archival/libunarchive/get_header_tar.c

/* vi: set sw=4 ts=4: */
/* Licensed under GPLv2 or later, see file LICENSE in this tarball for details.
 *
 *  FIXME:
 *    In privileged mode if uname and gname map to a uid and gid then use the
 *    mapped value instead of the uid/gid values in tar header
 *
 *  References:
 *    GNU tar and star man pages,
 *    Opengroup's ustar interchange format,
 *	http://www.opengroup.org/onlinepubs/007904975/utilities/pax.html
 */

#include "libbb.h"
#include "unarchive.h"

/*
 * GNU tar uses "base-256 encoding" for very large numbers (>8 billion).
 * Encoding is binary, with highest bit always set as a marker
 * and sign in next-highest bit:
 * 80 00 .. 00 - zero
 * bf ff .. ff - largest positive number
 * ff ff .. ff - minus 1
 * c0 00 .. 00 - smallest negative number
 *
 * We expect it only in size field, where negative numbers don't make sense.
 */
static off_t getBase256_len12(const char *str)
{
	off_t value;
	int len;

	/* if (*str & 0x40) error; - caller prevents this */

	if (sizeof(off_t) >= 12) {
		/* Probably 128-bit (16 byte) off_t. Can be optimized. */
		len = 12;
		value = *str++ & 0x3f;
		while (--len)
			value = (value << 8) + (unsigned char) *str++;
		return value;
	}

#ifdef CHECK_FOR_OVERFLOW
	/* Can be optimized to eat 32-bit chunks */
	char c = *str++ & 0x3f;
	len = 12;
	while (1) {
		if (c)
			bb_error_msg_and_die("overflow in base-256 encoded file size");
		if (--len == sizeof(off_t))
			break;
		c = *str++;
	}
#else
	str += (12 - sizeof(off_t));
#endif

/* Now str points to sizeof(off_t) least significant bytes.
 *
 * Example of tar file with 8914993153 (0x213600001) byte file.
 * Field starts at offset 7c:
 * 00070  30 30 30 00 30 30 30 30  30 30 30 00 80 00 00 00  |000.0000000.....|
 * 00080  00 00 00 02 13 60 00 01  31 31 31 32 30 33 33 36  |.....`..11120336|
 *
 * str is at offset 80 or 84 now (64-bit or 32-bit off_t).
 * We (ab)use the fact that value happens to be aligned,
 * and fetch it in one go:
 */
	if (sizeof(off_t) == 8) {
		value = *(off_t*)str;
		value = SWAP_BE64(value);
	} else if (sizeof(off_t) == 4) {
		value = *(off_t*)str;
		value = SWAP_BE32(value);
	} else {
		value = 0;
		len = sizeof(off_t);
		while (--len)
			value = (value << 8) + (unsigned char) *str++;
	}
	return value;
}

/* NB: _DESTROYS_ str[len] character! */
static unsigned long long getOctal(char *str, int len)
{
	unsigned long long v;
	/* NB: leading spaces are allowed. Using strtoull to handle that.
	 * The downside is that we accept e.g. "-123" too :(
	 */
	str[len] = '\0';
	v = strtoull(str, &str, 8);
	/* std: "Each numeric field is terminated by one or more
	 * <space> or NUL characters". We must support ' '! */
	if (*str != '\0' && *str != ' ')
		bb_error_msg_and_die("corrupted octal value in tar header");
	return v;
}
#define GET_OCTAL(a) getOctal((a), sizeof(a))

void BUG_tar_header_size(void);
char FAST_FUNC get_header_tar(archive_handle_t *archive_handle)
{
	file_header_t *file_header = archive_handle->file_header;
	struct {
		/* ustar header, Posix 1003.1 */
		char name[100];     /*   0-99 */
		char mode[8];       /* 100-107 */
		char uid[8];        /* 108-115 */
		char gid[8];        /* 116-123 */
		char size[12];      /* 124-135 */
		char mtime[12];     /* 136-147 */
		char chksum[8];     /* 148-155 */
		char typeflag;      /* 156-156 */
		char linkname[100]; /* 157-256 */
		/* POSIX:   "ustar" NUL "00" */
		/* GNU tar: "ustar  " NUL */
		/* Normally it's defined as magic[6] followed by
		 * version[2], but we put them together to simplify code
		 */
		char magic[8];      /* 257-264 */
		char uname[32];     /* 265-296 */
		char gname[32];     /* 297-328 */
		char devmajor[8];   /* 329-336 */
		char devminor[8];   /* 337-344 */
		char prefix[155];   /* 345-499 */
		char padding[12];   /* 500-512 */
	} tar;
	char *cp;
	int i, sum_u, sum;
#if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
	int sum_s;
#endif
	int parse_names;

