GCC Code Coverage Report
Directory: ./ Exec Total Coverage
File: bin/pax/file_subs.c Lines: 90 335 26.9 %
Date: 2016-12-06 Branches: 56 296 18.9 %

Line Branch Exec Source
1
/*	$OpenBSD: file_subs.c,v 1.48 2016/02/16 04:30:07 guenther Exp $	*/
2
/*	$NetBSD: file_subs.c,v 1.4 1995/03/21 09:07:18 cgd 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/time.h>
38
#include <sys/stat.h>
39
#include <err.h>
40
#include <errno.h>
41
#include <fcntl.h>
42
#include <limits.h>
43
#include <stdio.h>
44
#include <stdlib.h>
45
#include <string.h>
46
#include <unistd.h>
47
#include "pax.h"
48
#include "options.h"
49
#include "extern.h"
50
51
static int
52
mk_link(char *, struct stat *, char *, int);
53
54
/*
55
 * routines that deal with file operations such as: creating, removing;
56
 * and setting access modes, uid/gid and times of files
57
 */
58
59
/*
60
 * file_creat()
61
 *	Create and open a file.
62
 * Return:
63
 *	file descriptor or -1 for failure
64
 */
65
66
int
67
file_creat(ARCHD *arcn)
68
6
{
69
6
	int fd = -1;
70
	mode_t file_mode;
71
	int oerrno;
72
73
	/*
74
	 * Assume file doesn't exist, so just try to create it, most times this
75
	 * works. We have to take special handling when the file does exist. To
76
	 * detect this, we use O_EXCL. For example when trying to create a
77
	 * file and a character device or fifo exists with the same name, we
78
	 * can accidently open the device by mistake (or block waiting to open).
79
	 * If we find that the open has failed, then spend the effort to
80
	 * figure out why. This strategy was found to have better average
81
	 * performance in common use than checking the file (and the path)
82
	 * first with lstat.
83
	 */
84
6
	file_mode = arcn->sb.st_mode & FILEBITS;
85
6
	if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_EXCL,
86
	    file_mode)) >= 0)
87
6
		return(fd);
88
89
	/*
90
	 * the file seems to exist. First we try to get rid of it (found to be
91
	 * the second most common failure when traced). If this fails, only
92
	 * then we go to the expense to check and create the path to the file
93
	 */
94
	if (unlnk_exist(arcn->name, arcn->type) != 0)
95
		return(-1);
96
97
	for (;;) {
98
		/*
99
		 * try to open it again, if this fails, check all the nodes in
100
		 * the path and give it a final try. if chk_path() finds that
101
		 * it cannot fix anything, we will skip the last attempt
102
		 */
103
		if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
104
		    file_mode)) >= 0)
105
			break;
106
		oerrno = errno;
107
		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
108
			syswarn(1, oerrno, "Unable to create %s", arcn->name);
109
			return(-1);
110
		}
111
	}
112
	return(fd);
113
}
114
115
/*
116
 * file_close()
117
 *	Close file descriptor to a file just created by pax. Sets modes,
118
 *	ownership and times as required.
119
 * Return:
120
 *	0 for success, -1 for failure
121
 */
122
123
void
124
file_close(ARCHD *arcn, int fd)
125
6
{
126
6
	int res = 0;
127
128
6
	if (fd < 0)
129
		return;
130
131
	/*
132
	 * set owner/groups first as this may strip off mode bits we want
133
	 * then set file permission modes. Then set file access and
134
	 * modification times.
