2 * Copyright (c) 2003-2007 Tim Kientzle
3 * Copyright (c) 2011-2012 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: head/lib/libarchive/archive_read_support_format_tar.c 201161 2009-12-29 05:44:39Z kientzle $");
43 #include "archive_acl_private.h" /* For ACL parsing routines. */
44 #include "archive_entry.h"
45 #include "archive_entry_locale.h"
46 #include "archive_private.h"
47 #include "archive_read_private.h"
49 #define tar_min(a,b) ((a) < (b) ? (a) : (b))
52 * Layout of POSIX 'ustar' tar header.
54 struct archive_entry_header_ustar {
63 char linkname[100]; /* "old format" header ends here */
64 char magic[6]; /* For POSIX: "ustar\0" */
65 char version[2]; /* For POSIX: "00" */
74 * Structure of GNU tar header
81 struct archive_entry_header_gnutar {
91 char magic[8]; /* "ustar \0" (note blank/blank/null at end) */
101 struct gnu_sparse sparse[4];
105 * Old GNU format doesn't use POSIX 'prefix' field; they use
106 * the 'L' (longname) entry instead.
111 * Data specific to this format.
113 struct sparse_block {
114 struct sparse_block *next;
121 struct archive_string acl_text;
122 struct archive_string entry_pathname;
123 /* For "GNU.sparse.name" and other similar path extensions. */
124 struct archive_string entry_pathname_override;
125 struct archive_string entry_linkpath;
126 struct archive_string entry_uname;
127 struct archive_string entry_gname;
128 struct archive_string longlink;
129 struct archive_string longname;
130 struct archive_string pax_header;
131 struct archive_string pax_global;
132 struct archive_string line;
133 int pax_hdrcharset_binary;
134 int header_recursion_depth;
135 int64_t entry_bytes_remaining;
136 int64_t entry_offset;
137 int64_t entry_padding;
138 int64_t entry_bytes_unconsumed;
141 struct sparse_block *sparse_list;
142 struct sparse_block *sparse_last;
143 int64_t sparse_offset;
144 int64_t sparse_numbytes;
145 int sparse_gnu_major;
146 int sparse_gnu_minor;
147 char sparse_gnu_pending;
149 struct archive_string localname;
150 struct archive_string_conv *opt_sconv;
151 struct archive_string_conv *sconv;
152 struct archive_string_conv *sconv_acl;
153 struct archive_string_conv *sconv_default;
154 int init_default_conversion;
156 int process_mac_extensions;
157 int read_concatenated_archives;
158 int realsize_override;
161 static int archive_block_is_null(const char *p);
162 static char *base64_decode(const char *, size_t, size_t *);
163 static int gnu_add_sparse_entry(struct archive_read *, struct tar *,
164 int64_t offset, int64_t remaining);
166 static void gnu_clear_sparse_list(struct tar *);
167 static int gnu_sparse_old_read(struct archive_read *, struct tar *,
168 const struct archive_entry_header_gnutar *header, size_t *);
169 static int gnu_sparse_old_parse(struct archive_read *, struct tar *,
170 const struct gnu_sparse *sparse, int length);
171 static int gnu_sparse_01_parse(struct archive_read *, struct tar *,
173 static ssize_t gnu_sparse_10_read(struct archive_read *, struct tar *,
175 static int header_Solaris_ACL(struct archive_read *, struct tar *,
176 struct archive_entry *, const void *, size_t *);
177 static int header_common(struct archive_read *, struct tar *,
178 struct archive_entry *, const void *);
179 static int header_old_tar(struct archive_read *, struct tar *,
180 struct archive_entry *, const void *);
181 static int header_pax_extensions(struct archive_read *, struct tar *,
182 struct archive_entry *, const void *, size_t *);
183 static int header_pax_global(struct archive_read *, struct tar *,
184 struct archive_entry *, const void *h, size_t *);
185 static int header_longlink(struct archive_read *, struct tar *,
186 struct archive_entry *, const void *h, size_t *);
187 static int header_longname(struct archive_read *, struct tar *,
188 struct archive_entry *, const void *h, size_t *);
189 static int read_mac_metadata_blob(struct archive_read *, struct tar *,
190 struct archive_entry *, const void *h, size_t *);
191 static int header_volume(struct archive_read *, struct tar *,
192 struct archive_entry *, const void *h, size_t *);
193 static int header_ustar(struct archive_read *, struct tar *,
194 struct archive_entry *, const void *h);
195 static int header_gnutar(struct archive_read *, struct tar *,
196 struct archive_entry *, const void *h, size_t *);
197 static int archive_read_format_tar_bid(struct archive_read *, int);
198 static int archive_read_format_tar_options(struct archive_read *,
199 const char *, const char *);
200 static int archive_read_format_tar_cleanup(struct archive_read *);
201 static int archive_read_format_tar_read_data(struct archive_read *a,
202 const void **buff, size_t *size, int64_t *offset);
203 static int archive_read_format_tar_skip(struct archive_read *a);
204 static int archive_read_format_tar_read_header(struct archive_read *,
205 struct archive_entry *);
206 static int checksum(struct archive_read *, const void *);
207 static int pax_attribute(struct archive_read *, struct tar *,
208 struct archive_entry *, const char *key, const char *value,
209 size_t value_length);
210 static int pax_attribute_acl(struct archive_read *, struct tar *,
211 struct archive_entry *, const char *, int);
212 static int pax_attribute_xattr(struct archive_entry *, const char *,
214 static int pax_header(struct archive_read *, struct tar *,
215 struct archive_entry *, struct archive_string *);
216 static void pax_time(const char *, int64_t *sec, long *nanos);
217 static ssize_t readline(struct archive_read *, struct tar *, const char **,
218 ssize_t limit, size_t *);
219 static int read_body_to_string(struct archive_read *, struct tar *,
220 struct archive_string *, const void *h, size_t *);
221 static int solaris_sparse_parse(struct archive_read *, struct tar *,
222 struct archive_entry *, const char *);
223 static int64_t tar_atol(const char *, size_t);
224 static int64_t tar_atol10(const char *, size_t);
225 static int64_t tar_atol256(const char *, size_t);
226 static int64_t tar_atol8(const char *, size_t);
227 static int tar_read_header(struct archive_read *, struct tar *,
228 struct archive_entry *, size_t *);
229 static int tohex(int c);
230 static char *url_decode(const char *);
231 static void tar_flush_unconsumed(struct archive_read *, size_t *);
235 archive_read_support_format_gnutar(struct archive *a)
237 archive_check_magic(a, ARCHIVE_READ_MAGIC,
238 ARCHIVE_STATE_NEW, "archive_read_support_format_gnutar");
239 return (archive_read_support_format_tar(a));
244 archive_read_support_format_tar(struct archive *_a)
246 struct archive_read *a = (struct archive_read *)_a;
250 archive_check_magic(_a, ARCHIVE_READ_MAGIC,
251 ARCHIVE_STATE_NEW, "archive_read_support_format_tar");
253 tar = (struct tar *)calloc(1, sizeof(*tar));
255 archive_set_error(&a->archive, ENOMEM,
256 "Can't allocate tar data");
257 return (ARCHIVE_FATAL);
259 #ifdef HAVE_COPYFILE_H
260 /* Set this by default on Mac OS. */
261 tar->process_mac_extensions = 1;
264 r = __archive_read_register_format(a, tar, "tar",
265 archive_read_format_tar_bid,
266 archive_read_format_tar_options,
267 archive_read_format_tar_read_header,
268 archive_read_format_tar_read_data,
269 archive_read_format_tar_skip,
271 archive_read_format_tar_cleanup,
281 archive_read_format_tar_cleanup(struct archive_read *a)
285 tar = (struct tar *)(a->format->data);
286 gnu_clear_sparse_list(tar);
287 archive_string_free(&tar->acl_text);
288 archive_string_free(&tar->entry_pathname);
289 archive_string_free(&tar->entry_pathname_override);
290 archive_string_free(&tar->entry_linkpath);
291 archive_string_free(&tar->entry_uname);
292 archive_string_free(&tar->entry_gname);
293 archive_string_free(&tar->line);
294 archive_string_free(&tar->pax_global);
295 archive_string_free(&tar->pax_header);
296 archive_string_free(&tar->longname);
297 archive_string_free(&tar->longlink);
298 archive_string_free(&tar->localname);
300 (a->format->data) = NULL;
305 * Validate number field
307 * This has to be pretty lenient in order to accommodate the enormous
308 * variety of tar writers in the world:
309 * = POSIX (IEEE Std 1003.1-1988) ustar requires octal values with leading
310 * zeros and allows fields to be terminated with space or null characters
311 * = Many writers use different termination (in particular, libarchive
312 * omits terminator bytes to squeeze one or two more digits)
313 * = Many writers pad with space and omit leading zeros
314 * = GNU tar and star write base-256 values if numbers are too
315 * big to be represented in octal
317 * Examples of specific tar headers that we should support:
318 * = Perl Archive::Tar terminates uid, gid, devminor and devmajor with two
319 * null bytes, pads size with spaces and other numeric fields with zeroes
320 * = plexus-archiver prior to 2.6.3 (before switching to commons-compress)
321 * may have uid and gid fields filled with spaces without any octal digits
322 * at all and pads all numeric fields with spaces
324 * This should tolerate all variants in use. It will reject a field
325 * where the writer just left garbage after a trailing NUL.
328 validate_number_field(const char* p_field, size_t i_size)
330 unsigned char marker = (unsigned char)p_field[0];
331 if (marker == 128 || marker == 255 || marker == 0) {
332 /* Base-256 marker, there's nothing we can check. */
337 /* Skip any leading spaces */
338 while (i < i_size && p_field[i] == ' ') {
341 /* Skip octal digits. */
342 while (i < i_size && p_field[i] >= '0' && p_field[i] <= '7') {
345 /* Any remaining characters must be space or NUL padding. */
347 if (p_field[i] != ' ' && p_field[i] != 0) {
357 archive_read_format_tar_bid(struct archive_read *a, int best_bid)
361 const struct archive_entry_header_ustar *header;
363 (void)best_bid; /* UNUSED */
367 /* Now let's look at the actual header and see if it matches. */
368 h = __archive_read_ahead(a, 512, NULL);
372 /* If it's an end-of-archive mark, we can handle it. */
373 if (h[0] == 0 && archive_block_is_null(h)) {
375 * Usually, I bid the number of bits verified, but
376 * in this case, 4096 seems excessive so I picked 10 as
377 * an arbitrary but reasonable-seeming value.