	/* Our "private data" */
#if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
# define p_longname (archive_handle->tar__longname)
# define p_linkname (archive_handle->tar__linkname)
#else
# define p_longname 0
# define p_linkname 0
#endif

	if (sizeof(tar) != 512)
		BUG_tar_header_size();

#if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
 again:
#endif
	/* Align header */
	data_align(archive_handle, 512);

 again_after_align:

#if ENABLE_DESKTOP || ENABLE_FEATURE_TAR_AUTODETECT
	/* to prevent misdetection of bz2 sig */
	*(uint32_t*)(&tar) = 0;
	i = full_read(archive_handle->src_fd, &tar, 512);
	/* If GNU tar sees EOF in above read, it says:
	 * "tar: A lone zero block at N", where N = kilobyte
	 * where EOF was met (not EOF block, actual EOF!),
	 * and exits with EXIT_SUCCESS.
	 * We will mimic exit(EXIT_SUCCESS), although we will not mimic
	 * the message and we don't check whether we indeed
	 * saw zero block directly before this. */
	if (i == 0) {
		xfunc_error_retval = 0;
 short_read:
		bb_error_msg_and_die("short read");
	}
	if (i != 512) {
		IF_FEATURE_TAR_AUTODETECT(goto autodetect;)
		goto short_read;
	}

#else
	i = 512;
	xread(archive_handle->src_fd, &tar, i);
#endif
	archive_handle->offset += i;

	/* If there is no filename its an empty header */
	if (tar.name[0] == 0 && tar.prefix[0] == 0) {
		if (archive_handle->tar__end) {
			/* Second consecutive empty header - end of archive.
			 * Read until the end to empty the pipe from gz or bz2
			 */
			while (full_read(archive_handle->src_fd, &tar, 512) == 512)
				continue;
			return EXIT_FAILURE;
		}
		archive_handle->tar__end = 1;
		return EXIT_SUCCESS;
	}
	archive_handle->tar__end = 0;

	/* Check header has valid magic, "ustar" is for the proper tar,
	 * five NULs are for the old tar format  */
	if (strncmp(tar.magic, "ustar", 5) != 0
	 && (!ENABLE_FEATURE_TAR_OLDGNU_COMPATIBILITY
	     || memcmp(tar.magic, "\0\0\0\0", 5) != 0)
	) {
#if ENABLE_FEATURE_TAR_AUTODETECT
		char FAST_FUNC (*get_header_ptr)(archive_handle_t *);

 autodetect:
		/* tar gz/bz autodetect: check for gz/bz2 magic.
		 * If we see the magic, and it is the very first block,
		 * we can switch to get_header_tar_gz/bz2/lzma().
		 * Needs seekable fd. I wish recv(MSG_PEEK) works
		 * on any fd... */
#if ENABLE_FEATURE_SEAMLESS_GZ
		if (tar.name[0] == 0x1f && tar.name[1] == (char)0x8b) { /* gzip */
			get_header_ptr = get_header_tar_gz;
		} else
#endif
#if ENABLE_FEATURE_SEAMLESS_BZ2
		if (tar.name[0] == 'B' && tar.name[1] == 'Z'
		 && tar.name[2] == 'h' && isdigit(tar.name[3])
		) { /* bzip2 */
			get_header_ptr = get_header_tar_bz2;
		} else
#endif
			goto err;
		/* Two different causes for lseek() != 0:
		 * unseekable fd (would like to support that too, but...),
		 * or not first block (false positive, it's not .gz/.bz2!) */
		if (lseek(archive_handle->src_fd, -i, SEEK_CUR) != 0)
			goto err;
		while (get_header_ptr(archive_handle) == EXIT_SUCCESS)
			continue;
		return EXIT_FAILURE;
 err:
#endif /* FEATURE_TAR_AUTODETECT */
		bb_error_msg_and_die("invalid tar magic");
	}

