xref: /minix3/bin/pax/file_subs.c (revision b1c4ba4ab66acbca4eb1bab1b49a15e96feef93e)
1 /*	$NetBSD: file_subs.c,v 1.62 2009/04/07 19:52:35 perry Exp $	*/
2 
3 /*-
4  * Copyright (c) 1992 Keith Muller.
5  * Copyright (c) 1992, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Keith Muller of the University of California, San Diego.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  */
35 
36 #if HAVE_NBTOOL_CONFIG_H
37 #include "nbtool_config.h"
38 #endif
39 
40 #include <sys/cdefs.h>
41 #if !defined(lint)
42 #if 0
43 static char sccsid[] = "@(#)file_subs.c	8.1 (Berkeley) 5/31/93";
44 #else
45 __RCSID("$NetBSD: file_subs.c,v 1.62 2009/04/07 19:52:35 perry Exp $");
46 #endif
47 #endif /* not lint */
48 
49 #include <sys/types.h>
50 #include <sys/time.h>
51 #include <sys/stat.h>
52 #include <unistd.h>
53 #include <sys/param.h>
54 #include <fcntl.h>
55 #include <string.h>
56 #include <stdio.h>
57 #include <ctype.h>
58 #include <errno.h>
59 #include <sys/uio.h>
60 #include <stdlib.h>
61 #include "pax.h"
62 #include "extern.h"
63 #include "options.h"
64 
65 #ifdef __minix
66 #include <utime.h>
67 #endif
68 
69 char *xtmp_name;
70 
71 static int
72 mk_link(char *,struct stat *,char *, int);
73 
74 static int warn_broken;
75 
76 /*
77  * routines that deal with file operations such as: creating, removing;
78  * and setting access modes, uid/gid and times of files
79  */
80 #define SET_BITS		(S_ISUID | S_ISGID)
81 #define FILE_BITS		(S_IRWXU | S_IRWXG | S_IRWXO)
82 #define A_BITS			(FILE_BITS | SET_BITS | S_ISVTX)
83 
84 /*
85  * The S_ISVTX (sticky bit) can be set by non-superuser on directories
86  * but not other kinds of files.
87  */
88 #define FILEBITS(dir)		((dir) ? (FILE_BITS | S_ISVTX) : FILE_BITS)
89 #define SETBITS(dir)		((dir) ? SET_BITS : (SET_BITS | S_ISVTX))
90 
91 static mode_t
92 apply_umask(mode_t mode)
93 {
94 	static mode_t cached_umask;
95 	static int cached_umask_valid;
96 
97 	if (!cached_umask_valid) {
98 		cached_umask = umask(0);
99 		umask(cached_umask);
100 		cached_umask_valid = 1;
101 	}
102 
103 	return mode & ~cached_umask;
104 }
105 
106 /*
107  * file_creat()
108  *	Create and open a file.
109  * Return:
110  *	file descriptor or -1 for failure
111  */
112 
113 int
114 file_creat(ARCHD *arcn, int write_to_hardlink)
115 {
116 	int fd = -1;
117 	int oerrno;
118 
119 	/*
120 	 * Some horribly busted tar implementations, have directory nodes
121 	 * that end in a /, but they mark as files. Compensate for that
122 	 * by not creating a directory node at this point, but a file node,
123 	 * and not creating the temp file.
124 	 */
125 	if (arcn->nlen != 0 && arcn->name[arcn->nlen - 1] == '/') {
126 		if (!warn_broken) {
127 			tty_warn(0, "Archive was created with a broken tar;"
128 			    " file `%s' is a directory, but marked as plain.",
129 			    arcn->name);
130 			warn_broken = 1;
131 		}
132 		return -1;
133 	}
134 
135 	/*
136 	 * In "cpio" archives it's usually the last record of a set of
137 	 * hardlinks which includes the contents of the file. We cannot
138 	 * use a tempory file in that case because we couldn't link it
139 	 * with the existing other hardlinks after restoring the contents
140 	 * to it. And it's also useless to create the hardlink under a
141 	 * temporary name because the other hardlinks would have partial
142 	 * contents while restoring.
143 	 */
144 	if (write_to_hardlink)
145 		return (open(arcn->name, O_TRUNC | O_EXCL | O_RDWR, 0));
146 
147 	/*
148 	 * Create a temporary file name so that the file doesn't have partial
149 	 * contents while restoring.