382 /* If it's not an end-of-archive mark, it must have a valid checksum.*/
385 bid += 48; /* Checksum is usually 6 octal digits. */
387 header = (const struct archive_entry_header_ustar *)h;
389 /* Recognize POSIX formats. */
390 if ((memcmp(header->magic, "ustar\0", 6) == 0)
391 && (memcmp(header->version, "00", 2) == 0))
394 /* Recognize GNU tar format. */
395 if ((memcmp(header->magic, "ustar ", 6) == 0)
396 && (memcmp(header->version, " \0", 2) == 0))
399 /* Type flag must be null, digit or A-Z, a-z. */
400 if (header->typeflag[0] != 0 &&
401 !( header->typeflag[0] >= '0' && header->typeflag[0] <= '9') &&
402 !( header->typeflag[0] >= 'A' && header->typeflag[0] <= 'Z') &&
403 !( header->typeflag[0] >= 'a' && header->typeflag[0] <= 'z') )
405 bid += 2; /* 6 bits of variation in an 8-bit field leaves 2 bits. */
408 * Check format of mode/uid/gid/mtime/size/rdevmajor/rdevminor fields.
411 validate_number_field(header->mode, sizeof(header->mode)) == 0
412 || validate_number_field(header->uid, sizeof(header->uid)) == 0
413 || validate_number_field(header->gid, sizeof(header->gid)) == 0
414 || validate_number_field(header->mtime, sizeof(header->mtime)) == 0
415 || validate_number_field(header->size, sizeof(header->size)) == 0
416 || validate_number_field(header->rdevmajor, sizeof(header->rdevmajor)) == 0
417 || validate_number_field(header->rdevminor, sizeof(header->rdevminor)) == 0)) {
425 archive_read_format_tar_options(struct archive_read *a,
426 const char *key, const char *val)
429 int ret = ARCHIVE_FAILED;
431 tar = (struct tar *)(a->format->data);
432 if (strcmp(key, "compat-2x") == 0) {
433 /* Handle UTF-8 filenames as libarchive 2.x */
434 tar->compat_2x = (val != NULL && val[0] != 0);
435 tar->init_default_conversion = tar->compat_2x;
437 } else if (strcmp(key, "hdrcharset") == 0) {
438 if (val == NULL || val[0] == 0)
439 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
440 "tar: hdrcharset option needs a character-set name");
443 archive_string_conversion_from_charset(
444 &a->archive, val, 0);
445 if (tar->opt_sconv != NULL)
451 } else if (strcmp(key, "mac-ext") == 0) {
452 tar->process_mac_extensions = (val != NULL && val[0] != 0);
454 } else if (strcmp(key, "read_concatenated_archives") == 0) {
455 tar->read_concatenated_archives = (val != NULL && val[0] != 0);
459 /* Note: The "warn" return is just to inform the options
460 * supervisor that we didn't handle it. It will generate
461 * a suitable error if no one used this option. */
462 return (ARCHIVE_WARN);
465 /* utility function- this exists to centralize the logic of tracking
466 * how much unconsumed data we have floating around, and to consume
467 * anything outstanding since we're going to do read_aheads
470 tar_flush_unconsumed(struct archive_read *a, size_t *unconsumed)
474 void *data = (void *)__archive_read_ahead(a, *unconsumed, NULL);
475 * this block of code is to poison claimed unconsumed space, ensuring
476 * things break if it is in use still.
477 * currently it WILL break things, so enable it only for debugging this issue
479 memset(data, 0xff, *unconsumed);
482 __archive_read_consume(a, *unconsumed);
488 * The function invoked by archive_read_next_header(). This
489 * just sets up a few things and then calls the internal
490 * tar_read_header() function below.
493 archive_read_format_tar_read_header(struct archive_read *a,
494 struct archive_entry *entry)
497 * When converting tar archives to cpio archives, it is
498 * essential that each distinct file have a distinct inode
499 * number. To simplify this, we keep a static count here to
500 * assign fake dev/inode numbers to each tar entry. Note that
501 * pax format archives may overwrite this with something more
504 * Ideally, we would track every file read from the archive so
505 * that we could assign the same dev/ino pair to hardlinks,
506 * but the memory required to store a complete lookup table is
507 * probably not worthwhile just to support the relatively
508 * obscure tar->cpio conversion case.
510 static int default_inode;
511 static int default_dev;
516 size_t l, unconsumed = 0;
518 /* Assign default device/inode values. */
519 archive_entry_set_dev(entry, 1 + default_dev); /* Don't use zero. */
520 archive_entry_set_ino(entry, ++default_inode); /* Don't use zero. */
521 /* Limit generated st_ino number to 16 bits. */
522 if (default_inode >= 0xffff) {
527 tar = (struct tar *)(a->format->data);
528 tar->entry_offset = 0;
529 gnu_clear_sparse_list(tar);
530 tar->realsize = -1; /* Mark this as "unset" */
531 tar->realsize_override = 0;
533 /* Setup default string conversion. */
534 tar->sconv = tar->opt_sconv;
535 if (tar->sconv == NULL) {
536 if (!tar->init_default_conversion) {
538 archive_string_default_conversion_for_read(&(a->archive));
539 tar->init_default_conversion = 1;
541 tar->sconv = tar->sconv_default;
544 r = tar_read_header(a, tar, entry, &unconsumed);
546 tar_flush_unconsumed(a, &unconsumed);
549 * "non-sparse" files are really just sparse files with
552 if (tar->sparse_list == NULL) {
553 if (gnu_add_sparse_entry(a, tar, 0, tar->entry_bytes_remaining)
555 return (ARCHIVE_FATAL);
557 struct sparse_block *sb;
559 for (sb = tar->sparse_list; sb != NULL; sb = sb->next) {
561 archive_entry_sparse_add_entry(entry,
562 sb->offset, sb->remaining);
566 if (r == ARCHIVE_OK && archive_entry_filetype(entry) == AE_IFREG) {
568 * "Regular" entry with trailing '/' is really
569 * directory: This is needed for certain old tar
570 * variants and even for some broken newer ones.
572 if ((wp = archive_entry_pathname_w(entry)) != NULL) {
574 if (l > 0 && wp[l - 1] == L'/') {
575 archive_entry_set_filetype(entry, AE_IFDIR);
577 } else if ((p = archive_entry_pathname(entry)) != NULL) {
579 if (l > 0 && p[l - 1] == '/') {
580 archive_entry_set_filetype(entry, AE_IFDIR);
588 archive_read_format_tar_read_data(struct archive_read *a,
589 const void **buff, size_t *size, int64_t *offset)
593 struct sparse_block *p;
595 tar = (struct tar *)(a->format->data);
598 /* Remove exhausted entries from sparse list. */
599 while (tar->sparse_list != NULL &&
600 tar->sparse_list->remaining == 0) {
601 p = tar->sparse_list;
602 tar->sparse_list = p->next;
606 if (tar->entry_bytes_unconsumed) {
607 __archive_read_consume(a, tar->entry_bytes_unconsumed);
608 tar->entry_bytes_unconsumed = 0;
611 /* If we're at end of file, return EOF. */
612 if (tar->sparse_list == NULL ||
613 tar->entry_bytes_remaining == 0) {
614 if (__archive_read_consume(a, tar->entry_padding) < 0)
615 return (ARCHIVE_FATAL);
616 tar->entry_padding = 0;
619 *offset = tar->realsize;
620 return (ARCHIVE_EOF);
623 *buff = __archive_read_ahead(a, 1, &bytes_read);
625 return (ARCHIVE_FATAL);
627 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
628 "Truncated tar archive");
629 return (ARCHIVE_FATAL);
631 if (bytes_read > tar->entry_bytes_remaining)
632 bytes_read = (ssize_t)tar->entry_bytes_remaining;
633 /* Don't read more than is available in the
634 * current sparse block. */
635 if (tar->sparse_list->remaining < bytes_read)
636 bytes_read = (ssize_t)tar->sparse_list->remaining;
638 *offset = tar->sparse_list->offset;
639 tar->sparse_list->remaining -= bytes_read;
640 tar->sparse_list->offset += bytes_read;
641 tar->entry_bytes_remaining -= bytes_read;
642 tar->entry_bytes_unconsumed = bytes_read;
644 if (!tar->sparse_list->hole)
646 /* Current is hole data and skip this. */
651 archive_read_format_tar_skip(struct archive_read *a)
653 int64_t bytes_skipped;
655 struct sparse_block *p;
658 tar = (struct tar *)(a->format->data);
660 /* Do not consume the hole of a sparse file. */
662 for (p = tar->sparse_list; p != NULL; p = p->next) {
664 if (p->remaining >= INT64_MAX - request) {
665 return ARCHIVE_FATAL;
667 request += p->remaining;
670 if (request > tar->entry_bytes_remaining)
671 request = tar->entry_bytes_remaining;
672 request += tar->entry_padding + tar->entry_bytes_unconsumed;
674 bytes_skipped = __archive_read_consume(a, request);
675 if (bytes_skipped < 0)
676 return (ARCHIVE_FATAL);
678 tar->entry_bytes_remaining = 0;
679 tar->entry_bytes_unconsumed = 0;
680 tar->entry_padding = 0;
682 /* Free the sparse list. */
683 gnu_clear_sparse_list(tar);
689 * This function recursively interprets all of the headers associated
690 * with a single entry.
693 tar_read_header(struct archive_read *a, struct tar *tar,
694 struct archive_entry *entry, size_t *unconsumed)
699 const struct archive_entry_header_ustar *header;
700 const struct archive_entry_header_gnutar *gnuheader;
702 /* Loop until we find a workable header record. */
704 tar_flush_unconsumed(a, unconsumed);
706 /* Read 512-byte header record */
707 h = __archive_read_ahead(a, 512, &bytes);
710 if (bytes == 0) { /* EOF at a block boundary. */
711 /* Some writers do omit the block of nulls. <sigh> */
712 return (ARCHIVE_EOF);
714 if (bytes < 512) { /* Short block at EOF; this is bad. */
715 archive_set_error(&a->archive,
716 ARCHIVE_ERRNO_FILE_FORMAT,
717 "Truncated tar archive");
718 return (ARCHIVE_FATAL);
722 /* Header is workable if it's not an end-of-archive mark. */
723 if (h[0] != 0 || !archive_block_is_null(h))
726 /* Ensure format is set for archives with only null blocks. */
727 if (a->archive.archive_format_name == NULL) {
728 a->archive.archive_format = ARCHIVE_FORMAT_TAR;
729 a->archive.archive_format_name = "tar";
732 if (!tar->read_concatenated_archives) {
733 /* Try to consume a second all-null record, as well. */
734 tar_flush_unconsumed(a, unconsumed);
735 h = __archive_read_ahead(a, 512, NULL);
736 if (h != NULL && h[0] == 0 && archive_block_is_null(h))
737 __archive_read_consume(a, 512);
738 archive_clear_error(&a->archive);
739 return (ARCHIVE_EOF);
743 * We're reading concatenated archives, ignore this block and
744 * loop to get the next.