	/* Do checksum on headers.
	 * POSIX says that checksum is done on unsigned bytes, but
	 * Sun and HP-UX gets it wrong... more details in
	 * GNU tar source. */
#if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
	sum_s = ' ' * sizeof(tar.chksum);
#endif
	sum_u = ' ' * sizeof(tar.chksum);
	for (i = 0; i < 148; i++) {
		sum_u += ((unsigned char*)&tar)[i];
#if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
		sum_s += ((signed char*)&tar)[i];
#endif
	}
	for (i = 156; i < 512; i++) {
		sum_u += ((unsigned char*)&tar)[i];
#if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
		sum_s += ((signed char*)&tar)[i];
#endif
	}
	/* This field does not need special treatment (getOctal) */
	{
		char *endp; /* gcc likes temp var for &endp */
		sum = strtoul(tar.chksum, &endp, 8);
		if ((*endp != '\0' && *endp != ' ')
		 || (sum_u != sum IF_FEATURE_TAR_OLDSUN_COMPATIBILITY(&& sum_s != sum))
		) {
			bb_error_msg_and_die("invalid tar header checksum");
		}
	}
	/* don't use xstrtoul, tar.chksum may have leading spaces */
	sum = strtoul(tar.chksum, NULL, 8);
	if (sum_u != sum IF_FEATURE_TAR_OLDSUN_COMPATIBILITY(&& sum_s != sum)) {
		bb_error_msg_and_die("invalid tar header checksum");
	}

	/* 0 is reserved for high perf file, treat as normal file */
	if (!tar.typeflag) tar.typeflag = '0';
	parse_names = (tar.typeflag >= '0' && tar.typeflag <= '7');

	/* getOctal trashes subsequent field, therefore we call it
	 * on fields in reverse order */
	if (tar.devmajor[0]) {
		char t = tar.prefix[0];
		/* we trash prefix[0] here, but we DO need it later! */
		unsigned minor = GET_OCTAL(tar.devminor);
		unsigned major = GET_OCTAL(tar.devmajor);
		file_header->device = makedev(major, minor);
		tar.prefix[0] = t;
	}
	file_header->link_target = NULL;
	if (!p_linkname && parse_names && tar.linkname[0]) {
		file_header->link_target = xstrndup(tar.linkname, sizeof(tar.linkname));
		/* FIXME: what if we have non-link object with link_target? */
		/* Will link_target be free()ed? */
	}
#if ENABLE_FEATURE_TAR_UNAME_GNAME
	file_header->tar__uname = tar.uname[0] ? xstrndup(tar.uname, sizeof(tar.uname)) : NULL;
	file_header->tar__gname = tar.gname[0] ? xstrndup(tar.gname, sizeof(tar.gname)) : NULL;
#endif
	/* mtime: rudimentally handle GNU tar's "base256 encoding"
	 * People report tarballs with NEGATIVE unix times encoded that way */
	file_header->mtime = (tar.mtime[0] & 0x80) /* base256? */
			? 0 /* bogus */
			: GET_OCTAL(tar.mtime);
	/* size: handle GNU tar's "base256 encoding" */
	file_header->size = (tar.size[0] & 0xc0) == 0x80 /* positive base256? */
			? getBase256_len12(tar.size)
			: GET_OCTAL(tar.size);
	file_header->gid = GET_OCTAL(tar.gid);
	file_header->uid = GET_OCTAL(tar.uid);
	/* Set bits 0-11 of the files mode */
	file_header->mode = 07777 & GET_OCTAL(tar.mode);

	file_header->name = NULL;
	if (!p_longname && parse_names) {
		/* we trash mode[0] here, it's ok */
		//tar.name[sizeof(tar.name)] = '\0'; - gcc 4.3.0 would complain
		tar.mode[0] = '\0';
		if (tar.prefix[0]) {
			/* and padding[0] */
			//tar.prefix[sizeof(tar.prefix)] = '\0'; - gcc 4.3.0 would complain
			tar.padding[0] = '\0';
			file_header->name = concat_path_file(tar.prefix, tar.name);
		} else
			file_header->name = xstrdup(tar.name);
	}

