archive_read_support_format_tar.c revision 362133 
1/*-
2 * Copyright (c) 2003-2007 Tim Kientzle
3 * Copyright (c) 2011-2012 Michihiro NAKAJIMA
4 * Copyright (c) 2016 Martin Matuska
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
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.
15 *
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.
26 */
27
28#include "archive_platform.h"
29__FBSDID("$FreeBSD: stable/11/contrib/libarchive/libarchive/archive_read_support_format_tar.c 362133 2020-06-12 23:01:29Z mm $");
30
31#ifdef HAVE_ERRNO_H
32#include <errno.h>
33#endif
34#include <stddef.h>
35#ifdef HAVE_STDLIB_H
36#include <stdlib.h>
37#endif
38#ifdef HAVE_STRING_H
39#include <string.h>
40#endif
41
42#include "archive.h"
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"
48
49#define tar_min(a,b) ((a) < (b) ? (a) : (b))
50
51/*
52 * Layout of POSIX 'ustar' tar header.
53 */
54struct archive_entry_header_ustar {
55	char	name[100];
56	char	mode[8];
57	char	uid[8];
58	char	gid[8];
59	char	size[12];
60	char	mtime[12];
61	char	checksum[8];
62	char	typeflag[1];
63	char	linkname[100];	/* "old format" header ends here */
64	char	magic[6];	/* For POSIX: "ustar\0" */
65	char	version[2];	/* For POSIX: "00" */
66	char	uname[32];
67	char	gname[32];
68	char	rdevmajor[8];
69	char	rdevminor[8];
70	char	prefix[155];
71};
72
73/*
74 * Structure of GNU tar header
75 */
76struct gnu_sparse {
77	char	offset[12];
78	char	numbytes[12];
79};
80
81struct archive_entry_header_gnutar {
82	char	name[100];
83	char	mode[8];
84	char	uid[8];
85	char	gid[8];
86	char	size[12];
87	char	mtime[12];
88	char	checksum[8];
89	char	typeflag[1];
90	char	linkname[100];
91	char	magic[8];  /* "ustar  \0" (note blank/blank/null at end) */
92	char	uname[32];
93	char	gname[32];
94	char	rdevmajor[8];
95	char	rdevminor[8];
96	char	atime[12];
97	char	ctime[12];
98	char	offset[12];
99	char	longnames[4];
100	char	unused[1];
101	struct gnu_sparse sparse[4];
102	char	isextended[1];
103	char	realsize[12];
104	/*
105	 * Old GNU format doesn't use POSIX 'prefix' field; they use
106	 * the 'L' (longname) entry instead.
107	 */
108};
109
110/*
111 * Data specific to this format.
112 */
113struct sparse_block {
114	struct sparse_block	*next;
115	int64_t	offset;
116	int64_t	remaining;
117	int hole;
118};
119
120struct tar {
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;
139	int64_t			 realsize;
140	int			 sparse_allowed;
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;
148
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;
155	int			 compat_2x;
156	int			 process_mac_extensions;
157	int			 read_concatenated_archives;
158	int			 realsize_override;
159};
160
161static int	archive_block_is_null(const char *p);
162static char	*base64_decode(const char *, size_t, size_t *);
163static int	gnu_add_sparse_entry(struct archive_read *, struct tar *,
164		    int64_t offset, int64_t remaining);
165
166static void	gnu_clear_sparse_list(struct tar *);
167static int	gnu_sparse_old_read(struct archive_read *, struct tar *,
168		    const struct archive_entry_header_gnutar *header, size_t *);
169static int	gnu_sparse_old_parse(struct archive_read *, struct tar *,
170		    const struct gnu_sparse *sparse, int length);
171static int	gnu_sparse_01_parse(struct archive_read *, struct tar *,
172		    const char *);
173static ssize_t	gnu_sparse_10_read(struct archive_read *, struct tar *,
174			size_t *);
175static int	header_Solaris_ACL(struct archive_read *,  struct tar *,
176		    struct archive_entry *, const void *, size_t *);
177static int	header_common(struct archive_read *,  struct tar *,
178		    struct archive_entry *, const void *);
179static int	header_old_tar(struct archive_read *, struct tar *,
180		    struct archive_entry *, const void *);
181static int	header_pax_extensions(struct archive_read *, struct tar *,
182		    struct archive_entry *, const void *, size_t *);
183static int	header_pax_global(struct archive_read *, struct tar *,
184		    struct archive_entry *, const void *h, size_t *);
185static int	header_longlink(struct archive_read *, struct tar *,
186		    struct archive_entry *, const void *h, size_t *);
187static int	header_longname(struct archive_read *, struct tar *,
188		    struct archive_entry *, const void *h, size_t *);
189static int	read_mac_metadata_blob(struct archive_read *, struct tar *,
190		    struct archive_entry *, const void *h, size_t *);
191static int	header_volume(struct archive_read *, struct tar *,
192		    struct archive_entry *, const void *h, size_t *);
193static int	header_ustar(struct archive_read *, struct tar *,
194		    struct archive_entry *, const void *h);
195static int	header_gnutar(struct archive_read *, struct tar *,
196		    struct archive_entry *, const void *h, size_t *);
197static int	archive_read_format_tar_bid(struct archive_read *, int);
198static int	archive_read_format_tar_options(struct archive_read *,
199		    const char *, const char *);
200static int	archive_read_format_tar_cleanup(struct archive_read *);
201static int	archive_read_format_tar_read_data(struct archive_read *a,
202		    const void **buff, size_t *size, int64_t *offset);
203static int	archive_read_format_tar_skip(struct archive_read *a);
204static int	archive_read_format_tar_read_header(struct archive_read *,
205		    struct archive_entry *);
206static int	checksum(struct archive_read *, const void *);
207static int 	pax_attribute(struct archive_read *, struct tar *,
208		    struct archive_entry *, const char *key, const char *value,
209		    size_t value_length);
210static int	pax_attribute_acl(struct archive_read *, struct tar *,
211		    struct archive_entry *, const char *, int);
212static int	pax_attribute_xattr(struct archive_entry *, const char *,
213		    const char *);
214static int 	pax_header(struct archive_read *, struct tar *,
215		    struct archive_entry *, struct archive_string *);
216static void	pax_time(const char *, int64_t *sec, long *nanos);
217static ssize_t	readline(struct archive_read *, struct tar *, const char **,
218		    ssize_t limit, size_t *);
219static int	read_body_to_string(struct archive_read *, struct tar *,
220		    struct archive_string *, const void *h, size_t *);
221static int	solaris_sparse_parse(struct archive_read *, struct tar *,
222		    struct archive_entry *, const char *);
223static int64_t	tar_atol(const char *, size_t);
224static int64_t	tar_atol10(const char *, size_t);
225static int64_t	tar_atol256(const char *, size_t);
226static int64_t	tar_atol8(const char *, size_t);
227static int	tar_read_header(struct archive_read *, struct tar *,
228		    struct archive_entry *, size_t *);
229static int	tohex(int c);
230static char	*url_decode(const char *);
231static void	tar_flush_unconsumed(struct archive_read *, size_t *);
232
233
234int
235archive_read_support_format_gnutar(struct archive *a)
236{
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));
240}
241
242
243int
244archive_read_support_format_tar(struct archive *_a)
245{
246	struct archive_read *a = (struct archive_read *)_a;
247	struct tar *tar;
248	int r;
249
250	archive_check_magic(_a, ARCHIVE_READ_MAGIC,
251	    ARCHIVE_STATE_NEW, "archive_read_support_format_tar");
252
253	tar = (struct tar *)calloc(1, sizeof(*tar));
254	if (tar == NULL) {
255		archive_set_error(&a->archive, ENOMEM,
256		    "Can't allocate tar data");
257		return (ARCHIVE_FATAL);
258	}
259#ifdef HAVE_COPYFILE_H
260	/* Set this by default on Mac OS. */
261	tar->process_mac_extensions = 1;
262#endif
263
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,
270	    NULL,
271	    archive_read_format_tar_cleanup,
272	    NULL,
273	    NULL);
274
275	if (r != ARCHIVE_OK)
276		free(tar);
277	return (ARCHIVE_OK);
278}
279
280static int
281archive_read_format_tar_cleanup(struct archive_read *a)
282{
283	struct tar *tar;
284
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);
299	free(tar);
300	(a->format->data) = NULL;
301	return (ARCHIVE_OK);
302}
303
304/*
305 * Validate number field
306 *
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
316 *
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
323 *
324 * This should tolerate all variants in use.  It will reject a field
325 * where the writer just left garbage after a trailing NUL.
326 */
327static int
328validate_number_field(const char* p_field, size_t i_size)
329{
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. */
333		return 1;
334	} else {
335		/* Must be octal */
336		size_t i = 0;
337		/* Skip any leading spaces */
338		while (i < i_size && p_field[i] == ' ') {
339			++i;
340		}
341		/* Skip octal digits. */
342		while (i < i_size && p_field[i] >= '0' && p_field[i] <= '7') {
343			++i;
344		}
345		/* Any remaining characters must be space or NUL padding. */
346		while (i < i_size) {
347			if (p_field[i] != ' ' && p_field[i] != 0) {
348				return 0;
349			}
350			++i;
351		}
352		return 1;
353	}
354}
355
356static int
357archive_read_format_tar_bid(struct archive_read *a, int best_bid)
358{
359	int bid;
360	const char *h;
361	const struct archive_entry_header_ustar *header;
362
363	(void)best_bid; /* UNUSED */
364
365	bid = 0;
366
367	/* Now let's look at the actual header and see if it matches. */
368	h = __archive_read_ahead(a, 512, NULL);
369	if (h == NULL)
370		return (-1);
371
372	/* If it's an end-of-archive mark, we can handle it. */
373	if (h[0] == 0 && archive_block_is_null(h)) {
374		/*
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.
378		 */
379		return (10);
380	}
381
382	/* If it's not an end-of-archive mark, it must have a valid checksum.*/
383	if (!checksum(a, h))
384		return (0);
385	bid += 48;  /* Checksum is usually 6 octal digits. */
386
387	header = (const struct archive_entry_header_ustar *)h;
388
389	/* Recognize POSIX formats. */
390	if ((memcmp(header->magic, "ustar\0", 6) == 0)
391	    && (memcmp(header->version, "00", 2) == 0))
392		bid += 56;
393
394	/* Recognize GNU tar format. */
395	if ((memcmp(header->magic, "ustar ", 6) == 0)
396	    && (memcmp(header->version, " \0", 2) == 0))
397		bid += 56;
398
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') )
404		return (0);
405	bid += 2;  /* 6 bits of variation in an 8-bit field leaves 2 bits. */
406
407	/*
408	 * Check format of mode/uid/gid/mtime/size/rdevmajor/rdevminor fields.
409	 */
410	if (bid > 0 && (
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)) {
418		bid = 0;
419	}
420
421	return (bid);
422}
423
424static int
425archive_read_format_tar_options(struct archive_read *a,
426    const char *key, const char *val)
427{
428	struct tar *tar;
429	int ret = ARCHIVE_FAILED;
430
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;
436		return (ARCHIVE_OK);
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");
441		else {
442			tar->opt_sconv =
443			    archive_string_conversion_from_charset(
444				&a->archive, val, 0);
445			if (tar->opt_sconv != NULL)
446				ret = ARCHIVE_OK;
447			else
448				ret = ARCHIVE_FATAL;
449		}
450		return (ret);
451	} else if (strcmp(key, "mac-ext") == 0) {
452		tar->process_mac_extensions = (val != NULL && val[0] != 0);
453		return (ARCHIVE_OK);
454	} else if (strcmp(key, "read_concatenated_archives") == 0) {
455		tar->read_concatenated_archives = (val != NULL && val[0] != 0);
456		return (ARCHIVE_OK);
457	}
458
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);
463}
464
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
468 */
469static void
470tar_flush_unconsumed(struct archive_read *a, size_t *unconsumed)
471{
472	if (*unconsumed) {
473/*
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
478		if (data) {
479			memset(data, 0xff, *unconsumed);
480		}
481*/
482		__archive_read_consume(a, *unconsumed);
483		*unconsumed = 0;
484	}
485}
486
487/*
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.
491 */
492static int
493archive_read_format_tar_read_header(struct archive_read *a,
494    struct archive_entry *entry)
495{
496	/*
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
502	 * useful.
503	 *
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.