749 * Note: If the checksum fails and we return ARCHIVE_RETRY,
750 * then the client is likely to just retry. This is a very
751 * crude way to search for the next valid header!
753 * TODO: Improve this by implementing a real header scan.
755 if (!checksum(a, h)) {
756 tar_flush_unconsumed(a, unconsumed);
757 archive_set_error(&a->archive, EINVAL, "Damaged tar archive");
758 return (ARCHIVE_RETRY); /* Retryable: Invalid header */
761 if (++tar->header_recursion_depth > 32) {
762 tar_flush_unconsumed(a, unconsumed);
763 archive_set_error(&a->archive, EINVAL, "Too many special headers");
764 return (ARCHIVE_WARN);
767 /* Determine the format variant. */
768 header = (const struct archive_entry_header_ustar *)h;
770 switch(header->typeflag[0]) {
771 case 'A': /* Solaris tar ACL */
772 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
773 a->archive.archive_format_name = "Solaris tar";
774 err = header_Solaris_ACL(a, tar, entry, h, unconsumed);
776 case 'g': /* POSIX-standard 'g' header. */
777 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
778 a->archive.archive_format_name = "POSIX pax interchange format";
779 err = header_pax_global(a, tar, entry, h, unconsumed);
780 if (err == ARCHIVE_EOF)
783 case 'K': /* Long link name (GNU tar, others) */
784 err = header_longlink(a, tar, entry, h, unconsumed);
786 case 'L': /* Long filename (GNU tar, others) */
787 err = header_longname(a, tar, entry, h, unconsumed);
789 case 'V': /* GNU volume header */
790 err = header_volume(a, tar, entry, h, unconsumed);
792 case 'X': /* Used by SUN tar; same as 'x'. */
793 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
794 a->archive.archive_format_name =
795 "POSIX pax interchange format (Sun variant)";
796 err = header_pax_extensions(a, tar, entry, h, unconsumed);
798 case 'x': /* POSIX-standard 'x' header. */
799 a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
800 a->archive.archive_format_name = "POSIX pax interchange format";
801 err = header_pax_extensions(a, tar, entry, h, unconsumed);
804 gnuheader = (const struct archive_entry_header_gnutar *)h;
805 if (memcmp(gnuheader->magic, "ustar \0", 8) == 0) {
806 a->archive.archive_format = ARCHIVE_FORMAT_TAR_GNUTAR;
807 a->archive.archive_format_name = "GNU tar format";
808 err = header_gnutar(a, tar, entry, h, unconsumed);
809 } else if (memcmp(header->magic, "ustar", 5) == 0) {
810 if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) {
811 a->archive.archive_format = ARCHIVE_FORMAT_TAR_USTAR;
812 a->archive.archive_format_name = "POSIX ustar format";
814 err = header_ustar(a, tar, entry, h);
816 a->archive.archive_format = ARCHIVE_FORMAT_TAR;
817 a->archive.archive_format_name = "tar (non-POSIX)";
818 err = header_old_tar(a, tar, entry, h);
821 if (err == ARCHIVE_FATAL)
824 tar_flush_unconsumed(a, unconsumed);
829 --tar->header_recursion_depth;
830 /* Yuck. Apple's design here ends up storing long pathname
831 * extensions for both the AppleDouble extension entry and the
834 if ((err == ARCHIVE_WARN || err == ARCHIVE_OK) &&
835 tar->header_recursion_depth == 0 &&
836 tar->process_mac_extensions) {
837 int err2 = read_mac_metadata_blob(a, tar, entry, h, unconsumed);
842 /* We return warnings or success as-is. Anything else is fatal. */
843 if (err == ARCHIVE_WARN || err == ARCHIVE_OK) {
844 if (tar->sparse_gnu_pending) {
845 if (tar->sparse_gnu_major == 1 &&
846 tar->sparse_gnu_minor == 0) {
849 tar->sparse_gnu_pending = 0;
850 /* Read initial sparse map. */
851 bytes_read = gnu_sparse_10_read(a, tar, unconsumed);
853 return ((int)bytes_read);
854 tar->entry_bytes_remaining -= bytes_read;
856 archive_set_error(&a->archive,
858 "Unrecognized GNU sparse file format");
859 return (ARCHIVE_WARN);
861 tar->sparse_gnu_pending = 0;
865 if (err == ARCHIVE_EOF)
866 /* EOF when recursively reading a header is bad. */
867 archive_set_error(&a->archive, EINVAL, "Damaged tar archive");
868 return (ARCHIVE_FATAL);
872 * Return true if block checksum is correct.
875 checksum(struct archive_read *a, const void *h)
877 const unsigned char *bytes;
878 const struct archive_entry_header_ustar *header;
882 (void)a; /* UNUSED */
883 bytes = (const unsigned char *)h;
884 header = (const struct archive_entry_header_ustar *)h;
886 /* Checksum field must hold an octal number */
887 for (i = 0; i < sizeof(header->checksum); ++i) {
888 char c = header->checksum[i];
889 if (c != ' ' && c != '\0' && (c < '0' || c > '7'))
894 * Test the checksum. Note that POSIX specifies _unsigned_
895 * bytes for this calculation.
897 sum = (int)tar_atol(header->checksum, sizeof(header->checksum));
899 for (i = 0; i < 148; i++)
900 check += (unsigned char)bytes[i];
904 check += (unsigned char)bytes[i];
909 * Repeat test with _signed_ bytes, just in case this archive
910 * was created by an old BSD, Solaris, or HP-UX tar with a
911 * broken checksum calculation.
914 for (i = 0; i < 148; i++)
915 check += (signed char)bytes[i];
919 check += (signed char)bytes[i];
927 * Return true if this block contains only nulls.
930 archive_block_is_null(const char *p)
934 for (i = 0; i < 512; i++)
941 * Interpret 'A' Solaris ACL header
944 header_Solaris_ACL(struct archive_read *a, struct tar *tar,
945 struct archive_entry *entry, const void *h, size_t *unconsumed)
947 const struct archive_entry_header_ustar *header;
954 * read_body_to_string adds a NUL terminator, but we need a little
955 * more to make sure that we don't overrun acl_text later.
957 header = (const struct archive_entry_header_ustar *)h;
958 size = (size_t)tar_atol(header->size, sizeof(header->size));
959 err = read_body_to_string(a, tar, &(tar->acl_text), h, unconsumed);
960 if (err != ARCHIVE_OK)
963 /* Recursively read next header */
964 err = tar_read_header(a, tar, entry, unconsumed);
965 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
968 /* TODO: Examine the first characters to see if this
969 * is an AIX ACL descriptor. We'll likely never support
970 * them, but it would be polite to recognize and warn when
973 /* Leading octal number indicates ACL type and number of entries. */
974 p = acl = tar->acl_text.s;
976 while (*p != '\0' && p < acl + size) {
977 if (*p < '0' || *p > '7') {
978 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
979 "Malformed Solaris ACL attribute (invalid digit)");
980 return(ARCHIVE_WARN);
984 if (type > 077777777) {
985 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
986 "Malformed Solaris ACL attribute (count too large)");
987 return (ARCHIVE_WARN);
991 switch ((int)type & ~0777777) {
994 acl_type = ARCHIVE_ENTRY_ACL_TYPE_ACCESS;
998 acl_type = ARCHIVE_ENTRY_ACL_TYPE_NFS4;
1001 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1002 "Malformed Solaris ACL attribute (unsupported type %o)",
1004 return (ARCHIVE_WARN);
1008 if (p >= acl + size) {
1009 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1010 "Malformed Solaris ACL attribute (body overflow)");
1011 return(ARCHIVE_WARN);
1014 /* ACL text is null-terminated; find the end. */
1018 while (*p != '\0' && p < acl + size)
1021 if (tar->sconv_acl == NULL) {
1022 tar->sconv_acl = archive_string_conversion_from_charset(
1023 &(a->archive), "UTF-8", 1);
1024 if (tar->sconv_acl == NULL)
1025 return (ARCHIVE_FATAL);
1027 archive_strncpy(&(tar->localname), acl, p - acl);
1028 err = archive_acl_from_text_l(archive_entry_acl(entry),
1029 tar->localname.s, acl_type, tar->sconv_acl);
1030 if (err != ARCHIVE_OK) {
1031 if (errno == ENOMEM) {
1032 archive_set_error(&a->archive, ENOMEM,
1033 "Can't allocate memory for ACL");
1035 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1036 "Malformed Solaris ACL attribute (unparsable)");
1042 * Interpret 'K' long linkname header.
1045 header_longlink(struct archive_read *a, struct tar *tar,
1046 struct archive_entry *entry, const void *h, size_t *unconsumed)
1050 err = read_body_to_string(a, tar, &(tar->longlink), h, unconsumed);
1051 if (err != ARCHIVE_OK)
1053 err = tar_read_header(a, tar, entry, unconsumed);
1054 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1056 /* Set symlink if symlink already set, else hardlink. */
1057 archive_entry_copy_link(entry, tar->longlink.s);
1058 return (ARCHIVE_OK);
1062 set_conversion_failed_error(struct archive_read *a,
1063 struct archive_string_conv *sconv, const char *name)
1065 if (errno == ENOMEM) {
1066 archive_set_error(&a->archive, ENOMEM,
1067 "Can't allocate memory for %s", name);
1068 return (ARCHIVE_FATAL);
1070 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1071 "%s can't be converted from %s to current locale.",
1072 name, archive_string_conversion_charset_name(sconv));
1073 return (ARCHIVE_WARN);
1077 * Interpret 'L' long filename header.
1080 header_longname(struct archive_read *a, struct tar *tar,
1081 struct archive_entry *entry, const void *h, size_t *unconsumed)
1085 err = read_body_to_string(a, tar, &(tar->longname), h, unconsumed);
1086 if (err != ARCHIVE_OK)
1088 /* Read and parse "real" header, then override name. */
1089 err = tar_read_header(a, tar, entry, unconsumed);
1090 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1092 if (archive_entry_copy_pathname_l(entry, tar->longname.s,
1093 archive_strlen(&(tar->longname)), tar->sconv) != 0)
1094 err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1100 * Interpret 'V' GNU tar volume header.
1103 header_volume(struct archive_read *a, struct tar *tar,
1104 struct archive_entry *entry, const void *h, size_t *unconsumed)
1108 /* Just skip this and read the next header. */
1109 return (tar_read_header(a, tar, entry, unconsumed));
1113 * Read body of an archive entry into an archive_string object.