150 	 */
151 	arcn->tmp_name = malloc(arcn->nlen + 8);
152 	if (arcn->tmp_name == NULL) {
153 		syswarn(1, errno, "Cannot malloc %d bytes", arcn->nlen + 8);
154 		return -1;
155 	}
156 	if (xtmp_name != NULL)
157 		abort();
158 	xtmp_name = arcn->tmp_name;
159 
160 	for (;;) {
161 		/*
162 		 * try to create the temporary file we use to restore the
163 		 * contents info.  if this fails, keep checking all the nodes
164 		 * in the path until chk_path() finds that it cannot fix
165 		 * anything further.  if that happens we just give up.
166 		 */
167 #ifdef __minix
168 		{
169 			/* For minix, generate the temporary filename
170 			 * conservatively - just write Xes into the last component.
171 			 * There is space because of the malloc().
172 			 */
173 			int cc;
174 			char *last_slash;
175 			strcpy(arcn->tmp_name, arcn->name);
176 			if(!(last_slash = strrchr(arcn->tmp_name, '/')))
177 				strcpy(arcn->tmp_name, "XXXXXX");
178 			else	strcpy(last_slash+1, "XXXXXX");
179 		}
180 #else
181 		(void)snprintf(arcn->tmp_name, arcn->nlen + 8, "%s.XXXXXX",
182 		    arcn->name);
183 #endif
184 		fd = mkstemp(arcn->tmp_name);
185 		if (fd >= 0)
186 			break;
187 		oerrno = errno;
188 		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
189 			(void)fflush(listf);
190 			syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name);
191 			xtmp_name = NULL;
192 			free(arcn->tmp_name);
193 			arcn->tmp_name = NULL;
194 			return -1;
195 		}
196 	}
197 	return fd;
198 }
199 
200 /*
201  * file_close()
202  *	Close file descriptor to a file just created by pax. Sets modes,
203  *	ownership and times as required.
204  * Return:
205  *	0 for success, -1 for failure
206  */
207 
208 void
209 file_close(ARCHD *arcn, int fd)
210 {
211 	char *tmp_name;
212 	int res;
213 
214 	if (fd < 0)
215 		return;
216 
217 	tmp_name = (arcn->tmp_name != NULL) ? arcn->tmp_name : arcn->name;
218 
219 	if (close(fd) < 0)
220 		syswarn(0, errno, "Cannot close file descriptor on %s",
221 		    tmp_name);
222 
223 	/*
224 	 * set owner/groups first as this may strip off mode bits we want
225 	 * then set file permission modes. Then set file access and
226 	 * modification times.
227 	 */
228 	if (pids)
229 		res = set_ids(tmp_name, arcn->sb.st_uid, arcn->sb.st_gid);
230 	else
231 		res = 0;
232 
233 	/*
234 	 * IMPORTANT SECURITY NOTE:
235 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
236 	 * set uid/gid bits but restore the file modes (since mkstemp doesn't).
237 	 */
238 	if (!pmode || res)
239 		arcn->sb.st_mode &= ~SETBITS(0);
240 	if (pmode)
241 		set_pmode(tmp_name, arcn->sb.st_mode);
242 	else
243 		set_pmode(tmp_name,
244 		    apply_umask((arcn->sb.st_mode & FILEBITS(0))));
245 	if (patime || pmtime)
246 		set_ftime(tmp_name, arcn->sb.st_mtime,
247 		    arcn->sb.st_atime, 0, 0);
248 
249 	/* Did we write directly to the target file? */
250 	if (arcn->tmp_name == NULL)
251 		return;
252 
253 	/*
254 	 * Finally, now the temp file is fully instantiated rename it to
255 	 * the desired file name.
256 	 */
257 	if (rename(tmp_name, arcn->name) < 0) {
258 		syswarn(0, errno, "Cannot rename %s to %s",
259 		    tmp_name, arcn->name);
260 		(void)unlink(tmp_name);
261 	}
262 
263 #if HAVE_STRUCT_STAT_ST_FLAGS
264 	if (pfflags && arcn->type != PAX_SLK)
265 		set_chflags(arcn->name, arcn->sb.st_flags);
266 #endif
267 
268 	free(arcn->tmp_name);
269 	arcn->tmp_name = NULL;
270 	xtmp_name = NULL;
271 }
272 
273 /*
274  * lnk_creat()
275  *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
276  *	must exist;
277  * Return:
278  *	0 if ok, -1 otherwise
279  */
280 
281 int
282 lnk_creat(ARCHD *arcn, int *payload)
283 {
284 	struct stat sb;
285 
286 	/*
287 	 * Check if this hardlink carries the "payload". In "cpio" archives
288 	 * it's usually the last record of a set of hardlinks which includes
289 	 * the contents of the file.
