archive_read_support_format_tar.c revision 311042 
1/*-
2 * Copyright (c) 2003-2007 Tim Kientzle
3 * Copyright (c) 2011-2012 Michihiro NAKAJIMA
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 */
26
27#include "archive_platform.h"
28__FBSDID("$FreeBSD: stable/10/contrib/libarchive/libarchive/archive_read_support_format_tar.c 311042 2017-01-02 01:43:11Z mm $");
29
30#ifdef HAVE_ERRNO_H
31#include <errno.h>
32#endif
33#include <stddef.h>
34#ifdef HAVE_STDLIB_H
35#include <stdlib.h>
36#endif
37#ifdef HAVE_STRING_H
38#include <string.h>
39#endif
40
41#include "archive.h"
42#include "archive_acl_private.h" /* For ACL parsing routines. */
43#include "archive_entry.h"
44#include "archive_entry_locale.h"
45#include "archive_private.h"
46#include "archive_read_private.h"
47
48#define tar_min(a,b) ((a) < (b) ? (a) : (b))
49
50/*
51 * Layout of POSIX 'ustar' tar header.
52 */
53struct archive_entry_header_ustar {
54	char	name[100];
55	char	mode[8];
56	char	uid[8];
57	char	gid[8];
58	char	size[12];
59	char	mtime[12];
60	char	checksum[8];
61	char	typeflag[1];
62	char	linkname[100];	/* "old format" header ends here */
63	char	magic[6];	/* For POSIX: "ustar\0" */
64	char	version[2];	/* For POSIX: "00" */
65	char	uname[32];
66	char	gname[32];
67	char	rdevmajor[8];
68	char	rdevminor[8];
69	char	prefix[155];
70};
71
72/*
73 * Structure of GNU tar header
74 */
75struct gnu_sparse {
76	char	offset[12];
77	char	numbytes[12];
78};
79
80struct archive_entry_header_gnutar {
81	char	name[100];
82	char	mode[8];
83	char	uid[8];
84	char	gid[8];
85	char	size[12];
86	char	mtime[12];
87	char	checksum[8];
88	char	typeflag[1];
89	char	linkname[100];
90	char	magic[8];  /* "ustar  \0" (note blank/blank/null at end) */
91	char	uname[32];
92	char	gname[32];
93	char	rdevmajor[8];
94	char	rdevminor[8];
95	char	atime[12];
96	char	ctime[12];
97	char	offset[12];
98	char	longnames[4];
99	char	unused[1];
100	struct gnu_sparse sparse[4];
101	char	isextended[1];
102	char	realsize[12];
103	/*
104	 * Old GNU format doesn't use POSIX 'prefix' field; they use
105	 * the 'L' (longname) entry instead.
106	 */
107};
108
109/*
110 * Data specific to this format.
111 */
112struct sparse_block {
113	struct sparse_block	*next;
114	int64_t	offset;
115	int64_t	remaining;
116	int hole;
117};
118
119struct tar {
120	struct archive_string	 acl_text;
121	struct archive_string	 entry_pathname;
122	/* For "GNU.sparse.name" and other similar path extensions. */
123	struct archive_string	 entry_pathname_override;
124	struct archive_string	 entry_linkpath;
125	struct archive_string	 entry_uname;
126	struct archive_string	 entry_gname;
127	struct archive_string	 longlink;
128	struct archive_string	 longname;
129	struct archive_string	 pax_header;
130	struct archive_string	 pax_global;
131	struct archive_string	 line;
132	int			 pax_hdrcharset_binary;
133	int			 header_recursion_depth;
134	int64_t			 entry_bytes_remaining;
135	int64_t			 entry_offset;
136	int64_t			 entry_padding;
137	int64_t 		 entry_bytes_unconsumed;
138	int64_t			 realsize;
139	int			 sparse_allowed;
140	struct sparse_block	*sparse_list;
141	struct sparse_block	*sparse_last;
142	int64_t			 sparse_offset;
143	int64_t			 sparse_numbytes;
144	int			 sparse_gnu_major;
145	int			 sparse_gnu_minor;
146	char			 sparse_gnu_pending;
147
148	struct archive_string	 localname;
149	struct archive_string_conv *opt_sconv;
150	struct archive_string_conv *sconv;
151	struct archive_string_conv *sconv_acl;
152	struct archive_string_conv *sconv_default;
153	int			 init_default_conversion;
154	int			 compat_2x;
155	int			 process_mac_extensions;
156	int			 read_concatenated_archives;
157};
158
159static int	archive_block_is_null(const char *p);
160static char	*base64_decode(const char *, size_t, size_t *);
161static int	gnu_add_sparse_entry(struct archive_read *, struct tar *,
162		    int64_t offset, int64_t remaining);
163
164static void	gnu_clear_sparse_list(struct tar *);
165static int	gnu_sparse_old_read(struct archive_read *, struct tar *,
166		    const struct archive_entry_header_gnutar *header, size_t *);
167static int	gnu_sparse_old_parse(struct archive_read *, struct tar *,
168		    const struct gnu_sparse *sparse, int length);
169static int	gnu_sparse_01_parse(struct archive_read *, struct tar *,
170		    const char *);
171static ssize_t	gnu_sparse_10_read(struct archive_read *, struct tar *,
172			size_t *);
173static int	header_Solaris_ACL(struct archive_read *,  struct tar *,
174		    struct archive_entry *, const void *, size_t *);
175static int	header_common(struct archive_read *,  struct tar *,
176		    struct archive_entry *, const void *);
177static int	header_old_tar(struct archive_read *, struct tar *,
178		    struct archive_entry *, const void *);
179static int	header_pax_extensions(struct archive_read *, struct tar *,
180		    struct archive_entry *, const void *, size_t *);
181static int	header_pax_global(struct archive_read *, struct tar *,
182		    struct archive_entry *, const void *h, size_t *);
183static int	header_longlink(struct archive_read *, struct tar *,
184		    struct archive_entry *, const void *h, size_t *);
185static int	header_longname(struct archive_read *, struct tar *,
186		    struct archive_entry *, const void *h, size_t *);
187static int	read_mac_metadata_blob(struct archive_read *, struct tar *,
188		    struct archive_entry *, const void *h, size_t *);
189static int	header_volume(struct archive_read *, struct tar *,
190		    struct archive_entry *, const void *h, size_t *);
191static int	header_ustar(struct archive_read *, struct tar *,
192		    struct archive_entry *, const void *h);
193static int	header_gnutar(struct archive_read *, struct tar *,
194		    struct archive_entry *, const void *h, size_t *);
195static int	archive_read_format_tar_bid(struct archive_read *, int);
196static int	archive_read_format_tar_options(struct archive_read *,
197		    const char *, const char *);
198static int	archive_read_format_tar_cleanup(struct archive_read *);
199static int	archive_read_format_tar_read_data(struct archive_read *a,
200		    const void **buff, size_t *size, int64_t *offset);
201static int	archive_read_format_tar_skip(struct archive_read *a);
202static int	archive_read_format_tar_read_header(struct archive_read *,
203		    struct archive_entry *);
204static int	checksum(struct archive_read *, const void *);
205static int 	pax_attribute(struct archive_read *, struct tar *,
206		    struct archive_entry *, const char *key, const char *value);
207static int 	pax_header(struct archive_read *, struct tar *,
208		    struct archive_entry *, char *attr);
209static void	pax_time(const char *, int64_t *sec, long *nanos);
210static ssize_t	readline(struct archive_read *, struct tar *, const char **,
211		    ssize_t limit, size_t *);
212static int	read_body_to_string(struct archive_read *, struct tar *,
213		    struct archive_string *, const void *h, size_t *);
214static int	solaris_sparse_parse(struct archive_read *, struct tar *,
215		    struct archive_entry *, const char *);
216static int64_t	tar_atol(const char *, size_t);
217static int64_t	tar_atol10(const char *, size_t);
218static int64_t	tar_atol256(const char *, size_t);
219static int64_t	tar_atol8(const char *, size_t);
220static int	tar_read_header(struct archive_read *, struct tar *,
221		    struct archive_entry *, size_t *);
222static int	tohex(int c);
223static char	*url_decode(const char *);
224static void	tar_flush_unconsumed(struct archive_read *, size_t *);
225
226
227int
228archive_read_support_format_gnutar(struct archive *a)
229{
230	archive_check_magic(a, ARCHIVE_READ_MAGIC,
231	    ARCHIVE_STATE_NEW, "archive_read_support_format_gnutar");
232	return (archive_read_support_format_tar(a));
233}
234
235
236int
237archive_read_support_format_tar(struct archive *_a)
238{
239	struct archive_read *a = (struct archive_read *)_a;
240	struct tar *tar;
241	int r;
242
243	archive_check_magic(_a, ARCHIVE_READ_MAGIC,
244	    ARCHIVE_STATE_NEW, "archive_read_support_format_tar");
245
246	tar = (struct tar *)calloc(1, sizeof(*tar));
247#ifdef HAVE_COPYFILE_H
248	/* Set this by default on Mac OS. */
249	tar->process_mac_extensions = 1;
250#endif
251	if (tar == NULL) {
252		archive_set_error(&a->archive, ENOMEM,
253		    "Can't allocate tar data");
254		return (ARCHIVE_FATAL);
255	}
256
257	r = __archive_read_register_format(a, tar, "tar",
258	    archive_read_format_tar_bid,
259	    archive_read_format_tar_options,
260	    archive_read_format_tar_read_header,
261	    archive_read_format_tar_read_data,
262	    archive_read_format_tar_skip,
263	    NULL,
264	    archive_read_format_tar_cleanup,
265	    NULL,
266	    NULL);
267
268	if (r != ARCHIVE_OK)
269		free(tar);
270	return (ARCHIVE_OK);
271}
272
273static int
274archive_read_format_tar_cleanup(struct archive_read *a)
275{
276	struct tar *tar;
277
278	tar = (struct tar *)(a->format->data);
279	gnu_clear_sparse_list(tar);
280	archive_string_free(&tar->acl_text);
281	archive_string_free(&tar->entry_pathname);
282	archive_string_free(&tar->entry_pathname_override);
283	archive_string_free(&tar->entry_linkpath);
284	archive_string_free(&tar->entry_uname);
285	archive_string_free(&tar->entry_gname);
286	archive_string_free(&tar->line);
287	archive_string_free(&tar->pax_global);
288	archive_string_free(&tar->pax_header);
289	archive_string_free(&tar->longname);
290	archive_string_free(&tar->longlink);
291	archive_string_free(&tar->localname);
292	free(tar);
293	(a->format->data) = NULL;
294	return (ARCHIVE_OK);
295}
296
297/*
298 * Validate number field
299 *
300 * This has to be pretty lenient in order to accommodate the enormous
301 * variety of tar writers in the world:
302 *  = POSIX (IEEE Std 1003.1-1988) ustar requires octal values with leading
303 *    zeros and allows fields to be terminated with space or null characters
304 *  = Many writers use different termination (in particular, libarchive
305 *    omits terminator bytes to squeeze one or two more digits)
306 *  = Many writers pad with space and omit leading zeros
307 *  = GNU tar and star write base-256 values if numbers are too
308 *    big to be represented in octal
309 *
310 *  Examples of specific tar headers that we should support:
311 *  = Perl Archive::Tar terminates uid, gid, devminor and devmajor with two
312 *    null bytes, pads size with spaces and other numeric fields with zeroes
313 *  = plexus-archiver prior to 2.6.3 (before switching to commons-compress)
314 *    may have uid and gid fields filled with spaces without any octal digits
315 *    at all and pads all numeric fields with spaces
316 *
317 * This should tolerate all variants in use.  It will reject a field
318 * where the writer just left garbage after a trailing NUL.
319 */
320static int
321validate_number_field(const char* p_field, size_t i_size)
322{
323	unsigned char marker = (unsigned char)p_field[0];
324	if (marker == 128 || marker == 255 || marker == 0) {
325		/* Base-256 marker, there's nothing we can check. */
326		return 1;
327	} else {
328		/* Must be octal */
329		size_t i = 0;
330		/* Skip any leading spaces */
331		while (i < i_size && p_field[i] == ' ') {
332			++i;
333		}
334		/* Skip octal digits. */
335		while (i < i_size && p_field[i] >= '0' && p_field[i] <= '7') {
336			++i;
337		}
338		/* Any remaining characters must be space or NUL padding. */
339		while (i < i_size) {
340			if (p_field[i] != ' ' && p_field[i] != 0) {
341				return 0;
342			}
343			++i;
344		}
345		return 1;
346	}
347}
348
349static int
350archive_read_format_tar_bid(struct archive_read *a, int best_bid)
351{
352	int bid;
353	const char *h;
354	const struct archive_entry_header_ustar *header;
355
356	(void)best_bid; /* UNUSED */
357
358	bid = 0;
359
360	/* Now let's look at the actual header and see if it matches. */
361	h = __archive_read_ahead(a, 512, NULL);
362	if (h == NULL)
363		return (-1);
364
365	/* If it's an end-of-archive mark, we can handle it. */
366	if (h[0] == 0 && archive_block_is_null(h)) {
367		/*
368		 * Usually, I bid the number of bits verified, but
369		 * in this case, 4096 seems excessive so I picked 10 as
370		 * an arbitrary but reasonable-seeming value.