1116 read_body_to_string(struct archive_read *a, struct tar *tar,
1117 struct archive_string *as, const void *h, size_t *unconsumed)
1120 const struct archive_entry_header_ustar *header;
1123 (void)tar; /* UNUSED */
1124 header = (const struct archive_entry_header_ustar *)h;
1125 size = tar_atol(header->size, sizeof(header->size));
1126 if ((size > 1048576) || (size < 0)) {
1127 archive_set_error(&a->archive, EINVAL,
1128 "Special header too large");
1129 return (ARCHIVE_FATAL);
1132 /* Fail if we can't make our buffer big enough. */
1133 if (archive_string_ensure(as, (size_t)size+1) == NULL) {
1134 archive_set_error(&a->archive, ENOMEM,
1136 return (ARCHIVE_FATAL);
1139 tar_flush_unconsumed(a, unconsumed);
1141 /* Read the body into the string. */
1142 *unconsumed = (size_t)((size + 511) & ~ 511);
1143 src = __archive_read_ahead(a, *unconsumed, NULL);
1146 return (ARCHIVE_FATAL);
1148 memcpy(as->s, src, (size_t)size);
1150 as->length = (size_t)size;
1151 return (ARCHIVE_OK);
1155 * Parse out common header elements.
1157 * This would be the same as header_old_tar, except that the
1158 * filename is handled slightly differently for old and POSIX
1159 * entries (POSIX entries support a 'prefix'). This factoring
1160 * allows header_old_tar and header_ustar
1161 * to handle filenames differently, while still putting most of the
1162 * common parsing into one place.
1165 header_common(struct archive_read *a, struct tar *tar,
1166 struct archive_entry *entry, const void *h)
1168 const struct archive_entry_header_ustar *header;
1170 int err = ARCHIVE_OK;
1172 header = (const struct archive_entry_header_ustar *)h;
1173 if (header->linkname[0])
1174 archive_strncpy(&(tar->entry_linkpath),
1175 header->linkname, sizeof(header->linkname));
1177 archive_string_empty(&(tar->entry_linkpath));
1179 /* Parse out the numeric fields (all are octal) */
1180 archive_entry_set_mode(entry,
1181 (mode_t)tar_atol(header->mode, sizeof(header->mode)));
1182 archive_entry_set_uid(entry, tar_atol(header->uid, sizeof(header->uid)));
1183 archive_entry_set_gid(entry, tar_atol(header->gid, sizeof(header->gid)));
1184 tar->entry_bytes_remaining = tar_atol(header->size, sizeof(header->size));
1185 if (tar->entry_bytes_remaining < 0) {
1186 tar->entry_bytes_remaining = 0;
1187 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1188 "Tar entry has negative size");
1189 return (ARCHIVE_FATAL);
1191 if (tar->entry_bytes_remaining == INT64_MAX) {
1192 /* Note: tar_atol returns INT64_MAX on overflow */
1193 tar->entry_bytes_remaining = 0;
1194 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1195 "Tar entry size overflow");
1196 return (ARCHIVE_FATAL);
1198 tar->realsize = tar->entry_bytes_remaining;
1199 archive_entry_set_size(entry, tar->entry_bytes_remaining);
1200 archive_entry_set_mtime(entry, tar_atol(header->mtime, sizeof(header->mtime)), 0);
1202 /* Handle the tar type flag appropriately. */
1203 tartype = header->typeflag[0];
1206 case '1': /* Hard link */
1207 if (archive_entry_copy_hardlink_l(entry, tar->entry_linkpath.s,
1208 archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) {
1209 err = set_conversion_failed_error(a, tar->sconv,
1211 if (err == ARCHIVE_FATAL)
1215 * The following may seem odd, but: Technically, tar
1216 * does not store the file type for a "hard link"
1217 * entry, only the fact that it is a hard link. So, I
1218 * leave the type zero normally. But, pax interchange
1219 * format allows hard links to have data, which
1220 * implies that the underlying entry is a regular
1223 if (archive_entry_size(entry) > 0)
1224 archive_entry_set_filetype(entry, AE_IFREG);
1227 * A tricky point: Traditionally, tar readers have
1228 * ignored the size field when reading hardlink
1229 * entries, and some writers put non-zero sizes even
1230 * though the body is empty. POSIX blessed this
1231 * convention in the 1988 standard, but broke with
1232 * this tradition in 2001 by permitting hardlink
1233 * entries to store valid bodies in pax interchange
1234 * format, but not in ustar format. Since there is no
1235 * hard and fast way to distinguish pax interchange
1236 * from earlier archives (the 'x' and 'g' entries are
1237 * optional, after all), we need a heuristic.
1239 if (archive_entry_size(entry) == 0) {
1240 /* If the size is already zero, we're done. */
1241 } else if (a->archive.archive_format
1242 == ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) {
1243 /* Definitely pax extended; must obey hardlink size. */
1244 } else if (a->archive.archive_format == ARCHIVE_FORMAT_TAR
1245 || a->archive.archive_format == ARCHIVE_FORMAT_TAR_GNUTAR)
1247 /* Old-style or GNU tar: we must ignore the size. */
1248 archive_entry_set_size(entry, 0);
1249 tar->entry_bytes_remaining = 0;
1250 } else if (archive_read_format_tar_bid(a, 50) > 50) {
1252 * We don't know if it's pax: If the bid
1253 * function sees a valid ustar header
1254 * immediately following, then let's ignore
1255 * the hardlink size.
1257 archive_entry_set_size(entry, 0);
1258 tar->entry_bytes_remaining = 0;
1261 * TODO: There are still two cases I'd like to handle:
1262 * = a ustar non-pax archive with a hardlink entry at
1263 * end-of-archive. (Look for block of nulls following?)
1264 * = a pax archive that has not seen any pax headers
1265 * and has an entry which is a hardlink entry storing
1266 * a body containing an uncompressed tar archive.
1267 * The first is worth addressing; I don't see any reliable
1268 * way to deal with the second possibility.
1271 case '2': /* Symlink */
1272 archive_entry_set_filetype(entry, AE_IFLNK);
1273 archive_entry_set_size(entry, 0);
1274 tar->entry_bytes_remaining = 0;
1275 if (archive_entry_copy_symlink_l(entry, tar->entry_linkpath.s,
1276 archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) {
1277 err = set_conversion_failed_error(a, tar->sconv,
1279 if (err == ARCHIVE_FATAL)
1283 case '3': /* Character device */
1284 archive_entry_set_filetype(entry, AE_IFCHR);
1285 archive_entry_set_size(entry, 0);
1286 tar->entry_bytes_remaining = 0;
1288 case '4': /* Block device */
1289 archive_entry_set_filetype(entry, AE_IFBLK);
1290 archive_entry_set_size(entry, 0);
1291 tar->entry_bytes_remaining = 0;
1294 archive_entry_set_filetype(entry, AE_IFDIR);
1295 archive_entry_set_size(entry, 0);
1296 tar->entry_bytes_remaining = 0;
1298 case '6': /* FIFO device */
1299 archive_entry_set_filetype(entry, AE_IFIFO);
1300 archive_entry_set_size(entry, 0);
1301 tar->entry_bytes_remaining = 0;
1303 case 'D': /* GNU incremental directory type */
1305 * No special handling is actually required here.
1306 * It might be nice someday to preprocess the file list and
1307 * provide it to the client, though.
1309 archive_entry_set_filetype(entry, AE_IFDIR);
1311 case 'M': /* GNU "Multi-volume" (remainder of file from last archive)*/
1313 * As far as I can tell, this is just like a regular file
1314 * entry, except that the contents should be _appended_ to
1315 * the indicated file at the indicated offset. This may
1316 * require some API work to fully support.
1319 case 'N': /* Old GNU "long filename" entry. */
1320 /* The body of this entry is a script for renaming
1321 * previously-extracted entries. Ugh. It will never
1322 * be supported by libarchive. */
1323 archive_entry_set_filetype(entry, AE_IFREG);
1325 case 'S': /* GNU sparse files */
1327 * Sparse files are really just regular files with
1328 * sparse information in the extended area.
1333 * Enable sparse file "read" support only for regular
1334 * files and explicit GNU sparse files. However, we
1335 * don't allow non-standard file types to be sparse.
1337 tar->sparse_allowed = 1;
1339 default: /* Regular file and non-standard types */
1341 * Per POSIX: non-recognized types should always be
1342 * treated as regular files.
1344 archive_entry_set_filetype(entry, AE_IFREG);
1351 * Parse out header elements for "old-style" tar archives.
1354 header_old_tar(struct archive_read *a, struct tar *tar,
1355 struct archive_entry *entry, const void *h)
1357 const struct archive_entry_header_ustar *header;
1358 int err = ARCHIVE_OK, err2;
1360 /* Copy filename over (to ensure null termination). */
1361 header = (const struct archive_entry_header_ustar *)h;
1362 if (archive_entry_copy_pathname_l(entry,
1363 header->name, sizeof(header->name), tar->sconv) != 0) {
1364 err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1365 if (err == ARCHIVE_FATAL)
1369 /* Grab rest of common fields */
1370 err2 = header_common(a, tar, entry, h);
1374 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1379 * Read a Mac AppleDouble-encoded blob of file metadata,
1383 read_mac_metadata_blob(struct archive_read *a, struct tar *tar,
1384 struct archive_entry *entry, const void *h, size_t *unconsumed)
1388 const char *p, *name;
1389 const wchar_t *wp, *wname;
1391 (void)h; /* UNUSED */
1393 wname = wp = archive_entry_pathname_w(entry);
1395 /* Find the last path element. */
1396 for (; *wp != L'\0'; ++wp) {
1397 if (wp[0] == '/' && wp[1] != L'\0')
1401 * If last path element starts with "._", then
1402 * this is a Mac extension.
1404 if (wname[0] != L'.' || wname[1] != L'_' || wname[2] == L'\0')
1407 /* Find the last path element. */
1408 name = p = archive_entry_pathname(entry);
1410 return (ARCHIVE_FAILED);
1411 for (; *p != '\0'; ++p) {
1412 if (p[0] == '/' && p[1] != '\0')
1416 * If last path element starts with "._", then
1417 * this is a Mac extension.
1419 if (name[0] != '.' || name[1] != '_' || name[2] == '\0')
1423 /* Read the body as a Mac OS metadata blob. */
1424 size = archive_entry_size(entry);
1427 * TODO: Look beyond the body here to peek at the next header.
1428 * If it's a regular header (not an extension header)
1429 * that has the wrong name, just return the current
1430 * entry as-is, without consuming the body here.
1431 * That would reduce the risk of us mis-identifying
1432 * an ordinary file that just happened to have
1433 * a name starting with "._".
1435 * Q: Is the above idea really possible? Even
1436 * when there are GNU or pax extension entries?
