1 /* vi: set sw=4 ts=4: */
3 * Licensed under GPLv2 or later, see file LICENSE in this source tree.
6 * In privileged mode if uname and gname map to a uid and gid then use the
7 * mapped value instead of the uid/gid values in tar header
10 * GNU tar and star man pages,
11 * Opengroup's ustar interchange format,
12 * 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;
20 /* NB: _DESTROYS_ str[len] character! */
21 static unsigned long long getOctal(char *str, int len)
25 /* NB: leading spaces are allowed. Using strtoull to handle that.
26 * The downside is that we accept e.g. "-123" too :(
29 v = strtoull(str, &end, 8);
30 /* std: "Each numeric field is terminated by one or more
31 * <space> or NUL characters". We must support ' '! */
32 if (*end != '\0' && *end != ' ') {
33 int8_t first = str[0];
35 bb_simple_error_msg_and_die("corrupted octal value in tar header");
37 * GNU tar uses "base-256 encoding" for very large numbers.
38 * Encoding is binary, with highest bit always set as a marker
39 * and sign in next-highest bit:
41 * bf ff .. ff - largest positive number
42 * ff ff .. ff - minus 1
43 * c0 00 .. 00 - smallest negative number
45 * Example of tar file with 8914993153 (0x213600001) byte file.
46 * Field starts at offset 7c:
47 * 00070 30 30 30 00 30 30 30 30 30 30 30 00 80 00 00 00 |000.0000000.....|
48 * 00080 00 00 00 02 13 60 00 01 31 31 31 32 30 33 33 36 |.....`..11120336|
50 * NB: tarballs with NEGATIVE unix times encoded that way were seen!
52 /* Sign-extend 7bit 'first' to 64bit 'v' (that is, using 6th bit as sign): */
54 first >>= 1; /* now 7th bit = 6th bit */
55 v = first; /* sign-extend 8 bits to 64 */
57 v = (v << 8) + (uint8_t) *++str;
61 #define GET_OCTAL(a) getOctal((a), sizeof(a))
63 #define TAR_EXTD (ENABLE_FEATURE_TAR_GNU_EXTENSIONS || ENABLE_FEATURE_TAR_SELINUX)
65 #define process_pax_hdr(archive_handle, sz, global) \
66 process_pax_hdr(archive_handle, sz)
68 /* "global" is 0 or 1 */
69 static void process_pax_hdr(archive_handle_t *archive_handle, unsigned sz, int global)
72 unsigned blk_sz = (sz + 511) & (~511);
73 seek_by_read(archive_handle->src_fd, blk_sz);
75 unsigned blk_sz = (sz + 511) & (~511);
78 p = buf = xmalloc(blk_sz + 1);
79 xread(archive_handle->src_fd, buf, blk_sz);
80 archive_handle->offset += blk_sz;
82 /* prevent bb_strtou from running off the buffer */
89 /* Every record has this format: "LEN NAME=VALUE\n" */
90 len = bb_strtou(p, &end, 10);
91 /* expect errno to be EINVAL, because the character
92 * following the digits should be a space
97 /** (int)sz < 0 - not good enough for huge malicious VALUE of 2^32-1 */
98 (int)(sz|len) < 0 /* this works */
103 bb_simple_error_msg("malformed extended header, skipped");
104 // More verbose version:
105 //bb_error_msg("malformed extended header at %"OFF_FMT"d, skipped",
106 // archive_handle->offset - (sz + len));
109 /* overwrite the terminating newline with NUL
110 * (we do not bother to check that it *was* a newline)
115 # if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
117 if (is_prefixed_with(value, "path=")) {
118 value += sizeof("path=") - 1;
119 free(archive_handle->tar__longname);
120 archive_handle->tar__longname = xstrdup(value);
123 if (is_prefixed_with(value, "linkpath=")) {
124 value += sizeof("linkpath=") - 1;
125 free(archive_handle->tar__linkname);
126 archive_handle->tar__linkname = xstrdup(value);
132 # if ENABLE_FEATURE_TAR_SELINUX
133 /* Scan for SELinux contexts, via "RHT.security.selinux" keyword.
