1 /* vi: set sw=4 ts=4: */
2 /* Licensed under GPLv2 or later, see file LICENSE in this source tree.
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
9 * GNU tar and star man pages,
10 * Opengroup's ustar interchange format,
11 * http://www.opengroup.org/onlinepubs/007904975/utilities/pax.html
15 #include "bb_archive.h"
17 typedef uint32_t aliased_uint32_t FIX_ALIASING;
18 typedef off_t aliased_off_t FIX_ALIASING;
21 const char* FAST_FUNC strip_unsafe_prefix(const char *str)
30 if (strncmp(cp, "/../"+1, 3) == 0) {
34 cp2 = strstr(cp, "/../");
40 static smallint warned = 0;
43 bb_error_msg("removing leading '%.*s' from member names",
44 (int)(cp - str), str);
50 /* NB: _DESTROYS_ str[len] character! */
51 static unsigned long long getOctal(char *str, int len)
55 /* NB: leading spaces are allowed. Using strtoull to handle that.
56 * The downside is that we accept e.g. "-123" too :(
59 v = strtoull(str, &end, 8);
60 /* std: "Each numeric field is terminated by one or more
61 * <space> or NUL characters". We must support ' '! */
62 if (*end != '\0' && *end != ' ') {
63 int8_t first = str[0];
65 bb_error_msg_and_die("corrupted octal value in tar header");
67 * GNU tar uses "base-256 encoding" for very large numbers.
68 * Encoding is binary, with highest bit always set as a marker
69 * and sign in next-highest bit:
71 * bf ff .. ff - largest positive number
72 * ff ff .. ff - minus 1
73 * c0 00 .. 00 - smallest negative number
75 * Example of tar file with 8914993153 (0x213600001) byte file.
76 * Field starts at offset 7c:
77 * 00070 30 30 30 00 30 30 30 30 30 30 30 00 80 00 00 00 |000.0000000.....|
78 * 00080 00 00 00 02 13 60 00 01 31 31 31 32 30 33 33 36 |.....`..11120336|
80 * NB: tarballs with NEGATIVE unix times encoded that way were seen!
82 /* Sign-extend 7bit 'first' to 64bit 'v' (that is, using 6th bit as sign): */
84 first >>= 1; /* now 7th bit = 6th bit */
85 v = first; /* sign-extend 8 bits to 64 */
87 v = (v << 8) + (unsigned char) *str++;
91 #define GET_OCTAL(a) getOctal((a), sizeof(a))
93 /* "global" is 0 or 1 */
94 static void process_pax_hdr(archive_handle_t *archive_handle, unsigned sz, int global)
99 blk_sz = (sz + 511) & (~511);
100 p = buf = xmalloc(blk_sz + 1);
101 xread(archive_handle->src_fd, buf, blk_sz);
102 archive_handle->offset += blk_sz;
104 /* prevent bb_strtou from running off the buffer */
111 /* Every record has this format: "LEN NAME=VALUE\n" */
112 len = bb_strtou(p, &end, 10);
113 /* expect errno to be EINVAL, because the character
114 * following the digits should be a space
123 bb_error_msg("malformed extended header, skipped");
124 // More verbose version:
125 //bb_error_msg("malformed extended header at %"OFF_FMT"d, skipped",
126 // archive_handle->offset - (sz + len));
129 /* overwrite the terminating newline with NUL
130 * (we do not bother to check that it *was* a newline)
135 #if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
136 if (!global && strncmp(value, "path=", sizeof("path=") - 1) == 0) {
137 value += sizeof("path=") - 1;
138 free(archive_handle->tar__longname);
139 archive_handle->tar__longname = xstrdup(value);
144 #if ENABLE_FEATURE_TAR_SELINUX
145 /* Scan for SELinux contexts, via "RHT.security.selinux" keyword.
146 * This is what Red Hat's patched version of tar uses.