1438 data = __archive_read_ahead(a, (size_t)size, NULL);
1441 return (ARCHIVE_FATAL);
1443 archive_entry_copy_mac_metadata(entry, data, (size_t)size);
1444 *unconsumed = (size_t)((size + 511) & ~ 511);
1445 tar_flush_unconsumed(a, unconsumed);
1446 return (tar_read_header(a, tar, entry, unconsumed));
1450 * Parse a file header for a pax extended archive entry.
1453 header_pax_global(struct archive_read *a, struct tar *tar,
1454 struct archive_entry *entry, const void *h, size_t *unconsumed)
1458 err = read_body_to_string(a, tar, &(tar->pax_global), h, unconsumed);
1459 if (err != ARCHIVE_OK)
1461 err = tar_read_header(a, tar, entry, unconsumed);
1466 header_pax_extensions(struct archive_read *a, struct tar *tar,
1467 struct archive_entry *entry, const void *h, size_t *unconsumed)
1471 err = read_body_to_string(a, tar, &(tar->pax_header), h, unconsumed);
1472 if (err != ARCHIVE_OK)
1475 /* Parse the next header. */
1476 err = tar_read_header(a, tar, entry, unconsumed);
1477 if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1481 * TODO: Parse global/default options into 'entry' struct here
1482 * before handling file-specific options.
1484 * This design (parse standard header, then overwrite with pax
1485 * extended attribute data) usually works well, but isn't ideal;
1486 * it would be better to parse the pax extended attributes first
1487 * and then skip any fields in the standard header that were
1488 * defined in the pax header.
1490 err2 = pax_header(a, tar, entry, &tar->pax_header);
1491 err = err_combine(err, err2);
1492 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1498 * Parse a file header for a Posix "ustar" archive entry. This also
1499 * handles "pax" or "extended ustar" entries.
1502 header_ustar(struct archive_read *a, struct tar *tar,
1503 struct archive_entry *entry, const void *h)
1505 const struct archive_entry_header_ustar *header;
1506 struct archive_string *as;
1507 int err = ARCHIVE_OK, r;
1509 header = (const struct archive_entry_header_ustar *)h;
1511 /* Copy name into an internal buffer to ensure null-termination. */
1512 as = &(tar->entry_pathname);
1513 if (header->prefix[0]) {
1514 archive_strncpy(as, header->prefix, sizeof(header->prefix));
1515 if (as->s[archive_strlen(as) - 1] != '/')
1516 archive_strappend_char(as, '/');
1517 archive_strncat(as, header->name, sizeof(header->name));
1519 archive_strncpy(as, header->name, sizeof(header->name));
1521 if (archive_entry_copy_pathname_l(entry, as->s, archive_strlen(as),
1523 err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1524 if (err == ARCHIVE_FATAL)
1528 /* Handle rest of common fields. */
1529 r = header_common(a, tar, entry, h);
1530 if (r == ARCHIVE_FATAL)
1535 /* Handle POSIX ustar fields. */
1536 if (archive_entry_copy_uname_l(entry,
1537 header->uname, sizeof(header->uname), tar->sconv) != 0) {
1538 err = set_conversion_failed_error(a, tar->sconv, "Uname");
1539 if (err == ARCHIVE_FATAL)
1543 if (archive_entry_copy_gname_l(entry,
1544 header->gname, sizeof(header->gname), tar->sconv) != 0) {
1545 err = set_conversion_failed_error(a, tar->sconv, "Gname");
1546 if (err == ARCHIVE_FATAL)
1550 /* Parse out device numbers only for char and block specials. */
1551 if (header->typeflag[0] == '3' || header->typeflag[0] == '4') {
1552 archive_entry_set_rdevmajor(entry, (dev_t)
1553 tar_atol(header->rdevmajor, sizeof(header->rdevmajor)));
1554 archive_entry_set_rdevminor(entry, (dev_t)
1555 tar_atol(header->rdevminor, sizeof(header->rdevminor)));
1558 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1565 * Parse the pax extended attributes record.
1567 * Returns non-zero if there's an error in the data.
1570 pax_header(struct archive_read *a, struct tar *tar,
1571 struct archive_entry *entry, struct archive_string *in_as)
1573 size_t attr_length, l, line_length, value_length;
1576 struct archive_string *as;
1577 struct archive_string_conv *sconv;
1579 char *attr = in_as->s;
1581 attr_length = in_as->length;
1582 tar->pax_hdrcharset_binary = 0;
1583 archive_string_empty(&(tar->entry_gname));
1584 archive_string_empty(&(tar->entry_linkpath));
1585 archive_string_empty(&(tar->entry_pathname));
1586 archive_string_empty(&(tar->entry_pathname_override));
1587 archive_string_empty(&(tar->entry_uname));
1589 while (attr_length > 0) {
1590 /* Parse decimal length field at start of line. */
1593 p = attr; /* Record start of line. */
1600 if (*p < '0' || *p > '9') {
1601 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1602 "Ignoring malformed pax extended attributes");
1603 return (ARCHIVE_WARN);
1606 line_length += *p - '0';
1607 if (line_length > 999999) {
1608 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1609 "Rejecting pax extended attribute > 1MB");
1610 return (ARCHIVE_WARN);
1617 * Parsed length must be no bigger than available data,
1618 * at least 1, and the last character of the line must
1621 if (line_length > attr_length
1623 || attr[line_length - 1] != '\n')
1625 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1626 "Ignoring malformed pax extended attribute");
1627 return (ARCHIVE_WARN);
1630 /* Null-terminate the line. */
1631 attr[line_length - 1] = '\0';
1633 /* Find end of key and null terminate it. */
1637 while (*p && *p != '=')
1640 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1641 "Invalid pax extended attributes");
1642 return (ARCHIVE_WARN);
1648 /* Some values may be binary data */
1649 value_length = attr + line_length - 1 - value;
1651 /* Identify this attribute and set it in the entry. */
1652 err2 = pax_attribute(a, tar, entry, key, value, value_length);
1653 if (err2 == ARCHIVE_FATAL)
1655 err = err_combine(err, err2);
1657 /* Skip to next line */
1658 attr += line_length;
1659 attr_length -= line_length;
1663 * PAX format uses UTF-8 as default charset for its metadata
1664 * unless hdrcharset=BINARY is present in its header.
1665 * We apply the charset specified by the hdrcharset option only
1666 * when the hdrcharset attribute(in PAX header) is BINARY because
1667 * we respect the charset described in PAX header and BINARY also
1668 * means that metadata(filename,uname and gname) character-set
1671 if (tar->pax_hdrcharset_binary)
1672 sconv = tar->opt_sconv;
1674 sconv = archive_string_conversion_from_charset(
1675 &(a->archive), "UTF-8", 1);
1677 return (ARCHIVE_FATAL);
1679 archive_string_conversion_set_opt(sconv,
1680 SCONV_SET_OPT_UTF8_LIBARCHIVE2X);
1683 if (archive_strlen(&(tar->entry_gname)) > 0) {
1684 if (archive_entry_copy_gname_l(entry, tar->entry_gname.s,
1685 archive_strlen(&(tar->entry_gname)), sconv) != 0) {
1686 err = set_conversion_failed_error(a, sconv, "Gname");
1687 if (err == ARCHIVE_FATAL)
1689 /* Use a converted an original name. */
1690 archive_entry_copy_gname(entry, tar->entry_gname.s);
1693 if (archive_strlen(&(tar->entry_linkpath)) > 0) {
1694 if (archive_entry_copy_link_l(entry, tar->entry_linkpath.s,
1695 archive_strlen(&(tar->entry_linkpath)), sconv) != 0) {
1696 err = set_conversion_failed_error(a, sconv, "Linkname");
1697 if (err == ARCHIVE_FATAL)
1699 /* Use a converted an original name. */
1700 archive_entry_copy_link(entry, tar->entry_linkpath.s);
1704 * Some extensions (such as the GNU sparse file extensions)
1705 * deliberately store a synthetic name under the regular 'path'
1706 * attribute and the real file name under a different attribute.
1707 * Since we're supposed to not care about the order, we
1708 * have no choice but to store all of the various filenames
1709 * we find and figure it all out afterwards. This is the
1710 * figuring out part.
1713 if (archive_strlen(&(tar->entry_pathname_override)) > 0)
1714 as = &(tar->entry_pathname_override);
1715 else if (archive_strlen(&(tar->entry_pathname)) > 0)
1716 as = &(tar->entry_pathname);
1718 if (archive_entry_copy_pathname_l(entry, as->s,
1719 archive_strlen(as), sconv) != 0) {
1720 err = set_conversion_failed_error(a, sconv, "Pathname");
1721 if (err == ARCHIVE_FATAL)
1723 /* Use a converted an original name. */
1724 archive_entry_copy_pathname(entry, as->s);
1727 if (archive_strlen(&(tar->entry_uname)) > 0) {
1728 if (archive_entry_copy_uname_l(entry, tar->entry_uname.s,
1729 archive_strlen(&(tar->entry_uname)), sconv) != 0) {
1730 err = set_conversion_failed_error(a, sconv, "Uname");
1731 if (err == ARCHIVE_FATAL)
1733 /* Use a converted an original name. */
1734 archive_entry_copy_uname(entry, tar->entry_uname.s);
1741 pax_attribute_xattr(struct archive_entry *entry,
1742 const char *name, const char *value)
1745 void *value_decoded;
1748 if (strlen(name) < 18 || (memcmp(name, "LIBARCHIVE.xattr.", 17)) != 0)
1753 /* URL-decode name */
1754 name_decoded = url_decode(name);
1755 if (name_decoded == NULL)
1758 /* Base-64 decode value */
1759 value_decoded = base64_decode(value, strlen(value), &value_len);
1760 if (value_decoded == NULL) {
1765 archive_entry_xattr_add_entry(entry, name_decoded,
1766 value_decoded, value_len);
1769 free(value_decoded);
1774 pax_attribute_schily_xattr(struct archive_entry *entry,
1775 const char *name, const char *value, size_t value_length)
1777 if (strlen(name) < 14 || (memcmp(name, "SCHILY.xattr.", 13)) != 0)
1782 archive_entry_xattr_add_entry(entry, name, value, value_length);
1788 pax_attribute_acl(struct archive_read *a, struct tar *tar,
1789 struct archive_entry *entry, const char *value, int type)
1795 case ARCHIVE_ENTRY_ACL_TYPE_ACCESS:
1796 errstr = "SCHILY.acl.access";
1798 case ARCHIVE_ENTRY_ACL_TYPE_DEFAULT:
1799 errstr = "SCHILY.acl.default";
1801 case ARCHIVE_ENTRY_ACL_TYPE_NFS4:
1802 errstr = "SCHILY.acl.ace";
1805 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1806 "Unknown ACL type: %d", type);
1807 return(ARCHIVE_FATAL);
1810 if (tar->sconv_acl == NULL) {
1812 archive_string_conversion_from_charset(
1813 &(a->archive), "UTF-8", 1);
1814 if (tar->sconv_acl == NULL)
1815 return (ARCHIVE_FATAL);
1818 r = archive_acl_from_text_l(archive_entry_acl(entry), value, type,
1820 if (r != ARCHIVE_OK) {
1821 if (r == ARCHIVE_FATAL) {
1822 archive_set_error(&a->archive, ENOMEM,
1823 "%s %s", "Can't allocate memory for ",
1827 archive_set_error(&a->archive,
1828 ARCHIVE_ERRNO_MISC, "%s %s", "Parse error: ", errstr);
1834 * Parse a single key=value attribute. key/value pointers are
1835 * assumed to point into reasonably long-lived storage.