134 * This is what Red Hat's patched version of tar uses.
136 # define SELINUX_CONTEXT_KEYWORD "RHT.security.selinux"
137 if (is_prefixed_with(value, SELINUX_CONTEXT_KEYWORD"=")) {
138 value += sizeof(SELINUX_CONTEXT_KEYWORD"=") - 1;
139 free(archive_handle->tar__sctx[global]);
140 archive_handle->tar__sctx[global] = xstrdup(value);
150 char FAST_FUNC get_header_tar(archive_handle_t *archive_handle)
152 file_header_t *file_header = archive_handle->file_header;
153 struct tar_header_t tar;
155 int tar_typeflag; /* can be "char", "int" seems give smaller code */
157 #if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
162 /* Our "private data" */
163 #if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
164 # define p_longname (archive_handle->tar__longname)
165 # define p_linkname (archive_handle->tar__linkname)
167 # define p_longname 0
168 # define p_linkname 0
171 #if ENABLE_FEATURE_TAR_GNU_EXTENSIONS || ENABLE_FEATURE_TAR_SELINUX
175 data_align(archive_handle, 512);
179 #if ENABLE_DESKTOP || ENABLE_FEATURE_TAR_AUTODETECT
180 /* to prevent misdetection of bz2 sig */
181 *(aliased_uint32_t*)&tar = 0;
182 i = full_read(archive_handle->src_fd, &tar, 512);
183 /* If GNU tar sees EOF in above read, it says:
184 * "tar: A lone zero block at N", where N = kilobyte
185 * where EOF was met (not EOF block, actual EOF!),
186 * and exits with EXIT_SUCCESS.
187 * We will mimic exit(EXIT_SUCCESS), although we will not mimic
188 * the message and we don't check whether we indeed
189 * saw zero block directly before this. */
191 /* GNU tar 1.29 will be silent if tar archive ends abruptly
192 * (if there are no zero blocks at all, and last read returns zero,
193 * not short read 0 < len < 512). Complain only if
194 * the very first read fails. Grrr.
196 if (archive_handle->offset == 0)
197 bb_simple_error_msg("short read");
198 /* this merely signals end of archive, not exit(1): */
202 IF_FEATURE_TAR_AUTODETECT(goto autodetect;)
203 bb_simple_error_msg_and_die("short read");
208 xread(archive_handle->src_fd, &tar, i);
210 archive_handle->offset += i;
212 /* If there is no filename its an empty header */
213 if (tar.name[0] == 0 && tar.prefix[0] == 0
214 /* Have seen a tar archive with pax 'x' header supplying UTF8 filename,
215 * with actual file having all name fields NUL-filled. Check this: */
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)
224 return EXIT_FAILURE; /* "end of archive" */
226 archive_handle->tar__end = 1;
227 return EXIT_SUCCESS; /* "decoded one header" */
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 (!is_prefixed_with(tar.magic, "ustar")
234 && (!ENABLE_FEATURE_TAR_OLDGNU_COMPATIBILITY
235 || memcmp(tar.magic, "\0\0\0\0", 5) != 0)
237 #if ENABLE_FEATURE_TAR_AUTODETECT
239 /* Two different causes for lseek() != 0:
240 * unseekable fd (would like to support that too, but...),
241 * or not first block (false positive, it's not .gz/.bz2!) */
242 if (lseek(archive_handle->src_fd, -i, SEEK_CUR) != 0)
244 if (setup_unzip_on_fd(archive_handle->src_fd, /*fail_if_not_compressed:*/ 0) != 0)
246 bb_simple_error_msg_and_die("invalid tar magic");
247 archive_handle->offset = 0;
248 goto again_after_align;
250 bb_simple_error_msg_and_die("invalid tar magic");
253 /* Do checksum on headers.
