xref: /openbsd-src/bin/pax/pax.c (revision f2da64fbbbf1b03f09f390ab01267c93dfd77c4c)
1 /*	$OpenBSD: pax.c,v 1.47 2016/08/26 04:11:16 guenther Exp $	*/
2 /*	$NetBSD: pax.c,v 1.5 1996/03/26 23:54:20 mrg Exp $	*/
3 
4 /*-
5  * Copyright (c) 1992 Keith Muller.
6  * Copyright (c) 1992, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * Keith Muller of the University of California, San Diego.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include <sys/types.h>
38 #include <sys/stat.h>
39 #include <sys/resource.h>
40 #include <signal.h>
41 #include <unistd.h>
42 #include <stdlib.h>
43 #include <string.h>
44 #include <errno.h>
45 #include <err.h>
46 #include <fcntl.h>
47 #include <paths.h>
48 #include <stdio.h>
49 
50 #include "pax.h"
51 #include "extern.h"
52 static int gen_init(void);
53 
54 /*
55  * PAX main routines, general globals and some simple start up routines
56  */
57 
58 /*
59  * Variables that can be accessed by any routine within pax
60  */
61 int	act = DEFOP;		/* read/write/append/copy */
62 FSUB	*frmt = NULL;		/* archive format type */
63 int	cflag;			/* match all EXCEPT pattern/file */
64 int	cwdfd;			/* starting cwd */
65 int	dflag;			/* directory member match only  */
66 int	iflag;			/* interactive file/archive rename */
67 int	kflag;			/* do not overwrite existing files */
68 int	lflag;			/* use hard links when possible */
69 int	nflag;			/* select first archive member match */
70 int	tflag;			/* restore access time after read */
71 int	uflag;			/* ignore older modification time files */
72 int	vflag;			/* produce verbose output */
73 int	Dflag;			/* same as uflag except inode change time */
74 int	Hflag;			/* follow command line symlinks (write only) */
75 int	Lflag;			/* follow symlinks when writing */
76 int	Nflag;			/* only use numeric uid and gid */
77 int	Xflag;			/* archive files with same device id only */
78 int	Yflag;			/* same as Dflag except after name mode */
79 int	Zflag;			/* same as uflag except after name mode */
80 int	zeroflag;		/* use \0 as pathname terminator */
81 int	vfpart;			/* is partial verbose output in progress */
82 int	patime = 1;		/* preserve file access time */
83 int	pmtime = 1;		/* preserve file modification times */
84 int	nodirs;			/* do not create directories as needed */
85 int	pmode;			/* preserve file mode bits */
86 int	pids;			/* preserve file uid/gid */
87 int	rmleadslash = 0;	/* remove leading '/' from pathnames */
88 int	exit_val;		/* exit value */
89 int	docrc;			/* check/create file crc */
90 char	*dirptr;		/* destination dir in a copy */
91 char	*argv0;			/* root of argv[0] */
92 enum op_mode op_mode;		/* what program are we acting as? */
93 sigset_t s_mask;		/* signal mask for cleanup critical sect */
94 FILE	*listf = stderr;	/* file pointer to print file list to */
95 int	listfd = STDERR_FILENO;	/* fd matching listf, for sighandler output */
96 char	*tempfile;		/* tempfile to use for mkstemp(3) */
97 char	*tempbase;		/* basename of tempfile to use for mkstemp(3) */
98 
99 /*
100  *	PAX - Portable Archive Interchange
101  *
102  *	A utility to read, write, and write lists of the members of archive
103  *	files and copy directory hierarchies. A variety of archive formats
104  *	are supported (some are described in POSIX 1003.1 10.1):
105  *
106  *		ustar - 10.1.1 extended tar interchange format
107  *		cpio  - 10.1.2 extended cpio interchange format
108  *		tar - old BSD 4.3 tar format
109  *		binary cpio - old cpio with binary header format
110  *		sysVR4 cpio -  with and without CRC
111  *
112  * This version is a superset of IEEE Std 1003.2b-d3
113  *
114  * Summary of Extensions to the IEEE Standard:
115  *
116  * 1	READ ENHANCEMENTS
117  * 1.1	Operations which read archives will continue to operate even when
118  *	processing archives which may be damaged, truncated, or fail to meet
119  *	format specs in several different ways. Damaged sections of archives
120  *	are detected and avoided if possible. Attempts will be made to resync
121  *	archive read operations even with badly damaged media.
