2 * Copyright (c) 2003-2007 Tim Kientzle
3 * Copyright (c) 2010-2012 Michihiro NAKAJIMA
4 * Copyright (c) 2016 Martin Matuska
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19 * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 #include "archive_platform.h"
29 __FBSDID("$FreeBSD$");
42 #include "archive_entry.h"
43 #include "archive_entry_locale.h"
44 #include "archive_private.h"
45 #include "archive_write_private.h"
48 struct sparse_block *next;
55 uint64_t entry_bytes_remaining;
56 uint64_t entry_padding;
57 struct archive_string l_url_encoded_name;
58 struct archive_string pax_header;
59 struct archive_string sparse_map;
60 size_t sparse_map_padding;
61 struct sparse_block *sparse_list;
62 struct sparse_block *sparse_tail;
63 struct archive_string_conv *sconv_utf8;
67 static void add_pax_attr(struct archive_string *, const char *key,
69 static void add_pax_attr_int(struct archive_string *,
70 const char *key, int64_t value);
71 static void add_pax_attr_time(struct archive_string *,
72 const char *key, int64_t sec,
74 static int add_pax_acl(struct archive_write *,
75 struct archive_entry *, struct pax *, int);
76 static ssize_t archive_write_pax_data(struct archive_write *,
77 const void *, size_t);
78 static int archive_write_pax_close(struct archive_write *);
79 static int archive_write_pax_free(struct archive_write *);
80 static int archive_write_pax_finish_entry(struct archive_write *);
81 static int archive_write_pax_header(struct archive_write *,
82 struct archive_entry *);
83 static int archive_write_pax_options(struct archive_write *,
84 const char *, const char *);
85 static char *base64_encode(const char *src, size_t len);
86 static char *build_gnu_sparse_name(char *dest, const char *src);
87 static char *build_pax_attribute_name(char *dest, const char *src);
88 static char *build_ustar_entry_name(char *dest, const char *src,
89 size_t src_length, const char *insert);
90 static char *format_int(char *dest, int64_t);
91 static int has_non_ASCII(const char *);
92 static void sparse_list_clear(struct pax *);
93 static int sparse_list_add(struct pax *, int64_t, int64_t);
94 static char *url_encode(const char *in);
97 * Set output format to 'restricted pax' format.
99 * This is the same as normal 'pax', but tries to suppress
100 * the pax header whenever possible. This is the default for
101 * bsdtar, for instance.
104 archive_write_set_format_pax_restricted(struct archive *_a)
106 struct archive_write *a = (struct archive_write *)_a;
109 archive_check_magic(_a, ARCHIVE_WRITE_MAGIC,
110 ARCHIVE_STATE_NEW, "archive_write_set_format_pax_restricted");
112 r = archive_write_set_format_pax(&a->archive);
113 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_RESTRICTED;
114 a->archive.archive_format_name = "restricted POSIX pax interchange";
119 * Set output format to 'pax' format.
122 archive_write_set_format_pax(struct archive *_a)
124 struct archive_write *a = (struct archive_write *)_a;
127 archive_check_magic(_a, ARCHIVE_WRITE_MAGIC,
128 ARCHIVE_STATE_NEW, "archive_write_set_format_pax");
130 if (a->format_free != NULL)
133 pax = (struct pax *)calloc(1, sizeof(*pax));
135 archive_set_error(&a->archive, ENOMEM,
136 "Can't allocate pax data");
137 return (ARCHIVE_FATAL);
139 a->format_data = pax;
140 a->format_name = "pax";
141 a->format_options = archive_write_pax_options;
142 a->format_write_header = archive_write_pax_header;
143 a->format_write_data = archive_write_pax_data;
144 a->format_close = archive_write_pax_close;
145 a->format_free = archive_write_pax_free;
146 a->format_finish_entry = archive_write_pax_finish_entry;
147 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
148 a->archive.archive_format_name = "POSIX pax interchange";
153 archive_write_pax_options(struct archive_write *a, const char *key,
156 struct pax *pax = (struct pax *)a->format_data;
157 int ret = ARCHIVE_FAILED;
159 if (strcmp(key, "hdrcharset") == 0) {
161 * The character-set we can use are defined in
162 * IEEE Std 1003.1-2001
164 if (val == NULL || val[0] == 0)
165 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
166 "pax: hdrcharset option needs a character-set name");
167 else if (strcmp(val, "BINARY") == 0 ||
168 strcmp(val, "binary") == 0) {
170 * Specify binary mode. We will not convert
171 * filenames, uname and gname to any charsets.
175 } else if (strcmp(val, "UTF-8") == 0) {
177 * Specify UTF-8 character-set to be used for
178 * filenames. This is almost the test that
179 * running platform supports the string conversion.
180 * Especially libarchive_test needs this trick for
183 pax->sconv_utf8 = archive_string_conversion_to_charset(
184 &(a->archive), "UTF-8", 0);
185 if (pax->sconv_utf8 == NULL)
190 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
191 "pax: invalid charset name");
195 /* Note: The "warn" return is just to inform the options
196 * supervisor that we didn't handle it. It will generate
197 * a suitable error if no one used this option. */
198 return (ARCHIVE_WARN);
202 * Note: This code assumes that 'nanos' has the same sign as 'sec',
203 * which implies that sec=-1, nanos=200000000 represents -1.2 seconds
204 * and not -0.8 seconds. This is a pretty pedantic point, as we're
205 * unlikely to encounter many real files created before Jan 1, 1970,
206 * much less ones with timestamps recorded to sub-second resolution.
209 add_pax_attr_time(struct archive_string *as, const char *key,
210 int64_t sec, unsigned long nanos)
215 * Note that each byte contributes fewer than 3 base-10
216 * digits, so this will always be big enough.
218 char tmp[1 + 3*sizeof(sec) + 1 + 3*sizeof(nanos)];
220 tmp[sizeof(tmp) - 1] = 0;
221 t = tmp + sizeof(tmp) - 1;
223 /* Skip trailing zeros in the fractional part. */
224 for (digit = 0, i = 10; i > 0 && digit == 0; i--) {
229 /* Only format the fraction if it's non-zero. */
232 *--t = "0123456789"[digit];
239 t = format_int(t, sec);
241 add_pax_attr(as, key, t);
245 format_int(char *t, int64_t i)
250 ui = (i == INT64_MIN) ? (uint64_t)(INT64_MAX) + 1 : (uint64_t)(-i);
255 *--t = "0123456789"[ui % 10];
263 add_pax_attr_int(struct archive_string *as, const char *key, int64_t value)
265 char tmp[1 + 3 * sizeof(value)];
267 tmp[sizeof(tmp) - 1] = 0;
268 add_pax_attr(as, key, format_int(tmp + sizeof(tmp) - 1, value));
272 * Add a key/value attribute to the pax header. This function handles
273 * the length field and various other syntactic requirements.
276 add_pax_attr(struct archive_string *as, const char *key, const char *value)
278 int digits, i, len, next_ten;
279 char tmp[1 + 3 * sizeof(int)]; /* < 3 base-10 digits per byte */
282 * PAX attributes have the following layout:
283 * <len> <space> <key> <=> <value> <nl>
285 len = 1 + (int)strlen(key) + 1 + (int)strlen(value) + 1;
288 * The <len> field includes the length of the <len> field, so
289 * computing the correct length is tricky. I start by
290 * counting the number of base-10 digits in 'len' and
291 * computing the next higher power of 10.