148 # define SELINUX_CONTEXT_KEYWORD "RHT.security.selinux"
149 if (strncmp(value, SELINUX_CONTEXT_KEYWORD"=", sizeof(SELINUX_CONTEXT_KEYWORD"=") - 1) == 0) {
150 value += sizeof(SELINUX_CONTEXT_KEYWORD"=") - 1;
151 free(archive_handle->tar__sctx[global]);
152 archive_handle->tar__sctx[global] = xstrdup(value);
161 char FAST_FUNC get_header_tar(archive_handle_t *archive_handle)
163 file_header_t *file_header = archive_handle->file_header;
164 struct tar_header_t tar;
167 #if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
172 /* Our "private data" */
173 #if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
174 # define p_longname (archive_handle->tar__longname)
175 # define p_linkname (archive_handle->tar__linkname)
177 # define p_longname 0
178 # define p_linkname 0
181 #if ENABLE_FEATURE_TAR_GNU_EXTENSIONS || ENABLE_FEATURE_TAR_SELINUX
185 data_align(archive_handle, 512);
189 #if ENABLE_DESKTOP || ENABLE_FEATURE_TAR_AUTODETECT
190 /* to prevent misdetection of bz2 sig */
191 *(aliased_uint32_t*)&tar = 0;
192 i = full_read(archive_handle->src_fd, &tar, 512);
193 /* If GNU tar sees EOF in above read, it says:
194 * "tar: A lone zero block at N", where N = kilobyte
195 * where EOF was met (not EOF block, actual EOF!),
196 * and exits with EXIT_SUCCESS.
197 * We will mimic exit(EXIT_SUCCESS), although we will not mimic
198 * the message and we don't check whether we indeed
199 * saw zero block directly before this. */
201 xfunc_error_retval = 0;
203 bb_error_msg_and_die("short read");
206 IF_FEATURE_TAR_AUTODETECT(goto autodetect;)
212 xread(archive_handle->src_fd, &tar, i);
214 archive_handle->offset += i;
216 /* If there is no filename its an empty header */
217 if (tar.name[0] == 0 && tar.prefix[0] == 0) {
218 if (archive_handle->tar__end) {
219 /* Second consecutive empty header - end of archive.
220 * Read until the end to empty the pipe from gz or bz2
222 while (full_read(archive_handle->src_fd, &tar, 512) == 512)
226 archive_handle->tar__end = 1;
229 archive_handle->tar__end = 0;
231 /* Check header has valid magic, "ustar" is for the proper tar,
232 * five NULs are for the old tar format */
233 if (strncmp(tar.magic, "ustar", 5) != 0
234 && (!ENABLE_FEATURE_TAR_OLDGNU_COMPATIBILITY
235 || memcmp(tar.magic, "\0\0\0\0", 5) != 0)
237 #if ENABLE_FEATURE_TAR_AUTODETECT
238 char FAST_FUNC (*get_header_ptr)(archive_handle_t *);
242 magic2 = *(bb__aliased_uint16_t*)tar.name;
243 /* tar gz/bz autodetect: check for gz/bz2 magic.
244 * If we see the magic, and it is the very first block,
245 * we can switch to get_header_tar_gz/bz2/lzma().
246 * Needs seekable fd. I wish recv(MSG_PEEK) works
248 # if ENABLE_FEATURE_SEAMLESS_GZ
249 if (magic2 == GZIP_MAGIC) {
250 get_header_ptr = get_header_tar_gz;
253 # if ENABLE_FEATURE_SEAMLESS_BZ2
254 if (magic2 == BZIP2_MAGIC
255 && tar.name[2] == 'h' && isdigit(tar.name[3])
257 get_header_ptr = get_header_tar_bz2;
260 # if ENABLE_FEATURE_SEAMLESS_XZ
261 //TODO: if (magic2 == XZ_MAGIC1)...
265 /* Two different causes for lseek() != 0:
266 * unseekable fd (would like to support that too, but...),
267 * or not first block (false positive, it's not .gz/.bz2!) */
268 if (lseek(archive_handle->src_fd, -i, SEEK_CUR) != 0)
270 while (get_header_ptr(archive_handle) == EXIT_SUCCESS)
274 #endif /* FEATURE_TAR_AUTODETECT */
275 bb_error_msg_and_die("invalid tar magic");
278 /* Do checksum on headers.