122  * 1.2	Blocksize requirements are not strictly enforced on archive read.
123  *	Tapes which have variable sized records can be read without errors.
124  * 1.3	The user can specify via the non-standard option flag -E if error
125  *	resync operation should stop on a media error, try a specified number
126  *	of times to correct, or try to correct forever.
127  * 1.4	Sparse files (lseek holes) stored on the archive (but stored with blocks
128  *	of all zeros will be restored with holes appropriate for the target
129  *	filesystem
130  * 1.5	The user is notified whenever something is found during archive
131  *	read operations which violates spec (but the read will continue).
132  * 1.6	Multiple archive volumes can be read and may span over different
133  *	archive devices
134  * 1.7	Rigidly restores all file attributes exactly as they are stored on the
135  *	archive.
136  * 1.8	Modification change time ranges can be specified via multiple -T
137  *	options. These allow a user to select files whose modification time
138  *	lies within a specific time range.
139  * 1.9	Files can be selected based on owner (user name or uid) via one or more
140  *	-U options.
141  * 1.10	Files can be selected based on group (group name or gid) via one o
142  *	more -G options.
143  * 1.11	File modification time can be checked against existing file after
144  *	name modification (-Z)
145  *
146  * 2	WRITE ENHANCEMENTS
147  * 2.1	Write operation will stop instead of allowing a user to create a flawed
148  *	flawed archive (due to any problem).
149  * 2.2	Archives written by pax are forced to strictly conform to both the
150  *	archive and pax the specific format specifications.
151  * 2.3	Blocking size and format is rigidly enforced on writes.
152  * 2.4	Formats which may exhibit header overflow problems (they have fields
153  *	too small for large file systems, such as inode number storage), use
154  *	routines designed to repair this problem. These techniques still
155  *	conform to both pax and format specifications, but no longer truncate
156  *	these fields. This removes any restrictions on using these archive
157  *	formats on large file systems.
158  * 2.5	Multiple archive volumes can be written and may span over different
159  *	archive devices
160  * 2.6	A archive volume record limit allows the user to specify the number
161  *	of bytes stored on an archive volume. When reached the user is
162  *	prompted for the next archive volume. This is specified with the
163  *	non-standard -B flag. The limit is rounded up to the next blocksize.
164  * 2.7	All archive padding during write use zero filled sections. This makes
165  *	it much easier to pull data out of flawed archive during read
166  *	operations.
167  * 2.8	Access time reset with the -t applies to all file nodes (including
168  *	directories).
169  * 2.9	Symbolic links can be followed with -L (optional in the spec).
170  * 2.10	Modification or inode change time ranges can be specified via
171  *	multiple -T options. These allow a user to select files whose
172  *	modification or inode change time lies within a specific time range.
173  * 2.11	Files can be selected based on owner (user name or uid) via one or more
174  *	-U options.
175  * 2.12	Files can be selected based on group (group name or gid) via one o
176  *	more -G options.
177  * 2.13	Symlinks which appear on the command line can be followed (without
178  *	following other symlinks; -H flag)
179  *
180  * 3	COPY ENHANCEMENTS
181  * 3.1	Sparse files (lseek holes) can be copied without expanding the holes
182  *	into zero filled blocks. The file copy is created with holes which are
183  *	appropriate for the target filesystem
184  * 3.2	Access time as well as modification time on copied file trees can be
185  *	preserved with the appropriate -p options.
186  * 3.3	Access time reset with the -t applies to all file nodes (including
187  *	directories).
188  * 3.4	Symbolic links can be followed with -L (optional in the spec).
189  * 3.5	Modification or inode change time ranges can be specified via
190  *	multiple -T options. These allow a user to select files whose
191  *	modification or inode change time lies within a specific time range.
192  * 3.6	Files can be selected based on owner (user name or uid) via one or more
193  *	-U options.
194  * 3.7	Files can be selected based on group (group name or gid) via one o
195  *	more -G options.
196  * 3.8	Symlinks which appear on the command line can be followed (without
197  *	following other symlinks; -H flag)
198  * 3.9  File inode change time can be checked against existing file before
199  *	name modification (-D)
200  * 3.10 File inode change time can be checked against existing file after
201  *	name modification (-Y)
202  * 3.11	File modification time can be checked against existing file after
203  *	name modification (-Z)
204  *
205  * 4	GENERAL ENHANCEMENTS
206  * 4.1	Internal structure is designed to isolate format dependent and
207  *	independent functions. Formats are selected via a format driver table.