299 next_ten = next_ten * 10;
302 * For example, if string without the length field is 99
303 * chars, then adding the 2 digit length "99" will force the
304 * total length past 100, requiring an extra digit. The next
305 * statement adjusts for this effect.
307 if (len + digits >= next_ten)
310 /* Now, we have the right length so we can build the line. */
311 tmp[sizeof(tmp) - 1] = 0; /* Null-terminate the work area. */
312 archive_strcat(as, format_int(tmp + sizeof(tmp) - 1, len + digits));
313 archive_strappend_char(as, ' ');
314 archive_strcat(as, key);
315 archive_strappend_char(as, '=');
316 archive_strcat(as, value);
317 archive_strappend_char(as, '\n');
321 archive_write_pax_header_xattrs(struct archive_write *a,
322 struct pax *pax, struct archive_entry *entry)
324 struct archive_string s;
325 int i = archive_entry_xattr_reset(entry);
331 char *url_encoded_name = NULL, *encoded_name = NULL;
335 archive_entry_xattr_next(entry, &name, &value, &size);
336 url_encoded_name = url_encode(name);
337 if (url_encoded_name != NULL) {
338 /* Convert narrow-character to UTF-8. */
339 r = archive_strcpy_l(&(pax->l_url_encoded_name),
340 url_encoded_name, pax->sconv_utf8);
341 free(url_encoded_name); /* Done with this. */
343 encoded_name = pax->l_url_encoded_name.s;
344 else if (errno == ENOMEM) {
345 archive_set_error(&a->archive, ENOMEM,
346 "Can't allocate memory for Linkname");
347 return (ARCHIVE_FATAL);
351 encoded_value = base64_encode((const char *)value, size);
353 if (encoded_name != NULL && encoded_value != NULL) {
354 archive_string_init(&s);
355 archive_strcpy(&s, "LIBARCHIVE.xattr.");
356 archive_strcat(&s, encoded_name);
357 add_pax_attr(&(pax->pax_header), s.s, encoded_value);
358 archive_string_free(&s);
366 get_entry_hardlink(struct archive_write *a, struct archive_entry *entry,
367 const char **name, size_t *length, struct archive_string_conv *sc)
371 r = archive_entry_hardlink_l(entry, name, length, sc);
373 if (errno == ENOMEM) {
374 archive_set_error(&a->archive, ENOMEM,
375 "Can't allocate memory for Linkname");
376 return (ARCHIVE_FATAL);
378 return (ARCHIVE_WARN);
384 get_entry_pathname(struct archive_write *a, struct archive_entry *entry,
385 const char **name, size_t *length, struct archive_string_conv *sc)
389 r = archive_entry_pathname_l(entry, name, length, sc);
391 if (errno == ENOMEM) {
392 archive_set_error(&a->archive, ENOMEM,
393 "Can't allocate memory for Pathname");
394 return (ARCHIVE_FATAL);
396 return (ARCHIVE_WARN);
402 get_entry_uname(struct archive_write *a, struct archive_entry *entry,
403 const char **name, size_t *length, struct archive_string_conv *sc)
407 r = archive_entry_uname_l(entry, name, length, sc);
409 if (errno == ENOMEM) {
410 archive_set_error(&a->archive, ENOMEM,
411 "Can't allocate memory for Uname");
412 return (ARCHIVE_FATAL);
414 return (ARCHIVE_WARN);
420 get_entry_gname(struct archive_write *a, struct archive_entry *entry,
421 const char **name, size_t *length, struct archive_string_conv *sc)
425 r = archive_entry_gname_l(entry, name, length, sc);
427 if (errno == ENOMEM) {
428 archive_set_error(&a->archive, ENOMEM,
429 "Can't allocate memory for Gname");
430 return (ARCHIVE_FATAL);
432 return (ARCHIVE_WARN);
438 get_entry_symlink(struct archive_write *a, struct archive_entry *entry,
439 const char **name, size_t *length, struct archive_string_conv *sc)
443 r = archive_entry_symlink_l(entry, name, length, sc);
445 if (errno == ENOMEM) {
446 archive_set_error(&a->archive, ENOMEM,
447 "Can't allocate memory for Linkname");
448 return (ARCHIVE_FATAL);
450 return (ARCHIVE_WARN);
455 /* Add ACL to pax header */
457 add_pax_acl(struct archive_write *a,
458 struct archive_entry *entry, struct pax *pax, int flags)
464 acl_types = archive_entry_acl_types(entry);
466 if ((acl_types & ARCHIVE_ENTRY_ACL_TYPE_NFS4) != 0)
467 attr = "SCHILY.acl.ace";
468 else if ((flags & ARCHIVE_ENTRY_ACL_TYPE_ACCESS) != 0)
469 attr = "SCHILY.acl.access";
470 else if ((flags & ARCHIVE_ENTRY_ACL_TYPE_DEFAULT) != 0)
471 attr = "SCHILY.acl.default";
473 return (ARCHIVE_FATAL);
475 p = archive_entry_acl_to_text_l(entry, NULL, flags, pax->sconv_utf8);
477 if (errno == ENOMEM) {
478 archive_set_error(&a->archive, ENOMEM, "%s %s",
479 "Can't allocate memory for ", attr);
480 return (ARCHIVE_FATAL);
482 archive_set_error(&a->archive,
483 ARCHIVE_ERRNO_FILE_FORMAT, "%s %s %s",
484 "Can't translate ", attr, " to UTF-8");
485 return(ARCHIVE_WARN);
486 } else if (*p != '\0') {
487 add_pax_attr(&(pax->pax_header),
495 * TODO: Consider adding 'comment' and 'charset' fields to
496 * archive_entry so that clients can specify them. Also, consider
497 * adding generic key/value tags so clients can add arbitrary
500 * TODO: Break up this 700-line function!!!! Yowza!
503 archive_write_pax_header(struct archive_write *a,
504 struct archive_entry *entry_original)
506 struct archive_entry *entry_main;
509 int need_extension, r, ret;
512 uint64_t sparse_total, real_size;
514 const char *hardlink;
515 const char *path = NULL, *linkpath = NULL;
516 const char *uname = NULL, *gname = NULL;
517 const void *mac_metadata;
518 size_t mac_metadata_size;
519 struct archive_string_conv *sconv;
520 size_t hardlink_length, path_length, linkpath_length;
521 size_t uname_length, gname_length;
525 char ustar_entry_name[256];
526 char pax_entry_name[256];
527 char gnu_sparse_name[256];
528 struct archive_string entry_name;
532 pax = (struct pax *)a->format_data;
535 if (archive_entry_pathname(entry_original) == NULL) {
536 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
537 "Can't record entry in tar file without pathname");
538 return (ARCHIVE_FAILED);
542 * Choose a header encoding.
