archive_read_support_format_tar.c revision 328827 
1/*-
2 * Copyright (c) 2003-2007 Tim Kientzle
3 * Copyright (c) 2011-2012 Michihiro NAKAJIMA
4 * Copyright (c) 2016 Martin Matuska
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19 * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 */
27
28#include "archive_platform.h"
29__FBSDID("$FreeBSD: stable/11/contrib/libarchive/libarchive/archive_read_support_format_tar.c 328827 2018-02-03 02:17:04Z mm $");
30
31#ifdef HAVE_ERRNO_H
32#include <errno.h>
33#endif
34#include <stddef.h>
35#ifdef HAVE_STDLIB_H
36#include <stdlib.h>
37#endif
38#ifdef HAVE_STRING_H
39#include <string.h>
40#endif
41
42#include "archive.h"
43#include "archive_acl_private.h" /* For ACL parsing routines. */
44#include "archive_entry.h"
45#include "archive_entry_locale.h"
46#include "archive_private.h"
47#include "archive_read_private.h"
48
49#define tar_min(a,b) ((a) < (b) ? (a) : (b))
50
51/*
52 * Layout of POSIX 'ustar' tar header.
53 */
54struct archive_entry_header_ustar {
55	char	name[100];
56	char	mode[8];
57	char	uid[8];
58	char	gid[8];
59	char	size[12];
60	char	mtime[12];
61	char	checksum[8];
62	char	typeflag[1];
63	char	linkname[100];	/* "old format" header ends here */
64	char	magic[6];	/* For POSIX: "ustar\0" */
65	char	version[2];	/* For POSIX: "00" */
66	char	uname[32];
67	char	gname[32];
68	char	rdevmajor[8];
69	char	rdevminor[8];
70	char	prefix[155];
71};
72
73/*
74 * Structure of GNU tar header
75 */
76struct gnu_sparse {
77	char	offset[12];
78	char	numbytes[12];
79};
80
81struct archive_entry_header_gnutar {
82	char	name[100];
83	char	mode[8];
84	char	uid[8];
85	char	gid[8];
86	char	size[12];
87	char	mtime[12];
88	char	checksum[8];
89	char	typeflag[1];
90	char	linkname[100];
91	char	magic[8];  /* "ustar  \0" (note blank/blank/null at end) */
92	char	uname[32];
93	char	gname[32];
94	char	rdevmajor[8];
95	char	rdevminor[8];
96	char	atime[12];
97	char	ctime[12];
98	char	offset[12];
99	char	longnames[4];
100	char	unused[1];
101	struct gnu_sparse sparse[4];
102	char	isextended[1];
103	char	realsize[12];
104	/*
105	 * Old GNU format doesn't use POSIX 'prefix' field; they use
106	 * the 'L' (longname) entry instead.
107	 */
108};
109
110/*
111 * Data specific to this format.
112 */
113struct sparse_block {
114	struct sparse_block	*next;
115	int64_t	offset;
116	int64_t	remaining;
117	int hole;
118};
119
120struct tar {
121	struct archive_string	 acl_text;
122	struct archive_string	 entry_pathname;
123	/* For "GNU.sparse.name" and other similar path extensions. */
124	struct archive_string	 entry_pathname_override;
125	struct archive_string	 entry_linkpath;
126	struct archive_string	 entry_uname;
127	struct archive_string	 entry_gname;
128	struct archive_string	 longlink;
129	struct archive_string	 longname;
130	struct archive_string	 pax_header;
131	struct archive_string	 pax_global;
132	struct archive_string	 line;
133	int			 pax_hdrcharset_binary;
134	int			 header_recursion_depth;
135	int64_t			 entry_bytes_remaining;
136	int64_t			 entry_offset;
137	int64_t			 entry_padding;
138	int64_t 		 entry_bytes_unconsumed;
139	int64_t			 realsize;
140	int			 sparse_allowed;
141	struct sparse_block	*sparse_list;
142	struct sparse_block	*sparse_last;
143	int64_t			 sparse_offset;
144	int64_t			 sparse_numbytes;
145	int			 sparse_gnu_major;
146	int			 sparse_gnu_minor;
147	char			 sparse_gnu_pending;
148
149	struct archive_string	 localname;
150	struct archive_string_conv *opt_sconv;
151	struct archive_string_conv *sconv;
152	struct archive_string_conv *sconv_acl;
153	struct archive_string_conv *sconv_default;
154	int			 init_default_conversion;
155	int			 compat_2x;
156	int			 process_mac_extensions;
157	int			 read_concatenated_archives;
158	int			 realsize_override;
159};
160
161static int	archive_block_is_null(const char *p);
162static char	*base64_decode(const char *, size_t, size_t *);
163static int	gnu_add_sparse_entry(struct archive_read *, struct tar *,
164		    int64_t offset, int64_t remaining);
165
166static void	gnu_clear_sparse_list(struct tar *);
167static int	gnu_sparse_old_read(struct archive_read *, struct tar *,
168		    const struct archive_entry_header_gnutar *header, size_t *);
169static int	gnu_sparse_old_parse(struct archive_read *, struct tar *,
170		    const struct gnu_sparse *sparse, int length);
171static int	gnu_sparse_01_parse(struct archive_read *, struct tar *,
172		    const char *);
173static ssize_t	gnu_sparse_10_read(struct archive_read *, struct tar *,
174			size_t *);
175static int	header_Solaris_ACL(struct archive_read *,  struct tar *,
176		    struct archive_entry *, const void *, size_t *);
177static int	header_common(struct archive_read *,  struct tar *,
178		    struct archive_entry *, const void *);
179static int	header_old_tar(struct archive_read *, struct tar *,
180		    struct archive_entry *, const void *);
181static int	header_pax_extensions(struct archive_read *, struct tar *,
182		    struct archive_entry *, const void *, size_t *);
183static int	header_pax_global(struct archive_read *, struct tar *,
184		    struct archive_entry *, const void *h, size_t *);
185static int	header_longlink(struct archive_read *, struct tar *,
186		    struct archive_entry *, const void *h, size_t *);
187static int	header_longname(struct archive_read *, struct tar *,
188		    struct archive_entry *, const void *h, size_t *);
189static int	read_mac_metadata_blob(struct archive_read *, struct tar *,
190		    struct archive_entry *, const void *h, size_t *);
191static int	header_volume(struct archive_read *, struct tar *,
192		    struct archive_entry *, const void *h, size_t *);
193static int	header_ustar(struct archive_read *, struct tar *,
194		    struct archive_entry *, const void *h);
195static int	header_gnutar(struct archive_read *, struct tar *,
196		    struct archive_entry *, const void *h, size_t *);
197static int	archive_read_format_tar_bid(struct archive_read *, int);
198static int	archive_read_format_tar_options(struct archive_read *,
199		    const char *, const char *);
200static int	archive_read_format_tar_cleanup(struct archive_read *);
201static int	archive_read_format_tar_read_data(struct archive_read *a,
202		    const void **buff, size_t *size, int64_t *offset);
203static int	archive_read_format_tar_skip(struct archive_read *a);
204static int	archive_read_format_tar_read_header(struct archive_read *,
205		    struct archive_entry *);
206static int	checksum(struct archive_read *, const void *);
207static int 	pax_attribute(struct archive_read *, struct tar *,
208		    struct archive_entry *, const char *key, const char *value,
209		    size_t value_length);
210static int	pax_attribute_acl(struct archive_read *, struct tar *,
211		    struct archive_entry *, const char *, int);
212static int	pax_attribute_xattr(struct archive_entry *, const char *,
213		    const char *);
214static int 	pax_header(struct archive_read *, struct tar *,
215		    struct archive_entry *, struct archive_string *);
216static void	pax_time(const char *, int64_t *sec, long *nanos);
217static ssize_t	readline(struct archive_read *, struct tar *, const char **,
218		    ssize_t limit, size_t *);
219static int	read_body_to_string(struct archive_read *, struct tar *,
220		    struct archive_string *, const void *h, size_t *);
221static int	solaris_sparse_parse(struct archive_read *, struct tar *,
222		    struct archive_entry *, const char *);
223static int64_t	tar_atol(const char *, size_t);
224static int64_t	tar_atol10(const char *, size_t);
225static int64_t	tar_atol256(const char *, size_t);
226static int64_t	tar_atol8(const char *, size_t);
227static int	tar_read_header(struct archive_read *, struct tar *,
228		    struct archive_entry *, size_t *);
229static int	tohex(int c);
230static char	*url_decode(const char *);
231static void	tar_flush_unconsumed(struct archive_read *, size_t *);
232
233
234int
235archive_read_support_format_gnutar(struct archive *a)
236{
237	archive_check_magic(a, ARCHIVE_READ_MAGIC,
238	    ARCHIVE_STATE_NEW, "archive_read_support_format_gnutar");
239	return (archive_read_support_format_tar(a));
240}
241
242
243int
244archive_read_support_format_tar(struct archive *_a)
245{
246	struct archive_read *a = (struct archive_read *)_a;
247	struct tar *tar;
248	int r;
249
250	archive_check_magic(_a, ARCHIVE_READ_MAGIC,
251	    ARCHIVE_STATE_NEW, "archive_read_support_format_tar");
252
253	tar = (struct tar *)calloc(1, sizeof(*tar));
254	if (tar == NULL) {
255		archive_set_error(&a->archive, ENOMEM,
256		    "Can't allocate tar data");
257		return (ARCHIVE_FATAL);
258	}
259#ifdef HAVE_COPYFILE_H
260	/* Set this by default on Mac OS. */
261	tar->process_mac_extensions = 1;
262#endif
263
264	r = __archive_read_register_format(a, tar, "tar",
265	    archive_read_format_tar_bid,
266	    archive_read_format_tar_options,
267	    archive_read_format_tar_read_header,
268	    archive_read_format_tar_read_data,
269	    archive_read_format_tar_skip,
270	    NULL,
271	    archive_read_format_tar_cleanup,
272	    NULL,
273	    NULL);
274
275	if (r != ARCHIVE_OK)
276		free(tar);
277	return (ARCHIVE_OK);
278}
279
280static int
281archive_read_format_tar_cleanup(struct archive_read *a)
282{
283	struct tar *tar;
284
285	tar = (struct tar *)(a->format->data);
286	gnu_clear_sparse_list(tar);
287	archive_string_free(&tar->acl_text);
288	archive_string_free(&tar->entry_pathname);
289	archive_string_free(&tar->entry_pathname_override);
290	archive_string_free(&tar->entry_linkpath);
291	archive_string_free(&tar->entry_uname);
292	archive_string_free(&tar->entry_gname);
293	archive_string_free(&tar->line);
294	archive_string_free(&tar->pax_global);
295	archive_string_free(&tar->pax_header);
296	archive_string_free(&tar->longname);
297	archive_string_free(&tar->longlink);
298	archive_string_free(&tar->localname);
299	free(tar);
300	(a->format->data) = NULL;
301	return (ARCHIVE_OK);
302}
303
304/*
305 * Validate number field
306 *
307 * This has to be pretty lenient in order to accommodate the enormous
308 * variety of tar writers in the world:
309 *  = POSIX (IEEE Std 1003.1-1988) ustar requires octal values with leading
310 *    zeros and allows fields to be terminated with space or null characters
311 *  = Many writers use different termination (in particular, libarchive
312 *    omits terminator bytes to squeeze one or two more digits)
313 *  = Many writers pad with space and omit leading zeros
314 *  = GNU tar and star write base-256 values if numbers are too
315 *    big to be represented in octal
316 *
317 *  Examples of specific tar headers that we should support:
318 *  = Perl Archive::Tar terminates uid, gid, devminor and devmajor with two
319 *    null bytes, pads size with spaces and other numeric fields with zeroes
320 *  = plexus-archiver prior to 2.6.3 (before switching to commons-compress)
321 *    may have uid and gid fields filled with spaces without any octal digits
322 *    at all and pads all numeric fields with spaces
323 *
324 * This should tolerate all variants in use.  It will reject a field
325 * where the writer just left garbage after a trailing NUL.
326 */
327static int
328validate_number_field(const char* p_field, size_t i_size)
329{
330	unsigned char marker = (unsigned char)p_field[0];
331	if (marker == 128 || marker == 255 || marker == 0) {
332		/* Base-256 marker, there's nothing we can check. */
333		return 1;
334	} else {
335		/* Must be octal */
336		size_t i = 0;
337		/* Skip any leading spaces */
338		while (i < i_size && p_field[i] == ' ') {
339			++i;
340		}
341		/* Skip octal digits. */
342		while (i < i_size && p_field[i] >= '0' && p_field[i] <= '7') {
343			++i;
344		}
345		/* Any remaining characters must be space or NUL padding. */
346		while (i < i_size) {
347			if (p_field[i] != ' ' && p_field[i] != 0) {
348				return 0;
349			}
350			++i;
351		}
352		return 1;
353	}
354}
355
356static int
357archive_read_format_tar_bid(struct archive_read *a, int best_bid)
358{
359	int bid;
360	const char *h;
361	const struct archive_entry_header_ustar *header;
362
363	(void)best_bid; /* UNUSED */
364
365	bid = 0;
366
367	/* Now let's look at the actual header and see if it matches. */
368	h = __archive_read_ahead(a, 512, NULL);
369	if (h == NULL)
370		return (-1);
371
372	/* If it's an end-of-archive mark, we can handle it. */
373	if (h[0] == 0 && archive_block_is_null(h)) {
374		/*
375		 * Usually, I bid the number of bits verified, but
376		 * in this case, 4096 seems excessive so I picked 10 as
377		 * an arbitrary but reasonable-seeming value.
