archive_write_set_format_pax.c revision 232153
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/contrib/libarchive/libarchive/archive_write_set_format_pax.c 232153 2012-02-25 10:58:02Z mm $"); 29 30#ifdef HAVE_ERRNO_H 31#include <errno.h> 32#endif 33#ifdef HAVE_STDLIB_H 34#include <stdlib.h> 35#endif 36#ifdef HAVE_STRING_H 37#include <string.h> 38#endif 39 40#include "archive.h" 41#include "archive_entry.h" 42#include "archive_entry_locale.h" 43#include "archive_private.h" 44#include "archive_write_private.h" 45 46struct sparse_block { 47 struct sparse_block *next; 48 int is_hole; 49 uint64_t offset; 50 uint64_t remaining; 51}; 52 53struct pax { 54 uint64_t entry_bytes_remaining; 55 uint64_t entry_padding; 56 struct archive_string l_url_encoded_name; 57 struct archive_string pax_header; 58 struct archive_string sparse_map; 59 size_t sparse_map_padding; 60 struct sparse_block *sparse_list; 61 struct sparse_block *sparse_tail; 62 struct archive_string_conv *sconv_utf8; 63 int opt_binary; 64}; 65 66static void add_pax_attr(struct archive_string *, const char *key, 67 const char *value); 68static void add_pax_attr_int(struct archive_string *, 69 const char *key, int64_t value); 70static void add_pax_attr_time(struct archive_string *, 71 const char *key, int64_t sec, 72 unsigned long nanos); 73static ssize_t archive_write_pax_data(struct archive_write *, 74 const void *, size_t); 75static int archive_write_pax_close(struct archive_write *); 76static int archive_write_pax_free(struct archive_write *); 77static int archive_write_pax_finish_entry(struct archive_write *); 78static int archive_write_pax_header(struct archive_write *, 79 struct archive_entry *); 80static int archive_write_pax_options(struct archive_write *, 81 const char *, const char *); 82static char *base64_encode(const char *src, size_t len); 83static char *build_gnu_sparse_name(char *dest, const char *src); 84static char *build_pax_attribute_name(char *dest, const char *src); 85static char *build_ustar_entry_name(char *dest, const char *src, 86 size_t src_length, const char *insert); 87static char *format_int(char *dest, int64_t); 88static int has_non_ASCII(const char *); 89static void sparse_list_clear(struct pax *); 90static int sparse_list_add(struct pax *, int64_t, int64_t); 91static char *url_encode(const char *in); 92 93/* 94 * Set output format to 'restricted pax' format. 95 * 96 * This is the same as normal 'pax', but tries to suppress 97 * the pax header whenever possible. This is the default for 98 * bsdtar, for instance. 99 */ 100int 101archive_write_set_format_pax_restricted(struct archive *_a) 102{ 103 struct archive_write *a = (struct archive_write *)_a; 104 int r; 105 106 archive_check_magic(_a, ARCHIVE_WRITE_MAGIC, 107 ARCHIVE_STATE_NEW, "archive_write_set_format_pax_restricted"); 108 109 r = archive_write_set_format_pax(&a->archive); 110 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_RESTRICTED; 111 a->archive.archive_format_name = "restricted POSIX pax interchange"; 112 return (r); 113} 114 115/* 116 * Set output format to 'pax' format. 117 */ 118int 119archive_write_set_format_pax(struct archive *_a) 120{ 121 struct archive_write *a = (struct archive_write *)_a; 122 struct pax *pax; 123 124 archive_check_magic(_a, ARCHIVE_WRITE_MAGIC, 125 ARCHIVE_STATE_NEW, "archive_write_set_format_pax"); 126 127 if (a->format_free != NULL) 128 (a->format_free)(a); 129 130 pax = (struct pax *)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 150static int 151archive_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 */ 206static void 207add_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 242static char * 243format_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 260static void 261add_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 */ 273static void 274add_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 318static int 319archive_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_in_locale( 338 &(pax->l_url_encoded_name), 339 url_encoded_name, pax->sconv_utf8); 340 free(url_encoded_name); /* Done with this. */ 341 if (r == 0) 342 encoded_name = pax->l_url_encoded_name.s; 343 else if (errno == ENOMEM) { 344 archive_set_error(&a->archive, ENOMEM, 345 "Can't allocate memory for Linkname"); 346 return (ARCHIVE_FATAL); 347 } 348 } 349 350 encoded_value = base64_encode((const char *)value, size); 351 352 if (encoded_name != NULL && encoded_value != NULL) { 353 archive_string_init(&s); 354 archive_strcpy(&s, "LIBARCHIVE.xattr."); 355 archive_strcat(&s, encoded_name); 356 add_pax_attr(&(pax->pax_header), s.s, encoded_value); 357 archive_string_free(&s); 358 } 359 free(encoded_value); 360 } 361 return (ARCHIVE_OK); 362} 363 364static int 365get_entry_hardlink(struct archive_write *a, struct archive_entry *entry, 366 const char **name, size_t *length, struct archive_string_conv *sc) 367{ 368 int r; 369 370 r = archive_entry_hardlink_l(entry, name, length, sc); 371 if (r != 0) { 372 if (errno == ENOMEM) { 373 archive_set_error(&a->archive, ENOMEM, 374 "Can't allocate memory for Linkname"); 375 return (ARCHIVE_FATAL); 376 } 377 return (ARCHIVE_WARN); 378 } 379 return (ARCHIVE_OK); 380} 381 382static int 383get_entry_pathname(struct archive_write *a, struct archive_entry *entry, 384 const char **name, size_t *length, struct archive_string_conv *sc) 385{ 386 int r; 387 388 r = archive_entry_pathname_l(entry, name, length, sc); 389 if (r != 0) { 390 if (errno == ENOMEM) { 391 archive_set_error(&a->archive, ENOMEM, 392 "Can't allocate memory for Pathname"); 393 return (ARCHIVE_FATAL); 394 } 395 return (ARCHIVE_WARN); 396 } 397 return (ARCHIVE_OK); 398} 399 400static int 401get_entry_uname(struct archive_write *a, struct archive_entry *entry, 402 const char **name, size_t *length, struct archive_string_conv *sc) 403{ 404 int r; 405 406 r = archive_entry_uname_l(entry, name, length, sc); 407 if (r != 0) { 408 if (errno == ENOMEM) { 409 archive_set_error(&a->archive, ENOMEM, 410 "Can't allocate memory for Uname"); 411 return (ARCHIVE_FATAL); 412 } 413 return (ARCHIVE_WARN); 414 } 415 return (ARCHIVE_OK); 416} 417 418static int 419get_entry_gname(struct archive_write *a, struct archive_entry *entry, 420 const char **name, size_t *length, struct archive_string_conv *sc) 421{ 422 int r; 423 424 r = archive_entry_gname_l(entry, name, length, sc); 425 if (r != 0) { 426 if (errno == ENOMEM) { 427 archive_set_error(&a->archive, ENOMEM, 428 "Can't allocate memory for Gname"); 429 return (ARCHIVE_FATAL); 430 } 431 return (ARCHIVE_WARN); 432 } 433 return (ARCHIVE_OK); 434} 435 436static int 437get_entry_symlink(struct archive_write *a, struct archive_entry *entry, 438 const char **name, size_t *length, struct archive_string_conv *sc) 439{ 440 int r; 441 442 r = archive_entry_symlink_l(entry, name, length, sc); 443 if (r != 0) { 444 if (errno == ENOMEM) { 445 archive_set_error(&a->archive, ENOMEM, 446 "Can't allocate memory for Linkname"); 447 return (ARCHIVE_FATAL); 448 } 449 return (ARCHIVE_WARN); 450 } 451 return (ARCHIVE_OK); 452} 453 454/* 455 * TODO: Consider adding 'comment' and 'charset' fields to 456 * archive_entry so that clients can specify them. Also, consider 457 * adding generic key/value tags so clients can add arbitrary 458 * key/value data. 459 * 460 * TODO: Break up this 700-line function!!!! Yowza! 