279 * POSIX says that checksum is done on unsigned bytes, but
280 * Sun and HP-UX gets it wrong... more details in
282 #if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
283 sum_s = ' ' * sizeof(tar.chksum);
285 sum_u = ' ' * sizeof(tar.chksum);
286 for (i = 0; i < 148; i++) {
287 sum_u += ((unsigned char*)&tar)[i];
288 #if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
289 sum_s += ((signed char*)&tar)[i];
292 for (i = 156; i < 512; i++) {
293 sum_u += ((unsigned char*)&tar)[i];
294 #if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
295 sum_s += ((signed char*)&tar)[i];
298 /* This field does not need special treatment (getOctal) */
300 char *endp; /* gcc likes temp var for &endp */
301 sum = strtoul(tar.chksum, &endp, 8);
302 if ((*endp != '\0' && *endp != ' ')
303 || (sum_u != sum IF_FEATURE_TAR_OLDSUN_COMPATIBILITY(&& sum_s != sum))
305 bb_error_msg_and_die("invalid tar header checksum");
308 /* don't use xstrtoul, tar.chksum may have leading spaces */
309 sum = strtoul(tar.chksum, NULL, 8);
310 if (sum_u != sum IF_FEATURE_TAR_OLDSUN_COMPATIBILITY(&& sum_s != sum)) {
311 bb_error_msg_and_die("invalid tar header checksum");
314 /* 0 is reserved for high perf file, treat as normal file */
315 if (!tar.typeflag) tar.typeflag = '0';
316 parse_names = (tar.typeflag >= '0' && tar.typeflag <= '7');
318 /* getOctal trashes subsequent field, therefore we call it
319 * on fields in reverse order */
320 if (tar.devmajor[0]) {
321 char t = tar.prefix[0];
322 /* we trash prefix[0] here, but we DO need it later! */
323 unsigned minor = GET_OCTAL(tar.devminor);
324 unsigned major = GET_OCTAL(tar.devmajor);
325 file_header->device = makedev(major, minor);
328 file_header->link_target = NULL;
329 if (!p_linkname && parse_names && tar.linkname[0]) {
330 file_header->link_target = xstrndup(tar.linkname, sizeof(tar.linkname));
331 /* FIXME: what if we have non-link object with link_target? */
332 /* Will link_target be free()ed? */
334 #if ENABLE_FEATURE_TAR_UNAME_GNAME
335 file_header->tar__uname = tar.uname[0] ? xstrndup(tar.uname, sizeof(tar.uname)) : NULL;
336 file_header->tar__gname = tar.gname[0] ? xstrndup(tar.gname, sizeof(tar.gname)) : NULL;
338 file_header->mtime = GET_OCTAL(tar.mtime);
339 file_header->size = GET_OCTAL(tar.size);
340 file_header->gid = GET_OCTAL(tar.gid);
341 file_header->uid = GET_OCTAL(tar.uid);
342 /* Set bits 0-11 of the files mode */
343 file_header->mode = 07777 & GET_OCTAL(tar.mode);
345 file_header->name = NULL;
346 if (!p_longname && parse_names) {
347 /* we trash mode[0] here, it's ok */
348 //tar.name[sizeof(tar.name)] = '\0'; - gcc 4.3.0 would complain
352 //tar.prefix[sizeof(tar.prefix)] = '\0'; - gcc 4.3.0 would complain
353 tar.padding[0] = '\0';
354 file_header->name = concat_path_file(tar.prefix, tar.name);
356 file_header->name = xstrdup(tar.name);
359 /* Set bits 12-15 of the files mode */
360 /* (typeflag was not trashed because chksum does not use getOctal) */
361 switch (tar.typeflag) {
362 case '1': /* hardlink */
363 /* we mark hardlinks as regular files with zero size and a link name */
364 file_header->mode |= S_IFREG;
365 /* on size of link fields from star(4)
366 * ... For tar archives written by pre POSIX.1-1988
367 * implementations, the size field usually contains the size of
368 * the file and needs to be ignored as no data may follow this
369 * header type. For POSIX.1- 1988 compliant archives, the size
370 * field needs to be 0. For POSIX.1-2001 compliant archives,
371 * the size field may be non zero, indicating that file data is
372 * included in the archive.