378		 */
379		return (10);
380	}
381
382	/* If it's not an end-of-archive mark, it must have a valid checksum.*/
383	if (!checksum(a, h))
384		return (0);
385	bid += 48;  /* Checksum is usually 6 octal digits. */
386
387	header = (const struct archive_entry_header_ustar *)h;
388
389	/* Recognize POSIX formats. */
390	if ((memcmp(header->magic, "ustar\0", 6) == 0)
391	    && (memcmp(header->version, "00", 2) == 0))
392		bid += 56;
393
394	/* Recognize GNU tar format. */
395	if ((memcmp(header->magic, "ustar ", 6) == 0)
396	    && (memcmp(header->version, " \0", 2) == 0))
397		bid += 56;
398
399	/* Type flag must be null, digit or A-Z, a-z. */
400	if (header->typeflag[0] != 0 &&
401	    !( header->typeflag[0] >= '0' && header->typeflag[0] <= '9') &&
402	    !( header->typeflag[0] >= 'A' && header->typeflag[0] <= 'Z') &&
403	    !( header->typeflag[0] >= 'a' && header->typeflag[0] <= 'z') )
404		return (0);
405	bid += 2;  /* 6 bits of variation in an 8-bit field leaves 2 bits. */
406
407	/*
408	 * Check format of mode/uid/gid/mtime/size/rdevmajor/rdevminor fields.
409	 */
410	if (bid > 0 && (
411	    validate_number_field(header->mode, sizeof(header->mode)) == 0
412	    || validate_number_field(header->uid, sizeof(header->uid)) == 0
413	    || validate_number_field(header->gid, sizeof(header->gid)) == 0
414	    || validate_number_field(header->mtime, sizeof(header->mtime)) == 0
415	    || validate_number_field(header->size, sizeof(header->size)) == 0
416	    || validate_number_field(header->rdevmajor, sizeof(header->rdevmajor)) == 0
417	    || validate_number_field(header->rdevminor, sizeof(header->rdevminor)) == 0)) {
418		bid = 0;
419	}
420
421	return (bid);
422}
423
424static int
425archive_read_format_tar_options(struct archive_read *a,
426    const char *key, const char *val)
427{
428	struct tar *tar;
429	int ret = ARCHIVE_FAILED;
430
431	tar = (struct tar *)(a->format->data);
432	if (strcmp(key, "compat-2x")  == 0) {
433		/* Handle UTF-8 filenames as libarchive 2.x */
434		tar->compat_2x = (val != NULL && val[0] != 0);
435		tar->init_default_conversion = tar->compat_2x;
436		return (ARCHIVE_OK);
437	} else if (strcmp(key, "hdrcharset")  == 0) {
438		if (val == NULL || val[0] == 0)
439			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
440			    "tar: hdrcharset option needs a character-set name");
441		else {
442			tar->opt_sconv =
443			    archive_string_conversion_from_charset(
444				&a->archive, val, 0);
445			if (tar->opt_sconv != NULL)
446				ret = ARCHIVE_OK;
447			else
448				ret = ARCHIVE_FATAL;
449		}
450		return (ret);
451	} else if (strcmp(key, "mac-ext") == 0) {
452		tar->process_mac_extensions = (val != NULL && val[0] != 0);
453		return (ARCHIVE_OK);
454	} else if (strcmp(key, "read_concatenated_archives") == 0) {
455		tar->read_concatenated_archives = (val != NULL && val[0] != 0);
456		return (ARCHIVE_OK);
457	}
458
459	/* Note: The "warn" return is just to inform the options
460	 * supervisor that we didn't handle it.  It will generate
461	 * a suitable error if no one used this option. */
462	return (ARCHIVE_WARN);
463}
464
465/* utility function- this exists to centralize the logic of tracking
466 * how much unconsumed data we have floating around, and to consume
467 * anything outstanding since we're going to do read_aheads
468 */
469static void
470tar_flush_unconsumed(struct archive_read *a, size_t *unconsumed)
471{
472	if (*unconsumed) {
473/*
474		void *data = (void *)__archive_read_ahead(a, *unconsumed, NULL);
475		 * this block of code is to poison claimed unconsumed space, ensuring
476		 * things break if it is in use still.
477		 * currently it WILL break things, so enable it only for debugging this issue
478		if (data) {
479			memset(data, 0xff, *unconsumed);
480		}
481*/
482		__archive_read_consume(a, *unconsumed);
483		*unconsumed = 0;
484	}
485}
486
487/*
488 * The function invoked by archive_read_next_header().  This
489 * just sets up a few things and then calls the internal
490 * tar_read_header() function below.
491 */
492static int
493archive_read_format_tar_read_header(struct archive_read *a,
494    struct archive_entry *entry)
495{
496	/*
497	 * When converting tar archives to cpio archives, it is
498	 * essential that each distinct file have a distinct inode
499	 * number.  To simplify this, we keep a static count here to
500	 * assign fake dev/inode numbers to each tar entry.  Note that
501	 * pax format archives may overwrite this with something more
502	 * useful.
503	 *
504	 * Ideally, we would track every file read from the archive so
505	 * that we could assign the same dev/ino pair to hardlinks,
506	 * but the memory required to store a complete lookup table is
507	 * probably not worthwhile just to support the relatively
508	 * obscure tar->cpio conversion case.
509	 */
510	static int default_inode;
511	static int default_dev;
512	struct tar *tar;
513	const char *p;
514	const wchar_t *wp;
515	int r;
516	size_t l, unconsumed = 0;
517
518	/* Assign default device/inode values. */
519	archive_entry_set_dev(entry, 1 + default_dev); /* Don't use zero. */
520	archive_entry_set_ino(entry, ++default_inode); /* Don't use zero. */
521	/* Limit generated st_ino number to 16 bits. */
522	if (default_inode >= 0xffff) {
523		++default_dev;
524		default_inode = 0;
525	}
526
527	tar = (struct tar *)(a->format->data);
528	tar->entry_offset = 0;
529	gnu_clear_sparse_list(tar);
530	tar->realsize = -1; /* Mark this as "unset" */
531	tar->realsize_override = 0;
532
533	/* Setup default string conversion. */
534	tar->sconv = tar->opt_sconv;
535	if (tar->sconv == NULL) {
536		if (!tar->init_default_conversion) {
537			tar->sconv_default =
538			    archive_string_default_conversion_for_read(&(a->archive));
539			tar->init_default_conversion = 1;
540		}
541		tar->sconv = tar->sconv_default;
542	}
543
544	r = tar_read_header(a, tar, entry, &unconsumed);
545
546	tar_flush_unconsumed(a, &unconsumed);
547
548	/*
549	 * "non-sparse" files are really just sparse files with
550	 * a single block.
551	 */
552	if (tar->sparse_list == NULL) {
553		if (gnu_add_sparse_entry(a, tar, 0, tar->entry_bytes_remaining)
554		    != ARCHIVE_OK)
555			return (ARCHIVE_FATAL);
556	} else {
557		struct sparse_block *sb;
558
559		for (sb = tar->sparse_list; sb != NULL; sb = sb->next) {
560			if (!sb->hole)
561				archive_entry_sparse_add_entry(entry,
562				    sb->offset, sb->remaining);
563		}
564	}
565
566	if (r == ARCHIVE_OK && archive_entry_filetype(entry) == AE_IFREG) {
567		/*
568		 * "Regular" entry with trailing '/' is really
569		 * directory: This is needed for certain old tar
570		 * variants and even for some broken newer ones.
571		 */
572		if ((wp = archive_entry_pathname_w(entry)) != NULL) {
573			l = wcslen(wp);
574			if (l > 0 && wp[l - 1] == L'/') {
575				archive_entry_set_filetype(entry, AE_IFDIR);
576			}
577		} else if ((p = archive_entry_pathname(entry)) != NULL) {
578			l = strlen(p);
579			if (l > 0 && p[l - 1] == '/') {
580				archive_entry_set_filetype(entry, AE_IFDIR);
581			}
582		}
583	}
584	return (r);
585}
586
587static int
588archive_read_format_tar_read_data(struct archive_read *a,
589    const void **buff, size_t *size, int64_t *offset)
590{
591	ssize_t bytes_read;
592	struct tar *tar;
593	struct sparse_block *p;
594
595	tar = (struct tar *)(a->format->data);
596
597	for (;;) {
598		/* Remove exhausted entries from sparse list. */
599		while (tar->sparse_list != NULL &&
600		    tar->sparse_list->remaining == 0) {
601			p = tar->sparse_list;
602			tar->sparse_list = p->next;
603			free(p);
604		}
605
606		if (tar->entry_bytes_unconsumed) {
607			__archive_read_consume(a, tar->entry_bytes_unconsumed);
608			tar->entry_bytes_unconsumed = 0;
609		}
610
611		/* If we're at end of file, return EOF. */
612		if (tar->sparse_list == NULL ||
613		    tar->entry_bytes_remaining == 0) {
614			if (__archive_read_consume(a, tar->entry_padding) < 0)
615				return (ARCHIVE_FATAL);
616			tar->entry_padding = 0;
617			*buff = NULL;
618			*size = 0;
619			*offset = tar->realsize;
620			return (ARCHIVE_EOF);
621		}
622
623		*buff = __archive_read_ahead(a, 1, &bytes_read);
624		if (bytes_read < 0)
625			return (ARCHIVE_FATAL);
626		if (*buff == NULL) {
627			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
628			    "Truncated tar archive");
629			return (ARCHIVE_FATAL);
630		}
631		if (bytes_read > tar->entry_bytes_remaining)
632			bytes_read = (ssize_t)tar->entry_bytes_remaining;
633		/* Don't read more than is available in the
634		 * current sparse block. */
635		if (tar->sparse_list->remaining < bytes_read)
636			bytes_read = (ssize_t)tar->sparse_list->remaining;
637		*size = bytes_read;
638		*offset = tar->sparse_list->offset;
639		tar->sparse_list->remaining -= bytes_read;
640		tar->sparse_list->offset += bytes_read;
641		tar->entry_bytes_remaining -= bytes_read;
642		tar->entry_bytes_unconsumed = bytes_read;
643
644		if (!tar->sparse_list->hole)
645			return (ARCHIVE_OK);
646		/* Current is hole data and skip this. */
647	}
648}
649
650static int
651archive_read_format_tar_skip(struct archive_read *a)
652{
653	int64_t bytes_skipped;
654	int64_t request;
655	struct sparse_block *p;
656	struct tar* tar;
657
658	tar = (struct tar *)(a->format->data);
659
660	/* Do not consume the hole of a sparse file. */
661	request = 0;
662	for (p = tar->sparse_list; p != NULL; p = p->next) {
663		if (!p->hole) {
664			if (p->remaining >= INT64_MAX - request) {
665				return ARCHIVE_FATAL;
666			}
667			request += p->remaining;
668		}
669	}
670	if (request > tar->entry_bytes_remaining)
671		request = tar->entry_bytes_remaining;
672	request += tar->entry_padding + tar->entry_bytes_unconsumed;
673
674	bytes_skipped = __archive_read_consume(a, request);
675	if (bytes_skipped < 0)
676		return (ARCHIVE_FATAL);
677
678	tar->entry_bytes_remaining = 0;
679	tar->entry_bytes_unconsumed = 0;
680	tar->entry_padding = 0;
681
682	/* Free the sparse list. */
683	gnu_clear_sparse_list(tar);
684
685	return (ARCHIVE_OK);
686}
687
688/*
689 * This function recursively interprets all of the headers associated
690 * with a single entry.
691 */
692static int
693tar_read_header(struct archive_read *a, struct tar *tar,
694    struct archive_entry *entry, size_t *unconsumed)
695{
696	ssize_t bytes;
697	int err;
698	const char *h;
699	const struct archive_entry_header_ustar *header;
700	const struct archive_entry_header_gnutar *gnuheader;
701
702	/* Loop until we find a workable header record. */
703	for (;;) {
704		tar_flush_unconsumed(a, unconsumed);
705
706		/* Read 512-byte header record */
707		h = __archive_read_ahead(a, 512, &bytes);
708		if (bytes < 0)
709			return ((int)bytes);
710		if (bytes == 0) { /* EOF at a block boundary. */
711			/* Some writers do omit the block of nulls. <sigh> */
712			return (ARCHIVE_EOF);
713		}
714		if (bytes < 512) {  /* Short block at EOF; this is bad. */
715			archive_set_error(&a->archive,
716			    ARCHIVE_ERRNO_FILE_FORMAT,
717			    "Truncated tar archive");
718			return (ARCHIVE_FATAL);
719		}
720		*unconsumed = 512;
721
722		/* Header is workable if it's not an end-of-archive mark. */
723		if (h[0] != 0 || !archive_block_is_null(h))
724			break;
725
726		/* Ensure format is set for archives with only null blocks. */
727		if (a->archive.archive_format_name == NULL) {
728			a->archive.archive_format = ARCHIVE_FORMAT_TAR;
729			a->archive.archive_format_name = "tar";
730		}
731
732		if (!tar->read_concatenated_archives) {
733			/* Try to consume a second all-null record, as well. */
734			tar_flush_unconsumed(a, unconsumed);
735			h = __archive_read_ahead(a, 512, NULL);
736			if (h != NULL && h[0] == 0 && archive_block_is_null(h))
737				__archive_read_consume(a, 512);
738			archive_clear_error(&a->archive);
739			return (ARCHIVE_EOF);
740		}
741
742		/*
743		 * We're reading concatenated archives, ignore this block and
744		 * loop to get the next.