135
	 */
136
6
	if (pids)
137
		res = fset_ids(arcn->name, fd, arcn->sb.st_uid,
138
		    arcn->sb.st_gid);
139
140
	/*
141
	 * IMPORTANT SECURITY NOTE:
142
	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
143
	 * set uid/gid bits
144
	 */
145
6
	if (!pmode || res)
146
6
		arcn->sb.st_mode &= ~(SETBITS);
147
6
	if (pmode)
148
		fset_pmode(arcn->name, fd, arcn->sb.st_mode);
149

6
	if (patime || pmtime)
150
6
		fset_ftime(arcn->name, fd, &arcn->sb.st_mtim,
151
		    &arcn->sb.st_atim, 0);
152
6
	if (close(fd) < 0)
153
		syswarn(0, errno, "Unable to close file descriptor on %s",
154
		    arcn->name);
155
}
156
157
/*
158
 * lnk_creat()
159
 *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
160
 *	must exist;
161
 * Return:
162
 *	0 if ok, -1 otherwise
163
 */
164
165
int
166
lnk_creat(ARCHD *arcn)
167
4
{
168
	struct stat sb;
169
	int res;
170
171
	/*
172
	 * we may be running as root, so we have to be sure that link target
173
	 * is not a directory, so we lstat and check
174
	 */
175
4
	if (lstat(arcn->ln_name, &sb) < 0) {
176
		syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
177
		    arcn->name);
178
		return(-1);
179
	}
180
181
4
	if (S_ISDIR(sb.st_mode)) {
182
		paxwarn(1, "A hard link to the directory %s is not allowed",
183
		    arcn->ln_name);
184
		return(-1);
185
	}
186
187
4
	res = mk_link(arcn->ln_name, &sb, arcn->name, 0);
188
4
	if (res == 0) {
189
		/* check for a hardlink to a placeholder symlink */
190
4
		res = sltab_add_link(arcn->name, &sb);
191
192
4
		if (res < 0) {
193
			/* arrgh, it failed, clean up */
194
			unlink(arcn->name);
195
		}
196
	}
197
198
4
	return (res);
199
}
200
201
/*
202
 * cross_lnk()
203
 *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
204
 *	with the -l flag. No warning or error if this does not succeed (we will
205
 *	then just create the file)
206
 * Return:
207
 *	1 if copy() should try to create this file node
208
 *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
209
 */
210
211
int
212
cross_lnk(ARCHD *arcn)
213
{
214
	/*
215
	 * try to make a link to original file (-l flag in copy mode). make
216
	 * sure we do not try to link to directories in case we are running as
217
	 * root (and it might succeed).
218
	 */
219
	if (arcn->type == PAX_DIR)
220
		return(1);
221
	return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
222
}
223
224
/*
225
 * chk_same()
226
 *	In copy mode if we are not trying to make hard links between the src
227
 *	and destinations, make sure we are not going to overwrite ourselves by
228
 *	accident. This slows things down a little, but we have to protect all
229
 *	those people who make typing errors.
230
 * Return:
231
 *	1 the target does not exist, go ahead and copy
232
 *	0 skip it file exists (-k) or may be the same as source file
233
 */
234
235
int
236
chk_same(ARCHD *arcn)
237
{
238
	struct stat sb;
239
240
	/*
241
	 * if file does not exist, return. if file exists and -k, skip it
242
	 * quietly
243
	 */
244
	if (lstat(arcn->name, &sb) < 0)
245
		return(1);
246
	if (kflag)
247
		return(0);
248
249
	/*
250
	 * better make sure the user does not have src == dest by mistake
251
	 */
252
	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
253
		paxwarn(1, "Unable to copy %s, file would overwrite itself",
254
		    arcn->name);
255
		return(0);
256
	}
257
	return(1);
258
}
259
260
/*
261
 * mk_link()
262
 *	try to make a hard link between two files. if ign set, we do not
263
 *	complain.
264
 * Return:
265
 *	0 if successful (or we are done with this file but no error, such as
266
 *	finding the from file exists and the user has set -k).
267
 *	1 when ign was set to indicates we could not make the link but we
268
 *	should try to copy/extract the file as that might work (and is an
269
 *	allowed option). -1 an error occurred.
270
 */
271
272
static int
273
mk_link(char *to, struct stat *to_sb, char *from, int ign)
274
4
{
275
	struct stat sb;
276
	int oerrno;
277
278
	/*
279
	 * if from file exists, it has to be unlinked to make the link. If the
280
	 * file exists and -k is set, skip it quietly
281
	 */
282
4
	if (lstat(from, &sb) == 0) {
283
		if (kflag)
284
			return(0);
285
286
		/*
287
		 * make sure it is not the same file, protect the user
288
		 */
289
		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
290
			paxwarn(1, "Unable to link file %s to itself", to);
291
			return(-1);
292
		}
293
294
		/*
295
		 * try to get rid of the file, based on the type
296
		 */
297
		if (S_ISDIR(sb.st_mode)) {
298
			if (rmdir(from) < 0) {
299
				syswarn(1, errno, "Unable to remove %s", from);
300
				return(-1);
301
			}
302
			delete_dir(sb.st_dev, sb.st_ino);
303
		} else if (unlink(from) < 0) {
304
			if (!ign) {
305
				syswarn(1, errno, "Unable to remove %s", from);
306
				return(-1);
307
			}
308
			return(1);
309
		}
310
	}
311
312
	/*
313
	 * from file is gone (or did not exist), try to make the hard link.