545 sconv = NULL;/* Binary mode. */
547 /* Header encoding is UTF-8. */
548 if (pax->sconv_utf8 == NULL) {
549 /* Initialize the string conversion object
551 pax->sconv_utf8 = archive_string_conversion_to_charset(
552 &(a->archive), "UTF-8", 1);
553 if (pax->sconv_utf8 == NULL)
554 /* Couldn't allocate memory */
555 return (ARCHIVE_FAILED);
557 sconv = pax->sconv_utf8;
560 r = get_entry_hardlink(a, entry_original, &hardlink,
561 &hardlink_length, sconv);
562 if (r == ARCHIVE_FATAL)
564 else if (r != ARCHIVE_OK) {
565 r = get_entry_hardlink(a, entry_original, &hardlink,
566 &hardlink_length, NULL);
567 if (r == ARCHIVE_FATAL)
569 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
570 "Can't translate linkname '%s' to %s", hardlink,
571 archive_string_conversion_charset_name(sconv));
573 sconv = NULL;/* The header charset switches to binary mode. */
576 /* Make sure this is a type of entry that we can handle here */
577 if (hardlink == NULL) {
578 switch (archive_entry_filetype(entry_original)) {
588 * Ensure a trailing '/'. Modify the original
589 * entry so the client sees the change.
591 #if defined(_WIN32) && !defined(__CYGWIN__)
594 wp = archive_entry_pathname_w(entry_original);
595 if (wp != NULL && wp[wcslen(wp) -1] != L'/') {
596 struct archive_wstring ws;
598 archive_string_init(&ws);
599 path_length = wcslen(wp);
600 if (archive_wstring_ensure(&ws,
601 path_length + 2) == NULL) {
602 archive_set_error(&a->archive, ENOMEM,
603 "Can't allocate pax data");
604 archive_wstring_free(&ws);
605 return(ARCHIVE_FATAL);
607 /* Should we keep '\' ? */
608 if (wp[path_length -1] == L'\\')
610 archive_wstrncpy(&ws, wp, path_length);
611 archive_wstrappend_wchar(&ws, L'/');
612 archive_entry_copy_pathname_w(
613 entry_original, ws.s);
614 archive_wstring_free(&ws);
618 p = archive_entry_pathname(entry_original);
620 * On Windows, this is a backup operation just in
621 * case getting WCS failed. On POSIX, this is a
624 if (p != NULL && p[strlen(p) - 1] != '/') {
625 struct archive_string as;
627 archive_string_init(&as);
628 path_length = strlen(p);
629 if (archive_string_ensure(&as,
630 path_length + 2) == NULL) {
631 archive_set_error(&a->archive, ENOMEM,
632 "Can't allocate pax data");
633 archive_string_free(&as);
634 return(ARCHIVE_FATAL);
636 #if defined(_WIN32) && !defined(__CYGWIN__)
637 /* NOTE: This might break the pathname
638 * if the current code page is CP932 and
639 * the pathname includes a character '\'
640 * as a part of its multibyte pathname. */
641 if (p[strlen(p) -1] == '\\')
645 archive_strncpy(&as, p, path_length);
646 archive_strappend_char(&as, '/');
647 archive_entry_copy_pathname(
648 entry_original, as.s);
649 archive_string_free(&as);
654 archive_set_error(&a->archive,
655 ARCHIVE_ERRNO_FILE_FORMAT,
656 "tar format cannot archive socket");
657 return (ARCHIVE_FAILED);
659 archive_set_error(&a->archive,
660 ARCHIVE_ERRNO_FILE_FORMAT,
661 "tar format cannot archive this (type=0%lo)",
663 archive_entry_filetype(entry_original));
664 return (ARCHIVE_FAILED);
669 * If Mac OS metadata blob is here, recurse to write that
670 * as a separate entry. This is really a pretty poor design:
671 * In particular, it doubles the overhead for long filenames.
672 * TODO: Help Apple folks design something better and figure
673 * out how to transition from this legacy format.
675 * Note that this code is present on every platform; clients
676 * on non-Mac are unlikely to ever provide this data, but
677 * applications that copy entries from one archive to another
678 * should not lose data just because the local filesystem
682 archive_entry_mac_metadata(entry_original, &mac_metadata_size);
683 if (mac_metadata != NULL) {
687 struct archive_entry *extra = archive_entry_new2(&a->archive);
689 oname = archive_entry_pathname(entry_original);
690 name_length = strlen(oname);
691 name = malloc(name_length + 3);
692 if (name == NULL || extra == NULL) {
693 /* XXX error message */
694 archive_entry_free(extra);
696 return (ARCHIVE_FAILED);
699 /* Find last '/'; strip trailing '/' characters */
700 bname = strrchr(name, '/');
701 while (bname != NULL && bname[1] == '\0') {
703 bname = strrchr(name, '/');
706 memmove(name + 2, name, name_length + 1);
707 memmove(name, "._", 2);
710 memmove(bname + 2, bname, strlen(bname) + 1);
711 memmove(bname, "._", 2);
713 archive_entry_copy_pathname(extra, name);
716 archive_entry_set_size(extra, mac_metadata_size);
717 archive_entry_set_filetype(extra, AE_IFREG);
718 archive_entry_set_perm(extra,
719 archive_entry_perm(entry_original));
720 archive_entry_set_mtime(extra,
721 archive_entry_mtime(entry_original),
722 archive_entry_mtime_nsec(entry_original));
723 archive_entry_set_gid(extra,
724 archive_entry_gid(entry_original));
725 archive_entry_set_gname(extra,
726 archive_entry_gname(entry_original));
727 archive_entry_set_uid(extra,
728 archive_entry_uid(entry_original));
729 archive_entry_set_uname(extra,
730 archive_entry_uname(entry_original));
732 /* Recurse to write the special copyfile entry. */
733 r = archive_write_pax_header(a, extra);
734 archive_entry_free(extra);
735 if (r < ARCHIVE_WARN)
739 r = (int)archive_write_pax_data(a, mac_metadata,
741 if (r < ARCHIVE_WARN)
745 r = archive_write_pax_finish_entry(a);
746 if (r < ARCHIVE_WARN)
752 /* Copy entry so we can modify it as needed. */
753 #if defined(_WIN32) && !defined(__CYGWIN__)
754 /* Make sure the path separators in pathname, hardlink and symlink
755 * are all slash '/', not the Windows path separator '\'. */
756 entry_main = __la_win_entry_in_posix_pathseparator(entry_original);
757 if (entry_main == entry_original)
758 entry_main = archive_entry_clone(entry_original);
760 entry_main = archive_entry_clone(entry_original);
762 if (entry_main == NULL) {
763 archive_set_error(&a->archive, ENOMEM,
764 "Can't allocate pax data");
765 return(ARCHIVE_FATAL);
767 archive_string_empty(&(pax->pax_header)); /* Blank our work area. */
768 archive_string_empty(&(pax->sparse_map));
770 sparse_list_clear(pax);
772 if (hardlink == NULL &&
773 archive_entry_filetype(entry_main) == AE_IFREG)
774 sparse_count = archive_entry_sparse_reset(entry_main);
778 int64_t offset, length, last_offset = 0;
779 /* Get the last entry of sparse block. */
780 while (archive_entry_sparse_next(
781 entry_main, &offset, &length) == ARCHIVE_OK)
782 last_offset = offset + length;
784 /* If the last sparse block does not reach the end of file,
785 * We have to add a empty sparse block as the last entry to
786 * manage storing file data. */
787 if (last_offset < archive_entry_size(entry_main))
