1 /*- 2 * Copyright (c) 2003-2007 Tim Kientzle 3 * Copyright (c) 2010-2011 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 } else 191 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 192 "pax: unknown keyword ``%s''", key); 193 194 return (ret); 195 } 196 197 /* 198 * Note: This code assumes that 'nanos' has the same sign as 'sec', 199 * which implies that sec=-1, nanos=200000000 represents -1.2 seconds 200 * and not -0.8 seconds. This is a pretty pedantic point, as we're 201 * unlikely to encounter many real files created before Jan 1, 1970, 202 * much less ones with timestamps recorded to sub-second resolution. 203 */ 204 static void 205 add_pax_attr_time(struct archive_string *as, const char *key, 206 int64_t sec, unsigned long nanos) 207 { 208 int digit, i; 209 char *t; 210 /* 211 * Note that each byte contributes fewer than 3 base-10 212 * digits, so this will always be big enough. 213 */ 214 char tmp[1 + 3*sizeof(sec) + 1 + 3*sizeof(nanos)]; 215 216 tmp[sizeof(tmp) - 1] = 0; 217 t = tmp + sizeof(tmp) - 1; 218 219 /* Skip trailing zeros in the fractional part. */ 220 for (digit = 0, i = 10; i > 0 && digit == 0; i--) { 221 digit = nanos % 10; 222 nanos /= 10; 223 } 224 225 /* Only format the fraction if it's non-zero. */ 226 if (i > 0) { 227 while (i > 0) { 228 *--t = "0123456789"[digit]; 229 digit = nanos % 10; 230 nanos /= 10; 231 i--; 232 } 233 *--t = '.'; 234 } 235 t = format_int(t, sec); 236 237 add_pax_attr(as, key, t); 238 } 239 240 static char * 241 format_int(char *t, int64_t i) 242 { 243 uint64_t ui; 244 245 if (i < 0) 246 ui = (i == INT64_MIN) ? (uint64_t)(INT64_MAX) + 1 : (uint64_t)(-i); 247 else 248 ui = i; 249 250 do { 251 *--t = "0123456789"[ui % 10]; 252 } while (ui /= 10); 253 if (i < 0) 254 *--t = '-'; 255 return (t); 256 } 257 258 static void 259 add_pax_attr_int(struct archive_string *as, const char *key, int64_t value) 260 { 261 char tmp[1 + 3 * sizeof(value)]; 262 263 tmp[sizeof(tmp) - 1] = 0; 264 add_pax_attr(as, key, format_int(tmp + sizeof(tmp) - 1, value)); 265 } 266 267 /* 268 * Add a key/value attribute to the pax header. This function handles 269 * the length field and various other syntactic requirements. 270 */ 271 static void 272 add_pax_attr(struct archive_string *as, const char *key, const char *value) 273 { 274 int digits, i, len, next_ten; 275 char tmp[1 + 3 * sizeof(int)]; /* < 3 base-10 digits per byte */ 276 277 /*- 278 * PAX attributes have the following layout: 279 * <len> <space> <key> <=> <value> <nl> 280 */ 281 len = 1 + (int)strlen(key) + 1 + (int)strlen(value) + 1; 282 283 /* 284 * The <len> field includes the length of the <len> field, so 285 * computing the correct length is tricky. I start by 286 * counting the number of base-10 digits in 'len' and 287 * computing the next higher power of 10. 288 */ 289 next_ten = 1; 290 digits = 0; 291 i = len; 292 while (i > 0) { 293 i = i / 10; 294 digits++; 295 next_ten = next_ten * 10; 296 } 297 /* 298 * For example, if string without the length field is 99 299 * chars, then adding the 2 digit length "99" will force the 300 * total length past 100, requiring an extra digit. The next 301 * statement adjusts for this effect. 302 */ 303 if (len + digits >= next_ten) 304 digits++; 305 306 /* Now, we have the right length so we can build the line. */ 307 tmp[sizeof(tmp) - 1] = 0; /* Null-terminate the work area. */ 308 archive_strcat(as, format_int(tmp + sizeof(tmp) - 1, len + digits)); 309 archive_strappend_char(as, ' '); 310 archive_strcat(as, key); 311 archive_strappend_char(as, '='); 312 archive_strcat(as, value); 313 archive_strappend_char(as, '\n'); 314 } 315 316 static int 317 archive_write_pax_header_xattrs(struct archive_write *a, 318 struct pax *pax, struct archive_entry *entry) 319 { 320 struct archive_string s; 321 int i = archive_entry_xattr_reset(entry); 322 323 while (i--) { 324 const char *name; 325 const void *value; 326 char *encoded_value; 327 char *url_encoded_name = NULL, *encoded_name = NULL; 328 size_t size; 329 int r; 330 331 archive_entry_xattr_next(entry, &name, &value, &size); 332 url_encoded_name = url_encode(name); 333 if (url_encoded_name != NULL) { 334 /* Convert narrow-character to UTF-8. */ 335 r = archive_strcpy_in_locale( 336 &(pax->l_url_encoded_name), 337 url_encoded_name, pax->sconv_utf8); 338 free(url_encoded_name); /* Done with this. */ 339 if (r == 0) 340 encoded_name = pax->l_url_encoded_name.s; 341 else if (errno == ENOMEM) { 342 archive_set_error(&a->archive, ENOMEM, 343 "Can't allocate memory for Linkname"); 344 return (ARCHIVE_FATAL); 345 } 346 } 347 348 encoded_value = base64_encode((const char *)value, size); 349 350 if (encoded_name != NULL && encoded_value != NULL) { 351 archive_string_init(&s); 352 archive_strcpy(&s, "LIBARCHIVE.xattr."); 353 archive_strcat(&s, encoded_name); 354 add_pax_attr(&(pax->pax_header), s.s, encoded_value); 355 archive_string_free(&s); 356 } 357 free(encoded_value); 358 } 359 return (ARCHIVE_OK); 360 } 361 362 static int 363 get_entry_hardlink(struct archive_write *a, struct archive_entry *entry, 364 const char **name, size_t *length, struct archive_string_conv *sc) 365 { 366 int r; 367 368 r = archive_entry_hardlink_l(entry, name, length, sc); 369 if (r != 0) { 370 if (errno == ENOMEM) { 371 archive_set_error(&a->archive, ENOMEM, 372 "Can't allocate memory for Linkname"); 373 return (ARCHIVE_FATAL); 374 } 375 return (ARCHIVE_WARN); 376 } 377 return (ARCHIVE_OK); 378 } 379 380 static int 381 get_entry_pathname(struct archive_write *a, struct archive_entry *entry, 382 const char **name, size_t *length, struct archive_string_conv *sc) 383 { 384 int r; 385 386 r = archive_entry_pathname_l(entry, name, length, sc); 387 if (r != 0) { 388 if (errno == ENOMEM) { 389 archive_set_error(&a->archive, ENOMEM, 390 "Can't allocate memory for Pathname"); 391 return (ARCHIVE_FATAL); 392 } 393 return (ARCHIVE_WARN); 394 } 395 return (ARCHIVE_OK); 396 } 397 398 static int 399 get_entry_uname(struct archive_write *a, struct archive_entry *entry, 400 const char **name, size_t *length, struct archive_string_conv *sc) 401 { 402 int r; 403 404 r = archive_entry_uname_l(entry, name, length, sc); 405 if (r != 0) { 406 if (errno == ENOMEM) { 407 archive_set_error(&a->archive, ENOMEM, 408 "Can't allocate memory for