254 * POSIX says that checksum is done on unsigned bytes, but
255 * Sun and HP-UX gets it wrong... more details in
257 sum_u = ' ' * sizeof(tar.chksum);
258 #if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
261 for (i = 0; i < 148; i++) {
262 sum_u += ((unsigned char*)&tar)[i];
263 #if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
264 sum_s += ((signed char*)&tar)[i];
267 for (i = 156; i < 512; i++) {
268 sum_u += ((unsigned char*)&tar)[i];
269 #if ENABLE_FEATURE_TAR_OLDSUN_COMPATIBILITY
270 sum_s += ((signed char*)&tar)[i];
273 /* Most tarfiles have tar.chksum NUL or space terminated, but
274 * github.com decided to be "special" and have unterminated field:
275 * 0090: 30343300 30303031 33323731 30000000 |043.000132710...|
277 * Need to use GET_OCTAL. This overwrites tar.typeflag ---+
278 * (the '0' char immediately after chksum in example above) with NUL.
280 tar_typeflag = (uint8_t)tar.typeflag; /* save it */
281 sum = GET_OCTAL(tar.chksum);
283 IF_FEATURE_TAR_OLDSUN_COMPATIBILITY(&& sum_s != sum)
285 bb_simple_error_msg_and_die("invalid tar header checksum");
288 /* GET_OCTAL trashes subsequent field, therefore we call it
289 * on fields in reverse order */
290 if (tar.devmajor[0]) {
291 char t = tar.prefix[0];
292 /* we trash prefix[0] here, but we DO need it later! */
293 unsigned minor = GET_OCTAL(tar.devminor);
294 unsigned major = GET_OCTAL(tar.devmajor);
295 file_header->device = makedev(major, minor);
299 /* 0 is reserved for high perf file, treat as normal file */
300 if (tar_typeflag == '\0') tar_typeflag = '0';
301 parse_names = (tar_typeflag >= '0' && tar_typeflag <= '7');
303 file_header->link_target = NULL;
304 if (!p_linkname && parse_names && tar.linkname[0]) {
305 file_header->link_target = xstrndup(tar.linkname, sizeof(tar.linkname));
306 /* FIXME: what if we have non-link object with link_target? */
307 /* Will link_target be free()ed? */
309 #if ENABLE_FEATURE_TAR_UNAME_GNAME
310 file_header->tar__uname = tar.uname[0] ? xstrndup(tar.uname, sizeof(tar.uname)) : NULL;
311 file_header->tar__gname = tar.gname[0] ? xstrndup(tar.gname, sizeof(tar.gname)) : NULL;
313 file_header->mtime = GET_OCTAL(tar.mtime);
314 file_header->size = GET_OCTAL(tar.size);
315 file_header->gid = GET_OCTAL(tar.gid);
316 file_header->uid = GET_OCTAL(tar.uid);
317 /* Set bits 0-11 of the files mode */
318 file_header->mode = 07777 & GET_OCTAL(tar.mode);
320 file_header->name = NULL;
321 if (!p_longname && parse_names) {
322 /* we trash mode[0] here, it's ok */
323 //tar.name[sizeof(tar.name)] = '\0'; - gcc 4.3.0 would complain
327 //tar.prefix[sizeof(tar.prefix)] = '\0'; - gcc 4.3.0 would complain
328 tar.padding[0] = '\0';
329 file_header->name = concat_path_file(tar.prefix, tar.name);
331 file_header->name = xstrdup(tar.name);
334 /* Set bits 12-15 of the files mode */
335 /* (typeflag was not trashed because chksum does not use getOctal) */
336 switch (tar_typeflag) {
337 case '1': /* hardlink */
338 /* we mark hardlinks as regular files with zero size and a link name */
339 file_header->mode |= S_IFREG;
340 /* on size of link fields from star(4)
341 * ... For tar archives written by pre POSIX.1-1988
342 * implementations, the size field usually contains the size of
343 * the file and needs to be ignored as no data may follow this
344 * header type. For POSIX.1- 1988 compliant archives, the size
345 * field needs to be 0. For POSIX.1-2001 compliant archives,
346 * the size field may be non zero, indicating that file data is
347 * included in the archive.
348 * i.e; always assume this is zero for safety.