371		 */
372		return (10);
373	}
374
375	/* If it's not an end-of-archive mark, it must have a valid checksum.*/
376	if (!checksum(a, h))
377		return (0);
378	bid += 48;  /* Checksum is usually 6 octal digits. */
379
380	header = (const struct archive_entry_header_ustar *)h;
381
382	/* Recognize POSIX formats. */
383	if ((memcmp(header->magic, "ustar\0", 6) == 0)
384	    && (memcmp(header->version, "00", 2) == 0))
385		bid += 56;
386
387	/* Recognize GNU tar format. */
388	if ((memcmp(header->magic, "ustar ", 6) == 0)
389	    && (memcmp(header->version, " \0", 2) == 0))
390		bid += 56;
391
392	/* Type flag must be null, digit or A-Z, a-z. */
393	if (header->typeflag[0] != 0 &&
394	    !( header->typeflag[0] >= '0' && header->typeflag[0] <= '9') &&
395	    !( header->typeflag[0] >= 'A' && header->typeflag[0] <= 'Z') &&
396	    !( header->typeflag[0] >= 'a' && header->typeflag[0] <= 'z') )
397		return (0);
398	bid += 2;  /* 6 bits of variation in an 8-bit field leaves 2 bits. */
399
400	/*
401	 * Check format of mode/uid/gid/mtime/size/rdevmajor/rdevminor fields.
402	 */
403	if (bid > 0 && (
404	    validate_number_field(header->mode, sizeof(header->mode)) == 0
405	    || validate_number_field(header->uid, sizeof(header->uid)) == 0
406	    || validate_number_field(header->gid, sizeof(header->gid)) == 0
407	    || validate_number_field(header->mtime, sizeof(header->mtime)) == 0
408	    || validate_number_field(header->size, sizeof(header->size)) == 0
409	    || validate_number_field(header->rdevmajor, sizeof(header->rdevmajor)) == 0
410	    || validate_number_field(header->rdevminor, sizeof(header->rdevminor)) == 0)) {
411		bid = 0;
412	}
413
414	return (bid);
415}
416
417static int
418archive_read_format_tar_options(struct archive_read *a,
419    const char *key, const char *val)
420{
421	struct tar *tar;
422	int ret = ARCHIVE_FAILED;
423
424	tar = (struct tar *)(a->format->data);
425	if (strcmp(key, "compat-2x")  == 0) {
426		/* Handle UTF-8 filenames as libarchive 2.x */
427		tar->compat_2x = (val != NULL && val[0] != 0);
428		tar->init_default_conversion = tar->compat_2x;
429		return (ARCHIVE_OK);
430	} else if (strcmp(key, "hdrcharset")  == 0) {
431		if (val == NULL || val[0] == 0)
432			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
433			    "tar: hdrcharset option needs a character-set name");
434		else {
435			tar->opt_sconv =
436			    archive_string_conversion_from_charset(
437				&a->archive, val, 0);
438			if (tar->opt_sconv != NULL)
439				ret = ARCHIVE_OK;
440			else
441				ret = ARCHIVE_FATAL;
442		}
443		return (ret);
444	} else if (strcmp(key, "mac-ext") == 0) {
445		tar->process_mac_extensions = (val != NULL && val[0] != 0);
446		return (ARCHIVE_OK);
447	} else if (strcmp(key, "read_concatenated_archives") == 0) {
448		tar->read_concatenated_archives = (val != NULL && val[0] != 0);
449		return (ARCHIVE_OK);
450	}
451
452	/* Note: The "warn" return is just to inform the options
453	 * supervisor that we didn't handle it.  It will generate
454	 * a suitable error if no one used this option. */
455	return (ARCHIVE_WARN);
456}
457
458/* utility function- this exists to centralize the logic of tracking
459 * how much unconsumed data we have floating around, and to consume
460 * anything outstanding since we're going to do read_aheads
461 */
462static void
463tar_flush_unconsumed(struct archive_read *a, size_t *unconsumed)
464{
465	if (*unconsumed) {
466/*
467		void *data = (void *)__archive_read_ahead(a, *unconsumed, NULL);
468		 * this block of code is to poison claimed unconsumed space, ensuring
469		 * things break if it is in use still.
470		 * currently it WILL break things, so enable it only for debugging this issue
471		if (data) {
472			memset(data, 0xff, *unconsumed);
473		}
474*/
475		__archive_read_consume(a, *unconsumed);
476		*unconsumed = 0;
477	}
478}
479
480/*
481 * The function invoked by archive_read_next_header().  This
482 * just sets up a few things and then calls the internal
483 * tar_read_header() function below.
484 */
485static int
486archive_read_format_tar_read_header(struct archive_read *a,
487    struct archive_entry *entry)
488{
489	/*
490	 * When converting tar archives to cpio archives, it is
491	 * essential that each distinct file have a distinct inode
492	 * number.  To simplify this, we keep a static count here to
493	 * assign fake dev/inode numbers to each tar entry.  Note that
494	 * pax format archives may overwrite this with something more
495	 * useful.
496	 *
497	 * Ideally, we would track every file read from the archive so
498	 * that we could assign the same dev/ino pair to hardlinks,
499	 * but the memory required to store a complete lookup table is
500	 * probably not worthwhile just to support the relatively
501	 * obscure tar->cpio conversion case.
502	 */
503	static int default_inode;
504	static int default_dev;
505	struct tar *tar;
506	const char *p;
507	const wchar_t *wp;
508	int r;
509	size_t l, unconsumed = 0;
510
511	/* Assign default device/inode values. */
512	archive_entry_set_dev(entry, 1 + default_dev); /* Don't use zero. */
513	archive_entry_set_ino(entry, ++default_inode); /* Don't use zero. */
514	/* Limit generated st_ino number to 16 bits. */
515	if (default_inode >= 0xffff) {
516		++default_dev;
517		default_inode = 0;
518	}
519
520	tar = (struct tar *)(a->format->data);
521	tar->entry_offset = 0;
522	gnu_clear_sparse_list(tar);
523	tar->realsize = -1; /* Mark this as "unset" */
524
525	/* Setup default string conversion. */
526	tar->sconv = tar->opt_sconv;
527	if (tar->sconv == NULL) {
528		if (!tar->init_default_conversion) {
529			tar->sconv_default =
530			    archive_string_default_conversion_for_read(&(a->archive));
531			tar->init_default_conversion = 1;
532		}
533		tar->sconv = tar->sconv_default;
534	}
535
536	r = tar_read_header(a, tar, entry, &unconsumed);
537
538	tar_flush_unconsumed(a, &unconsumed);
539
540	/*
541	 * "non-sparse" files are really just sparse files with
542	 * a single block.
543	 */
544	if (tar->sparse_list == NULL) {
545		if (gnu_add_sparse_entry(a, tar, 0, tar->entry_bytes_remaining)
546		    != ARCHIVE_OK)
547			return (ARCHIVE_FATAL);
548	} else {
549		struct sparse_block *sb;
550
551		for (sb = tar->sparse_list; sb != NULL; sb = sb->next) {
552			if (!sb->hole)
553				archive_entry_sparse_add_entry(entry,
554				    sb->offset, sb->remaining);
555		}
556	}
557
558	if (r == ARCHIVE_OK && archive_entry_filetype(entry) == AE_IFREG) {
559		/*
560		 * "Regular" entry with trailing '/' is really
561		 * directory: This is needed for certain old tar
562		 * variants and even for some broken newer ones.
563		 */
564		if ((wp = archive_entry_pathname_w(entry)) != NULL) {
565			l = wcslen(wp);
566			if (l > 0 && wp[l - 1] == L'/') {
567				archive_entry_set_filetype(entry, AE_IFDIR);
568			}
569		} else if ((p = archive_entry_pathname(entry)) != NULL) {
570			l = strlen(p);
571			if (l > 0 && p[l - 1] == '/') {
572				archive_entry_set_filetype(entry, AE_IFDIR);
573			}
574		}
575	}
576	return (r);
577}
578
579static int
580archive_read_format_tar_read_data(struct archive_read *a,
581    const void **buff, size_t *size, int64_t *offset)
582{
583	ssize_t bytes_read;
584	struct tar *tar;
585	struct sparse_block *p;
586
587	tar = (struct tar *)(a->format->data);
588
589	for (;;) {
590		/* Remove exhausted entries from sparse list. */
591		while (tar->sparse_list != NULL &&
592		    tar->sparse_list->remaining == 0) {
593			p = tar->sparse_list;
594			tar->sparse_list = p->next;
595			free(p);
596		}
597
598		if (tar->entry_bytes_unconsumed) {
599			__archive_read_consume(a, tar->entry_bytes_unconsumed);
600			tar->entry_bytes_unconsumed = 0;
601		}
602
603		/* If we're at end of file, return EOF. */
604		if (tar->sparse_list == NULL ||
605		    tar->entry_bytes_remaining == 0) {
606			if (__archive_read_consume(a, tar->entry_padding) < 0)
607				return (ARCHIVE_FATAL);
608			tar->entry_padding = 0;
609			*buff = NULL;
610			*size = 0;
611			*offset = tar->realsize;
612			return (ARCHIVE_EOF);
613		}
614
615		*buff = __archive_read_ahead(a, 1, &bytes_read);
616		if (bytes_read < 0)
617			return (ARCHIVE_FATAL);
618		if (*buff == NULL) {
619			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
620			    "Truncated tar archive");
621			return (ARCHIVE_FATAL);
622		}
623		if (bytes_read > tar->entry_bytes_remaining)
624			bytes_read = (ssize_t)tar->entry_bytes_remaining;
625		/* Don't read more than is available in the
626		 * current sparse block. */
627		if (tar->sparse_list->remaining < bytes_read)
628			bytes_read = (ssize_t)tar->sparse_list->remaining;
629		*size = bytes_read;
630		*offset = tar->sparse_list->offset;
631		tar->sparse_list->remaining -= bytes_read;
632		tar->sparse_list->offset += bytes_read;
633		tar->entry_bytes_remaining -= bytes_read;
634		tar->entry_bytes_unconsumed = bytes_read;
635
636		if (!tar->sparse_list->hole)
637			return (ARCHIVE_OK);
638		/* Current is hole data and skip this. */
639	}
640}
641
642static int
643archive_read_format_tar_skip(struct archive_read *a)
644{
645	int64_t bytes_skipped;
646	int64_t request;
647	struct sparse_block *p;
648	struct tar* tar;
649
650	tar = (struct tar *)(a->format->data);
651
652	/* Do not consume the hole of a sparse file. */
653	request = 0;
654	for (p = tar->sparse_list; p != NULL; p = p->next) {
655		if (!p->hole) {
656			if (p->remaining >= INT64_MAX - request) {
657				return ARCHIVE_FATAL;
658			}
659			request += p->remaining;
660		}
661	}
662	if (request > tar->entry_bytes_remaining)
663		request = tar->entry_bytes_remaining;
664	request += tar->entry_padding + tar->entry_bytes_unconsumed;
665
666	bytes_skipped = __archive_read_consume(a, request);
667	if (bytes_skipped < 0)
668		return (ARCHIVE_FATAL);
669
670	tar->entry_bytes_remaining = 0;
671	tar->entry_bytes_unconsumed = 0;
672	tar->entry_padding = 0;
673
674	/* Free the sparse list. */
675	gnu_clear_sparse_list(tar);
676
677	return (ARCHIVE_OK);
678}
679
680/*
681 * This function recursively interprets all of the headers associated
682 * with a single entry.
683 */
684static int
685tar_read_header(struct archive_read *a, struct tar *tar,
686    struct archive_entry *entry, size_t *unconsumed)
687{
688	ssize_t bytes;
689	int err;
690	const char *h;
691	const struct archive_entry_header_ustar *header;
692	const struct archive_entry_header_gnutar *gnuheader;
693
694	/* Loop until we find a workable header record. */
695	for (;;) {
696		tar_flush_unconsumed(a, unconsumed);
697
698		/* Read 512-byte header record */
699		h = __archive_read_ahead(a, 512, &bytes);
700		if (bytes < 0)
701			return ((int)bytes);
702		if (bytes == 0) { /* EOF at a block boundary. */
703			/* Some writers do omit the block of nulls. <sigh> */
704			return (ARCHIVE_EOF);
705		}
706		if (bytes < 512) {  /* Short block at EOF; this is bad. */
707			archive_set_error(&a->archive,
708			    ARCHIVE_ERRNO_FILE_FORMAT,
709			    "Truncated tar archive");
710			return (ARCHIVE_FATAL);
711		}
712		*unconsumed = 512;
713
714		/* Header is workable if it's not an end-of-archive mark. */
715		if (h[0] != 0 || !archive_block_is_null(h))
716			break;
717
718		/* Ensure format is set for archives with only null blocks. */
719		if (a->archive.archive_format_name == NULL) {
720			a->archive.archive_format = ARCHIVE_FORMAT_TAR;
721			a->archive.archive_format_name = "tar";
722		}
723
724		if (!tar->read_concatenated_archives) {
725			/* Try to consume a second all-null record, as well. */
726			tar_flush_unconsumed(a, unconsumed);
727			h = __archive_read_ahead(a, 512, NULL);
728			if (h != NULL && h[0] == 0 && archive_block_is_null(h))
729				__archive_read_consume(a, 512);
730			archive_clear_error(&a->archive);
731			return (ARCHIVE_EOF);
732		}
733
734		/*
735		 * We're reading concatenated archives, ignore this block and
736		 * loop to get the next.