Uname"); 409 return (ARCHIVE_FATAL); 410 } 411 return (ARCHIVE_WARN); 412 } 413 return (ARCHIVE_OK); 414 } 415 416 static int 417 get_entry_gname(struct archive_write *a, struct archive_entry *entry, 418 const char **name, size_t *length, struct archive_string_conv *sc) 419 { 420 int r; 421 422 r = archive_entry_gname_l(entry, name, length, sc); 423 if (r != 0) { 424 if (errno == ENOMEM) { 425 archive_set_error(&a->archive, ENOMEM, 426 "Can't allocate memory for Gname"); 427 return (ARCHIVE_FATAL); 428 } 429 return (ARCHIVE_WARN); 430 } 431 return (ARCHIVE_OK); 432 } 433 434 static int 435 get_entry_symlink(struct archive_write *a, struct archive_entry *entry, 436 const char **name, size_t *length, struct archive_string_conv *sc) 437 { 438 int r; 439 440 r = archive_entry_symlink_l(entry, name, length, sc); 441 if (r != 0) { 442 if (errno == ENOMEM) { 443 archive_set_error(&a->archive, ENOMEM, 444 "Can't allocate memory for Linkname"); 445 return (ARCHIVE_FATAL); 446 } 447 return (ARCHIVE_WARN); 448 } 449 return (ARCHIVE_OK); 450 } 451 452 /* 453 * TODO: Consider adding 'comment' and 'charset' fields to 454 * archive_entry so that clients can specify them. Also, consider 455 * adding generic key/value tags so clients can add arbitrary 456 * key/value data. 457 * 458 * TODO: Break up this 700-line function!!!! Yowza! 459 */ 460 static int 461 archive_write_pax_header(struct archive_write *a, 462 struct archive_entry *entry_original) 463 { 464 struct archive_entry *entry_main; 465 const char *p; 466 char *t; 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 * Ensure a trailing '/'. Modify the original 546 * entry so the client sees the change. 547 */ 548 p = archive_entry_pathname(entry_original); 549 if (p[strlen(p) - 1] != '/') { 550 t = (char *)malloc(strlen(p) + 2); 551 if (t == NULL) { 552 archive_set_error(&a->archive, ENOMEM, 553 "Can't allocate pax data"); 554 return(ARCHIVE_FATAL); 555 } 556 strcpy(t, p); 557 strcat(t, "/"); 558 archive_entry_copy_pathname(entry_original, t); 559 free(t); 560 } 561 break; 562 case AE_IFSOCK: 563 archive_set_error(&a->archive, 564 ARCHIVE_ERRNO_FILE_FORMAT, 565 "tar format cannot archive socket"); 566 return (ARCHIVE_FAILED); 567 default: 568 archive_set_error(&a->archive, 569 ARCHIVE_ERRNO_FILE_FORMAT, 570 "tar format cannot archive this (type=0%lo)", 571 (unsigned long) 572 archive_entry_filetype(entry_original)); 573 return (ARCHIVE_FAILED); 574 } 575 } 576 577 /* 578 * If Mac OS metadata blob is here, recurse to write that 579 * as a separate entry. This is really a pretty poor design: 580 * In particular, it doubles the overhead for long filenames. 581 * TODO: Help Apple folks design something better and figure 582 * out how to transition from this legacy format. 583 * 584 * Note that this code is present on every platform; clients 585 * on non-Mac are unlikely to ever provide this data, but 586 * applications that copy entries from one archive to another 587 * should not lose data just because the local filesystem 588 * can't store it. 589 */ 590 mac_metadata = 591 archive_entry_mac_metadata(entry_original, &mac_metadata_size); 592 if (mac_metadata != NULL) { 593 const char *oname; 594 char *name, *bname; 595 size_t name_length; 596 struct archive_entry *extra = archive_entry_new2(&a->archive); 597 598 oname = archive_entry_pathname(entry_original); 599 name_length = strlen(oname); 600 name = malloc(name_length + 3); 601 if (name == NULL) { 602 /* XXX error message */ 603 return (ARCHIVE_FAILED); 604 } 605 strcpy(name, oname); 606 /* Find last '/'; strip trailing '/' characters */ 607 bname = strrchr(name, '/'); 608 while (bname != NULL && bname[1] == '\0') { 609 *bname = '\0'; 610 bname = strrchr(name, '/'); 611 } 612 if (bname == NULL) { 613 memmove(name + 2, name, name_length + 1); 614 memmove(name, "._", 2); 615 } else { 616 bname += 1; 617 memmove(bname + 2, bname, strlen(bname) + 1); 618 memmove(bname, "._", 2); 619 } 620 archive_entry_copy_pathname(extra, name); 621 free(name); 622 623 archive_entry_set_size(extra, mac_metadata_size); 624 archive_entry_set_filetype(extra, AE_IFREG); 625 archive_entry_set_perm(extra, 626 archive_entry_perm(entry_original)); 627 archive_entry_set_mtime(extra, 628 archive_entry_mtime(entry_original), 629 archive_entry_mtime_nsec(entry_original)); 630 archive_entry_set_gid(extra, 631 archive_entry_gid(entry_original)); 632 archive_entry_set_gname(extra, 633 archive_entry_gname(entry_original)); 634 archive_entry_set_uid(extra, 635 archive_entry_uid(entry_original)); 636 archive_entry_set_uname(extra, 637 archive_entry_uname(entry_original)); 638 639 /* Recurse to write the special copyfile entry. */ 640 r = archive_write_pax_header(a, extra); 641 if (r < ARCHIVE_WARN) 642 return (r); 643 if (r < ret) 644 ret = r; 645 r = archive_write_pax_data(a, mac_metadata, mac_metadata_size); 646 if (r < ARCHIVE_WARN) 647 return (r); 648 if (r < ret) 649 ret = r; 650 r = archive_write_pax_finish_entry(a); 651 if (r < ARCHIVE_WARN) 652 return (r); 653 if (r < ret) 654 ret = r; 655 } 656 657 /* Copy entry so we can modify it as needed. */ 658 entry_main = archive_entry_clone(entry_original); 659 archive_string_empty(&(pax->pax_header)); /* Blank our work area. */ 660 archive_string_empty(&(pax->sparse_map)); 661 sparse_total = 0; 662 sparse_list_clear(pax); 663 664 if (hardlink == NULL && 665 archive_entry_filetype(entry_main) == AE_IFREG) 666 sparse_count = archive_entry_sparse_reset(entry_main); 667 else 668 sparse_count = 0; 669 if (sparse_count) { 670 int64_t offset, length, last_offset = 0; 671 /* Get the last entry of sparse block. */ 672 while (archive_entry_sparse_next( 673 entry_main, &offset, &length) == ARCHIVE_OK) 674 last_offset = offset + length; 675 676 /* If the last sparse block does not reach the end of file, 677 * We have to add a empty sparse block as the last entry to 678 * manage storing file data. */ 679 if (last_offset < archive_entry_size(entry_main)) 680 archive_entry_sparse_add_entry(entry_main, 681 archive_entry_size(entry_main), 0); 682 sparse_count = archive_entry_sparse_reset(entry_main); 683 } 684 685 /* 686 * First, check the name fields and see if any of them 687 * require binary coding. If any of them does, then all of 688 * them do. 689 */ 690 r = get_entry_pathname(a, entry_main, &path, &path_length, sconv); 691 if (r == ARCHIVE_FATAL) 692 return (r); 693 else if (r != ARCHIVE_OK) { 694 r = get_entry_pathname(a, entry_main, &path, 695 &path_length, NULL); 696 if (r == ARCHIVE_FATAL) 697 return (r); 698 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 699 "Can't translate pathname '%s' to %s", path, 700 archive_string_conversion_charset_name(sconv)); 701 ret = ARCHIVE_WARN; 702 sconv = NULL;/* The header charset switches to binary mode. */ 703 } 704 r = get_entry_uname(a, entry_main, &uname, &uname_length, sconv); 705 if (r == ARCHIVE_FATAL) 706 return (r); 707 else if (r != ARCHIVE_OK) { 708 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL); 709 if (r == ARCHIVE_FATAL) 710 return (r); 711 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 712 "Can't translate uname '%s' to %s", uname, 713 archive_string_conversion_charset_name(sconv)); 714 ret = ARCHIVE_WARN; 715 sconv = NULL;/* The header charset switches to binary mode. */ 716 } 717 r = get_entry_gname(a, entry_main, &gname, &gname_length, sconv); 718 if (r == ARCHIVE_FATAL) 719 return (r); 720 else if (r != ARCHIVE_OK) { 721 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL); 722 if (r == ARCHIVE_FATAL) 723 return (r); 724 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, 725 "Can't translate gname '%s' to %s", gname, 726 archive_string_conversion_charset_name(sconv)); 727 ret = ARCHIVE_WARN; 728 sconv = NULL;/* The header charset switches to binary mode. */ 729 } 730 linkpath = hardlink; 731 linkpath_length = hardlink_length; 732 if (linkpath == NULL) { 733 r = get_entry_symlink(a, entry_main, &linkpath, 734 &linkpath_length, sconv); 735 if (r == ARCHIVE_FATAL) 736 return (r); 737 else if (r != ARCHIVE_OK) { 738 r = get_entry_symlink(a, entry_main, &linkpath, 739 &linkpath_length, NULL); 740 if (r == ARCHIVE_FATAL) 741 return (r); 742 archive_set_error(&a->archive, 743 ARCHIVE_ERRNO_FILE_FORMAT, 744 "Can't translate linkname '%s' to %s", linkpath, 745 archive_string_conversion_charset_name(sconv)); 746 ret = ARCHIVE_WARN; 747 sconv = NULL; 748 } 749 } 750 751 /* If any string conversions failed, get all attributes 752 * in binary-mode. */ 753 if (sconv == NULL && !pax->opt_binary) { 754 if (hardlink != NULL) { 755 r = get_entry_hardlink(a, entry_main, &hardlink, 756 &hardlink_length, NULL); 757 if (r == ARCHIVE_FATAL) 758 return (r); 759 linkpath = hardlink; 760 linkpath_length = hardlink_length; 761 } 762 r = get_entry_pathname(a, entry_main, &path, 763 &path_length, NULL); 764 if (r == ARCHIVE_FATAL) 765 return (r); 766 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL); 767 if (r == ARCHIVE_FATAL) 768 return (r); 769 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL); 770 if (r == ARCHIVE_FATAL) 771 return (r); 772 } 773 774 /* Store the header encoding first, to be nice to readers. */ 775 if (sconv == NULL) 776 add_pax_attr(&(pax->pax_header), "hdrcharset", "BINARY"); 777 778 779 /* 780 * If name is too long, or has non-ASCII characters, add 781 * 'path' to pax extended attrs. (Note that an unconvertible 782 * name must have non-ASCII characters.) 783 */ 784 if (has_non_ASCII(path)) { 785 /* We have non-ASCII characters. */ 786 add_pax_attr(&(pax->pax_header), "path", path); 787 archive_entry_set_pathname(entry_main, 788 build_ustar_entry_name(ustar_entry_name, 789 path, path_length, NULL)); 790 need_extension = 1; 791 } else { 792 /* We have an all-ASCII path; we'd like to just store 793 * it in the ustar header if it will fit. Yes, this 794 * duplicates some of the logic in 795 * archive_write_set_format_ustar.c 796 */ 797 if (path_length <= 100) { 798 /* Fits in the old 100-char tar name field. */ 799 } else { 800 /* Find largest suffix that will fit. */ 801 /* Note: strlen() > 100, so strlen() - 100 - 1 >= 0 */ 802 suffix = strchr(path + path_length - 100 - 1, '/'); 803 /* Don't attempt an empty prefix. */ 804 if (suffix == path) 805 suffix = strchr(suffix + 1, '/'); 806 /* We can put it in the ustar header if it's 807 * all ASCII and it's either <= 100 characters 808 * or can be split at a '/' into a prefix <= 809 * 155 chars and a suffix <= 100 chars. (Note 810 * the strchr() above will return NULL exactly 811 * when the path can't be split.) 812 */ 813 if (suffix == NULL /* Suffix > 100 chars. */ 814 || suffix[1] == '\0' /* empty suffix */ 815 || suffix - path > 155) /* Prefix > 155 chars */ 816 { 817 add_pax_attr(&(pax->pax_header), "path", path); 818 archive_entry_set_pathname(entry_main, 819 build_ustar_entry_name(ustar_entry_name, 820 path, path_length, NULL)); 821 need_extension = 1; 822 } 823 } 824 } 825 826 if (linkpath != NULL) { 827 /* If link name is too long or has non-ASCII characters, add 828 * 'linkpath' to pax extended attrs. */ 829 if (linkpath_length > 100 || has_non_ASCII(linkpath)) { 830 add_pax_attr(&(pax->pax_header), "linkpath", linkpath); 831 if (linkpath_length > 100) { 832 if (hardlink != NULL) 833 archive_entry_set_hardlink(entry_main, 834 "././@LongHardLink"); 835 else 836 archive_entry_set_symlink(entry_main, 837 "././@LongSymLink"); 838 } 839 need_extension = 1; 840 } 841 } 842 /* Save a pathname since it will be renamed if `entry_main` has 843 * sparse blocks. */ 844 archive_string_init(&entry_name); 845 archive_strcpy(&entry_name, archive_entry_pathname(entry_main)); 846 847 /* If file size is too large, add 'size' to pax extended attrs. */ 848 if (archive_entry_size(entry_main) >= (((int64_t)1) << 33)) { 849 add_pax_attr_int(&(pax->pax_header), "size", 850 archive_entry_size(entry_main)); 851 need_extension = 1; 852 } 853 854 /* If numeric GID is too large, add 'gid' to pax extended attrs. */ 855 if ((unsigned int)archive_entry_gid(entry_main) >= (1 << 18)) { 856 add_pax_attr_int(&(pax->pax_header), "gid", 857 archive_entry_gid(entry_main)); 858 need_extension = 1; 859 } 860 861 /* If group name is too large or has non-ASCII characters, add 862 * 'gname' to pax extended attrs. */ 863 if (gname != NULL) { 864 if (gname_length > 31 || has_non_ASCII(gname)) { 865 add_pax_attr(&(pax->pax_header), "gname", gname); 866 need_extension = 1; 867 } 868 } 869 870 /* If numeric UID is too large, add 'uid' to pax extended attrs. */ 871 if ((unsigned int)archive_entry_uid(entry_main) >= (1 << 18)) { 872 add_pax_attr_int(&(pax->pax_header), "uid", 873 archive_entry_uid(entry_main)); 874 need_extension = 1; 875 } 876 877 /* Add 'uname' to pax extended attrs if necessary. */ 878 if (uname != NULL) { 879 if (uname_length > 31 || has_non_ASCII(uname)) { 880 add_pax_attr(&(pax->pax_header), "uname", uname); 881 need_extension = 1; 882 } 883 } 884 885 /* 886 * POSIX/SUSv3 doesn't provide a standard key for large device 887 * numbers. I use the same keys here that Joerg Schilling 888 * used for 'star.' (Which, somewhat confusingly, are called 889 * "devXXX" even though they code "rdev" values.) No doubt, 890 * other implementations use other keys. Note that there's no 891 * reason we can't write the same information into a number of 892 * different keys. 893 * 894 * Of course, this is only needed for block or char device entries. 