xref: /openbsd/bin/pax/pax.c (revision d89ec533)
1 /*	$OpenBSD: pax.c,v 1.53 2019/06/28 13:34:59 deraadt 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 <grp.h>
48 #include <paths.h>
49 #include <pwd.h>
50 #include <stdio.h>
51 
52 #include "pax.h"
53 #include "extern.h"
54 static int gen_init(void);
55 
56 /*
57  * PAX main routines, general globals and some simple start up routines
58  */
59 
60 /*
61  * Variables that can be accessed by any routine within pax
62  */
63 int	act = DEFOP;		/* read/write/append/copy */
64 FSUB	*frmt = NULL;		/* archive format type */
65 int	cflag;			/* match all EXCEPT pattern/file */
66 int	cwdfd;			/* starting cwd */
67 int	dflag;			/* directory member match only  */
68 int	iflag;			/* interactive file/archive rename */
69 int	kflag;			/* do not overwrite existing files */
70 int	lflag;			/* use hard links when possible */
71 int	nflag;			/* select first archive member match */
72 int	tflag;			/* restore access time after read */
73 int	uflag;			/* ignore older modification time files */
74 int	vflag;			/* produce verbose output */
75 int	Dflag;			/* same as uflag except inode change time */
76 int	Hflag;			/* follow command line symlinks (write only) */
77 int	Lflag;			/* follow symlinks when writing */
78 int	Nflag;			/* only use numeric uid and gid */
79 int	Xflag;			/* archive files with same device id only */
80 int	Yflag;			/* same as Dflag except after name mode */
81 int	Zflag;			/* same as uflag except after name mode */
82 int	zeroflag;		/* use \0 as pathname terminator */
83 int	vfpart;			/* is partial verbose output in progress */
84 int	patime = 1;		/* preserve file access time */
85 int	pmtime = 1;		/* preserve file modification times */
86 int	nodirs;			/* do not create directories as needed */
87 int	pmode;			/* preserve file mode bits */
88 int	pids;			/* preserve file uid/gid */
89 int	rmleadslash = 0;	/* remove leading '/' from pathnames */
90 int	exit_val;		/* exit value */
91 int	docrc;			/* check/create file crc */
92 char	*dirptr;		/* destination dir in a copy */
93 char	*argv0;			/* root of argv[0] */
94 enum op_mode op_mode;		/* what program are we acting as? */
95 sigset_t s_mask;		/* signal mask for cleanup critical sect */
96 FILE	*listf = stderr;	/* file pointer to print file list to */
97 int	listfd = STDERR_FILENO;	/* fd matching listf, for sighandler output */
98 char	*tempfile;		/* tempfile to use for mkstemp(3) */
99 char	*tempbase;		/* basename of tempfile to use for mkstemp(3) */
100 
101 /*
102  *	PAX - Portable Archive Interchange
103  *
104  *	A utility to read, write, and write lists of the members of archive
105  *	files and copy directory hierarchies. A variety of archive formats
106  *	are supported (some are described in POSIX 1003.1 10.1):
107  *
108  *		ustar - 10.1.1 extended tar interchange format
109  *		cpio  - 10.1.2 extended cpio interchange format
110  *		tar - old BSD 4.3 tar format
111  *		binary cpio - old cpio with binary header format
112  *		sysVR4 cpio -  with and without CRC
113  *
114  * This version is a superset of IEEE Std 1003.2b-d3
115  *
116  * Summary of Extensions to the IEEE Standard:
117  *
118  * 1	READ ENHANCEMENTS
119  * 1.1	Operations which read archives will continue to operate even when
120  *	processing archives which may be damaged, truncated, or fail to meet
121  *	format specs in several different ways. Damaged sections of archives
122  *	are detected and avoided if possible. Attempts will be made to resync
123  *	archive read operations even with badly damaged media.
124  * 1.2	Blocksize requirements are not strictly enforced on archive read.
125  *	Tapes which have variable sized records can be read without errors.
126  * 1.3	The user can specify via the non-standard option flag -E if error
127  *	resync operation should stop on a media error, try a specified number
128  *	of times to correct, or try to correct forever.