788 archive_entry_sparse_add_entry(entry_main,
789 archive_entry_size(entry_main), 0);
790 sparse_count = archive_entry_sparse_reset(entry_main);
794 * First, check the name fields and see if any of them
795 * require binary coding. If any of them does, then all of
798 r = get_entry_pathname(a, entry_main, &path, &path_length, sconv);
799 if (r == ARCHIVE_FATAL)
801 else if (r != ARCHIVE_OK) {
802 r = get_entry_pathname(a, entry_main, &path,
804 if (r == ARCHIVE_FATAL)
806 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
807 "Can't translate pathname '%s' to %s", path,
808 archive_string_conversion_charset_name(sconv));
810 sconv = NULL;/* The header charset switches to binary mode. */
812 r = get_entry_uname(a, entry_main, &uname, &uname_length, sconv);
813 if (r == ARCHIVE_FATAL)
815 else if (r != ARCHIVE_OK) {
816 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL);
817 if (r == ARCHIVE_FATAL)
819 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
820 "Can't translate uname '%s' to %s", uname,
821 archive_string_conversion_charset_name(sconv));
823 sconv = NULL;/* The header charset switches to binary mode. */
825 r = get_entry_gname(a, entry_main, &gname, &gname_length, sconv);
826 if (r == ARCHIVE_FATAL)
828 else if (r != ARCHIVE_OK) {
829 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL);
830 if (r == ARCHIVE_FATAL)
832 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
833 "Can't translate gname '%s' to %s", gname,
834 archive_string_conversion_charset_name(sconv));
836 sconv = NULL;/* The header charset switches to binary mode. */
839 linkpath_length = hardlink_length;
840 if (linkpath == NULL) {
841 r = get_entry_symlink(a, entry_main, &linkpath,
842 &linkpath_length, sconv);
843 if (r == ARCHIVE_FATAL)
845 else if (r != ARCHIVE_OK) {
846 r = get_entry_symlink(a, entry_main, &linkpath,
847 &linkpath_length, NULL);
848 if (r == ARCHIVE_FATAL)
850 archive_set_error(&a->archive,
851 ARCHIVE_ERRNO_FILE_FORMAT,
852 "Can't translate linkname '%s' to %s", linkpath,
853 archive_string_conversion_charset_name(sconv));
859 /* If any string conversions failed, get all attributes
861 if (sconv == NULL && !pax->opt_binary) {
862 if (hardlink != NULL) {
863 r = get_entry_hardlink(a, entry_main, &hardlink,
864 &hardlink_length, NULL);
865 if (r == ARCHIVE_FATAL)
868 linkpath_length = hardlink_length;
870 r = get_entry_pathname(a, entry_main, &path,
872 if (r == ARCHIVE_FATAL)
874 r = get_entry_uname(a, entry_main, &uname, &uname_length, NULL);
875 if (r == ARCHIVE_FATAL)
877 r = get_entry_gname(a, entry_main, &gname, &gname_length, NULL);
878 if (r == ARCHIVE_FATAL)
882 /* Store the header encoding first, to be nice to readers. */
884 add_pax_attr(&(pax->pax_header), "hdrcharset", "BINARY");
888 * If name is too long, or has non-ASCII characters, add
889 * 'path' to pax extended attrs. (Note that an unconvertible
890 * name must have non-ASCII characters.)
892 if (has_non_ASCII(path)) {
893 /* We have non-ASCII characters. */
894 add_pax_attr(&(pax->pax_header), "path", path);
895 archive_entry_set_pathname(entry_main,
896 build_ustar_entry_name(ustar_entry_name,
897 path, path_length, NULL));
900 /* We have an all-ASCII path; we'd like to just store
901 * it in the ustar header if it will fit. Yes, this
902 * duplicates some of the logic in
903 * archive_write_set_format_ustar.c
905 if (path_length <= 100) {
906 /* Fits in the old 100-char tar name field. */
908 /* Find largest suffix that will fit. */
909 /* Note: strlen() > 100, so strlen() - 100 - 1 >= 0 */
910 suffix = strchr(path + path_length - 100 - 1, '/');
911 /* Don't attempt an empty prefix. */
913 suffix = strchr(suffix + 1, '/');
914 /* We can put it in the ustar header if it's
915 * all ASCII and it's either <= 100 characters
916 * or can be split at a '/' into a prefix <=
917 * 155 chars and a suffix <= 100 chars. (Note
918 * the strchr() above will return NULL exactly
919 * when the path can't be split.)
921 if (suffix == NULL /* Suffix > 100 chars. */
922 || suffix[1] == '\0' /* empty suffix */
923 || suffix - path > 155) /* Prefix > 155 chars */
925 add_pax_attr(&(pax->pax_header), "path", path);
926 archive_entry_set_pathname(entry_main,
927 build_ustar_entry_name(ustar_entry_name,
928 path, path_length, NULL));
934 if (linkpath != NULL) {
935 /* If link name is too long or has non-ASCII characters, add
936 * 'linkpath' to pax extended attrs. */
937 if (linkpath_length > 100 || has_non_ASCII(linkpath)) {
938 add_pax_attr(&(pax->pax_header), "linkpath", linkpath);
939 if (linkpath_length > 100) {
940 if (hardlink != NULL)
941 archive_entry_set_hardlink(entry_main,
942 "././@LongHardLink");
944 archive_entry_set_symlink(entry_main,
950 /* Save a pathname since it will be renamed if `entry_main` has
952 archive_string_init(&entry_name);
953 archive_strcpy(&entry_name, archive_entry_pathname(entry_main));
955 /* If file size is too large, add 'size' to pax extended attrs. */
956 if (archive_entry_size(entry_main) >= (((int64_t)1) << 33)) {
957 add_pax_attr_int(&(pax->pax_header), "size",
958 archive_entry_size(entry_main));
962 /* If numeric GID is too large, add 'gid' to pax extended attrs. */
963 if ((unsigned int)archive_entry_gid(entry_main) >= (1 << 18)) {
964 add_pax_attr_int(&(pax->pax_header), "gid",
965 archive_entry_gid(entry_main));
969 /* If group name is too large or has non-ASCII characters, add
970 * 'gname' to pax extended attrs. */
972 if (gname_length > 31 || has_non_ASCII(gname)) {
973 add_pax_attr(&(pax->pax_header), "gname", gname);
978 /* If numeric UID is too large, add 'uid' to pax extended attrs. */
979 if ((unsigned int)archive_entry_uid(entry_main) >= (1 << 18)) {
980 add_pax_attr_int(&(pax->pax_header), "uid",
981 archive_entry_uid(entry_main));
985 /* Add 'uname' to pax extended attrs if necessary. */
987 if (uname_length > 31 || has_non_ASCII(uname)) {
988 add_pax_attr(&(pax->pax_header), "uname", uname);
994 * POSIX/SUSv3 doesn't provide a standard key for large device
995 * numbers. I use the same keys here that Joerg Schilling
996 * used for 'star.' (Which, somewhat confusingly, are called
997 * "devXXX" even though they code "rdev" values.) No doubt,
998 * other implementations use other keys. Note that there's no
999 * reason we can't write the same information into a number of
1002 * Of course, this is only needed for block or char device entries.
1004 if (archive_entry_filetype(entry_main) == AE_IFBLK
1005 || archive_entry_filetype(entry_main) == AE_IFCHR) {
1007 * If rdevmajor is too large, add 'SCHILY.devmajor' to
1008 * extended attributes.