737		 */
738	}
739
740	/*
741	 * Note: If the checksum fails and we return ARCHIVE_RETRY,
742	 * then the client is likely to just retry.  This is a very
743	 * crude way to search for the next valid header!
744	 *
745	 * TODO: Improve this by implementing a real header scan.
746	 */
747	if (!checksum(a, h)) {
748		tar_flush_unconsumed(a, unconsumed);
749		archive_set_error(&a->archive, EINVAL, "Damaged tar archive");
750		return (ARCHIVE_RETRY); /* Retryable: Invalid header */
751	}
752
753	if (++tar->header_recursion_depth > 32) {
754		tar_flush_unconsumed(a, unconsumed);
755		archive_set_error(&a->archive, EINVAL, "Too many special headers");
756		return (ARCHIVE_WARN);
757	}
758
759	/* Determine the format variant. */
760	header = (const struct archive_entry_header_ustar *)h;
761
762	switch(header->typeflag[0]) {
763	case 'A': /* Solaris tar ACL */
764		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
765		a->archive.archive_format_name = "Solaris tar";
766		err = header_Solaris_ACL(a, tar, entry, h, unconsumed);
767		break;
768	case 'g': /* POSIX-standard 'g' header. */
769		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
770		a->archive.archive_format_name = "POSIX pax interchange format";
771		err = header_pax_global(a, tar, entry, h, unconsumed);
772		if (err == ARCHIVE_EOF)
773			return (err);
774		break;
775	case 'K': /* Long link name (GNU tar, others) */
776		err = header_longlink(a, tar, entry, h, unconsumed);
777		break;
778	case 'L': /* Long filename (GNU tar, others) */
779		err = header_longname(a, tar, entry, h, unconsumed);
780		break;
781	case 'V': /* GNU volume header */
782		err = header_volume(a, tar, entry, h, unconsumed);
783		break;
784	case 'X': /* Used by SUN tar; same as 'x'. */
785		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
786		a->archive.archive_format_name =
787		    "POSIX pax interchange format (Sun variant)";
788		err = header_pax_extensions(a, tar, entry, h, unconsumed);
789		break;
790	case 'x': /* POSIX-standard 'x' header. */
791		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
792		a->archive.archive_format_name = "POSIX pax interchange format";
793		err = header_pax_extensions(a, tar, entry, h, unconsumed);
794		break;
795	default:
796		gnuheader = (const struct archive_entry_header_gnutar *)h;
797		if (memcmp(gnuheader->magic, "ustar  \0", 8) == 0) {
798			a->archive.archive_format = ARCHIVE_FORMAT_TAR_GNUTAR;
799			a->archive.archive_format_name = "GNU tar format";
800			err = header_gnutar(a, tar, entry, h, unconsumed);
801		} else if (memcmp(header->magic, "ustar", 5) == 0) {
802			if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) {
803				a->archive.archive_format = ARCHIVE_FORMAT_TAR_USTAR;
804				a->archive.archive_format_name = "POSIX ustar format";
805			}
806			err = header_ustar(a, tar, entry, h);
807		} else {
808			a->archive.archive_format = ARCHIVE_FORMAT_TAR;
809			a->archive.archive_format_name = "tar (non-POSIX)";
810			err = header_old_tar(a, tar, entry, h);
811		}
812	}
813	if (err == ARCHIVE_FATAL)
814		return (err);
815
816	tar_flush_unconsumed(a, unconsumed);
817
818	h = NULL;
819	header = NULL;
820
821	--tar->header_recursion_depth;
822	/* Yuck.  Apple's design here ends up storing long pathname
823	 * extensions for both the AppleDouble extension entry and the
824	 * regular entry.
825	 */
826	if ((err == ARCHIVE_WARN || err == ARCHIVE_OK) &&
827	    tar->header_recursion_depth == 0 &&
828	    tar->process_mac_extensions) {
829		int err2 = read_mac_metadata_blob(a, tar, entry, h, unconsumed);
830		if (err2 < err)
831			err = err2;
832	}
833
834	/* We return warnings or success as-is.  Anything else is fatal. */
835	if (err == ARCHIVE_WARN || err == ARCHIVE_OK) {
836		if (tar->sparse_gnu_pending) {
837			if (tar->sparse_gnu_major == 1 &&
838			    tar->sparse_gnu_minor == 0) {
839				ssize_t bytes_read;
840
841				tar->sparse_gnu_pending = 0;
842				/* Read initial sparse map. */
843				bytes_read = gnu_sparse_10_read(a, tar, unconsumed);
844				tar->entry_bytes_remaining -= bytes_read;
845				if (bytes_read < 0)
846					return ((int)bytes_read);
847			} else {
848				archive_set_error(&a->archive,
849				    ARCHIVE_ERRNO_MISC,
850				    "Unrecognized GNU sparse file format");
851				return (ARCHIVE_WARN);
852			}
853			tar->sparse_gnu_pending = 0;
854		}
855		return (err);
856	}
857	if (err == ARCHIVE_EOF)
858		/* EOF when recursively reading a header is bad. */
859		archive_set_error(&a->archive, EINVAL, "Damaged tar archive");
860	return (ARCHIVE_FATAL);
861}
862
863/*
864 * Return true if block checksum is correct.
865 */
866static int
867checksum(struct archive_read *a, const void *h)
868{
869	const unsigned char *bytes;
870	const struct archive_entry_header_ustar	*header;
871	int check, sum;
872	size_t i;
873
874	(void)a; /* UNUSED */
875	bytes = (const unsigned char *)h;
876	header = (const struct archive_entry_header_ustar *)h;
877
878	/* Checksum field must hold an octal number */
879	for (i = 0; i < sizeof(header->checksum); ++i) {
880		char c = header->checksum[i];
881		if (c != ' ' && c != '\0' && (c < '0' || c > '7'))
882			return 0;
883	}
884
885	/*
886	 * Test the checksum.  Note that POSIX specifies _unsigned_
887	 * bytes for this calculation.
888	 */
889	sum = (int)tar_atol(header->checksum, sizeof(header->checksum));
890	check = 0;
891	for (i = 0; i < 148; i++)
892		check += (unsigned char)bytes[i];
893	for (; i < 156; i++)
894		check += 32;
895	for (; i < 512; i++)
896		check += (unsigned char)bytes[i];
897	if (sum == check)
898		return (1);
899
900	/*
901	 * Repeat test with _signed_ bytes, just in case this archive
902	 * was created by an old BSD, Solaris, or HP-UX tar with a
903	 * broken checksum calculation.
904	 */
905	check = 0;
906	for (i = 0; i < 148; i++)
907		check += (signed char)bytes[i];
908	for (; i < 156; i++)
909		check += 32;
910	for (; i < 512; i++)
911		check += (signed char)bytes[i];
912	if (sum == check)
913		return (1);
914
915	return (0);
916}
917
918/*
919 * Return true if this block contains only nulls.
920 */
921static int
922archive_block_is_null(const char *p)
923{
924	unsigned i;
925
926	for (i = 0; i < 512; i++)
927		if (*p++)
928			return (0);
929	return (1);
930}
931
932/*
933 * Interpret 'A' Solaris ACL header
934 */
935static int
936header_Solaris_ACL(struct archive_read *a, struct tar *tar,
937    struct archive_entry *entry, const void *h, size_t *unconsumed)
938{
939	const struct archive_entry_header_ustar *header;
940	size_t size;
941	int err;
942	int64_t type;
943	char *acl, *p;
944
945	/*
946	 * read_body_to_string adds a NUL terminator, but we need a little
947	 * more to make sure that we don't overrun acl_text later.
948	 */
949	header = (const struct archive_entry_header_ustar *)h;
950	size = (size_t)tar_atol(header->size, sizeof(header->size));
951	err = read_body_to_string(a, tar, &(tar->acl_text), h, unconsumed);
952	if (err != ARCHIVE_OK)
953		return (err);
954
955	/* Recursively read next header */
956	err = tar_read_header(a, tar, entry, unconsumed);
957	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
958		return (err);
959
960	/* TODO: Examine the first characters to see if this
961	 * is an AIX ACL descriptor.  We'll likely never support
962	 * them, but it would be polite to recognize and warn when
963	 * we do see them. */
964
965	/* Leading octal number indicates ACL type and number of entries. */
966	p = acl = tar->acl_text.s;
967	type = 0;
968	while (*p != '\0' && p < acl + size) {
969		if (*p < '0' || *p > '7') {
970			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
971			    "Malformed Solaris ACL attribute (invalid digit)");
972			return(ARCHIVE_WARN);
973		}
974		type <<= 3;
975		type += *p - '0';
976		if (type > 077777777) {
977			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
978			    "Malformed Solaris ACL attribute (count too large)");
979			return (ARCHIVE_WARN);
980		}
981		p++;
982	}
983	switch ((int)type & ~0777777) {
984	case 01000000:
985		/* POSIX.1e ACL */
986		break;
987	case 03000000:
988		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
989		    "Solaris NFSv4 ACLs not supported");
990		return (ARCHIVE_WARN);
991	default:
992		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
993		    "Malformed Solaris ACL attribute (unsupported type %o)",
994		    (int)type);
995		return (ARCHIVE_WARN);
996	}
997	p++;
998
999	if (p >= acl + size) {
1000		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1001		    "Malformed Solaris ACL attribute (body overflow)");
1002		return(ARCHIVE_WARN);
1003	}
1004
1005	/* ACL text is null-terminated; find the end. */
1006	size -= (p - acl);
1007	acl = p;
1008
1009	while (*p != '\0' && p < acl + size)
1010		p++;
1011
1012	if (tar->sconv_acl == NULL) {
1013		tar->sconv_acl = archive_string_conversion_from_charset(
1014		    &(a->archive), "UTF-8", 1);
1015		if (tar->sconv_acl == NULL)
1016			return (ARCHIVE_FATAL);
1017	}
1018	archive_strncpy(&(tar->localname), acl, p - acl);
1019	err = archive_acl_parse_l(archive_entry_acl(entry),
1020	    tar->localname.s, ARCHIVE_ENTRY_ACL_TYPE_ACCESS, tar->sconv_acl);
1021	if (err != ARCHIVE_OK) {
1022		if (errno == ENOMEM) {
1023			archive_set_error(&a->archive, ENOMEM,
1024			    "Can't allocate memory for ACL");
1025		} else
1026			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1027			    "Malformed Solaris ACL attribute (unparsable)");
1028	}
1029	return (err);
1030}
1031
1032/*
1033 * Interpret 'K' long linkname header.
1034 */
1035static int
1036header_longlink(struct archive_read *a, struct tar *tar,
1037    struct archive_entry *entry, const void *h, size_t *unconsumed)
1038{
1039	int err;
1040
1041	err = read_body_to_string(a, tar, &(tar->longlink), h, unconsumed);
1042	if (err != ARCHIVE_OK)
1043		return (err);
1044	err = tar_read_header(a, tar, entry, unconsumed);
1045	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1046		return (err);
1047	/* Set symlink if symlink already set, else hardlink. */
1048	archive_entry_copy_link(entry, tar->longlink.s);
1049	return (ARCHIVE_OK);
1050}
1051
1052static int
1053set_conversion_failed_error(struct archive_read *a,
1054    struct archive_string_conv *sconv, const char *name)
1055{
1056	if (errno == ENOMEM) {
1057		archive_set_error(&a->archive, ENOMEM,
1058		    "Can't allocate memory for %s", name);
1059		return (ARCHIVE_FATAL);
1060	}
1061	archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1062	    "%s can't be converted from %s to current locale.",
1063	    name, archive_string_conversion_charset_name(sconv));
1064	return (ARCHIVE_WARN);
1065}
1066
1067/*
1068 * Interpret 'L' long filename header.
1069 */
1070static int
1071header_longname(struct archive_read *a, struct tar *tar,
1072    struct archive_entry *entry, const void *h, size_t *unconsumed)
1073{
1074	int err;
1075
1076	err = read_body_to_string(a, tar, &(tar->longname), h, unconsumed);
1077	if (err != ARCHIVE_OK)
1078		return (err);
1079	/* Read and parse "real" header, then override name. */
1080	err = tar_read_header(a, tar, entry, unconsumed);
1081	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1082		return (err);
1083	if (archive_entry_copy_pathname_l(entry, tar->longname.s,
1084	    archive_strlen(&(tar->longname)), tar->sconv) != 0)
1085		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1086	return (err);
1087}
1088
1089
1090/*
1091 * Interpret 'V' GNU tar volume header.
1092 */
1093static int
1094header_volume(struct archive_read *a, struct tar *tar,
1095    struct archive_entry *entry, const void *h, size_t *unconsumed)
1096{
1097	(void)h;
1098
1099	/* Just skip this and read the next header. */
1100	return (tar_read_header(a, tar, entry, unconsumed));
1101}
1102
1103/*
1104 * Read body of an archive entry into an archive_string object.