290 	 *
291 	 */
292 	*payload = S_ISREG(arcn->sb.st_mode) &&
293 	    (arcn->sb.st_size > 0) && (arcn->sb.st_size <= arcn->skip);
294 
295 	/*
296 	 * We may be running as root, so we have to be sure that link target
297 	 * is not a directory, so we lstat and check. XXX: This is still racy.
298 	 */
299 	if (lstat(arcn->ln_name, &sb) != -1 && S_ISDIR(sb.st_mode)) {
300 		tty_warn(1, "A hard link to the directory %s is not allowed",
301 		    arcn->ln_name);
302 		return -1;
303 	}
304 
305 	return mk_link(arcn->ln_name, &sb, arcn->name, 0);
306 }
307 
308 /*
309  * cross_lnk()
310  *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
311  *	with the -l flag. No warning or error if this does not succeed (we will
312  *	then just create the file)
313  * Return:
314  *	1 if copy() should try to create this file node
315  *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
316  */
317 
318 int
319 cross_lnk(ARCHD *arcn)
320 {
321 	/*
322 	 * try to make a link to original file (-l flag in copy mode). make
323 	 * sure we do not try to link to directories in case we are running as
324 	 * root (and it might succeed).
325 	 */
326 	if (arcn->type == PAX_DIR)
327 		return 1;
328 	return mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1);
329 }
330 
331 /*
332  * chk_same()
333  *	In copy mode if we are not trying to make hard links between the src
334  *	and destinations, make sure we are not going to overwrite ourselves by
335  *	accident. This slows things down a little, but we have to protect all
336  *	those people who make typing errors.
337  * Return:
338  *	1 the target does not exist, go ahead and copy
339  *	0 skip it file exists (-k) or may be the same as source file
340  */
341 
342 int
343 chk_same(ARCHD *arcn)
344 {
345 	struct stat sb;
346 
347 	/*
348 	 * if file does not exist, return. if file exists and -k, skip it
349 	 * quietly
350 	 */
351 	if (lstat(arcn->name, &sb) < 0)
352 		return 1;
353 	if (kflag)
354 		return 0;
355 
356 	/*
357 	 * better make sure the user does not have src == dest by mistake
358 	 */
359 	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
360 		tty_warn(1, "Unable to copy %s, file would overwrite itself",
361 		    arcn->name);
362 		return 0;
363 	}
364 	return 1;
365 }
366 
367 /*
368  * mk_link()
369  *	try to make a hard link between two files. if ign set, we do not
370  *	complain.
371  * Return:
372  *	0 if successful (or we are done with this file but no error, such as
373  *	finding the from file exists and the user has set -k).
374  *	1 when ign was set to indicates we could not make the link but we
375  *	should try to copy/extract the file as that might work (and is an
376  *	allowed option). -1 an error occurred.
377  */
378 
379 static int
380 mk_link(char *to, struct stat *to_sb, char *from, int ign)
381 {
382 	struct stat sb;
383 	int oerrno;
384 
385 	/*
386 	 * if from file exists, it has to be unlinked to make the link. If the
387 	 * file exists and -k is set, skip it quietly
388 	 */
389 	if (lstat(from, &sb) == 0) {
390 		if (kflag)
391 			return 0;
392 
393 		/*
394 		 * make sure it is not the same file, protect the user
395 		 */
396 		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
397 			tty_warn(1, "Cannot link file %s to itself", to);
398 			return -1;
399 		}
400 
401 		/*
402 		 * try to get rid of the file, based on the type
403 		 */
404 		if (S_ISDIR(sb.st_mode) && strcmp(from, ".") != 0) {
405 			if (rmdir(from) < 0) {
406 				syswarn(1, errno, "Cannot remove %s", from);
407 				return -1;
408 			}
409 		} else if (unlink(from) < 0) {
410 			if (!ign) {
411 				syswarn(1, errno, "Cannot remove %s", from);
412 				return -1;
413 			}
414 			return 1;
415 		}
416 	}
417 
418 	/*
419 	 * from file is gone (or did not exist), try to make the hard link.
420 	 * if it fails, check the path and try it again (if chk_path() says to
421 	 * try again)
422 	 */
423 	for (;;) {
424 		if (link(to, from) == 0)
425 			break;
426 		oerrno = errno;
427 		if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
428 			continue;
429 		if (!ign) {
430 			syswarn(1, oerrno, "Cannot link to %s from %s", to,
431 			    from);
432 			return -1;
433 		}
434 		return 1;
435 	}
436 
437 	/*
438 	 * all right the link was made
439 	 */
440 	return 0;
441 }
442 
443 /*
444  * node_creat()
445  *	create an entry in the file system (other than a file or hard link).