1010 int rdevmajor, rdevminor;
1011 rdevmajor = archive_entry_rdevmajor(entry_main);
1012 rdevminor = archive_entry_rdevminor(entry_main);
1013 if (rdevmajor >= (1 << 18)) {
1014 add_pax_attr_int(&(pax->pax_header), "SCHILY.devmajor",
1017 * Non-strict formatting below means we don't
1018 * have to truncate here. Not truncating improves
1019 * the chance that some more modern tar archivers
1020 * (such as GNU tar 1.13) can restore the full
1021 * value even if they don't understand the pax
1022 * extended attributes. See my rant below about
1023 * file size fields for additional details.
1025 /* archive_entry_set_rdevmajor(entry_main,
1026 rdevmajor & ((1 << 18) - 1)); */
1031 * If devminor is too large, add 'SCHILY.devminor' to
1032 * extended attributes.
1034 if (rdevminor >= (1 << 18)) {
1035 add_pax_attr_int(&(pax->pax_header), "SCHILY.devminor",
1037 /* Truncation is not necessary here, either. */
1038 /* archive_entry_set_rdevminor(entry_main,
1039 rdevminor & ((1 << 18) - 1)); */
1045 * Technically, the mtime field in the ustar header can
1046 * support 33 bits, but many platforms use signed 32-bit time
1047 * values. The cutoff of 0x7fffffff here is a compromise.
1048 * Yes, this check is duplicated just below; this helps to
1049 * avoid writing an mtime attribute just to handle a
1050 * high-resolution timestamp in "restricted pax" mode.
1052 if (!need_extension &&
1053 ((archive_entry_mtime(entry_main) < 0)
1054 || (archive_entry_mtime(entry_main) >= 0x7fffffff)))
1057 /* I use a star-compatible file flag attribute. */
1058 p = archive_entry_fflags_text(entry_main);
1059 if (!need_extension && p != NULL && *p != '\0')
1062 /* If there are extended attributes, we need an extension */
1063 if (!need_extension && archive_entry_xattr_count(entry_original) > 0)
1066 /* If there are sparse info, we need an extension */
1067 if (!need_extension && sparse_count > 0)
1070 acl_types = archive_entry_acl_types(entry_original);
1072 /* If there are any ACL entries, we need an extension */
1073 if (!need_extension && acl_types != 0)
1077 * Libarchive used to include these in extended headers for
1078 * restricted pax format, but that confused people who
1079 * expected ustar-like time semantics. So now we only include
1080 * them in full pax format.
1082 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_RESTRICTED) {
1083 if (archive_entry_ctime(entry_main) != 0 ||
1084 archive_entry_ctime_nsec(entry_main) != 0)
1085 add_pax_attr_time(&(pax->pax_header), "ctime",
1086 archive_entry_ctime(entry_main),
1087 archive_entry_ctime_nsec(entry_main));
1089 if (archive_entry_atime(entry_main) != 0 ||
1090 archive_entry_atime_nsec(entry_main) != 0)
1091 add_pax_attr_time(&(pax->pax_header), "atime",
1092 archive_entry_atime(entry_main),
1093 archive_entry_atime_nsec(entry_main));
1095 /* Store birth/creationtime only if it's earlier than mtime */
1096 if (archive_entry_birthtime_is_set(entry_main) &&
1097 archive_entry_birthtime(entry_main)
1098 < archive_entry_mtime(entry_main))
1099 add_pax_attr_time(&(pax->pax_header),
1100 "LIBARCHIVE.creationtime",
1101 archive_entry_birthtime(entry_main),
1102 archive_entry_birthtime_nsec(entry_main));
1106 * The following items are handled differently in "pax
1107 * restricted" format. In particular, in "pax restricted"
1108 * format they won't be added unless need_extension is
1109 * already set (we're already generating an extended header, so
1110 * may as well include these).
1112 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_RESTRICTED ||
1114 if (archive_entry_mtime(entry_main) < 0 ||
1115 archive_entry_mtime(entry_main) >= 0x7fffffff ||
1116 archive_entry_mtime_nsec(entry_main) != 0)
1117 add_pax_attr_time(&(pax->pax_header), "mtime",
1118 archive_entry_mtime(entry_main),
1119 archive_entry_mtime_nsec(entry_main));
1121 /* I use a star-compatible file flag attribute. */
1122 p = archive_entry_fflags_text(entry_main);
1123 if (p != NULL && *p != '\0')
1124 add_pax_attr(&(pax->pax_header), "SCHILY.fflags", p);
1126 /* I use star-compatible ACL attributes. */
1127 if ((acl_types & ARCHIVE_ENTRY_ACL_TYPE_NFS4) != 0) {
1128 ret = add_pax_acl(a, entry_original, pax,
1129 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID |
1130 ARCHIVE_ENTRY_ACL_STYLE_SEPARATOR_COMMA);
1131 if (ret == ARCHIVE_FATAL)
1132 return (ARCHIVE_FATAL);
1134 if (acl_types & ARCHIVE_ENTRY_ACL_TYPE_ACCESS) {
1135 ret = add_pax_acl(a, entry_original, pax,
1136 ARCHIVE_ENTRY_ACL_TYPE_ACCESS |
1137 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID |
1138 ARCHIVE_ENTRY_ACL_STYLE_SEPARATOR_COMMA);
1139 if (ret == ARCHIVE_FATAL)
1140 return (ARCHIVE_FATAL);
1142 if (acl_types & ARCHIVE_ENTRY_ACL_TYPE_DEFAULT) {
1143 ret = add_pax_acl(a, entry_original, pax,
1144 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT |
1145 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID |
1146 ARCHIVE_ENTRY_ACL_STYLE_SEPARATOR_COMMA);
1147 if (ret == ARCHIVE_FATAL)
1148 return (ARCHIVE_FATAL);
1151 /* We use GNU-tar-compatible sparse attributes. */
1152 if (sparse_count > 0) {
1153 int64_t soffset, slength;
1155 add_pax_attr_int(&(pax->pax_header),
1156 "GNU.sparse.major", 1);
1157 add_pax_attr_int(&(pax->pax_header),
1158 "GNU.sparse.minor", 0);
1159 add_pax_attr(&(pax->pax_header),
1160 "GNU.sparse.name", entry_name.s);
1161 add_pax_attr_int(&(pax->pax_header),
1162 "GNU.sparse.realsize",
1163 archive_entry_size(entry_main));
1165 /* Rename the file name which will be used for
1166 * ustar header to a special name, which GNU
1167 * PAX Format 1.0 requires */
1168 archive_entry_set_pathname(entry_main,
1169 build_gnu_sparse_name(gnu_sparse_name,
1173 * - Make a sparse map, which will precede a file data.
1174 * - Get the total size of available data of sparse.
1176 archive_string_sprintf(&(pax->sparse_map), "%d\n",
1178 while (archive_entry_sparse_next(entry_main,
1179 &soffset, &slength) == ARCHIVE_OK) {
1180 archive_string_sprintf(&(pax->sparse_map),
1184 sparse_total += slength;
1185 if (sparse_list_add(pax, soffset, slength)
1187 archive_set_error(&a->archive,
1189 "Can't allocate memory");
1190 archive_entry_free(entry_main);
1191 archive_string_free(&entry_name);
1192 return (ARCHIVE_FATAL);
1197 /* Store extended attributes */
1198 if (archive_write_pax_header_xattrs(a, pax, entry_original)
1200 archive_entry_free(entry_main);
1201 archive_string_free(&entry_name);
1202 return (ARCHIVE_FATAL);
1206 /* Only regular files have data. */
1207 if (archive_entry_filetype(entry_main) != AE_IFREG)
1208 archive_entry_set_size(entry_main, 0);
1211 * Pax-restricted does not store data for hardlinks, in order
1212 * to improve compatibility with ustar.