1105 */
1106static int
1107read_body_to_string(struct archive_read *a, struct tar *tar,
1108    struct archive_string *as, const void *h, size_t *unconsumed)
1109{
1110	int64_t size;
1111	const struct archive_entry_header_ustar *header;
1112	const void *src;
1113
1114	(void)tar; /* UNUSED */
1115	header = (const struct archive_entry_header_ustar *)h;
1116	size  = tar_atol(header->size, sizeof(header->size));
1117	if ((size > 1048576) || (size < 0)) {
1118		archive_set_error(&a->archive, EINVAL,
1119		    "Special header too large");
1120		return (ARCHIVE_FATAL);
1121	}
1122
1123	/* Fail if we can't make our buffer big enough. */
1124	if (archive_string_ensure(as, (size_t)size+1) == NULL) {
1125		archive_set_error(&a->archive, ENOMEM,
1126		    "No memory");
1127		return (ARCHIVE_FATAL);
1128	}
1129
1130	tar_flush_unconsumed(a, unconsumed);
1131
1132	/* Read the body into the string. */
1133	*unconsumed = (size_t)((size + 511) & ~ 511);
1134	src = __archive_read_ahead(a, *unconsumed, NULL);
1135	if (src == NULL) {
1136		*unconsumed = 0;
1137		return (ARCHIVE_FATAL);
1138	}
1139	memcpy(as->s, src, (size_t)size);
1140	as->s[size] = '\0';
1141	as->length = (size_t)size;
1142	return (ARCHIVE_OK);
1143}
1144
1145/*
1146 * Parse out common header elements.
1147 *
1148 * This would be the same as header_old_tar, except that the
1149 * filename is handled slightly differently for old and POSIX
1150 * entries  (POSIX entries support a 'prefix').  This factoring
1151 * allows header_old_tar and header_ustar
1152 * to handle filenames differently, while still putting most of the
1153 * common parsing into one place.
1154 */
1155static int
1156header_common(struct archive_read *a, struct tar *tar,
1157    struct archive_entry *entry, const void *h)
1158{
1159	const struct archive_entry_header_ustar	*header;
1160	char	tartype;
1161	int     err = ARCHIVE_OK;
1162
1163	header = (const struct archive_entry_header_ustar *)h;
1164	if (header->linkname[0])
1165		archive_strncpy(&(tar->entry_linkpath),
1166		    header->linkname, sizeof(header->linkname));
1167	else
1168		archive_string_empty(&(tar->entry_linkpath));
1169
1170	/* Parse out the numeric fields (all are octal) */
1171	archive_entry_set_mode(entry,
1172		(mode_t)tar_atol(header->mode, sizeof(header->mode)));
1173	archive_entry_set_uid(entry, tar_atol(header->uid, sizeof(header->uid)));
1174	archive_entry_set_gid(entry, tar_atol(header->gid, sizeof(header->gid)));
1175	tar->entry_bytes_remaining = tar_atol(header->size, sizeof(header->size));
1176	if (tar->entry_bytes_remaining < 0) {
1177		tar->entry_bytes_remaining = 0;
1178		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1179		    "Tar entry has negative size");
1180		return (ARCHIVE_FATAL);
1181	}
1182	if (tar->entry_bytes_remaining == INT64_MAX) {
1183		/* Note: tar_atol returns INT64_MAX on overflow */
1184		tar->entry_bytes_remaining = 0;
1185		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1186		    "Tar entry size overflow");
1187		return (ARCHIVE_FATAL);
1188	}
1189	tar->realsize = tar->entry_bytes_remaining;
1190	archive_entry_set_size(entry, tar->entry_bytes_remaining);
1191	archive_entry_set_mtime(entry, tar_atol(header->mtime, sizeof(header->mtime)), 0);
1192
1193	/* Handle the tar type flag appropriately. */
1194	tartype = header->typeflag[0];
1195
1196	switch (tartype) {
1197	case '1': /* Hard link */
1198		if (archive_entry_copy_hardlink_l(entry, tar->entry_linkpath.s,
1199		    archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) {
1200			err = set_conversion_failed_error(a, tar->sconv,
1201			    "Linkname");
1202			if (err == ARCHIVE_FATAL)
1203				return (err);
1204		}
1205		/*
1206		 * The following may seem odd, but: Technically, tar
1207		 * does not store the file type for a "hard link"
1208		 * entry, only the fact that it is a hard link.  So, I
1209		 * leave the type zero normally.  But, pax interchange
1210		 * format allows hard links to have data, which
1211		 * implies that the underlying entry is a regular
1212		 * file.
1213		 */
1214		if (archive_entry_size(entry) > 0)
1215			archive_entry_set_filetype(entry, AE_IFREG);
1216
1217		/*
1218		 * A tricky point: Traditionally, tar readers have
1219		 * ignored the size field when reading hardlink
1220		 * entries, and some writers put non-zero sizes even
1221		 * though the body is empty.  POSIX blessed this
1222		 * convention in the 1988 standard, but broke with
1223		 * this tradition in 2001 by permitting hardlink
1224		 * entries to store valid bodies in pax interchange
1225		 * format, but not in ustar format.  Since there is no
1226		 * hard and fast way to distinguish pax interchange
1227		 * from earlier archives (the 'x' and 'g' entries are
1228		 * optional, after all), we need a heuristic.
1229		 */
1230		if (archive_entry_size(entry) == 0) {
1231			/* If the size is already zero, we're done. */
1232		}  else if (a->archive.archive_format
1233		    == ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) {
1234			/* Definitely pax extended; must obey hardlink size. */
1235		} else if (a->archive.archive_format == ARCHIVE_FORMAT_TAR
1236		    || a->archive.archive_format == ARCHIVE_FORMAT_TAR_GNUTAR)
1237		{
1238			/* Old-style or GNU tar: we must ignore the size. */
1239			archive_entry_set_size(entry, 0);
1240			tar->entry_bytes_remaining = 0;
1241		} else if (archive_read_format_tar_bid(a, 50) > 50) {
1242			/*
1243			 * We don't know if it's pax: If the bid
1244			 * function sees a valid ustar header
1245			 * immediately following, then let's ignore
1246			 * the hardlink size.
1247			 */
1248			archive_entry_set_size(entry, 0);
1249			tar->entry_bytes_remaining = 0;
1250		}
1251		/*
1252		 * TODO: There are still two cases I'd like to handle:
1253		 *   = a ustar non-pax archive with a hardlink entry at
1254		 *     end-of-archive.  (Look for block of nulls following?)
1255		 *   = a pax archive that has not seen any pax headers
1256		 *     and has an entry which is a hardlink entry storing
1257		 *     a body containing an uncompressed tar archive.
1258		 * The first is worth addressing; I don't see any reliable
1259		 * way to deal with the second possibility.
1260		 */
1261		break;
1262	case '2': /* Symlink */
1263		archive_entry_set_filetype(entry, AE_IFLNK);
1264		archive_entry_set_size(entry, 0);
1265		tar->entry_bytes_remaining = 0;
1266		if (archive_entry_copy_symlink_l(entry, tar->entry_linkpath.s,
1267		    archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) {
1268			err = set_conversion_failed_error(a, tar->sconv,
1269			    "Linkname");
1270			if (err == ARCHIVE_FATAL)
1271				return (err);
1272		}
1273		break;
1274	case '3': /* Character device */
1275		archive_entry_set_filetype(entry, AE_IFCHR);
1276		archive_entry_set_size(entry, 0);
1277		tar->entry_bytes_remaining = 0;
1278		break;
1279	case '4': /* Block device */
1280		archive_entry_set_filetype(entry, AE_IFBLK);
1281		archive_entry_set_size(entry, 0);
1282		tar->entry_bytes_remaining = 0;
1283		break;
1284	case '5': /* Dir */
1285		archive_entry_set_filetype(entry, AE_IFDIR);
1286		archive_entry_set_size(entry, 0);
1287		tar->entry_bytes_remaining = 0;
1288		break;
1289	case '6': /* FIFO device */
1290		archive_entry_set_filetype(entry, AE_IFIFO);
1291		archive_entry_set_size(entry, 0);
1292		tar->entry_bytes_remaining = 0;
1293		break;
1294	case 'D': /* GNU incremental directory type */
1295		/*
1296		 * No special handling is actually required here.
1297		 * It might be nice someday to preprocess the file list and
1298		 * provide it to the client, though.
1299		 */
1300		archive_entry_set_filetype(entry, AE_IFDIR);
1301		break;
1302	case 'M': /* GNU "Multi-volume" (remainder of file from last archive)*/
1303		/*
1304		 * As far as I can tell, this is just like a regular file
1305		 * entry, except that the contents should be _appended_ to
1306		 * the indicated file at the indicated offset.  This may
1307		 * require some API work to fully support.
1308		 */
1309		break;
1310	case 'N': /* Old GNU "long filename" entry. */
1311		/* The body of this entry is a script for renaming
1312		 * previously-extracted entries.  Ugh.  It will never
1313		 * be supported by libarchive. */
1314		archive_entry_set_filetype(entry, AE_IFREG);
1315		break;
1316	case 'S': /* GNU sparse files */
1317		/*
1318		 * Sparse files are really just regular files with
1319		 * sparse information in the extended area.
1320		 */
1321		/* FALLTHROUGH */
1322	case '0':
1323		/*
1324		 * Enable sparse file "read" support only for regular
1325		 * files and explicit GNU sparse files.  However, we
1326		 * don't allow non-standard file types to be sparse.
1327		 */
1328		tar->sparse_allowed = 1;
1329		/* FALLTHROUGH */
1330	default: /* Regular file  and non-standard types */
1331		/*
1332		 * Per POSIX: non-recognized types should always be
1333		 * treated as regular files.
1334		 */
1335		archive_entry_set_filetype(entry, AE_IFREG);
1336		break;
1337	}
1338	return (err);
1339}
1340
1341/*
1342 * Parse out header elements for "old-style" tar archives.
1343 */
1344static int
1345header_old_tar(struct archive_read *a, struct tar *tar,
1346    struct archive_entry *entry, const void *h)
1347{
1348	const struct archive_entry_header_ustar	*header;
1349	int err = ARCHIVE_OK, err2;
1350
1351	/* Copy filename over (to ensure null termination). */
1352	header = (const struct archive_entry_header_ustar *)h;
1353	if (archive_entry_copy_pathname_l(entry,
1354	    header->name, sizeof(header->name), tar->sconv) != 0) {
1355		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1356		if (err == ARCHIVE_FATAL)
1357			return (err);
1358	}
1359
1360	/* Grab rest of common fields */
1361	err2 = header_common(a, tar, entry, h);
1362	if (err > err2)
1363		err = err2;
1364
1365	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1366	return (err);
1367}
1368
1369/*
1370 * Read a Mac AppleDouble-encoded blob of file metadata,
1371 * if there is one.
1372 */
1373static int
1374read_mac_metadata_blob(struct archive_read *a, struct tar *tar,
1375    struct archive_entry *entry, const void *h, size_t *unconsumed)
1376{
1377	int64_t size;
1378	const void *data;
1379	const char *p, *name;
1380	const wchar_t *wp, *wname;
1381
1382	(void)h; /* UNUSED */
1383
1384	wname = wp = archive_entry_pathname_w(entry);
1385	if (wp != NULL) {
1386		/* Find the last path element. */
1387		for (; *wp != L'\0'; ++wp) {
1388			if (wp[0] == '/' && wp[1] != L'\0')
1389				wname = wp + 1;
1390		}
1391		/*
1392		 * If last path element starts with "._", then
1393		 * this is a Mac extension.
1394		 */
1395		if (wname[0] != L'.' || wname[1] != L'_' || wname[2] == L'\0')
1396			return ARCHIVE_OK;
1397	} else {
1398		/* Find the last path element. */
1399		name = p = archive_entry_pathname(entry);
1400		if (p == NULL)
1401			return (ARCHIVE_FAILED);
1402		for (; *p != '\0'; ++p) {
1403			if (p[0] == '/' && p[1] != '\0')
1404				name = p + 1;
1405		}
1406		/*
1407		 * If last path element starts with "._", then
1408		 * this is a Mac extension.
1409		 */
1410		if (name[0] != '.' || name[1] != '_' || name[2] == '\0')
1411			return ARCHIVE_OK;
1412	}
1413
1414 	/* Read the body as a Mac OS metadata blob. */
1415	size = archive_entry_size(entry);
1416
1417	/*
1418	 * TODO: Look beyond the body here to peek at the next header.
1419	 * If it's a regular header (not an extension header)
1420	 * that has the wrong name, just return the current
1421	 * entry as-is, without consuming the body here.
1422	 * That would reduce the risk of us mis-identifying
1423	 * an ordinary file that just happened to have
1424	 * a name starting with "._".
1425	 *
1426	 * Q: Is the above idea really possible?  Even
1427	 * when there are GNU or pax extension entries?