129  * 1.4	Sparse files (lseek holes) stored on the archive (but stored with blocks
130  *	of all zeros will be restored with holes appropriate for the target
131  *	filesystem
132  * 1.5	The user is notified whenever something is found during archive
133  *	read operations which violates spec (but the read will continue).
134  * 1.6	Multiple archive volumes can be read and may span over different
135  *	archive devices
136  * 1.7	Rigidly restores all file attributes exactly as they are stored on the
137  *	archive.
138  * 1.8	Modification change time ranges can be specified via multiple -T
139  *	options. These allow a user to select files whose modification time
140  *	lies within a specific time range.
141  * 1.9	Files can be selected based on owner (user name or uid) via one or more
142  *	-U options.
143  * 1.10	Files can be selected based on group (group name or gid) via one o
144  *	more -G options.
145  * 1.11	File modification time can be checked against existing file after
146  *	name modification (-Z)
147  *
148  * 2	WRITE ENHANCEMENTS
149  * 2.1	Write operation will stop instead of allowing a user to create a flawed
150  *	flawed archive (due to any problem).
151  * 2.2	Archives written by pax are forced to strictly conform to both the
152  *	archive and pax the specific format specifications.
153  * 2.3	Blocking size and format is rigidly enforced on writes.
154  * 2.4	Formats which may exhibit header overflow problems (they have fields
155  *	too small for large file systems, such as inode number storage), use
156  *	routines designed to repair this problem. These techniques still
157  *	conform to both pax and format specifications, but no longer truncate
158  *	these fields. This removes any restrictions on using these archive
159  *	formats on large file systems.
160  * 2.5	Multiple archive volumes can be written and may span over different
161  *	archive devices
162  * 2.6	A archive volume record limit allows the user to specify the number
163  *	of bytes stored on an archive volume. When reached the user is
164  *	prompted for the next archive volume. This is specified with the
165  *	non-standard -B flag. The limit is rounded up to the next blocksize.
166  * 2.7	All archive padding during write use zero filled sections. This makes
167  *	it much easier to pull data out of flawed archive during read
168  *	operations.
169  * 2.8	Access time reset with the -t applies to all file nodes (including
170  *	directories).
171  * 2.9	Symbolic links can be followed with -L (optional in the spec).
172  * 2.10	Modification or inode change time ranges can be specified via
173  *	multiple -T options. These allow a user to select files whose
174  *	modification or inode change time lies within a specific time range.
175  * 2.11	Files can be selected based on owner (user name or uid) via one or more
176  *	-U options.
177  * 2.12	Files can be selected based on group (group name or gid) via one o
178  *	more -G options.
179  * 2.13	Symlinks which appear on the command line can be followed (without
180  *	following other symlinks; -H flag)
181  *
182  * 3	COPY ENHANCEMENTS
183  * 3.1	Sparse files (lseek holes) can be copied without expanding the holes
184  *	into zero filled blocks. The file copy is created with holes which are
185  *	appropriate for the target filesystem
186  * 3.2	Access time as well as modification time on copied file trees can be
187  *	preserved with the appropriate -p options.
188  * 3.3	Access time reset with the -t applies to all file nodes (including
189  *	directories).
190  * 3.4	Symbolic links can be followed with -L (optional in the spec).
191  * 3.5	Modification or inode change time ranges can be specified via
192  *	multiple -T options. These allow a user to select files whose
193  *	modification or inode change time lies within a specific time range.
194  * 3.6	Files can be selected based on owner (user name or uid) via one or more
195  *	-U options.
196  * 3.7	Files can be selected based on group (group name or gid) via one o
197  *	more -G options.
198  * 3.8	Symlinks which appear on the command line can be followed (without
199  *	following other symlinks; -H flag)
200  * 3.9  File inode change time can be checked against existing file before
201  *	name modification (-D)
202  * 3.10 File inode change time can be checked against existing file after
203  *	name modification (-Y)
204  * 3.11	File modification time can be checked against existing file after
205  *	name modification (-Z)
206  *
207  * 4	GENERAL ENHANCEMENTS
208  * 4.1	Internal structure is designed to isolate format dependent and
209  *	independent functions. Formats are selected via a format driver table.