745		 */
746	}
747
748	/*
749	 * Note: If the checksum fails and we return ARCHIVE_RETRY,
750	 * then the client is likely to just retry.  This is a very
751	 * crude way to search for the next valid header!
752	 *
753	 * TODO: Improve this by implementing a real header scan.
754	 */
755	if (!checksum(a, h)) {
756		tar_flush_unconsumed(a, unconsumed);
757		archive_set_error(&a->archive, EINVAL, "Damaged tar archive");
758		return (ARCHIVE_RETRY); /* Retryable: Invalid header */
759	}
760
761	if (++tar->header_recursion_depth > 32) {
762		tar_flush_unconsumed(a, unconsumed);
763		archive_set_error(&a->archive, EINVAL, "Too many special headers");
764		return (ARCHIVE_WARN);
765	}
766
767	/* Determine the format variant. */
768	header = (const struct archive_entry_header_ustar *)h;
769
770	switch(header->typeflag[0]) {
771	case 'A': /* Solaris tar ACL */
772		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
773		a->archive.archive_format_name = "Solaris tar";
774		err = header_Solaris_ACL(a, tar, entry, h, unconsumed);
775		break;
776	case 'g': /* POSIX-standard 'g' header. */
777		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
778		a->archive.archive_format_name = "POSIX pax interchange format";
779		err = header_pax_global(a, tar, entry, h, unconsumed);
780		if (err == ARCHIVE_EOF)
781			return (err);
782		break;
783	case 'K': /* Long link name (GNU tar, others) */
784		err = header_longlink(a, tar, entry, h, unconsumed);
785		break;
786	case 'L': /* Long filename (GNU tar, others) */
787		err = header_longname(a, tar, entry, h, unconsumed);
788		break;
789	case 'V': /* GNU volume header */
790		err = header_volume(a, tar, entry, h, unconsumed);
791		break;
792	case 'X': /* Used by SUN tar; same as 'x'. */
793		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
794		a->archive.archive_format_name =
795		    "POSIX pax interchange format (Sun variant)";
796		err = header_pax_extensions(a, tar, entry, h, unconsumed);
797		break;
798	case 'x': /* POSIX-standard 'x' header. */
799		a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
800		a->archive.archive_format_name = "POSIX pax interchange format";
801		err = header_pax_extensions(a, tar, entry, h, unconsumed);
802		break;
803	default:
804		gnuheader = (const struct archive_entry_header_gnutar *)h;
805		if (memcmp(gnuheader->magic, "ustar  \0", 8) == 0) {
806			a->archive.archive_format = ARCHIVE_FORMAT_TAR_GNUTAR;
807			a->archive.archive_format_name = "GNU tar format";
808			err = header_gnutar(a, tar, entry, h, unconsumed);
809		} else if (memcmp(header->magic, "ustar", 5) == 0) {
810			if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) {
811				a->archive.archive_format = ARCHIVE_FORMAT_TAR_USTAR;
812				a->archive.archive_format_name = "POSIX ustar format";
813			}
814			err = header_ustar(a, tar, entry, h);
815		} else {
816			a->archive.archive_format = ARCHIVE_FORMAT_TAR;
817			a->archive.archive_format_name = "tar (non-POSIX)";
818			err = header_old_tar(a, tar, entry, h);
819		}
820	}
821	if (err == ARCHIVE_FATAL)
822		return (err);
823
824	tar_flush_unconsumed(a, unconsumed);
825
826	h = NULL;
827	header = NULL;
828
829	--tar->header_recursion_depth;
830	/* Yuck.  Apple's design here ends up storing long pathname
831	 * extensions for both the AppleDouble extension entry and the
832	 * regular entry.
833	 */
834	if ((err == ARCHIVE_WARN || err == ARCHIVE_OK) &&
835	    tar->header_recursion_depth == 0 &&
836	    tar->process_mac_extensions) {
837		int err2 = read_mac_metadata_blob(a, tar, entry, h, unconsumed);
838		if (err2 < err)
839			err = err2;
840	}
841
842	/* We return warnings or success as-is.  Anything else is fatal. */
843	if (err == ARCHIVE_WARN || err == ARCHIVE_OK) {
844		if (tar->sparse_gnu_pending) {
845			if (tar->sparse_gnu_major == 1 &&
846			    tar->sparse_gnu_minor == 0) {
847				ssize_t bytes_read;
848
849				tar->sparse_gnu_pending = 0;
850				/* Read initial sparse map. */
851				bytes_read = gnu_sparse_10_read(a, tar, unconsumed);
852				if (bytes_read < 0)
853					return ((int)bytes_read);
854				tar->entry_bytes_remaining -= bytes_read;
855			} else {
856				archive_set_error(&a->archive,
857				    ARCHIVE_ERRNO_MISC,
858				    "Unrecognized GNU sparse file format");
859				return (ARCHIVE_WARN);
860			}
861			tar->sparse_gnu_pending = 0;
862		}
863		return (err);
864	}
865	if (err == ARCHIVE_EOF)
866		/* EOF when recursively reading a header is bad. */
867		archive_set_error(&a->archive, EINVAL, "Damaged tar archive");
868	return (ARCHIVE_FATAL);
869}
870
871/*
872 * Return true if block checksum is correct.
873 */
874static int
875checksum(struct archive_read *a, const void *h)
876{
877	const unsigned char *bytes;
878	const struct archive_entry_header_ustar	*header;
879	int check, sum;
880	size_t i;
881
882	(void)a; /* UNUSED */
883	bytes = (const unsigned char *)h;
884	header = (const struct archive_entry_header_ustar *)h;
885
886	/* Checksum field must hold an octal number */
887	for (i = 0; i < sizeof(header->checksum); ++i) {
888		char c = header->checksum[i];
889		if (c != ' ' && c != '\0' && (c < '0' || c > '7'))
890			return 0;
891	}
892
893	/*
894	 * Test the checksum.  Note that POSIX specifies _unsigned_
895	 * bytes for this calculation.
896	 */
897	sum = (int)tar_atol(header->checksum, sizeof(header->checksum));
898	check = 0;
899	for (i = 0; i < 148; i++)
900		check += (unsigned char)bytes[i];
901	for (; i < 156; i++)
902		check += 32;
903	for (; i < 512; i++)
904		check += (unsigned char)bytes[i];
905	if (sum == check)
906		return (1);
907
908	/*
909	 * Repeat test with _signed_ bytes, just in case this archive
910	 * was created by an old BSD, Solaris, or HP-UX tar with a
911	 * broken checksum calculation.
912	 */
913	check = 0;
914	for (i = 0; i < 148; i++)
915		check += (signed char)bytes[i];
916	for (; i < 156; i++)
917		check += 32;
918	for (; i < 512; i++)
919		check += (signed char)bytes[i];
920	if (sum == check)
921		return (1);
922
923	return (0);
924}
925
926/*
927 * Return true if this block contains only nulls.
928 */
929static int
930archive_block_is_null(const char *p)
931{
932	unsigned i;
933
934	for (i = 0; i < 512; i++)
935		if (*p++)
936			return (0);
937	return (1);
938}
939
940/*
941 * Interpret 'A' Solaris ACL header
942 */
943static int
944header_Solaris_ACL(struct archive_read *a, struct tar *tar,
945    struct archive_entry *entry, const void *h, size_t *unconsumed)
946{
947	const struct archive_entry_header_ustar *header;
948	size_t size;
949	int err, acl_type;
950	int64_t type;
951	char *acl, *p;
952
953	/*
954	 * read_body_to_string adds a NUL terminator, but we need a little
955	 * more to make sure that we don't overrun acl_text later.
956	 */
957	header = (const struct archive_entry_header_ustar *)h;
958	size = (size_t)tar_atol(header->size, sizeof(header->size));
959	err = read_body_to_string(a, tar, &(tar->acl_text), h, unconsumed);
960	if (err != ARCHIVE_OK)
961		return (err);
962
963	/* Recursively read next header */
964	err = tar_read_header(a, tar, entry, unconsumed);
965	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
966		return (err);
967
968	/* TODO: Examine the first characters to see if this
969	 * is an AIX ACL descriptor.  We'll likely never support
970	 * them, but it would be polite to recognize and warn when
971	 * we do see them. */
972
973	/* Leading octal number indicates ACL type and number of entries. */
974	p = acl = tar->acl_text.s;
975	type = 0;
976	while (*p != '\0' && p < acl + size) {
977		if (*p < '0' || *p > '7') {
978			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
979			    "Malformed Solaris ACL attribute (invalid digit)");
980			return(ARCHIVE_WARN);
981		}
982		type <<= 3;
983		type += *p - '0';
984		if (type > 077777777) {
985			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
986			    "Malformed Solaris ACL attribute (count too large)");
987			return (ARCHIVE_WARN);
988		}
989		p++;
990	}
991	switch ((int)type & ~0777777) {
992	case 01000000:
993		/* POSIX.1e ACL */
994		acl_type = ARCHIVE_ENTRY_ACL_TYPE_ACCESS;
995		break;
996	case 03000000:
997		/* NFSv4 ACL */
998		acl_type = ARCHIVE_ENTRY_ACL_TYPE_NFS4;
999		break;
1000	default:
1001		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1002		    "Malformed Solaris ACL attribute (unsupported type %o)",
1003		    (int)type);
1004		return (ARCHIVE_WARN);
1005	}
1006	p++;
1007
1008	if (p >= acl + size) {
1009		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1010		    "Malformed Solaris ACL attribute (body overflow)");
1011		return(ARCHIVE_WARN);
1012	}
1013
1014	/* ACL text is null-terminated; find the end. */
1015	size -= (p - acl);
1016	acl = p;
1017
1018	while (*p != '\0' && p < acl + size)
1019		p++;
1020
1021	if (tar->sconv_acl == NULL) {
1022		tar->sconv_acl = archive_string_conversion_from_charset(
1023		    &(a->archive), "UTF-8", 1);
1024		if (tar->sconv_acl == NULL)
1025			return (ARCHIVE_FATAL);
1026	}
1027	archive_strncpy(&(tar->localname), acl, p - acl);
1028	err = archive_acl_from_text_l(archive_entry_acl(entry),
1029	    tar->localname.s, acl_type, tar->sconv_acl);
1030	if (err != ARCHIVE_OK) {
1031		if (errno == ENOMEM) {
1032			archive_set_error(&a->archive, ENOMEM,
1033			    "Can't allocate memory for ACL");
1034		} else
1035			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1036			    "Malformed Solaris ACL attribute (unparsable)");
1037	}
1038	return (err);
1039}
1040
1041/*
1042 * Interpret 'K' long linkname header.
1043 */
1044static int
1045header_longlink(struct archive_read *a, struct tar *tar,
1046    struct archive_entry *entry, const void *h, size_t *unconsumed)
1047{
1048	int err;
1049
1050	err = read_body_to_string(a, tar, &(tar->longlink), h, unconsumed);
1051	if (err != ARCHIVE_OK)
1052		return (err);
1053	err = tar_read_header(a, tar, entry, unconsumed);
1054	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1055		return (err);
1056	/* Set symlink if symlink already set, else hardlink. */
1057	archive_entry_copy_link(entry, tar->longlink.s);
1058	return (ARCHIVE_OK);
1059}
1060
1061static int
1062set_conversion_failed_error(struct archive_read *a,
1063    struct archive_string_conv *sconv, const char *name)
1064{
1065	if (errno == ENOMEM) {
1066		archive_set_error(&a->archive, ENOMEM,
1067		    "Can't allocate memory for %s", name);
1068		return (ARCHIVE_FATAL);
1069	}
1070	archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1071	    "%s can't be converted from %s to current locale.",
1072	    name, archive_string_conversion_charset_name(sconv));
1073	return (ARCHIVE_WARN);
1074}
1075
1076/*
1077 * Interpret 'L' long filename header.
1078 */
1079static int
1080header_longname(struct archive_read *a, struct tar *tar,
1081    struct archive_entry *entry, const void *h, size_t *unconsumed)
1082{
1083	int err;
1084
1085	err = read_body_to_string(a, tar, &(tar->longname), h, unconsumed);
1086	if (err != ARCHIVE_OK)
1087		return (err);
1088	/* Read and parse "real" header, then override name. */
1089	err = tar_read_header(a, tar, entry, unconsumed);
1090	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1091		return (err);
1092	if (archive_entry_copy_pathname_l(entry, tar->longname.s,
1093	    archive_strlen(&(tar->longname)), tar->sconv) != 0)
1094		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1095	return (err);
1096}
1097
1098
1099/*
1100 * Interpret 'V' GNU tar volume header.
1101 */
1102static int
1103header_volume(struct archive_read *a, struct tar *tar,
1104    struct archive_entry *entry, const void *h, size_t *unconsumed)
1105{
1106	(void)h;
1107
1108	/* Just skip this and read the next header. */
1109	return (tar_read_header(a, tar, entry, unconsumed));
1110}
1111
1112/*
1113 * Read body of an archive entry into an archive_string object.