895 */ 896 if (archive_entry_filetype(entry_main) == AE_IFBLK 897 || archive_entry_filetype(entry_main) == AE_IFCHR) { 898 /* 899 * If rdevmajor is too large, add 'SCHILY.devmajor' to 900 * extended attributes. 901 */ 902 int rdevmajor, rdevminor; 903 rdevmajor = archive_entry_rdevmajor(entry_main); 904 rdevminor = archive_entry_rdevminor(entry_main); 905 if (rdevmajor >= (1 << 18)) { 906 add_pax_attr_int(&(pax->pax_header), "SCHILY.devmajor", 907 rdevmajor); 908 /* 909 * Non-strict formatting below means we don't 910 * have to truncate here. Not truncating improves 911 * the chance that some more modern tar archivers 912 * (such as GNU tar 1.13) can restore the full 913 * value even if they don't understand the pax 914 * extended attributes. See my rant below about 915 * file size fields for additional details. 916 */ 917 /* archive_entry_set_rdevmajor(entry_main, 918 rdevmajor & ((1 << 18) - 1)); */ 919 need_extension = 1; 920 } 921 922 /* 923 * If devminor is too large, add 'SCHILY.devminor' to 924 * extended attributes. 925 */ 926 if (rdevminor >= (1 << 18)) { 927 add_pax_attr_int(&(pax->pax_header), "SCHILY.devminor", 928 rdevminor); 929 /* Truncation is not necessary here, either. */ 930 /* archive_entry_set_rdevminor(entry_main, 931 rdevminor & ((1 << 18) - 1)); */ 932 need_extension = 1; 933 } 934 } 935 936 /* 937 * Technically, the mtime field in the ustar header can 938 * support 33 bits, but many platforms use signed 32-bit time 939 * values. The cutoff of 0x7fffffff here is a compromise. 940 * Yes, this check is duplicated just below; this helps to 941 * avoid writing an mtime attribute just to handle a 942 * high-resolution timestamp in "restricted pax" mode. 943 */ 944 if (!need_extension && 945 ((archive_entry_mtime(entry_main) < 0) 946 || (archive_entry_mtime(entry_main) >= 0x7fffffff))) 947 need_extension = 1; 948 949 /* I use a star-compatible file flag attribute. */ 950 p = archive_entry_fflags_text(entry_main); 951 if (!need_extension && p != NULL && *p != '\0') 952 need_extension = 1; 953 954 /* If there are non-trivial ACL entries, we need an extension. */ 955 if (!need_extension && archive_entry_acl_count(entry_original, 956 ARCHIVE_ENTRY_ACL_TYPE_ACCESS) > 0) 957 need_extension = 1; 958 959 /* If there are non-trivial ACL entries, we need an extension. */ 960 if (!need_extension && archive_entry_acl_count(entry_original, 961 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT) > 0) 962 need_extension = 1; 963 964 /* If there are extended attributes, we need an extension */ 965 if (!need_extension && archive_entry_xattr_count(entry_original) > 0) 966 need_extension = 1; 967 968 /* If there are sparse info, we need an extension */ 969 if (!need_extension && sparse_count > 0) 970 need_extension = 1; 971 972 /* 973 * The following items are handled differently in "pax 974 * restricted" format. In particular, in "pax restricted" 975 * format they won't be added unless need_extension is 976 * already set (we're already generating an extended header, so 977 * may as well include these). 978 */ 979 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_RESTRICTED || 980 need_extension) { 981 982 if (archive_entry_mtime(entry_main) < 0 || 983 archive_entry_mtime(entry_main) >= 0x7fffffff || 984 archive_entry_mtime_nsec(entry_main) != 0) 985 add_pax_attr_time(&(pax->pax_header), "mtime", 986 archive_entry_mtime(entry_main), 987 archive_entry_mtime_nsec(entry_main)); 988 989 if (archive_entry_ctime(entry_main) != 0 || 990 archive_entry_ctime_nsec(entry_main) != 0) 991 add_pax_attr_time(&(pax->pax_header), "ctime", 992 archive_entry_ctime(entry_main), 993 archive_entry_ctime_nsec(entry_main)); 994 995 if (archive_entry_atime(entry_main) != 0 || 996 archive_entry_atime_nsec(entry_main) != 0) 997 add_pax_attr_time(&(pax->pax_header), "atime", 998 archive_entry_atime(entry_main), 999 archive_entry_atime_nsec(entry_main)); 1000 1001 /* Store birth/creationtime only if it's earlier than mtime */ 1002 if (archive_entry_birthtime_is_set(entry_main) && 1003 archive_entry_birthtime(entry_main) 1004 < archive_entry_mtime(entry_main)) 1005 add_pax_attr_time(&(pax->pax_header), 1006 "LIBARCHIVE.creationtime", 1007 archive_entry_birthtime(entry_main), 1008 archive_entry_birthtime_nsec(entry_main)); 1009 1010 /* I use a star-compatible file flag attribute. */ 1011 p = archive_entry_fflags_text(entry_main); 1012 if (p != NULL && *p != '\0') 1013 add_pax_attr(&(pax->pax_header), "SCHILY.fflags", p); 1014 1015 /* I use star-compatible ACL attributes. */ 1016 r = archive_entry_acl_text_l(entry_original, 1017 ARCHIVE_ENTRY_ACL_TYPE_ACCESS | 1018 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID, 1019 &p, NULL, pax->sconv_utf8); 1020 if (r != 0) { 1021 if (errno == ENOMEM) { 1022 archive_set_error(&a->archive, ENOMEM, 1023 "Can't allocate memory for " 1024 "ACL.access"); 1025 return (ARCHIVE_FATAL); 1026 } 1027 archive_set_error(&a->archive, 1028 ARCHIVE_ERRNO_FILE_FORMAT, 1029 "Can't translate ACL.access to UTF-8"); 1030 ret = ARCHIVE_WARN; 1031 } else if (p != NULL && *p != '\0') { 1032 add_pax_attr(&(pax->pax_header), 1033 "SCHILY.acl.access", p); 1034 } 1035 r = archive_entry_acl_text_l(entry_original, 1036 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT | 1037 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID, 1038 &p, NULL, pax->sconv_utf8); 1039 if (r != 0) { 1040 if (errno == ENOMEM) { 1041 archive_set_error(&a->archive, ENOMEM, 