210  *	This encourages the addition of new archive formats by only having to
211  *	write those routines which id, read and write the archive header.
212  */
213 
214 /*
215  * main()
216  *	parse options, set up and operate as specified by the user.
217  *	any operational flaw will set exit_val to non-zero
218  * Return: 0 if ok, 1 otherwise
219  */
220 
221 int
222 main(int argc, char **argv)
223 {
224 	char *tmpdir;
225 	size_t tdlen;
226 
227 	/*
228 	 * Keep a reference to cwd, so we can always come back home.
229 	 */
230 	cwdfd = open(".", O_RDONLY | O_CLOEXEC);
231 	if (cwdfd == -1) {
232 		syswarn(1, errno, "Can't open current working directory.");
233 		return(exit_val);
234 	}
235 
236 	/*
237 	 * Where should we put temporary files?
238 	 */
239 	if ((tmpdir = getenv("TMPDIR")) == NULL || *tmpdir == '\0')
240 		tmpdir = _PATH_TMP;
241 	tdlen = strlen(tmpdir);
242 	while (tdlen > 0 && tmpdir[tdlen - 1] == '/')
243 		tdlen--;
244 	tempfile = malloc(tdlen + 1 + sizeof(_TFILE_BASE));
245 	if (tempfile == NULL) {
246 		paxwarn(1, "Cannot allocate memory for temp file name.");
247 		return(exit_val);
248 	}
249 	if (tdlen)
250 		memcpy(tempfile, tmpdir, tdlen);
251 	tempbase = tempfile + tdlen;
252 	*tempbase++ = '/';
253 
254 	/*
255 	 * keep passwd and group files open for faster lookups.
256 	 */
257 	setpassent(1);
258 	setgroupent(1);
259 
260 	/*
261 	 * parse options, determine operational mode, general init
262 	 */
263 	options(argc, argv);
264 	if ((gen_init() < 0) || (tty_init() < 0))
265 		return(exit_val);
266 
267 	/*
268 	 * pmode needs to restore setugid bits when extracting or copying,
269 	 * so can't pledge at all then.
270 	 */
271 	if (pmode == 0 || (act != EXTRACT && act != COPY)) {
272 		if (pledge("stdio rpath wpath cpath fattr dpath getpw proc exec tape",
273 		    NULL) == -1)
274 			err(1, "pledge");
275 
276 		/* Copy mode, or no gzip -- don't need to fork/exec. */
277 		if (gzip_program == NULL || act == COPY) {
278 			if (pledge("stdio rpath wpath cpath fattr dpath getpw tape",
279 			    NULL) == -1)
280 				err(1, "pledge");
281 		}
282 	}
283 
284 	/*
285 	 * select a primary operation mode
286 	 */
287 	switch (act) {
288 	case EXTRACT:
289 		extract();
290 		break;
291 	case ARCHIVE:
292 		archive();
293 		break;
294 	case APPND:
295 		if (gzip_program != NULL)
296 			errx(1, "can not gzip while appending");
297 		append();
298 		break;
299 	case COPY:
300 		copy();
301 		break;
302 	default:
303 	case LIST:
304 		list();
305 		break;
306 	}
307 	return(exit_val);
308 }
309 
310 /*
311  * sig_cleanup()
312  *	when interrupted we try to do whatever delayed processing we can.
313  *	This is not critical, but we really ought to limit our damage when we
314  *	are aborted by the user.
315  * Return:
316  *	never....
317  */
318 
319 void
320 sig_cleanup(int which_sig)
321 {
322 	/*
323 	 * restore modes and times for any dirs we may have created
324 	 * or any dirs we may have read.
325 	 */
326 
327 	/* paxwarn() uses stdio; fake it as well as we can */
328 	if (which_sig == SIGXCPU)
329 		dprintf(STDERR_FILENO, "\nCPU time limit reached, cleaning up.\n");
330 	else
331 		dprintf(STDERR_FILENO, "\nSignal caught, cleaning up.\n");
332 
333 	ar_close(1);
334 	sltab_process(1);
335 	proc_dir(1);
336 	if (tflag)
337 		atdir_end();
338 	_exit(1);
339 }
340 
341 /*
342  * setup_sig()
343  *	set a signal to be caught, but only if it isn't being ignored already
344  */
345 
346 static int
347 setup_sig(int sig, const struct sigaction *n_hand)
348 {
349 	struct sigaction o_hand;
350 
351 	if (sigaction(sig, NULL, &o_hand) == -1)
352 		return (-1);
353 
354 	if (o_hand.sa_handler == SIG_IGN)
355 		return (0);
356 
357 	return (sigaction(sig, n_hand, NULL));
358 }
359 
360 /*
361  * gen_init()
362  *	general setup routines. Not all are required, but they really help
363  *	when dealing with a medium to large sized archives.