509	 */
510	static int default_inode;
511	static int default_dev;
512	struct tar *tar;
513	const char *p;
514	const wchar_t *wp;
515	int r;
516	size_t l, unconsumed = 0;
517
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) {
523		++default_dev;
524		default_inode = 0;
525	}
526
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;
532
533	/* Setup default string conversion. */
534	tar->sconv = tar->opt_sconv;
535	if (tar->sconv == NULL) {
536		if (!tar->init_default_conversion) {
537			tar->sconv_default =
538			    archive_string_default_conversion_for_read(&(a->archive));
539			tar->init_default_conversion = 1;
540		}
541		tar->sconv = tar->sconv_default;
542	}
543
544	r = tar_read_header(a, tar, entry, &unconsumed);
545
546	tar_flush_unconsumed(a, &unconsumed);
547
548	/*
549	 * "non-sparse" files are really just sparse files with
550	 * a single block.
551	 */
552	if (tar->sparse_list == NULL) {
553		if (gnu_add_sparse_entry(a, tar, 0, tar->entry_bytes_remaining)
554		    != ARCHIVE_OK)
555			return (ARCHIVE_FATAL);
556	} else {
557		struct sparse_block *sb;
558
559		for (sb = tar->sparse_list; sb != NULL; sb = sb->next) {
560			if (!sb->hole)
561				archive_entry_sparse_add_entry(entry,
562				    sb->offset, sb->remaining);
563		}
564	}
565
566	if (r == ARCHIVE_OK && archive_entry_filetype(entry) == AE_IFREG) {
567		/*
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.
571		 */
572		if ((wp = archive_entry_pathname_w(entry)) != NULL) {
573			l = wcslen(wp);
574			if (l > 0 && wp[l - 1] == L'/') {
575				archive_entry_set_filetype(entry, AE_IFDIR);
576			}
577		} else if ((p = archive_entry_pathname(entry)) != NULL) {
578			l = strlen(p);
579			if (l > 0 && p[l - 1] == '/') {
580				archive_entry_set_filetype(entry, AE_IFDIR);
581			}
582		}
583	}
584	return (r);
585}
586
587static int
588archive_read_format_tar_read_data(struct archive_read *a,
589    const void **buff, size_t *size, int64_t *offset)
590{
591	ssize_t bytes_read;
592	struct tar *tar;
593	struct sparse_block *p;
594
595	tar = (struct tar *)(a->format->data);
596
597	for (;;) {
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;
603			free(p);
604		}
605
606		if (tar->entry_bytes_unconsumed) {
607			__archive_read_consume(a, tar->entry_bytes_unconsumed);
608			tar->entry_bytes_unconsumed = 0;
609		}
610
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;
617			*buff = NULL;
618			*size = 0;
619			*offset = tar->realsize;
620			return (ARCHIVE_EOF);
621		}
622
623		*buff = __archive_read_ahead(a, 1, &bytes_read);
624		if (bytes_read < 0)
625			return (ARCHIVE_FATAL);
626		if (*buff == NULL) {
627			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
628			    "Truncated tar archive");
629			return (ARCHIVE_FATAL);
630		}
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;
637		*size = bytes_read;
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;
643
644		if (!tar->sparse_list->hole)
645			return (ARCHIVE_OK);
646		/* Current is hole data and skip this. */
647	}
648}
649
650static int
651archive_read_format_tar_skip(struct archive_read *a)
652{
653	int64_t bytes_skipped;
654	int64_t request;
655	struct sparse_block *p;
656	struct tar* tar;
657
658	tar = (struct tar *)(a->format->data);
659
660	/* Do not consume the hole of a sparse file. */
661	request = 0;
662	for (p = tar->sparse_list; p != NULL; p = p->next) {
663		if (!p->hole) {
664			if (p->remaining >= INT64_MAX - request) {
665				return ARCHIVE_FATAL;
666			}
667			request += p->remaining;
668		}
669	}
670	if (request > tar->entry_bytes_remaining)
671		request = tar->entry_bytes_remaining;
672	request += tar->entry_padding + tar->entry_bytes_unconsumed;
673
674	bytes_skipped = __archive_read_consume(a, request);
675	if (bytes_skipped < 0)
676		return (ARCHIVE_FATAL);
677
678	tar->entry_bytes_remaining = 0;
679	tar->entry_bytes_unconsumed = 0;
680	tar->entry_padding = 0;
681
682	/* Free the sparse list. */
683	gnu_clear_sparse_list(tar);
684
685	return (ARCHIVE_OK);
686}
687
688/*
689 * This function recursively interprets all of the headers associated
690 * with a single entry.
691 */
692static int
693tar_read_header(struct archive_read *a, struct tar *tar,
694    struct archive_entry *entry, size_t *unconsumed)
695{
696	ssize_t bytes;
697	int err, eof_vol_header;
698	const char *h;
699	const struct archive_entry_header_ustar *header;
700	const struct archive_entry_header_gnutar *gnuheader;
701
702	eof_vol_header = 0;
703
704	/* Loop until we find a workable header record. */
705	for (;;) {
706		tar_flush_unconsumed(a, unconsumed);
707
708		/* Read 512-byte header record */
709		h = __archive_read_ahead(a, 512, &bytes);
710		if (bytes < 0)
711			return ((int)bytes);
712		if (bytes == 0) { /* EOF at a block boundary. */
713			/* Some writers do omit the block of nulls. <sigh> */
714			return (ARCHIVE_EOF);
715		}
716		if (bytes < 512) {  /* Short block at EOF; this is bad. */
717			archive_set_error(&a->archive,
718			    ARCHIVE_ERRNO_FILE_FORMAT,
719			    "Truncated tar archive");
720			return (ARCHIVE_FATAL);
721		}
722		*unconsumed = 512;
723
724		/* Header is workable if it's not an end-of-archive mark. */
725		if (h[0] != 0 || !archive_block_is_null(h))
726			break;
727
728		/* Ensure format is set for archives with only null blocks. */
729		if (a->archive.archive_format_name == NULL) {
730			a->archive.archive_format = ARCHIVE_FORMAT_TAR;
731			a->archive.archive_format_name = "tar";
732		}
733
734		if (!tar->read_concatenated_archives) {
735			/* Try to consume a second all-null record, as well. */
736			tar_flush_unconsumed(a, unconsumed);
737			h = __archive_read_ahead(a, 512, NULL);
738			if (h != NULL && h[0] == 0 && archive_block_is_null(h))
739				__archive_read_consume(a, 512);
740			archive_clear_error(&a->archive);
741			return (ARCHIVE_EOF);
742		}
743
744		/*
745		 * We're reading concatenated archives, ignore this block and
746		 * loop to get the next.
747		 */
748	}
749
750	/*
751	 * Note: If the checksum fails and we return ARCHIVE_RETRY,
752	 * then the client is likely to just retry.  This is a very
753	 * crude way to search for the next valid header!
754	 *
755	 * TODO: Improve this by implementing a real header scan.
756	 */
757	if (!checksum(a, h)) {
758		tar_flush_unconsumed(a, unconsumed);
759		archive_set_error(&a->archive, EINVAL, "Damaged tar archive");
760		return (ARCHIVE_RETRY); /* Retryable: Invalid header */
761	}
762
763	if (++tar->header_recursion_depth > 32) {
764		tar_flush_unconsumed(a, unconsumed);
765		archive_set_error(&a->archive, EINVAL, "Too many special headers");
766		return (ARCHIVE_WARN);
767	}
768
769	/* Determine the format variant. */
770	header = (const struct archive_entry_header_ustar *)h;
771
772	switch(header->typeflag[0]) {
773	case 'A': /* Solaris tar ACL */
774		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
775		a->archive.archive_format_name = "Solaris tar";
776		err = header_Solaris_ACL(a, tar, entry, h, unconsumed);
777		break;
778	case 'g': /* POSIX-standard 'g' header. */
779		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
780		a->archive.archive_format_name = "POSIX pax interchange format";
781		err = header_pax_global(a, tar, entry, h, unconsumed);
782		if (err == ARCHIVE_EOF)
783			return (err);
784		break;
785	case 'K': /* Long link name (GNU tar, others) */
786		err = header_longlink(a, tar, entry, h, unconsumed);
787		break;
788	case 'L': /* Long filename (GNU tar, others) */
789		err = header_longname(a, tar, entry, h, unconsumed);
790		break;
791	case 'V': /* GNU volume header */
792		err = header_volume(a, tar, entry, h, unconsumed);
793		if (err == ARCHIVE_EOF)
794			eof_vol_header = 1;
795		break;
796	case 'X': /* Used by SUN tar; same as 'x'. */
797		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
798		a->archive.archive_format_name =
799		    "POSIX pax interchange format (Sun variant)";
800		err = header_pax_extensions(a, tar, entry, h, unconsumed);
801		break;
802	case 'x': /* POSIX-standard 'x' header. */
803		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
804		a->archive.archive_format_name = "POSIX pax interchange format";
805		err = header_pax_extensions(a, tar, entry, h, unconsumed);
806		break;
807	default:
808		gnuheader = (const struct archive_entry_header_gnutar *)h;
809		if (memcmp(gnuheader->magic, "ustar  \0", 8) == 0) {
810			a->archive.archive_format = ARCHIVE_FORMAT_TAR_GNUTAR;
811			a->archive.archive_format_name = "GNU tar format";
812			err = header_gnutar(a, tar, entry, h, unconsumed);
813		} else if (memcmp(header->magic, "ustar", 5) == 0) {
814			if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) {
815				a->archive.archive_format = ARCHIVE_FORMAT_TAR_USTAR;
816				a->archive.archive_format_name = "POSIX ustar format";
817			}
818			err = header_ustar(a, tar, entry, h);
819		} else {
820			a->archive.archive_format = ARCHIVE_FORMAT_TAR;
821			a->archive.archive_format_name = "tar (non-POSIX)";
822			err = header_old_tar(a, tar, entry, h);
823		}
824	}
825	if (err == ARCHIVE_FATAL)
826		return (err);
827
828	tar_flush_unconsumed(a, unconsumed);
829
830	h = NULL;
831	header = NULL;
832
833	--tar->header_recursion_depth;
834	/* Yuck.  Apple's design here ends up storing long pathname
835	 * extensions for both the AppleDouble extension entry and the
836	 * regular entry.
837	 */
838	if ((err == ARCHIVE_WARN || err == ARCHIVE_OK) &&
839	    tar->header_recursion_depth == 0 &&
840	    tar->process_mac_extensions) {
841		int err2 = read_mac_metadata_blob(a, tar, entry, h, unconsumed);
842		if (err2 < err)
843			err = err2;
844	}
845
846	/* We return warnings or success as-is.  Anything else is fatal. */
847	if (err == ARCHIVE_WARN || err == ARCHIVE_OK) {
848		if (tar->sparse_gnu_pending) {
849			if (tar->sparse_gnu_major == 1 &&
850			    tar->sparse_gnu_minor == 0) {
851				ssize_t bytes_read;
852
853				tar->sparse_gnu_pending = 0;
854				/* Read initial sparse map. */
855				bytes_read = gnu_sparse_10_read(a, tar, unconsumed);
856				if (bytes_read < 0)
857					return ((int)bytes_read);
858				tar->entry_bytes_remaining -= bytes_read;
859			} else {
860				archive_set_error(&a->archive,
861				    ARCHIVE_ERRNO_MISC,
862				    "Unrecognized GNU sparse file format");
863				return (ARCHIVE_WARN);
864			}
865			tar->sparse_gnu_pending = 0;
866		}
867		return (err);
868	}
869	if (err == ARCHIVE_EOF) {
870		if (!eof_vol_header) {
871			/* EOF when recursively reading a header is bad. */
872			archive_set_error(&a->archive, EINVAL,
873			    "Damaged tar archive");
874		} else {
875			/* If we encounter just a GNU volume header treat
876			 * this situation as an empty archive */
877			return (ARCHIVE_EOF);
878		}
879	}
880	return (ARCHIVE_FATAL);
881}
882
883/*
884 * Return true if block checksum is correct.
885 */
886static int
887checksum(struct archive_read *a, const void *h)
888{
889	const unsigned char *bytes;
890	const struct archive_entry_header_ustar	*header;
891	int check, sum;
892	size_t i;
893
894	(void)a; /* UNUSED */
895	bytes = (const unsigned char *)h;
896	header = (const struct archive_entry_header_ustar *)h;
897
898	/* Checksum field must hold an octal number */
899	for (i = 0; i < sizeof(header->checksum); ++i) {
900		char c = header->checksum[i];
901		if (c != ' ' && c != '\0' && (c < '0' || c > '7'))
902			return 0;
903	}
904
905	/*
906	 * Test the checksum.  Note that POSIX specifies _unsigned_
907	 * bytes for this calculation.