1114 */
1115static int
1116read_body_to_string(struct archive_read *a, struct tar *tar,
1117    struct archive_string *as, const void *h, size_t *unconsumed)
1118{
1119	int64_t size;
1120	const struct archive_entry_header_ustar *header;
1121	const void *src;
1122
1123	(void)tar; /* UNUSED */
1124	header = (const struct archive_entry_header_ustar *)h;
1125	size  = tar_atol(header->size, sizeof(header->size));
1126	if ((size > 1048576) || (size < 0)) {
1127		archive_set_error(&a->archive, EINVAL,
1128		    "Special header too large");
1129		return (ARCHIVE_FATAL);
1130	}
1131
1132	/* Fail if we can't make our buffer big enough. */
1133	if (archive_string_ensure(as, (size_t)size+1) == NULL) {
1134		archive_set_error(&a->archive, ENOMEM,
1135		    "No memory");
1136		return (ARCHIVE_FATAL);
1137	}
1138
1139	tar_flush_unconsumed(a, unconsumed);
1140
1141	/* Read the body into the string. */
1142	*unconsumed = (size_t)((size + 511) & ~ 511);
1143	src = __archive_read_ahead(a, *unconsumed, NULL);
1144	if (src == NULL) {
1145		*unconsumed = 0;
1146		return (ARCHIVE_FATAL);
1147	}
1148	memcpy(as->s, src, (size_t)size);
1149	as->s[size] = '\0';
1150	as->length = (size_t)size;
1151	return (ARCHIVE_OK);
1152}
1153
1154/*
1155 * Parse out common header elements.
1156 *
1157 * This would be the same as header_old_tar, except that the
1158 * filename is handled slightly differently for old and POSIX
1159 * entries  (POSIX entries support a 'prefix').  This factoring
1160 * allows header_old_tar and header_ustar
1161 * to handle filenames differently, while still putting most of the
1162 * common parsing into one place.
1163 */
1164static int
1165header_common(struct archive_read *a, struct tar *tar,
1166    struct archive_entry *entry, const void *h)
1167{
1168	const struct archive_entry_header_ustar	*header;
1169	char	tartype;
1170	int     err = ARCHIVE_OK;
1171
1172	header = (const struct archive_entry_header_ustar *)h;
1173	if (header->linkname[0])
1174		archive_strncpy(&(tar->entry_linkpath),
1175		    header->linkname, sizeof(header->linkname));
1176	else
1177		archive_string_empty(&(tar->entry_linkpath));
1178
1179	/* Parse out the numeric fields (all are octal) */
1180	archive_entry_set_mode(entry,
1181		(mode_t)tar_atol(header->mode, sizeof(header->mode)));
1182	archive_entry_set_uid(entry, tar_atol(header->uid, sizeof(header->uid)));
1183	archive_entry_set_gid(entry, tar_atol(header->gid, sizeof(header->gid)));
1184	tar->entry_bytes_remaining = tar_atol(header->size, sizeof(header->size));
1185	if (tar->entry_bytes_remaining < 0) {
1186		tar->entry_bytes_remaining = 0;
1187		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1188		    "Tar entry has negative size");
1189		return (ARCHIVE_FATAL);
1190	}
1191	if (tar->entry_bytes_remaining == INT64_MAX) {
1192		/* Note: tar_atol returns INT64_MAX on overflow */
1193		tar->entry_bytes_remaining = 0;
1194		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1195		    "Tar entry size overflow");
1196		return (ARCHIVE_FATAL);
1197	}
1198	tar->realsize = tar->entry_bytes_remaining;
1199	archive_entry_set_size(entry, tar->entry_bytes_remaining);
1200	archive_entry_set_mtime(entry, tar_atol(header->mtime, sizeof(header->mtime)), 0);
1201
1202	/* Handle the tar type flag appropriately. */
1203	tartype = header->typeflag[0];
1204
1205	switch (tartype) {
1206	case '1': /* Hard link */
1207		if (archive_entry_copy_hardlink_l(entry, tar->entry_linkpath.s,
1208		    archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) {
1209			err = set_conversion_failed_error(a, tar->sconv,
1210			    "Linkname");
1211			if (err == ARCHIVE_FATAL)
1212				return (err);
1213		}
1214		/*
1215		 * The following may seem odd, but: Technically, tar
1216		 * does not store the file type for a "hard link"
1217		 * entry, only the fact that it is a hard link.  So, I
1218		 * leave the type zero normally.  But, pax interchange
1219		 * format allows hard links to have data, which
1220		 * implies that the underlying entry is a regular
1221		 * file.
1222		 */
1223		if (archive_entry_size(entry) > 0)
1224			archive_entry_set_filetype(entry, AE_IFREG);
1225
1226		/*
1227		 * A tricky point: Traditionally, tar readers have
1228		 * ignored the size field when reading hardlink
1229		 * entries, and some writers put non-zero sizes even
1230		 * though the body is empty.  POSIX blessed this
1231		 * convention in the 1988 standard, but broke with
1232		 * this tradition in 2001 by permitting hardlink
1233		 * entries to store valid bodies in pax interchange
1234		 * format, but not in ustar format.  Since there is no
1235		 * hard and fast way to distinguish pax interchange
1236		 * from earlier archives (the 'x' and 'g' entries are
1237		 * optional, after all), we need a heuristic.
1238		 */
1239		if (archive_entry_size(entry) == 0) {
1240			/* If the size is already zero, we're done. */
1241		}  else if (a->archive.archive_format
1242		    == ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE) {
1243			/* Definitely pax extended; must obey hardlink size. */
1244		} else if (a->archive.archive_format == ARCHIVE_FORMAT_TAR
1245		    || a->archive.archive_format == ARCHIVE_FORMAT_TAR_GNUTAR)
1246		{
1247			/* Old-style or GNU tar: we must ignore the size. */
1248			archive_entry_set_size(entry, 0);
1249			tar->entry_bytes_remaining = 0;
1250		} else if (archive_read_format_tar_bid(a, 50) > 50) {
1251			/*
1252			 * We don't know if it's pax: If the bid
1253			 * function sees a valid ustar header
1254			 * immediately following, then let's ignore
1255			 * the hardlink size.
1256			 */
1257			archive_entry_set_size(entry, 0);
1258			tar->entry_bytes_remaining = 0;
1259		}
1260		/*
1261		 * TODO: There are still two cases I'd like to handle:
1262		 *   = a ustar non-pax archive with a hardlink entry at
1263		 *     end-of-archive.  (Look for block of nulls following?)
1264		 *   = a pax archive that has not seen any pax headers
1265		 *     and has an entry which is a hardlink entry storing
1266		 *     a body containing an uncompressed tar archive.
1267		 * The first is worth addressing; I don't see any reliable
1268		 * way to deal with the second possibility.
1269		 */
1270		break;
1271	case '2': /* Symlink */
1272		archive_entry_set_filetype(entry, AE_IFLNK);
1273		archive_entry_set_size(entry, 0);
1274		tar->entry_bytes_remaining = 0;
1275		if (archive_entry_copy_symlink_l(entry, tar->entry_linkpath.s,
1276		    archive_strlen(&(tar->entry_linkpath)), tar->sconv) != 0) {
1277			err = set_conversion_failed_error(a, tar->sconv,
1278			    "Linkname");
1279			if (err == ARCHIVE_FATAL)
1280				return (err);
1281		}
1282		break;
1283	case '3': /* Character device */
1284		archive_entry_set_filetype(entry, AE_IFCHR);
1285		archive_entry_set_size(entry, 0);
1286		tar->entry_bytes_remaining = 0;
1287		break;
1288	case '4': /* Block device */
1289		archive_entry_set_filetype(entry, AE_IFBLK);
1290		archive_entry_set_size(entry, 0);
1291		tar->entry_bytes_remaining = 0;
1292		break;
1293	case '5': /* Dir */
1294		archive_entry_set_filetype(entry, AE_IFDIR);
1295		archive_entry_set_size(entry, 0);
1296		tar->entry_bytes_remaining = 0;
1297		break;
1298	case '6': /* FIFO device */
1299		archive_entry_set_filetype(entry, AE_IFIFO);
1300		archive_entry_set_size(entry, 0);
1301		tar->entry_bytes_remaining = 0;
1302		break;
1303	case 'D': /* GNU incremental directory type */
1304		/*
1305		 * No special handling is actually required here.
1306		 * It might be nice someday to preprocess the file list and
1307		 * provide it to the client, though.
1308		 */
1309		archive_entry_set_filetype(entry, AE_IFDIR);
1310		break;
1311	case 'M': /* GNU "Multi-volume" (remainder of file from last archive)*/
1312		/*
1313		 * As far as I can tell, this is just like a regular file
1314		 * entry, except that the contents should be _appended_ to
1315		 * the indicated file at the indicated offset.  This may
1316		 * require some API work to fully support.
1317		 */
1318		break;
1319	case 'N': /* Old GNU "long filename" entry. */
1320		/* The body of this entry is a script for renaming
1321		 * previously-extracted entries.  Ugh.  It will never
1322		 * be supported by libarchive. */
1323		archive_entry_set_filetype(entry, AE_IFREG);
1324		break;
1325	case 'S': /* GNU sparse files */
1326		/*
1327		 * Sparse files are really just regular files with
1328		 * sparse information in the extended area.
1329		 */
1330		/* FALLTHROUGH */
1331	case '0':
1332		/*
1333		 * Enable sparse file "read" support only for regular
1334		 * files and explicit GNU sparse files.  However, we
1335		 * don't allow non-standard file types to be sparse.
1336		 */
1337		tar->sparse_allowed = 1;
1338		/* FALLTHROUGH */
1339	default: /* Regular file  and non-standard types */
1340		/*
1341		 * Per POSIX: non-recognized types should always be
1342		 * treated as regular files.
1343		 */
1344		archive_entry_set_filetype(entry, AE_IFREG);
1345		break;
1346	}
1347	return (err);
1348}
1349
1350/*
1351 * Parse out header elements for "old-style" tar archives.
1352 */
1353static int
1354header_old_tar(struct archive_read *a, struct tar *tar,
1355    struct archive_entry *entry, const void *h)
1356{
1357	const struct archive_entry_header_ustar	*header;
1358	int err = ARCHIVE_OK, err2;
1359
1360	/* Copy filename over (to ensure null termination). */
1361	header = (const struct archive_entry_header_ustar *)h;
1362	if (archive_entry_copy_pathname_l(entry,
1363	    header->name, sizeof(header->name), tar->sconv) != 0) {
1364		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1365		if (err == ARCHIVE_FATAL)
1366			return (err);
1367	}
1368
1369	/* Grab rest of common fields */
1370	err2 = header_common(a, tar, entry, h);
1371	if (err > err2)
1372		err = err2;
1373
1374	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1375	return (err);
1376}
1377
1378/*
1379 * Read a Mac AppleDouble-encoded blob of file metadata,
1380 * if there is one.
1381 */
1382static int
1383read_mac_metadata_blob(struct archive_read *a, struct tar *tar,
1384    struct archive_entry *entry, const void *h, size_t *unconsumed)
1385{
1386	int64_t size;
1387	const void *data;
1388	const char *p, *name;
1389	const wchar_t *wp, *wname;
1390
1391	(void)h; /* UNUSED */
1392
1393	wname = wp = archive_entry_pathname_w(entry);
1394	if (wp != NULL) {
1395		/* Find the last path element. */
1396		for (; *wp != L'\0'; ++wp) {
1397			if (wp[0] == '/' && wp[1] != L'\0')
1398				wname = wp + 1;
1399		}
1400		/*
1401		 * If last path element starts with "._", then
1402		 * this is a Mac extension.
1403		 */
1404		if (wname[0] != L'.' || wname[1] != L'_' || wname[2] == L'\0')
1405			return ARCHIVE_OK;
1406	} else {
1407		/* Find the last path element. */
1408		name = p = archive_entry_pathname(entry);
1409		if (p == NULL)
1410			return (ARCHIVE_FAILED);
1411		for (; *p != '\0'; ++p) {
1412			if (p[0] == '/' && p[1] != '\0')
1413				name = p + 1;
1414		}
1415		/*
1416		 * If last path element starts with "._", then
1417		 * this is a Mac extension.
1418		 */
1419		if (name[0] != '.' || name[1] != '_' || name[2] == '\0')
1420			return ARCHIVE_OK;
1421	}
1422
1423 	/* Read the body as a Mac OS metadata blob. */
1424	size = archive_entry_size(entry);
1425
1426	/*
1427	 * TODO: Look beyond the body here to peek at the next header.
1428	 * If it's a regular header (not an extension header)
1429	 * that has the wrong name, just return the current
1430	 * entry as-is, without consuming the body here.
1431	 * That would reduce the risk of us mis-identifying
1432	 * an ordinary file that just happened to have
1433	 * a name starting with "._".
1434	 *
1435	 * Q: Is the above idea really possible?  Even
1436	 * when there are GNU or pax extension entries?
1437	 */
1438	data = __archive_read_ahead(a, (size_t)size, NULL);
1439	if (data == NULL) {
1440		*unconsumed = 0;
1441		return (ARCHIVE_FATAL);
1442	}
1443	archive_entry_copy_mac_metadata(entry, data, (size_t)size);
1444	*unconsumed = (size_t)((size + 511) & ~ 511);
1445	tar_flush_unconsumed(a, unconsumed);
1446	return (tar_read_header(a, tar, entry, unconsumed));
1447}
1448
1449/*
1450 * Parse a file header for a pax extended archive entry.