1214 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE &&
1216 archive_entry_set_size(entry_main, 0);
1219 * XXX Full pax interchange format does permit a hardlink
1220 * entry to have data associated with it. I'm not supporting
1221 * that here because the client expects me to tell them whether
1222 * or not this format expects data for hardlinks. If I
1223 * don't check here, then every pax archive will end up with
1224 * duplicated data for hardlinks. Someday, there may be
1225 * need to select this behavior, in which case the following
1226 * will need to be revisited. XXX
1228 if (hardlink != NULL)
1229 archive_entry_set_size(entry_main, 0);
1231 /* Save a real file size. */
1232 real_size = archive_entry_size(entry_main);
1234 * Overwrite a file size by the total size of sparse blocks and
1235 * the size of sparse map info. That file size is the length of
1236 * the data, which we will exactly store into an archive file.
1238 if (archive_strlen(&(pax->sparse_map))) {
1239 size_t mapsize = archive_strlen(&(pax->sparse_map));
1240 pax->sparse_map_padding = 0x1ff & (-(ssize_t)mapsize);
1241 archive_entry_set_size(entry_main,
1242 mapsize + pax->sparse_map_padding + sparse_total);
1245 /* Format 'ustar' header for main entry.
1247 * The trouble with file size: If the reader can't understand
1248 * the file size, they may not be able to locate the next
1249 * entry and the rest of the archive is toast. Pax-compliant
1250 * readers are supposed to ignore the file size in the main
1251 * header, so the question becomes how to maximize portability
1252 * for readers that don't support pax attribute extensions.
1253 * For maximum compatibility, I permit numeric extensions in
1254 * the main header so that the file size stored will always be
1255 * correct, even if it's in a format that only some
1256 * implementations understand. The technique used here is:
1258 * a) If possible, follow the standard exactly. This handles
1259 * files up to 8 gigabytes minus 1.
1261 * b) If that fails, try octal but omit the field terminator.
1262 * That handles files up to 64 gigabytes minus 1.
1264 * c) Otherwise, use base-256 extensions. That handles files
1265 * up to 2^63 in this implementation, with the potential to
1266 * go up to 2^94. That should hold us for a while. ;-)
1268 * The non-strict formatter uses similar logic for other
1269 * numeric fields, though they're less critical.
1271 if (__archive_write_format_header_ustar(a, ustarbuff, entry_main, -1, 0,
1272 NULL) == ARCHIVE_FATAL)
1273 return (ARCHIVE_FATAL);
1275 /* If we built any extended attributes, write that entry first. */
1276 if (archive_strlen(&(pax->pax_header)) > 0) {
1277 struct archive_entry *pax_attr_entry;
1282 pax_attr_entry = archive_entry_new2(&a->archive);
1284 archive_entry_set_pathname(pax_attr_entry,
1285 build_pax_attribute_name(pax_entry_name, p));
1286 archive_entry_set_size(pax_attr_entry,
1287 archive_strlen(&(pax->pax_header)));
1288 /* Copy uid/gid (but clip to ustar limits). */
1289 uid = archive_entry_uid(entry_main);
1291 uid = (1 << 18) - 1;
1292 archive_entry_set_uid(pax_attr_entry, uid);
1293 gid = archive_entry_gid(entry_main);
1295 gid = (1 << 18) - 1;
1296 archive_entry_set_gid(pax_attr_entry, gid);
1297 /* Copy mode over (but not setuid/setgid bits) */
1298 mode = archive_entry_mode(entry_main);
1308 archive_entry_set_mode(pax_attr_entry, mode);
1310 /* Copy uname/gname. */
1311 archive_entry_set_uname(pax_attr_entry,
1312 archive_entry_uname(entry_main));
1313 archive_entry_set_gname(pax_attr_entry,
1314 archive_entry_gname(entry_main));
1316 /* Copy mtime, but clip to ustar limits. */
1317 s = archive_entry_mtime(entry_main);
1318 if (s < 0) { s = 0; }
1319 if (s >= 0x7fffffff) { s = 0x7fffffff; }
1320 archive_entry_set_mtime(pax_attr_entry, s, 0);
1322 /* Standard ustar doesn't support atime. */
1323 archive_entry_set_atime(pax_attr_entry, 0, 0);
1325 /* Standard ustar doesn't support ctime. */
1326 archive_entry_set_ctime(pax_attr_entry, 0, 0);
1328 r = __archive_write_format_header_ustar(a, paxbuff,
1329 pax_attr_entry, 'x', 1, NULL);
1331 archive_entry_free(pax_attr_entry);
1333 /* Note that the 'x' header shouldn't ever fail to format */
1334 if (r < ARCHIVE_WARN) {
1335 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1336 "archive_write_pax_header: "
1337 "'x' header failed?! This can't happen.\n");
1338 return (ARCHIVE_FATAL);
1341 r = __archive_write_output(a, paxbuff, 512);
1342 if (r != ARCHIVE_OK) {
1343 sparse_list_clear(pax);
1344 pax->entry_bytes_remaining = 0;
1345 pax->entry_padding = 0;
1346 return (ARCHIVE_FATAL);
1349 pax->entry_bytes_remaining = archive_strlen(&(pax->pax_header));
1350 pax->entry_padding =
1351 0x1ff & (-(int64_t)pax->entry_bytes_remaining);
1353 r = __archive_write_output(a, pax->pax_header.s,
1354 archive_strlen(&(pax->pax_header)));
1355 if (r != ARCHIVE_OK) {
1356 /* If a write fails, we're pretty much toast. */
1357 return (ARCHIVE_FATAL);
1359 /* Pad out the end of the entry. */
1360 r = __archive_write_nulls(a, (size_t)pax->entry_padding);
1361 if (r != ARCHIVE_OK) {
1362 /* If a write fails, we're pretty much toast. */
1363 return (ARCHIVE_FATAL);
1365 pax->entry_bytes_remaining = pax->entry_padding = 0;
1368 /* Write the header for main entry. */
1369 r = __archive_write_output(a, ustarbuff, 512);
1370 if (r != ARCHIVE_OK)
1374 * Inform the client of the on-disk size we're using, so
1375 * they can avoid unnecessarily writing a body for something
1376 * that we're just going to ignore.
1378 archive_entry_set_size(entry_original, real_size);
1379 if (pax->sparse_list == NULL && real_size > 0) {
1380 /* This is not a sparse file but we handle its data as
1381 * a sparse block. */
1382 sparse_list_add(pax, 0, real_size);
1383 sparse_total = real_size;
1385 pax->entry_padding = 0x1ff & (-(int64_t)sparse_total);
1386 archive_entry_free(entry_main);
1387 archive_string_free(&entry_name);
1393 * We need a valid name for the regular 'ustar' entry. This routine
1394 * tries to hack something more-or-less reasonable.
1396 * The approach here tries to preserve leading dir names. We do so by
1397 * working with four sections:
1398 * 1) "prefix" directory names,
1399 * 2) "suffix" directory names,
1400 * 3) inserted dir name (optional),
1403 * These sections must satisfy the following requirements:
1404 * * Parts 1 & 2 together form an initial portion of the dir name.