908	 */
909	sum = (int)tar_atol(header->checksum, sizeof(header->checksum));
910	check = 0;
911	for (i = 0; i < 148; i++)
912		check += (unsigned char)bytes[i];
913	for (; i < 156; i++)
914		check += 32;
915	for (; i < 512; i++)
916		check += (unsigned char)bytes[i];
917	if (sum == check)
918		return (1);
919
920	/*
921	 * Repeat test with _signed_ bytes, just in case this archive
922	 * was created by an old BSD, Solaris, or HP-UX tar with a
923	 * broken checksum calculation.
924	 */
925	check = 0;
926	for (i = 0; i < 148; i++)
927		check += (signed char)bytes[i];
928	for (; i < 156; i++)
929		check += 32;
930	for (; i < 512; i++)
931		check += (signed char)bytes[i];
932	if (sum == check)
933		return (1);
934
935	return (0);
936}
937
938/*
939 * Return true if this block contains only nulls.
940 */
941static int
942archive_block_is_null(const char *p)
943{
944	unsigned i;
945
946	for (i = 0; i < 512; i++)
947		if (*p++)
948			return (0);
949	return (1);
950}
951
952/*
953 * Interpret 'A' Solaris ACL header
954 */
955static int
956header_Solaris_ACL(struct archive_read *a, struct tar *tar,
957    struct archive_entry *entry, const void *h, size_t *unconsumed)
958{
959	const struct archive_entry_header_ustar *header;
960	size_t size;
961	int err, acl_type;
962	int64_t type;
963	char *acl, *p;
964
965	/*
966	 * read_body_to_string adds a NUL terminator, but we need a little
967	 * more to make sure that we don't overrun acl_text later.
968	 */
969	header = (const struct archive_entry_header_ustar *)h;
970	size = (size_t)tar_atol(header->size, sizeof(header->size));
971	err = read_body_to_string(a, tar, &(tar->acl_text), h, unconsumed);
972	if (err != ARCHIVE_OK)
973		return (err);
974
975	/* Recursively read next header */
976	err = tar_read_header(a, tar, entry, unconsumed);
977	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
978		return (err);
979
980	/* TODO: Examine the first characters to see if this
981	 * is an AIX ACL descriptor.  We'll likely never support
982	 * them, but it would be polite to recognize and warn when
983	 * we do see them. */
984
985	/* Leading octal number indicates ACL type and number of entries. */
986	p = acl = tar->acl_text.s;
987	type = 0;
988	while (*p != '\0' && p < acl + size) {
989		if (*p < '0' || *p > '7') {
990			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
991			    "Malformed Solaris ACL attribute (invalid digit)");
992			return(ARCHIVE_WARN);
993		}
994		type <<= 3;
995		type += *p - '0';
996		if (type > 077777777) {
997			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
998			    "Malformed Solaris ACL attribute (count too large)");
999			return (ARCHIVE_WARN);
1000		}
1001		p++;
1002	}
1003	switch ((int)type & ~0777777) {
1004	case 01000000:
1005		/* POSIX.1e ACL */
1006		acl_type = ARCHIVE_ENTRY_ACL_TYPE_ACCESS;
1007		break;
1008	case 03000000:
1009		/* NFSv4 ACL */
1010		acl_type = ARCHIVE_ENTRY_ACL_TYPE_NFS4;
1011		break;
1012	default:
1013		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1014		    "Malformed Solaris ACL attribute (unsupported type %o)",
1015		    (int)type);
1016		return (ARCHIVE_WARN);
1017	}
1018	p++;
1019
1020	if (p >= acl + size) {
1021		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1022		    "Malformed Solaris ACL attribute (body overflow)");
1023		return(ARCHIVE_WARN);
1024	}
1025
1026	/* ACL text is null-terminated; find the end. */
1027	size -= (p - acl);
1028	acl = p;
1029
1030	while (*p != '\0' && p < acl + size)
1031		p++;
1032
1033	if (tar->sconv_acl == NULL) {
1034		tar->sconv_acl = archive_string_conversion_from_charset(
1035		    &(a->archive), "UTF-8", 1);
1036		if (tar->sconv_acl == NULL)
1037			return (ARCHIVE_FATAL);
1038	}
1039	archive_strncpy(&(tar->localname), acl, p - acl);
1040	err = archive_acl_from_text_l(archive_entry_acl(entry),
1041	    tar->localname.s, acl_type, tar->sconv_acl);
1042	if (err != ARCHIVE_OK) {
1043		if (errno == ENOMEM) {
1044			archive_set_error(&a->archive, ENOMEM,
1045			    "Can't allocate memory for ACL");
1046		} else
1047			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1048			    "Malformed Solaris ACL attribute (unparsable)");
1049	}
1050	return (err);
1051}
1052
1053/*
1054 * Interpret 'K' long linkname header.
1055 */
1056static int
1057header_longlink(struct archive_read *a, struct tar *tar,
1058    struct archive_entry *entry, const void *h, size_t *unconsumed)
1059{
1060	int err;
1061
1062	err = read_body_to_string(a, tar, &(tar->longlink), h, unconsumed);
1063	if (err != ARCHIVE_OK)
1064		return (err);
1065	err = tar_read_header(a, tar, entry, unconsumed);
1066	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1067		return (err);
1068	/* Set symlink if symlink already set, else hardlink. */
1069	archive_entry_copy_link(entry, tar->longlink.s);
1070	return (ARCHIVE_OK);
1071}
1072
1073static int
1074set_conversion_failed_error(struct archive_read *a,
1075    struct archive_string_conv *sconv, const char *name)
1076{
1077	if (errno == ENOMEM) {
1078		archive_set_error(&a->archive, ENOMEM,
1079		    "Can't allocate memory for %s", name);
1080		return (ARCHIVE_FATAL);
1081	}
1082	archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1083	    "%s can't be converted from %s to current locale.",
1084	    name, archive_string_conversion_charset_name(sconv));
1085	return (ARCHIVE_WARN);
1086}
1087
1088/*
1089 * Interpret 'L' long filename header.
1090 */
1091static int
1092header_longname(struct archive_read *a, struct tar *tar,
1093    struct archive_entry *entry, const void *h, size_t *unconsumed)
1094{
1095	int err;
1096
1097	err = read_body_to_string(a, tar, &(tar->longname), h, unconsumed);
1098	if (err != ARCHIVE_OK)
1099		return (err);
1100	/* Read and parse "real" header, then override name. */
1101	err = tar_read_header(a, tar, entry, unconsumed);
1102	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1103		return (err);
1104	if (archive_entry_copy_pathname_l(entry, tar->longname.s,
1105	    archive_strlen(&(tar->longname)), tar->sconv) != 0)
1106		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1107	return (err);
1108}
1109
1110
1111/*
1112 * Interpret 'V' GNU tar volume header.
1113 */
1114static int
1115header_volume(struct archive_read *a, struct tar *tar,
1116    struct archive_entry *entry, const void *h, size_t *unconsumed)
1117{
1118	(void)h;
1119
1120	/* Just skip this and read the next header. */
1121	return (tar_read_header(a, tar, entry, unconsumed));
1122}
1123
1124/*
1125 * Read body of an archive entry into an archive_string object.
1126 */
1127static int
1128read_body_to_string(struct archive_read *a, struct tar *tar,
1129    struct archive_string *as, const void *h, size_t *unconsumed)
1130{
1131	int64_t size;
1132	const struct archive_entry_header_ustar *header;
1133	const void *src;
1134
1135	(void)tar; /* UNUSED */
1136	header = (const struct archive_entry_header_ustar *)h;
1137	size  = tar_atol(header->size, sizeof(header->size));
1138	if ((size > 1048576) || (size < 0)) {
1139		archive_set_error(&a->archive, EINVAL,
1140		    "Special header too large");
1141		return (ARCHIVE_FATAL);
1142	}
1143
1144	/* Fail if we can't make our buffer big enough. */
1145	if (archive_string_ensure(as, (size_t)size+1) == NULL) {
1146		archive_set_error(&a->archive, ENOMEM,
1147		    "No memory");
1148		return (ARCHIVE_FATAL);
1149	}
1150
1151	tar_flush_unconsumed(a, unconsumed);
1152
1153	/* Read the body into the string. */
1154	*unconsumed = (size_t)((size + 511) & ~ 511);
1155	src = __archive_read_ahead(a, *unconsumed, NULL);
1156	if (src == NULL) {
1157		*unconsumed = 0;
1158		return (ARCHIVE_FATAL);
1159	}
1160	memcpy(as->s, src, (size_t)size);
1161	as->s[size] = '\0';
1162	as->length = (size_t)size;
1163	return (ARCHIVE_OK);
1164}
1165
1166/*
1167 * Parse out common header elements.
1168 *
1169 * This would be the same as header_old_tar, except that the
1170 * filename is handled slightly differently for old and POSIX
1171 * entries  (POSIX entries support a 'prefix').  This factoring
1172 * allows header_old_tar and header_ustar
1173 * to handle filenames differently, while still putting most of the
1174 * common parsing into one place.
1175 */
1176static int
1177header_common(struct archive_read *a, struct tar *tar,
1178    struct archive_entry *entry, const void *h)
1179{
1180	const struct archive_entry_header_ustar	*header;
1181	char	tartype;
1182	int     err = ARCHIVE_OK;
1183
1184	header = (const struct archive_entry_header_ustar *)h;
1185	if (header->linkname[0])
1186		archive_strncpy(&(tar->entry_linkpath),
1187		    header->linkname, sizeof(header->linkname));
1188	else
1189		archive_string_empty(&(tar->entry_linkpath));
1190
1191	/* Parse out the numeric fields (all are octal) */
1192	archive_entry_set_mode(entry,
1193		(mode_t)tar_atol(header->mode, sizeof(header->mode)));
1194	archive_entry_set_uid(entry, tar_atol(header->uid, sizeof(header->uid)));
1195	archive_entry_set_gid(entry, tar_atol(header->gid, sizeof(header->gid)));
1196	tar->entry_bytes_remaining = tar_atol(header->size, sizeof(header->size));
1197	if (tar->entry_bytes_remaining < 0) {
1198		tar->entry_bytes_remaining = 0;
1199		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1200		    "Tar entry has negative size");
1201		return (ARCHIVE_FATAL);
1202	}
1203	if (tar->entry_bytes_remaining == INT64_MAX) {
1204		/* Note: tar_atol returns INT64_MAX on overflow */
1205		tar->entry_bytes_remaining = 0;
1206		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1207		    "Tar entry size overflow");
1208		return (ARCHIVE_FATAL);
1209	}
1210	tar->realsize = tar->entry_bytes_remaining;
1211	archive_entry_set_size(entry, tar->entry_bytes_remaining);
1212	archive_entry_set_mtime(entry, tar_atol(header->mtime, sizeof(header->mtime)), 0);
1213
1214	/* Handle the tar type flag appropriately. */
1215	tartype = header->typeflag[0];
1216
1217	switch (tartype) {
1218	case '1': /* Hard link */
1219		if (archive_entry_copy_hardlink_l(entry, tar->entry_linkpath.s,
1220		    archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) {
1221			err = set_conversion_failed_error(a, tar->sconv,
1222			    "Linkname");
1223			if (err == ARCHIVE_FATAL)
1224				return (err);
1225		}
1226		/*
1227		 * The following may seem odd, but: Technically, tar
1228		 * does not store the file type for a "hard link"
1229		 * entry, only the fact that it is a hard link.  So, I
1230		 * leave the type zero normally.  But, pax interchange
1231		 * format allows hard links to have data, which
1232		 * implies that the underlying entry is a regular
1233		 * file.
1234		 */
1235		if (archive_entry_size(entry) > 0)
1236			archive_entry_set_filetype(entry, AE_IFREG);
1237
1238		/*
1239		 * A tricky point: Traditionally, tar readers have
1240		 * ignored the size field when reading hardlink
1241		 * entries, and some writers put non-zero sizes even
1242		 * though the body is empty.  POSIX blessed this
1243		 * convention in the 1988 standard, but broke with
1244		 * this tradition in 2001 by permitting hardlink
1245		 * entries to store valid bodies in pax interchange
1246		 * format, but not in ustar format.  Since there is no
1247		 * hard and fast way to distinguish pax interchange
1248		 * from earlier archives (the 'x' and 'g' entries are
1249		 * optional, after all), we need a heuristic.