461 */ 462static int 463archive_write_pax_header(struct archive_write *a, 464 struct archive_entry *entry_original) 465{ 466 struct archive_entry *entry_main; 467 const char *p; 468 const char *suffix; 469 int need_extension, r, ret; 470 int sparse_count; 471 uint64_t sparse_total, real_size; 472 struct pax *pax; 473 const char *hardlink; 474 const char *path = NULL, *linkpath = NULL; 475 const char *uname = NULL, *gname = NULL; 476 const void *mac_metadata; 477 size_t mac_metadata_size; 478 struct archive_string_conv *sconv; 479 size_t hardlink_length, path_length, linkpath_length; 480 size_t uname_length, gname_length; 481 482 char paxbuff[512]; 483 char ustarbuff[512]; 484 char ustar_entry_name[256]; 485 char pax_entry_name[256]; 486 char gnu_sparse_name[256]; 487 struct archive_string entry_name; 488 489 ret = ARCHIVE_OK; 490 need_extension = 0; 491 pax = (struct pax *)a->format_data; 492 493 /* Sanity check. */ 494 if (archive_entry_pathname(entry_original) == NULL) { 495 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 496 "Can't record entry in tar file without pathname"); 497 return (ARCHIVE_FAILED); 498 } 499 500 /* 501 * Choose a header encoding. 502 */ 503 if (pax->opt_binary) 504 sconv = NULL;/* Binary mode. */ 505 else { 506 /* Header encoding is UTF-8. */ 507 if (pax->sconv_utf8 == NULL) { 508 /* Initialize the string conversion object 509 * we must need */ 510 pax->sconv_utf8 = archive_string_conversion_to_charset( 511 &(a->archive), "UTF-8", 1); 512 if (pax->sconv_utf8 == NULL) 513 /* Couldn't allocate memory */ 514 return (ARCHIVE_FAILED); 515 } 516 sconv = pax->sconv_utf8; 517 } 518 519 r = get_entry_hardlink(a, entry_original, &hardlink, 520 &hardlink_length, sconv); 521 if (r == ARCHIVE_FATAL) 522 return (r); 523 else if (r != ARCHIVE_OK) { 524 r = get_entry_hardlink(a, entry_original, &hardlink, 525 &hardlink_length, NULL); 526 if (r == ARCHIVE_FATAL) 527 return (r); 528 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 529 "Can't translate linkname '%s' to %s", hardlink, 530 archive_string_conversion_charset_name(sconv)); 531 ret = ARCHIVE_WARN; 532 sconv = NULL;/* The header charset switches to binary mode. */ 533 } 534 535 /* Make sure this is a type of entry that we can handle here */ 536 if (hardlink == NULL) { 537 switch (archive_entry_filetype(entry_original)) { 538 case AE_IFBLK: 539 case AE_IFCHR: 540 case AE_IFIFO: 541 case AE_IFLNK: 542 case AE_IFREG: 543 break; 544 case AE_IFDIR: 545 { 546 /* 547 * Ensure a trailing '/'. Modify the original 548 * entry so the client sees the change. 549 */ 550#if defined(_WIN32) && !defined(__CYGWIN__) 551 const wchar_t *wp; 552 553 wp = archive_entry_pathname_w(entry_original); 554 if (wp != NULL && wp[wcslen(wp) -1] != L'/') { 555 struct archive_wstring ws; 556 557 archive_string_init(&ws); 558 path_length = wcslen(wp); 559 if (archive_wstring_ensure(&ws, 560 path_length + 2) == NULL) { 561 archive_set_error(&a->archive, ENOMEM, 562 "Can't allocate pax data"); 563 archive_wstring_free(&ws); 564 return(ARCHIVE_FATAL); 565 } 566 /* Should we keep '\' ? */ 567 if (wp[path_length -1] == L'\\') 568 path_length--; 569 archive_wstrncpy(&ws, wp, path_length); 570 archive_wstrappend_wchar(&ws, L'/'); 571 archive_entry_copy_pathname_w( 572 entry_original, ws.s); 573 archive_wstring_free(&ws); 574 p = NULL; 575 } else 576#endif 577 p = archive_entry_pathname(entry_original); 578 /* 579 * On Windows, this is a backup operation just in 580 * case getting WCS failed. On POSIX, this is a 581 * normal operation. 582 */ 583 if (p != NULL && p[strlen(p) - 1] != '/') { 584 struct archive_string as; 585 586 archive_string_init(&as); 587 path_length = strlen(p); 588 if (archive_string_ensure(&as, 589 path_length + 2) == NULL) { 590 archive_set_error(&a->archive, ENOMEM, 591 "Can't allocate pax data"); 592 archive_string_free(&as); 593 return(ARCHIVE_FATAL); 594 } 595#if defined(_WIN32) && !defined(__CYGWIN__) 596 /* NOTE: This might break the pathname 597 * if the current code page is CP932 and 598 * the pathname includes a character '\' 599 * as a part of its multibyte pathname. */ 600 if (p[strlen(p) -1] == '\\') 601 path_length--; 602 else 603#endif 604 archive_strncpy(&as, p, path_length); 605 archive_strappend_char(&as, '/'); 606 archive_entry_copy_pathname( 607 entry_original, as.s); 608 archive_string_free(&as); 609 } 610 break; 611 } 612 case AE_IFSOCK: 613 archive_set_error(&a->archive, 614 ARCHIVE_ERRNO_FILE_FORMAT, 615 "tar format cannot archive socket"); 616 return (ARCHIVE_FAILED); 617 default: 618 archive_set_error(&a->archive, 619 ARCHIVE_ERRNO_FILE_FORMAT, 620 "tar format cannot archive this (type=0%lo)", 621 (unsigned long) 622 archive_entry_filetype(entry_original)); 623 return (ARCHIVE_FAILED); 624 } 625 } 626 627 /* 628 * If Mac OS metadata blob is here, recurse to write that 629 * as a separate entry. This is really a pretty poor design: 630 * In particular, it doubles the overhead for long filenames. 631 * TODO: Help Apple folks design something better and figure 632 * out how to transition from this legacy format. 633 * 634 * Note that this code is present on every platform; clients 635 * on non-Mac are unlikely to ever provide this data, but 636 * applications that copy entries from one archive to another 637 * should not lose data just because the local filesystem 638 * can't store it. 639 */ 640 mac_metadata = 641 archive_entry_mac_metadata(entry_original, &mac_metadata_size); 642 if (mac_metadata != NULL) { 643 const char *oname; 644 char *name, *bname; 645 size_t name_length; 646 struct archive_entry *extra = archive_entry_new2(&a->archive); 647 648 oname = archive_entry_pathname(entry_original); 649 name_length = strlen(oname); 650 name = malloc(name_length + 3); 651 if (name == NULL) { 652 /* XXX error message */ 653 return (ARCHIVE_FAILED); 654 } 655 strcpy(name, oname); 656 /* Find last '/'; strip trailing '/' characters */ 657 bname = strrchr(name, '/'); 658 while (bname != NULL && bname[1] == '\0') { 659 *bname = '\0'; 660 bname = strrchr(name, '/'); 661 } 662 if (bname == NULL) { 663 memmove(name + 2, name, name_length + 1); 664 memmove(name, "._", 2); 665 } else { 666 bname += 1; 667 memmove(bname + 2, bname, strlen(bname) + 1); 668 memmove(bname, "._", 2); 669 } 670 archive_entry_copy_pathname(extra, name); 671 free(name); 672 673 archive_entry_set_size(extra, mac_metadata_size); 674 archive_entry_set_filetype(extra, AE_IFREG); 675 archive_entry_set_perm(extra, 676 archive_entry_perm(entry_original)); 677 archive_entry_set_mtime(extra, 678 archive_entry_mtime(entry_original), 679 archive_entry_mtime_nsec(entry_original)); 680 archive_entry_set_gid(extra, 681 archive_entry_gid(entry_original)); 682 archive_entry_set_gname(extra, 683 archive_entry_gname(entry_original)); 684 archive_entry_set_uid(extra, 685 archive_entry_uid(entry_original)); 686 archive_entry_set_uname(extra, 687 archive_entry_uname(entry_original)); 688 689 /* Recurse to write the special copyfile entry. */ 690 r = archive_write_pax_header(a, extra); 691 if (r < ARCHIVE_WARN) 692 return (r); 693 if (r < ret) 694 ret = r; 695 r = archive_write_pax_data(a, mac_metadata, mac_metadata_size); 696 if (r < ARCHIVE_WARN) 697 return (r); 698 if (r < ret) 699 ret = r; 700 r = archive_write_pax_finish_entry(a); 701 if (r < ARCHIVE_WARN) 702 return (r); 703 if (r < ret) 704 ret = r; 705 } 706 707 /* Copy entry so we can modify it as needed. */ 708#if defined(_WIN32) && !defined(__CYGWIN__) 709 /* Make sure the path separators in pahtname, hardlink and symlink 710 * are all slash '/', not the Windows path separator '\'. */ 711 entry_main = __la_win_entry_in_posix_pathseparator(entry_original); 712 if (entry_main == entry_original) 713 entry_main = archive_entry_clone(entry_original); 714#else 715 entry_main = archive_entry_clone(entry_original); 716#endif 717 if (entry_main == NULL) { 718 archive_set_error(&a->archive, ENOMEM, 719 "Can't allocate pax data"); 720 return(ARCHIVE_FATAL); 721 } 722 archive_string_empty(&(pax->pax_header)); /* Blank our work area. */ 723 archive_string_empty(&(pax->sparse_map)); 724 sparse_total = 0; 725 sparse_list_clear(pax); 726 727 if (hardlink == NULL && 728 archive_entry_filetype(entry_main) == AE_IFREG) 729 sparse_count = archive_entry_sparse_reset(entry_main); 730 else 731 sparse_count = 0; 732 if (sparse_count) { 733 int64_t offset, length, last_offset = 0; 734 /* Get the last entry of sparse block. */ 735 while (archive_entry_sparse_next( 736 entry_main, &offset, &length) == ARCHIVE_OK) 737 last_offset = offset + length; 738 739 /* If the last sparse block does not reach the end of file, 740 * We have to add a empty sparse block as the last entry to 741 * manage storing file data. */ 742 if (last_offset < archive_entry_size(entry_main)) 743 archive_entry_sparse_add_entry(entry_main, 744 archive_entry_size(entry_main), 0); 745 sparse_count = archive_entry_sparse_reset(entry_main); 746 } 747 748 /* 749 * First, check the name fields and see if any of them 750 * require binary coding. If any of them does, then all of 751 * them do. 752 */ 753 r = get_entry_pathname(a, entry_main, &path, &path_length, sconv); 754 if (r == ARCHIVE_FATAL) 755 return (r); 756 else if (r != ARCHIVE_OK) { 757 r = get_entry_pathname(a, entry_main, &path, 758 &path_length, NULL); 759 if (r == ARCHIVE_FATAL) 760 return (r); 761 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 762 "Can't translate pathname '%s' to %s", path, 763 archive_string_conversion_charset_name(sconv)); 764 ret = ARCHIVE_WARN; 765 sconv = NULL;/* The header charset switches to binary mode. */ 766 } 767 r = get_entry_uname(a, entry_main, &uname, &uname_length, sconv); 768 if (r == ARCHIVE_FATAL) 769 return (r); 770 else if (r != ARCHIVE_OK) { 771 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL); 772 if (r == ARCHIVE_FATAL) 773 return (r); 774 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 775 "Can't translate uname '%s' to %s", uname, 776 archive_string_conversion_charset_name(sconv)); 777 ret = ARCHIVE_WARN; 778 sconv = NULL;/* The header charset switches to binary mode. */ 779 } 780 r = get_entry_gname(a, entry_main, &gname, &gname_length, sconv); 781 if (r == ARCHIVE_FATAL) 782 return (r); 783 else if (r != ARCHIVE_OK) { 784 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL); 785 if (r == ARCHIVE_FATAL) 786 return (r); 787 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 788 "Can't translate gname '%s' to %s", gname, 789 archive_string_conversion_charset_name(sconv)); 790 ret = ARCHIVE_WARN; 791 sconv = NULL;/* The header charset switches to binary mode. */ 792 } 793 linkpath = hardlink; 794 linkpath_length = hardlink_length; 795 if (linkpath == NULL) { 796 r = get_entry_symlink(a, entry_main, &linkpath, 797 &linkpath_length, sconv); 798 if (r == ARCHIVE_FATAL) 799 return (r); 800 else if (r != ARCHIVE_OK) { 801 r = get_entry_symlink(a, entry_main, &linkpath, 802 &linkpath_length, NULL); 803 if (r == ARCHIVE_FATAL) 804 return (r); 805 archive_set_error(&a->archive, 806 ARCHIVE_ERRNO_FILE_FORMAT, 807 "Can't translate linkname '%s' to %s", linkpath, 808 archive_string_conversion_charset_name(sconv)); 809 ret = ARCHIVE_WARN; 810 sconv = NULL; 811 } 812 } 813 814 /* If any string conversions failed, get all attributes 815 * in binary-mode. */ 816 if (sconv == NULL && !pax->opt_binary) { 817 if (hardlink != NULL) { 818 r = get_entry_hardlink(a, entry_main, &hardlink, 819 &hardlink_length, NULL); 820 if (r == ARCHIVE_FATAL) 821 return (r); 822 linkpath = hardlink; 823 linkpath_length = hardlink_length; 824 } 825 r = get_entry_pathname(a, entry_main, &path, 826 &path_length, NULL); 827 if (r == ARCHIVE_FATAL) 828 return (r); 829 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL); 830 if (r == ARCHIVE_FATAL) 831 return (r); 832 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL); 833 if (r == ARCHIVE_FATAL) 834 return (r); 835 } 836 837 /* Store the header encoding first, to be nice to readers. */ 838 if (sconv == NULL) 839 add_pax_attr(&(pax->pax_header), "hdrcharset", "BINARY"); 840 841 842 /* 843 * If name is too long, or has non-ASCII characters, add 844 * 'path' to pax extended attrs. (Note that an unconvertible 845 * name must have non-ASCII characters.) 846 */ 847 if (has_non_ASCII(path)) { 848 /* We have non-ASCII characters. */ 849 add_pax_attr(&(pax->pax_header), "path", path); 850 archive_entry_set_pathname(entry_main, 851 build_ustar_entry_name(ustar_entry_name, 852 path, path_length, NULL)); 853 need_extension = 1; 854 } else { 855 /* We have an all-ASCII path; we'd like to just store 856 * it in the ustar header if it will fit. Yes, this 857 * duplicates some of the logic in 858 * archive_write_set_format_ustar.c 859 */ 860 if (path_length <= 100) { 861 /* Fits in the old 100-char tar name field. */ 862 } else { 863 /* Find largest suffix that will fit. */ 864 /* Note: strlen() > 100, so strlen() - 100 - 1 >= 0 */ 865 suffix = strchr(path + path_length - 100 - 1, '/'); 866 /* Don't attempt an empty prefix. */ 867 if (suffix == path) 868 suffix = strchr(suffix + 1, '/'); 869 /* We can put it in the ustar header if it's 870 * all ASCII and it's either <= 100 characters 871 * or can be split at a '/' into a prefix <= 872 * 155 chars and a suffix <= 100 chars. (Note 873 * the strchr() above will return NULL exactly 874 * when the path can't be split.) 