208  *	This encourages the addition of new archive formats by only having to
209  *	write those routines which id, read and write the archive header.
210  */
211 
212 /*
213  * main()
214  *	parse options, set up and operate as specified by the user.
215  *	any operational flaw will set exit_val to non-zero
216  * Return: 0 if ok, 1 otherwise
217  */
218 
219 int
220 main(int argc, char **argv)
221 {
222 	char *tmpdir;
223 	size_t tdlen;
224 
225 	/*
226 	 * Keep a reference to cwd, so we can always come back home.
227 	 */
228 	cwdfd = open(".", O_RDONLY | O_CLOEXEC);
229 	if (cwdfd < 0) {
230 		syswarn(1, errno, "Can't open current working directory.");
231 		return(exit_val);
232 	}
233 
234 	/*
235 	 * Where should we put temporary files?
236 	 */
237 	if ((tmpdir = getenv("TMPDIR")) == NULL || *tmpdir == '\0')
238 		tmpdir = _PATH_TMP;
239 	tdlen = strlen(tmpdir);
240 	while (tdlen > 0 && tmpdir[tdlen - 1] == '/')
241 		tdlen--;
242 	tempfile = malloc(tdlen + 1 + sizeof(_TFILE_BASE));
243 	if (tempfile == NULL) {
244 		paxwarn(1, "Cannot allocate memory for temp file name.");
245 		return(exit_val);
246 	}
247 	if (tdlen)
248 		memcpy(tempfile, tmpdir, tdlen);
249 	tempbase = tempfile + tdlen;
250 	*tempbase++ = '/';
251 
252 	/*
253 	 * parse options, determine operational mode, general init
254 	 */
255 	options(argc, argv);
256 	if ((gen_init() < 0) || (tty_init() < 0))
257 		return(exit_val);
258 
259 	/*
260 	 * pmode needs to restore setugid bits when extracting or copying,
261 	 * so can't pledge at all then.
262 	 */
263 	if (pmode == 0 || (act != EXTRACT && act != COPY)) {
264 		if (pledge("stdio rpath wpath cpath fattr dpath getpw ioctl proc exec",
265 		    NULL) == -1)
266 			err(1, "pledge");
267 
268 		/* Copy mode, or no gzip -- don't need to fork/exec. */
269 		if (gzip_program == NULL || act == COPY) {
270 			if (pledge("stdio rpath wpath cpath fattr dpath getpw ioctl",
271 			    NULL) == -1)
272 				err(1, "pledge");
273 		}
274 	}
275 
276 	/*
277 	 * select a primary operation mode
278 	 */
279 	switch (act) {
280 	case EXTRACT:
281 		extract();
282 		break;
283 	case ARCHIVE:
284 		archive();
285 		break;
286 	case APPND:
287 		if (gzip_program != NULL)
288 			errx(1, "can not gzip while appending");
289 		append();
290 		break;
291 	case COPY:
292 		copy();
293 		break;
294 	default:
295 	case LIST:
296 		list();
297 		break;
298 	}
299 	return(exit_val);
300 }
301 
302 /*
303  * sig_cleanup()
304  *	when interrupted we try to do whatever delayed processing we can.
305  *	This is not critical, but we really ought to limit our damage when we
306  *	are aborted by the user.
307  * Return:
308  *	never....
309  */
310 
311 void
312 sig_cleanup(int which_sig)
313 {
314 	char errbuf[80];
315 
316 	/*
317 	 * restore modes and times for any dirs we may have created
318 	 * or any dirs we may have read.