1428	 */
1429	data = __archive_read_ahead(a, (size_t)size, NULL);
1430	if (data == NULL) {
1431		*unconsumed = 0;
1432		return (ARCHIVE_FATAL);
1433	}
1434	archive_entry_copy_mac_metadata(entry, data, (size_t)size);
1435	*unconsumed = (size_t)((size + 511) & ~ 511);
1436	tar_flush_unconsumed(a, unconsumed);
1437	return (tar_read_header(a, tar, entry, unconsumed));
1438}
1439
1440/*
1441 * Parse a file header for a pax extended archive entry.
1442 */
1443static int
1444header_pax_global(struct archive_read *a, struct tar *tar,
1445    struct archive_entry *entry, const void *h, size_t *unconsumed)
1446{
1447	int err;
1448
1449	err = read_body_to_string(a, tar, &(tar->pax_global), h, unconsumed);
1450	if (err != ARCHIVE_OK)
1451		return (err);
1452	err = tar_read_header(a, tar, entry, unconsumed);
1453	return (err);
1454}
1455
1456static int
1457header_pax_extensions(struct archive_read *a, struct tar *tar,
1458    struct archive_entry *entry, const void *h, size_t *unconsumed)
1459{
1460	int err, err2;
1461
1462	err = read_body_to_string(a, tar, &(tar->pax_header), h, unconsumed);
1463	if (err != ARCHIVE_OK)
1464		return (err);
1465
1466	/* Parse the next header. */
1467	err = tar_read_header(a, tar, entry, unconsumed);
1468	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1469		return (err);
1470
1471	/*
1472	 * TODO: Parse global/default options into 'entry' struct here
1473	 * before handling file-specific options.
1474	 *
1475	 * This design (parse standard header, then overwrite with pax
1476	 * extended attribute data) usually works well, but isn't ideal;
1477	 * it would be better to parse the pax extended attributes first
1478	 * and then skip any fields in the standard header that were
1479	 * defined in the pax header.
1480	 */
1481	err2 = pax_header(a, tar, entry, tar->pax_header.s);
1482	err =  err_combine(err, err2);
1483	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1484	return (err);
1485}
1486
1487
1488/*
1489 * Parse a file header for a Posix "ustar" archive entry.  This also
1490 * handles "pax" or "extended ustar" entries.
1491 */
1492static int
1493header_ustar(struct archive_read *a, struct tar *tar,
1494    struct archive_entry *entry, const void *h)
1495{
1496	const struct archive_entry_header_ustar	*header;
1497	struct archive_string *as;
1498	int err = ARCHIVE_OK, r;
1499
1500	header = (const struct archive_entry_header_ustar *)h;
1501
1502	/* Copy name into an internal buffer to ensure null-termination. */
1503	as = &(tar->entry_pathname);
1504	if (header->prefix[0]) {
1505		archive_strncpy(as, header->prefix, sizeof(header->prefix));
1506		if (as->s[archive_strlen(as) - 1] != '/')
1507			archive_strappend_char(as, '/');
1508		archive_strncat(as, header->name, sizeof(header->name));
1509	} else {
1510		archive_strncpy(as, header->name, sizeof(header->name));
1511	}
1512	if (archive_entry_copy_pathname_l(entry, as->s, archive_strlen(as),
1513	    tar->sconv) != 0) {
1514		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1515		if (err == ARCHIVE_FATAL)
1516			return (err);
1517	}
1518
1519	/* Handle rest of common fields. */
1520	r = header_common(a, tar, entry, h);
1521	if (r == ARCHIVE_FATAL)
1522		return (r);
1523	if (r < err)
1524		err = r;
1525
1526	/* Handle POSIX ustar fields. */
1527	if (archive_entry_copy_uname_l(entry,
1528	    header->uname, sizeof(header->uname), tar->sconv) != 0) {
1529		err = set_conversion_failed_error(a, tar->sconv, "Uname");
1530		if (err == ARCHIVE_FATAL)
1531			return (err);
1532	}
1533
1534	if (archive_entry_copy_gname_l(entry,
1535	    header->gname, sizeof(header->gname), tar->sconv) != 0) {
1536		err = set_conversion_failed_error(a, tar->sconv, "Gname");
1537		if (err == ARCHIVE_FATAL)
1538			return (err);
1539	}
1540
1541	/* Parse out device numbers only for char and block specials. */
1542	if (header->typeflag[0] == '3' || header->typeflag[0] == '4') {
1543		archive_entry_set_rdevmajor(entry, (dev_t)
1544		    tar_atol(header->rdevmajor, sizeof(header->rdevmajor)));
1545		archive_entry_set_rdevminor(entry, (dev_t)
1546		    tar_atol(header->rdevminor, sizeof(header->rdevminor)));
1547	}
1548
1549	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1550
1551	return (err);
1552}
1553
1554
1555/*
1556 * Parse the pax extended attributes record.
1557 *
1558 * Returns non-zero if there's an error in the data.
1559 */
1560static int
1561pax_header(struct archive_read *a, struct tar *tar,
1562    struct archive_entry *entry, char *attr)
1563{
1564	size_t attr_length, l, line_length;
1565	char *p;
1566	char *key, *value;
1567	struct archive_string *as;
1568	struct archive_string_conv *sconv;
1569	int err, err2;
1570
1571	attr_length = strlen(attr);
1572	tar->pax_hdrcharset_binary = 0;
1573	archive_string_empty(&(tar->entry_gname));
1574	archive_string_empty(&(tar->entry_linkpath));
1575	archive_string_empty(&(tar->entry_pathname));
1576	archive_string_empty(&(tar->entry_pathname_override));
1577	archive_string_empty(&(tar->entry_uname));
1578	err = ARCHIVE_OK;
1579	while (attr_length > 0) {
1580		/* Parse decimal length field at start of line. */
1581		line_length = 0;
1582		l = attr_length;
1583		p = attr; /* Record start of line. */
1584		while (l>0) {
1585			if (*p == ' ') {
1586				p++;
1587				l--;
1588				break;
1589			}
1590			if (*p < '0' || *p > '9') {
1591				archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1592				    "Ignoring malformed pax extended attributes");
1593				return (ARCHIVE_WARN);
1594			}
1595			line_length *= 10;
1596			line_length += *p - '0';
1597			if (line_length > 999999) {
1598				archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1599				    "Rejecting pax extended attribute > 1MB");
1600				return (ARCHIVE_WARN);
1601			}
1602			p++;
1603			l--;
1604		}
1605
1606		/*
1607		 * Parsed length must be no bigger than available data,
1608		 * at least 1, and the last character of the line must
1609		 * be '\n'.
1610		 */
1611		if (line_length > attr_length
1612		    || line_length < 1
1613		    || attr[line_length - 1] != '\n')
1614		{
1615			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1616			    "Ignoring malformed pax extended attribute");
1617			return (ARCHIVE_WARN);
1618		}
1619
1620		/* Null-terminate the line. */
1621		attr[line_length - 1] = '\0';
1622
1623		/* Find end of key and null terminate it. */
1624		key = p;
1625		if (key[0] == '=')
1626			return (-1);
1627		while (*p && *p != '=')
1628			++p;
1629		if (*p == '\0') {
1630			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1631			    "Invalid pax extended attributes");
1632			return (ARCHIVE_WARN);
1633		}
1634		*p = '\0';
1635
1636		/* Identify null-terminated 'value' portion. */
1637		value = p + 1;
1638
1639		/* Identify this attribute and set it in the entry. */
1640		err2 = pax_attribute(a, tar, entry, key, value);
1641		if (err2 == ARCHIVE_FATAL)
1642			return (err2);
1643		err = err_combine(err, err2);
1644
1645		/* Skip to next line */
1646		attr += line_length;
1647		attr_length -= line_length;
1648	}
1649
1650	/*
1651	 * PAX format uses UTF-8 as default charset for its metadata
1652	 * unless hdrcharset=BINARY is present in its header.
1653	 * We apply the charset specified by the hdrcharset option only
1654	 * when the hdrcharset attribute(in PAX header) is BINARY because
1655	 * we respect the charset described in PAX header and BINARY also
1656	 * means that metadata(filename,uname and gname) character-set
1657	 * is unknown.
1658	 */
1659	if (tar->pax_hdrcharset_binary)
1660		sconv = tar->opt_sconv;
1661	else {
1662		sconv = archive_string_conversion_from_charset(
1663		    &(a->archive), "UTF-8", 1);
1664		if (sconv == NULL)
1665			return (ARCHIVE_FATAL);
1666		if (tar->compat_2x)
1667			archive_string_conversion_set_opt(sconv,
1668			    SCONV_SET_OPT_UTF8_LIBARCHIVE2X);
1669	}
1670
1671	if (archive_strlen(&(tar->entry_gname)) > 0) {
1672		if (archive_entry_copy_gname_l(entry, tar->entry_gname.s,
1673		    archive_strlen(&(tar->entry_gname)), sconv) != 0) {
1674			err = set_conversion_failed_error(a, sconv, "Gname");
1675			if (err == ARCHIVE_FATAL)
1676				return (err);
1677			/* Use a converted an original name. */
1678			archive_entry_copy_gname(entry, tar->entry_gname.s);
1679		}
1680	}
1681	if (archive_strlen(&(tar->entry_linkpath)) > 0) {
1682		if (archive_entry_copy_link_l(entry, tar->entry_linkpath.s,
1683		    archive_strlen(&(tar->entry_linkpath)), sconv) != 0) {
1684			err = set_conversion_failed_error(a, sconv, "Linkname");
1685			if (err == ARCHIVE_FATAL)
1686				return (err);
1687			/* Use a converted an original name. */
1688			archive_entry_copy_link(entry, tar->entry_linkpath.s);
1689		}
1690	}
1691	/*
1692	 * Some extensions (such as the GNU sparse file extensions)
1693	 * deliberately store a synthetic name under the regular 'path'
1694	 * attribute and the real file name under a different attribute.
1695	 * Since we're supposed to not care about the order, we
1696	 * have no choice but to store all of the various filenames
1697	 * we find and figure it all out afterwards.  This is the
1698	 * figuring out part.
1699	 */
1700	as = NULL;
1701	if (archive_strlen(&(tar->entry_pathname_override)) > 0)
1702		as = &(tar->entry_pathname_override);
1703	else if (archive_strlen(&(tar->entry_pathname)) > 0)
1704		as = &(tar->entry_pathname);
1705	if (as != NULL) {
1706		if (archive_entry_copy_pathname_l(entry, as->s,
1707		    archive_strlen(as), sconv) != 0) {
1708			err = set_conversion_failed_error(a, sconv, "Pathname");
1709			if (err == ARCHIVE_FATAL)
1710				return (err);
1711			/* Use a converted an original name. */
1712			archive_entry_copy_pathname(entry, as->s);
1713		}
1714	}
1715	if (archive_strlen(&(tar->entry_uname)) > 0) {
1716		if (archive_entry_copy_uname_l(entry, tar->entry_uname.s,
1717		    archive_strlen(&(tar->entry_uname)), sconv) != 0) {
1718			err = set_conversion_failed_error(a, sconv, "Uname");
1719			if (err == ARCHIVE_FATAL)
1720				return (err);
1721			/* Use a converted an original name. */
1722			archive_entry_copy_uname(entry, tar->entry_uname.s);
1723		}
1724	}
1725	return (err);
1726}
1727
1728static int
1729pax_attribute_xattr(struct archive_entry *entry,
1730	const char *name, const char *value)
1731{
1732	char *name_decoded;
1733	void *value_decoded;
1734	size_t value_len;
1735
1736	if (strlen(name) < 18 || (memcmp(name, "LIBARCHIVE.xattr.", 17)) != 0)
1737		return 3;
1738
1739	name += 17;
1740
1741	/* URL-decode name */
1742	name_decoded = url_decode(name);
1743	if (name_decoded == NULL)
1744		return 2;
1745
1746	/* Base-64 decode value */
1747	value_decoded = base64_decode(value, strlen(value), &value_len);
1748	if (value_decoded == NULL) {
1749		free(name_decoded);
1750		return 1;
1751	}
1752
1753	archive_entry_xattr_add_entry(entry, name_decoded,
1754		value_decoded, value_len);
1755
1756	free(name_decoded);
1757	free(value_decoded);
1758	return 0;
1759}
1760
1761/*
1762 * Parse a single key=value attribute.  key/value pointers are
1763 * assumed to point into reasonably long-lived storage.
1764 *
1765 * Note that POSIX reserves all-lowercase keywords.  Vendor-specific
1766 * extensions should always have keywords of the form "VENDOR.attribute"
1767 * In particular, it's quite feasible to support many different
1768 * vendor extensions here.  I'm using "LIBARCHIVE" for extensions
1769 * unique to this library.
1770 *
1771 * Investigate other vendor-specific extensions and see if
1772 * any of them look useful.