875 */ 876 if (suffix == NULL /* Suffix > 100 chars. */ 877 || suffix[1] == '\0' /* empty suffix */ 878 || suffix - path > 155) /* Prefix > 155 chars */ 879 { 880 add_pax_attr(&(pax->pax_header), "path", path); 881 archive_entry_set_pathname(entry_main, 882 build_ustar_entry_name(ustar_entry_name, 883 path, path_length, NULL)); 884 need_extension = 1; 885 } 886 } 887 } 888 889 if (linkpath != NULL) { 890 /* If link name is too long or has non-ASCII characters, add 891 * 'linkpath' to pax extended attrs. */ 892 if (linkpath_length > 100 || has_non_ASCII(linkpath)) { 893 add_pax_attr(&(pax->pax_header), "linkpath", linkpath); 894 if (linkpath_length > 100) { 895 if (hardlink != NULL) 896 archive_entry_set_hardlink(entry_main, 897 "././@LongHardLink"); 898 else 899 archive_entry_set_symlink(entry_main, 900 "././@LongSymLink"); 901 } 902 need_extension = 1; 903 } 904 } 905 /* Save a pathname since it will be renamed if `entry_main` has 906 * sparse blocks. */ 907 archive_string_init(&entry_name); 908 archive_strcpy(&entry_name, archive_entry_pathname(entry_main)); 909 910 /* If file size is too large, add 'size' to pax extended attrs. */ 911 if (archive_entry_size(entry_main) >= (((int64_t)1) << 33)) { 912 add_pax_attr_int(&(pax->pax_header), "size", 913 archive_entry_size(entry_main)); 914 need_extension = 1; 915 } 916 917 /* If numeric GID is too large, add 'gid' to pax extended attrs. */ 918 if ((unsigned int)archive_entry_gid(entry_main) >= (1 << 18)) { 919 add_pax_attr_int(&(pax->pax_header), "gid", 920 archive_entry_gid(entry_main)); 921 need_extension = 1; 922 } 923 924 /* If group name is too large or has non-ASCII characters, add 925 * 'gname' to pax extended attrs. */ 926 if (gname != NULL) { 927 if (gname_length > 31 || has_non_ASCII(gname)) { 928 add_pax_attr(&(pax->pax_header), "gname", gname); 929 need_extension = 1; 930 } 931 } 932 933 /* If numeric UID is too large, add 'uid' to pax extended attrs. */ 934 if ((unsigned int)archive_entry_uid(entry_main) >= (1 << 18)) { 935 add_pax_attr_int(&(pax->pax_header), "uid", 936 archive_entry_uid(entry_main)); 937 need_extension = 1; 938 } 939 940 /* Add 'uname' to pax extended attrs if necessary. */ 941 if (uname != NULL) { 942 if (uname_length > 31 || has_non_ASCII(uname)) { 943 add_pax_attr(&(pax->pax_header), "uname", uname); 944 need_extension = 1; 945 } 946 } 947 948 /* 949 * POSIX/SUSv3 doesn't provide a standard key for large device 950 * numbers. I use the same keys here that Joerg Schilling 951 * used for 'star.' (Which, somewhat confusingly, are called 952 * "devXXX" even though they code "rdev" values.) No doubt, 953 * other implementations use other keys. Note that there's no 954 * reason we can't write the same information into a number of 955 * different keys. 956 * 957 * Of course, this is only needed for block or char device entries. 958 */ 959 if (archive_entry_filetype(entry_main) == AE_IFBLK 960 || archive_entry_filetype(entry_main) == AE_IFCHR) { 961 /* 962 * If rdevmajor is too large, add 'SCHILY.devmajor' to 963 * extended attributes. 964 */ 965 int rdevmajor, rdevminor; 966 rdevmajor = archive_entry_rdevmajor(entry_main); 967 rdevminor = archive_entry_rdevminor(entry_main); 968 if (rdevmajor >= (1 << 18)) { 969 add_pax_attr_int(&(pax->pax_header), "SCHILY.devmajor", 970 rdevmajor); 971 /* 972 * Non-strict formatting below means we don't 973 * have to truncate here. Not truncating improves 974 * the chance that some more modern tar archivers 975 * (such as GNU tar 1.13) can restore the full 976 * value even if they don't understand the pax 977 * extended attributes. See my rant below about 978 * file size fields for additional details. 979 */ 980 /* archive_entry_set_rdevmajor(entry_main, 981 rdevmajor & ((1 << 18) - 1)); */ 982 need_extension = 1; 983 } 984 985 /* 986 * If devminor is too large, add 'SCHILY.devminor' to 987 * extended attributes. 988 */ 989 if (rdevminor >= (1 << 18)) { 990 add_pax_attr_int(&(pax->pax_header), "SCHILY.devminor", 991 rdevminor); 992 /* Truncation is not necessary here, either. */ 993 /* archive_entry_set_rdevminor(entry_main, 994 rdevminor & ((1 << 18) - 1)); */ 995 need_extension = 1; 996 } 997 } 998 999 /* 1000 * Technically, the mtime field in the ustar header can 1001 * support 33 bits, but many platforms use signed 32-bit time 1002 * values. The cutoff of 0x7fffffff here is a compromise. 1003 * Yes, this check is duplicated just below; this helps to 1004 * avoid writing an mtime attribute just to handle a 1005 * high-resolution timestamp in "restricted pax" mode. 1006 */ 1007 if (!need_extension && 1008 ((archive_entry_mtime(entry_main) < 0) 1009 || (archive_entry_mtime(entry_main) >= 0x7fffffff))) 1010 need_extension = 1; 1011 1012 /* I use a star-compatible file flag attribute. */ 1013 p = archive_entry_fflags_text(entry_main); 1014 if (!need_extension && p != NULL && *p != '\0') 1015 need_extension = 1; 1016 1017 /* If there are non-trivial ACL entries, we need an extension. */ 1018 if (!need_extension && archive_entry_acl_count(entry_original, 1019 ARCHIVE_ENTRY_ACL_TYPE_ACCESS) > 0) 1020 need_extension = 1; 1021 1022 /* If there are non-trivial ACL entries, we need an extension. */ 1023 if (!need_extension && archive_entry_acl_count(entry_original, 1024 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT) > 0) 1025 need_extension = 1; 1026 1027 /* If there are extended attributes, we need an extension */ 1028 if (!need_extension && archive_entry_xattr_count(entry_original) > 0) 1029 need_extension = 1; 1030 1031 /* If there are sparse info, we need an extension */ 1032 if (!need_extension && sparse_count > 0) 1033 need_extension = 1; 1034 1035 /* 1036 * The following items are handled differently in "pax 1037 * restricted" format. In particular, in "pax restricted" 1038 * format they won't be added unless need_extension is 1039 * already set (we're already generating an extended header, so 1040 * may as well include these). 