1250		 */
1251		if (archive_entry_size(entry) == 0) {
1252			/* If the size is already zero, we're done. */
1253		}  else if (a->archive.archive_format
1254		    == ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) {
1255			/* Definitely pax extended; must obey hardlink size. */
1256		} else if (a->archive.archive_format == ARCHIVE_FORMAT_TAR
1257		    || a->archive.archive_format == ARCHIVE_FORMAT_TAR_GNUTAR)
1258		{
1259			/* Old-style or GNU tar: we must ignore the size. */
1260			archive_entry_set_size(entry, 0);
1261			tar->entry_bytes_remaining = 0;
1262		} else if (archive_read_format_tar_bid(a, 50) > 50) {
1263			/*
1264			 * We don't know if it's pax: If the bid
1265			 * function sees a valid ustar header
1266			 * immediately following, then let's ignore
1267			 * the hardlink size.
1268			 */
1269			archive_entry_set_size(entry, 0);
1270			tar->entry_bytes_remaining = 0;
1271		}
1272		/*
1273		 * TODO: There are still two cases I'd like to handle:
1274		 *   = a ustar non-pax archive with a hardlink entry at
1275		 *     end-of-archive.  (Look for block of nulls following?)
1276		 *   = a pax archive that has not seen any pax headers
1277		 *     and has an entry which is a hardlink entry storing
1278		 *     a body containing an uncompressed tar archive.
1279		 * The first is worth addressing; I don't see any reliable
1280		 * way to deal with the second possibility.
1281		 */
1282		break;
1283	case '2': /* Symlink */
1284		archive_entry_set_filetype(entry, AE_IFLNK);
1285		archive_entry_set_size(entry, 0);
1286		tar->entry_bytes_remaining = 0;
1287		if (archive_entry_copy_symlink_l(entry, tar->entry_linkpath.s,
1288		    archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) {
1289			err = set_conversion_failed_error(a, tar->sconv,
1290			    "Linkname");
1291			if (err == ARCHIVE_FATAL)
1292				return (err);
1293		}
1294		break;
1295	case '3': /* Character device */
1296		archive_entry_set_filetype(entry, AE_IFCHR);
1297		archive_entry_set_size(entry, 0);
1298		tar->entry_bytes_remaining = 0;
1299		break;
1300	case '4': /* Block device */
1301		archive_entry_set_filetype(entry, AE_IFBLK);
1302		archive_entry_set_size(entry, 0);
1303		tar->entry_bytes_remaining = 0;
1304		break;
1305	case '5': /* Dir */
1306		archive_entry_set_filetype(entry, AE_IFDIR);
1307		archive_entry_set_size(entry, 0);
1308		tar->entry_bytes_remaining = 0;
1309		break;
1310	case '6': /* FIFO device */
1311		archive_entry_set_filetype(entry, AE_IFIFO);
1312		archive_entry_set_size(entry, 0);
1313		tar->entry_bytes_remaining = 0;
1314		break;
1315	case 'D': /* GNU incremental directory type */
1316		/*
1317		 * No special handling is actually required here.
1318		 * It might be nice someday to preprocess the file list and
1319		 * provide it to the client, though.
1320		 */
1321		archive_entry_set_filetype(entry, AE_IFDIR);
1322		break;
1323	case 'M': /* GNU "Multi-volume" (remainder of file from last archive)*/
1324		/*
1325		 * As far as I can tell, this is just like a regular file
1326		 * entry, except that the contents should be _appended_ to
1327		 * the indicated file at the indicated offset.  This may
1328		 * require some API work to fully support.
1329		 */
1330		break;
1331	case 'N': /* Old GNU "long filename" entry. */
1332		/* The body of this entry is a script for renaming
1333		 * previously-extracted entries.  Ugh.  It will never
1334		 * be supported by libarchive. */
1335		archive_entry_set_filetype(entry, AE_IFREG);
1336		break;
1337	case 'S': /* GNU sparse files */
1338		/*
1339		 * Sparse files are really just regular files with
1340		 * sparse information in the extended area.
1341		 */
1342		/* FALLTHROUGH */
1343	case '0':
1344		/*
1345		 * Enable sparse file "read" support only for regular
1346		 * files and explicit GNU sparse files.  However, we
1347		 * don't allow non-standard file types to be sparse.
1348		 */
1349		tar->sparse_allowed = 1;
1350		/* FALLTHROUGH */
1351	default: /* Regular file  and non-standard types */
1352		/*
1353		 * Per POSIX: non-recognized types should always be
1354		 * treated as regular files.
1355		 */
1356		archive_entry_set_filetype(entry, AE_IFREG);
1357		break;
1358	}
1359	return (err);
1360}
1361
1362/*
1363 * Parse out header elements for "old-style" tar archives.
1364 */
1365static int
1366header_old_tar(struct archive_read *a, struct tar *tar,
1367    struct archive_entry *entry, const void *h)
1368{
1369	const struct archive_entry_header_ustar	*header;
1370	int err = ARCHIVE_OK, err2;
1371
1372	/* Copy filename over (to ensure null termination). */
1373	header = (const struct archive_entry_header_ustar *)h;
1374	if (archive_entry_copy_pathname_l(entry,
1375	    header->name, sizeof(header->name), tar->sconv) != 0) {
1376		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1377		if (err == ARCHIVE_FATAL)
1378			return (err);
1379	}
1380
1381	/* Grab rest of common fields */
1382	err2 = header_common(a, tar, entry, h);
1383	if (err > err2)
1384		err = err2;
1385
1386	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1387	return (err);
1388}
1389
1390/*
1391 * Read a Mac AppleDouble-encoded blob of file metadata,
1392 * if there is one.
1393 */
1394static int
1395read_mac_metadata_blob(struct archive_read *a, struct tar *tar,
1396    struct archive_entry *entry, const void *h, size_t *unconsumed)
1397{
1398	int64_t size;
1399	const void *data;
1400	const char *p, *name;
1401	const wchar_t *wp, *wname;
1402
1403	(void)h; /* UNUSED */
1404
1405	wname = wp = archive_entry_pathname_w(entry);
1406	if (wp != NULL) {
1407		/* Find the last path element. */
1408		for (; *wp != L'\0'; ++wp) {
1409			if (wp[0] == '/' && wp[1] != L'\0')
1410				wname = wp + 1;
1411		}
1412		/*
1413		 * If last path element starts with "._", then
1414		 * this is a Mac extension.
1415		 */
1416		if (wname[0] != L'.' || wname[1] != L'_' || wname[2] == L'\0')
1417			return ARCHIVE_OK;
1418	} else {
1419		/* Find the last path element. */
1420		name = p = archive_entry_pathname(entry);
1421		if (p == NULL)
1422			return (ARCHIVE_FAILED);
1423		for (; *p != '\0'; ++p) {
1424			if (p[0] == '/' && p[1] != '\0')
1425				name = p + 1;
1426		}
1427		/*
1428		 * If last path element starts with "._", then
1429		 * this is a Mac extension.
1430		 */
1431		if (name[0] != '.' || name[1] != '_' || name[2] == '\0')
1432			return ARCHIVE_OK;
1433	}
1434
1435 	/* Read the body as a Mac OS metadata blob. */
1436	size = archive_entry_size(entry);
1437
1438	/*
1439	 * TODO: Look beyond the body here to peek at the next header.
1440	 * If it's a regular header (not an extension header)
1441	 * that has the wrong name, just return the current
1442	 * entry as-is, without consuming the body here.
1443	 * That would reduce the risk of us mis-identifying
1444	 * an ordinary file that just happened to have
1445	 * a name starting with "._".
1446	 *
1447	 * Q: Is the above idea really possible?  Even
1448	 * when there are GNU or pax extension entries?
1449	 */
1450	data = __archive_read_ahead(a, (size_t)size, NULL);
1451	if (data == NULL) {
1452		*unconsumed = 0;
1453		return (ARCHIVE_FATAL);
1454	}
1455	archive_entry_copy_mac_metadata(entry, data, (size_t)size);
1456	*unconsumed = (size_t)((size + 511) & ~ 511);
1457	tar_flush_unconsumed(a, unconsumed);
1458	return (tar_read_header(a, tar, entry, unconsumed));
1459}
1460
1461/*
1462 * Parse a file header for a pax extended archive entry.
1463 */
1464static int
1465header_pax_global(struct archive_read *a, struct tar *tar,
1466    struct archive_entry *entry, const void *h, size_t *unconsumed)
1467{
1468	int err;
1469
1470	err = read_body_to_string(a, tar, &(tar->pax_global), h, unconsumed);
1471	if (err != ARCHIVE_OK)
1472		return (err);
1473	err = tar_read_header(a, tar, entry, unconsumed);
1474	return (err);
1475}
1476
1477static int
1478header_pax_extensions(struct archive_read *a, struct tar *tar,
1479    struct archive_entry *entry, const void *h, size_t *unconsumed)
1480{
1481	int err, err2;
1482
1483	err = read_body_to_string(a, tar, &(tar->pax_header), h, unconsumed);
1484	if (err != ARCHIVE_OK)
1485		return (err);
1486
1487	/* Parse the next header. */
1488	err = tar_read_header(a, tar, entry, unconsumed);
1489	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1490		return (err);
1491
1492	/*
1493	 * TODO: Parse global/default options into 'entry' struct here
1494	 * before handling file-specific options.
1495	 *
1496	 * This design (parse standard header, then overwrite with pax
1497	 * extended attribute data) usually works well, but isn't ideal;
1498	 * it would be better to parse the pax extended attributes first
1499	 * and then skip any fields in the standard header that were
1500	 * defined in the pax header.
1501	 */
1502	err2 = pax_header(a, tar, entry, &tar->pax_header);
1503	err =  err_combine(err, err2);
1504	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1505	return (err);
1506}
1507
1508
1509/*
1510 * Parse a file header for a Posix "ustar" archive entry.  This also
1511 * handles "pax" or "extended ustar" entries.
1512 */
1513static int
1514header_ustar(struct archive_read *a, struct tar *tar,
1515    struct archive_entry *entry, const void *h)
1516{
1517	const struct archive_entry_header_ustar	*header;
1518	struct archive_string *as;
1519	int err = ARCHIVE_OK, r;
1520
1521	header = (const struct archive_entry_header_ustar *)h;
1522
1523	/* Copy name into an internal buffer to ensure null-termination. */
1524	as = &(tar->entry_pathname);
1525	if (header->prefix[0]) {
1526		archive_strncpy(as, header->prefix, sizeof(header->prefix));
1527		if (as->s[archive_strlen(as) - 1] != '/')
1528			archive_strappend_char(as, '/');
1529		archive_strncat(as, header->name, sizeof(header->name));
1530	} else {
1531		archive_strncpy(as, header->name, sizeof(header->name));
1532	}
1533	if (archive_entry_copy_pathname_l(entry, as->s, archive_strlen(as),
1534	    tar->sconv) != 0) {
1535		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1536		if (err == ARCHIVE_FATAL)
1537			return (err);
1538	}
1539
1540	/* Handle rest of common fields. */
1541	r = header_common(a, tar, entry, h);
1542	if (r == ARCHIVE_FATAL)
1543		return (r);
1544	if (r < err)
1545		err = r;
1546
1547	/* Handle POSIX ustar fields. */
1548	if (archive_entry_copy_uname_l(entry,
1549	    header->uname, sizeof(header->uname), tar->sconv) != 0) {
1550		err = set_conversion_failed_error(a, tar->sconv, "Uname");
1551		if (err == ARCHIVE_FATAL)
1552			return (err);
1553	}
1554
1555	if (archive_entry_copy_gname_l(entry,
1556	    header->gname, sizeof(header->gname), tar->sconv) != 0) {
1557		err = set_conversion_failed_error(a, tar->sconv, "Gname");
1558		if (err == ARCHIVE_FATAL)
1559			return (err);
1560	}
1561
1562	/* Parse out device numbers only for char and block specials. */
1563	if (header->typeflag[0] == '3' || header->typeflag[0] == '4') {
1564		archive_entry_set_rdevmajor(entry, (dev_t)
1565		    tar_atol(header->rdevmajor, sizeof(header->rdevmajor)));
1566		archive_entry_set_rdevminor(entry, (dev_t)
1567		    tar_atol(header->rdevminor, sizeof(header->rdevminor)));
1568	}
1569
1570	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1571
1572	return (err);
1573}
1574
1575
1576/*
1577 * Parse the pax extended attributes record.
1578 *
1579 * Returns non-zero if there's an error in the data.