314
	 * if it fails, check the path and try it again (if chk_path() says to
315
	 * try again)
316
	 */
317
	for (;;) {
318
4
		if (linkat(AT_FDCWD, to, AT_FDCWD, from, 0) == 0)
319
4
			break;
320
		oerrno = errno;
321
		if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
322
			continue;
323
		if (!ign) {
324
			syswarn(1, oerrno, "Could not link to %s from %s", to,
325
			    from);
326
			return(-1);
327
		}
328
		return(1);
329
	}
330
331
	/*
332
	 * all right the link was made
333
	 */
334
4
	return(0);
335
}
336
337
/*
338
 * node_creat()
339
 *	create an entry in the file system (other than a file or hard link).
340
 *	If successful, sets uid/gid modes and times as required.
341
 * Return:
342
 *	0 if ok, -1 otherwise
343
 */
344
345
int
346
node_creat(ARCHD *arcn)
347
4
{
348
	int res;
349
4
	int ign = 0;
350
	int oerrno;
351
4
	int pass = 0;
352
	mode_t file_mode;
353
	struct stat sb;
354
	char target[PATH_MAX];
355
4
	char *nm = arcn->name;
356
4
	int len, defer_pmode = 0;
357
358
	/*
359
	 * create node based on type, if that fails try to unlink the node and
360
	 * try again. finally check the path and try again. As noted in the
361
	 * file and link creation routines, this method seems to exhibit the
362
	 * best performance in general use workloads.
363
	 */
364
4
	file_mode = arcn->sb.st_mode & FILEBITS;
365
366
	for (;;) {
367

4
		switch (arcn->type) {
368
		case PAX_DIR:
369
			/*
370
			 * If -h (or -L) was given in tar-mode, follow the
371
			 * potential symlink chain before trying to create the
372
			 * directory.
373
			 */
374

4
			if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
375
				while (lstat(nm, &sb) == 0 &&
376
				    S_ISLNK(sb.st_mode)) {
377
					len = readlink(nm, target,
378
					    sizeof target - 1);
379
					if (len == -1) {
380
						syswarn(0, errno,
381
						   "cannot follow symlink %s in chain for %s",
382
						    nm, arcn->name);
383
						res = -1;
384
						goto badlink;
385
					}
386
					target[len] = '\0';
387
					nm = target;
388
				}
389
			}
390
4
			res = mkdir(nm, file_mode);
391
392
4
badlink:
393
4
			if (ign)
394
				res = 0;
395
			break;
396
		case PAX_CHR:
397
			file_mode |= S_IFCHR;
398
			res = mknod(nm, file_mode, arcn->sb.st_rdev);
399
			break;
400
		case PAX_BLK:
401
			file_mode |= S_IFBLK;
402
			res = mknod(nm, file_mode, arcn->sb.st_rdev);
403
			break;
404
		case PAX_FIF:
405
			res = mkfifo(nm, file_mode);
406
			break;
407
		case PAX_SCK:
408
			/*
409
			 * Skip sockets, operation has no meaning under BSD
410
			 */
411
			paxwarn(0,
412
			    "%s skipped. Sockets cannot be copied or extracted",
413
			    nm);
414
			return(-1);
415
		case PAX_SLK:
416
			if (arcn->ln_name[0] != '/' &&
417
			    !has_dotdot(arcn->ln_name))
418
				res = symlink(arcn->ln_name, nm);
419
			else {
420
				/*
421
				 * absolute symlinks and symlinks with ".."
422
				 * have to be deferred to prevent the archive
423
				 * from bootstrapping itself to outside the
424
				 * working directory.
425
				 */
426
				res = sltab_add_sym(nm, arcn->ln_name,
427
				    arcn->sb.st_mode);
428
				if (res == 0)
429
					defer_pmode = 1;
430
			}
431
			break;
432
		case PAX_CTG:
433
		case PAX_HLK:
434
		case PAX_HRG:
435
		case PAX_REG:
436
		default:
437
			/*
438
			 * we should never get here
439
			 */
440
			paxwarn(0, "%s has an unknown file type, skipping",
441
				nm);
442
			return(-1);
443
		}
444
445
		/*
446
		 * if we were able to create the node break out of the loop,
447
		 * otherwise try to unlink the node and try again. if that
448
		 * fails check the full path and try a final time.