1041 */ 1042 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_RESTRICTED || 1043 need_extension) { 1044 1045 if (archive_entry_mtime(entry_main) < 0 || 1046 archive_entry_mtime(entry_main) >= 0x7fffffff || 1047 archive_entry_mtime_nsec(entry_main) != 0) 1048 add_pax_attr_time(&(pax->pax_header), "mtime", 1049 archive_entry_mtime(entry_main), 1050 archive_entry_mtime_nsec(entry_main)); 1051 1052 if (archive_entry_ctime(entry_main) != 0 || 1053 archive_entry_ctime_nsec(entry_main) != 0) 1054 add_pax_attr_time(&(pax->pax_header), "ctime", 1055 archive_entry_ctime(entry_main), 1056 archive_entry_ctime_nsec(entry_main)); 1057 1058 if (archive_entry_atime(entry_main) != 0 || 1059 archive_entry_atime_nsec(entry_main) != 0) 1060 add_pax_attr_time(&(pax->pax_header), "atime", 1061 archive_entry_atime(entry_main), 1062 archive_entry_atime_nsec(entry_main)); 1063 1064 /* Store birth/creationtime only if it's earlier than mtime */ 1065 if (archive_entry_birthtime_is_set(entry_main) && 1066 archive_entry_birthtime(entry_main) 1067 < archive_entry_mtime(entry_main)) 1068 add_pax_attr_time(&(pax->pax_header), 1069 "LIBARCHIVE.creationtime", 1070 archive_entry_birthtime(entry_main), 1071 archive_entry_birthtime_nsec(entry_main)); 1072 1073 /* I use a star-compatible file flag attribute. */ 1074 p = archive_entry_fflags_text(entry_main); 1075 if (p != NULL && *p != '\0') 1076 add_pax_attr(&(pax->pax_header), "SCHILY.fflags", p); 1077 1078 /* I use star-compatible ACL attributes. */ 1079 r = archive_entry_acl_text_l(entry_original, 1080 ARCHIVE_ENTRY_ACL_TYPE_ACCESS | 1081 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID, 1082 &p, NULL, pax->sconv_utf8); 1083 if (r != 0) { 1084 if (errno == ENOMEM) { 1085 archive_set_error(&a->archive, ENOMEM, 1086 "Can't allocate memory for " 1087 "ACL.access"); 1088 return (ARCHIVE_FATAL); 1089 } 1090 archive_set_error(&a->archive, 1091 ARCHIVE_ERRNO_FILE_FORMAT, 1092 "Can't translate ACL.access to UTF-8"); 1093 ret = ARCHIVE_WARN; 1094 } else if (p != NULL && *p != '\0') { 1095 add_pax_attr(&(pax->pax_header), 1096 "SCHILY.acl.access", p); 1097 } 1098 r = archive_entry_acl_text_l(entry_original, 1099 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT | 1100 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID, 1101 &p, NULL, pax->sconv_utf8); 1102 if (r != 0) { 1103 if (errno == ENOMEM) { 1104 archive_set_error(&a->archive, ENOMEM, 1105 "Can't allocate memory for " 1106 "ACL.default"); 1107 return (ARCHIVE_FATAL); 1108 } 1109 archive_set_error(&a->archive, 1110 ARCHIVE_ERRNO_FILE_FORMAT, 1111 "Can't translate ACL.default to UTF-8"); 1112 ret = ARCHIVE_WARN; 1113 } else if (p != NULL && *p != '\0') { 1114 add_pax_attr(&(pax->pax_header), 1115 "SCHILY.acl.default", p); 1116 } 1117 1118 /* We use GNU-tar-compatible sparse attributes. */ 1119 if (sparse_count > 0) { 1120 int64_t soffset, slength; 1121 1122 add_pax_attr_int(&(pax->pax_header), 1123 "GNU.sparse.major", 1); 1124 add_pax_attr_int(&(pax->pax_header), 1125 "GNU.sparse.minor", 0); 1126 add_pax_attr(&(pax->pax_header), 1127 "GNU.sparse.name", entry_name.s); 1128 add_pax_attr_int(&(pax->pax_header), 1129 "GNU.sparse.realsize", 1130 archive_entry_size(entry_main)); 1131 1132 /* Rename the file name which will be used for 1133 * ustar header to a special name, which GNU 1134 * PAX Format 1.0 requires */ 1135 archive_entry_set_pathname(entry_main, 1136 build_gnu_sparse_name(gnu_sparse_name, 1137 entry_name.s)); 1138 1139 /* 1140 * - Make a sparse map, which will precede a file data. 1141 * - Get the total size of available data of sparse. 1142 */ 1143 archive_string_sprintf(&(pax->sparse_map), "%d\n", 1144 sparse_count); 1145 while (archive_entry_sparse_next(entry_main, 1146 &soffset, &slength) == ARCHIVE_OK) { 1147 archive_string_sprintf(&(pax->sparse_map), 1148 "%jd\n%jd\n", 1149 (intmax_t)soffset, 1150 (intmax_t)slength); 1151 sparse_total += slength; 1152 if (sparse_list_add(pax, soffset, slength) 1153 != ARCHIVE_OK) { 1154 archive_set_error(&a->archive, 1155 ENOMEM, 1156 "Can't allocate memory"); 1157 archive_entry_free(entry_main); 1158 archive_string_free(&entry_name); 1159 return (ARCHIVE_FATAL); 1160 } 1161 } 1162 } 1163 1164 /* Store extended attributes */ 1165 if (archive_write_pax_header_xattrs(a, pax, entry_original) 1166 == ARCHIVE_FATAL) { 1167 archive_entry_free(entry_main); 1168 archive_string_free(&entry_name); 1169 return (ARCHIVE_FATAL); 1170 } 1171 } 1172 1173 /* Only regular files have data. */ 1174 if (archive_entry_filetype(entry_main) != AE_IFREG) 1175 archive_entry_set_size(entry_main, 0); 1176 1177 /* 1178 * Pax-restricted does not store data for hardlinks, in order 1179 * to improve compatibility with ustar. 1180 */ 1181 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE && 1182 hardlink != NULL) 1183 archive_entry_set_size(entry_main, 0); 1184 1185 /* 1186 * XXX Full pax interchange format does permit a hardlink 1187 * entry to have data associated with it. I'm not supporting 1188 * that here because the client expects me to tell them whether 1189 * or not this format expects data for hardlinks. If I 1190 * don't check here, then every pax archive will end up with 1191 * duplicated data for hardlinks. Someday, there may be 1192 * need to select this behavior, in which case the following 1193 * will need to be revisited. XXX 1194 */ 1195 if (hardlink != NULL) 1196 archive_entry_set_size(entry_main, 0); 1197 1198 /* Save a real file size. */ 1199 real_size = archive_entry_size(entry_main); 1200 /* 1201 * Overwrite a file size by the total size of sparse blocks and 1202 * the size of sparse map info. That file size is the length of 1203 * the data, which we will exactly store into an archive file. 1204 */ 1205 if (archive_strlen(&(pax->sparse_map))) { 1206 size_t mapsize = archive_strlen(&(pax->sparse_map)); 1207 pax->sparse_map_padding = 0x1ff & (-(ssize_t)mapsize); 1208 archive_entry_set_size(entry_main, 1209 mapsize + pax->sparse_map_padding + sparse_total); 1210 } 1211 1212 /* Format 'ustar' header for main entry. 1213 * 1214 * The trouble with file size: If the reader can't understand 1215 * the file size, they may not be able to locate the next 1216 * entry and the rest of the archive is toast. Pax-compliant 1217 * readers are supposed to ignore the file size in the main 1218 * header, so the question becomes how to maximize portability 1219 * for readers that don't support pax attribute extensions. 1220 * For maximum compatibility, I permit numeric extensions in 1221 * the main header so that the file size stored will always be 1222 * correct, even if it's in a format that only some 1223 * implementations understand. The technique used here is: 1224 * 1225 * a) If possible, follow the standard exactly. This handles 1226 * files up to 8 gigabytes minus 1. 1227 * 1228 * b) If that fails, try octal but omit the field terminator. 1229 * That handles files up to 64 gigabytes minus 1. 1230 * 1231 * c) Otherwise, use base-256 extensions. That handles files 1232 * up to 2^63 in this implementation, with the potential to 1233 * go up to 2^94. That should hold us for a while. ;-) 1234 * 1235 * The non-strict formatter uses similar logic for other 1236 * numeric fields, though they're less critical. 