (dest); 1525 } 1526 1527 /* Prune final '/' and other unwanted final elements. */ 1528 p = src + strlen(src); 1529 for (;;) { 1530 /* Ends in "/", remove the '/' */ 1531 if (p > src && p[-1] == '/') { 1532 --p; 1533 continue; 1534 } 1535 /* Ends in "/.", remove the '.' */ 1536 if (p > src + 1 && p[-1] == '.' 1537 && p[-2] == '/') { 1538 --p; 1539 continue; 1540 } 1541 break; 1542 } 1543 1544 /* Pathological case: After above, there was nothing left. 1545 * This includes "/." "/./." "/.//./." etc. */ 1546 if (p == src) { 1547 strcpy(dest, "/PaxHeader/rootdir"); 1548 return (dest); 1549 } 1550 1551 /* Convert unadorned "." into a suitable filename. */ 1552 if (*src == '.' && p == src + 1) { 1553 strcpy(dest, "PaxHeader/currentdir"); 1554 return (dest); 1555 } 1556 1557 /* 1558 * TODO: Push this string into the 'pax' structure to avoid 1559 * recomputing it every time. That will also open the door 1560 * to having clients override it. 1561 */ 1562 #if HAVE_GETPID && 0 /* Disable this for now; see above comment. */ 1563 sprintf(buff, "PaxHeader.%d", getpid()); 1564 #else 1565 /* If the platform can't fetch the pid, don't include it. */ 1566 strcpy(buff, "PaxHeader"); 1567 #endif 1568 /* General case: build a ustar-compatible name adding 1569 * "/PaxHeader/". */ 1570 build_ustar_entry_name(dest, src, p - src, buff); 1571 1572 return (dest); 1573 } 1574 1575 /* 1576 * GNU PAX Format 1.0 requires the special name, which pattern is: 1577 * <dir>/GNUSparseFile.<pid>/<original file name> 1578 * 1579 * This function is used for only Sparse file, a file type of which 1580 * is regular file. 1581 */ 1582 static char * 1583 build_gnu_sparse_name(char *dest, const char *src) 1584 { 1585 char buff[64]; 1586 const char *p; 1587 1588 /* Handle the null filename case. */ 1589 if (src == NULL || *src == '\0') { 1590 strcpy(dest, "GNUSparseFile/blank"); 1591 return (dest); 1592 } 1593 1594 /* Prune final '/' and other unwanted final elements. */ 1595 p = src + strlen(src); 1596 for (;;) { 1597 /* Ends in "/", remove the '/' */ 1598 if (p > src && p[-1] == '/') { 1599 --p; 1600 continue; 1601 } 1602 /* Ends in "/.", remove the '.' */ 1603 if (p > src + 1 && p[-1] == '.' 1604 && p[-2] == '/') { 1605 --p; 1606 continue; 1607 } 1608 break; 1609 } 1610 1611 #if HAVE_GETPID && 0 /* Disable this as pax attribute name. */ 1612 sprintf(buff, "GNUSparseFile.%d", getpid()); 1613 #else 1614 /* If the platform can't fetch the pid, don't include it. */ 1615 strcpy(buff, "GNUSparseFile"); 1616 #endif 1617 /* General case: build a ustar-compatible name adding 1618 * "/GNUSparseFile/". */ 1619 build_ustar_entry_name(dest, src, p - src, buff); 1620 1621 return (dest); 1622 } 1623 1624 /* Write two null blocks for the end of archive */ 1625 static int 1626 archive_write_pax_close(struct archive_write *a) 1627 { 1628 return (__archive_write_nulls(a, 512 * 2)); 1629 } 1630 1631 static int 1632 archive_write_pax_free(struct archive_write *a) 1633 { 1634 struct pax *pax; 1635 1636 pax = (struct pax *)a->format_data; 1637 if (pax == NULL) 1638 return (ARCHIVE_OK); 1639 1640 archive_string_free(&pax->pax_header); 1641 archive_string_free(&pax->sparse_map); 1642 archive_string_free(&pax->l_url_encoded_name); 1643 sparse_list_clear(pax); 1644 free(pax); 1645 a->format_data = NULL; 1646 return (ARCHIVE_OK); 1647 } 1648 1649 static int 1650 archive_write_pax_finish_entry(struct archive_write *a) 1651 { 1652 struct pax *pax; 1653 uint64_t remaining; 1654 int ret; 1655 1656 pax = (struct pax *)a->format_data; 1657 remaining = pax->entry_bytes_remaining; 1658 if (remaining == 0) { 1659 while (pax->sparse_list) { 1660 struct sparse_block *sb; 1661 if (!pax->sparse_list->is_hole) 1662 remaining += pax->sparse_list->remaining; 1663 sb = pax->sparse_list->next; 1664 free(pax->sparse_list); 1665 pax->sparse_list = sb; 1666 } 1667 } 1668 ret = __archive_write_nulls(a, (size_t)(remaining + pax->entry_padding)); 1669 pax->entry_bytes_remaining = pax->entry_padding = 0; 1670 return (ret); 1671 } 1672 1673 static ssize_t 1674 archive_write_pax_data(struct archive_write *a, const void *buff, size_t s) 1675 { 1676 struct pax *pax; 1677 size_t ws; 1678 size_t total; 1679 int ret; 1680 1681 pax = (struct pax *)a->format_data; 1682 1683 /* 1684 * According to GNU PAX format 1.0, write a sparse map 1685 * before the body. 1686 */ 1687 if (archive_strlen(&(pax->sparse_map))) { 1688 ret = __archive_write_output(a, pax->sparse_map.s, 1689 archive_strlen(&(pax->sparse_map))); 1690 if (ret != ARCHIVE_OK) 1691 return (ret); 1692 ret = __archive_write_nulls(a, pax->sparse_map_padding); 1693 if (ret != ARCHIVE_OK) 1694 return (ret); 1695 archive_string_empty(&(pax->sparse_map)); 1696 } 1697 1698 total = 0; 1699 while (total < s) { 1700 const unsigned char *p; 1701 1702 while (pax->sparse_list != NULL && 1703 pax->sparse_list->remaining == 0) { 1704 struct sparse_block *sb = pax->sparse_list->next; 1705 free(pax->sparse_list); 1706 pax->sparse_list = sb; 1707 } 1708 1709 if (pax->sparse_list == NULL) 1710 return (total); 1711 1712 p = ((const unsigned char *)buff) + total; 1713 ws = s - total; 1714 if (ws > pax->sparse_list->remaining) 1715 ws = (size_t)pax->sparse_list->remaining; 1716 1717 if (pax->sparse_list->is_hole) { 1718 /* Current block is hole thus we do not write 1719 * the body. */ 1720 pax->sparse_list->remaining -= ws; 1721 total += ws; 1722 continue; 1723 } 1724 1725 ret = __archive_write_output(a, p, ws); 1726 pax->sparse_list->remaining -= ws; 1727 total += ws; 1728 if (ret != ARCHIVE_OK) 1729 return (ret); 1730 } 1731 return (total); 1732 } 1733 1734 static int 1735 has_non_ASCII(const char *_p) 1736 { 1737 const unsigned char *p = (const unsigned char *)_p; 1738 1739 if (p == NULL) 1740 return (1); 1741 while (*p != '\0' && *p < 128) 1742 p++; 1743 return (*p != '\0'); 1744 } 1745 1746 /* 1747 * Used by extended attribute support; encodes the name 1748 * so that there will be no '=' characters in the result. 1749 */ 1750 static char * 1751 url_encode(const char *in) 1752 { 1753 const char *s; 1754 char *d; 1755 int out_len = 0; 1756 char *out; 1757 1758 for (s = in; *s != '\0'; s++) { 1759 if (*s < 33 || *s > 126 || *s == '%' || *s == '=') 1760 out_len += 3; 1761 else 1762 out_len++; 1763 } 1764 1765 out = (char *)malloc(out_len + 1); 1766 if (out == NULL) 1767 return (NULL); 1768 1769 for (s = in, d = out; *s != '\0'; s++) { 1770 /* encode any non-printable ASCII character or '%' or '=' */ 1771 if (*s < 33 || *s > 126 || *s == '%' || *s == '=') { 1772 /* URL encoding is '%' followed by two hex digits */ 1773 *d++ = '%'; 1774 *d++ = "0123456789ABCDEF"[0x0f & (*s >> 4)]; 1775 *d++ = "0123456789ABCDEF"[0x0f & *s]; 1776 } else { 1777 *d++ = *s; 1778 } 1779 } 1780 *d = '\0'; 1781 return (out); 1782 } 1783 1784 /* 1785 * Encode a sequence of bytes into a C string using base-64 encoding. 