449
		 */
450
4
		if (res == 0)
451
4
			break;
452
453
		/*
454
		 * we failed to make the node
455
		 */
456
		oerrno = errno;
457
		if ((ign = unlnk_exist(nm, arcn->type)) < 0)
458
			return(-1);
459
460
		if (++pass <= 1)
461
			continue;
462
463
		if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
464
			syswarn(1, oerrno, "Could not create: %s", nm);
465
			return(-1);
466
		}
467
	}
468
469
	/*
470
	 * we were able to create the node. set uid/gid, modes and times
471
	 */
472
4
	if (pids)
473
		res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
474
	else
475
4
		res = 0;
476
477
	/*
478
	 * IMPORTANT SECURITY NOTE:
479
	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
480
	 * set uid/gid bits
481
	 */
482
4
	if (!pmode || res)
483
4
		arcn->sb.st_mode &= ~(SETBITS);
484
4
	if (pmode && !defer_pmode)
485
		set_pmode(nm, arcn->sb.st_mode);
486
487

4
	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
488
		/*
489
		 * Dirs must be processed again at end of extract to set times
490
		 * and modes to agree with those stored in the archive. However
491
		 * to allow extract to continue, we may have to also set owner
492
		 * rights. This allows nodes in the archive that are children
493
		 * of this directory to be extracted without failure. Both time
494
		 * and modes will be fixed after the entire archive is read and
495
		 * before pax exits.  To do that safely, we want the dev+ino
496
		 * of the directory we created.
497
		 */
498
4
		if (lstat(nm, &sb) < 0) {
499
			syswarn(0, errno,"Could not access %s (stat)", nm);
500
4
		} else if (access(nm, R_OK | W_OK | X_OK) < 0) {
501
			/*
502
			 * We have to add rights to the dir, so we make
503
			 * sure to restore the mode. The mode must be
504
			 * restored AS CREATED and not as stored if
505
			 * pmode is not set.
506
			 */
507
			set_pmode(nm,
508
			    ((sb.st_mode & FILEBITS) | S_IRWXU));
509
			if (!pmode)
510
				arcn->sb.st_mode = sb.st_mode;
511
512
			/*
513
			 * we have to force the mode to what was set
514
			 * here, since we changed it from the default
515
			 * as created.
516
			 */
517
			arcn->sb.st_dev = sb.st_dev;
518
			arcn->sb.st_ino = sb.st_ino;
519
			add_dir(nm, &(arcn->sb), 1);
520

4
		} else if (pmode || patime || pmtime) {
521
4
			arcn->sb.st_dev = sb.st_dev;
522
4
			arcn->sb.st_ino = sb.st_ino;
523
4
			add_dir(nm, &(arcn->sb), 0);
524
		}
525
	}
526
527

4
	if (patime || pmtime)
528
4
		set_ftime(nm, &arcn->sb.st_mtim, &arcn->sb.st_atim, 0);
529
4
	return(0);
530
}
531
532
/*
533
 * unlnk_exist()
534
 *	Remove node from file system with the specified name. We pass the type
535
 *	of the node that is going to replace it. When we try to create a
536
 *	directory and find that it already exists, we allow processing to
537
 *	continue as proper modes etc will always be set for it later on.
538
 * Return:
539
 *	0 is ok to proceed, no file with the specified name exists
540
 *	-1 we were unable to remove the node, or we should not remove it (-k)
541
 *	1 we found a directory and we were going to create a directory.
542
 */
543
544
int
545
unlnk_exist(char *name, int type)
546
{
547
	struct stat sb;
548
549
	/*
550
	 * the file does not exist, or -k we are done
551
	 */
552
	if (lstat(name, &sb) < 0)
553
		return(0);
554
	if (kflag)
555
		return(-1);
556
557
	if (S_ISDIR(sb.st_mode)) {
558
		/*
559
		 * try to remove a directory, if it fails and we were going to
560
		 * create a directory anyway, tell the caller (return a 1)
561
		 */
562
		if (rmdir(name) < 0) {
563
			if (type == PAX_DIR)
564
				return(1);
565
			syswarn(1,errno,"Unable to remove directory %s", name);
566
			return(-1);
567
		}
568
		delete_dir(sb.st_dev, sb.st_ino);
569
		return(0);
570
	}
571
572
	/*
573
	 * try to get rid of all non-directory type nodes
574
	 */
575
	if (unlink(name) < 0) {
576
		syswarn(1, errno, "Could not unlink %s", name);
577
		return(-1);
578
	}
579
	return(0);
580
}
581
582
/*
583
 * chk_path()
584
 *	We were trying to create some kind of node in the file system and it
585
 *	failed. chk_path() makes sure the path up to the node exists and is
586
 *	writeable. When we have to create a directory that is missing along the
587
 *	path somewhere, the directory we create will be set to the same
588
 *	uid/gid as the file has (when uid and gid are being preserved).