1580 */
1581static int
1582pax_header(struct archive_read *a, struct tar *tar,
1583    struct archive_entry *entry, struct archive_string *in_as)
1584{
1585	size_t attr_length, l, line_length, value_length;
1586	char *p;
1587	char *key, *value;
1588	struct archive_string *as;
1589	struct archive_string_conv *sconv;
1590	int err, err2;
1591	char *attr = in_as->s;
1592
1593	attr_length = in_as->length;
1594	tar->pax_hdrcharset_binary = 0;
1595	archive_string_empty(&(tar->entry_gname));
1596	archive_string_empty(&(tar->entry_linkpath));
1597	archive_string_empty(&(tar->entry_pathname));
1598	archive_string_empty(&(tar->entry_pathname_override));
1599	archive_string_empty(&(tar->entry_uname));
1600	err = ARCHIVE_OK;
1601	while (attr_length > 0) {
1602		/* Parse decimal length field at start of line. */
1603		line_length = 0;
1604		l = attr_length;
1605		p = attr; /* Record start of line. */
1606		while (l>0) {
1607			if (*p == ' ') {
1608				p++;
1609				l--;
1610				break;
1611			}
1612			if (*p < '0' || *p > '9') {
1613				archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1614				    "Ignoring malformed pax extended attributes");
1615				return (ARCHIVE_WARN);
1616			}
1617			line_length *= 10;
1618			line_length += *p - '0';
1619			if (line_length > 999999) {
1620				archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1621				    "Rejecting pax extended attribute > 1MB");
1622				return (ARCHIVE_WARN);
1623			}
1624			p++;
1625			l--;
1626		}
1627
1628		/*
1629		 * Parsed length must be no bigger than available data,
1630		 * at least 1, and the last character of the line must
1631		 * be '\n'.
1632		 */
1633		if (line_length > attr_length
1634		    || line_length < 1
1635		    || attr[line_length - 1] != '\n')
1636		{
1637			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1638			    "Ignoring malformed pax extended attribute");
1639			return (ARCHIVE_WARN);
1640		}
1641
1642		/* Null-terminate the line. */
1643		attr[line_length - 1] = '\0';
1644
1645		/* Find end of key and null terminate it. */
1646		key = p;
1647		if (key[0] == '=')
1648			return (-1);
1649		while (*p && *p != '=')
1650			++p;
1651		if (*p == '\0') {
1652			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1653			    "Invalid pax extended attributes");
1654			return (ARCHIVE_WARN);
1655		}
1656		*p = '\0';
1657
1658		value = p + 1;
1659
1660		/* Some values may be binary data */
1661		value_length = attr + line_length - 1 - value;
1662
1663		/* Identify this attribute and set it in the entry. */
1664		err2 = pax_attribute(a, tar, entry, key, value, value_length);
1665		if (err2 == ARCHIVE_FATAL)
1666			return (err2);
1667		err = err_combine(err, err2);
1668
1669		/* Skip to next line */
1670		attr += line_length;
1671		attr_length -= line_length;
1672	}
1673
1674	/*
1675	 * PAX format uses UTF-8 as default charset for its metadata
1676	 * unless hdrcharset=BINARY is present in its header.
1677	 * We apply the charset specified by the hdrcharset option only
1678	 * when the hdrcharset attribute(in PAX header) is BINARY because
1679	 * we respect the charset described in PAX header and BINARY also
1680	 * means that metadata(filename,uname and gname) character-set
1681	 * is unknown.
1682	 */
1683	if (tar->pax_hdrcharset_binary)
1684		sconv = tar->opt_sconv;
1685	else {
1686		sconv = archive_string_conversion_from_charset(
1687		    &(a->archive), "UTF-8", 1);
1688		if (sconv == NULL)
1689			return (ARCHIVE_FATAL);
1690		if (tar->compat_2x)
1691			archive_string_conversion_set_opt(sconv,
1692			    SCONV_SET_OPT_UTF8_LIBARCHIVE2X);
1693	}
1694
1695	if (archive_strlen(&(tar->entry_gname)) > 0) {
1696		if (archive_entry_copy_gname_l(entry, tar->entry_gname.s,
1697		    archive_strlen(&(tar->entry_gname)), sconv) != 0) {
1698			err = set_conversion_failed_error(a, sconv, "Gname");
1699			if (err == ARCHIVE_FATAL)
1700				return (err);
1701			/* Use a converted an original name. */
1702			archive_entry_copy_gname(entry, tar->entry_gname.s);
1703		}
1704	}
1705	if (archive_strlen(&(tar->entry_linkpath)) > 0) {
1706		if (archive_entry_copy_link_l(entry, tar->entry_linkpath.s,
1707		    archive_strlen(&(tar->entry_linkpath)), sconv) != 0) {
1708			err = set_conversion_failed_error(a, sconv, "Linkname");
1709			if (err == ARCHIVE_FATAL)
1710				return (err);
1711			/* Use a converted an original name. */
1712			archive_entry_copy_link(entry, tar->entry_linkpath.s);
1713		}
1714	}
1715	/*
1716	 * Some extensions (such as the GNU sparse file extensions)
1717	 * deliberately store a synthetic name under the regular 'path'
1718	 * attribute and the real file name under a different attribute.
1719	 * Since we're supposed to not care about the order, we
1720	 * have no choice but to store all of the various filenames
1721	 * we find and figure it all out afterwards.  This is the
1722	 * figuring out part.
1723	 */
1724	as = NULL;
1725	if (archive_strlen(&(tar->entry_pathname_override)) > 0)
1726		as = &(tar->entry_pathname_override);
1727	else if (archive_strlen(&(tar->entry_pathname)) > 0)
1728		as = &(tar->entry_pathname);
1729	if (as != NULL) {
1730		if (archive_entry_copy_pathname_l(entry, as->s,
1731		    archive_strlen(as), sconv) != 0) {
1732			err = set_conversion_failed_error(a, sconv, "Pathname");
1733			if (err == ARCHIVE_FATAL)
1734				return (err);
1735			/* Use a converted an original name. */
1736			archive_entry_copy_pathname(entry, as->s);
1737		}
1738	}
1739	if (archive_strlen(&(tar->entry_uname)) > 0) {
1740		if (archive_entry_copy_uname_l(entry, tar->entry_uname.s,
1741		    archive_strlen(&(tar->entry_uname)), sconv) != 0) {
1742			err = set_conversion_failed_error(a, sconv, "Uname");
1743			if (err == ARCHIVE_FATAL)
1744				return (err);
1745			/* Use a converted an original name. */
1746			archive_entry_copy_uname(entry, tar->entry_uname.s);
1747		}
1748	}
1749	return (err);
1750}
1751
1752static int
1753pax_attribute_xattr(struct archive_entry *entry,
1754	const char *name, const char *value)
1755{
1756	char *name_decoded;
1757	void *value_decoded;
1758	size_t value_len;
1759
1760	if (strlen(name) < 18 || (memcmp(name, "LIBARCHIVE.xattr.", 17)) != 0)
1761		return 3;
1762
1763	name += 17;
1764
1765	/* URL-decode name */
1766	name_decoded = url_decode(name);
1767	if (name_decoded == NULL)
1768		return 2;
1769
1770	/* Base-64 decode value */
1771	value_decoded = base64_decode(value, strlen(value), &value_len);
1772	if (value_decoded == NULL) {
1773		free(name_decoded);
1774		return 1;
1775	}
1776
1777	archive_entry_xattr_add_entry(entry, name_decoded,
1778		value_decoded, value_len);
1779
1780	free(name_decoded);
1781	free(value_decoded);
1782	return 0;
1783}
1784
1785static int
1786pax_attribute_schily_xattr(struct archive_entry *entry,
1787	const char *name, const char *value, size_t value_length)
1788{
1789	if (strlen(name) < 14 || (memcmp(name, "SCHILY.xattr.", 13)) != 0)
1790		return 1;
1791
1792	name += 13;
1793
1794	archive_entry_xattr_add_entry(entry, name, value, value_length);
1795
1796	return 0;
1797}
1798
1799static int
1800pax_attribute_rht_security_selinux(struct archive_entry *entry,
1801	const char *value, size_t value_length)
1802{
1803	archive_entry_xattr_add_entry(entry, "security.selinux",
1804            value, value_length);
1805
1806	return 0;
1807}
1808
1809static int
1810pax_attribute_acl(struct archive_read *a, struct tar *tar,
1811    struct archive_entry *entry, const char *value, int type)
1812{
1813	int r;
1814	const char* errstr;
1815
1816	switch (type) {
1817	case ARCHIVE_ENTRY_ACL_TYPE_ACCESS:
1818		errstr = "SCHILY.acl.access";
1819		break;
1820	case ARCHIVE_ENTRY_ACL_TYPE_DEFAULT:
1821		errstr = "SCHILY.acl.default";
1822		break;
1823	case ARCHIVE_ENTRY_ACL_TYPE_NFS4:
1824		errstr = "SCHILY.acl.ace";
1825		break;
1826	default:
1827		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1828		    "Unknown ACL type: %d", type);
1829		return(ARCHIVE_FATAL);
1830	}
1831
1832	if (tar->sconv_acl == NULL) {
1833		tar->sconv_acl =
1834		    archive_string_conversion_from_charset(
1835			&(a->archive), "UTF-8", 1);
1836		if (tar->sconv_acl == NULL)
1837			return (ARCHIVE_FATAL);
1838	}
1839
1840	r = archive_acl_from_text_l(archive_entry_acl(entry), value, type,
1841	    tar->sconv_acl);
1842	if (r != ARCHIVE_OK) {
1843		if (r == ARCHIVE_FATAL) {
1844			archive_set_error(&a->archive, ENOMEM,
1845			    "%s %s", "Can't allocate memory for ",
1846			    errstr);
1847			return (r);
1848		}
1849		archive_set_error(&a->archive,
1850		    ARCHIVE_ERRNO_MISC, "%s %s", "Parse error: ", errstr);
1851	}
1852	return (r);
1853}
1854
1855/*
1856 * Parse a single key=value attribute.  key/value pointers are
1857 * assumed to point into reasonably long-lived storage.
1858 *
1859 * Note that POSIX reserves all-lowercase keywords.  Vendor-specific
1860 * extensions should always have keywords of the form "VENDOR.attribute"
1861 * In particular, it's quite feasible to support many different
1862 * vendor extensions here.  I'm using "LIBARCHIVE" for extensions
1863 * unique to this library.
1864 *
1865 * Investigate other vendor-specific extensions and see if
1866 * any of them look useful.