	/* Set bits 12-15 of the files mode */
	/* (typeflag was not trashed because chksum does not use getOctal) */
	switch (tar.typeflag) {
	/* busybox identifies hard links as being regular files with 0 size and a link name */
	case '1':
		file_header->mode |= S_IFREG;
		break;
	case '7':
	/* case 0: */
	case '0':
#if ENABLE_FEATURE_TAR_OLDGNU_COMPATIBILITY
		if (last_char_is(file_header->name, '/')) {
			goto set_dir;
		}
#endif
		file_header->mode |= S_IFREG;
		break;
	case '2':
		file_header->mode |= S_IFLNK;
		/* have seen tarballs with size field containing
		 * the size of the link target's name */
 size0:
		file_header->size = 0;
		break;
	case '3':
		file_header->mode |= S_IFCHR;
		goto size0; /* paranoia */
	case '4':
		file_header->mode |= S_IFBLK;
		goto size0;
	case '5':
 IF_FEATURE_TAR_OLDGNU_COMPATIBILITY(set_dir:)
		file_header->mode |= S_IFDIR;
		goto size0;
	case '6':
		file_header->mode |= S_IFIFO;
		goto size0;
#if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
	case 'L':
		/* free: paranoia: tar with several consecutive longnames */
		free(p_longname);
		/* For paranoia reasons we allocate extra NUL char */
		p_longname = xzalloc(file_header->size + 1);
		/* We read ASCIZ string, including NUL */
		xread(archive_handle->src_fd, p_longname, file_header->size);
		archive_handle->offset += file_header->size;
		/* return get_header_tar(archive_handle); */
		/* gcc 4.1.1 didn't optimize it into jump */
		/* so we will do it ourself, this also saves stack */
		goto again;
	case 'K':
		free(p_linkname);
		p_linkname = xzalloc(file_header->size + 1);
		xread(archive_handle->src_fd, p_linkname, file_header->size);
		archive_handle->offset += file_header->size;
		/* return get_header_tar(archive_handle); */
		goto again;
	case 'D':	/* GNU dump dir */
	case 'M':	/* Continuation of multi volume archive */
	case 'N':	/* Old GNU for names > 100 characters */
	case 'S':	/* Sparse file */
	case 'V':	/* Volume header */
#endif
	case 'g':	/* pax global header */
	case 'x': {	/* pax extended header */
		off_t sz;
		bb_error_msg("warning: skipping header '%c'", tar.typeflag);
		sz = (file_header->size + 511) & ~(off_t)511;
		archive_handle->offset += sz;
		sz >>= 9; /* sz /= 512 but w/o contortions for signed div */
		while (sz--)
			xread(archive_handle->src_fd, &tar, 512);
		/* return get_header_tar(archive_handle); */
		goto again_after_align;
	}
	default:
		bb_error_msg_and_die("unknown typeflag: 0x%x", tar.typeflag);
	}

#if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
	if (p_longname) {
		file_header->name = p_longname;
		p_longname = NULL;
	}
	if (p_linkname) {
		file_header->link_target = p_linkname;
		p_linkname = NULL;
	}
#endif
	if (strncmp(file_header->name, "/../"+1, 3) == 0
	 || strstr(file_header->name, "/../")
	) {
		bb_error_msg_and_die("name with '..' encountered: '%s'",
				file_header->name);
	}

	/* Strip trailing '/' in directories */
	/* Must be done after mode is set as '/' is used to check if it's a directory */
	cp = last_char_is(file_header->name, '/');

	if (archive_handle->filter(archive_handle) == EXIT_SUCCESS) {
		archive_handle->action_header(/*archive_handle->*/ file_header);
		/* Note that we kill the '/' only after action_header() */
		/* (like GNU tar 1.15.1: verbose mode outputs "dir/dir/") */
		if (cp) *cp = '\0';
		archive_handle->ah_flags |= ARCHIVE_EXTRACT_QUIET;
		archive_handle->action_data(archive_handle);
		llist_add_to(&(archive_handle->passed), file_header->name);
	} else {
		data_skip(archive_handle);
		free(file_header->name);
	}
	archive_handle->offset += file_header->size;