1773 */
1774static int
1775pax_attribute(struct archive_read *a, struct tar *tar,
1776    struct archive_entry *entry, const char *key, const char *value)
1777{
1778	int64_t s;
1779	long n;
1780	int err = ARCHIVE_OK, r;
1781
1782#ifndef __FreeBSD__
1783	if (value == NULL)
1784		value = "";	/* Disable compiler warning; do not pass
1785				 * NULL pointer to strlen().  */
1786#endif
1787	switch (key[0]) {
1788	case 'G':
1789		/* Reject GNU.sparse.* headers on non-regular files. */
1790		if (strncmp(key, "GNU.sparse", 10) == 0 &&
1791		    !tar->sparse_allowed) {
1792			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1793			    "Non-regular file cannot be sparse");
1794			return (ARCHIVE_FATAL);
1795		}
1796
1797		/* GNU "0.0" sparse pax format. */
1798		if (strcmp(key, "GNU.sparse.numblocks") == 0) {
1799			tar->sparse_offset = -1;
1800			tar->sparse_numbytes = -1;
1801			tar->sparse_gnu_major = 0;
1802			tar->sparse_gnu_minor = 0;
1803		}
1804		if (strcmp(key, "GNU.sparse.offset") == 0) {
1805			tar->sparse_offset = tar_atol10(value, strlen(value));
1806			if (tar->sparse_numbytes != -1) {
1807				if (gnu_add_sparse_entry(a, tar,
1808				    tar->sparse_offset, tar->sparse_numbytes)
1809				    != ARCHIVE_OK)
1810					return (ARCHIVE_FATAL);
1811				tar->sparse_offset = -1;
1812				tar->sparse_numbytes = -1;
1813			}
1814		}
1815		if (strcmp(key, "GNU.sparse.numbytes") == 0) {
1816			tar->sparse_numbytes = tar_atol10(value, strlen(value));
1817			if (tar->sparse_numbytes != -1) {
1818				if (gnu_add_sparse_entry(a, tar,
1819				    tar->sparse_offset, tar->sparse_numbytes)
1820				    != ARCHIVE_OK)
1821					return (ARCHIVE_FATAL);
1822				tar->sparse_offset = -1;
1823				tar->sparse_numbytes = -1;
1824			}
1825		}
1826		if (strcmp(key, "GNU.sparse.size") == 0) {
1827			tar->realsize = tar_atol10(value, strlen(value));
1828			archive_entry_set_size(entry, tar->realsize);
1829		}
1830
1831		/* GNU "0.1" sparse pax format. */
1832		if (strcmp(key, "GNU.sparse.map") == 0) {
1833			tar->sparse_gnu_major = 0;
1834			tar->sparse_gnu_minor = 1;
1835			if (gnu_sparse_01_parse(a, tar, value) != ARCHIVE_OK)
1836				return (ARCHIVE_WARN);
1837		}
1838
1839		/* GNU "1.0" sparse pax format */
1840		if (strcmp(key, "GNU.sparse.major") == 0) {
1841			tar->sparse_gnu_major = (int)tar_atol10(value, strlen(value));
1842			tar->sparse_gnu_pending = 1;
1843		}
1844		if (strcmp(key, "GNU.sparse.minor") == 0) {
1845			tar->sparse_gnu_minor = (int)tar_atol10(value, strlen(value));
1846			tar->sparse_gnu_pending = 1;
1847		}
1848		if (strcmp(key, "GNU.sparse.name") == 0) {
1849			/*
1850			 * The real filename; when storing sparse
1851			 * files, GNU tar puts a synthesized name into
1852			 * the regular 'path' attribute in an attempt
1853			 * to limit confusion. ;-)
1854			 */
1855			archive_strcpy(&(tar->entry_pathname_override), value);
1856		}
1857		if (strcmp(key, "GNU.sparse.realsize") == 0) {
1858			tar->realsize = tar_atol10(value, strlen(value));
1859			archive_entry_set_size(entry, tar->realsize);
1860		}
1861		break;
1862	case 'L':
1863		/* Our extensions */
1864/* TODO: Handle arbitrary extended attributes... */
1865/*
1866		if (strcmp(key, "LIBARCHIVE.xxxxxxx") == 0)
1867			archive_entry_set_xxxxxx(entry, value);
1868*/
1869		if (strcmp(key, "LIBARCHIVE.creationtime") == 0) {
1870			pax_time(value, &s, &n);
1871			archive_entry_set_birthtime(entry, s, n);
1872		}
1873		if (memcmp(key, "LIBARCHIVE.xattr.", 17) == 0)
1874			pax_attribute_xattr(entry, key, value);
1875		break;
1876	case 'S':
1877		/* We support some keys used by the "star" archiver */
1878		if (strcmp(key, "SCHILY.acl.access") == 0) {
1879			if (tar->sconv_acl == NULL) {
1880				tar->sconv_acl =
1881				    archive_string_conversion_from_charset(
1882					&(a->archive), "UTF-8", 1);
1883				if (tar->sconv_acl == NULL)
1884					return (ARCHIVE_FATAL);
1885			}
1886
1887			r = archive_acl_parse_l(archive_entry_acl(entry),
1888			    value, ARCHIVE_ENTRY_ACL_TYPE_ACCESS,
1889			    tar->sconv_acl);
1890			if (r != ARCHIVE_OK) {
1891				err = r;
1892				if (err == ARCHIVE_FATAL) {
1893					archive_set_error(&a->archive, ENOMEM,
1894					    "Can't allocate memory for "
1895					    "SCHILY.acl.access");
1896					return (err);
1897				}
1898				archive_set_error(&a->archive,
1899				    ARCHIVE_ERRNO_MISC,
1900				    "Parse error: SCHILY.acl.access");
1901			}
1902		} else if (strcmp(key, "SCHILY.acl.default") == 0) {
1903			if (tar->sconv_acl == NULL) {
1904				tar->sconv_acl =
1905				    archive_string_conversion_from_charset(
1906					&(a->archive), "UTF-8", 1);
1907				if (tar->sconv_acl == NULL)
1908					return (ARCHIVE_FATAL);
1909			}
1910
1911			r = archive_acl_parse_l(archive_entry_acl(entry),
1912			    value, ARCHIVE_ENTRY_ACL_TYPE_DEFAULT,
1913			    tar->sconv_acl);
1914			if (r != ARCHIVE_OK) {
1915				err = r;
1916				if (err == ARCHIVE_FATAL) {
1917					archive_set_error(&a->archive, ENOMEM,
1918					    "Can't allocate memory for "
1919					    "SCHILY.acl.default");
1920					return (err);
1921				}
1922				archive_set_error(&a->archive,
1923				    ARCHIVE_ERRNO_MISC,
1924				    "Parse error: SCHILY.acl.default");
1925			}
1926		} else if (strcmp(key, "SCHILY.devmajor") == 0) {
1927			archive_entry_set_rdevmajor(entry,
1928			    (dev_t)tar_atol10(value, strlen(value)));
1929		} else if (strcmp(key, "SCHILY.devminor") == 0) {
1930			archive_entry_set_rdevminor(entry,
1931			    (dev_t)tar_atol10(value, strlen(value)));
1932		} else if (strcmp(key, "SCHILY.fflags") == 0) {
1933			archive_entry_copy_fflags_text(entry, value);
1934		} else if (strcmp(key, "SCHILY.dev") == 0) {
1935			archive_entry_set_dev(entry,
1936			    (dev_t)tar_atol10(value, strlen(value)));
1937		} else if (strcmp(key, "SCHILY.ino") == 0) {
1938			archive_entry_set_ino(entry,
1939			    tar_atol10(value, strlen(value)));
1940		} else if (strcmp(key, "SCHILY.nlink") == 0) {
1941			archive_entry_set_nlink(entry, (unsigned)
1942			    tar_atol10(value, strlen(value)));
1943		} else if (strcmp(key, "SCHILY.realsize") == 0) {
1944			tar->realsize = tar_atol10(value, strlen(value));
1945			archive_entry_set_size(entry, tar->realsize);
1946		} else if (strcmp(key, "SUN.holesdata") == 0) {
1947			/* A Solaris extension for sparse. */
1948			r = solaris_sparse_parse(a, tar, entry, value);
1949			if (r < err) {
1950				if (r == ARCHIVE_FATAL)
1951					return (r);
1952				err = r;
1953				archive_set_error(&a->archive,
1954				    ARCHIVE_ERRNO_MISC,
1955				    "Parse error: SUN.holesdata");
1956			}
1957		}
1958		break;
1959	case 'a':
1960		if (strcmp(key, "atime") == 0) {
1961			pax_time(value, &s, &n);
1962			archive_entry_set_atime(entry, s, n);
1963		}
1964		break;
1965	case 'c':
1966		if (strcmp(key, "ctime") == 0) {
1967			pax_time(value, &s, &n);
1968			archive_entry_set_ctime(entry, s, n);
1969		} else if (strcmp(key, "charset") == 0) {
1970			/* TODO: Publish charset information in entry. */
1971		} else if (strcmp(key, "comment") == 0) {
1972			/* TODO: Publish comment in entry. */
1973		}
1974		break;
1975	case 'g':
1976		if (strcmp(key, "gid") == 0) {
1977			archive_entry_set_gid(entry,
1978			    tar_atol10(value, strlen(value)));
1979		} else if (strcmp(key, "gname") == 0) {
1980			archive_strcpy(&(tar->entry_gname), value);
1981		}
1982		break;
1983	case 'h':
1984		if (strcmp(key, "hdrcharset") == 0) {
1985			if (strcmp(value, "BINARY") == 0)
1986				/* Binary  mode. */
1987				tar->pax_hdrcharset_binary = 1;
1988			else if (strcmp(value, "ISO-IR 10646 2000 UTF-8") == 0)
1989				tar->pax_hdrcharset_binary = 0;
1990		}
1991		break;
1992	case 'l':
1993		/* pax interchange doesn't distinguish hardlink vs. symlink. */
1994		if (strcmp(key, "linkpath") == 0) {
1995			archive_strcpy(&(tar->entry_linkpath), value);
1996		}
1997		break;
1998	case 'm':
1999		if (strcmp(key, "mtime") == 0) {
2000			pax_time(value, &s, &n);
2001			archive_entry_set_mtime(entry, s, n);
2002		}
2003		break;
2004	case 'p':
2005		if (strcmp(key, "path") == 0) {
2006			archive_strcpy(&(tar->entry_pathname), value);
2007		}
2008		break;
2009	case 'r':
2010		/* POSIX has reserved 'realtime.*' */
2011		break;
2012	case 's':
2013		/* POSIX has reserved 'security.*' */
2014		/* Someday: if (strcmp(key, "security.acl") == 0) { ... } */
2015		if (strcmp(key, "size") == 0) {
2016			/* "size" is the size of the data in the entry. */
2017			tar->entry_bytes_remaining
2018			    = tar_atol10(value, strlen(value));
2019			/*
2020			 * But, "size" is not necessarily the size of
2021			 * the file on disk; if this is a sparse file,
2022			 * the disk size may have already been set from
2023			 * GNU.sparse.realsize or GNU.sparse.size or
2024			 * an old GNU header field or SCHILY.realsize
2025			 * or ....
2026			 */
2027			if (tar->realsize < 0) {
2028				archive_entry_set_size(entry,
2029				    tar->entry_bytes_remaining);
2030				tar->realsize
2031				    = tar->entry_bytes_remaining;
2032			}
2033		}
2034		break;
2035	case 'u':
2036		if (strcmp(key, "uid") == 0) {
2037			archive_entry_set_uid(entry,
2038			    tar_atol10(value, strlen(value)));
2039		} else if (strcmp(key, "uname") == 0) {
2040			archive_strcpy(&(tar->entry_uname), value);
2041		}
2042		break;
2043	}
2044	return (err);
2045}
2046
2047
2048
2049/*
2050 * parse a decimal time value, which may include a fractional portion
2051 */
2052static void
2053pax_time(const char *p, int64_t *ps, long *pn)
2054{
2055	char digit;
2056	int64_t	s;
2057	unsigned long l;
2058	int sign;
2059	int64_t limit, last_digit_limit;
2060
2061	limit = INT64_MAX / 10;
2062	last_digit_limit = INT64_MAX % 10;
2063
2064	s = 0;
2065	sign = 1;
2066	if (*p == '-') {
2067		sign = -1;
2068		p++;
2069	}
2070	while (*p >= '0' && *p <= '9') {
2071		digit = *p - '0';
2072		if (s > limit ||
2073		    (s == limit && digit > last_digit_limit)) {
2074			s = INT64_MAX;
2075			break;
2076		}
2077		s = (s * 10) + digit;
2078		++p;
2079	}
2080
2081	*ps = s * sign;
2082
2083	/* Calculate nanoseconds. */
2084	*pn = 0;
2085
2086	if (*p != '.')
2087		return;
2088
2089	l = 100000000UL;
2090	do {
2091		++p;
2092		if (*p >= '0' && *p <= '9')
2093			*pn += (*p - '0') * l;
2094		else
2095			break;
2096	} while (l /= 10);
2097}
2098
2099/*
2100 * Parse GNU tar header
2101 */
2102static int
2103header_gnutar(struct archive_read *a, struct tar *tar,
2104    struct archive_entry *entry, const void *h, size_t *unconsumed)
2105{
2106	const struct archive_entry_header_gnutar *header;
2107	int64_t t;
2108	int err = ARCHIVE_OK;
2109
2110	/*
2111	 * GNU header is like POSIX ustar, except 'prefix' is
2112	 * replaced with some other fields. This also means the
2113	 * filename is stored as in old-style archives.