1451 */
1452static int
1453header_pax_global(struct archive_read *a, struct tar *tar,
1454    struct archive_entry *entry, const void *h, size_t *unconsumed)
1455{
1456	int err;
1457
1458	err = read_body_to_string(a, tar, &(tar->pax_global), h, unconsumed);
1459	if (err != ARCHIVE_OK)
1460		return (err);
1461	err = tar_read_header(a, tar, entry, unconsumed);
1462	return (err);
1463}
1464
1465static int
1466header_pax_extensions(struct archive_read *a, struct tar *tar,
1467    struct archive_entry *entry, const void *h, size_t *unconsumed)
1468{
1469	int err, err2;
1470
1471	err = read_body_to_string(a, tar, &(tar->pax_header), h, unconsumed);
1472	if (err != ARCHIVE_OK)
1473		return (err);
1474
1475	/* Parse the next header. */
1476	err = tar_read_header(a, tar, entry, unconsumed);
1477	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
1478		return (err);
1479
1480	/*
1481	 * TODO: Parse global/default options into 'entry' struct here
1482	 * before handling file-specific options.
1483	 *
1484	 * This design (parse standard header, then overwrite with pax
1485	 * extended attribute data) usually works well, but isn't ideal;
1486	 * it would be better to parse the pax extended attributes first
1487	 * and then skip any fields in the standard header that were
1488	 * defined in the pax header.
1489	 */
1490	err2 = pax_header(a, tar, entry, &tar->pax_header);
1491	err =  err_combine(err, err2);
1492	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1493	return (err);
1494}
1495
1496
1497/*
1498 * Parse a file header for a Posix "ustar" archive entry.  This also
1499 * handles "pax" or "extended ustar" entries.
1500 */
1501static int
1502header_ustar(struct archive_read *a, struct tar *tar,
1503    struct archive_entry *entry, const void *h)
1504{
1505	const struct archive_entry_header_ustar	*header;
1506	struct archive_string *as;
1507	int err = ARCHIVE_OK, r;
1508
1509	header = (const struct archive_entry_header_ustar *)h;
1510
1511	/* Copy name into an internal buffer to ensure null-termination. */
1512	as = &(tar->entry_pathname);
1513	if (header->prefix[0]) {
1514		archive_strncpy(as, header->prefix, sizeof(header->prefix));
1515		if (as->s[archive_strlen(as) - 1] != '/')
1516			archive_strappend_char(as, '/');
1517		archive_strncat(as, header->name, sizeof(header->name));
1518	} else {
1519		archive_strncpy(as, header->name, sizeof(header->name));
1520	}
1521	if (archive_entry_copy_pathname_l(entry, as->s, archive_strlen(as),
1522	    tar->sconv) != 0) {
1523		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
1524		if (err == ARCHIVE_FATAL)
1525			return (err);
1526	}
1527
1528	/* Handle rest of common fields. */
1529	r = header_common(a, tar, entry, h);
1530	if (r == ARCHIVE_FATAL)
1531		return (r);
1532	if (r < err)
1533		err = r;
1534
1535	/* Handle POSIX ustar fields. */
1536	if (archive_entry_copy_uname_l(entry,
1537	    header->uname, sizeof(header->uname), tar->sconv) != 0) {
1538		err = set_conversion_failed_error(a, tar->sconv, "Uname");
1539		if (err == ARCHIVE_FATAL)
1540			return (err);
1541	}
1542
1543	if (archive_entry_copy_gname_l(entry,
1544	    header->gname, sizeof(header->gname), tar->sconv) != 0) {
1545		err = set_conversion_failed_error(a, tar->sconv, "Gname");
1546		if (err == ARCHIVE_FATAL)
1547			return (err);
1548	}
1549
1550	/* Parse out device numbers only for char and block specials. */
1551	if (header->typeflag[0] == '3' || header->typeflag[0] == '4') {
1552		archive_entry_set_rdevmajor(entry, (dev_t)
1553		    tar_atol(header->rdevmajor, sizeof(header->rdevmajor)));
1554		archive_entry_set_rdevminor(entry, (dev_t)
1555		    tar_atol(header->rdevminor, sizeof(header->rdevminor)));
1556	}
1557
1558	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
1559
1560	return (err);
1561}
1562
1563
1564/*
1565 * Parse the pax extended attributes record.
1566 *
1567 * Returns non-zero if there's an error in the data.
1568 */
1569static int
1570pax_header(struct archive_read *a, struct tar *tar,
1571    struct archive_entry *entry, struct archive_string *in_as)
1572{
1573	size_t attr_length, l, line_length, value_length;
1574	char *p;
1575	char *key, *value;
1576	struct archive_string *as;
1577	struct archive_string_conv *sconv;
1578	int err, err2;
1579	char *attr = in_as->s;
1580
1581	attr_length = in_as->length;
1582	tar->pax_hdrcharset_binary = 0;
1583	archive_string_empty(&(tar->entry_gname));
1584	archive_string_empty(&(tar->entry_linkpath));
1585	archive_string_empty(&(tar->entry_pathname));
1586	archive_string_empty(&(tar->entry_pathname_override));
1587	archive_string_empty(&(tar->entry_uname));
1588	err = ARCHIVE_OK;
1589	while (attr_length > 0) {
1590		/* Parse decimal length field at start of line. */
1591		line_length = 0;
1592		l = attr_length;
1593		p = attr; /* Record start of line. */
1594		while (l>0) {
1595			if (*p == ' ') {
1596				p++;
1597				l--;
1598				break;
1599			}
1600			if (*p < '0' || *p > '9') {
1601				archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1602				    "Ignoring malformed pax extended attributes");
1603				return (ARCHIVE_WARN);
1604			}
1605			line_length *= 10;
1606			line_length += *p - '0';
1607			if (line_length > 999999) {
1608				archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1609				    "Rejecting pax extended attribute > 1MB");
1610				return (ARCHIVE_WARN);
1611			}
1612			p++;
1613			l--;
1614		}
1615
1616		/*
1617		 * Parsed length must be no bigger than available data,
1618		 * at least 1, and the last character of the line must
1619		 * be '\n'.
1620		 */
1621		if (line_length > attr_length
1622		    || line_length < 1
1623		    || attr[line_length - 1] != '\n')
1624		{
1625			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1626			    "Ignoring malformed pax extended attribute");
1627			return (ARCHIVE_WARN);
1628		}
1629
1630		/* Null-terminate the line. */
1631		attr[line_length - 1] = '\0';
1632
1633		/* Find end of key and null terminate it. */
1634		key = p;
1635		if (key[0] == '=')
1636			return (-1);
1637		while (*p && *p != '=')
1638			++p;
1639		if (*p == '\0') {
1640			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1641			    "Invalid pax extended attributes");
1642			return (ARCHIVE_WARN);
1643		}
1644		*p = '\0';
1645
1646		value = p + 1;
1647
1648		/* Some values may be binary data */
1649		value_length = attr + line_length - 1 - value;
1650
1651		/* Identify this attribute and set it in the entry. */
1652		err2 = pax_attribute(a, tar, entry, key, value, value_length);
1653		if (err2 == ARCHIVE_FATAL)
1654			return (err2);
1655		err = err_combine(err, err2);
1656
1657		/* Skip to next line */
1658		attr += line_length;
1659		attr_length -= line_length;
1660	}
1661
1662	/*
1663	 * PAX format uses UTF-8 as default charset for its metadata
1664	 * unless hdrcharset=BINARY is present in its header.
1665	 * We apply the charset specified by the hdrcharset option only
1666	 * when the hdrcharset attribute(in PAX header) is BINARY because
1667	 * we respect the charset described in PAX header and BINARY also
1668	 * means that metadata(filename,uname and gname) character-set
1669	 * is unknown.
1670	 */
1671	if (tar->pax_hdrcharset_binary)
1672		sconv = tar->opt_sconv;
1673	else {
1674		sconv = archive_string_conversion_from_charset(
1675		    &(a->archive), "UTF-8", 1);
1676		if (sconv == NULL)
1677			return (ARCHIVE_FATAL);
1678		if (tar->compat_2x)
1679			archive_string_conversion_set_opt(sconv,
1680			    SCONV_SET_OPT_UTF8_LIBARCHIVE2X);
1681	}
1682
1683	if (archive_strlen(&(tar->entry_gname)) > 0) {
1684		if (archive_entry_copy_gname_l(entry, tar->entry_gname.s,
1685		    archive_strlen(&(tar->entry_gname)), sconv) != 0) {
1686			err = set_conversion_failed_error(a, sconv, "Gname");
1687			if (err == ARCHIVE_FATAL)
1688				return (err);
1689			/* Use a converted an original name. */
1690			archive_entry_copy_gname(entry, tar->entry_gname.s);
1691		}
1692	}
1693	if (archive_strlen(&(tar->entry_linkpath)) > 0) {
1694		if (archive_entry_copy_link_l(entry, tar->entry_linkpath.s,
1695		    archive_strlen(&(tar->entry_linkpath)), sconv) != 0) {
1696			err = set_conversion_failed_error(a, sconv, "Linkname");
1697			if (err == ARCHIVE_FATAL)
1698				return (err);
1699			/* Use a converted an original name. */
1700			archive_entry_copy_link(entry, tar->entry_linkpath.s);
1701		}
1702	}
1703	/*
1704	 * Some extensions (such as the GNU sparse file extensions)
1705	 * deliberately store a synthetic name under the regular 'path'
1706	 * attribute and the real file name under a different attribute.
1707	 * Since we're supposed to not care about the order, we
1708	 * have no choice but to store all of the various filenames
1709	 * we find and figure it all out afterwards.  This is the
1710	 * figuring out part.
1711	 */
1712	as = NULL;
1713	if (archive_strlen(&(tar->entry_pathname_override)) > 0)
1714		as = &(tar->entry_pathname_override);
1715	else if (archive_strlen(&(tar->entry_pathname)) > 0)
1716		as = &(tar->entry_pathname);
1717	if (as != NULL) {
1718		if (archive_entry_copy_pathname_l(entry, as->s,
1719		    archive_strlen(as), sconv) != 0) {
1720			err = set_conversion_failed_error(a, sconv, "Pathname");
1721			if (err == ARCHIVE_FATAL)
1722				return (err);
1723			/* Use a converted an original name. */
1724			archive_entry_copy_pathname(entry, as->s);
1725		}
1726	}
1727	if (archive_strlen(&(tar->entry_uname)) > 0) {
1728		if (archive_entry_copy_uname_l(entry, tar->entry_uname.s,
1729		    archive_strlen(&(tar->entry_uname)), sconv) != 0) {
1730			err = set_conversion_failed_error(a, sconv, "Uname");
1731			if (err == ARCHIVE_FATAL)
1732				return (err);
1733			/* Use a converted an original name. */
1734			archive_entry_copy_uname(entry, tar->entry_uname.s);
1735		}
1736	}
1737	return (err);
1738}
1739
1740static int
1741pax_attribute_xattr(struct archive_entry *entry,
1742	const char *name, const char *value)
1743{
1744	char *name_decoded;
1745	void *value_decoded;
1746	size_t value_len;
1747
1748	if (strlen(name) < 18 || (memcmp(name, "LIBARCHIVE.xattr.", 17)) != 0)
1749		return 3;
1750
1751	name += 17;
1752
1753	/* URL-decode name */
1754	name_decoded = url_decode(name);
1755	if (name_decoded == NULL)
1756		return 2;
1757
1758	/* Base-64 decode value */
1759	value_decoded = base64_decode(value, strlen(value), &value_len);
1760	if (value_decoded == NULL) {
1761		free(name_decoded);
1762		return 1;
1763	}
1764
1765	archive_entry_xattr_add_entry(entry, name_decoded,
1766		value_decoded, value_len);
1767
1768	free(name_decoded);
1769	free(value_decoded);
1770	return 0;
1771}
1772
1773static int
1774pax_attribute_schily_xattr(struct archive_entry *entry,
1775	const char *name, const char *value, size_t value_length)
1776{
1777	if (strlen(name) < 14 || (memcmp(name, "SCHILY.xattr.", 13)) != 0)
1778		return 1;
1779
1780	name += 13;
1781
1782	archive_entry_xattr_add_entry(entry, name, value, value_length);
1783
1784	return 0;
1785}
1786
1787static int
1788pax_attribute_acl(struct archive_read *a, struct tar *tar,
1789    struct archive_entry *entry, const char *value, int type)
1790{
1791	int r;
1792	const char* errstr;
1793
1794	switch (type) {
1795	case ARCHIVE_ENTRY_ACL_TYPE_ACCESS:
1796		errstr = "SCHILY.acl.access";
1797		break;
1798	case ARCHIVE_ENTRY_ACL_TYPE_DEFAULT:
1799		errstr = "SCHILY.acl.default";
1800		break;
1801	case ARCHIVE_ENTRY_ACL_TYPE_NFS4:
1802		errstr = "SCHILY.acl.ace";
1803		break;
1804	default:
1805		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1806		    "Unknown ACL type: %d", type);
1807		return(ARCHIVE_FATAL);
1808	}
1809
1810	if (tar->sconv_acl == NULL) {
1811		tar->sconv_acl =
1812		    archive_string_conversion_from_charset(
1813			&(a->archive), "UTF-8", 1);
1814		if (tar->sconv_acl == NULL)
1815			return (ARCHIVE_FATAL);
1816	}
1817
1818	r = archive_acl_from_text_l(archive_entry_acl(entry), value, type,
1819	    tar->sconv_acl);
1820	if (r != ARCHIVE_OK) {
1821		if (r == ARCHIVE_FATAL) {
1822			archive_set_error(&a->archive, ENOMEM,
1823			    "%s %s", "Can't allocate memory for ",
1824			    errstr);
1825			return (r);
1826		}
1827		archive_set_error(&a->archive,
1828		    ARCHIVE_ERRNO_MISC, "%s %s", "Parse error: ", errstr);
1829	}
1830	return (r);
1831}
1832
1833/*
1834 * Parse a single key=value attribute.  key/value pointers are
1835 * assumed to point into reasonably long-lived storage.