	free(file_header->link_target);
	/* Do not free(file_header->name)! (why?) */
#if ENABLE_FEATURE_TAR_UNAME_GNAME
	free(file_header->tar__uname);
	free(file_header->tar__gname);
#endif
	return EXIT_SUCCESS;
}
1	/* vi: set sw=4 ts=4: */
2	/* Licensed under GPLv2 or later, see file LICENSE in this tarball for details.
3	*
4	* FIXME:
5	* In privileged mode if uname and gname map to a uid and gid then use the
6	* mapped value instead of the uid/gid values in tar header
7	*
8	* References:
9	* GNU tar and star man pages,
10	* Opengroup's ustar interchange format,
11	* http://www.opengroup.org/onlinepubs/007904975/utilities/pax.html
12	*/
13
14	#include "libbb.h"
15	#include "unarchive.h"
16
17	/*
18	* GNU tar uses "base-256 encoding" for very large numbers (>8 billion).
19	* Encoding is binary, with highest bit always set as a marker
20	* and sign in next-highest bit:
21	* 80 00 .. 00 - zero
22	* bf ff .. ff - largest positive number
23	* ff ff .. ff - minus 1
24	* c0 00 .. 00 - smallest negative number
25	*
26	* We expect it only in size field, where negative numbers don't make sense.
27	*/
28	static off_t getBase256_len12(const char *str)
29	{
30	off_t value;
31	int len;
32
33	/* if (str & 0x40) error; - caller prevents this /
34
35	if (sizeof(off_t) >= 12) {
36	/* Probably 128-bit (16 byte) off_t. Can be optimized. */
37	len = 12;
38	value = *str++ & 0x3f;
39	while (--len)
40	value = (value << 8) + (unsigned char) *str++;
41	return value;
42	}
43
44	#ifdef CHECK_FOR_OVERFLOW
45	/* Can be optimized to eat 32-bit chunks */
46	char c = *str++ & 0x3f;
47	len = 12;
48	while (1) {
49	if (c)
50	bb_error_msg_and_die("overflow in base-256 encoded file size");
51	if (--len == sizeof(off_t))
52	break;
53	c = *str++;
54	}
55	#else
56	str += (12 - sizeof(off_t));
57	#endif
58
59	/* Now str points to sizeof(off_t) least significant bytes.
60	*
61	* Example of tar file with 8914993153 (0x213600001) byte file.
62	* Field starts at offset 7c:
63	* 00070 30 30 30 00 30 30 30 30 30 30 30 00 80 00 00 00 \|000.0000000.....\|
64	* 00080 00 00 00 02 13 60 00 01 31 31 31 32 30 33 33 36 \|.....`..11120336\|
65	*
66	* str is at offset 80 or 84 now (64-bit or 32-bit off_t).
67	* We (ab)use the fact that value happens to be aligned,
68	* and fetch it in one go:
69	*/
70	if (sizeof(off_t) == 8) {
71	value = (off_t)str;
72	value = SWAP_BE64(value);
73	} else if (sizeof(off_t) == 4) {
74	value = (off_t)str;
75	value = SWAP_BE32(value);
76	} else {
77	value = 0;
78	len = sizeof(off_t);
79	while (--len)
80	value = (value << 8) + (unsigned char) *str++;
81	}
82	return value;
83	}
84
85	/* NB: _DESTROYS_ str[len] character! */
86	static unsigned long long getOctal(char *str, int len)
87	{
88	unsigned long long v;
89	/* NB: leading spaces are allowed. Using strtoull to handle that.
90	* The downside is that we accept e.g. "-123" too :(
91	*/
92	str[len] = '\0';
93	v = strtoull(str, &str, 8);
94	/* std: "Each numeric field is terminated by one or more
95	* <space> or NUL characters". We must support ' '! */
96	if (str != '\0' && str != ' ')
97	bb_error_msg_and_die("corrupted octal value in tar header");
98	return v;
99	}
100	#define GET_OCTAL(a) getOctal((a), sizeof(a))
101
102	void BUG_tar_header_size(void);
103	char FAST_FUNC get_header_tar(archive_handle_t *archive_handle)
104	{
105	file_header_t *file_header = archive_handle->file_header;
106	struct {
107	/* ustar header, Posix 1003.1 */
108	char name[100]; /* 0-99 */
109	char mode[8]; /* 100-107 */
110	char uid[8]; /* 108-115 */
111	char gid[8]; /* 116-123 */
112	char size[12]; /* 124-135 */
113	char mtime[12]; /* 136-147 */
114	char chksum[8]; /* 148-155 */
115	char typeflag; /* 156-156 */
116	char linkname[100]; /* 157-256 */
117	/* POSIX: "ustar" NUL "00" */
118	/* GNU tar: "ustar " NUL */
119	/* Normally it's defined as magic[6] followed by