589
 *	NOTE: this routine is a real performance loss. It is only used as a
590
 *	last resort when trying to create entries in the file system.
591
 * Return:
592
 *	-1 when it could find nothing it is allowed to fix.
593
 *	0 otherwise
594
 */
595
596
int
597
chk_path(char *name, uid_t st_uid, gid_t st_gid)
598
{
599
	char *spt = name;
600
	char *next;
601
	struct stat sb;
602
	int retval = -1;
603
604
	/*
605
	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
606
	 */
607
	while (*spt == '/')
608
		++spt;
609
610
	for (;;) {
611
		/*
612
		 * work forward from the first / and check each part of the path
613
		 */
614
		spt = strchr(spt, '/');
615
		if (spt == NULL)
616
			break;
617
618
		/*
619
		 * skip over duplicate slashes; stop if there're only
620
		 * trailing slashes left
621
		 */
622
		next = spt + 1;
623
		while (*next == '/')
624
			next++;
625
		if (*next == '\0')
626
			break;
627
628
		*spt = '\0';
629
630
		/*
631
		 * if it exists we assume it is a directory, it is not within
632
		 * the spec (at least it seems to read that way) to alter the
633
		 * file system for nodes NOT EXPLICITLY stored on the archive.
634
		 * If that assumption is changed, you would test the node here
635
		 * and figure out how to get rid of it (probably like some
636
		 * recursive unlink()) or fix up the directory permissions if
637
		 * required (do an access()).
638
		 */
639
		if (lstat(name, &sb) == 0) {
640
			*spt = '/';
641
			spt = next;
642
			continue;
643
		}
644
645
		/*
646
		 * the path fails at this point, see if we can create the
647
		 * needed directory and continue on
648
		 */
649
		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
650
			*spt = '/';
651
			retval = -1;
652
			break;
653
		}
654
655
		/*
656
		 * we were able to create the directory. We will tell the
657
		 * caller that we found something to fix, and it is ok to try
658
		 * and create the node again.
659
		 */
660
		retval = 0;
661
		if (pids)
662
			(void)set_ids(name, st_uid, st_gid);
663
664
		/*
665
		 * make sure the user doesn't have some strange umask that
666
		 * causes this newly created directory to be unusable. We fix
667
		 * the modes and restore them back to the creation default at
668
		 * the end of pax
669
		 */
670
		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
671
		    (lstat(name, &sb) == 0)) {
672
			set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
673
			add_dir(name, &sb, 1);
674
		}
675
		*spt = '/';
676
		spt = next;
677
		continue;
678
	}
679
	return(retval);
680
}
681
682
/*
683
 * set_ftime()
684
 *	Set the access time and modification time for a named file. If frc
685
 *	is non-zero we force these times to be set even if the user did not
686
 *	request access and/or modification time preservation (this is also
687
 *	used by -t to reset access times).
688
 *	When ign is zero, only those times the user has asked for are set, the
689
 *	other ones are left alone.
690
 */
691
692
void
693
set_ftime(const char *fnm, const struct timespec *mtimp,
694
    const struct timespec *atimp, int frc)
695
4
{
696
	struct timespec tv[2];
697
698
4
	tv[0] = *atimp;
699
4
	tv[1] = *mtimp;
700
701
4
	if (!frc) {
702
		/*
703
		 * if we are not forcing, only set those times the user wants
704
		 * set.
705
		 */
706
4
		if (!patime)
707
			tv[0].tv_nsec = UTIME_OMIT;
708
4
		if (!pmtime)
709
			tv[1].tv_nsec = UTIME_OMIT;
710
	}
711
712
	/*
713
	 * set the times
714
	 */
715
4
	if (utimensat(AT_FDCWD, fnm, tv, AT_SYMLINK_NOFOLLOW) < 0)
716
		syswarn(1, errno, "Access/modification time set failed on: %s",
717
		    fnm);
718
4
}
719
720
void
721
fset_ftime(const char *fnm, int fd, const struct timespec *mtimp,
722
    const struct timespec *atimp, int frc)
723
8
{
724
	struct timespec tv[2];
725
726
727
8
	tv[0] = *atimp;
728
8
	tv[1] = *mtimp;
729
730
8
	if (!frc) {
731
		/*
732
		 * if we are not forcing, only set those times the user wants
733
		 * set.