2114	 */
2115
2116	/* Grab fields common to all tar variants. */
2117	err = header_common(a, tar, entry, h);
2118	if (err == ARCHIVE_FATAL)
2119		return (err);
2120
2121	/* Copy filename over (to ensure null termination). */
2122	header = (const struct archive_entry_header_gnutar *)h;
2123	if (archive_entry_copy_pathname_l(entry,
2124	    header->name, sizeof(header->name), tar->sconv) != 0) {
2125		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
2126		if (err == ARCHIVE_FATAL)
2127			return (err);
2128	}
2129
2130	/* Fields common to ustar and GNU */
2131	/* XXX Can the following be factored out since it's common
2132	 * to ustar and gnu tar?  Is it okay to move it down into
2133	 * header_common, perhaps?  */
2134	if (archive_entry_copy_uname_l(entry,
2135	    header->uname, sizeof(header->uname), tar->sconv) != 0) {
2136		err = set_conversion_failed_error(a, tar->sconv, "Uname");
2137		if (err == ARCHIVE_FATAL)
2138			return (err);
2139	}
2140
2141	if (archive_entry_copy_gname_l(entry,
2142	    header->gname, sizeof(header->gname), tar->sconv) != 0) {
2143		err = set_conversion_failed_error(a, tar->sconv, "Gname");
2144		if (err == ARCHIVE_FATAL)
2145			return (err);
2146	}
2147
2148	/* Parse out device numbers only for char and block specials */
2149	if (header->typeflag[0] == '3' || header->typeflag[0] == '4') {
2150		archive_entry_set_rdevmajor(entry, (dev_t)
2151		    tar_atol(header->rdevmajor, sizeof(header->rdevmajor)));
2152		archive_entry_set_rdevminor(entry, (dev_t)
2153		    tar_atol(header->rdevminor, sizeof(header->rdevminor)));
2154	} else
2155		archive_entry_set_rdev(entry, 0);
2156
2157	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
2158
2159	/* Grab GNU-specific fields. */
2160	t = tar_atol(header->atime, sizeof(header->atime));
2161	if (t > 0)
2162		archive_entry_set_atime(entry, t, 0);
2163	t = tar_atol(header->ctime, sizeof(header->ctime));
2164	if (t > 0)
2165		archive_entry_set_ctime(entry, t, 0);
2166
2167	if (header->realsize[0] != 0) {
2168		tar->realsize
2169		    = tar_atol(header->realsize, sizeof(header->realsize));
2170		archive_entry_set_size(entry, tar->realsize);
2171	}
2172
2173	if (header->sparse[0].offset[0] != 0) {
2174		if (gnu_sparse_old_read(a, tar, header, unconsumed)
2175		    != ARCHIVE_OK)
2176			return (ARCHIVE_FATAL);
2177	} else {
2178		if (header->isextended[0] != 0) {
2179			/* XXX WTF? XXX */
2180		}
2181	}
2182
2183	return (err);
2184}
2185
2186static int
2187gnu_add_sparse_entry(struct archive_read *a, struct tar *tar,
2188    int64_t offset, int64_t remaining)
2189{
2190	struct sparse_block *p;
2191
2192	p = (struct sparse_block *)calloc(1, sizeof(*p));
2193	if (p == NULL) {
2194		archive_set_error(&a->archive, ENOMEM, "Out of memory");
2195		return (ARCHIVE_FATAL);
2196	}
2197	if (tar->sparse_last != NULL)
2198		tar->sparse_last->next = p;
2199	else
2200		tar->sparse_list = p;
2201	tar->sparse_last = p;
2202	if (remaining < 0 || offset < 0) {
2203		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, "Malformed sparse map data");
2204		return (ARCHIVE_FATAL);
2205	}
2206	p->offset = offset;
2207	p->remaining = remaining;
2208	return (ARCHIVE_OK);
2209}
2210
2211static void
2212gnu_clear_sparse_list(struct tar *tar)
2213{
2214	struct sparse_block *p;
2215
2216	while (tar->sparse_list != NULL) {
2217		p = tar->sparse_list;
2218		tar->sparse_list = p->next;
2219		free(p);
2220	}
2221	tar->sparse_last = NULL;
2222}
2223
2224/*
2225 * GNU tar old-format sparse data.
2226 *
2227 * GNU old-format sparse data is stored in a fixed-field
2228 * format.  Offset/size values are 11-byte octal fields (same
2229 * format as 'size' field in ustart header).  These are
2230 * stored in the header, allocating subsequent header blocks
2231 * as needed.  Extending the header in this way is a pretty
2232 * severe POSIX violation; this design has earned GNU tar a
2233 * lot of criticism.
2234 */
2235
2236static int
2237gnu_sparse_old_read(struct archive_read *a, struct tar *tar,
2238    const struct archive_entry_header_gnutar *header, size_t *unconsumed)
2239{
2240	ssize_t bytes_read;
2241	const void *data;
2242	struct extended {
2243		struct gnu_sparse sparse[21];
2244		char	isextended[1];
2245		char	padding[7];
2246	};
2247	const struct extended *ext;
2248
2249	if (gnu_sparse_old_parse(a, tar, header->sparse, 4) != ARCHIVE_OK)
2250		return (ARCHIVE_FATAL);
2251	if (header->isextended[0] == 0)
2252		return (ARCHIVE_OK);
2253
2254	do {
2255		tar_flush_unconsumed(a, unconsumed);
2256		data = __archive_read_ahead(a, 512, &bytes_read);
2257		if (bytes_read < 0)
2258			return (ARCHIVE_FATAL);
2259		if (bytes_read < 512) {
2260			archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2261			    "Truncated tar archive "
2262			    "detected while reading sparse file data");
2263			return (ARCHIVE_FATAL);
2264		}
2265		*unconsumed = 512;
2266		ext = (const struct extended *)data;
2267		if (gnu_sparse_old_parse(a, tar, ext->sparse, 21) != ARCHIVE_OK)
2268			return (ARCHIVE_FATAL);
2269	} while (ext->isextended[0] != 0);
2270	if (tar->sparse_list != NULL)
2271		tar->entry_offset = tar->sparse_list->offset;
2272	return (ARCHIVE_OK);
2273}
2274
2275static int
2276gnu_sparse_old_parse(struct archive_read *a, struct tar *tar,
2277    const struct gnu_sparse *sparse, int length)
2278{
2279	while (length > 0 && sparse->offset[0] != 0) {
2280		if (gnu_add_sparse_entry(a, tar,
2281		    tar_atol(sparse->offset, sizeof(sparse->offset)),
2282		    tar_atol(sparse->numbytes, sizeof(sparse->numbytes)))
2283		    != ARCHIVE_OK)
2284			return (ARCHIVE_FATAL);
2285		sparse++;
2286		length--;
2287	}
2288	return (ARCHIVE_OK);
2289}
2290
2291/*
2292 * GNU tar sparse format 0.0
2293 *
2294 * Beginning with GNU tar 1.15, sparse files are stored using
2295 * information in the pax extended header.  The GNU tar maintainers
2296 * have gone through a number of variations in the process of working
2297 * out this scheme; fortunately, they're all numbered.
2298 *
2299 * Sparse format 0.0 uses attribute GNU.sparse.numblocks to store the
2300 * number of blocks, and GNU.sparse.offset/GNU.sparse.numbytes to
2301 * store offset/size for each block.  The repeated instances of these
2302 * latter fields violate the pax specification (which frowns on
2303 * duplicate keys), so this format was quickly replaced.
2304 */
2305
2306/*
2307 * GNU tar sparse format 0.1
2308 *
2309 * This version replaced the offset/numbytes attributes with
2310 * a single "map" attribute that stored a list of integers.  This
2311 * format had two problems: First, the "map" attribute could be very
2312 * long, which caused problems for some implementations.  More
2313 * importantly, the sparse data was lost when extracted by archivers
2314 * that didn't recognize this extension.
2315 */
2316
2317static int
2318gnu_sparse_01_parse(struct archive_read *a, struct tar *tar, const char *p)
2319{
2320	const char *e;
2321	int64_t offset = -1, size = -1;
2322
2323	for (;;) {
2324		e = p;
2325		while (*e != '\0' && *e != ',') {
2326			if (*e < '0' || *e > '9')
2327				return (ARCHIVE_WARN);
2328			e++;
2329		}
2330		if (offset < 0) {
2331			offset = tar_atol10(p, e - p);
2332			if (offset < 0)
2333				return (ARCHIVE_WARN);
2334		} else {
2335			size = tar_atol10(p, e - p);
2336			if (size < 0)
2337				return (ARCHIVE_WARN);
2338			if (gnu_add_sparse_entry(a, tar, offset, size)
2339			    != ARCHIVE_OK)
2340				return (ARCHIVE_FATAL);
2341			offset = -1;
2342		}
2343		if (*e == '\0')
2344			return (ARCHIVE_OK);
2345		p = e + 1;
2346	}
2347}
2348
2349/*
2350 * GNU tar sparse format 1.0
2351 *
2352 * The idea: The offset/size data is stored as a series of base-10
2353 * ASCII numbers prepended to the file data, so that dearchivers that
2354 * don't support this format will extract the block map along with the
2355 * data and a separate post-process can restore the sparseness.
2356 *
2357 * Unfortunately, GNU tar 1.16 had a bug that added unnecessary
2358 * padding to the body of the file when using this format.  GNU tar
2359 * 1.17 corrected this bug without bumping the version number, so
2360 * it's not possible to support both variants.  This code supports
2361 * the later variant at the expense of not supporting the former.
2362 *
2363 * This variant also replaced GNU.sparse.size with GNU.sparse.realsize
2364 * and introduced the GNU.sparse.major/GNU.sparse.minor attributes.
2365 */
2366
2367/*
2368 * Read the next line from the input, and parse it as a decimal
2369 * integer followed by '\n'.  Returns positive integer value or
2370 * negative on error.
2371 */
2372static int64_t
2373gnu_sparse_10_atol(struct archive_read *a, struct tar *tar,
2374    int64_t *remaining, size_t *unconsumed)
2375{
2376	int64_t l, limit, last_digit_limit;
2377	const char *p;
2378	ssize_t bytes_read;
2379	int base, digit;
2380
2381	base = 10;
2382	limit = INT64_MAX / base;
2383	last_digit_limit = INT64_MAX % base;
2384
2385	/*
2386	 * Skip any lines starting with '#'; GNU tar specs
2387	 * don't require this, but they should.
2388	 */
2389	do {
2390		bytes_read = readline(a, tar, &p,
2391			(ssize_t)tar_min(*remaining, 100), unconsumed);
2392		if (bytes_read <= 0)
2393			return (ARCHIVE_FATAL);
2394		*remaining -= bytes_read;
2395	} while (p[0] == '#');
2396
2397	l = 0;
2398	while (bytes_read > 0) {
2399		if (*p == '\n')
2400			return (l);
2401		if (*p < '0' || *p >= '0' + base)
2402			return (ARCHIVE_WARN);
2403		digit = *p - '0';
2404		if (l > limit || (l == limit && digit > last_digit_limit))
2405			l = INT64_MAX; /* Truncate on overflow. */
2406		else
2407			l = (l * base) + digit;
2408		p++;
2409		bytes_read--;
2410	}
2411	/* TODO: Error message. */
2412	return (ARCHIVE_WARN);
2413}
2414
2415/*
2416 * Returns length (in bytes) of the sparse data description
2417 * that was read.
2418 */
2419static ssize_t
2420gnu_sparse_10_read(struct archive_read *a, struct tar *tar, size_t *unconsumed)
2421{
2422	ssize_t bytes_read;
2423	int entries;
2424	int64_t offset, size, to_skip, remaining;
2425
2426	/* Clear out the existing sparse list. */
2427	gnu_clear_sparse_list(tar);
2428
2429	remaining = tar->entry_bytes_remaining;
2430
2431	/* Parse entries. */
2432	entries = (int)gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2433	if (entries < 0)
2434		return (ARCHIVE_FATAL);
2435	/* Parse the individual entries. */
2436	while (entries-- > 0) {
2437		/* Parse offset/size */
2438		offset = gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2439		if (offset < 0)
2440			return (ARCHIVE_FATAL);
2441		size = gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2442		if (size < 0)
2443			return (ARCHIVE_FATAL);
2444		/* Add a new sparse entry. */
2445		if (gnu_add_sparse_entry(a, tar, offset, size) != ARCHIVE_OK)
2446			return (ARCHIVE_FATAL);
2447	}
2448	/* Skip rest of block... */
2449	tar_flush_unconsumed(a, unconsumed);
2450	bytes_read = (ssize_t)(tar->entry_bytes_remaining - remaining);
2451	to_skip = 0x1ff & -bytes_read;
2452	if (to_skip != __archive_read_consume(a, to_skip))
2453		return (ARCHIVE_FATAL);
2454	return ((ssize_t)(bytes_read + to_skip));
2455}
2456
2457/*
2458 * Solaris pax extension for a sparse file. This is recorded with the
2459 * data and hole pairs. The way recording sparse information by Solaris'
2460 * pax simply indicates where data and sparse are, so the stored contents
2461 * consist of both data and hole.