1867 */
1868static int
1869pax_attribute(struct archive_read *a, struct tar *tar,
1870    struct archive_entry *entry, const char *key, const char *value, size_t value_length)
1871{
1872	int64_t s;
1873	long n;
1874	int err = ARCHIVE_OK, r;
1875
1876#ifndef __FreeBSD__
1877	if (value == NULL)
1878		value = "";	/* Disable compiler warning; do not pass
1879				 * NULL pointer to strlen().  */
1880#endif
1881	switch (key[0]) {
1882	case 'G':
1883		/* Reject GNU.sparse.* headers on non-regular files. */
1884		if (strncmp(key, "GNU.sparse", 10) == 0 &&
1885		    !tar->sparse_allowed) {
1886			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1887			    "Non-regular file cannot be sparse");
1888			return (ARCHIVE_FATAL);
1889		}
1890
1891		/* GNU "0.0" sparse pax format. */
1892		if (strcmp(key, "GNU.sparse.numblocks") == 0) {
1893			tar->sparse_offset = -1;
1894			tar->sparse_numbytes = -1;
1895			tar->sparse_gnu_major = 0;
1896			tar->sparse_gnu_minor = 0;
1897		}
1898		if (strcmp(key, "GNU.sparse.offset") == 0) {
1899			tar->sparse_offset = tar_atol10(value, strlen(value));
1900			if (tar->sparse_numbytes != -1) {
1901				if (gnu_add_sparse_entry(a, tar,
1902				    tar->sparse_offset, tar->sparse_numbytes)
1903				    != ARCHIVE_OK)
1904					return (ARCHIVE_FATAL);
1905				tar->sparse_offset = -1;
1906				tar->sparse_numbytes = -1;
1907			}
1908		}
1909		if (strcmp(key, "GNU.sparse.numbytes") == 0) {
1910			tar->sparse_numbytes = tar_atol10(value, strlen(value));
1911			if (tar->sparse_numbytes != -1) {
1912				if (gnu_add_sparse_entry(a, tar,
1913				    tar->sparse_offset, tar->sparse_numbytes)
1914				    != ARCHIVE_OK)
1915					return (ARCHIVE_FATAL);
1916				tar->sparse_offset = -1;
1917				tar->sparse_numbytes = -1;
1918			}
1919		}
1920		if (strcmp(key, "GNU.sparse.size") == 0) {
1921			tar->realsize = tar_atol10(value, strlen(value));
1922			archive_entry_set_size(entry, tar->realsize);
1923			tar->realsize_override = 1;
1924		}
1925
1926		/* GNU "0.1" sparse pax format. */
1927		if (strcmp(key, "GNU.sparse.map") == 0) {
1928			tar->sparse_gnu_major = 0;
1929			tar->sparse_gnu_minor = 1;
1930			if (gnu_sparse_01_parse(a, tar, value) != ARCHIVE_OK)
1931				return (ARCHIVE_WARN);
1932		}
1933
1934		/* GNU "1.0" sparse pax format */
1935		if (strcmp(key, "GNU.sparse.major") == 0) {
1936			tar->sparse_gnu_major = (int)tar_atol10(value, strlen(value));
1937			tar->sparse_gnu_pending = 1;
1938		}
1939		if (strcmp(key, "GNU.sparse.minor") == 0) {
1940			tar->sparse_gnu_minor = (int)tar_atol10(value, strlen(value));
1941			tar->sparse_gnu_pending = 1;
1942		}
1943		if (strcmp(key, "GNU.sparse.name") == 0) {
1944			/*
1945			 * The real filename; when storing sparse
1946			 * files, GNU tar puts a synthesized name into
1947			 * the regular 'path' attribute in an attempt
1948			 * to limit confusion. ;-)
1949			 */
1950			archive_strcpy(&(tar->entry_pathname_override), value);
1951		}
1952		if (strcmp(key, "GNU.sparse.realsize") == 0) {
1953			tar->realsize = tar_atol10(value, strlen(value));
1954			archive_entry_set_size(entry, tar->realsize);
1955			tar->realsize_override = 1;
1956		}
1957		break;
1958	case 'L':
1959		/* Our extensions */
1960/* TODO: Handle arbitrary extended attributes... */
1961/*
1962		if (strcmp(key, "LIBARCHIVE.xxxxxxx") == 0)
1963			archive_entry_set_xxxxxx(entry, value);
1964*/
1965		if (strcmp(key, "LIBARCHIVE.creationtime") == 0) {
1966			pax_time(value, &s, &n);
1967			archive_entry_set_birthtime(entry, s, n);
1968		}
1969		if (strcmp(key, "LIBARCHIVE.symlinktype") == 0) {
1970			if (strcmp(value, "file") == 0) {
1971				archive_entry_set_symlink_type(entry,
1972				    AE_SYMLINK_TYPE_FILE);
1973			} else if (strcmp(value, "dir") == 0) {
1974				archive_entry_set_symlink_type(entry,
1975				    AE_SYMLINK_TYPE_DIRECTORY);
1976			}
1977		}
1978		if (memcmp(key, "LIBARCHIVE.xattr.", 17) == 0)
1979			pax_attribute_xattr(entry, key, value);
1980		break;
1981	case 'R':
1982		/* GNU tar uses RHT.security header to store SELinux xattrs
1983		 * SCHILY.xattr.security.selinux == RHT.security.selinux */
1984		if (strcmp(key, "RHT.security.selinux") == 0) {
1985			pax_attribute_rht_security_selinux(entry, value,
1986			    value_length);
1987			}
1988		break;
1989	case 'S':
1990		/* We support some keys used by the "star" archiver */
1991		if (strcmp(key, "SCHILY.acl.access") == 0) {
1992			r = pax_attribute_acl(a, tar, entry, value,
1993			    ARCHIVE_ENTRY_ACL_TYPE_ACCESS);
1994			if (r == ARCHIVE_FATAL)
1995				return (r);
1996		} else if (strcmp(key, "SCHILY.acl.default") == 0) {
1997			r = pax_attribute_acl(a, tar, entry, value,
1998			    ARCHIVE_ENTRY_ACL_TYPE_DEFAULT);
1999			if (r == ARCHIVE_FATAL)
2000				return (r);
2001		} else if (strcmp(key, "SCHILY.acl.ace") == 0) {
2002			r = pax_attribute_acl(a, tar, entry, value,
2003			    ARCHIVE_ENTRY_ACL_TYPE_NFS4);
2004			if (r == ARCHIVE_FATAL)
2005				return (r);
2006		} else if (strcmp(key, "SCHILY.devmajor") == 0) {
2007			archive_entry_set_rdevmajor(entry,
2008			    (dev_t)tar_atol10(value, strlen(value)));
2009		} else if (strcmp(key, "SCHILY.devminor") == 0) {
2010			archive_entry_set_rdevminor(entry,
2011			    (dev_t)tar_atol10(value, strlen(value)));
2012		} else if (strcmp(key, "SCHILY.fflags") == 0) {
2013			archive_entry_copy_fflags_text(entry, value);
2014		} else if (strcmp(key, "SCHILY.dev") == 0) {
2015			archive_entry_set_dev(entry,
2016			    (dev_t)tar_atol10(value, strlen(value)));
2017		} else if (strcmp(key, "SCHILY.ino") == 0) {
2018			archive_entry_set_ino(entry,
2019			    tar_atol10(value, strlen(value)));
2020		} else if (strcmp(key, "SCHILY.nlink") == 0) {
2021			archive_entry_set_nlink(entry, (unsigned)
2022			    tar_atol10(value, strlen(value)));
2023		} else if (strcmp(key, "SCHILY.realsize") == 0) {
2024			tar->realsize = tar_atol10(value, strlen(value));
2025			tar->realsize_override = 1;
2026			archive_entry_set_size(entry, tar->realsize);
2027		} else if (strncmp(key, "SCHILY.xattr.", 13) == 0) {
2028			pax_attribute_schily_xattr(entry, key, value,
2029			    value_length);
2030		} else if (strcmp(key, "SUN.holesdata") == 0) {
2031			/* A Solaris extension for sparse. */
2032			r = solaris_sparse_parse(a, tar, entry, value);
2033			if (r < err) {
2034				if (r == ARCHIVE_FATAL)
2035					return (r);
2036				err = r;
2037				archive_set_error(&a->archive,
2038				    ARCHIVE_ERRNO_MISC,
2039				    "Parse error: SUN.holesdata");
2040			}
2041		}
2042		break;
2043	case 'a':
2044		if (strcmp(key, "atime") == 0) {
2045			pax_time(value, &s, &n);
2046			archive_entry_set_atime(entry, s, n);
2047		}
2048		break;
2049	case 'c':
2050		if (strcmp(key, "ctime") == 0) {
2051			pax_time(value, &s, &n);
2052			archive_entry_set_ctime(entry, s, n);
2053		} else if (strcmp(key, "charset") == 0) {
2054			/* TODO: Publish charset information in entry. */
2055		} else if (strcmp(key, "comment") == 0) {
2056			/* TODO: Publish comment in entry. */
2057		}
2058		break;
2059	case 'g':
2060		if (strcmp(key, "gid") == 0) {
2061			archive_entry_set_gid(entry,
2062			    tar_atol10(value, strlen(value)));
2063		} else if (strcmp(key, "gname") == 0) {
2064			archive_strcpy(&(tar->entry_gname), value);
2065		}
2066		break;
2067	case 'h':
2068		if (strcmp(key, "hdrcharset") == 0) {
2069			if (strcmp(value, "BINARY") == 0)
2070				/* Binary  mode. */
2071				tar->pax_hdrcharset_binary = 1;
2072			else if (strcmp(value, "ISO-IR 10646 2000 UTF-8") == 0)
2073				tar->pax_hdrcharset_binary = 0;
2074		}
2075		break;
2076	case 'l':
2077		/* pax interchange doesn't distinguish hardlink vs. symlink. */
2078		if (strcmp(key, "linkpath") == 0) {
2079			archive_strcpy(&(tar->entry_linkpath), value);
2080		}
2081		break;
2082	case 'm':
2083		if (strcmp(key, "mtime") == 0) {
2084			pax_time(value, &s, &n);
2085			archive_entry_set_mtime(entry, s, n);
2086		}
2087		break;
2088	case 'p':
2089		if (strcmp(key, "path") == 0) {
2090			archive_strcpy(&(tar->entry_pathname), value);
2091		}
2092		break;
2093	case 'r':
2094		/* POSIX has reserved 'realtime.*' */
2095		break;
2096	case 's':
2097		/* POSIX has reserved 'security.*' */
2098		/* Someday: if (strcmp(key, "security.acl") == 0) { ... } */
2099		if (strcmp(key, "size") == 0) {
2100			/* "size" is the size of the data in the entry. */
2101			tar->entry_bytes_remaining
2102			    = tar_atol10(value, strlen(value));
2103			/*
2104			 * The "size" pax header keyword always overrides the
2105			 * "size" field in the tar header.
2106			 * GNU.sparse.realsize, GNU.sparse.size and
2107			 * SCHILY.realsize override this value.
2108			 */
2109			if (!tar->realsize_override) {
2110				archive_entry_set_size(entry,
2111				    tar->entry_bytes_remaining);
2112				tar->realsize
2113				    = tar->entry_bytes_remaining;
2114			}
2115		}
2116		break;
2117	case 'u':
2118		if (strcmp(key, "uid") == 0) {
2119			archive_entry_set_uid(entry,
2120			    tar_atol10(value, strlen(value)));
2121		} else if (strcmp(key, "uname") == 0) {
2122			archive_strcpy(&(tar->entry_uname), value);
2123		}
2124		break;
2125	}
2126	return (err);
2127}
2128
2129
2130
2131/*
2132 * parse a decimal time value, which may include a fractional portion
2133 */
2134static void
2135pax_time(const char *p, int64_t *ps, long *pn)
2136{
2137	char digit;
2138	int64_t	s;
2139	unsigned long l;
2140	int sign;
2141	int64_t limit, last_digit_limit;
2142
2143	limit = INT64_MAX / 10;
2144	last_digit_limit = INT64_MAX % 10;
2145
2146	s = 0;
2147	sign = 1;
2148	if (*p == '-') {
2149		sign = -1;
2150		p++;
2151	}
2152	while (*p >= '0' && *p <= '9') {
2153		digit = *p - '0';
2154		if (s > limit ||
2155		    (s == limit && digit > last_digit_limit)) {
2156			s = INT64_MAX;
2157			break;
2158		}
2159		s = (s * 10) + digit;
2160		++p;
2161	}
2162
2163	*ps = s * sign;
2164
2165	/* Calculate nanoseconds. */
2166	*pn = 0;
2167
2168	if (*p != '.')
2169		return;
2170
2171	l = 100000000UL;
2172	do {
2173		++p;
2174		if (*p >= '0' && *p <= '9')
2175			*pn += (*p - '0') * l;
2176		else
2177			break;
2178	} while (l /= 10);
2179}
2180
2181/*
2182 * Parse GNU tar header
2183 */
2184static int
2185header_gnutar(struct archive_read *a, struct tar *tar,
2186    struct archive_entry *entry, const void *h, size_t *unconsumed)
2187{
2188	const struct archive_entry_header_gnutar *header;
2189	int64_t t;
2190	int err = ARCHIVE_OK;
2191
2192	/*
2193	 * GNU header is like POSIX ustar, except 'prefix' is
2194	 * replaced with some other fields. This also means the
2195	 * filename is stored as in old-style archives.