1786 * 1787 * Returns a null-terminated C string allocated with malloc(); caller 1788 * is responsible for freeing the result. 1789 */ 1790 static char * 1791 base64_encode(const char *s, size_t len) 1792 { 1793 static const char digits[64] = 1794 { 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O', 1795 'P','Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d', 1796 'e','f','g','h','i','j','k','l','m','n','o','p','q','r','s', 1797 't','u','v','w','x','y','z','0','1','2','3','4','5','6','7', 1798 '8','9','+','/' }; 1799 int v; 1800 char *d, *out; 1801 1802 /* 3 bytes becomes 4 chars, but round up and allow for trailing NUL */ 1803 out = (char *)malloc((len * 4 + 2) / 3 + 1); 1804 if (out == NULL) 1805 return (NULL); 1806 d = out; 1807 1808 /* Convert each group of 3 bytes into 4 characters. */ 1809 while (len >= 3) { 1810 v = (((int)s[0] << 16) & 0xff0000) 1811 | (((int)s[1] << 8) & 0xff00) 1812 | (((int)s[2]) & 0x00ff); 1813 s += 3; 1814 len -= 3; 1815 *d++ = digits[(v >> 18) & 0x3f]; 1816 *d++ = digits[(v >> 12) & 0x3f]; 1817 *d++ = digits[(v >> 6) & 0x3f]; 1818 *d++ = digits[(v) & 0x3f]; 1819 } 1820 /* Handle final group of 1 byte (2 chars) or 2 bytes (3 chars). */ 1821 switch (len) { 1822 case 0: break; 1823 case 1: 1824 v = (((int)s[0] << 16) & 0xff0000); 1825 *d++ = digits[(v >> 18) & 0x3f]; 1826 *d++ = digits[(v >> 12) & 0x3f]; 1827 break; 1828 case 2: 1829 v = (((int)s[0] << 16) & 0xff0000) 1830 | (((int)s[1] << 8) & 0xff00); 1831 *d++ = digits[(v >> 18) & 0x3f]; 1832 *d++ = digits[(v >> 12) & 0x3f]; 1833 *d++ = digits[(v >> 6) & 0x3f]; 1834 break; 1835 } 1836 /* Add trailing NUL character so output is a valid C string. */ 1837 *d = '\0'; 1838 return (out); 1839 } 1840 1841 static void 1842 sparse_list_clear(struct pax *pax) 1843 { 1844 while (pax->sparse_list != NULL) { 1845 struct sparse_block *sb = pax->sparse_list; 1846 pax->sparse_list = sb->next; 1847 free(sb); 1848 } 1849 pax->sparse_tail = NULL; 1850 } 1851 1852 static int 1853 _sparse_list_add_block(struct pax *pax, int64_t offset, int64_t length, 1854 int is_hole) 1855 { 1856 struct sparse_block *sb; 1857 1858 sb = (struct sparse_block *)malloc(sizeof(*sb)); 1859 if (sb == NULL) 1860 return (ARCHIVE_FATAL); 1861 sb->next = NULL; 1862 sb->is_hole = is_hole; 1863 sb->offset = offset; 1864 sb->remaining = length; 1865 if (pax->sparse_list == NULL) 1866 pax->sparse_list = pax->sparse_tail = sb; 1867 else { 1868 pax->sparse_tail->next = sb; 1869 pax->sparse_tail = sb; 1870 } 1871 return (ARCHIVE_OK); 1872 } 1873 1874 static int 1875 sparse_list_add(struct pax *pax, int64_t offset, int64_t length) 1876 { 1877 int64_t last_offset; 1878 int r; 1879 1880 if (pax->sparse_tail == NULL) 1881 last_offset = 0; 1882 else { 1883 last_offset = pax->sparse_tail->offset + 1884 pax->sparse_tail->remaining; 1885 } 1886 if (last_offset < offset) { 1887 /* Add a hole block. */ 1888 r = _sparse_list_add_block(pax, last_offset, 1889 offset - last_offset, 1); 1890 if (r != ARCHIVE_OK) 1891 return (r); 1892 } 1893 /* Add data block. */ 1894 return (_sparse_list_add_block(pax, offset, length, 0)); 1895 } 1896 1897