1836 *
1837 * Note that POSIX reserves all-lowercase keywords.  Vendor-specific
1838 * extensions should always have keywords of the form "VENDOR.attribute"
1839 * In particular, it's quite feasible to support many different
1840 * vendor extensions here.  I'm using "LIBARCHIVE" for extensions
1841 * unique to this library.
1842 *
1843 * Investigate other vendor-specific extensions and see if
1844 * any of them look useful.
1845 */
1846static int
1847pax_attribute(struct archive_read *a, struct tar *tar,
1848    struct archive_entry *entry, const char *key, const char *value, size_t value_length)
1849{
1850	int64_t s;
1851	long n;
1852	int err = ARCHIVE_OK, r;
1853
1854#ifndef __FreeBSD__
1855	if (value == NULL)
1856		value = "";	/* Disable compiler warning; do not pass
1857				 * NULL pointer to strlen().  */
1858#endif
1859	switch (key[0]) {
1860	case 'G':
1861		/* Reject GNU.sparse.* headers on non-regular files. */
1862		if (strncmp(key, "GNU.sparse", 10) == 0 &&
1863		    !tar->sparse_allowed) {
1864			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
1865			    "Non-regular file cannot be sparse");
1866			return (ARCHIVE_FATAL);
1867		}
1868
1869		/* GNU "0.0" sparse pax format. */
1870		if (strcmp(key, "GNU.sparse.numblocks") == 0) {
1871			tar->sparse_offset = -1;
1872			tar->sparse_numbytes = -1;
1873			tar->sparse_gnu_major = 0;
1874			tar->sparse_gnu_minor = 0;
1875		}
1876		if (strcmp(key, "GNU.sparse.offset") == 0) {
1877			tar->sparse_offset = tar_atol10(value, strlen(value));
1878			if (tar->sparse_numbytes != -1) {
1879				if (gnu_add_sparse_entry(a, tar,
1880				    tar->sparse_offset, tar->sparse_numbytes)
1881				    != ARCHIVE_OK)
1882					return (ARCHIVE_FATAL);
1883				tar->sparse_offset = -1;
1884				tar->sparse_numbytes = -1;
1885			}
1886		}
1887		if (strcmp(key, "GNU.sparse.numbytes") == 0) {
1888			tar->sparse_numbytes = tar_atol10(value, strlen(value));
1889			if (tar->sparse_numbytes != -1) {
1890				if (gnu_add_sparse_entry(a, tar,
1891				    tar->sparse_offset, tar->sparse_numbytes)
1892				    != ARCHIVE_OK)
1893					return (ARCHIVE_FATAL);
1894				tar->sparse_offset = -1;
1895				tar->sparse_numbytes = -1;
1896			}
1897		}
1898		if (strcmp(key, "GNU.sparse.size") == 0) {
1899			tar->realsize = tar_atol10(value, strlen(value));
1900			archive_entry_set_size(entry, tar->realsize);
1901			tar->realsize_override = 1;
1902		}
1903
1904		/* GNU "0.1" sparse pax format. */
1905		if (strcmp(key, "GNU.sparse.map") == 0) {
1906			tar->sparse_gnu_major = 0;
1907			tar->sparse_gnu_minor = 1;
1908			if (gnu_sparse_01_parse(a, tar, value) != ARCHIVE_OK)
1909				return (ARCHIVE_WARN);
1910		}
1911
1912		/* GNU "1.0" sparse pax format */
1913		if (strcmp(key, "GNU.sparse.major") == 0) {
1914			tar->sparse_gnu_major = (int)tar_atol10(value, strlen(value));
1915			tar->sparse_gnu_pending = 1;
1916		}
1917		if (strcmp(key, "GNU.sparse.minor") == 0) {
1918			tar->sparse_gnu_minor = (int)tar_atol10(value, strlen(value));
1919			tar->sparse_gnu_pending = 1;
1920		}
1921		if (strcmp(key, "GNU.sparse.name") == 0) {
1922			/*
1923			 * The real filename; when storing sparse
1924			 * files, GNU tar puts a synthesized name into
1925			 * the regular 'path' attribute in an attempt
1926			 * to limit confusion. ;-)
1927			 */
1928			archive_strcpy(&(tar->entry_pathname_override), value);
1929		}
1930		if (strcmp(key, "GNU.sparse.realsize") == 0) {
1931			tar->realsize = tar_atol10(value, strlen(value));
1932			archive_entry_set_size(entry, tar->realsize);
1933			tar->realsize_override = 1;
1934		}
1935		break;
1936	case 'L':
1937		/* Our extensions */
1938/* TODO: Handle arbitrary extended attributes... */
1939/*
1940		if (strcmp(key, "LIBARCHIVE.xxxxxxx") == 0)
1941			archive_entry_set_xxxxxx(entry, value);
1942*/
1943		if (strcmp(key, "LIBARCHIVE.creationtime") == 0) {
1944			pax_time(value, &s, &n);
1945			archive_entry_set_birthtime(entry, s, n);
1946		}
1947		if (memcmp(key, "LIBARCHIVE.xattr.", 17) == 0)
1948			pax_attribute_xattr(entry, key, value);
1949		break;
1950	case 'S':
1951		/* We support some keys used by the "star" archiver */
1952		if (strcmp(key, "SCHILY.acl.access") == 0) {
1953			r = pax_attribute_acl(a, tar, entry, value,
1954			    ARCHIVE_ENTRY_ACL_TYPE_ACCESS);
1955			if (r == ARCHIVE_FATAL)
1956				return (r);
1957		} else if (strcmp(key, "SCHILY.acl.default") == 0) {
1958			r = pax_attribute_acl(a, tar, entry, value,
1959			    ARCHIVE_ENTRY_ACL_TYPE_DEFAULT);
1960			if (r == ARCHIVE_FATAL)
1961				return (r);
1962		} else if (strcmp(key, "SCHILY.acl.ace") == 0) {
1963			r = pax_attribute_acl(a, tar, entry, value,
1964			    ARCHIVE_ENTRY_ACL_TYPE_NFS4);
1965			if (r == ARCHIVE_FATAL)
1966				return (r);
1967		} else if (strcmp(key, "SCHILY.devmajor") == 0) {
1968			archive_entry_set_rdevmajor(entry,
1969			    (dev_t)tar_atol10(value, strlen(value)));
1970		} else if (strcmp(key, "SCHILY.devminor") == 0) {
1971			archive_entry_set_rdevminor(entry,
1972			    (dev_t)tar_atol10(value, strlen(value)));
1973		} else if (strcmp(key, "SCHILY.fflags") == 0) {
1974			archive_entry_copy_fflags_text(entry, value);
1975		} else if (strcmp(key, "SCHILY.dev") == 0) {
1976			archive_entry_set_dev(entry,
1977			    (dev_t)tar_atol10(value, strlen(value)));
1978		} else if (strcmp(key, "SCHILY.ino") == 0) {
1979			archive_entry_set_ino(entry,
1980			    tar_atol10(value, strlen(value)));
1981		} else if (strcmp(key, "SCHILY.nlink") == 0) {
1982			archive_entry_set_nlink(entry, (unsigned)
1983			    tar_atol10(value, strlen(value)));
1984		} else if (strcmp(key, "SCHILY.realsize") == 0) {
1985			tar->realsize = tar_atol10(value, strlen(value));
1986			tar->realsize_override = 1;
1987			archive_entry_set_size(entry, tar->realsize);
1988		} else if (strncmp(key, "SCHILY.xattr.", 13) == 0) {
1989			pax_attribute_schily_xattr(entry, key, value,
1990			    value_length);
1991		} else if (strcmp(key, "SUN.holesdata") == 0) {
1992			/* A Solaris extension for sparse. */
1993			r = solaris_sparse_parse(a, tar, entry, value);
1994			if (r < err) {
1995				if (r == ARCHIVE_FATAL)
1996					return (r);
1997				err = r;
1998				archive_set_error(&a->archive,
1999				    ARCHIVE_ERRNO_MISC,
2000				    "Parse error: SUN.holesdata");
2001			}
2002		}
2003		break;
2004	case 'a':
2005		if (strcmp(key, "atime") == 0) {
2006			pax_time(value, &s, &n);
2007			archive_entry_set_atime(entry, s, n);
2008		}
2009		break;
2010	case 'c':
2011		if (strcmp(key, "ctime") == 0) {
2012			pax_time(value, &s, &n);
2013			archive_entry_set_ctime(entry, s, n);
2014		} else if (strcmp(key, "charset") == 0) {
2015			/* TODO: Publish charset information in entry. */
2016		} else if (strcmp(key, "comment") == 0) {
2017			/* TODO: Publish comment in entry. */
2018		}
2019		break;
2020	case 'g':
2021		if (strcmp(key, "gid") == 0) {
2022			archive_entry_set_gid(entry,
2023			    tar_atol10(value, strlen(value)));
2024		} else if (strcmp(key, "gname") == 0) {
2025			archive_strcpy(&(tar->entry_gname), value);
2026		}
2027		break;
2028	case 'h':
2029		if (strcmp(key, "hdrcharset") == 0) {
2030			if (strcmp(value, "BINARY") == 0)
2031				/* Binary  mode. */
2032				tar->pax_hdrcharset_binary = 1;
2033			else if (strcmp(value, "ISO-IR 10646 2000 UTF-8") == 0)
2034				tar->pax_hdrcharset_binary = 0;
2035		}
2036		break;
2037	case 'l':
2038		/* pax interchange doesn't distinguish hardlink vs. symlink. */
2039		if (strcmp(key, "linkpath") == 0) {
2040			archive_strcpy(&(tar->entry_linkpath), value);
2041		}
2042		break;
2043	case 'm':
2044		if (strcmp(key, "mtime") == 0) {
2045			pax_time(value, &s, &n);
2046			archive_entry_set_mtime(entry, s, n);
2047		}
2048		break;
2049	case 'p':
2050		if (strcmp(key, "path") == 0) {
2051			archive_strcpy(&(tar->entry_pathname), value);
2052		}
2053		break;
2054	case 'r':
2055		/* POSIX has reserved 'realtime.*' */
2056		break;
2057	case 's':
2058		/* POSIX has reserved 'security.*' */
2059		/* Someday: if (strcmp(key, "security.acl") == 0) { ... } */
2060		if (strcmp(key, "size") == 0) {
2061			/* "size" is the size of the data in the entry. */
2062			tar->entry_bytes_remaining
2063			    = tar_atol10(value, strlen(value));
2064			/*
2065			 * The "size" pax header keyword always overrides the
2066			 * "size" field in the tar header.
2067			 * GNU.sparse.realsize, GNU.sparse.size and
2068			 * SCHILY.realsize override this value.
2069			 */
2070			if (!tar->realsize_override) {
2071				archive_entry_set_size(entry,
2072				    tar->entry_bytes_remaining);
2073				tar->realsize
2074				    = tar->entry_bytes_remaining;
2075			}
2076		}
2077		break;
2078	case 'u':
2079		if (strcmp(key, "uid") == 0) {
2080			archive_entry_set_uid(entry,
2081			    tar_atol10(value, strlen(value)));
2082		} else if (strcmp(key, "uname") == 0) {
2083			archive_strcpy(&(tar->entry_uname), value);
2084		}
2085		break;
2086	}
2087	return (err);
2088}
2089
2090
2091
2092/*
2093 * parse a decimal time value, which may include a fractional portion
2094 */
2095static void
2096pax_time(const char *p, int64_t *ps, long *pn)
2097{
2098	char digit;
2099	int64_t	s;
2100	unsigned long l;
2101	int sign;
2102	int64_t limit, last_digit_limit;
2103
2104	limit = INT64_MAX / 10;
2105	last_digit_limit = INT64_MAX % 10;
2106
2107	s = 0;
2108	sign = 1;
2109	if (*p == '-') {
2110		sign = -1;
2111		p++;
2112	}
2113	while (*p >= '0' && *p <= '9') {
2114		digit = *p - '0';
2115		if (s > limit ||
2116		    (s == limit && digit > last_digit_limit)) {
2117			s = INT64_MAX;
2118			break;
2119		}
2120		s = (s * 10) + digit;
2121		++p;
2122	}
2123
2124	*ps = s * sign;
2125
2126	/* Calculate nanoseconds. */
2127	*pn = 0;
2128
2129	if (*p != '.')
2130		return;
2131
2132	l = 100000000UL;
2133	do {
2134		++p;
2135		if (*p >= '0' && *p <= '9')
2136			*pn += (*p - '0') * l;
2137		else
2138			break;
2139	} while (l /= 10);
2140}
2141
2142/*
2143 * Parse GNU tar header
2144 */
2145static int
2146header_gnutar(struct archive_read *a, struct tar *tar,
2147    struct archive_entry *entry, const void *h, size_t *unconsumed)
2148{
2149	const struct archive_entry_header_gnutar *header;
2150	int64_t t;
2151	int err = ARCHIVE_OK;
2152
2153	/*
2154	 * GNU header is like POSIX ustar, except 'prefix' is
2155	 * replaced with some other fields. This also means the
2156	 * filename is stored as in old-style archives.