120	* version[2], but we put them together to simplify code
121	*/
122	char magic[8]; /* 257-264 */
123	char uname[32]; /* 265-296 */
124	char gname[32]; /* 297-328 */
125	char devmajor[8]; /* 329-336 */
126	char devminor[8]; /* 337-344 */
127	char prefix[155]; /* 345-499 */
128	char padding[12]; /* 500-512 */
129	} tar;
130	char *cp;
131	int i, sum_u, sum;
132	#if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
133	int sum_s;
134	#endif
135	int parse_names;
136
137	/* Our "private data" */
138	#if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
139	# define p_longname (archive_handle->tar__longname)
140	# define p_linkname (archive_handle->tar__linkname)
141	#else
142	# define p_longname 0
143	# define p_linkname 0
144	#endif
145
146	if (sizeof(tar) != 512)
147	BUG_tar_header_size();
148
149	#if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
150	again:
151	#endif
152	/* Align header */
153	data_align(archive_handle, 512);
154
155	again_after_align:
156
157	#if ENABLE_DESKTOP \|\| ENABLE_FEATURE_TAR_AUTODETECT
158	/* to prevent misdetection of bz2 sig */
159	(uint32_t)(&tar) = 0;
160	i = full_read(archive_handle->src_fd, &tar, 512);
161	/* If GNU tar sees EOF in above read, it says:
162	* "tar: A lone zero block at N", where N = kilobyte
163	* where EOF was met (not EOF block, actual EOF!),
164	* and exits with EXIT_SUCCESS.
165	* We will mimic exit(EXIT_SUCCESS), although we will not mimic
166	* the message and we don't check whether we indeed
167	* saw zero block directly before this. */
168	if (i == 0) {
169	xfunc_error_retval = 0;
170	short_read:
171	bb_error_msg_and_die("short read");
172	}
173	if (i != 512) {
174	IF_FEATURE_TAR_AUTODETECT(goto autodetect;)
175	goto short_read;
176	}
177
178	#else
179	i = 512;
180	xread(archive_handle->src_fd, &tar, i);
181	#endif
182	archive_handle->offset += i;
183
184	/* If there is no filename its an empty header */
185	if (tar.name[0] == 0 && tar.prefix[0] == 0) {
186	if (archive_handle->tar__end) {
187	/* Second consecutive empty header - end of archive.
188	* Read until the end to empty the pipe from gz or bz2
189	*/
190	while (full_read(archive_handle->src_fd, &tar, 512) == 512)
191	continue;
192	return EXIT_FAILURE;
193	}
194	archive_handle->tar__end = 1;
195	return EXIT_SUCCESS;
196	}
197	archive_handle->tar__end = 0;
198
199	/* Check header has valid magic, "ustar" is for the proper tar,
200	* five NULs are for the old tar format */
201	if (strncmp(tar.magic, "ustar", 5) != 0
202	&& (!ENABLE_FEATURE_TAR_OLDGNU_COMPATIBILITY
203	\|\| memcmp(tar.magic, "\0\0\0\0", 5) != 0)
204	) {
205	#if ENABLE_FEATURE_TAR_AUTODETECT
206	char FAST_FUNC (get_header_ptr)(archive_handle_t );
207
208	autodetect:
209	/* tar gz/bz autodetect: check for gz/bz2 magic.
210	* If we see the magic, and it is the very first block,
211	* we can switch to get_header_tar_gz/bz2/lzma().
212	* Needs seekable fd. I wish recv(MSG_PEEK) works
213	* on any fd... */
214	#if ENABLE_FEATURE_SEAMLESS_GZ
215	if (tar.name[0] == 0x1f && tar.name[1] == (char)0x8b) { /* gzip */
216	get_header_ptr = get_header_tar_gz;
217	} else
218	#endif
219	#if ENABLE_FEATURE_SEAMLESS_BZ2
220	if (tar.name[0] == 'B' && tar.name[1] == 'Z'
221	&& tar.name[2] == 'h' && isdigit(tar.name[3])
222	) { /* bzip2 */
223	get_header_ptr = get_header_tar_bz2;
224	} else
225	#endif
226	goto err;
227	/* Two different causes for lseek() != 0:
228	* unseekable fd (would like to support that too, but...),
229	* or not first block (false positive, it's not .gz/.bz2!) */
230	if (lseek(archive_handle->src_fd, -i, SEEK_CUR) != 0)
231	goto err;
232	while (get_header_ptr(archive_handle) == EXIT_SUCCESS)
233	continue;
234	return EXIT_FAILURE;
235	err:
236	#endif /* FEATURE_TAR_AUTODETECT */
237	bb_error_msg_and_die("invalid tar magic");