1042 "Can't allocate memory for " 1043 "ACL.default"); 1044 return (ARCHIVE_FATAL); 1045 } 1046 archive_set_error(&a->archive, 1047 ARCHIVE_ERRNO_FILE_FORMAT, 1048 "Can't translate ACL.default to UTF-8"); 1049 ret = ARCHIVE_WARN; 1050 } else if (p != NULL && *p != '\0') { 1051 add_pax_attr(&(pax->pax_header), 1052 "SCHILY.acl.default", p); 1053 } 1054 1055 /* We use GNU-tar-compatible sparse attributes. */ 1056 if (sparse_count > 0) { 1057 int64_t soffset, slength; 1058 1059 add_pax_attr_int(&(pax->pax_header), 1060 "GNU.sparse.major", 1); 1061 add_pax_attr_int(&(pax->pax_header), 1062 "GNU.sparse.minor", 0); 1063 add_pax_attr(&(pax->pax_header), 1064 "GNU.sparse.name", entry_name.s); 1065 add_pax_attr_int(&(pax->pax_header), 1066 "GNU.sparse.realsize", 1067 archive_entry_size(entry_main)); 1068 1069 /* Rename the file name which will be used for 1070 * ustar header to a special name, which GNU 1071 * PAX Format 1.0 requires */ 1072 archive_entry_set_pathname(entry_main, 1073 build_gnu_sparse_name(gnu_sparse_name, 1074 entry_name.s)); 1075 1076 /* 1077 * - Make a sparse map, which will precede a file data. 1078 * - Get the total size of available data of sparse. 1079 */ 1080 archive_string_sprintf(&(pax->sparse_map), "%d\n", 1081 sparse_count); 1082 while (archive_entry_sparse_next(entry_main, 1083 &soffset, &slength) == ARCHIVE_OK) { 1084 archive_string_sprintf(&(pax->sparse_map), 1085 "%jd\n%jd\n", 1086 (intmax_t)soffset, 1087 (intmax_t)slength); 1088 sparse_total += slength; 1089 if (sparse_list_add(pax, soffset, slength) 1090 != ARCHIVE_OK) { 1091 archive_set_error(&a->archive, 1092 ENOMEM, 1093 "Can't allocate memory"); 1094 archive_entry_free(entry_main); 1095 archive_string_free(&entry_name); 1096 return (ARCHIVE_FATAL); 1097 } 1098 } 1099 } 1100 1101 /* Store extended attributes */ 1102 if (archive_write_pax_header_xattrs(a, pax, entry_original) 1103 == ARCHIVE_FATAL) { 1104 archive_entry_free(entry_main); 1105 archive_string_free(&entry_name); 1106 return (ARCHIVE_FATAL); 1107 } 1108 } 1109 1110 /* Only regular files have data. */ 1111 if (archive_entry_filetype(entry_main) != AE_IFREG) 1112 archive_entry_set_size(entry_main, 0); 1113 1114 /* 1115 * Pax-restricted does not store data for hardlinks, in order 1116 * to improve compatibility with ustar. 1117 */ 1118 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE && 1119 hardlink != NULL) 1120 archive_entry_set_size(entry_main, 0); 1121 1122 /* 1123 * XXX Full pax interchange format does permit a hardlink 1124 * entry to have data associated with it. I'm not supporting 1125 * that here because the client expects me to tell them whether 1126 * or not this format expects data for hardlinks. If I 1127 * don't check here, then every pax archive will end up with 1128 * duplicated data for hardlinks. Someday, there may be 1129 * need to select this behavior, in which case the following 1130 * will need to be revisited. XXX 1131 */ 1132 if (hardlink != NULL) 1133 archive_entry_set_size(entry_main, 0); 1134 1135 /* Save a real file size. */ 1136 real_size = archive_entry_size(entry_main); 1137 /* 1138 * Overwrite a file size by the total size of sparse blocks and 1139 * the size of sparse map info. That file size is the length of 1140 * the data, which we will exactly store into an archive file. 1141 */ 1142 if (archive_strlen(&(pax->sparse_map))) { 1143 size_t mapsize = archive_strlen(&(pax->sparse_map)); 1144 pax->sparse_map_padding = 0x1ff & (-(ssize_t)mapsize); 1145 archive_entry_set_size(entry_main, 1146 mapsize + pax->sparse_map_padding + sparse_total); 1147 } 1148 1149 /* Format 'ustar' header for main entry. 1150 * 1151 * The trouble with file size: If the reader can't understand 1152 * the file size, they may not be able to locate the next 1153 * entry and the rest of the archive is toast. Pax-compliant 1154 * readers are supposed to ignore the file size in the main 1155 * header, so the question becomes how to maximize portability 1156 * for readers that don't support pax attribute extensions. 1157 * For maximum compatibility, I permit numeric extensions in 1158 * the main header so that the file size stored will always be 1159 * correct, even if it's in a format that only some 1160 * implementations understand. The technique used here is: 1161 * 1162 * a) If possible, follow the standard exactly. This handles 1163 * files up to 8 gigabytes minus 1. 1164 * 1165 * b) If that fails, try octal but omit the field terminator. 1166 * That handles files up to 64 gigabytes minus 1. 1167 * 1168 * c) Otherwise, use base-256 extensions. That handles files 1169 * up to 2^63 in this implementation, with the potential to 1170 * go up to 2^94. That should hold us for a while. ;-) 1171 * 1172 * The non-strict formatter uses similar logic for other 1173 * numeric fields, though they're less critical. 1174 */ 1175 if (__archive_write_format_header_ustar(a, ustarbuff, entry_main, -1, 0, 1176 NULL) == ARCHIVE_FATAL) 1177 return (ARCHIVE_FATAL); 1178 1179 /* If we built any extended attributes, write that entry first. */ 1180 if (archive_strlen(&(pax->pax_header)) > 0) { 1181 struct archive_entry *pax_attr_entry; 1182 time_t s; 1183 int64_t uid, gid; 1184 int mode; 1185 1186 pax_attr_entry = archive_entry_new2(&a->archive); 1187 p = entry_name.s; 1188 archive_entry_set_pathname(pax_attr_entry, 1189 build_pax_attribute_name(pax_entry_name, p)); 1190 archive_entry_set_size(pax_attr_entry, 1191 archive_strlen(&(pax->pax_header))); 1192 /* Copy uid/gid (but clip to ustar limits). */ 1193 uid = archive_entry_uid(entry_main); 1194 if (uid >= 1 << 18) 1195 uid = (1 << 18) - 1; 1196 archive_entry_set_uid(pax_attr_entry, uid); 1197 gid = archive_entry_gid(entry_main); 1198 if (gid >= 1 << 18) 1199 gid = (1 << 18) - 1; 1200 archive_entry_set_gid(pax_attr_entry, gid); 1201 /* Copy mode over (but not setuid/setgid bits) */ 1202 mode = archive_entry_mode(entry_main); 1203 #ifdef S_ISUID 1204 mode &= ~S_ISUID; 1205 #endif 1206 #ifdef S_ISGID 1207 mode &= ~S_ISGID; 1208 #endif 1209 #ifdef S_ISVTX 1210 mode &= ~S_ISVTX; 1211 #endif 1212 archive_entry_set_mode(pax_attr_entry, mode); 1213 1214 /* Copy uname/gname. */ 1215 archive_entry_set_uname(pax_attr_entry, 1216 archive_entry_uname(entry_main)); 1217 archive_entry_set_gname(pax_attr_entry, 1218 archive_entry_gname(entry_main)); 1219 1220 /* Copy mtime, but clip to ustar limits. */ 1221 s = archive_entry_mtime(entry_main); 1222 if (s < 0) { s = 0; } 1223 if (s >= 0x7fffffff) { s = 0x7fffffff; } 1224 archive_entry_set_mtime(pax_attr_entry, s, 0); 1225 1226 /* Standard ustar doesn't support atime. */ 1227 archive_entry_set_atime(pax_attr_entry, 0, 0); 1228 1229 /* Standard ustar doesn't support ctime. */ 1230 archive_entry_set_ctime(pax_attr_entry, 0, 0); 1231 1232 r = __archive_write_format_header_ustar(a, paxbuff, 1233 pax_attr_entry, 'x', 1, NULL); 1234 1235 archive_entry_free(pax_attr_entry); 1236 1237 /* Note that the 'x' header shouldn't ever fail to format */ 1238 if (r < ARCHIVE_WARN) { 1239 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, 1240 "archive_write_pax_header: " 1241 "'x' header failed?! This can't happen.