319 	 */
320 
321 	/* paxwarn() uses stdio; fake it as well as we can */
322 	if (which_sig == SIGXCPU)
323 		strlcpy(errbuf, "\nCPU time limit reached, cleaning up.\n",
324 		    sizeof errbuf);
325 	else
326 		strlcpy(errbuf, "\nSignal caught, cleaning up.\n",
327 		    sizeof errbuf);
328 	(void) write(STDERR_FILENO, errbuf, strlen(errbuf));
329 
330 	ar_close(1);
331 	sltab_process(1);
332 	proc_dir(1);
333 	if (tflag)
334 		atdir_end();
335 	_exit(1);
336 }
337 
338 /*
339  * setup_sig()
340  *	set a signal to be caught, but only if it isn't being ignored already
341  */
342 
343 static int
344 setup_sig(int sig, const struct sigaction *n_hand)
345 {
346 	struct sigaction o_hand;
347 
348 	if (sigaction(sig, NULL, &o_hand) < 0)
349 		return (-1);
350 
351 	if (o_hand.sa_handler == SIG_IGN)
352 		return (0);
353 
354 	return (sigaction(sig, n_hand, NULL));
355 }
356 
357 /*
358  * gen_init()
359  *	general setup routines. Not all are required, but they really help
360  *	when dealing with a medium to large sized archives.
361  */
362 
363 static int
364 gen_init(void)
365 {
366 	struct rlimit reslimit;
367 	struct sigaction n_hand;
368 
369 	/*
370 	 * Really needed to handle large archives. We can run out of memory for
371 	 * internal tables really fast when we have a whole lot of files...
372 	 */
373 	if (getrlimit(RLIMIT_DATA , &reslimit) == 0){
374 		reslimit.rlim_cur = reslimit.rlim_max;
375 		(void)setrlimit(RLIMIT_DATA , &reslimit);
376 	}
377 
378 	/*
379 	 * should file size limits be waived? if the os limits us, this is
380 	 * needed if we want to write a large archive
381 	 */
382 	if (getrlimit(RLIMIT_FSIZE , &reslimit) == 0){
383 		reslimit.rlim_cur = reslimit.rlim_max;
384 		(void)setrlimit(RLIMIT_FSIZE , &reslimit);
385 	}
386 
387 	/*
388 	 * increase the size the stack can grow to
389 	 */
390 	if (getrlimit(RLIMIT_STACK , &reslimit) == 0){
391 		reslimit.rlim_cur = reslimit.rlim_max;
392 		(void)setrlimit(RLIMIT_STACK , &reslimit);
393 	}
394 
395 	/*
396 	 * not really needed, but doesn't hurt
397 	 */
398 	if (getrlimit(RLIMIT_RSS , &reslimit) == 0){
399 		reslimit.rlim_cur = reslimit.rlim_max;
400 		(void)setrlimit(RLIMIT_RSS , &reslimit);
401 	}
402 
403 	/*
404 	 * signal handling to reset stored directory times and modes. Since
405 	 * we deal with broken pipes via failed writes we ignore it. We also
406 	 * deal with any file size limit through failed writes. Cpu time
407 	 * limits are caught and a cleanup is forced.
408 	 */
409 	if ((sigemptyset(&s_mask) < 0) || (sigaddset(&s_mask, SIGTERM) < 0) ||
410 	    (sigaddset(&s_mask,SIGINT) < 0)||(sigaddset(&s_mask,SIGHUP) < 0) ||
411 	    (sigaddset(&s_mask,SIGPIPE) < 0)||(sigaddset(&s_mask,SIGQUIT)<0) ||
412 	    (sigaddset(&s_mask,SIGXCPU) < 0)||(sigaddset(&s_mask,SIGXFSZ)<0)) {
413 		paxwarn(1, "Unable to set up signal mask");
414 		return(-1);
415 	}
416 
417 	/* snag the fd to be used from the signal handler */
418 	listfd = fileno(listf);
419 
420 	memset(&n_hand, 0, sizeof n_hand);
421 	n_hand.sa_mask = s_mask;
422 	n_hand.sa_flags = 0;
423 	n_hand.sa_handler = sig_cleanup;
424 
425 	if (setup_sig(SIGHUP,  &n_hand) ||
426 	    setup_sig(SIGTERM, &n_hand) ||
427 	    setup_sig(SIGINT,  &n_hand) ||
428 	    setup_sig(SIGQUIT, &n_hand) ||
429 	    setup_sig(SIGXCPU, &n_hand))
430 		goto out;
431 
432 	n_hand.sa_handler = SIG_IGN;
433 	if ((sigaction(SIGPIPE, &n_hand, NULL) < 0) ||
434 	    (sigaction(SIGXFSZ, &n_hand, NULL) < 0))
435 		goto out;
436 	return(0);
437 
438     out:
439 	syswarn(1, errno, "Unable to set up signal handler");
440 	return(-1);
441 }
442