238	}
239
240	/* Do checksum on headers.
241	* POSIX says that checksum is done on unsigned bytes, but
242	* Sun and HP-UX gets it wrong... more details in
243	* GNU tar source. */
244	#if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
245	sum_s = ' ' * sizeof(tar.chksum);
246	#endif
247	sum_u = ' ' * sizeof(tar.chksum);
248	for (i = 0; i < 148; i++) {
249	sum_u += ((unsigned char*)&tar)[i];
250	#if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
251	sum_s += ((signed char*)&tar)[i];
252	#endif
253	}
254	for (i = 156; i < 512; i++) {
255	sum_u += ((unsigned char*)&tar)[i];
256	#if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
257	sum_s += ((signed char*)&tar)[i];
258	#endif
259	}
260	/* This field does not need special treatment (getOctal) */
261	{
262	char endp; / gcc likes temp var for &endp */
263	sum = strtoul(tar.chksum, &endp, 8);
264	if ((endp != '\0' && endp != ' ')
265	\|\| (sum_u != sum IF_FEATURE_TAR_OLDSUN_COMPATIBILITY(&& sum_s != sum))
266	) {
267	bb_error_msg_and_die("invalid tar header checksum");
268	}
269	}
270	/* don't use xstrtoul, tar.chksum may have leading spaces */
271	sum = strtoul(tar.chksum, NULL, 8);
272	if (sum_u != sum IF_FEATURE_TAR_OLDSUN_COMPATIBILITY(&& sum_s != sum)) {
273	bb_error_msg_and_die("invalid tar header checksum");
274	}
275
276	/* 0 is reserved for high perf file, treat as normal file */
277	if (!tar.typeflag) tar.typeflag = '0';
278	parse_names = (tar.typeflag >= '0' && tar.typeflag <= '7');
279
280	/* getOctal trashes subsequent field, therefore we call it
281	* on fields in reverse order */
282	if (tar.devmajor[0]) {
283	char t = tar.prefix[0];
284	/* we trash prefix[0] here, but we DO need it later! */
285	unsigned minor = GET_OCTAL(tar.devminor);
286	unsigned major = GET_OCTAL(tar.devmajor);
287	file_header->device = makedev(major, minor);
288	tar.prefix[0] = t;
289	}
290	file_header->link_target = NULL;
291	if (!p_linkname && parse_names && tar.linkname[0]) {
292	file_header->link_target = xstrndup(tar.linkname, sizeof(tar.linkname));
293	/* FIXME: what if we have non-link object with link_target? */
294	/* Will link_target be free()ed? */
295	}
296	#if ENABLE_FEATURE_TAR_UNAME_GNAME
297	file_header->tar__uname = tar.uname[0] ? xstrndup(tar.uname, sizeof(tar.uname)) : NULL;
298	file_header->tar__gname = tar.gname[0] ? xstrndup(tar.gname, sizeof(tar.gname)) : NULL;
299	#endif
300	/* mtime: rudimentally handle GNU tar's "base256 encoding"
301	* People report tarballs with NEGATIVE unix times encoded that way */
302	file_header->mtime = (tar.mtime[0] & 0x80) /* base256? */
303	? 0 /* bogus */
304	: GET_OCTAL(tar.mtime);
305	/* size: handle GNU tar's "base256 encoding" */
306	file_header->size = (tar.size[0] & 0xc0) == 0x80 /* positive base256? */
307	? getBase256_len12(tar.size)
308	: GET_OCTAL(tar.size);
309	file_header->gid = GET_OCTAL(tar.gid);
310	file_header->uid = GET_OCTAL(tar.uid);
311	/* Set bits 0-11 of the files mode */
312	file_header->mode = 07777 & GET_OCTAL(tar.mode);
313
314	file_header->name = NULL;
315	if (!p_longname && parse_names) {
316	/* we trash mode[0] here, it's ok */
317	//tar.name[sizeof(tar.name)] = '\0'; - gcc 4.3.0 would complain
318	tar.mode[0] = '\0';
319	if (tar.prefix[0]) {
320	/* and padding[0] */
321	//tar.prefix[sizeof(tar.prefix)] = '\0'; - gcc 4.3.0 would complain
322	tar.padding[0] = '\0';
323	file_header->name = concat_path_file(tar.prefix, tar.name);
324	} else
325	file_header->name = xstrdup(tar.name);
326	}
327
328	/* Set bits 12-15 of the files mode */
329	/* (typeflag was not trashed because chksum does not use getOctal) */
330	switch (tar.typeflag) {
331	/* busybox identifies hard links as being regular files with 0 size and a link name */
332	case '1':
333	file_header->mode \|= S_IFREG;
334	break;
335	case '7':