1237 */ 1238 if (__archive_write_format_header_ustar(a, ustarbuff, entry_main, -1, 0, 1239 NULL) == ARCHIVE_FATAL) 1240 return (ARCHIVE_FATAL); 1241 1242 /* If we built any extended attributes, write that entry first. */ 1243 if (archive_strlen(&(pax->pax_header)) > 0) { 1244 struct archive_entry *pax_attr_entry; 1245 time_t s; 1246 int64_t uid, gid; 1247 int mode; 1248 1249 pax_attr_entry = archive_entry_new2(&a->archive); 1250 p = entry_name.s; 1251 archive_entry_set_pathname(pax_attr_entry, 1252 build_pax_attribute_name(pax_entry_name, p)); 1253 archive_entry_set_size(pax_attr_entry, 1254 archive_strlen(&(pax->pax_header))); 1255 /* Copy uid/gid (but clip to ustar limits). */ 1256 uid = archive_entry_uid(entry_main); 1257 if (uid >= 1 << 18) 1258 uid = (1 << 18) - 1; 1259 archive_entry_set_uid(pax_attr_entry, uid); 1260 gid = archive_entry_gid(entry_main); 1261 if (gid >= 1 << 18) 1262 gid = (1 << 18) - 1; 1263 archive_entry_set_gid(pax_attr_entry, gid); 1264 /* Copy mode over (but not setuid/setgid bits) */ 1265 mode = archive_entry_mode(entry_main); 1266#ifdef S_ISUID 1267 mode &= ~S_ISUID; 1268#endif 1269#ifdef S_ISGID 1270 mode &= ~S_ISGID; 1271#endif 1272#ifdef S_ISVTX 1273 mode &= ~S_ISVTX; 1274#endif 1275 archive_entry_set_mode(pax_attr_entry, mode); 1276 1277 /* Copy uname/gname. */ 1278 archive_entry_set_uname(pax_attr_entry, 1279 archive_entry_uname(entry_main)); 1280 archive_entry_set_gname(pax_attr_entry, 1281 archive_entry_gname(entry_main)); 1282 1283 /* Copy mtime, but clip to ustar limits. */ 1284 s = archive_entry_mtime(entry_main); 1285 if (s < 0) { s = 0; } 1286 if (s >= 0x7fffffff) { s = 0x7fffffff; } 1287 archive_entry_set_mtime(pax_attr_entry, s, 0); 1288 1289 /* Standard ustar doesn't support atime. */ 1290 archive_entry_set_atime(pax_attr_entry, 0, 0); 1291 1292 /* Standard ustar doesn't support ctime. */ 1293 archive_entry_set_ctime(pax_attr_entry, 0, 0); 1294 1295 r = __archive_write_format_header_ustar(a, paxbuff, 1296 pax_attr_entry, 'x', 1, NULL); 1297 1298 archive_entry_free(pax_attr_entry); 1299 1300 /* Note that the 'x' header shouldn't ever fail to format */ 1301 if (r < ARCHIVE_WARN) { 1302 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1303 "archive_write_pax_header: " 1304 "'x' header failed?! This can't happen.\n"); 1305 return (ARCHIVE_FATAL); 1306 } else if (r < ret) 1307 ret = r; 1308 r = __archive_write_output(a, paxbuff, 512); 1309 if (r != ARCHIVE_OK) { 1310 sparse_list_clear(pax); 1311 pax->entry_bytes_remaining = 0; 1312 pax->entry_padding = 0; 1313 return (ARCHIVE_FATAL); 1314 } 1315 1316 pax->entry_bytes_remaining = archive_strlen(&(pax->pax_header)); 1317 pax->entry_padding = 1318 0x1ff & (-(int64_t)pax->entry_bytes_remaining); 1319 1320 r = __archive_write_output(a, pax->pax_header.s, 1321 archive_strlen(&(pax->pax_header))); 1322 if (r != ARCHIVE_OK) { 1323 /* If a write fails, we're pretty much toast. */ 1324 return (ARCHIVE_FATAL); 1325 } 1326 /* Pad out the end of the entry. */ 1327 r = __archive_write_nulls(a, pax->entry_padding); 1328 if (r != ARCHIVE_OK) { 1329 /* If a write fails, we're pretty much toast. */ 1330 return (ARCHIVE_FATAL); 1331 } 1332 pax->entry_bytes_remaining = pax->entry_padding = 0; 1333 } 1334 1335 /* Write the header for main entry. */ 1336 r = __archive_write_output(a, ustarbuff, 512); 1337 if (r != ARCHIVE_OK) 1338 return (r); 1339 1340 /* 1341 * Inform the client of the on-disk size we're using, so 1342 * they can avoid unnecessarily writing a body for something 1343 * that we're just going to ignore. 1344 */ 1345 archive_entry_set_size(entry_original, real_size); 1346 if (pax->sparse_list == NULL && real_size > 0) { 1347 /* This is not a sparse file but we handle its data as 1348 * a sparse block. */ 1349 sparse_list_add(pax, 0, real_size); 1350 sparse_total = real_size; 1351 } 1352 pax->entry_padding = 0x1ff & (-(int64_t)sparse_total); 1353 archive_entry_free(entry_main); 1354 archive_string_free(&entry_name); 1355 1356 return (ret); 1357} 1358 1359/* 1360 * We need a valid name for the regular 'ustar' entry. This routine 1361 * tries to hack something more-or-less reasonable. 1362 * 1363 * The approach here tries to preserve leading dir names. We do so by 1364 * working with four sections: 1365 * 1) "prefix" directory names, 1366 * 2) "suffix" directory names, 1367 * 3) inserted dir name (optional), 1368 * 4) filename. 1369 * 1370 * These sections must satisfy the following requirements: 1371 * * Parts 1 & 2 together form an initial portion of the dir name. 1372 * * Part 3 is specified by the caller. (It should not contain a leading 1373 * or trailing '/'.) 1374 * * Part 4 forms an initial portion of the base filename. 1375 * * The filename must be <= 99 chars to fit the ustar 'name' field. 1376 * * Parts 2, 3, 4 together must be <= 99 chars to fit the ustar 'name' fld. 1377 * * Part 1 must be <= 155 chars to fit the ustar 'prefix' field. 1378 * * If the original name ends in a '/', the new name must also end in a '/' 1379 * * Trailing '/.' sequences may be stripped. 1380 * 1381 * Note: Recall that the ustar format does not store the '/' separating 1382 * parts 1 & 2, but does store the '/' separating parts 2 & 3. 1383 */ 1384static char * 1385build_ustar_entry_name(char *dest, const char *src, size_t src_length, 1386 const char *insert) 1387{ 1388 const char *prefix, *prefix_end; 1389 const char *suffix, *suffix_end; 1390 const char *filename, *filename_end; 1391 char *p; 1392 int need_slash = 0; /* Was there a trailing slash? */ 1393 size_t suffix_length = 99; 1394 size_t insert_length; 1395 1396 /* Length of additional dir element to be added. */ 1397 if (insert == NULL) 1398 insert_length = 0; 1399 else 1400 /* +2 here allows for '/' before and after the insert. */ 1401 insert_length = strlen(insert) + 2; 1402 1403 /* Step 0: Quick bailout in a common case. */ 1404 if (src_length < 100 && insert == NULL) { 1405 strncpy(dest, src, src_length); 1406 dest[src_length] = '\0'; 1407 return (dest); 1408 } 1409 1410 /* Step 1: Locate filename and enforce the length restriction. */ 1411 filename_end = src + src_length; 1412 /* Remove trailing '/' chars and '/.' pairs. */ 1413 for (;;) { 1414 if (filename_end > src && filename_end[-1] == '/') { 1415 filename_end --; 1416 need_slash = 1; /* Remember to restore trailing '/'. */ 1417 continue; 1418 } 1419 if (filename_end > src + 1 && filename_end[-1] == '.' 