1405 * * Part 3 is specified by the caller. (It should not contain a leading
1407 * * Part 4 forms an initial portion of the base filename.
1408 * * The filename must be <= 99 chars to fit the ustar 'name' field.
1409 * * Parts 2, 3, 4 together must be <= 99 chars to fit the ustar 'name' fld.
1410 * * Part 1 must be <= 155 chars to fit the ustar 'prefix' field.
1411 * * If the original name ends in a '/', the new name must also end in a '/'
1412 * * Trailing '/.' sequences may be stripped.
1414 * Note: Recall that the ustar format does not store the '/' separating
1415 * parts 1 & 2, but does store the '/' separating parts 2 & 3.
1418 build_ustar_entry_name(char *dest, const char *src, size_t src_length,
1421 const char *prefix, *prefix_end;
1422 const char *suffix, *suffix_end;
1423 const char *filename, *filename_end;
1425 int need_slash = 0; /* Was there a trailing slash? */
1426 size_t suffix_length = 99;
1427 size_t insert_length;
1429 /* Length of additional dir element to be added. */
1433 /* +2 here allows for '/' before and after the insert. */
1434 insert_length = strlen(insert) + 2;
1436 /* Step 0: Quick bailout in a common case. */
1437 if (src_length < 100 && insert == NULL) {
1438 strncpy(dest, src, src_length);
1439 dest[src_length] = '\0';
1443 /* Step 1: Locate filename and enforce the length restriction. */
1444 filename_end = src + src_length;
1445 /* Remove trailing '/' chars and '/.' pairs. */
1447 if (filename_end > src && filename_end[-1] == '/') {
1449 need_slash = 1; /* Remember to restore trailing '/'. */
1452 if (filename_end > src + 1 && filename_end[-1] == '.'
1453 && filename_end[-2] == '/') {
1455 need_slash = 1; /* "foo/." will become "foo/" */
1462 /* Find start of filename. */
1463 filename = filename_end - 1;
1464 while ((filename > src) && (*filename != '/'))
1466 if ((*filename == '/') && (filename < filename_end - 1))
1468 /* Adjust filename_end so that filename + insert fits in 99 chars. */
1469 suffix_length -= insert_length;
1470 if (filename_end > filename + suffix_length)
1471 filename_end = filename + suffix_length;
1472 /* Calculate max size for "suffix" section (#3 above). */
1473 suffix_length -= filename_end - filename;
1475 /* Step 2: Locate the "prefix" section of the dirname, including
1478 prefix_end = prefix + 155;
1479 if (prefix_end > filename)
1480 prefix_end = filename;
1481 while (prefix_end > prefix && *prefix_end != '/')
1483 if ((prefix_end < filename) && (*prefix_end == '/'))
1486 /* Step 3: Locate the "suffix" section of the dirname,
1487 * including trailing '/'. */
1488 suffix = prefix_end;
1489 suffix_end = suffix + suffix_length; /* Enforce limit. */
1490 if (suffix_end > filename)
1491 suffix_end = filename;
1492 if (suffix_end < suffix)
1493 suffix_end = suffix;
1494 while (suffix_end > suffix && *suffix_end != '/')
1496 if ((suffix_end < filename) && (*suffix_end == '/'))
1499 /* Step 4: Build the new name. */
1500 /* The OpenBSD strlcpy function is safer, but less portable. */
1501 /* Rather than maintain two versions, just use the strncpy version. */
1503 if (prefix_end > prefix) {
1504 strncpy(p, prefix, prefix_end - prefix);
1505 p += prefix_end - prefix;
1507 if (suffix_end > suffix) {
1508 strncpy(p, suffix, suffix_end - suffix);
1509 p += suffix_end - suffix;
1511 if (insert != NULL) {
1512 /* Note: assume insert does not have leading or trailing '/' */
1514 p += strlen(insert);
1517 strncpy(p, filename, filename_end - filename);
1518 p += filename_end - filename;
1527 * The ustar header for the pax extended attributes must have a
1528 * reasonable name: SUSv3 requires 'dirname'/PaxHeader.'pid'/'filename'
1529 * where 'pid' is the PID of the archiving process. Unfortunately,
1530 * that makes testing a pain since the output varies for each run,
1531 * so I'm sticking with the simpler 'dirname'/PaxHeader/'filename'
1532 * for now. (Someday, I'll make this settable. Then I can use the
1533 * SUS recommendation as default and test harnesses can override it
1534 * to get predictable results.)
1536 * Joerg Schilling has argued that this is unnecessary because, in
1537 * practice, if the pax extended attributes get extracted as regular
1538 * files, no one is going to bother reading those attributes to
1539 * manually restore them. Based on this, 'star' uses
1540 * /tmp/PaxHeader/'basename' as the ustar header name. This is a
1541 * tempting argument, in part because it's simpler than the SUSv3
1542 * recommendation, but I'm not entirely convinced. I'm also
1543 * uncomfortable with the fact that "/tmp" is a Unix-ism.
1545 * The following routine leverages build_ustar_entry_name() above and
1546 * so is simpler than you might think. It just needs to provide the
1547 * additional path element and handle a few pathological cases).
1550 build_pax_attribute_name(char *dest, const char *src)
1555 /* Handle the null filename case. */
1556 if (src == NULL || *src == '\0') {
1557 strcpy(dest, "PaxHeader/blank");
1561 /* Prune final '/' and other unwanted final elements. */
1562 p = src + strlen(src);
1564 /* Ends in "/", remove the '/' */
1565 if (p > src && p[-1] == '/') {
1569 /* Ends in "/.", remove the '.' */
1570 if (p > src + 1 && p[-1] == '.'
1578 /* Pathological case: After above, there was nothing left.
1579 * This includes "/." "/./." "/.//./." etc. */
1581 strcpy(dest, "/PaxHeader/rootdir");
1585 /* Convert unadorned "." into a suitable filename. */
1586 if (*src == '.' && p == src + 1) {
1587 strcpy(dest, "PaxHeader/currentdir");
1592 * TODO: Push this string into the 'pax' structure to avoid
1593 * recomputing it every time. That will also open the door
1594 * to having clients override it.
1596 #if HAVE_GETPID && 0 /* Disable this for now; see above comment. */
1597 sprintf(buff, "PaxHeader.%d", getpid());
1599 /* If the platform can't fetch the pid, don't include it. */
1600 strcpy(buff, "PaxHeader");
1602 /* General case: build a ustar-compatible name adding
1604 build_ustar_entry_name(dest, src, p - src, buff);
1610 * GNU PAX Format 1.0 requires the special name, which pattern is:
1611 * <dir>/GNUSparseFile.<pid>/<original file name>
1613 * This function is used for only Sparse file, a file type of which
1617 build_gnu_sparse_name(char *dest, const char *src)
1622 /* Handle the null filename case. */
1623 if (src == NULL || *src == '\0') {
1624 strcpy(dest, "GNUSparseFile/blank");
1628 /* Prune final '/' and other unwanted final elements. */
1629 p = src + strlen(src);
1631 /* Ends in "/", remove the '/' */
1632 if (p > src && p[-1] == '/') {
1636 /* Ends in "/.", remove the '.' */
1637 if (p > src + 1 && p[-1] == '.'