2462 */
2463static int
2464solaris_sparse_parse(struct archive_read *a, struct tar *tar,
2465    struct archive_entry *entry, const char *p)
2466{
2467	const char *e;
2468	int64_t start, end;
2469	int hole = 1;
2470
2471	(void)entry; /* UNUSED */
2472
2473	end = 0;
2474	if (*p == ' ')
2475		p++;
2476	else
2477		return (ARCHIVE_WARN);
2478	for (;;) {
2479		e = p;
2480		while (*e != '\0' && *e != ' ') {
2481			if (*e < '0' || *e > '9')
2482				return (ARCHIVE_WARN);
2483			e++;
2484		}
2485		start = end;
2486		end = tar_atol10(p, e - p);
2487		if (end < 0)
2488			return (ARCHIVE_WARN);
2489		if (start < end) {
2490			if (gnu_add_sparse_entry(a, tar, start,
2491			    end - start) != ARCHIVE_OK)
2492				return (ARCHIVE_FATAL);
2493			tar->sparse_last->hole = hole;
2494		}
2495		if (*e == '\0')
2496			return (ARCHIVE_OK);
2497		p = e + 1;
2498		hole = hole == 0;
2499	}
2500}
2501
2502/*-
2503 * Convert text->integer.
2504 *
2505 * Traditional tar formats (including POSIX) specify base-8 for
2506 * all of the standard numeric fields.  This is a significant limitation
2507 * in practice:
2508 *   = file size is limited to 8GB
2509 *   = rdevmajor and rdevminor are limited to 21 bits
2510 *   = uid/gid are limited to 21 bits
2511 *
2512 * There are two workarounds for this:
2513 *   = pax extended headers, which use variable-length string fields
2514 *   = GNU tar and STAR both allow either base-8 or base-256 in
2515 *      most fields.  The high bit is set to indicate base-256.
2516 *
2517 * On read, this implementation supports both extensions.
2518 */
2519static int64_t
2520tar_atol(const char *p, size_t char_cnt)
2521{
2522	/*
2523	 * Technically, GNU tar considers a field to be in base-256
2524	 * only if the first byte is 0xff or 0x80.
2525	 */
2526	if (*p & 0x80)
2527		return (tar_atol256(p, char_cnt));
2528	return (tar_atol8(p, char_cnt));
2529}
2530
2531/*
2532 * Note that this implementation does not (and should not!) obey
2533 * locale settings; you cannot simply substitute strtol here, since
2534 * it does obey locale.
2535 */
2536static int64_t
2537tar_atol_base_n(const char *p, size_t char_cnt, int base)
2538{
2539	int64_t	l, maxval, limit, last_digit_limit;
2540	int digit, sign;
2541
2542	maxval = INT64_MAX;
2543	limit = INT64_MAX / base;
2544	last_digit_limit = INT64_MAX % base;
2545
2546	/* the pointer will not be dereferenced if char_cnt is zero
2547	 * due to the way the && operator is evaluated.
2548	 */
2549	while (char_cnt != 0 && (*p == ' ' || *p == '\t')) {
2550		p++;
2551		char_cnt--;
2552	}
2553
2554	sign = 1;
2555	if (char_cnt != 0 && *p == '-') {
2556		sign = -1;
2557		p++;
2558		char_cnt--;
2559
2560		maxval = INT64_MIN;
2561		limit = -(INT64_MIN / base);
2562		last_digit_limit = INT64_MIN % base;
2563	}
2564
2565	l = 0;
2566	if (char_cnt != 0) {
2567		digit = *p - '0';
2568		while (digit >= 0 && digit < base  && char_cnt != 0) {
2569			if (l>limit || (l == limit && digit > last_digit_limit)) {
2570				return maxval; /* Truncate on overflow. */
2571			}
2572			l = (l * base) + digit;
2573			digit = *++p - '0';
2574			char_cnt--;
2575		}
2576	}
2577	return (sign < 0) ? -l : l;
2578}
2579
2580static int64_t
2581tar_atol8(const char *p, size_t char_cnt)
2582{
2583	return tar_atol_base_n(p, char_cnt, 8);
2584}
2585
2586static int64_t
2587tar_atol10(const char *p, size_t char_cnt)
2588{
2589	return tar_atol_base_n(p, char_cnt, 10);
2590}
2591
2592/*
2593 * Parse a base-256 integer.  This is just a variable-length
2594 * twos-complement signed binary value in big-endian order, except
2595 * that the high-order bit is ignored.  The values here can be up to
2596 * 12 bytes, so we need to be careful about overflowing 64-bit
2597 * (8-byte) integers.
2598 *
2599 * This code unashamedly assumes that the local machine uses 8-bit
2600 * bytes and twos-complement arithmetic.
2601 */
2602static int64_t
2603tar_atol256(const char *_p, size_t char_cnt)
2604{
2605	uint64_t l;
2606	const unsigned char *p = (const unsigned char *)_p;
2607	unsigned char c, neg;
2608
2609	/* Extend 7-bit 2s-comp to 8-bit 2s-comp, decide sign. */
2610	c = *p;
2611	if (c & 0x40) {
2612		neg = 0xff;
2613		c |= 0x80;
2614		l = ~ARCHIVE_LITERAL_ULL(0);
2615	} else {
2616		neg = 0;
2617		c &= 0x7f;
2618		l = 0;
2619	}
2620
2621	/* If more than 8 bytes, check that we can ignore
2622	 * high-order bits without overflow. */
2623	while (char_cnt > sizeof(int64_t)) {
2624		--char_cnt;
2625		if (c != neg)
2626			return neg ? INT64_MIN : INT64_MAX;
2627		c = *++p;
2628	}
2629
2630	/* c is first byte that fits; if sign mismatch, return overflow */
2631	if ((c ^ neg) & 0x80) {
2632		return neg ? INT64_MIN : INT64_MAX;
2633	}
2634
2635	/* Accumulate remaining bytes. */
2636	while (--char_cnt > 0) {
2637		l = (l << 8) | c;
2638		c = *++p;
2639	}
2640	l = (l << 8) | c;
2641	/* Return signed twos-complement value. */
2642	return (int64_t)(l);
2643}
2644
2645/*
2646 * Returns length of line (including trailing newline)
2647 * or negative on error.  'start' argument is updated to
2648 * point to first character of line.  This avoids copying
2649 * when possible.
2650 */
2651static ssize_t
2652readline(struct archive_read *a, struct tar *tar, const char **start,
2653    ssize_t limit, size_t *unconsumed)
2654{
2655	ssize_t bytes_read;
2656	ssize_t total_size = 0;
2657	const void *t;
2658	const char *s;
2659	void *p;
2660
2661	tar_flush_unconsumed(a, unconsumed);
2662
2663	t = __archive_read_ahead(a, 1, &bytes_read);
2664	if (bytes_read <= 0)
2665		return (ARCHIVE_FATAL);
2666	s = t;  /* Start of line? */
2667	p = memchr(t, '\n', bytes_read);
2668	/* If we found '\n' in the read buffer, return pointer to that. */
2669	if (p != NULL) {
2670		bytes_read = 1 + ((const char *)p) - s;
2671		if (bytes_read > limit) {
2672			archive_set_error(&a->archive,
2673			    ARCHIVE_ERRNO_FILE_FORMAT,
2674			    "Line too long");
2675			return (ARCHIVE_FATAL);
2676		}
2677		*unconsumed = bytes_read;
2678		*start = s;
2679		return (bytes_read);
2680	}
2681	*unconsumed = bytes_read;
2682	/* Otherwise, we need to accumulate in a line buffer. */
2683	for (;;) {
2684		if (total_size + bytes_read > limit) {
2685			archive_set_error(&a->archive,
2686			    ARCHIVE_ERRNO_FILE_FORMAT,
2687			    "Line too long");
2688			return (ARCHIVE_FATAL);
2689		}
2690		if (archive_string_ensure(&tar->line, total_size + bytes_read) == NULL) {
2691			archive_set_error(&a->archive, ENOMEM,
2692			    "Can't allocate working buffer");
2693			return (ARCHIVE_FATAL);
2694		}
2695		memcpy(tar->line.s + total_size, t, bytes_read);
2696		tar_flush_unconsumed(a, unconsumed);
2697		total_size += bytes_read;
2698		/* If we found '\n', clean up and return. */
2699		if (p != NULL) {
2700			*start = tar->line.s;
2701			return (total_size);
2702		}
2703		/* Read some more. */
2704		t = __archive_read_ahead(a, 1, &bytes_read);
2705		if (bytes_read <= 0)
2706			return (ARCHIVE_FATAL);
2707		s = t;  /* Start of line? */
2708		p = memchr(t, '\n', bytes_read);
2709		/* If we found '\n', trim the read. */
2710		if (p != NULL) {
2711			bytes_read = 1 + ((const char *)p) - s;
2712		}
2713		*unconsumed = bytes_read;
2714	}
2715}
2716
2717/*
2718 * base64_decode - Base64 decode
2719 *
2720 * This accepts most variations of base-64 encoding, including:
2721 *    * with or without line breaks
2722 *    * with or without the final group padded with '=' or '_' characters
2723 * (The most economical Base-64 variant does not pad the last group and
2724 * omits line breaks; RFC1341 used for MIME requires both.)
2725 */
2726static char *
2727base64_decode(const char *s, size_t len, size_t *out_len)
2728{
2729	static const unsigned char digits[64] = {
2730		'A','B','C','D','E','F','G','H','I','J','K','L','M','N',
2731		'O','P','Q','R','S','T','U','V','W','X','Y','Z','a','b',
2732		'c','d','e','f','g','h','i','j','k','l','m','n','o','p',
2733		'q','r','s','t','u','v','w','x','y','z','0','1','2','3',
2734		'4','5','6','7','8','9','+','/' };
2735	static unsigned char decode_table[128];
2736	char *out, *d;
2737	const unsigned char *src = (const unsigned char *)s;
2738
2739	/* If the decode table is not yet initialized, prepare it. */
2740	if (decode_table[digits[1]] != 1) {
2741		unsigned i;
2742		memset(decode_table, 0xff, sizeof(decode_table));
2743		for (i = 0; i < sizeof(digits); i++)
2744			decode_table[digits[i]] = i;
2745	}
2746
2747	/* Allocate enough space to hold the entire output. */
2748	/* Note that we may not use all of this... */
2749	out = (char *)malloc(len - len / 4 + 1);
2750	if (out == NULL) {
2751		*out_len = 0;
2752		return (NULL);
2753	}
2754	d = out;
2755
2756	while (len > 0) {
2757		/* Collect the next group of (up to) four characters. */
2758		int v = 0;
2759		int group_size = 0;
2760		while (group_size < 4 && len > 0) {
2761			/* '=' or '_' padding indicates final group. */
2762			if (*src == '=' || *src == '_') {
2763				len = 0;
2764				break;
2765			}
2766			/* Skip illegal characters (including line breaks) */
2767			if (*src > 127 || *src < 32
2768			    || decode_table[*src] == 0xff) {
2769				len--;
2770				src++;
2771				continue;
2772			}
2773			v <<= 6;
2774			v |= decode_table[*src++];
2775			len --;
2776			group_size++;
2777		}
2778		/* Align a short group properly. */
2779		v <<= 6 * (4 - group_size);
2780		/* Unpack the group we just collected. */
2781		switch (group_size) {
2782		case 4: d[2] = v & 0xff;
2783			/* FALLTHROUGH */
2784		case 3: d[1] = (v >> 8) & 0xff;
2785			/* FALLTHROUGH */
2786		case 2: d[0] = (v >> 16) & 0xff;
2787			break;
2788		case 1: /* this is invalid! */
2789			break;
2790		}
2791		d += group_size * 3 / 4;
2792	}
2793
2794	*out_len = d - out;
2795	return (out);
2796}
2797
2798static char *
2799url_decode(const char *in)
2800{
2801	char *out, *d;
2802	const char *s;
2803
2804	out = (char *)malloc(strlen(in) + 1);
2805	if (out == NULL)
2806		return (NULL);
2807	for (s = in, d = out; *s != '\0'; ) {
2808		if (s[0] == '%' && s[1] != '\0' && s[2] != '\0') {
2809			/* Try to convert % escape */
2810			int digit1 = tohex(s[1]);
2811			int digit2 = tohex(s[2]);
2812			if (digit1 >= 0 && digit2 >= 0) {
2813				/* Looks good, consume three chars */
2814				s += 3;
2815				/* Convert output */
2816				*d++ = ((digit1 << 4) | digit2);
2817				continue;
2818			}
2819			/* Else fall through and treat '%' as normal char */
2820		}
2821		*d++ = *s++;
2822	}
2823	*d = '\0';
2824	return (out);
2825}
2826
2827static int
2828tohex(int c)
2829{
2830	if (c >= '0' && c <= '9')
2831		return (c - '0');
2832	else if (c >= 'A' && c <= 'F')
2833		return (c - 'A' + 10);
2834	else if (c >= 'a' && c <= 'f')
2835		return (c - 'a' + 10);
2836	else
2837		return (-1);
2838}
2839