734
		 */
735
8
		if (!patime)
736
			tv[0].tv_nsec = UTIME_OMIT;
737
8
		if (!pmtime)
738
			tv[1].tv_nsec = UTIME_OMIT;
739
	}
740
	/*
741
	 * set the times
742
	 */
743
8
	if (futimens(fd, tv) < 0)
744
		syswarn(1, errno, "Access/modification time set failed on: %s",
745
		    fnm);
746
8
}
747
748
/*
749
 * set_ids()
750
 *	set the uid and gid of a file system node
751
 * Return:
752
 *	0 when set, -1 on failure
753
 */
754
755
int
756
set_ids(char *fnm, uid_t uid, gid_t gid)
757
{
758
	if (fchownat(AT_FDCWD, fnm, uid, gid, AT_SYMLINK_NOFOLLOW) < 0) {
759
		/*
760
		 * ignore EPERM unless in verbose mode or being run by root.
761
		 * if running as pax, POSIX requires a warning.
762
		 */
763
		if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
764
		    geteuid() == 0)
765
			syswarn(1, errno, "Unable to set file uid/gid of %s",
766
			    fnm);
767
		return(-1);
768
	}
769
	return(0);
770
}
771
772
int
773
fset_ids(char *fnm, int fd, uid_t uid, gid_t gid)
774
{
775
	if (fchown(fd, uid, gid) < 0) {
776
		/*
777
		 * ignore EPERM unless in verbose mode or being run by root.
778
		 * if running as pax, POSIX requires a warning.
779
		 */
780
		if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
781
		    geteuid() == 0)
782
			syswarn(1, errno, "Unable to set file uid/gid of %s",
783
			    fnm);
784
		return(-1);
785
	}
786
	return(0);
787
}
788
789
/*
790
 * set_pmode()
791
 *	Set file access mode
792
 */
793
794
void
795
set_pmode(char *fnm, mode_t mode)
796
{
797
	mode &= ABITS;
798
	if (fchmodat(AT_FDCWD, fnm, mode, AT_SYMLINK_NOFOLLOW) < 0)
799
		syswarn(1, errno, "Could not set permissions on %s", fnm);
800
}
801
802
void
803
fset_pmode(char *fnm, int fd, mode_t mode)
804
{
805
	mode &= ABITS;
806
	if (fchmod(fd, mode) < 0)
807
		syswarn(1, errno, "Could not set permissions on %s", fnm);
808
}
809
810
/*
811
 * set_attr()
812
 *	Given a DIRDATA, restore the mode and times as indicated, but
813
 *	only after verifying that it's the directory that we wanted.
814
 */
815
int
816
set_attr(const struct file_times *ft, int force_times, mode_t mode,
817
    int do_mode, int in_sig)
818
4
{
819
	struct stat sb;
820
	int fd, r;
821
822

4
	if (!do_mode && !force_times && !patime && !pmtime)
823
		return (0);
824
825
	/*
826
	 * We could legitimately go through a symlink here,
827
	 * so do *not* use O_NOFOLLOW.  The dev+ino check will
828
	 * protect us from evil.
829
	 */
830
4
	fd = open(ft->ft_name, O_RDONLY | O_DIRECTORY);
831
4
	if (fd == -1) {
832
		if (!in_sig)
833
			syswarn(1, errno, "Unable to restore mode and times"
834
			    " for directory: %s", ft->ft_name);
835
		return (-1);
836
	}
837
838
4
	if (fstat(fd, &sb) == -1) {
839
		if (!in_sig)
840
			syswarn(1, errno, "Unable to stat directory: %s",
841
			    ft->ft_name);
842
		r = -1;
843

4
	} else if (ft->ft_ino != sb.st_ino || ft->ft_dev != sb.st_dev) {
844
		if (!in_sig)
845
			paxwarn(1, "Directory vanished before restoring"
846
			    " mode and times: %s", ft->ft_name);
847
		r = -1;
848
	} else {
849
		/* Whew, it's a match!  Is there anything to change? */
850

4
		if (do_mode && (mode & ABITS) != (sb.st_mode & ABITS))
851
			fset_pmode(ft->ft_name, fd, mode);
852





4
		if (((force_times || patime) &&
853
		    timespeccmp(&ft->ft_atim, &sb.st_atim, !=)) ||
854
		    ((force_times || pmtime) &&
855
		    timespeccmp(&ft->ft_mtim, &sb.st_mtim, !=)))
856
2
			fset_ftime(ft->ft_name, fd, &ft->ft_mtim,
857
			    &ft->ft_atim, force_times);
858
4
		r = 0;
859
	}
860
4
	close(fd);
861
862
4
	return (r);
863
}
864
865
866
/*
867
 * file_write()
868
 *	Write/copy a file (during copy or archive extract). This routine knows
869
 *	how to copy files with lseek holes in it. (Which are read as file
870
 *	blocks containing all 0's but do not have any file blocks associated
871
 *	with the data). Typical examples of these are files created by dbm
872
 *	variants (.pag files). While the file size of these files are huge, the
873
 *	actual storage is quite small (the files are sparse). The problem is
874
 *	the holes read as all zeros so are probably stored on the archive that
875
 *	way (there is no way to determine if the file block is really a hole,
876
 *	we only know that a file block of all zero's can be a hole).