354 #if ENABLE_FEATURE_TAR_OLDGNU_COMPATIBILITY
355 if (last_char_is(file_header->name, '/')) {
359 file_header->mode |= S_IFREG;
362 file_header->mode |= S_IFLNK;
363 /* have seen tarballs with size field containing
364 * the size of the link target's name */
366 file_header->size = 0;
369 file_header->mode |= S_IFCHR;
370 goto size0; /* paranoia */
372 file_header->mode |= S_IFBLK;
375 IF_FEATURE_TAR_OLDGNU_COMPATIBILITY(set_dir:)
376 file_header->mode |= S_IFDIR;
379 file_header->mode |= S_IFIFO;
381 case 'g': /* pax global header */
382 case 'x': { /* pax extended header */
383 if ((uoff_t)file_header->size > 0xfffff) /* paranoia */
385 process_pax_hdr(archive_handle, file_header->size, (tar_typeflag == 'g'));
386 goto again_after_align;
387 #if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
388 /* See http://www.gnu.org/software/tar/manual/html_node/Extensions.html */
390 /* free: paranoia: tar with several consecutive longnames */
392 /* For paranoia reasons we allocate extra NUL char */
393 p_longname = xzalloc(file_header->size + 1);
394 /* We read ASCIZ string, including NUL */
395 xread(archive_handle->src_fd, p_longname, file_header->size);
396 archive_handle->offset += file_header->size;
397 /* return get_header_tar(archive_handle); */
398 /* gcc 4.1.1 didn't optimize it into jump */
399 /* so we will do it ourself, this also saves stack */
403 p_linkname = xzalloc(file_header->size + 1);
404 xread(archive_handle->src_fd, p_linkname, file_header->size);
405 archive_handle->offset += file_header->size;
406 /* return get_header_tar(archive_handle); */
409 * case 'S': // Sparse file
410 * Was seen in the wild. Not supported (yet?).
411 * See https://www.gnu.org/software/tar/manual/html_section/tar_92.html
412 * for the format. (An "Old GNU Format" was seen, not PAX formats).
414 // case 'D': /* GNU dump dir */
415 // case 'M': /* Continuation of multi volume archive */
416 // case 'N': /* Old GNU for names > 100 characters */
417 case 'V': /* Volume header */
418 ; /* Fall through to skip it */
424 bb_error_msg("warning: skipping header '%c'", tar_typeflag);
425 sz = (file_header->size + 511) & ~(off_t)511;
426 archive_handle->offset += sz;
427 sz >>= 9; /* sz /= 512 but w/o contortions for signed div */
429 xread(archive_handle->src_fd, &tar, 512);
430 /* return get_header_tar(archive_handle); */
431 goto again_after_align;
434 bb_error_msg_and_die("unknown typeflag: 0x%x", tar_typeflag);
437 #if ENABLE_FEATURE_TAR_GNU_EXTENSIONS
439 file_header->name = p_longname;
443 file_header->link_target = p_linkname;
448 /* Everything up to and including last ".." component is stripped */
449 overlapping_strcpy(file_header->name, strip_unsafe_prefix(file_header->name));
450 //TODO: do the same for file_header->link_target?
452 /* Strip trailing '/' in directories */
453 /* Must be done after mode is set as '/' is used to check if it's a directory */
454 cp = last_char_is(file_header->name, '/');
456 if (archive_handle->filter(archive_handle) == EXIT_SUCCESS) {
457 archive_handle->action_header(/*archive_handle->*/ file_header);
458 /* Note that we kill the '/' only after action_header() */
459 /* (like GNU tar 1.15.1: verbose mode outputs "dir/dir/") */
462 archive_handle->action_data(archive_handle);
463 if (archive_handle->accept || archive_handle->reject
464 || (archive_handle->ah_flags & ARCHIVE_REMEMBER_NAMES)
466 llist_add_to(&archive_handle->passed, file_header->name);
467 } else /* Caller isn't interested in list of unpacked files */
468 free(file_header->name);
470 data_skip(archive_handle);
471 free(file_header->name);
473 archive_handle->offset += file_header->size;
475 free(file_header->link_target);
476 /* Do not free(file_header->name)!
477 * It might be inserted in archive_handle->passed - see above */
478 #if ENABLE_FEATURE_TAR_UNAME_GNAME
479 free(file_header->tar__uname);
480 free(file_header->tar__gname);
482 return EXIT_SUCCESS; /* "decoded one header" */