1420 && filename_end[-2] == '/') { 1421 filename_end -= 2; 1422 need_slash = 1; /* "foo/." will become "foo/" */ 1423 continue; 1424 } 1425 break; 1426 } 1427 if (need_slash) 1428 suffix_length--; 1429 /* Find start of filename. */ 1430 filename = filename_end - 1; 1431 while ((filename > src) && (*filename != '/')) 1432 filename --; 1433 if ((*filename == '/') && (filename < filename_end - 1)) 1434 filename ++; 1435 /* Adjust filename_end so that filename + insert fits in 99 chars. */ 1436 suffix_length -= insert_length; 1437 if (filename_end > filename + suffix_length) 1438 filename_end = filename + suffix_length; 1439 /* Calculate max size for "suffix" section (#3 above). */ 1440 suffix_length -= filename_end - filename; 1441 1442 /* Step 2: Locate the "prefix" section of the dirname, including 1443 * trailing '/'. */ 1444 prefix = src; 1445 prefix_end = prefix + 155; 1446 if (prefix_end > filename) 1447 prefix_end = filename; 1448 while (prefix_end > prefix && *prefix_end != '/') 1449 prefix_end--; 1450 if ((prefix_end < filename) && (*prefix_end == '/')) 1451 prefix_end++; 1452 1453 /* Step 3: Locate the "suffix" section of the dirname, 1454 * including trailing '/'. */ 1455 suffix = prefix_end; 1456 suffix_end = suffix + suffix_length; /* Enforce limit. */ 1457 if (suffix_end > filename) 1458 suffix_end = filename; 1459 if (suffix_end < suffix) 1460 suffix_end = suffix; 1461 while (suffix_end > suffix && *suffix_end != '/') 1462 suffix_end--; 1463 if ((suffix_end < filename) && (*suffix_end == '/')) 1464 suffix_end++; 1465 1466 /* Step 4: Build the new name. */ 1467 /* The OpenBSD strlcpy function is safer, but less portable. */ 1468 /* Rather than maintain two versions, just use the strncpy version. */ 1469 p = dest; 1470 if (prefix_end > prefix) { 1471 strncpy(p, prefix, prefix_end - prefix); 1472 p += prefix_end - prefix; 1473 } 1474 if (suffix_end > suffix) { 1475 strncpy(p, suffix, suffix_end - suffix); 1476 p += suffix_end - suffix; 1477 } 1478 if (insert != NULL) { 1479 /* Note: assume insert does not have leading or trailing '/' */ 1480 strcpy(p, insert); 1481 p += strlen(insert); 1482 *p++ = '/'; 1483 } 1484 strncpy(p, filename, filename_end - filename); 1485 p += filename_end - filename; 1486 if (need_slash) 1487 *p++ = '/'; 1488 *p = '\0'; 1489 1490 return (dest); 1491} 1492 1493/* 1494 * The ustar header for the pax extended attributes must have a 1495 * reasonable name: SUSv3 requires 'dirname'/PaxHeader.'pid'/'filename' 1496 * where 'pid' is the PID of the archiving process. Unfortunately, 1497 * that makes testing a pain since the output varies for each run, 1498 * so I'm sticking with the simpler 'dirname'/PaxHeader/'filename' 1499 * for now. (Someday, I'll make this settable. Then I can use the 1500 * SUS recommendation as default and test harnesses can override it 1501 * to get predictable results.) 1502 * 1503 * Joerg Schilling has argued that this is unnecessary because, in 1504 * practice, if the pax extended attributes get extracted as regular 1505 * files, no one is going to bother reading those attributes to 1506 * manually restore them. Based on this, 'star' uses 1507 * /tmp/PaxHeader/'basename' as the ustar header name. This is a 1508 * tempting argument, in part because it's simpler than the SUSv3 1509 * recommendation, but I'm not entirely convinced. I'm also 1510 * uncomfortable with the fact that "/tmp" is a Unix-ism. 1511 * 1512 * The following routine leverages build_ustar_entry_name() above and 1513 * so is simpler than you might think. It just needs to provide the 1514 * additional path element and handle a few pathological cases). 1515 */ 1516static char * 1517build_pax_attribute_name(char *dest, const char *src) 1518{ 1519 char buff[64]; 1520 const char *p; 1521 1522 /* Handle the null filename case. */ 1523 if (src == NULL || *src == '\0') { 1524 strcpy(dest, "PaxHeader/blank"); 1525 return (dest); 1526 } 1527 1528 /* Prune final '/' and other unwanted final elements. */ 1529 p = src + strlen(src); 1530 for (;;) { 1531 /* Ends in "/", remove the '/' */ 1532 if (p > src && p[-1] == '/') { 1533 --p; 1534 continue; 1535 } 1536 /* Ends in "/.", remove the '.' */ 1537 if (p > src + 1 && p[-1] == '.' 1538 && p[-2] == '/') { 1539 --p; 1540 continue; 1541 } 1542 break; 1543 } 1544 1545 /* Pathological case: After above, there was nothing left. 1546 * This includes "/." "/./." "/.//./." etc. */ 1547 if (p == src) { 1548 strcpy(dest, "/PaxHeader/rootdir"); 1549 return (dest); 1550 } 1551 1552 /* Convert unadorned "." into a suitable filename. */ 1553 if (*src == '.' && p == src + 1) { 1554 strcpy(dest, "PaxHeader/currentdir"); 1555 return (dest); 1556 } 1557 1558 /* 1559 * TODO: Push this string into the 'pax' structure to avoid 1560 * recomputing it every time. That will also open the door 1561 * to having clients override it. 1562 */ 1563#if HAVE_GETPID && 0 /* Disable this for now; see above comment. */ 1564 sprintf(buff, "PaxHeader.%d", getpid()); 1565#else 1566 /* If the platform can't fetch the pid, don't include it. */ 1567 strcpy(buff, "PaxHeader"); 1568#endif 1569 /* General case: build a ustar-compatible name adding 1570 * "/PaxHeader/". */ 1571 build_ustar_entry_name(dest, src, p - src, buff); 1572 1573 return (dest); 1574} 1575 1576/* 1577 * GNU PAX Format 1.0 requires the special name, which pattern is: 1578 * <dir>/GNUSparseFile.<pid>/<original file name> 1579 * 1580 * This function is used for only Sparse file, a file type of which 1581 * is regular file. 1582 */ 1583static char * 1584build_gnu_sparse_name(char *dest, const char *src) 1585{ 1586 char buff[64]; 1587 const char *p; 1588 1589 /* Handle the null filename case. */ 1590 if (src == NULL || *src == '\0') { 1591 strcpy(dest, "GNUSparseFile/blank"); 1592 return (dest); 1593 } 1594 1595 /* Prune final '/' and other unwanted final elements. */ 1596 p = src + strlen(src); 1597 for (;;) { 1598 /* Ends in "/", remove the '/' */ 1599 if (p > src && p[-1] == '/') { 1600 --p; 1601 continue; 1602 } 1603 /* Ends in "/.", remove the '.' */ 1604 if (p > src + 1 && p[-1] == '.' 1605 && p[-2] == '/') { 1606 --p; 1607 continue; 1608 } 1609 break; 1610 } 1611 1612#if HAVE_GETPID && 0 /* Disable this as pax attribute name. */ 1613 sprintf(buff, "GNUSparseFile.