877
 *	At this writing, no major archive format knows how to archive files
878
 *	with holes. However, on extraction (or during copy, -rw) we have to
879
 *	deal with these files. Without detecting the holes, the files can
880
 *	consume a lot of file space if just written to disk. This replacement
881
 *	for write when passed the basic allocation size of a file system block,
882
 *	uses lseek whenever it detects the input data is all 0 within that
883
 *	file block. In more detail, the strategy is as follows:
884
 *	While the input is all zero keep doing an lseek. Keep track of when we
885
 *	pass over file block boundaries. Only write when we hit a non zero
886
 *	input. once we have written a file block, we continue to write it to
887
 *	the end (we stop looking at the input). When we reach the start of the
888
 *	next file block, start checking for zero blocks again. Working on file
889
 *	block boundaries significantly reduces the overhead when copying files
890
 *	that are NOT very sparse. This overhead (when compared to a write) is
891
 *	almost below the measurement resolution on many systems. Without it,
892
 *	files with holes cannot be safely copied. It does has a side effect as
893
 *	it can put holes into files that did not have them before, but that is
894
 *	not a problem since the file contents are unchanged (in fact it saves
895
 *	file space). (Except on paging files for diskless clients. But since we
896
 *	cannot determine one of those file from here, we ignore them). If this
897
 *	ever ends up on a system where CTG files are supported and the holes
898
 *	are not desired, just do a conditional test in those routines that
899
 *	call file_write() and have it call write() instead. BEFORE CLOSING THE
900
 *	FILE, make sure to call file_flush() when the last write finishes with
901
 *	an empty block. A lot of file systems will not create an lseek hole at
902
 *	the end. In this case we drop a single 0 at the end to force the
903
 *	trailing 0's in the file.
904
 *	---Parameters---
905
 *	rem: how many bytes left in this file system block
906
 *	isempt: have we written to the file block yet (is it empty)
907
 *	sz: basic file block allocation size
908
 *	cnt: number of bytes on this write
909
 *	str: buffer to write
910
 * Return:
911
 *	number of bytes written, -1 on write (or lseek) error.
912
 */
913
914
int
915
file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
916
	char *name)
917
{
918
	char *pt;
919
	char *end;
920
	int wcnt;
921
	char *st = str;
922
	char **strp;
923
924
	/*
925
	 * while we have data to process
926
	 */
927
	while (cnt) {
928
		if (!*rem) {
929
			/*
930
			 * We are now at the start of file system block again
931
			 * (or what we think one is...). start looking for
932
			 * empty blocks again
933
			 */
934
			*isempt = 1;
935
			*rem = sz;
936
		}
937
938
		/*
939
		 * only examine up to the end of the current file block or
940
		 * remaining characters to write, whatever is smaller
941
		 */
942
		wcnt = MINIMUM(cnt, *rem);
943
		cnt -= wcnt;
944
		*rem -= wcnt;
945
		if (*isempt) {
946
			/*
947
			 * have not written to this block yet, so we keep
948
			 * looking for zero's
949
			 */
950
			pt = st;
951
			end = st + wcnt;
952
953
			/*
954
			 * look for a zero filled buffer
955
			 */
956
			while ((pt < end) && (*pt == '\0'))
957
				++pt;
958
959
			if (pt == end) {
960
				/*
961
				 * skip, buf is empty so far
962
				 */
963
				if (fd > -1 &&
964
				    lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
965
					syswarn(1,errno,"File seek on %s",
966
					    name);
967
					return(-1);
968
				}
969
				st = pt;
970
				continue;
971
			}
972
			/*
973
			 * drat, the buf is not zero filled
974
			 */
975
			*isempt = 0;
976
		}
977
978
		/*
979
		 * have non-zero data in this file system block, have to write
980
		 */
981
		switch (fd) {
982
		case -1:
983
			strp = &gnu_name_string;
984
			break;
985
		case -2:
986
			strp = &gnu_link_string;
987
			break;
988
		default:
989
			strp = NULL;
990
			break;
991
		}
992
		if (strp) {
993
			if (*strp)
994
				err(1, "WARNING! Major Internal Error! GNU hack Failing!");
995
			*strp = malloc(wcnt + 1);
996
			if (*strp == NULL) {
997
				paxwarn(1, "Out of memory");
998
				return(-1);
999
			}
1000
			memcpy(*strp, st, wcnt);
1001
			(*strp)[wcnt] = '\0';
1002
			break;
1003
		} else if (write(fd, st, wcnt) != wcnt) {
1004
			syswarn(1, errno, "Failed write to file %s", name);
1005
			return(-1);
1006
		}
1007
		st += wcnt;
1008
	}
1009
	return(st - str);
1010
}
1011
1012
/*
1013
 * file_flush()
1014
 *	when the last file block in a file is zero, many file systems will not
1015
 *	let us create a hole at the end. To get the last block with zeros, we
1016
 *	write the last BYTE with a zero (back up one byte and write a zero).