446  *	If successful, sets uid/gid modes and times as required.
447  * Return:
448  *	0 if ok, -1 otherwise
449  */
450 
451 int
452 node_creat(ARCHD *arcn)
453 {
454 	int res;
455 	int ign = 0;
456 	int oerrno;
457 	int pass = 0;
458 	mode_t file_mode;
459 	struct stat sb;
460 	char target[MAXPATHLEN];
461 	char *nm = arcn->name;
462 	int len;
463 
464 	/*
465 	 * create node based on type, if that fails try to unlink the node and
466 	 * try again. finally check the path and try again. As noted in the
467 	 * file and link creation routines, this method seems to exhibit the
468 	 * best performance in general use workloads.
469 	 */
470 	file_mode = arcn->sb.st_mode & FILEBITS(arcn->type == PAX_DIR);
471 
472 	for (;;) {
473 		switch (arcn->type) {
474 		case PAX_DIR:
475 			/*
476 			 * If -h (or -L) was given in tar-mode, follow the
477 			 * potential symlink chain before trying to create the
478 			 * directory.
479 			 */
480 			if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
481 				while (lstat(nm, &sb) == 0 &&
482 				    S_ISLNK(sb.st_mode)) {
483 					len = readlink(nm, target,
484 					    sizeof target - 1);
485 					if (len == -1) {
486 						syswarn(0, errno,
487 						   "cannot follow symlink %s "
488 						   "in chain for %s",
489 						    nm, arcn->name);
490 						res = -1;
491 						goto badlink;
492 					}
493 					target[len] = '\0';
494 					nm = target;
495 				}
496 			}
497 			res = domkdir(nm, file_mode);
498 badlink:
499 			if (ign)
500 				res = 0;
501 			break;
502 		case PAX_CHR:
503 			file_mode |= S_IFCHR;
504 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
505 			break;
506 		case PAX_BLK:
507 			file_mode |= S_IFBLK;
508 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
509 			break;
510 		case PAX_FIF:
511 			res = mkfifo(nm, file_mode);
512 			break;
513 		case PAX_SCK:
514 			/*
515 			 * Skip sockets, operation has no meaning under BSD
516 			 */
517 			tty_warn(0,
518 			    "%s skipped. Sockets cannot be copied or extracted",
519 			    nm);
520 			return (-1);
521 		case PAX_SLK:
522 			res = symlink(arcn->ln_name, nm);
523 			break;
524 		case PAX_CTG:
525 		case PAX_HLK:
526 		case PAX_HRG:
527 		case PAX_REG:
528 		default:
529 			/*
530 			 * we should never get here
531 			 */
532 			tty_warn(0, "%s has an unknown file type, skipping",
533 			    nm);
534 			return (-1);
535 		}
536 
537 		/*
538 		 * if we were able to create the node break out of the loop,
539 		 * otherwise try to unlink the node and try again. if that
540 		 * fails check the full path and try a final time.
541 		 */
542 		if (res == 0)
543 			break;
544 
545 		/*
546 		 * we failed to make the node
547 		 */
548 		oerrno = errno;
549 		switch (pass++) {
550 		case 0:
551 			if ((ign = unlnk_exist(nm, arcn->type)) < 0)
552 				return (-1);
553 			continue;
554 
555 		case 1:
556 			if (nodirs ||
557 			    chk_path(nm, arcn->sb.st_uid,
558 			    arcn->sb.st_gid) < 0) {
559 				syswarn(1, oerrno, "Cannot create %s", nm);
560 				return (-1);
561 			}
562 			continue;
563 		}
564 
565 		/*
566 		 * it must be a file that exists but we can't create or
567 		 * remove, but we must avoid the infinite loop.
568 		 */
569 		break;
570 	}
571 
572 	/*
573 	 * we were able to create the node. set uid/gid, modes and times
574 	 */
575 	if (pids)
576 		res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
577 	else
578 		res = 0;
579 
580 	/*
581 	 * IMPORTANT SECURITY NOTE:
582 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
583 	 * set uid/gid bits
584 	 */
585 	if (!pmode || res)
586 		arcn->sb.st_mode &= ~SETBITS(arcn->type == PAX_DIR);
587 	if (pmode)
588 		set_pmode(arcn->name, arcn->sb.st_mode);
589 
590 	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
591 		/*
592 		 * Dirs must be processed again at end of extract to set times
593 		 * and modes to agree with those stored in the archive. However
594 		 * to allow extract to continue, we may have to also set owner
595 		 * rights. This allows nodes in the archive that are children
596 		 * of this directory to be extracted without failure. Both time
597 		 * and modes will be fixed after the entire archive is read and
598 		 * before pax exits.