364  */
365 
366 static int
367 gen_init(void)
368 {
369 	struct rlimit reslimit;
370 	struct sigaction n_hand;
371 
372 	/*
373 	 * Really needed to handle large archives. We can run out of memory for
374 	 * internal tables really fast when we have a whole lot of files...
375 	 */
376 	if (getrlimit(RLIMIT_DATA , &reslimit) == 0){
377 		reslimit.rlim_cur = reslimit.rlim_max;
378 		(void)setrlimit(RLIMIT_DATA , &reslimit);
379 	}
380 
381 	/*
382 	 * should file size limits be waived? if the os limits us, this is
383 	 * needed if we want to write a large archive
384 	 */
385 	if (getrlimit(RLIMIT_FSIZE , &reslimit) == 0){
386 		reslimit.rlim_cur = reslimit.rlim_max;
387 		(void)setrlimit(RLIMIT_FSIZE , &reslimit);
388 	}
389 
390 	/*
391 	 * increase the size the stack can grow to
392 	 */
393 	if (getrlimit(RLIMIT_STACK , &reslimit) == 0){
394 		reslimit.rlim_cur = reslimit.rlim_max;
395 		(void)setrlimit(RLIMIT_STACK , &reslimit);
396 	}
397 
398 	/*
399 	 * not really needed, but doesn't hurt
400 	 */
401 	if (getrlimit(RLIMIT_RSS , &reslimit) == 0){
402 		reslimit.rlim_cur = reslimit.rlim_max;
403 		(void)setrlimit(RLIMIT_RSS , &reslimit);
404 	}
405 
406 	/*
407 	 * signal handling to reset stored directory times and modes. Since
408 	 * we deal with broken pipes via failed writes we ignore it. We also
409 	 * deal with any file size limit through failed writes. Cpu time
410 	 * limits are caught and a cleanup is forced.
411 	 */
412 	if ((sigemptyset(&s_mask) < 0) || (sigaddset(&s_mask, SIGTERM) < 0) ||
413 	    (sigaddset(&s_mask,SIGINT) < 0)||(sigaddset(&s_mask,SIGHUP) < 0) ||
414 	    (sigaddset(&s_mask,SIGPIPE) < 0)||(sigaddset(&s_mask,SIGQUIT)<0) ||
415 	    (sigaddset(&s_mask,SIGXCPU) < 0)||(sigaddset(&s_mask,SIGXFSZ)<0)) {
416 		paxwarn(1, "Unable to set up signal mask");
417 		return(-1);
418 	}
419 
420 	/* snag the fd to be used from the signal handler */
421 	listfd = fileno(listf);
422 
423 	memset(&n_hand, 0, sizeof n_hand);
424 	n_hand.sa_mask = s_mask;
425 	n_hand.sa_flags = 0;
426 	n_hand.sa_handler = sig_cleanup;
427 
428 	if (setup_sig(SIGHUP,  &n_hand) ||
429 	    setup_sig(SIGTERM, &n_hand) ||
430 	    setup_sig(SIGINT,  &n_hand) ||
431 	    setup_sig(SIGQUIT, &n_hand) ||
432 	    setup_sig(SIGXCPU, &n_hand))
433 		goto out;
434 
435 	n_hand.sa_handler = SIG_IGN;
436 	if ((sigaction(SIGPIPE, &n_hand, NULL) == -1) ||
437 	    (sigaction(SIGXFSZ, &n_hand, NULL) == -1))
438 		goto out;
439 	return(0);
440 
441     out:
442 	syswarn(1, errno, "Unable to set up signal handler");
443 	return(-1);
444 }
445