1837 * Note that POSIX reserves all-lowercase keywords. Vendor-specific
1838 * extensions should always have keywords of the form "VENDOR.attribute"
1839 * In particular, it's quite feasible to support many different
1840 * vendor extensions here. I'm using "LIBARCHIVE" for extensions
1841 * unique to this library.
1843 * Investigate other vendor-specific extensions and see if
1844 * any of them look useful.
1847 pax_attribute(struct archive_read *a, struct tar *tar,
1848 struct archive_entry *entry, const char *key, const char *value, size_t value_length)
1852 int err = ARCHIVE_OK, r;
1855 value = ""; /* Disable compiler warning; do not pass
1856 * NULL pointer to strlen(). */
1859 /* Reject GNU.sparse.* headers on non-regular files. */
1860 if (strncmp(key, "GNU.sparse", 10) == 0 &&
1861 !tar->sparse_allowed) {
1862 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1863 "Non-regular file cannot be sparse");
1864 return (ARCHIVE_FATAL);
1867 /* GNU "0.0" sparse pax format. */
1868 if (strcmp(key, "GNU.sparse.numblocks") == 0) {
1869 tar->sparse_offset = -1;
1870 tar->sparse_numbytes = -1;
1871 tar->sparse_gnu_major = 0;
1872 tar->sparse_gnu_minor = 0;
1874 if (strcmp(key, "GNU.sparse.offset") == 0) {
1875 tar->sparse_offset = tar_atol10(value, strlen(value));
1876 if (tar->sparse_numbytes != -1) {
1877 if (gnu_add_sparse_entry(a, tar,
1878 tar->sparse_offset, tar->sparse_numbytes)
1880 return (ARCHIVE_FATAL);
1881 tar->sparse_offset = -1;
1882 tar->sparse_numbytes = -1;
1885 if (strcmp(key, "GNU.sparse.numbytes") == 0) {
1886 tar->sparse_numbytes = tar_atol10(value, strlen(value));
1887 if (tar->sparse_numbytes != -1) {
1888 if (gnu_add_sparse_entry(a, tar,
1889 tar->sparse_offset, tar->sparse_numbytes)
1891 return (ARCHIVE_FATAL);
1892 tar->sparse_offset = -1;
1893 tar->sparse_numbytes = -1;
1896 if (strcmp(key, "GNU.sparse.size") == 0) {
1897 tar->realsize = tar_atol10(value, strlen(value));
1898 archive_entry_set_size(entry, tar->realsize);
1899 tar->realsize_override = 1;
1902 /* GNU "0.1" sparse pax format. */
1903 if (strcmp(key, "GNU.sparse.map") == 0) {
1904 tar->sparse_gnu_major = 0;
1905 tar->sparse_gnu_minor = 1;
1906 if (gnu_sparse_01_parse(a, tar, value) != ARCHIVE_OK)
1907 return (ARCHIVE_WARN);
1910 /* GNU "1.0" sparse pax format */
1911 if (strcmp(key, "GNU.sparse.major") == 0) {
1912 tar->sparse_gnu_major = (int)tar_atol10(value, strlen(value));
1913 tar->sparse_gnu_pending = 1;
1915 if (strcmp(key, "GNU.sparse.minor") == 0) {
1916 tar->sparse_gnu_minor = (int)tar_atol10(value, strlen(value));
1917 tar->sparse_gnu_pending = 1;
1919 if (strcmp(key, "GNU.sparse.name") == 0) {
1921 * The real filename; when storing sparse
1922 * files, GNU tar puts a synthesized name into
1923 * the regular 'path' attribute in an attempt
1924 * to limit confusion. ;-)
1926 archive_strcpy(&(tar->entry_pathname_override), value);
1928 if (strcmp(key, "GNU.sparse.realsize") == 0) {
1929 tar->realsize = tar_atol10(value, strlen(value));
1930 archive_entry_set_size(entry, tar->realsize);
1931 tar->realsize_override = 1;
1935 /* Our extensions */
1936 /* TODO: Handle arbitrary extended attributes... */
1938 if (strcmp(key, "LIBARCHIVE.xxxxxxx") == 0)
1939 archive_entry_set_xxxxxx(entry, value);
1941 if (strcmp(key, "LIBARCHIVE.creationtime") == 0) {
1942 pax_time(value, &s, &n);
1943 archive_entry_set_birthtime(entry, s, n);
1945 if (memcmp(key, "LIBARCHIVE.xattr.", 17) == 0)
1946 pax_attribute_xattr(entry, key, value);
1949 /* We support some keys used by the "star" archiver */
1950 if (strcmp(key, "SCHILY.acl.access") == 0) {
1951 r = pax_attribute_acl(a, tar, entry, value,
1952 ARCHIVE_ENTRY_ACL_TYPE_ACCESS);
1953 if (r == ARCHIVE_FATAL)
1955 } else if (strcmp(key, "SCHILY.acl.default") == 0) {
1956 r = pax_attribute_acl(a, tar, entry, value,
1957 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT);
1958 if (r == ARCHIVE_FATAL)
1960 } else if (strcmp(key, "SCHILY.acl.ace") == 0) {
1961 r = pax_attribute_acl(a, tar, entry, value,
1962 ARCHIVE_ENTRY_ACL_TYPE_NFS4);
1963 if (r == ARCHIVE_FATAL)
1965 } else if (strcmp(key, "SCHILY.devmajor") == 0) {
1966 archive_entry_set_rdevmajor(entry,
1967 (dev_t)tar_atol10(value, strlen(value)));
1968 } else if (strcmp(key, "SCHILY.devminor") == 0) {
1969 archive_entry_set_rdevminor(entry,
1970 (dev_t)tar_atol10(value, strlen(value)));
1971 } else if (strcmp(key, "SCHILY.fflags") == 0) {
1972 archive_entry_copy_fflags_text(entry, value);
1973 } else if (strcmp(key, "SCHILY.dev") == 0) {
1974 archive_entry_set_dev(entry,
1975 (dev_t)tar_atol10(value, strlen(value)));
1976 } else if (strcmp(key, "SCHILY.ino") == 0) {
1977 archive_entry_set_ino(entry,
1978 tar_atol10(value, strlen(value)));
1979 } else if (strcmp(key, "SCHILY.nlink") == 0) {
1980 archive_entry_set_nlink(entry, (unsigned)
1981 tar_atol10(value, strlen(value)));
1982 } else if (strcmp(key, "SCHILY.realsize") == 0) {
1983 tar->realsize = tar_atol10(value, strlen(value));
1984 tar->realsize_override = 1;
1985 archive_entry_set_size(entry, tar->realsize);
1986 } else if (strncmp(key, "SCHILY.xattr.", 13) == 0) {
1987 pax_attribute_schily_xattr(entry, key, value,
1989 } else if (strcmp(key, "SUN.holesdata") == 0) {
1990 /* A Solaris extension for sparse. */
1991 r = solaris_sparse_parse(a, tar, entry, value);
1993 if (r == ARCHIVE_FATAL)
1996 archive_set_error(&a->archive,
1998 "Parse error: SUN.holesdata");
2003 if (strcmp(key, "atime") == 0) {
2004 pax_time(value, &s, &n);
2005 archive_entry_set_atime(entry, s, n);
2009 if (strcmp(key, "ctime") == 0) {
2010 pax_time(value, &s, &n);
2011 archive_entry_set_ctime(entry, s, n);
2012 } else if (strcmp(key, "charset") == 0) {
2013 /* TODO: Publish charset information in entry. */
2014 } else if (strcmp(key, "comment") == 0) {
2015 /* TODO: Publish comment in entry. */
2019 if (strcmp(key, "gid") == 0) {
2020 archive_entry_set_gid(entry,
2021 tar_atol10(value, strlen(value)));
2022 } else if (strcmp(key, "gname") == 0) {
2023 archive_strcpy(&(tar->entry_gname), value);
2027 if (strcmp(key, "hdrcharset") == 0) {
2028 if (strcmp(value, "BINARY") == 0)
2030 tar->pax_hdrcharset_binary = 1;
2031 else if (strcmp(value, "ISO-IR 10646 2000 UTF-8") == 0)
2032 tar->pax_hdrcharset_binary = 0;
2036 /* pax interchange doesn't distinguish hardlink vs. symlink. */
2037 if (strcmp(key, "linkpath") == 0) {
2038 archive_strcpy(&(tar->entry_linkpath), value);
2042 if (strcmp(key, "mtime") == 0) {
2043 pax_time(value, &s, &n);
2044 archive_entry_set_mtime(entry, s, n);
2048 if (strcmp(key, "path") == 0) {
2049 archive_strcpy(&(tar->entry_pathname), value);
2053 /* POSIX has reserved 'realtime.*' */
2056 /* POSIX has reserved 'security.*' */
2057 /* Someday: if (strcmp(key, "security.acl") == 0) { ... } */
2058 if (strcmp(key, "size") == 0) {
2059 /* "size" is the size of the data in the entry. */
2060 tar->entry_bytes_remaining
2061 = tar_atol10(value, strlen(value));
2063 * The "size" pax header keyword always overrides the
2064 * "size" field in the tar header.
2065 * GNU.sparse.realsize, GNU.sparse.size and
2066 * SCHILY.realsize override this value.