1645 #if HAVE_GETPID && 0 /* Disable this as pax attribute name. */
1646 sprintf(buff, "GNUSparseFile.%d", getpid());
1648 /* If the platform can't fetch the pid, don't include it. */
1649 strcpy(buff, "GNUSparseFile");
1651 /* General case: build a ustar-compatible name adding
1652 * "/GNUSparseFile/". */
1653 build_ustar_entry_name(dest, src, p - src, buff);
1658 /* Write two null blocks for the end of archive */
1660 archive_write_pax_close(struct archive_write *a)
1662 return (__archive_write_nulls(a, 512 * 2));
1666 archive_write_pax_free(struct archive_write *a)
1670 pax = (struct pax *)a->format_data;
1672 return (ARCHIVE_OK);
1674 archive_string_free(&pax->pax_header);
1675 archive_string_free(&pax->sparse_map);
1676 archive_string_free(&pax->l_url_encoded_name);
1677 sparse_list_clear(pax);
1679 a->format_data = NULL;
1680 return (ARCHIVE_OK);
1684 archive_write_pax_finish_entry(struct archive_write *a)
1690 pax = (struct pax *)a->format_data;
1691 remaining = pax->entry_bytes_remaining;
1692 if (remaining == 0) {
1693 while (pax->sparse_list) {
1694 struct sparse_block *sb;
1695 if (!pax->sparse_list->is_hole)
1696 remaining += pax->sparse_list->remaining;
1697 sb = pax->sparse_list->next;
1698 free(pax->sparse_list);
1699 pax->sparse_list = sb;
1702 ret = __archive_write_nulls(a, (size_t)(remaining + pax->entry_padding));
1703 pax->entry_bytes_remaining = pax->entry_padding = 0;
1708 archive_write_pax_data(struct archive_write *a, const void *buff, size_t s)
1715 pax = (struct pax *)a->format_data;
1718 * According to GNU PAX format 1.0, write a sparse map
1721 if (archive_strlen(&(pax->sparse_map))) {
1722 ret = __archive_write_output(a, pax->sparse_map.s,
1723 archive_strlen(&(pax->sparse_map)));
1724 if (ret != ARCHIVE_OK)
1726 ret = __archive_write_nulls(a, pax->sparse_map_padding);
1727 if (ret != ARCHIVE_OK)
1729 archive_string_empty(&(pax->sparse_map));
1734 const unsigned char *p;
1736 while (pax->sparse_list != NULL &&
1737 pax->sparse_list->remaining == 0) {
1738 struct sparse_block *sb = pax->sparse_list->next;
1739 free(pax->sparse_list);
1740 pax->sparse_list = sb;
1743 if (pax->sparse_list == NULL)
1746 p = ((const unsigned char *)buff) + total;
1748 if (ws > pax->sparse_list->remaining)
1749 ws = (size_t)pax->sparse_list->remaining;
1751 if (pax->sparse_list->is_hole) {
1752 /* Current block is hole thus we do not write
1754 pax->sparse_list->remaining -= ws;
1759 ret = __archive_write_output(a, p, ws);
1760 pax->sparse_list->remaining -= ws;
1762 if (ret != ARCHIVE_OK)
1769 has_non_ASCII(const char *_p)
1771 const unsigned char *p = (const unsigned char *)_p;
1775 while (*p != '\0' && *p < 128)
1777 return (*p != '\0');
1781 * Used by extended attribute support; encodes the name
1782 * so that there will be no '=' characters in the result.
1785 url_encode(const char *in)
1792 for (s = in; *s != '\0'; s++) {
1793 if (*s < 33 || *s > 126 || *s == '%' || *s == '=')
1799 out = (char *)malloc(out_len + 1);
1803 for (s = in, d = out; *s != '\0'; s++) {
1804 /* encode any non-printable ASCII character or '%' or '=' */
1805 if (*s < 33 || *s > 126 || *s == '%' || *s == '=') {
1806 /* URL encoding is '%' followed by two hex digits */
1808 *d++ = "0123456789ABCDEF"[0x0f & (*s >> 4)];
1809 *d++ = "0123456789ABCDEF"[0x0f & *s];
1819 * Encode a sequence of bytes into a C string using base-64 encoding.
1821 * Returns a null-terminated C string allocated with malloc(); caller
1822 * is responsible for freeing the result.
1825 base64_encode(const char *s, size_t len)
1827 static const char digits[64] =
1828 { 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O',
1829 'P','Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d',
1830 'e','f','g','h','i','j','k','l','m','n','o','p','q','r','s',
1831 't','u','v','w','x','y','z','0','1','2','3','4','5','6','7',
1836 /* 3 bytes becomes 4 chars, but round up and allow for trailing NUL */
1837 out = (char *)malloc((len * 4 + 2) / 3 + 1);
1842 /* Convert each group of 3 bytes into 4 characters. */
1844 v = (((int)s[0] << 16) & 0xff0000)
1845 | (((int)s[1] << 8) & 0xff00)
1846 | (((int)s[2]) & 0x00ff);
1849 *d++ = digits[(v >> 18) & 0x3f];
1850 *d++ = digits[(v >> 12) & 0x3f];
1851 *d++ = digits[(v >> 6) & 0x3f];
1852 *d++ = digits[(v) & 0x3f];
1854 /* Handle final group of 1 byte (2 chars) or 2 bytes (3 chars). */
1858 v = (((int)s[0] << 16) & 0xff0000);
1859 *d++ = digits[(v >> 18) & 0x3f];
1860 *d++ = digits[(v >> 12) & 0x3f];
1863 v = (((int)s[0] << 16) & 0xff0000)
1864 | (((int)s[1] << 8) & 0xff00);
1865 *d++ = digits[(v >> 18) & 0x3f];
1866 *d++ = digits[(v >> 12) & 0x3f];
1867 *d++ = digits[(v >> 6) & 0x3f];
1870 /* Add trailing NUL character so output is a valid C string. */
1876 sparse_list_clear(struct pax *pax)
1878 while (pax->sparse_list != NULL) {
1879 struct sparse_block *sb = pax->sparse_list;
1880 pax->sparse_list = sb->next;
1883 pax->sparse_tail = NULL;
1887 _sparse_list_add_block(struct pax *pax, int64_t offset, int64_t length,
1890 struct sparse_block *sb;
1892 sb = (struct sparse_block *)malloc(sizeof(*sb));
1894 return (ARCHIVE_FATAL);
1896 sb->is_hole = is_hole;
1897 sb->offset = offset;
1898 sb->remaining = length;
1899 if (pax->sparse_list == NULL || pax->sparse_tail == NULL)
1900 pax->sparse_list = pax->sparse_tail = sb;
1902 pax->sparse_tail->next = sb;
1903 pax->sparse_tail = sb;
1905 return (ARCHIVE_OK);
1909 sparse_list_add(struct pax *pax, int64_t offset, int64_t length)
1911 int64_t last_offset;
1914 if (pax->sparse_tail == NULL)
1917 last_offset = pax->sparse_tail->offset +
1918 pax->sparse_tail->remaining;
1920 if (last_offset < offset) {
1921 /* Add a hole block. */
1922 r = _sparse_list_add_block(pax, last_offset,
1923 offset - last_offset, 1);
1924 if (r != ARCHIVE_OK)
1927 /* Add data block. */
1928 return (_sparse_list_add_block(pax, offset, length, 0));