2196	 */
2197
2198	/* Grab fields common to all tar variants. */
2199	err = header_common(a, tar, entry, h);
2200	if (err == ARCHIVE_FATAL)
2201		return (err);
2202
2203	/* Copy filename over (to ensure null termination). */
2204	header = (const struct archive_entry_header_gnutar *)h;
2205	if (archive_entry_copy_pathname_l(entry,
2206	    header->name, sizeof(header->name), tar->sconv) != 0) {
2207		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
2208		if (err == ARCHIVE_FATAL)
2209			return (err);
2210	}
2211
2212	/* Fields common to ustar and GNU */
2213	/* XXX Can the following be factored out since it's common
2214	 * to ustar and gnu tar?  Is it okay to move it down into
2215	 * header_common, perhaps?  */
2216	if (archive_entry_copy_uname_l(entry,
2217	    header->uname, sizeof(header->uname), tar->sconv) != 0) {
2218		err = set_conversion_failed_error(a, tar->sconv, "Uname");
2219		if (err == ARCHIVE_FATAL)
2220			return (err);
2221	}
2222
2223	if (archive_entry_copy_gname_l(entry,
2224	    header->gname, sizeof(header->gname), tar->sconv) != 0) {
2225		err = set_conversion_failed_error(a, tar->sconv, "Gname");
2226		if (err == ARCHIVE_FATAL)
2227			return (err);
2228	}
2229
2230	/* Parse out device numbers only for char and block specials */
2231	if (header->typeflag[0] == '3' || header->typeflag[0] == '4') {
2232		archive_entry_set_rdevmajor(entry, (dev_t)
2233		    tar_atol(header->rdevmajor, sizeof(header->rdevmajor)));
2234		archive_entry_set_rdevminor(entry, (dev_t)
2235		    tar_atol(header->rdevminor, sizeof(header->rdevminor)));
2236	} else
2237		archive_entry_set_rdev(entry, 0);
2238
2239	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
2240
2241	/* Grab GNU-specific fields. */
2242	t = tar_atol(header->atime, sizeof(header->atime));
2243	if (t > 0)
2244		archive_entry_set_atime(entry, t, 0);
2245	t = tar_atol(header->ctime, sizeof(header->ctime));
2246	if (t > 0)
2247		archive_entry_set_ctime(entry, t, 0);
2248
2249	if (header->realsize[0] != 0) {
2250		tar->realsize
2251		    = tar_atol(header->realsize, sizeof(header->realsize));
2252		archive_entry_set_size(entry, tar->realsize);
2253		tar->realsize_override = 1;
2254	}
2255
2256	if (header->sparse[0].offset[0] != 0) {
2257		if (gnu_sparse_old_read(a, tar, header, unconsumed)
2258		    != ARCHIVE_OK)
2259			return (ARCHIVE_FATAL);
2260	} else {
2261		if (header->isextended[0] != 0) {
2262			/* XXX WTF? XXX */
2263		}
2264	}
2265
2266	return (err);
2267}
2268
2269static int
2270gnu_add_sparse_entry(struct archive_read *a, struct tar *tar,
2271    int64_t offset, int64_t remaining)
2272{
2273	struct sparse_block *p;
2274
2275	p = (struct sparse_block *)calloc(1, sizeof(*p));
2276	if (p == NULL) {
2277		archive_set_error(&a->archive, ENOMEM, "Out of memory");
2278		return (ARCHIVE_FATAL);
2279	}
2280	if (tar->sparse_last != NULL)
2281		tar->sparse_last->next = p;
2282	else
2283		tar->sparse_list = p;
2284	tar->sparse_last = p;
2285	if (remaining < 0 || offset < 0 || offset > INT64_MAX - remaining) {
2286		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, "Malformed sparse map data");
2287		return (ARCHIVE_FATAL);
2288	}
2289	p->offset = offset;
2290	p->remaining = remaining;
2291	return (ARCHIVE_OK);
2292}
2293
2294static void
2295gnu_clear_sparse_list(struct tar *tar)
2296{
2297	struct sparse_block *p;
2298
2299	while (tar->sparse_list != NULL) {
2300		p = tar->sparse_list;
2301		tar->sparse_list = p->next;
2302		free(p);
2303	}
2304	tar->sparse_last = NULL;
2305}
2306
2307/*
2308 * GNU tar old-format sparse data.
2309 *
2310 * GNU old-format sparse data is stored in a fixed-field
2311 * format.  Offset/size values are 11-byte octal fields (same
2312 * format as 'size' field in ustart header).  These are
2313 * stored in the header, allocating subsequent header blocks
2314 * as needed.  Extending the header in this way is a pretty
2315 * severe POSIX violation; this design has earned GNU tar a
2316 * lot of criticism.
2317 */
2318
2319static int
2320gnu_sparse_old_read(struct archive_read *a, struct tar *tar,
2321    const struct archive_entry_header_gnutar *header, size_t *unconsumed)
2322{
2323	ssize_t bytes_read;
2324	const void *data;
2325	struct extended {
2326		struct gnu_sparse sparse[21];
2327		char	isextended[1];
2328		char	padding[7];
2329	};
2330	const struct extended *ext;
2331
2332	if (gnu_sparse_old_parse(a, tar, header->sparse, 4) != ARCHIVE_OK)
2333		return (ARCHIVE_FATAL);
2334	if (header->isextended[0] == 0)
2335		return (ARCHIVE_OK);
2336
2337	do {
2338		tar_flush_unconsumed(a, unconsumed);
2339		data = __archive_read_ahead(a, 512, &bytes_read);
2340		if (bytes_read < 0)
2341			return (ARCHIVE_FATAL);
2342		if (bytes_read < 512) {
2343			archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2344			    "Truncated tar archive "
2345			    "detected while reading sparse file data");
2346			return (ARCHIVE_FATAL);
2347		}
2348		*unconsumed = 512;
2349		ext = (const struct extended *)data;
2350		if (gnu_sparse_old_parse(a, tar, ext->sparse, 21) != ARCHIVE_OK)
2351			return (ARCHIVE_FATAL);
2352	} while (ext->isextended[0] != 0);
2353	if (tar->sparse_list != NULL)
2354		tar->entry_offset = tar->sparse_list->offset;
2355	return (ARCHIVE_OK);
2356}
2357
2358static int
2359gnu_sparse_old_parse(struct archive_read *a, struct tar *tar,
2360    const struct gnu_sparse *sparse, int length)
2361{
2362	while (length > 0 && sparse->offset[0] != 0) {
2363		if (gnu_add_sparse_entry(a, tar,
2364		    tar_atol(sparse->offset, sizeof(sparse->offset)),
2365		    tar_atol(sparse->numbytes, sizeof(sparse->numbytes)))
2366		    != ARCHIVE_OK)
2367			return (ARCHIVE_FATAL);
2368		sparse++;
2369		length--;
2370	}
2371	return (ARCHIVE_OK);
2372}
2373
2374/*
2375 * GNU tar sparse format 0.0
2376 *
2377 * Beginning with GNU tar 1.15, sparse files are stored using
2378 * information in the pax extended header.  The GNU tar maintainers
2379 * have gone through a number of variations in the process of working
2380 * out this scheme; fortunately, they're all numbered.
2381 *
2382 * Sparse format 0.0 uses attribute GNU.sparse.numblocks to store the
2383 * number of blocks, and GNU.sparse.offset/GNU.sparse.numbytes to
2384 * store offset/size for each block.  The repeated instances of these
2385 * latter fields violate the pax specification (which frowns on
2386 * duplicate keys), so this format was quickly replaced.
2387 */
2388
2389/*
2390 * GNU tar sparse format 0.1
2391 *
2392 * This version replaced the offset/numbytes attributes with
2393 * a single "map" attribute that stored a list of integers.  This
2394 * format had two problems: First, the "map" attribute could be very
2395 * long, which caused problems for some implementations.  More
2396 * importantly, the sparse data was lost when extracted by archivers
2397 * that didn't recognize this extension.
2398 */
2399
2400static int
2401gnu_sparse_01_parse(struct archive_read *a, struct tar *tar, const char *p)
2402{
2403	const char *e;
2404	int64_t offset = -1, size = -1;
2405
2406	for (;;) {
2407		e = p;
2408		while (*e != '\0' && *e != ',') {
2409			if (*e < '0' || *e > '9')
2410				return (ARCHIVE_WARN);
2411			e++;
2412		}
2413		if (offset < 0) {
2414			offset = tar_atol10(p, e - p);
2415			if (offset < 0)
2416				return (ARCHIVE_WARN);
2417		} else {
2418			size = tar_atol10(p, e - p);
2419			if (size < 0)
2420				return (ARCHIVE_WARN);
2421			if (gnu_add_sparse_entry(a, tar, offset, size)
2422			    != ARCHIVE_OK)
2423				return (ARCHIVE_FATAL);
2424			offset = -1;
2425		}
2426		if (*e == '\0')
2427			return (ARCHIVE_OK);
2428		p = e + 1;
2429	}
2430}
2431
2432/*
2433 * GNU tar sparse format 1.0
2434 *
2435 * The idea: The offset/size data is stored as a series of base-10
2436 * ASCII numbers prepended to the file data, so that dearchivers that
2437 * don't support this format will extract the block map along with the
2438 * data and a separate post-process can restore the sparseness.
2439 *
2440 * Unfortunately, GNU tar 1.16 had a bug that added unnecessary
2441 * padding to the body of the file when using this format.  GNU tar
2442 * 1.17 corrected this bug without bumping the version number, so
2443 * it's not possible to support both variants.  This code supports
2444 * the later variant at the expense of not supporting the former.
2445 *
2446 * This variant also replaced GNU.sparse.size with GNU.sparse.realsize
2447 * and introduced the GNU.sparse.major/GNU.sparse.minor attributes.
2448 */
2449
2450/*
2451 * Read the next line from the input, and parse it as a decimal
2452 * integer followed by '\n'.  Returns positive integer value or
2453 * negative on error.
2454 */
2455static int64_t
2456gnu_sparse_10_atol(struct archive_read *a, struct tar *tar,
2457    int64_t *remaining, size_t *unconsumed)
2458{
2459	int64_t l, limit, last_digit_limit;
2460	const char *p;
2461	ssize_t bytes_read;
2462	int base, digit;
2463
2464	base = 10;
2465	limit = INT64_MAX / base;
2466	last_digit_limit = INT64_MAX % base;
2467
2468	/*
2469	 * Skip any lines starting with '#'; GNU tar specs
2470	 * don't require this, but they should.
2471	 */
2472	do {
2473		bytes_read = readline(a, tar, &p,
2474			(ssize_t)tar_min(*remaining, 100), unconsumed);
2475		if (bytes_read <= 0)
2476			return (ARCHIVE_FATAL);
2477		*remaining -= bytes_read;
2478	} while (p[0] == '#');
2479
2480	l = 0;
2481	while (bytes_read > 0) {
2482		if (*p == '\n')
2483			return (l);
2484		if (*p < '0' || *p >= '0' + base)
2485			return (ARCHIVE_WARN);
2486		digit = *p - '0';
2487		if (l > limit || (l == limit && digit > last_digit_limit))
2488			l = INT64_MAX; /* Truncate on overflow. */
2489		else
2490			l = (l * base) + digit;
2491		p++;
2492		bytes_read--;
2493	}
2494	/* TODO: Error message. */
2495	return (ARCHIVE_WARN);
2496}
2497
2498/*
2499 * Returns length (in bytes) of the sparse data description
2500 * that was read.
2501 */
2502static ssize_t
2503gnu_sparse_10_read(struct archive_read *a, struct tar *tar, size_t *unconsumed)
2504{
2505	ssize_t bytes_read;
2506	int entries;
2507	int64_t offset, size, to_skip, remaining;
2508
2509	/* Clear out the existing sparse list. */
2510	gnu_clear_sparse_list(tar);
2511
2512	remaining = tar->entry_bytes_remaining;
2513
2514	/* Parse entries. */
2515	entries = (int)gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2516	if (entries < 0)
2517		return (ARCHIVE_FATAL);
2518	/* Parse the individual entries. */
2519	while (entries-- > 0) {
2520		/* Parse offset/size */
2521		offset = gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2522		if (offset < 0)
2523			return (ARCHIVE_FATAL);
2524		size = gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2525		if (size < 0)
2526			return (ARCHIVE_FATAL);
2527		/* Add a new sparse entry. */
2528		if (gnu_add_sparse_entry(a, tar, offset, size) != ARCHIVE_OK)
2529			return (ARCHIVE_FATAL);
2530	}
2531	/* Skip rest of block... */
2532	tar_flush_unconsumed(a, unconsumed);
2533	bytes_read = (ssize_t)(tar->entry_bytes_remaining - remaining);
2534	to_skip = 0x1ff & -bytes_read;
2535	/* Fail if tar->entry_bytes_remaing would get negative */
2536	if (to_skip > remaining)
2537		return (ARCHIVE_FATAL);
2538	if (to_skip != __archive_read_consume(a, to_skip))
2539		return (ARCHIVE_FATAL);
2540	return ((ssize_t)(bytes_read + to_skip));
2541}
2542
2543/*
2544 * Solaris pax extension for a sparse file. This is recorded with the
2545 * data and hole pairs. The way recording sparse information by Solaris'
2546 * pax simply indicates where data and sparse are, so the stored contents
2547 * consist of both data and hole.