%d", getpid()); 1614#else 1615 /* If the platform can't fetch the pid, don't include it. */ 1616 strcpy(buff, "GNUSparseFile"); 1617#endif 1618 /* General case: build a ustar-compatible name adding 1619 * "/GNUSparseFile/". */ 1620 build_ustar_entry_name(dest, src, p - src, buff); 1621 1622 return (dest); 1623} 1624 1625/* Write two null blocks for the end of archive */ 1626static int 1627archive_write_pax_close(struct archive_write *a) 1628{ 1629 return (__archive_write_nulls(a, 512 * 2)); 1630} 1631 1632static int 1633archive_write_pax_free(struct archive_write *a) 1634{ 1635 struct pax *pax; 1636 1637 pax = (struct pax *)a->format_data; 1638 if (pax == NULL) 1639 return (ARCHIVE_OK); 1640 1641 archive_string_free(&pax->pax_header); 1642 archive_string_free(&pax->sparse_map); 1643 archive_string_free(&pax->l_url_encoded_name); 1644 sparse_list_clear(pax); 1645 free(pax); 1646 a->format_data = NULL; 1647 return (ARCHIVE_OK); 1648} 1649 1650static int 1651archive_write_pax_finish_entry(struct archive_write *a) 1652{ 1653 struct pax *pax; 1654 uint64_t remaining; 1655 int ret; 1656 1657 pax = (struct pax *)a->format_data; 1658 remaining = pax->entry_bytes_remaining; 1659 if (remaining == 0) { 1660 while (pax->sparse_list) { 1661 struct sparse_block *sb; 1662 if (!pax->sparse_list->is_hole) 1663 remaining += pax->sparse_list->remaining; 1664 sb = pax->sparse_list->next; 1665 free(pax->sparse_list); 1666 pax->sparse_list = sb; 1667 } 1668 } 1669 ret = __archive_write_nulls(a, remaining + pax->entry_padding); 1670 pax->entry_bytes_remaining = pax->entry_padding = 0; 1671 return (ret); 1672} 1673 1674static ssize_t 1675archive_write_pax_data(struct archive_write *a, const void *buff, size_t s) 1676{ 1677 struct pax *pax; 1678 size_t ws; 1679 size_t total; 1680 int ret; 1681 1682 pax = (struct pax *)a->format_data; 1683 1684 /* 1685 * According to GNU PAX format 1.0, write a sparse map 1686 * before the body. 1687 */ 1688 if (archive_strlen(&(pax->sparse_map))) { 1689 ret = __archive_write_output(a, pax->sparse_map.s, 1690 archive_strlen(&(pax->sparse_map))); 1691 if (ret != ARCHIVE_OK) 1692 return (ret); 1693 ret = __archive_write_nulls(a, pax->sparse_map_padding); 1694 if (ret != ARCHIVE_OK) 1695 return (ret); 1696 archive_string_empty(&(pax->sparse_map)); 1697 } 1698 1699 total = 0; 1700 while (total < s) { 1701 const unsigned char *p; 1702 1703 while (pax->sparse_list != NULL && 1704 pax->sparse_list->remaining == 0) { 1705 struct sparse_block *sb = pax->sparse_list->next; 1706 free(pax->sparse_list); 1707 pax->sparse_list = sb; 1708 } 1709 1710 if (pax->sparse_list == NULL) 1711 return (total); 1712 1713 p = ((const unsigned char *)buff) + total; 1714 ws = s - total; 1715 if (ws > pax->sparse_list->remaining) 1716 ws = pax->sparse_list->remaining; 1717 1718 if (pax->sparse_list->is_hole) { 1719 /* Current block is hole thus we do not write 1720 * the body. */ 1721 pax->sparse_list->remaining -= ws; 1722 total += ws; 1723 continue; 1724 } 1725 1726 ret = __archive_write_output(a, p, ws); 1727 pax->sparse_list->remaining -= ws; 1728 total += ws; 1729 if (ret != ARCHIVE_OK) 1730 return (ret); 1731 } 1732 return (total); 1733} 1734 1735static int 1736has_non_ASCII(const char *_p) 1737{ 1738 const unsigned char *p = (const unsigned char *)_p; 1739 1740 if (p == NULL) 1741 return (1); 1742 while (*p != '\0' && *p < 128) 1743 p++; 1744 return (*p != '\0'); 1745} 1746 1747/* 1748 * Used by extended attribute support; encodes the name 1749 * so that there will be no '=' characters in the result. 1750 */ 1751static char * 1752url_encode(const char *in) 1753{ 1754 const char *s; 1755 char *d; 1756 int out_len = 0; 1757 char *out; 1758 1759 for (s = in; *s != '\0'; s++) { 1760 if (*s < 33 || *s > 126 || *s == '%' || *s == '=') 1761 out_len += 3; 1762 else 1763 out_len++; 1764 } 1765 1766 out = (char *)malloc(out_len + 1); 1767 if (out == NULL) 1768 return (NULL); 1769 1770 for (s = in, d = out; *s != '\0'; s++) { 1771 /* encode any non-printable ASCII character or '%' or '=' */ 1772 if (*s < 33 || *s > 126 || *s == '%' || *s == '=') { 1773 /* URL encoding is '%' followed by two hex digits */ 1774 *d++ = '%'; 1775 *d++ = "0123456789ABCDEF"[0x0f & (*s >> 4)]; 1776 *d++ = "0123456789ABCDEF"[0x0f & *s]; 1777 } else { 1778 *d++ = *s; 1779 } 1780 } 1781 *d = '\0'; 1782 return (out); 1783} 1784 1785/* 1786 * Encode a sequence of bytes into a C string using base-64 encoding. 1787 * 1788 * Returns a null-terminated C string allocated with malloc(); caller 1789 * is responsible for freeing the result. 1790 */ 1791static char * 1792base64_encode(const char *s, size_t len) 1793{ 1794 static const char digits[64] = 1795 { 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O', 1796 'P','Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d', 1797 'e','f','g','h','i','j','k','l','m','n','o','p','q','r','s', 1798 't','u','v','w','x','y','z','0','1','2','3','4','5','6','7', 1799 '8','9','+','/' }; 1800 int v; 1801 char *d, *out; 1802 1803 /* 3 bytes becomes 4 chars, but round up and allow for trailing NUL */ 1804 out = (char *)malloc((len * 4 + 2) / 3 + 1); 1805 if (out == NULL) 1806 return (NULL); 1807 d = out; 1808 1809 /* Convert each group of 3 bytes into 4 characters. */ 1810 while (len >= 3) { 1811 v = (((int)s[0] << 16) & 0xff0000) 1812 | (((int)s[1] << 8) & 0xff00) 1813 | (((int)s[2]) & 0x00ff); 1814 s += 3; 1815 len -= 3; 1816 *d++ = digits[(v >> 18) & 0x3f]; 1817 *d++ = digits[(v >> 12) & 0x3f]; 1818 *d++ = digits[(v >> 6) & 0x3f]; 1819 *d++ = digits[(v) & 0x3f]; 1820 } 1821 /* Handle final group of 1 byte (2 chars) or 2 bytes (3 chars). */ 1822 switch (len) { 1823 case 0: break; 1824 case 1: 1825 v = (((int)s[0] << 16) & 0xff0000); 1826 *d++ = digits[(v >> 18) & 0x3f]; 1827 *d++ = digits[(v >> 12) & 0x3f]; 1828 break; 1829 case 2: 1830 v = (((int)s[0] << 16) & 0xff0000) 1831 | (((int)s[1] << 8) & 0xff00); 1832 *d++ = digits[(v >> 18) & 0x3f]; 1833 *d++ = digits[(v >> 12) & 0x3f]; 1834 *d++ = digits[(v >> 6) & 0x3f]; 1835 break; 1836 } 1837 /* Add trailing NUL character so output is a valid C string. */ 1838 *d = '\0'; 1839 return (out); 1840} 1841 1842static void 1843sparse_list_clear(struct pax *pax) 1844{ 1845 while (pax->sparse_list != NULL) { 1846 struct sparse_block *sb = pax->sparse_list; 1847 pax->sparse_list = sb->next; 1848 free(sb); 1849 } 1850 pax->sparse_tail = NULL; 1851} 1852 1853static int 1854_sparse_list_add_block(struct pax *pax, int64_t offset, int64_t length, 1855 int is_hole) 1856{ 1857 struct sparse_block *sb; 1858 1859 sb = (struct sparse_block *)malloc(sizeof(*sb)); 1860 if (sb == NULL) 1861 return (ARCHIVE_FATAL); 1862 sb->next = NULL; 1863 sb->is_hole = is_hole; 1864 sb->offset = offset; 1865 sb->remaining = length; 1866 if (pax->sparse_list == NULL) 1867 pax->sparse_list = pax->sparse_tail = sb; 1868 else { 1869 pax->sparse_tail->next = sb; 1870 pax->sparse_tail = sb; 1871 } 1872 return (ARCHIVE_OK); 1873} 1874 1875static int 1876sparse_list_add(struct pax *pax, int64_t offset, int64_t length) 1877{ 1878 int64_t last_offset; 1879 int r; 1880 1881 if (pax->sparse_tail == NULL) 1882 last_offset = 0; 1883 else { 1884 last_offset = pax->sparse_tail->offset + 1885 pax->sparse_tail->remaining; 1886 } 1887 if (last_offset < offset) { 1888 /* Add a hole block. */ 1889 r = _sparse_list_add_block(pax, last_offset, 1890 offset - last_offset, 1); 1891 if (r != ARCHIVE_OK) 1892 return (r); 1893 } 1894 /* Add data block. */ 1895 return (_sparse_list_add_block(pax, offset, length, 0)); 1896} 1897 1898