2157	 */
2158
2159	/* Grab fields common to all tar variants. */
2160	err = header_common(a, tar, entry, h);
2161	if (err == ARCHIVE_FATAL)
2162		return (err);
2163
2164	/* Copy filename over (to ensure null termination). */
2165	header = (const struct archive_entry_header_gnutar *)h;
2166	if (archive_entry_copy_pathname_l(entry,
2167	    header->name, sizeof(header->name), tar->sconv) != 0) {
2168		err = set_conversion_failed_error(a, tar->sconv, "Pathname");
2169		if (err == ARCHIVE_FATAL)
2170			return (err);
2171	}
2172
2173	/* Fields common to ustar and GNU */
2174	/* XXX Can the following be factored out since it's common
2175	 * to ustar and gnu tar?  Is it okay to move it down into
2176	 * header_common, perhaps?  */
2177	if (archive_entry_copy_uname_l(entry,
2178	    header->uname, sizeof(header->uname), tar->sconv) != 0) {
2179		err = set_conversion_failed_error(a, tar->sconv, "Uname");
2180		if (err == ARCHIVE_FATAL)
2181			return (err);
2182	}
2183
2184	if (archive_entry_copy_gname_l(entry,
2185	    header->gname, sizeof(header->gname), tar->sconv) != 0) {
2186		err = set_conversion_failed_error(a, tar->sconv, "Gname");
2187		if (err == ARCHIVE_FATAL)
2188			return (err);
2189	}
2190
2191	/* Parse out device numbers only for char and block specials */
2192	if (header->typeflag[0] == '3' || header->typeflag[0] == '4') {
2193		archive_entry_set_rdevmajor(entry, (dev_t)
2194		    tar_atol(header->rdevmajor, sizeof(header->rdevmajor)));
2195		archive_entry_set_rdevminor(entry, (dev_t)
2196		    tar_atol(header->rdevminor, sizeof(header->rdevminor)));
2197	} else
2198		archive_entry_set_rdev(entry, 0);
2199
2200	tar->entry_padding = 0x1ff & (-tar->entry_bytes_remaining);
2201
2202	/* Grab GNU-specific fields. */
2203	t = tar_atol(header->atime, sizeof(header->atime));
2204	if (t > 0)
2205		archive_entry_set_atime(entry, t, 0);
2206	t = tar_atol(header->ctime, sizeof(header->ctime));
2207	if (t > 0)
2208		archive_entry_set_ctime(entry, t, 0);
2209
2210	if (header->realsize[0] != 0) {
2211		tar->realsize
2212		    = tar_atol(header->realsize, sizeof(header->realsize));
2213		archive_entry_set_size(entry, tar->realsize);
2214		tar->realsize_override = 1;
2215	}
2216
2217	if (header->sparse[0].offset[0] != 0) {
2218		if (gnu_sparse_old_read(a, tar, header, unconsumed)
2219		    != ARCHIVE_OK)
2220			return (ARCHIVE_FATAL);
2221	} else {
2222		if (header->isextended[0] != 0) {
2223			/* XXX WTF? XXX */
2224		}
2225	}
2226
2227	return (err);
2228}
2229
2230static int
2231gnu_add_sparse_entry(struct archive_read *a, struct tar *tar,
2232    int64_t offset, int64_t remaining)
2233{
2234	struct sparse_block *p;
2235
2236	p = (struct sparse_block *)calloc(1, sizeof(*p));
2237	if (p == NULL) {
2238		archive_set_error(&a->archive, ENOMEM, "Out of memory");
2239		return (ARCHIVE_FATAL);
2240	}
2241	if (tar->sparse_last != NULL)
2242		tar->sparse_last->next = p;
2243	else
2244		tar->sparse_list = p;
2245	tar->sparse_last = p;
2246	if (remaining < 0 || offset < 0 || offset > INT64_MAX - remaining) {
2247		archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC, "Malformed sparse map data");
2248		return (ARCHIVE_FATAL);
2249	}
2250	p->offset = offset;
2251	p->remaining = remaining;
2252	return (ARCHIVE_OK);
2253}
2254
2255static void
2256gnu_clear_sparse_list(struct tar *tar)
2257{
2258	struct sparse_block *p;
2259
2260	while (tar->sparse_list != NULL) {
2261		p = tar->sparse_list;
2262		tar->sparse_list = p->next;
2263		free(p);
2264	}
2265	tar->sparse_last = NULL;
2266}
2267
2268/*
2269 * GNU tar old-format sparse data.
2270 *
2271 * GNU old-format sparse data is stored in a fixed-field
2272 * format.  Offset/size values are 11-byte octal fields (same
2273 * format as 'size' field in ustart header).  These are
2274 * stored in the header, allocating subsequent header blocks
2275 * as needed.  Extending the header in this way is a pretty
2276 * severe POSIX violation; this design has earned GNU tar a
2277 * lot of criticism.
2278 */
2279
2280static int
2281gnu_sparse_old_read(struct archive_read *a, struct tar *tar,
2282    const struct archive_entry_header_gnutar *header, size_t *unconsumed)
2283{
2284	ssize_t bytes_read;
2285	const void *data;
2286	struct extended {
2287		struct gnu_sparse sparse[21];
2288		char	isextended[1];
2289		char	padding[7];
2290	};
2291	const struct extended *ext;
2292
2293	if (gnu_sparse_old_parse(a, tar, header->sparse, 4) != ARCHIVE_OK)
2294		return (ARCHIVE_FATAL);
2295	if (header->isextended[0] == 0)
2296		return (ARCHIVE_OK);
2297
2298	do {
2299		tar_flush_unconsumed(a, unconsumed);
2300		data = __archive_read_ahead(a, 512, &bytes_read);
2301		if (bytes_read < 0)
2302			return (ARCHIVE_FATAL);
2303		if (bytes_read < 512) {
2304			archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2305			    "Truncated tar archive "
2306			    "detected while reading sparse file data");
2307			return (ARCHIVE_FATAL);
2308		}
2309		*unconsumed = 512;
2310		ext = (const struct extended *)data;
2311		if (gnu_sparse_old_parse(a, tar, ext->sparse, 21) != ARCHIVE_OK)
2312			return (ARCHIVE_FATAL);
2313	} while (ext->isextended[0] != 0);
2314	if (tar->sparse_list != NULL)
2315		tar->entry_offset = tar->sparse_list->offset;
2316	return (ARCHIVE_OK);
2317}
2318
2319static int
2320gnu_sparse_old_parse(struct archive_read *a, struct tar *tar,
2321    const struct gnu_sparse *sparse, int length)
2322{
2323	while (length > 0 && sparse->offset[0] != 0) {
2324		if (gnu_add_sparse_entry(a, tar,
2325		    tar_atol(sparse->offset, sizeof(sparse->offset)),
2326		    tar_atol(sparse->numbytes, sizeof(sparse->numbytes)))
2327		    != ARCHIVE_OK)
2328			return (ARCHIVE_FATAL);
2329		sparse++;
2330		length--;
2331	}
2332	return (ARCHIVE_OK);
2333}
2334
2335/*
2336 * GNU tar sparse format 0.0
2337 *
2338 * Beginning with GNU tar 1.15, sparse files are stored using
2339 * information in the pax extended header.  The GNU tar maintainers
2340 * have gone through a number of variations in the process of working
2341 * out this scheme; fortunately, they're all numbered.
2342 *
2343 * Sparse format 0.0 uses attribute GNU.sparse.numblocks to store the
2344 * number of blocks, and GNU.sparse.offset/GNU.sparse.numbytes to
2345 * store offset/size for each block.  The repeated instances of these
2346 * latter fields violate the pax specification (which frowns on
2347 * duplicate keys), so this format was quickly replaced.
2348 */
2349
2350/*
2351 * GNU tar sparse format 0.1
2352 *
2353 * This version replaced the offset/numbytes attributes with
2354 * a single "map" attribute that stored a list of integers.  This
2355 * format had two problems: First, the "map" attribute could be very
2356 * long, which caused problems for some implementations.  More
2357 * importantly, the sparse data was lost when extracted by archivers
2358 * that didn't recognize this extension.
2359 */
2360
2361static int
2362gnu_sparse_01_parse(struct archive_read *a, struct tar *tar, const char *p)
2363{
2364	const char *e;
2365	int64_t offset = -1, size = -1;
2366
2367	for (;;) {
2368		e = p;
2369		while (*e != '\0' && *e != ',') {
2370			if (*e < '0' || *e > '9')
2371				return (ARCHIVE_WARN);
2372			e++;
2373		}
2374		if (offset < 0) {
2375			offset = tar_atol10(p, e - p);
2376			if (offset < 0)
2377				return (ARCHIVE_WARN);
2378		} else {
2379			size = tar_atol10(p, e - p);
2380			if (size < 0)
2381				return (ARCHIVE_WARN);
2382			if (gnu_add_sparse_entry(a, tar, offset, size)
2383			    != ARCHIVE_OK)
2384				return (ARCHIVE_FATAL);
2385			offset = -1;
2386		}
2387		if (*e == '\0')
2388			return (ARCHIVE_OK);
2389		p = e + 1;
2390	}
2391}
2392
2393/*
2394 * GNU tar sparse format 1.0
2395 *
2396 * The idea: The offset/size data is stored as a series of base-10
2397 * ASCII numbers prepended to the file data, so that dearchivers that
2398 * don't support this format will extract the block map along with the
2399 * data and a separate post-process can restore the sparseness.
2400 *
2401 * Unfortunately, GNU tar 1.16 had a bug that added unnecessary
2402 * padding to the body of the file when using this format.  GNU tar
2403 * 1.17 corrected this bug without bumping the version number, so
2404 * it's not possible to support both variants.  This code supports
2405 * the later variant at the expense of not supporting the former.
2406 *
2407 * This variant also replaced GNU.sparse.size with GNU.sparse.realsize
2408 * and introduced the GNU.sparse.major/GNU.sparse.minor attributes.
2409 */
2410
2411/*
2412 * Read the next line from the input, and parse it as a decimal
2413 * integer followed by '\n'.  Returns positive integer value or
2414 * negative on error.
2415 */
2416static int64_t
2417gnu_sparse_10_atol(struct archive_read *a, struct tar *tar,
2418    int64_t *remaining, size_t *unconsumed)
2419{
2420	int64_t l, limit, last_digit_limit;
2421	const char *p;
2422	ssize_t bytes_read;
2423	int base, digit;
2424
2425	base = 10;
2426	limit = INT64_MAX / base;
2427	last_digit_limit = INT64_MAX % base;
2428
2429	/*
2430	 * Skip any lines starting with '#'; GNU tar specs
2431	 * don't require this, but they should.
2432	 */
2433	do {
2434		bytes_read = readline(a, tar, &p,
2435			(ssize_t)tar_min(*remaining, 100), unconsumed);
2436		if (bytes_read <= 0)
2437			return (ARCHIVE_FATAL);
2438		*remaining -= bytes_read;
2439	} while (p[0] == '#');
2440
2441	l = 0;
2442	while (bytes_read > 0) {
2443		if (*p == '\n')
2444			return (l);
2445		if (*p < '0' || *p >= '0' + base)
2446			return (ARCHIVE_WARN);
2447		digit = *p - '0';
2448		if (l > limit || (l == limit && digit > last_digit_limit))
2449			l = INT64_MAX; /* Truncate on overflow. */
2450		else
2451			l = (l * base) + digit;
2452		p++;
2453		bytes_read--;
2454	}
2455	/* TODO: Error message. */
2456	return (ARCHIVE_WARN);
2457}
2458
2459/*
2460 * Returns length (in bytes) of the sparse data description
2461 * that was read.
2462 */
2463static ssize_t
2464gnu_sparse_10_read(struct archive_read *a, struct tar *tar, size_t *unconsumed)
2465{
2466	ssize_t bytes_read;
2467	int entries;
2468	int64_t offset, size, to_skip, remaining;
2469
2470	/* Clear out the existing sparse list. */
2471	gnu_clear_sparse_list(tar);
2472
2473	remaining = tar->entry_bytes_remaining;
2474
2475	/* Parse entries. */
2476	entries = (int)gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2477	if (entries < 0)
2478		return (ARCHIVE_FATAL);
2479	/* Parse the individual entries. */
2480	while (entries-- > 0) {
2481		/* Parse offset/size */
2482		offset = gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2483		if (offset < 0)
2484			return (ARCHIVE_FATAL);
2485		size = gnu_sparse_10_atol(a, tar, &remaining, unconsumed);
2486		if (size < 0)
2487			return (ARCHIVE_FATAL);
2488		/* Add a new sparse entry. */
2489		if (gnu_add_sparse_entry(a, tar, offset, size) != ARCHIVE_OK)
2490			return (ARCHIVE_FATAL);
2491	}
2492	/* Skip rest of block... */
2493	tar_flush_unconsumed(a, unconsumed);
2494	bytes_read = (ssize_t)(tar->entry_bytes_remaining - remaining);
2495	to_skip = 0x1ff & -bytes_read;
2496	/* Fail if tar->entry_bytes_remaing would get negative */
2497	if (to_skip > remaining)
2498		return (ARCHIVE_FATAL);
2499	if (to_skip != __archive_read_consume(a, to_skip))
2500		return (ARCHIVE_FATAL);
2501	return ((ssize_t)(bytes_read + to_skip));
2502}
2503
2504/*
2505 * Solaris pax extension for a sparse file. This is recorded with the
2506 * data and hole pairs. The way recording sparse information by Solaris'
2507 * pax simply indicates where data and sparse are, so the stored contents
2508 * consist of both data and hole.