2548 */
2549static int
2550solaris_sparse_parse(struct archive_read *a, struct tar *tar,
2551    struct archive_entry *entry, const char *p)
2552{
2553	const char *e;
2554	int64_t start, end;
2555	int hole = 1;
2556
2557	(void)entry; /* UNUSED */
2558
2559	end = 0;
2560	if (*p == ' ')
2561		p++;
2562	else
2563		return (ARCHIVE_WARN);
2564	for (;;) {
2565		e = p;
2566		while (*e != '\0' && *e != ' ') {
2567			if (*e < '0' || *e > '9')
2568				return (ARCHIVE_WARN);
2569			e++;
2570		}
2571		start = end;
2572		end = tar_atol10(p, e - p);
2573		if (end < 0)
2574			return (ARCHIVE_WARN);
2575		if (start < end) {
2576			if (gnu_add_sparse_entry(a, tar, start,
2577			    end - start) != ARCHIVE_OK)
2578				return (ARCHIVE_FATAL);
2579			tar->sparse_last->hole = hole;
2580		}
2581		if (*e == '\0')
2582			return (ARCHIVE_OK);
2583		p = e + 1;
2584		hole = hole == 0;
2585	}
2586}
2587
2588/*-
2589 * Convert text->integer.
2590 *
2591 * Traditional tar formats (including POSIX) specify base-8 for
2592 * all of the standard numeric fields.  This is a significant limitation
2593 * in practice:
2594 *   = file size is limited to 8GB
2595 *   = rdevmajor and rdevminor are limited to 21 bits
2596 *   = uid/gid are limited to 21 bits
2597 *
2598 * There are two workarounds for this:
2599 *   = pax extended headers, which use variable-length string fields
2600 *   = GNU tar and STAR both allow either base-8 or base-256 in
2601 *      most fields.  The high bit is set to indicate base-256.
2602 *
2603 * On read, this implementation supports both extensions.
2604 */
2605static int64_t
2606tar_atol(const char *p, size_t char_cnt)
2607{
2608	/*
2609	 * Technically, GNU tar considers a field to be in base-256
2610	 * only if the first byte is 0xff or 0x80.
2611	 */
2612	if (*p & 0x80)
2613		return (tar_atol256(p, char_cnt));
2614	return (tar_atol8(p, char_cnt));
2615}
2616
2617/*
2618 * Note that this implementation does not (and should not!) obey
2619 * locale settings; you cannot simply substitute strtol here, since
2620 * it does obey locale.
2621 */
2622static int64_t
2623tar_atol_base_n(const char *p, size_t char_cnt, int base)
2624{
2625	int64_t	l, maxval, limit, last_digit_limit;
2626	int digit, sign;
2627
2628	maxval = INT64_MAX;
2629	limit = INT64_MAX / base;
2630	last_digit_limit = INT64_MAX % base;
2631
2632	/* the pointer will not be dereferenced if char_cnt is zero
2633	 * due to the way the && operator is evaluated.
2634	 */
2635	while (char_cnt != 0 && (*p == ' ' || *p == '\t')) {
2636		p++;
2637		char_cnt--;
2638	}
2639
2640	sign = 1;
2641	if (char_cnt != 0 && *p == '-') {
2642		sign = -1;
2643		p++;
2644		char_cnt--;
2645
2646		maxval = INT64_MIN;
2647		limit = -(INT64_MIN / base);
2648		last_digit_limit = INT64_MIN % base;
2649	}
2650
2651	l = 0;
2652	if (char_cnt != 0) {
2653		digit = *p - '0';
2654		while (digit >= 0 && digit < base  && char_cnt != 0) {
2655			if (l>limit || (l == limit && digit > last_digit_limit)) {
2656				return maxval; /* Truncate on overflow. */
2657			}
2658			l = (l * base) + digit;
2659			digit = *++p - '0';
2660			char_cnt--;
2661		}
2662	}
2663	return (sign < 0) ? -l : l;
2664}
2665
2666static int64_t
2667tar_atol8(const char *p, size_t char_cnt)
2668{
2669	return tar_atol_base_n(p, char_cnt, 8);
2670}
2671
2672static int64_t
2673tar_atol10(const char *p, size_t char_cnt)
2674{
2675	return tar_atol_base_n(p, char_cnt, 10);
2676}
2677
2678/*
2679 * Parse a base-256 integer.  This is just a variable-length
2680 * twos-complement signed binary value in big-endian order, except
2681 * that the high-order bit is ignored.  The values here can be up to
2682 * 12 bytes, so we need to be careful about overflowing 64-bit
2683 * (8-byte) integers.
2684 *
2685 * This code unashamedly assumes that the local machine uses 8-bit
2686 * bytes and twos-complement arithmetic.
2687 */
2688static int64_t
2689tar_atol256(const char *_p, size_t char_cnt)
2690{
2691	uint64_t l;
2692	const unsigned char *p = (const unsigned char *)_p;
2693	unsigned char c, neg;
2694
2695	/* Extend 7-bit 2s-comp to 8-bit 2s-comp, decide sign. */
2696	c = *p;
2697	if (c & 0x40) {
2698		neg = 0xff;
2699		c |= 0x80;
2700		l = ~ARCHIVE_LITERAL_ULL(0);
2701	} else {
2702		neg = 0;
2703		c &= 0x7f;
2704		l = 0;
2705	}
2706
2707	/* If more than 8 bytes, check that we can ignore
2708	 * high-order bits without overflow. */
2709	while (char_cnt > sizeof(int64_t)) {
2710		--char_cnt;
2711		if (c != neg)
2712			return neg ? INT64_MIN : INT64_MAX;
2713		c = *++p;
2714	}
2715
2716	/* c is first byte that fits; if sign mismatch, return overflow */
2717	if ((c ^ neg) & 0x80) {
2718		return neg ? INT64_MIN : INT64_MAX;
2719	}
2720
2721	/* Accumulate remaining bytes. */
2722	while (--char_cnt > 0) {
2723		l = (l << 8) | c;
2724		c = *++p;
2725	}
2726	l = (l << 8) | c;
2727	/* Return signed twos-complement value. */
2728	return (int64_t)(l);
2729}
2730
2731/*
2732 * Returns length of line (including trailing newline)
2733 * or negative on error.  'start' argument is updated to
2734 * point to first character of line.  This avoids copying
2735 * when possible.
2736 */
2737static ssize_t
2738readline(struct archive_read *a, struct tar *tar, const char **start,
2739    ssize_t limit, size_t *unconsumed)
2740{
2741	ssize_t bytes_read;
2742	ssize_t total_size = 0;
2743	const void *t;
2744	const char *s;
2745	void *p;
2746
2747	tar_flush_unconsumed(a, unconsumed);
2748
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' in the read buffer, return pointer to that. */
2755	if (p != NULL) {
2756		bytes_read = 1 + ((const char *)p) - s;
2757		if (bytes_read > limit) {
2758			archive_set_error(&a->archive,
2759			    ARCHIVE_ERRNO_FILE_FORMAT,
2760			    "Line too long");
2761			return (ARCHIVE_FATAL);
2762		}
2763		*unconsumed = bytes_read;
2764		*start = s;
2765		return (bytes_read);
2766	}
2767	*unconsumed = bytes_read;
2768	/* Otherwise, we need to accumulate in a line buffer. */
2769	for (;;) {
2770		if (total_size + bytes_read > limit) {
2771			archive_set_error(&a->archive,
2772			    ARCHIVE_ERRNO_FILE_FORMAT,
2773			    "Line too long");
2774			return (ARCHIVE_FATAL);
2775		}
2776		if (archive_string_ensure(&tar->line, total_size + bytes_read) == NULL) {
2777			archive_set_error(&a->archive, ENOMEM,
2778			    "Can't allocate working buffer");
2779			return (ARCHIVE_FATAL);
2780		}
2781		memcpy(tar->line.s + total_size, t, bytes_read);
2782		tar_flush_unconsumed(a, unconsumed);
2783		total_size += bytes_read;
2784		/* If we found '\n', clean up and return. */
2785		if (p != NULL) {
2786			*start = tar->line.s;
2787			return (total_size);
2788		}
2789		/* Read some more. */
2790		t = __archive_read_ahead(a, 1, &bytes_read);
2791		if (bytes_read <= 0)
2792			return (ARCHIVE_FATAL);
2793		s = t;  /* Start of line? */
2794		p = memchr(t, '\n', bytes_read);
2795		/* If we found '\n', trim the read. */
2796		if (p != NULL) {
2797			bytes_read = 1 + ((const char *)p) - s;
2798		}
2799		*unconsumed = bytes_read;
2800	}
2801}
2802
2803/*
2804 * base64_decode - Base64 decode
2805 *
2806 * This accepts most variations of base-64 encoding, including:
2807 *    * with or without line breaks
2808 *    * with or without the final group padded with '=' or '_' characters
2809 * (The most economical Base-64 variant does not pad the last group and
2810 * omits line breaks; RFC1341 used for MIME requires both.)
2811 */
2812static char *
2813base64_decode(const char *s, size_t len, size_t *out_len)
2814{
2815	static const unsigned char digits[64] = {
2816		'A','B','C','D','E','F','G','H','I','J','K','L','M','N',
2817		'O','P','Q','R','S','T','U','V','W','X','Y','Z','a','b',
2818		'c','d','e','f','g','h','i','j','k','l','m','n','o','p',
2819		'q','r','s','t','u','v','w','x','y','z','0','1','2','3',
2820		'4','5','6','7','8','9','+','/' };
2821	static unsigned char decode_table[128];
2822	char *out, *d;
2823	const unsigned char *src = (const unsigned char *)s;
2824
2825	/* If the decode table is not yet initialized, prepare it. */
2826	if (decode_table[digits[1]] != 1) {
2827		unsigned i;
2828		memset(decode_table, 0xff, sizeof(decode_table));
2829		for (i = 0; i < sizeof(digits); i++)
2830			decode_table[digits[i]] = i;
2831	}
2832
2833	/* Allocate enough space to hold the entire output. */
2834	/* Note that we may not use all of this... */
2835	out = (char *)malloc(len - len / 4 + 1);
2836	if (out == NULL) {
2837		*out_len = 0;
2838		return (NULL);
2839	}
2840	d = out;
2841
2842	while (len > 0) {
2843		/* Collect the next group of (up to) four characters. */
2844		int v = 0;
2845		int group_size = 0;
2846		while (group_size < 4 && len > 0) {
2847			/* '=' or '_' padding indicates final group. */
2848			if (*src == '=' || *src == '_') {
2849				len = 0;
2850				break;
2851			}
2852			/* Skip illegal characters (including line breaks) */
2853			if (*src > 127 || *src < 32
2854			    || decode_table[*src] == 0xff) {
2855				len--;
2856				src++;
2857				continue;
2858			}
2859			v <<= 6;
2860			v |= decode_table[*src++];
2861			len --;
2862			group_size++;
2863		}
2864		/* Align a short group properly. */
2865		v <<= 6 * (4 - group_size);
2866		/* Unpack the group we just collected. */
2867		switch (group_size) {
2868		case 4: d[2] = v & 0xff;
2869			/* FALLTHROUGH */
2870		case 3: d[1] = (v >> 8) & 0xff;
2871			/* FALLTHROUGH */
2872		case 2: d[0] = (v >> 16) & 0xff;
2873			break;
2874		case 1: /* this is invalid! */
2875			break;
2876		}
2877		d += group_size * 3 / 4;
2878	}
2879
2880	*out_len = d - out;
2881	return (out);
2882}
2883
2884static char *
2885url_decode(const char *in)
2886{
2887	char *out, *d;
2888	const char *s;
2889
2890	out = (char *)malloc(strlen(in) + 1);
2891	if (out == NULL)
2892		return (NULL);
2893	for (s = in, d = out; *s != '\0'; ) {
2894		if (s[0] == '%' && s[1] != '\0' && s[2] != '\0') {
2895			/* Try to convert % escape */
2896			int digit1 = tohex(s[1]);
2897			int digit2 = tohex(s[2]);
2898			if (digit1 >= 0 && digit2 >= 0) {
2899				/* Looks good, consume three chars */
2900				s += 3;
2901				/* Convert output */
2902				*d++ = ((digit1 << 4) | digit2);
2903				continue;
2904			}
2905			/* Else fall through and treat '%' as normal char */
2906		}
2907		*d++ = *s++;
2908	}
2909	*d = '\0';
2910	return (out);
2911}
2912
2913static int
2914tohex(int c)
2915{
2916	if (c >= '0' && c <= '9')
2917		return (c - '0');
2918	else if (c >= 'A' && c <= 'F')
2919		return (c - 'A' + 10);
2920	else if (c >= 'a' && c <= 'f')
2921		return (c - 'a' + 10);
2922	else
2923		return (-1);
2924}
2925