2068 if (!tar->realsize_override) {
2069 archive_entry_set_size(entry,
2070 tar->entry_bytes_remaining);
2072 = tar->entry_bytes_remaining;
2077 if (strcmp(key, "uid") == 0) {
2078 archive_entry_set_uid(entry,
2079 tar_atol10(value, strlen(value)));
2080 } else if (strcmp(key, "uname") == 0) {
2081 archive_strcpy(&(tar->entry_uname), value);
2091 * parse a decimal time value, which may include a fractional portion
2094 pax_time(const char *p, int64_t *ps, long *pn)
2100 int64_t limit, last_digit_limit;
2102 limit = INT64_MAX / 10;
2103 last_digit_limit = INT64_MAX % 10;
2111 while (*p >= '0' && *p <= '9') {
2114 (s == limit && digit > last_digit_limit)) {
2118 s = (s * 10) + digit;
2124 /* Calculate nanoseconds. */
2133 if (*p >= '0' && *p <= '9')
2134 *pn += (*p - '0') * l;
2141 * Parse GNU tar header
2144 header_gnutar(struct archive_read *a, struct tar *tar,
2145 struct archive_entry *entry, const void *h, size_t *unconsumed)
2147 const struct archive_entry_header_gnutar *header;
2149 int err = ARCHIVE_OK;
2152 * GNU header is like POSIX ustar, except 'prefix' is
2153 * replaced with some other fields. This also means the
2154 * filename is stored as in old-style archives.
2157 /* Grab fields common to all tar variants. */
2158 err = header_common(a, tar, entry, h);
2159 if (err == ARCHIVE_FATAL)
2162 /* Copy filename over (to ensure null termination). */
2163 header = (const struct archive_entry_header_gnutar *)h;
2164 if (archive_entry_copy_pathname_l(entry,
2165 header->name, sizeof(header->name), tar->sconv) != 0) {
2166 err = set_conversion_failed_error(a, tar->sconv, "Pathname");
2167 if (err == ARCHIVE_FATAL)
2171 /* Fields common to ustar and GNU */
2172 /* XXX Can the following be factored out since it's common
2173 * to ustar and gnu tar? Is it okay to move it down into
2174 * header_common, perhaps? */
2175 if (archive_entry_copy_uname_l(entry,
2176 header->uname, sizeof(header->uname), tar->sconv) != 0) {
2177 err = set_conversion_failed_error(a, tar->sconv, "Uname");
2178 if (err == ARCHIVE_FATAL)
2182 if (archive_entry_copy_gname_l(entry,
2183 header->gname, sizeof(header->gname), tar->sconv) != 0) {
2184 err = set_conversion_failed_error(a, tar->sconv, "Gname");
2185 if (err == ARCHIVE_FATAL)
2189 /* Parse out device numbers only for char and block specials */
2190 if (header->typeflag[0] == '3' || header->typeflag[0] == '4') {
2191 archive_entry_set_rdevmajor(entry, (dev_t)
2192 tar_atol(header->rdevmajor, sizeof(header->rdevmajor)));
2193 archive_entry_set_rdevminor(entry, (dev_t)
2194 tar_atol(header->rdevminor, sizeof(header->rdevminor)));
2196 archive_entry_set_rdev(entry, 0);
2198 tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
2200 /* Grab GNU-specific fields. */
2201 t = tar_atol(header->atime, sizeof(header->atime));
2203 archive_entry_set_atime(entry, t, 0);
2204 t = tar_atol(header->ctime, sizeof(header->ctime));
2206 archive_entry_set_ctime(entry, t, 0);
2208 if (header->realsize[0] != 0) {
2210 = tar_atol(header->realsize, sizeof(header->realsize));
2211 archive_entry_set_size(entry, tar->realsize);
2212 tar->realsize_override = 1;
2215 if (header->sparse[0].offset[0] != 0) {
2216 if (gnu_sparse_old_read(a, tar, header, unconsumed)
2218 return (ARCHIVE_FATAL);
2220 if (header->isextended[0] != 0) {
2229 gnu_add_sparse_entry(struct archive_read *a, struct tar *tar,
2230 int64_t offset, int64_t remaining)
2232 struct sparse_block *p;
2234 p = (struct sparse_block *)calloc(1, sizeof(*p));
2236 archive_set_error(&a->archive, ENOMEM, "Out of memory");
2237 return (ARCHIVE_FATAL);
2239 if (tar->sparse_last != NULL)
2240 tar->sparse_last->next = p;
2242 tar->sparse_list = p;
2243 tar->sparse_last = p;
2244 if (remaining < 0 || offset < 0 || offset > INT64_MAX - remaining) {
2245 archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, "Malformed sparse map data");
2246 return (ARCHIVE_FATAL);
2249 p->remaining = remaining;
2250 return (ARCHIVE_OK);
2254 gnu_clear_sparse_list(struct tar *tar)
2256 struct sparse_block *p;
2258 while (tar->sparse_list != NULL) {
2259 p = tar->sparse_list;
2260 tar->sparse_list = p->next;
2263 tar->sparse_last = NULL;
2267 * GNU tar old-format sparse data.
2269 * GNU old-format sparse data is stored in a fixed-field
2270 * format. Offset/size values are 11-byte octal fields (same
2271 * format as 'size' field in ustart header). These are
2272 * stored in the header, allocating subsequent header blocks
2273 * as needed. Extending the header in this way is a pretty
2274 * severe POSIX violation; this design has earned GNU tar a
2279 gnu_sparse_old_read(struct archive_read *a, struct tar *tar,
2280 const struct archive_entry_header_gnutar *header, size_t *unconsumed)
2285 struct gnu_sparse sparse[21];
2289 const struct extended *ext;
2291 if (gnu_sparse_old_parse(a, tar, header->sparse, 4) != ARCHIVE_OK)
2292 return (ARCHIVE_FATAL);
2293 if (header->isextended[0] == 0)
2294 return (ARCHIVE_OK);
2297 tar_flush_unconsumed(a, unconsumed);
2298 data = __archive_read_ahead(a, 512, &bytes_read);
2300 return (ARCHIVE_FATAL);
2301 if (bytes_read < 512) {
2302 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2303 "Truncated tar archive "
2304 "detected while reading sparse file data");
2305 return (ARCHIVE_FATAL);
2308 ext = (const struct extended *)data;
2309 if (gnu_sparse_old_parse(a, tar, ext->sparse, 21) != ARCHIVE_OK)
2310 return (ARCHIVE_FATAL);
2311 } while (ext->isextended[0] != 0);
2312 if (tar->sparse_list != NULL)
2313 tar->entry_offset = tar->sparse_list->offset;
2314 return (ARCHIVE_OK);
2318 gnu_sparse_old_parse(struct archive_read *a, struct tar *tar,
2319 const struct gnu_sparse *sparse, int length)
2321 while (length > 0 && sparse->offset[0] != 0) {
2322 if (gnu_add_sparse_entry(a, tar,
2323 tar_atol(sparse->offset, sizeof(sparse->offset)),
2324 tar_atol(sparse->numbytes, sizeof(sparse->numbytes)))
2326 return (ARCHIVE_FATAL);
2330 return (ARCHIVE_OK);
2334 * GNU tar sparse format 0.0
2336 * Beginning with GNU tar 1.15, sparse files are stored using
2337 * information in the pax extended header. The GNU tar maintainers
2338 * have gone through a number of variations in the process of working
2339 * out this scheme; fortunately, they're all numbered.
2341 * Sparse format 0.0 uses attribute GNU.sparse.numblocks to store the
2342 * number of blocks, and GNU.sparse.offset/GNU.sparse.numbytes to
2343 * store offset/size for each block. The repeated instances of these
2344 * latter fields violate the pax specification (which frowns on
2345 * duplicate keys), so this format was quickly replaced.
2349 * GNU tar sparse format 0.1
2351 * This version replaced the offset/numbytes attributes with
2352 * a single "map" attribute that stored a list of integers. This
2353 * format had two problems: First, the "map" attribute could be very
2354 * long, which caused problems for some implementations. More
2355 * importantly, the sparse data was lost when extracted by archivers
2356 * that didn't recognize this extension.
2360 gnu_sparse_01_parse(struct archive_read *a, struct tar *tar, const char *p)
2363 int64_t offset = -1, size = -1;
2367 while (*e != '\0' && *e != ',') {
2368 if (*e < '0' || *e > '9')
2369 return (ARCHIVE_WARN);
2373 offset = tar_atol10(p, e - p);
2375 return (ARCHIVE_WARN);
2377 size = tar_atol10(p, e - p);
2379 return (ARCHIVE_WARN);
2380 if (gnu_add_sparse_entry(a, tar, offset, size)
2382 return (ARCHIVE_FATAL);
2386 return (ARCHIVE_OK);
2392 * GNU tar sparse format 1.0
2394 * The idea: The offset/size data is stored as a series of base-10
2395 * ASCII numbers prepended to the file data, so that dearchivers that
2396 * don't support this format will extract the block map along with the
2397 * data and a separate post-process can restore the sparseness.
2399 * Unfortunately, GNU tar 1.16 had a bug that added unnecessary
2400 * padding to the body of the file when using this format. GNU tar
2401 * 1.17 corrected this bug without bumping the version number, so
2402 * it's not possible to support both variants. This code supports
2403 * the later variant at the expense of not supporting the former.
2405 * This variant also replaced GNU.sparse.size with GNU.sparse.realsize
2406 * and introduced the GNU.sparse.major/GNU.sparse.minor attributes.
2410 * Read the next line from the input, and parse it as a decimal
2411 * integer followed by '\n'. Returns positive integer value or
2412 * negative on error.
2415 gnu_sparse_10_atol(struct archive_read *a, struct tar *tar,
2416 int64_t *remaining, size_t *unconsumed)
2418 int64_t l, limit, last_digit_limit;
2424 limit = INT64_MAX / base;
2425 last_digit_limit = INT64_MAX % base;
2428 * Skip any lines starting with '#'; GNU tar specs
2429 * don't require this, but they should.