2509 */
2510static int
2511solaris_sparse_parse(struct archive_read *a, struct tar *tar,
2512    struct archive_entry *entry, const char *p)
2513{
2514	const char *e;
2515	int64_t start, end;
2516	int hole = 1;
2517
2518	(void)entry; /* UNUSED */
2519
2520	end = 0;
2521	if (*p == ' ')
2522		p++;
2523	else
2524		return (ARCHIVE_WARN);
2525	for (;;) {
2526		e = p;
2527		while (*e != '\0' && *e != ' ') {
2528			if (*e < '0' || *e > '9')
2529				return (ARCHIVE_WARN);
2530			e++;
2531		}
2532		start = end;
2533		end = tar_atol10(p, e - p);
2534		if (end < 0)
2535			return (ARCHIVE_WARN);
2536		if (start < end) {
2537			if (gnu_add_sparse_entry(a, tar, start,
2538			    end - start) != ARCHIVE_OK)
2539				return (ARCHIVE_FATAL);
2540			tar->sparse_last->hole = hole;
2541		}
2542		if (*e == '\0')
2543			return (ARCHIVE_OK);
2544		p = e + 1;
2545		hole = hole == 0;
2546	}
2547}
2548
2549/*-
2550 * Convert text->integer.
2551 *
2552 * Traditional tar formats (including POSIX) specify base-8 for
2553 * all of the standard numeric fields.  This is a significant limitation
2554 * in practice:
2555 *   = file size is limited to 8GB
2556 *   = rdevmajor and rdevminor are limited to 21 bits
2557 *   = uid/gid are limited to 21 bits
2558 *
2559 * There are two workarounds for this:
2560 *   = pax extended headers, which use variable-length string fields
2561 *   = GNU tar and STAR both allow either base-8 or base-256 in
2562 *      most fields.  The high bit is set to indicate base-256.
2563 *
2564 * On read, this implementation supports both extensions.
2565 */
2566static int64_t
2567tar_atol(const char *p, size_t char_cnt)
2568{
2569	/*
2570	 * Technically, GNU tar considers a field to be in base-256
2571	 * only if the first byte is 0xff or 0x80.
2572	 */
2573	if (*p & 0x80)
2574		return (tar_atol256(p, char_cnt));
2575	return (tar_atol8(p, char_cnt));
2576}
2577
2578/*
2579 * Note that this implementation does not (and should not!) obey
2580 * locale settings; you cannot simply substitute strtol here, since
2581 * it does obey locale.
2582 */
2583static int64_t
2584tar_atol_base_n(const char *p, size_t char_cnt, int base)
2585{
2586	int64_t	l, maxval, limit, last_digit_limit;
2587	int digit, sign;
2588
2589	maxval = INT64_MAX;
2590	limit = INT64_MAX / base;
2591	last_digit_limit = INT64_MAX % base;
2592
2593	/* the pointer will not be dereferenced if char_cnt is zero
2594	 * due to the way the && operator is evaluated.
2595	 */
2596	while (char_cnt != 0 && (*p == ' ' || *p == '\t')) {
2597		p++;
2598		char_cnt--;
2599	}
2600
2601	sign = 1;
2602	if (char_cnt != 0 && *p == '-') {
2603		sign = -1;
2604		p++;
2605		char_cnt--;
2606
2607		maxval = INT64_MIN;
2608		limit = -(INT64_MIN / base);
2609		last_digit_limit = INT64_MIN % base;
2610	}
2611
2612	l = 0;
2613	if (char_cnt != 0) {
2614		digit = *p - '0';
2615		while (digit >= 0 && digit < base  && char_cnt != 0) {
2616			if (l>limit || (l == limit && digit > last_digit_limit)) {
2617				return maxval; /* Truncate on overflow. */
2618			}
2619			l = (l * base) + digit;
2620			digit = *++p - '0';
2621			char_cnt--;
2622		}
2623	}
2624	return (sign < 0) ? -l : l;
2625}
2626
2627static int64_t
2628tar_atol8(const char *p, size_t char_cnt)
2629{
2630	return tar_atol_base_n(p, char_cnt, 8);
2631}
2632
2633static int64_t
2634tar_atol10(const char *p, size_t char_cnt)
2635{
2636	return tar_atol_base_n(p, char_cnt, 10);
2637}
2638
2639/*
2640 * Parse a base-256 integer.  This is just a variable-length
2641 * twos-complement signed binary value in big-endian order, except
2642 * that the high-order bit is ignored.  The values here can be up to
2643 * 12 bytes, so we need to be careful about overflowing 64-bit
2644 * (8-byte) integers.
2645 *
2646 * This code unashamedly assumes that the local machine uses 8-bit
2647 * bytes and twos-complement arithmetic.
2648 */
2649static int64_t
2650tar_atol256(const char *_p, size_t char_cnt)
2651{
2652	uint64_t l;
2653	const unsigned char *p = (const unsigned char *)_p;
2654	unsigned char c, neg;
2655
2656	/* Extend 7-bit 2s-comp to 8-bit 2s-comp, decide sign. */
2657	c = *p;
2658	if (c & 0x40) {
2659		neg = 0xff;
2660		c |= 0x80;
2661		l = ~ARCHIVE_LITERAL_ULL(0);
2662	} else {
2663		neg = 0;
2664		c &= 0x7f;
2665		l = 0;
2666	}
2667
2668	/* If more than 8 bytes, check that we can ignore
2669	 * high-order bits without overflow. */
2670	while (char_cnt > sizeof(int64_t)) {
2671		--char_cnt;
2672		if (c != neg)
2673			return neg ? INT64_MIN : INT64_MAX;
2674		c = *++p;
2675	}
2676
2677	/* c is first byte that fits; if sign mismatch, return overflow */
2678	if ((c ^ neg) & 0x80) {
2679		return neg ? INT64_MIN : INT64_MAX;
2680	}
2681
2682	/* Accumulate remaining bytes. */
2683	while (--char_cnt > 0) {
2684		l = (l << 8) | c;
2685		c = *++p;
2686	}
2687	l = (l << 8) | c;
2688	/* Return signed twos-complement value. */
2689	return (int64_t)(l);
2690}
2691
2692/*
2693 * Returns length of line (including trailing newline)
2694 * or negative on error.  'start' argument is updated to
2695 * point to first character of line.  This avoids copying
2696 * when possible.
2697 */
2698static ssize_t
2699readline(struct archive_read *a, struct tar *tar, const char **start,
2700    ssize_t limit, size_t *unconsumed)
2701{
2702	ssize_t bytes_read;
2703	ssize_t total_size = 0;
2704	const void *t;
2705	const char *s;
2706	void *p;
2707
2708	tar_flush_unconsumed(a, unconsumed);
2709
2710	t = __archive_read_ahead(a, 1, &bytes_read);
2711	if (bytes_read <= 0)
2712		return (ARCHIVE_FATAL);
2713	s = t;  /* Start of line? */
2714	p = memchr(t, '\n', bytes_read);
2715	/* If we found '\n' in the read buffer, return pointer to that. */
2716	if (p != NULL) {
2717		bytes_read = 1 + ((const char *)p) - s;
2718		if (bytes_read > limit) {
2719			archive_set_error(&a->archive,
2720			    ARCHIVE_ERRNO_FILE_FORMAT,
2721			    "Line too long");
2722			return (ARCHIVE_FATAL);
2723		}
2724		*unconsumed = bytes_read;
2725		*start = s;
2726		return (bytes_read);
2727	}
2728	*unconsumed = bytes_read;
2729	/* Otherwise, we need to accumulate in a line buffer. */
2730	for (;;) {
2731		if (total_size + bytes_read > limit) {
2732			archive_set_error(&a->archive,
2733			    ARCHIVE_ERRNO_FILE_FORMAT,
2734			    "Line too long");
2735			return (ARCHIVE_FATAL);
2736		}
2737		if (archive_string_ensure(&tar->line, total_size + bytes_read) == NULL) {
2738			archive_set_error(&a->archive, ENOMEM,
2739			    "Can't allocate working buffer");
2740			return (ARCHIVE_FATAL);
2741		}
2742		memcpy(tar->line.s + total_size, t, bytes_read);
2743		tar_flush_unconsumed(a, unconsumed);
2744		total_size += bytes_read;
2745		/* If we found '\n', clean up and return. */
2746		if (p != NULL) {
2747			*start = tar->line.s;
2748			return (total_size);
2749		}
2750		/* Read some more. */
2751		t = __archive_read_ahead(a, 1, &bytes_read);
2752		if (bytes_read <= 0)
2753			return (ARCHIVE_FATAL);
2754		s = t;  /* Start of line? */
2755		p = memchr(t, '\n', bytes_read);
2756		/* If we found '\n', trim the read. */
2757		if (p != NULL) {
2758			bytes_read = 1 + ((const char *)p) - s;
2759		}
2760		*unconsumed = bytes_read;
2761	}
2762}
2763
2764/*
2765 * base64_decode - Base64 decode
2766 *
2767 * This accepts most variations of base-64 encoding, including:
2768 *    * with or without line breaks
2769 *    * with or without the final group padded with '=' or '_' characters
2770 * (The most economical Base-64 variant does not pad the last group and
2771 * omits line breaks; RFC1341 used for MIME requires both.)
2772 */
2773static char *
2774base64_decode(const char *s, size_t len, size_t *out_len)
2775{
2776	static const unsigned char digits[64] = {
2777		'A','B','C','D','E','F','G','H','I','J','K','L','M','N',
2778		'O','P','Q','R','S','T','U','V','W','X','Y','Z','a','b',
2779		'c','d','e','f','g','h','i','j','k','l','m','n','o','p',
2780		'q','r','s','t','u','v','w','x','y','z','0','1','2','3',
2781		'4','5','6','7','8','9','+','/' };
2782	static unsigned char decode_table[128];
2783	char *out, *d;
2784	const unsigned char *src = (const unsigned char *)s;
2785
2786	/* If the decode table is not yet initialized, prepare it. */
2787	if (decode_table[digits[1]] != 1) {
2788		unsigned i;
2789		memset(decode_table, 0xff, sizeof(decode_table));
2790		for (i = 0; i < sizeof(digits); i++)
2791			decode_table[digits[i]] = i;
2792	}
2793
2794	/* Allocate enough space to hold the entire output. */
2795	/* Note that we may not use all of this... */
2796	out = (char *)malloc(len - len / 4 + 1);
2797	if (out == NULL) {
2798		*out_len = 0;
2799		return (NULL);
2800	}
2801	d = out;
2802
2803	while (len > 0) {
2804		/* Collect the next group of (up to) four characters. */
2805		int v = 0;
2806		int group_size = 0;
2807		while (group_size < 4 && len > 0) {
2808			/* '=' or '_' padding indicates final group. */
2809			if (*src == '=' || *src == '_') {
2810				len = 0;
2811				break;
2812			}
2813			/* Skip illegal characters (including line breaks) */
2814			if (*src > 127 || *src < 32
2815			    || decode_table[*src] == 0xff) {
2816				len--;
2817				src++;
2818				continue;
2819			}
2820			v <<= 6;
2821			v |= decode_table[*src++];
2822			len --;
2823			group_size++;
2824		}
2825		/* Align a short group properly. */
2826		v <<= 6 * (4 - group_size);
2827		/* Unpack the group we just collected. */
2828		switch (group_size) {
2829		case 4: d[2] = v & 0xff;
2830			/* FALLTHROUGH */
2831		case 3: d[1] = (v >> 8) & 0xff;
2832			/* FALLTHROUGH */
2833		case 2: d[0] = (v >> 16) & 0xff;
2834			break;
2835		case 1: /* this is invalid! */
2836			break;
2837		}
2838		d += group_size * 3 / 4;
2839	}
2840
2841	*out_len = d - out;
2842	return (out);
2843}
2844
2845static char *
2846url_decode(const char *in)
2847{
2848	char *out, *d;
2849	const char *s;
2850
2851	out = (char *)malloc(strlen(in) + 1);
2852	if (out == NULL)
2853		return (NULL);
2854	for (s = in, d = out; *s != '\0'; ) {
2855		if (s[0] == '%' && s[1] != '\0' && s[2] != '\0') {
2856			/* Try to convert % escape */
2857			int digit1 = tohex(s[1]);
2858			int digit2 = tohex(s[2]);
2859			if (digit1 >= 0 && digit2 >= 0) {
2860				/* Looks good, consume three chars */
2861				s += 3;
2862				/* Convert output */
2863				*d++ = ((digit1 << 4) | digit2);
2864				continue;
2865			}
2866			/* Else fall through and treat '%' as normal char */
2867		}
2868		*d++ = *s++;
2869	}
2870	*d = '\0';
2871	return (out);
2872}
2873
2874static int
2875tohex(int c)
2876{
2877	if (c >= '0' && c <= '9')
2878		return (c - '0');
2879	else if (c >= 'A' && c <= 'F')
2880		return (c - 'A' + 10);
2881	else if (c >= 'a' && c <= 'f')
2882		return (c - 'a' + 10);
2883	else
2884		return (-1);
2885}
2886