373 * i.e; always assume this is zero for safety.
379 #if ENABLE_FEATURE_TAR_OLDGNU_COMPATIBILITY
380 if (last_char_is(file_header->name, '/')) {
384 file_header->mode |= S_IFREG;
387 file_header->mode |= S_IFLNK;
388 /* have seen tarballs with size field containing
389 * the size of the link target's name */
391 file_header->size = 0;
394 file_header->mode |= S_IFCHR;
395 goto size0; /* paranoia */
397 file_header->mode |= S_IFBLK;
400 IF_FEATURE_TAR_OLDGNU_COMPATIBILITY(set_dir:)
401 file_header->mode |= S_IFDIR;
404 file_header->mode |= S_IFIFO;
406 #if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
408 /* free: paranoia: tar with several consecutive longnames */
410 /* For paranoia reasons we allocate extra NUL char */
411 p_longname = xzalloc(file_header->size + 1);
412 /* We read ASCIZ string, including NUL */
413 xread(archive_handle->src_fd, p_longname, file_header->size);
414 archive_handle->offset += file_header->size;
415 /* return get_header_tar(archive_handle); */
416 /* gcc 4.1.1 didn't optimize it into jump */
417 /* so we will do it ourself, this also saves stack */
421 p_linkname = xzalloc(file_header->size + 1);
422 xread(archive_handle->src_fd, p_linkname, file_header->size);
423 archive_handle->offset += file_header->size;
424 /* return get_header_tar(archive_handle); */
426 case 'D': /* GNU dump dir */
427 case 'M': /* Continuation of multi volume archive */
428 case 'N': /* Old GNU for names > 100 characters */
429 case 'S': /* Sparse file */
430 case 'V': /* Volume header */
432 case 'g': /* pax global header */
433 case 'x': { /* pax extended header */
434 if ((uoff_t)file_header->size > 0xfffff) /* paranoia */
436 process_pax_hdr(archive_handle, file_header->size, (tar.typeflag == 'g'));
437 goto again_after_align;
442 bb_error_msg("warning: skipping header '%c'", tar.typeflag);
443 sz = (file_header->size + 511) & ~(off_t)511;
444 archive_handle->offset += sz;
445 sz >>= 9; /* sz /= 512 but w/o contortions for signed div */
447 xread(archive_handle->src_fd, &tar, 512);
448 /* return get_header_tar(archive_handle); */
449 goto again_after_align;
452 bb_error_msg_and_die("unknown typeflag: 0x%x", tar.typeflag);
455 #if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
457 file_header->name = p_longname;
461 file_header->link_target = p_linkname;
466 /* Everything up to and including last ".." component is stripped */
467 overlapping_strcpy(file_header->name, strip_unsafe_prefix(file_header->name));
469 /* Strip trailing '/' in directories */
470 /* Must be done after mode is set as '/' is used to check if it's a directory */
471 cp = last_char_is(file_header->name, '/');
473 if (archive_handle->filter(archive_handle) == EXIT_SUCCESS) {
474 archive_handle->action_header(/*archive_handle->*/ file_header);
475 /* Note that we kill the '/' only after action_header() */
476 /* (like GNU tar 1.15.1: verbose mode outputs "dir/dir/") */
479 archive_handle->action_data(archive_handle);
480 if (archive_handle->accept || archive_handle->reject)
481 llist_add_to(&archive_handle->passed, file_header->name);
482 else /* Caller isn't interested in list of unpacked files */
483 free(file_header->name);
485 data_skip(archive_handle);
486 free(file_header->name);
488 archive_handle->offset += file_header->size;
490 free(file_header->link_target);
491 /* Do not free(file_header->name)!
492 * It might be inserted in archive_handle->passed - see above */
493 #if ENABLE_FEATURE_TAR_UNAME_GNAME
494 free(file_header->tar__uname);
495 free(file_header->tar__gname);