\n"); 1242 return (ARCHIVE_FATAL); 1243 } else if (r < ret) 1244 ret = r; 1245 r = __archive_write_output(a, paxbuff, 512); 1246 if (r != ARCHIVE_OK) { 1247 sparse_list_clear(pax); 1248 pax->entry_bytes_remaining = 0; 1249 pax->entry_padding = 0; 1250 return (ARCHIVE_FATAL); 1251 } 1252 1253 pax->entry_bytes_remaining = archive_strlen(&(pax->pax_header)); 1254 pax->entry_padding = 1255 0x1ff & (-(int64_t)pax->entry_bytes_remaining); 1256 1257 r = __archive_write_output(a, pax->pax_header.s, 1258 archive_strlen(&(pax->pax_header))); 1259 if (r != ARCHIVE_OK) { 1260 /* If a write fails, we're pretty much toast. */ 1261 return (ARCHIVE_FATAL); 1262 } 1263 /* Pad out the end of the entry. */ 1264 r = __archive_write_nulls(a, pax->entry_padding); 1265 if (r != ARCHIVE_OK) { 1266 /* If a write fails, we're pretty much toast. */ 1267 return (ARCHIVE_FATAL); 1268 } 1269 pax->entry_bytes_remaining = pax->entry_padding = 0; 1270 } 1271 1272 /* Write the header for main entry. */ 1273 r = __archive_write_output(a, ustarbuff, 512); 1274 if (r != ARCHIVE_OK) 1275 return (r); 1276 1277 /* 1278 * Inform the client of the on-disk size we're using, so 1279 * they can avoid unnecessarily writing a body for something 1280 * that we're just going to ignore. 1281 */ 1282 archive_entry_set_size(entry_original, real_size); 1283 if (pax->sparse_list == NULL && real_size > 0) { 1284 /* This is not a sparse file but we handle its data as 1285 * a sparse block. */ 1286 sparse_list_add(pax, 0, real_size); 1287 sparse_total = real_size; 1288 } 1289 pax->entry_padding = 0x1ff & (-(int64_t)sparse_total); 1290 archive_entry_free(entry_main); 1291 archive_string_free(&entry_name); 1292 1293 return (ret); 1294 } 1295 1296 /* 1297 * We need a valid name for the regular 'ustar' entry. This routine 1298 * tries to hack something more-or-less reasonable. 1299 * 1300 * The approach here tries to preserve leading dir names. We do so by 1301 * working with four sections: 1302 * 1) "prefix" directory names, 1303 * 2) "suffix" directory names, 1304 * 3) inserted dir name (optional), 1305 * 4) filename. 1306 * 1307 * These sections must satisfy the following requirements: 1308 * * Parts 1 & 2 together form an initial portion of the dir name. 1309 * * Part 3 is specified by the caller. (It should not contain a leading 1310 * or trailing '/'.) 1311 * * Part 4 forms an initial portion of the base filename. 1312 * * The filename must be <= 99 chars to fit the ustar 'name' field. 1313 * * Parts 2, 3, 4 together must be <= 99 chars to fit the ustar 'name' fld. 1314 * * Part 1 must be <= 155 chars to fit the ustar 'prefix' field. 1315 * * If the original name ends in a '/', the new name must also end in a '/' 1316 * * Trailing '/.' sequences may be stripped. 1317 * 1318 * Note: Recall that the ustar format does not store the '/' separating 1319 * parts 1 & 2, but does store the '/' separating parts 2 & 3. 1320 */ 1321 static char * 1322 build_ustar_entry_name(char *dest, const char *src, size_t src_length, 1323 const char *insert) 1324 { 1325 const char *prefix, *prefix_end; 1326 const char *suffix, *suffix_end; 1327 const char *filename, *filename_end; 1328 char *p; 1329 int need_slash = 0; /* Was there a trailing slash? */ 1330 size_t suffix_length = 99; 1331 size_t insert_length; 1332 1333 /* Length of additional dir element to be added. */ 1334 if (insert == NULL) 1335 insert_length = 0; 1336 else 1337 /* +2 here allows for '/' before and after the insert. */ 1338 insert_length = strlen(insert) + 2; 1339 1340 /* Step 0: Quick bailout in a common case. */ 1341 if (src_length < 100 && insert == NULL) { 1342 strncpy(dest, src, src_length); 1343 dest[src_length] = '\0'; 1344 return (dest); 1345 } 1346 1347 /* Step 1: Locate filename and enforce the length restriction. */ 1348 filename_end = src + src_length; 1349 /* Remove trailing '/' chars and '/.' pairs. */ 1350 for (;;) { 1351 if (filename_end > src && filename_end[-1] == '/') { 1352 filename_end --; 1353 need_slash = 1; /* Remember to restore trailing '/'. */ 1354 continue; 1355 } 1356 if (filename_end > src + 1 && filename_end[-1] == '.' 1357 && filename_end[-2] == '/') { 1358 filename_end -= 2; 1359 need_slash = 1; /* "foo/." will become "foo/" */ 1360 continue; 1361 } 1362 break; 1363 } 1364 if (need_slash) 1365 suffix_length--; 1366 /* Find start of filename. */ 1367 filename = filename_end - 1; 1368 while ((filename > src) && (*filename != '/')) 1369 filename --; 1370 if ((*filename == '/') && (filename < filename_end - 1)) 1371 filename ++; 1372 /* Adjust filename_end so that filename + insert fits in 99 chars. */ 1373 suffix_length -= insert_length; 1374 if (filename_end > filename + suffix_length) 1375 filename_end = filename + suffix_length; 1376 /* Calculate max size for "suffix" section (#3 above). */ 1377 suffix_length -= filename_end - filename; 1378 1379 /* Step 2: Locate the "prefix" section of the dirname, including 1380 * trailing '/'. */ 1381 prefix = src; 1382 prefix_end = prefix + 155; 1383 if (prefix_end > filename) 1384 prefix_end = filename; 1385 while (prefix_end > prefix && *prefix_end != '/') 1386 prefix_end--; 1387 if ((prefix_end < filename) && (*prefix_end == '/')) 1388 prefix_end++; 1389 1390 /* Step 3: Locate the "suffix" section of the dirname, 1391 * including trailing '/'. */ 1392 suffix = prefix_end; 1393 suffix_end = suffix + suffix_length; /* Enforce limit. */ 1394 if (suffix_end > filename) 1395 suffix_end = filename; 1396 if (suffix_end < suffix) 1397 suffix_end = suffix; 1398 while (suffix_end > suffix && *suffix_end != '/') 1399 suffix_end--; 1400 if ((suffix_end < filename) && (*suffix_end == '/')) 1401 suffix_end++; 1402 1403 /* Step 4: Build the new name. */ 1404 /* The OpenBSD strlcpy function is safer, but less portable. */ 1405 /* Rather than maintain two versions, just use the strncpy version. */ 1406 p = dest; 1407 if (prefix_end > prefix) { 1408 strncpy(p, prefix, prefix_end - prefix); 1409 p += prefix_end - prefix; 1410 } 1411 if (suffix_end > suffix) { 1412 strncpy(p, suffix, suffix_end - suffix); 1413 p += suffix_end - suffix; 1414 } 1415 if (insert != NULL) { 1416 /* Note: assume insert does not have leading or trailing '/' */ 1417 strcpy(p, insert); 1418 p += strlen(insert); 1419 *p++ = '/'; 1420 } 1421 strncpy(p, filename, filename_end - filename); 1422 p += filename_end - filename; 1423 if (need_slash) 1424 *p++ = '/'; 1425 *p = '\0'; 1426 1427 return (dest); 1428 } 1429 1430 /* 1431 * The ustar header for the pax extended attributes must have a 1432 * reasonable name: SUSv3 requires 'dirname'/PaxHeader.'pid'/'filename' 1433 * where 'pid' is the PID of the archiving process. Unfortunately, 1434 * that makes testing a pain since the output varies for each run, 1435 * so I'm sticking with the simpler 'dirname'/PaxHeader/'filename' 1436 * for now. (Someday, I'll make this settable. Then I can use the 1437 * SUS recommendation as default and test harnesses can override it 1438 * to get predictable results.) 1439 * 1440 * Joerg Schilling has argued that this is unnecessary because, in 1441 * practice, if the pax extended attributes get extracted as regular 1442 * files, no one is going to bother reading those attributes to 1443 * manually restore them. Based on this, 'star' uses 1444 * /tmp/PaxHeader/'basename' as the ustar header name. This is a 1445 * tempting argument, in part because it's simpler than the SUSv3 1446 * recommendation, but I'm not entirely convinced. I'm also 1447 * uncomfortable with the fact that "/tmp" is a Unix-ism. 1448 * 1449 * The following routine leverages build_ustar_entry_name() above and 1450 * so is simpler than you might think. It just needs to provide the 1451 * additional path element and handle a few pathological cases). 1452 */ 1453 static char * 1454 build_pax_attribute_name(char *dest, const char *src) 1455 { 1456 char buff[64]; 1457 const char *p; 1458 1459 /* Handle the null filename case. */ 1460 if (src == NULL || *src == '\0') { 1461 strcpy(dest, "PaxHeader/blank"); 1462 return (dest); 1463 } 1464 1465 /* Prune final '/' and other unwanted final elements. */ 1466 p = src + strlen(src); 1467 for (;;) { 1468 /* Ends in "/", remove the '/' */ 1469 if (p > src && p[-1] == '/') { 1470 --p; 1471 continue; 1472 } 1473 /* Ends in "/.", remove the '.' */ 1474 if (p > src + 1 && p[-1] == '.' 1475 && p[-2] == '/') { 1476 --p; 1477 continue; 1478 } 1479 break; 1480 } 1481 1482 /* Pathological case: After above, there was nothing left. 1483 * This includes "/." "/./." "/.//./." etc. */ 1484 if (p == src) { 1485 strcpy(dest, "/PaxHeader/rootdir"); 1486 return (dest); 1487 } 1488 1489 /* Convert unadorned "." into a suitable filename. */ 1490 if (*src == '.' && p == src + 1) { 1491 strcpy(dest, "PaxHeader/currentdir"); 1492 return (dest); 1493 } 1494 1495 /* 1496 * TODO: Push this string into the 'pax' structure to avoid 1497 * recomputing it every time. That will also open the door 1498 * to having clients override it. 1499 */ 1500 #if HAVE_GETPID && 0 /* Disable this for now; see above comment. */ 1501 sprintf(buff, "PaxHeader.%d", getpid()); 1502 #else 1503 /* If the platform can't fetch the pid, don't include it. */ 1504 strcpy(buff, "PaxHeader"); 1505 #endif 1506 /* General case: build a ustar-compatible name adding 1507 * "/PaxHeader/". */ 1508 build_ustar_entry_name(dest, src, p - src, buff); 1509 1510 return (dest); 1511 } 1512 1513 /* 1514 * GNU PAX Format 1.0 requires the special name, which pattern is: 1515 * <dir>/GNUSparseFile.<pid>/<original file name> 1516 * 1517 * This function is used for only Sparse file, a file type of which 1518 * is regular file. 1519 */ 1520 static char * 1521 build_gnu_sparse_name(char *dest, const char *src) 1522 { 1523 char buff[64]; 1524 const char *p; 1525 1526 /* Handle the null filename case. */ 1527 if (src == NULL || *src == '\0') { 1528 strcpy(dest, "GNUSparseFile/blank"); 1529 return (dest); 1530 } 1531 1532 /* Prune final '/' and other unwanted final elements. */ 1533 p = src + strlen(src); 1534 for (;;) { 1535 /* Ends in "/", remove the '/' */ 1536 if (p > src && p[-1] == '/') { 1537 --p; 1538 continue; 1539 } 1540 /* Ends in "/.", remove the '.' */ 1541 if (p > src + 1 && p[-1] == '.' 1542 && p[-2] == '/') { 1543 --p; 1544 continue; 1545 } 1546 break; 1547 } 1548 1549 #if HAVE_GETPID && 0 /* Disable this as pax attribute name. */ 1550 sprintf(buff, "GNUSparseFile.%d", getpid()); 1551 #else 1552 /* If the platform can't fetch the pid, don't include it. */ 1553 strcpy(buff, "GNUSparseFile"); 1554 #endif 1555 /* General case: build a ustar-compatible name adding 1556 * "/GNUSparseFile/". */ 1557 build_ustar_entry_name(dest, src, p - src, buff); 1558 1559 return (dest); 1560 } 1561 1562 /* Write two null blocks for the end of archive */ 1563 static int 1564 archive_write_pax_close(struct archive_write *a) 1565 { 1566 return (__archive_write_nulls(a, 512 * 2)); 1567 } 1568 1569 static int 1570 archive_write_pax_free(struct archive_write *a) 1571 { 1572 struct pax *pax; 1573 1574 pax = (struct pax *)a->format_data; 1575 if (pax == NULL) 1576 return (ARCHIVE_OK); 1577 1578 archive_string_free(&pax->pax_header); 1579 archive_string_free(&pax->sparse_map); 1580 archive_string_free(&pax->l_url_encoded_name); 1581 sparse_list_clear(pax); 1582 free(pax); 1583 a->format_data = NULL; 1584 return (ARCHIVE_OK); 1585 } 1586 1587 static int 1588 archive_write_pax_finish_entry(struct archive_write *a) 1589 { 1590 struct pax *pax; 1591 uint64_t remaining; 1592 int ret; 1593 1594 pax = (struct pax *)a->format_data; 1595 remaining = pax->entry_bytes_remaining; 1596 if (remaining == 0) { 1597 while (pax->sparse_list) { 1598 struct sparse_block *sb; 1599 if (!pax->sparse_list->is_hole) 1600 remaining += pax->sparse_list->remaining; 1601 sb = pax->sparse_list->next; 1602 free(pax->sparse_list); 1603 pax->sparse_list = sb; 1604 } 1605 } 1606 ret = __archive_write_nulls(a, remaining + pax->entry_padding); 1607 pax->entry_bytes_remaining = pax->entry_padding = 0; 1608 return (ret); 1609 } 1610 1611 static ssize_t 1612 archive_write_pax_data(struct archive_write *a, const void *buff, size_t s) 1613 { 1614 struct pax *pax; 1615 size_t ws; 1616 size_t total; 1617 int ret; 1618 1619 pax = (struct pax *)a->format_data; 1620 1621 /* 1622 * According to GNU PAX format 1.0, write a sparse map 1623 * before the body. 1624 */ 1625 if (archive_strlen(&(pax->sparse_map))) { 1626 ret = __archive_write_output(a, pax->sparse_map.s, 1627 archive_strlen(&(pax->sparse_map))); 1628 if (ret != ARCHIVE_OK) 1629 return (ret); 1630 ret = __archive_write_nulls(a, pax->sparse_map_padding); 1631 if (ret != ARCHIVE_OK) 1632 return (ret); 1633 archive_string_empty(&(pax->sparse_map)); 1634 } 1635 1636 total = 0; 1637 while (total < s) { 1638 const unsigned char *p; 1639 1640 while (pax->sparse_list != NULL && 1641 pax->sparse_list->remaining == 0) { 1642 struct sparse_block *sb = pax->sparse_list->next; 1643 free(pax->sparse_list); 1644 pax->sparse_list = sb; 1645 } 1646 1647 if (pax->sparse_list == NULL) 1648 return (total); 1649 1650 p = ((const unsigned char *)buff) + total; 1651 ws = s - total; 1652 if (ws > pax->sparse_list->remaining) 1653 ws = pax->sparse_list->remaining; 1654 1655 if (pax->sparse_list->is_hole) { 1656 /* Current block is hole thus we do not write 1657 * the body. */ 1658 pax->sparse_list->remaining -= ws; 1659 total += ws; 1660 continue; 1661 } 1662 1663 ret = __archive_write_output(a, p, ws); 1664 pax->sparse_list->remaining -= ws; 1665 total += ws; 1666 if (ret != ARCHIVE_OK) 1667 return (ret); 1668 } 1669 return (total); 1670 } 1671 1672 static int 1673 has_non_ASCII(const char *_p) 1674 { 1675 const unsigned char *p = (const unsigned char *)_p; 1676 1677 if (p == NULL) 1678 return (1); 1679 while (*p != '\0' && *p < 128) 1680 p++; 1681 return (*p != '\0'); 1682 } 1683 1684 /* 1685 * Used by extended attribute support; encodes the name 1686 * so that there will be no '=' characters in the result. 1687 */ 1688 static char * 1689 url_encode(const char *in) 1690 { 1691 const char *s; 1692 char *d; 1693 int out_len = 0; 1694 char *out; 1695 1696 for (s = in; *s != '\0'; s++) { 1697 if (*s < 33 || *s > 126 || *s == '%' || *s == '=') 1698 out_len += 3; 1699 else 1700 out_len++; 1701 } 1702 1703 out = (char *)malloc(out_len + 1); 1704 if (out == NULL) 1705 return (NULL); 1706 1707 for (s = in, d = out; *s != '\0'; s++) { 1708 /* encode any non-printable ASCII character or '%' or '=' */ 1709 if (*s < 33 || *s > 126 || *s == '%' || *s == '=') { 1710 /* URL encoding is '%' followed by two hex digits */ 1711 *d++ = '%'; 1712 *d++ = "0123456789ABCDEF"[0x0f & (*s >> 4)]; 1713 *d++ = "0123456789ABCDEF"[0x0f & *s]; 1714 } else { 1715 *d++ = *s; 1716 } 1717 } 1718 *d = '\0'; 1719 return (out); 1720 } 1721 1722 /* 1723 * Encode a sequence of bytes into a C string using base-64 encoding. 1724 * 1725 * Returns a null-terminated C string allocated with malloc(); caller 1726 * is responsible for freeing the result. 1727 */ 1728 static char * 1729 base64_encode(const char *s, size_t len) 1730 { 1731 static const char digits[64] = 1732 { 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O', 1733 'P','Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d', 1734 'e','f','g','h','i','j','k','l','m','n','o','p','q','r','s', 1735 't','u','v','w','x','y','z','0','1','2','3','4','5','6','7', 1736 '8','9','+','/' }; 1737 int v; 1738 char *d, *out; 1739 1740 /* 3 bytes becomes 4 chars, but round up and allow for trailing NUL */ 1741 out = (char *)malloc((len * 4 + 2) / 3 + 1); 1742 if (out == NULL) 1743 return (NULL); 1744 d = out; 1745 1746 /* Convert each group of 3 bytes into 4 characters. */ 1747 while (len >= 3) { 1748 v = (((int)s[0] << 16) & 0xff0000) 1749 | (((int)s[1] << 8) & 0xff00) 1750 | (((int)s[2]) & 0x00ff); 1751 s += 3; 1752 len -= 3; 1753 *d++ = digits[(v >> 18) & 0x3f]; 1754 *d++ = digits[(v >> 12) & 0x3f]; 1755 *d++ = digits[(v >> 6) & 0x3f]; 1756 *d++ = digits[(v) & 0x3f]; 1757 } 1758 /* Handle final group of 1 byte (2 chars) or 2 bytes (3 chars). */ 1759 switch (len) { 1760 case 0: break; 1761 case 1: 1762 v = (((int)s[0] << 16) & 0xff0000); 1763 *d++ = digits[(v >> 18) & 0x3f]; 1764 *d++ = digits[(v >> 12) & 0x3f]; 1765 break; 1766 case 2: 1767 v = (((int)s[0] << 16) & 0xff0000) 1768 | (((int)s[1] << 8) & 0xff00); 1769 *d++ = digits[(v >> 18) & 0x3f]; 1770 *d++ = digits[(v >> 12) & 0x3f]; 1771 *d++ = digits[(v >> 6) & 0x3f]; 1772 break; 1773 } 1774 /* Add trailing NUL character so output is a valid C string. */ 1775 *d = '\0'; 1776 return (out); 1777 } 1778 1779 static void 1780 sparse_list_clear(struct pax *pax) 1781 { 1782 while (pax->sparse_list != NULL) { 1783 struct sparse_block *sb = pax->sparse_list; 1784 pax->sparse_list = sb->next; 1785 free(sb); 1786 } 1787 pax->sparse_tail = NULL; 1788 } 1789 1790 static int 1791 _sparse_list_add_block(struct pax *pax, int64_t offset, int64_t length, 1792 int is_hole) 1793 { 1794 struct sparse_block *sb; 1795 1796 sb = (struct sparse_block *)malloc(sizeof(*sb)); 1797 if (sb == NULL) 1798 return (ARCHIVE_FATAL); 1799 sb->next = NULL; 1800 sb->is_hole = is_hole; 1801 sb->offset = offset; 1802 sb->remaining = length; 1803 if (pax->sparse_list == NULL) 1804 pax->sparse_list = pax->sparse_tail = sb; 1805 else { 1806 pax->sparse_tail->next = sb; 1807 pax->sparse_tail = sb; 1808 } 1809 return (ARCHIVE_OK); 1810 } 1811 1812 static int 1813 sparse_list_add(struct pax *pax, int64_t offset, int64_t length) 1814 { 1815 int64_t last_offset; 1816 int r; 1817 1818 if (pax->sparse_tail == NULL) 1819 last_offset = 0; 1820 else { 1821 last_offset = pax->sparse_tail->offset + 1822 pax->sparse_tail->remaining; 1823 } 1824 if (last_offset < offset) { 1825 /* Add a hole block. */ 1826 r = _sparse_list_add_block(pax, last_offset, 1827 offset - last_offset, 1); 1828 if (r != ARCHIVE_OK) 1829 return (r); 1830 } 1831 /* Add data block. */ 1832 return (_sparse_list_add_block(pax, offset, length, 0)); 1833 } 1834 1835