2432 bytes_read = readline(a, tar, &p,
2433 (ssize_t)tar_min(*remaining, 100), unconsumed);
2434 if (bytes_read <= 0)
2435 return (ARCHIVE_FATAL);
2436 *remaining -= bytes_read;
2437 } while (p[0] == '#');
2440 while (bytes_read > 0) {
2443 if (*p < '0' || *p >= '0' + base)
2444 return (ARCHIVE_WARN);
2446 if (l > limit || (l == limit && digit > last_digit_limit))
2447 l = INT64_MAX; /* Truncate on overflow. */
2449 l = (l * base) + digit;
2453 /* TODO: Error message. */
2454 return (ARCHIVE_WARN);
2458 * Returns length (in bytes) of the sparse data description
2462 gnu_sparse_10_read(struct archive_read *a, struct tar *tar, size_t *unconsumed)
2466 int64_t offset, size, to_skip, remaining;
2468 /* Clear out the existing sparse list. */
2469 gnu_clear_sparse_list(tar);
2471 remaining = tar->entry_bytes_remaining;
2473 /* Parse entries. */
2474 entries = (int)gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2476 return (ARCHIVE_FATAL);
2477 /* Parse the individual entries. */
2478 while (entries-- > 0) {
2479 /* Parse offset/size */
2480 offset = gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2482 return (ARCHIVE_FATAL);
2483 size = gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2485 return (ARCHIVE_FATAL);
2486 /* Add a new sparse entry. */
2487 if (gnu_add_sparse_entry(a, tar, offset, size) != ARCHIVE_OK)
2488 return (ARCHIVE_FATAL);
2490 /* Skip rest of block... */
2491 tar_flush_unconsumed(a, unconsumed);
2492 bytes_read = (ssize_t)(tar->entry_bytes_remaining - remaining);
2493 to_skip = 0x1ff & -bytes_read;
2494 /* Fail if tar->entry_bytes_remaing would get negative */
2495 if (to_skip > remaining)
2496 return (ARCHIVE_FATAL);
2497 if (to_skip != __archive_read_consume(a, to_skip))
2498 return (ARCHIVE_FATAL);
2499 return ((ssize_t)(bytes_read + to_skip));
2503 * Solaris pax extension for a sparse file. This is recorded with the
2504 * data and hole pairs. The way recording sparse information by Solaris'
2505 * pax simply indicates where data and sparse are, so the stored contents
2506 * consist of both data and hole.
2509 solaris_sparse_parse(struct archive_read *a, struct tar *tar,
2510 struct archive_entry *entry, const char *p)
2516 (void)entry; /* UNUSED */
2522 return (ARCHIVE_WARN);
2525 while (*e != '\0' && *e != ' ') {
2526 if (*e < '0' || *e > '9')
2527 return (ARCHIVE_WARN);
2531 end = tar_atol10(p, e - p);
2533 return (ARCHIVE_WARN);
2535 if (gnu_add_sparse_entry(a, tar, start,
2536 end - start) != ARCHIVE_OK)
2537 return (ARCHIVE_FATAL);
2538 tar->sparse_last->hole = hole;
2541 return (ARCHIVE_OK);
2548 * Convert text->integer.
2550 * Traditional tar formats (including POSIX) specify base-8 for
2551 * all of the standard numeric fields. This is a significant limitation
2553 * = file size is limited to 8GB
2554 * = rdevmajor and rdevminor are limited to 21 bits
2555 * = uid/gid are limited to 21 bits
2557 * There are two workarounds for this:
2558 * = pax extended headers, which use variable-length string fields
2559 * = GNU tar and STAR both allow either base-8 or base-256 in
2560 * most fields. The high bit is set to indicate base-256.
2562 * On read, this implementation supports both extensions.
2565 tar_atol(const char *p, size_t char_cnt)
2568 * Technically, GNU tar considers a field to be in base-256
2569 * only if the first byte is 0xff or 0x80.
2572 return (tar_atol256(p, char_cnt));
2573 return (tar_atol8(p, char_cnt));
2577 * Note that this implementation does not (and should not!) obey
2578 * locale settings; you cannot simply substitute strtol here, since
2579 * it does obey locale.
2582 tar_atol_base_n(const char *p, size_t char_cnt, int base)
2584 int64_t l, maxval, limit, last_digit_limit;
2588 limit = INT64_MAX / base;
2589 last_digit_limit = INT64_MAX % base;
2591 /* the pointer will not be dereferenced if char_cnt is zero
2592 * due to the way the && operator is evaluated.
2594 while (char_cnt != 0 && (*p == ' ' || *p == '\t')) {
2600 if (char_cnt != 0 && *p == '-') {
2606 limit = -(INT64_MIN / base);
2607 last_digit_limit = INT64_MIN % base;
2611 if (char_cnt != 0) {
2613 while (digit >= 0 && digit < base && char_cnt != 0) {
2614 if (l>limit || (l == limit && digit > last_digit_limit)) {
2615 return maxval; /* Truncate on overflow. */
2617 l = (l * base) + digit;
2622 return (sign < 0) ? -l : l;
2626 tar_atol8(const char *p, size_t char_cnt)
2628 return tar_atol_base_n(p, char_cnt, 8);
2632 tar_atol10(const char *p, size_t char_cnt)
2634 return tar_atol_base_n(p, char_cnt, 10);
2638 * Parse a base-256 integer. This is just a variable-length
2639 * twos-complement signed binary value in big-endian order, except
2640 * that the high-order bit is ignored. The values here can be up to
2641 * 12 bytes, so we need to be careful about overflowing 64-bit
2642 * (8-byte) integers.
2644 * This code unashamedly assumes that the local machine uses 8-bit
2645 * bytes and twos-complement arithmetic.
2648 tar_atol256(const char *_p, size_t char_cnt)
2651 const unsigned char *p = (const unsigned char *)_p;
2652 unsigned char c, neg;
2654 /* Extend 7-bit 2s-comp to 8-bit 2s-comp, decide sign. */
2659 l = ~ARCHIVE_LITERAL_ULL(0);
2666 /* If more than 8 bytes, check that we can ignore
2667 * high-order bits without overflow. */
2668 while (char_cnt > sizeof(int64_t)) {
2671 return neg ? INT64_MIN : INT64_MAX;
2675 /* c is first byte that fits; if sign mismatch, return overflow */
2676 if ((c ^ neg) & 0x80) {
2677 return neg ? INT64_MIN : INT64_MAX;
2680 /* Accumulate remaining bytes. */
2681 while (--char_cnt > 0) {
2686 /* Return signed twos-complement value. */
2687 return (int64_t)(l);
2691 * Returns length of line (including trailing newline)
2692 * or negative on error. 'start' argument is updated to
2693 * point to first character of line. This avoids copying
2697 readline(struct archive_read *a, struct tar *tar, const char **start,
2698 ssize_t limit, size_t *unconsumed)
2701 ssize_t total_size = 0;
2706 tar_flush_unconsumed(a, unconsumed);
2708 t = __archive_read_ahead(a, 1, &bytes_read);
2709 if (bytes_read <= 0)
2710 return (ARCHIVE_FATAL);
2711 s = t; /* Start of line? */
2712 p = memchr(t, '\n', bytes_read);
2713 /* If we found '\n' in the read buffer, return pointer to that. */
2715 bytes_read = 1 + ((const char *)p) - s;
2716 if (bytes_read > limit) {
2717 archive_set_error(&a->archive,
2718 ARCHIVE_ERRNO_FILE_FORMAT,
2720 return (ARCHIVE_FATAL);
2722 *unconsumed = bytes_read;
2724 return (bytes_read);
2726 *unconsumed = bytes_read;
2727 /* Otherwise, we need to accumulate in a line buffer. */
2729 if (total_size + bytes_read > limit) {
2730 archive_set_error(&a->archive,
2731 ARCHIVE_ERRNO_FILE_FORMAT,
2733 return (ARCHIVE_FATAL);
2735 if (archive_string_ensure(&tar->line, total_size + bytes_read) == NULL) {
2736 archive_set_error(&a->archive, ENOMEM,
2737 "Can't allocate working buffer");
2738 return (ARCHIVE_FATAL);
2740 memcpy(tar->line.s + total_size, t, bytes_read);
2741 tar_flush_unconsumed(a, unconsumed);
2742 total_size += bytes_read;
2743 /* If we found '\n', clean up and return. */
2745 *start = tar->line.s;
2746 return (total_size);
2748 /* Read some more. */
2749 t = __archive_read_ahead(a, 1, &bytes_read);
2750 if (bytes_read <= 0)
2751 return (ARCHIVE_FATAL);
2752 s = t; /* Start of line? */
2753 p = memchr(t, '\n', bytes_read);
2754 /* If we found '\n', trim the read. */
2756 bytes_read = 1 + ((const char *)p) - s;
2758 *unconsumed = bytes_read;
2763 * base64_decode - Base64 decode
2765 * This accepts most variations of base-64 encoding, including:
2766 * * with or without line breaks
2767 * * with or without the final group padded with '=' or '_' characters
2768 * (The most economical Base-64 variant does not pad the last group and
2769 * omits line breaks; RFC1341 used for MIME requires both.)
2772 base64_decode(const char *s, size_t len, size_t *out_len)
2774 static const unsigned char digits[64] = {
2775 'A','B','C','D','E','F','G','H','I','J','K','L','M','N',
2776 'O','P','Q','R','S','T','U','V','W','X','Y','Z','a','b',
2777 'c','d','e','f','g','h','i','j','k','l','m','n','o','p',
2778 'q','r','s','t','u','v','w','x','y','z','0','1','2','3',
2779 '4','5','6','7','8','9','+','/' };
2780 static unsigned char decode_table[128];
2782 const unsigned char *src = (const unsigned char *)s;
2784 /* If the decode table is not yet initialized, prepare it. */
2785 if (decode_table[digits[1]] != 1) {
2787 memset(decode_table, 0xff, sizeof(decode_table));
2788 for (i = 0; i < sizeof(digits); i++)
2789 decode_table[digits[i]] = i;
2792 /* Allocate enough space to hold the entire output. */
2793 /* Note that we may not use all of this... */
2794 out = (char *)malloc(len - len / 4 + 1);
2802 /* Collect the next group of (up to) four characters. */
2805 while (group_size < 4 && len > 0) {
2806 /* '=' or '_' padding indicates final group. */
2807 if (*src == '=' || *src == '_') {
2811 /* Skip illegal characters (including line breaks) */
2812 if (*src > 127 || *src < 32
2813 || decode_table[*src] == 0xff) {
2819 v |= decode_table[*src++];
2823 /* Align a short group properly. */
2824 v <<= 6 * (4 - group_size);
2825 /* Unpack the group we just collected. */
2826 switch (group_size) {
2827 case 4: d[2] = v & 0xff;
2829 case 3: d[1] = (v >> 8) & 0xff;
2831 case 2: d[0] = (v >> 16) & 0xff;
2833 case 1: /* this is invalid! */
2836 d += group_size * 3 / 4;
2844 url_decode(const char *in)
2849 out = (char *)malloc(strlen(in) + 1);
2852 for (s = in, d = out; *s != '\0'; ) {
2853 if (s[0] == '%' && s[1] != '\0' && s[2] != '\0') {
2854 /* Try to convert % escape */
2855 int digit1 = tohex(s[1]);
2856 int digit2 = tohex(s[2]);
2857 if (digit1 >= 0 && digit2 >= 0) {
2858 /* Looks good, consume three chars */
2860 /* Convert output */
2861 *d++ = ((digit1 << 4) | digit2);
2864 /* Else fall through and treat '%' as normal char */
2875 if (c >= '0' && c <= '9')
2877 else if (c >= 'A' && c <= 'F')
2878 return (c - 'A' + 10);
2879 else if (c >= 'a' && c <= 'f')
2880 return (c - 'a' + 10);