1017
 */
1018
1019
void
1020
file_flush(int fd, char *fname, int isempt)
1021
{
1022
	static char blnk[] = "\0";
1023
1024
	/*
1025
	 * silly test, but make sure we are only called when the last block is
1026
	 * filled with all zeros.
1027
	 */
1028
	if (!isempt)
1029
		return;
1030
1031
	/*
1032
	 * move back one byte and write a zero
1033
	 */
1034
	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1035
		syswarn(1, errno, "Failed seek on file %s", fname);
1036
		return;
1037
	}
1038
1039
	if (write(fd, blnk, 1) < 0)
1040
		syswarn(1, errno, "Failed write to file %s", fname);
1041
}
1042
1043
/*
1044
 * rdfile_close()
1045
 *	close a file we have been reading (to copy or archive). If we have to
1046
 *	reset access time (tflag) do so (the times are stored in arcn).
1047
 */
1048
1049
void
1050
rdfile_close(ARCHD *arcn, int *fd)
1051
18
{
1052
	/*
1053
	 * make sure the file is open
1054
	 */
1055
18
	if (*fd < 0)
1056
4
		return;
1057
1058
	/*
1059
	 * user wants last access time reset
1060
	 */
1061
14
	if (tflag)
1062
		fset_ftime(arcn->org_name, *fd, &arcn->sb.st_mtim,
1063
		    &arcn->sb.st_atim, 1);
1064
1065
14
	(void)close(*fd);
1066
14
	*fd = -1;
1067
}
1068
1069
/*
1070
 * set_crc()
1071
 *	read a file to calculate its crc. This is a real drag. Archive formats
1072
 *	that have this, end up reading the file twice (we have to write the
1073
 *	header WITH the crc before writing the file contents. Oh well...
1074
 * Return:
1075
 *	0 if was able to calculate the crc, -1 otherwise
1076
 */
1077
1078
int
1079
set_crc(ARCHD *arcn, int fd)
1080
{
1081
	int i;
1082
	int res;
1083
	off_t cpcnt = 0L;
1084
	u_long size;
1085
	u_int32_t crc = 0;
1086
	char tbuf[FILEBLK];
1087
	struct stat sb;
1088
1089
	if (fd < 0) {
1090
		/*
1091
		 * hmm, no fd, should never happen. well no crc then.
1092
		 */
1093
		arcn->crc = 0L;
1094
		return(0);
1095
	}
1096
1097
	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1098
		size = (u_long)sizeof(tbuf);
1099
1100
	/*
1101
	 * read all the bytes we think that there are in the file. If the user
1102
	 * is trying to archive an active file, forget this file.
1103
	 */
1104
	for (;;) {
1105
		if ((res = read(fd, tbuf, size)) <= 0)
1106
			break;
1107
		cpcnt += res;
1108
		for (i = 0; i < res; ++i)
1109
			crc += (tbuf[i] & 0xff);
1110
	}
1111
1112
	/*
1113
	 * safety check. we want to avoid archiving files that are active as
1114
	 * they can create inconsistent archive copies.
1115
	 */
1116
	if (cpcnt != arcn->sb.st_size)
1117
		paxwarn(1, "File changed size %s", arcn->org_name);
1118
	else if (fstat(fd, &sb) < 0)
1119
		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1120
	else if (timespeccmp(&arcn->sb.st_mtim, &sb.st_mtim, !=))
1121
		paxwarn(1, "File %s was modified during read", arcn->org_name);
1122
	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1123
		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1124
	else {
1125
		arcn->crc = crc;
1126
		return(0);
1127
	}
1128
	return(-1);
1129
}