336	/* case 0: */
337	case '0':
338	#if ENABLE_FEATURE_TAR_OLDGNU_COMPATIBILITY
339	if (last_char_is(file_header->name, '/')) {
340	goto set_dir;
341	}
342	#endif
343	file_header->mode \|= S_IFREG;
344	break;
345	case '2':
346	file_header->mode \|= S_IFLNK;
347	/* have seen tarballs with size field containing
348	* the size of the link target's name */
349	size0:
350	file_header->size = 0;
351	break;
352	case '3':
353	file_header->mode \|= S_IFCHR;
354	goto size0; /* paranoia */
355	case '4':
356	file_header->mode \|= S_IFBLK;
357	goto size0;
358	case '5':
359	IF_FEATURE_TAR_OLDGNU_COMPATIBILITY(set_dir:)
360	file_header->mode \|= S_IFDIR;
361	goto size0;
362	case '6':
363	file_header->mode \|= S_IFIFO;
364	goto size0;
365	#if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
366	case 'L':
367	/* free: paranoia: tar with several consecutive longnames */
368	free(p_longname);
369	/* For paranoia reasons we allocate extra NUL char */
370	p_longname = xzalloc(file_header->size + 1);
371	/* We read ASCIZ string, including NUL */
372	xread(archive_handle->src_fd, p_longname, file_header->size);
373	archive_handle->offset += file_header->size;
374	/* return get_header_tar(archive_handle); */
375	/* gcc 4.1.1 didn't optimize it into jump */
376	/* so we will do it ourself, this also saves stack */
377	goto again;
378	case 'K':
379	free(p_linkname);
380	p_linkname = xzalloc(file_header->size + 1);
381	xread(archive_handle->src_fd, p_linkname, file_header->size);
382	archive_handle->offset += file_header->size;
383	/* return get_header_tar(archive_handle); */
384	goto again;
385	case 'D': /* GNU dump dir */
386	case 'M': /* Continuation of multi volume archive */
387	case 'N': /* Old GNU for names > 100 characters */
388	case 'S': /* Sparse file */
389	case 'V': /* Volume header */
390	#endif
391	case 'g': /* pax global header */
392	case 'x': { /* pax extended header */
393	off_t sz;
394	bb_error_msg("warning: skipping header '%c'", tar.typeflag);
395	sz = (file_header->size + 511) & ~(off_t)511;
396	archive_handle->offset += sz;
397	sz >>= 9; /* sz /= 512 but w/o contortions for signed div */
398	while (sz--)
399	xread(archive_handle->src_fd, &tar, 512);
400	/* return get_header_tar(archive_handle); */
401	goto again_after_align;
402	}
403	default:
404	bb_error_msg_and_die("unknown typeflag: 0x%x", tar.typeflag);
405	}
406
407	#if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
408	if (p_longname) {
409	file_header->name = p_longname;
410	p_longname = NULL;
411	}
412	if (p_linkname) {
413	file_header->link_target = p_linkname;
414	p_linkname = NULL;
415	}
416	#endif
417	if (strncmp(file_header->name, "/../"+1, 3) == 0
418	\|\| strstr(file_header->name, "/../")
419	) {
420	bb_error_msg_and_die("name with '..' encountered: '%s'",
421	file_header->name);
422	}
423
424	/* Strip trailing '/' in directories */
425	/* Must be done after mode is set as '/' is used to check if it's a directory */
426	cp = last_char_is(file_header->name, '/');
427
428	if (archive_handle->filter(archive_handle) == EXIT_SUCCESS) {
429	archive_handle->action_header(/archive_handle->/ file_header);
430	/* Note that we kill the '/' only after action_header() */
431	/* (like GNU tar 1.15.1: verbose mode outputs "dir/dir/") */
432	if (cp) *cp = '\0';
433	archive_handle->ah_flags \|= ARCHIVE_EXTRACT_QUIET;
434	archive_handle->action_data(archive_handle);
435	llist_add_to(&(archive_handle->passed), file_header->name);
436	} else {
437	data_skip(archive_handle);
438	free(file_header->name);
439	}
440	archive_handle->offset += file_header->size;
441
442	free(file_header->link_target);
443	/* Do not free(file_header->name)! (why?) */
444	#if ENABLE_FEATURE_TAR_UNAME_GNAME
445	free(file_header->tar__uname);
446	free(file_header->tar__gname);
447	#endif
448	return EXIT_SUCCESS;
449	}