599 		 */
600 		if (access(nm, R_OK | W_OK | X_OK) < 0) {
601 			if (lstat(nm, &sb) < 0) {
602 				syswarn(0, errno,"Cannot access %s (stat)",
603 				    arcn->name);
604 				set_pmode(nm,file_mode | S_IRWXU);
605 			} else {
606 				/*
607 				 * We have to add rights to the dir, so we make
608 				 * sure to restore the mode. The mode must be
609 				 * restored AS CREATED and not as stored if
610 				 * pmode is not set.
611 				 */
612 				set_pmode(nm, ((sb.st_mode &
613 				    FILEBITS(arcn->type == PAX_DIR)) |
614 				    S_IRWXU));
615 				if (!pmode)
616 					arcn->sb.st_mode = sb.st_mode;
617 			}
618 
619 			/*
620 			 * we have to force the mode to what was set here,
621 			 * since we changed it from the default as created.
622 			 */
623 			add_dir(nm, arcn->nlen, &(arcn->sb), 1);
624 		} else if (pmode || patime || pmtime)
625 			add_dir(nm, arcn->nlen, &(arcn->sb), 0);
626 	}
627 
628 	if (patime || pmtime)
629 		set_ftime(arcn->name, arcn->sb.st_mtime,
630 		    arcn->sb.st_atime, 0, (arcn->type == PAX_SLK) ? 1 : 0);
631 
632 #if HAVE_STRUCT_STAT_ST_FLAGS
633 	if (pfflags && arcn->type != PAX_SLK)
634 		set_chflags(arcn->name, arcn->sb.st_flags);
635 #endif
636 	return 0;
637 }
638 
639 /*
640  * unlnk_exist()
641  *	Remove node from file system with the specified name. We pass the type
642  *	of the node that is going to replace it. When we try to create a
643  *	directory and find that it already exists, we allow processing to
644  *	continue as proper modes etc will always be set for it later on.
645  * Return:
646  *	0 is ok to proceed, no file with the specified name exists
647  *	-1 we were unable to remove the node, or we should not remove it (-k)
648  *	1 we found a directory and we were going to create a directory.
649  */
650 
651 int
652 unlnk_exist(char *name, int type)
653 {
654 	struct stat sb;
655 
656 	/*
657 	 * the file does not exist, or -k we are done
658 	 */
659 	if (lstat(name, &sb) < 0)
660 		return 0;
661 	if (kflag)
662 		return -1;
663 
664 	if (S_ISDIR(sb.st_mode)) {
665 		/*
666 		 * try to remove a directory, if it fails and we were going to
667 		 * create a directory anyway, tell the caller (return a 1).
668 		 *
669 		 * don't try to remove the directory if the name is "."
670 		 * otherwise later file/directory creation fails.
671 		 */
672 		if (strcmp(name, ".") == 0)
673 			return 1;
674 		if (rmdir(name) < 0) {
675 			if (type == PAX_DIR)
676 				return 1;
677 			syswarn(1, errno, "Cannot remove directory %s", name);
678 			return -1;
679 		}
680 		return 0;
681 	}
682 
683 	/*
684 	 * try to get rid of all non-directory type nodes
685 	 */
686 	if (unlink(name) < 0) {
687 		(void)fflush(listf);
688 		syswarn(1, errno, "Cannot unlink %s", name);
689 		return -1;
690 	}
691 	return 0;
692 }
693 
694 /*
695  * chk_path()
696  *	We were trying to create some kind of node in the file system and it
697  *	failed. chk_path() makes sure the path up to the node exists and is
698  *	writable. When we have to create a directory that is missing along the
699  *	path somewhere, the directory we create will be set to the same
700  *	uid/gid as the file has (when uid and gid are being preserved).
701  *	NOTE: this routine is a real performance loss. It is only used as a
702  *	last resort when trying to create entries in the file system.
703  * Return:
704  *	-1 when it could find nothing it is allowed to fix.
705  *	0 otherwise
706  */
707 
708 int
709 chk_path(char *name, uid_t st_uid, gid_t st_gid)
710 {
711 	char *spt = name;
712 	struct stat sb;
713 	int retval = -1;
714 
715 	/*
716 	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
717 	 */
718 	if (*spt == '/')
719 		++spt;
720 
721 	for(;;) {
722 		/*
723 		 * work forward from the first / and check each part of
724 		 * the path
725 		 */
726 		spt = strchr(spt, '/');
727 		if (spt == NULL)
728 			break;
729 		*spt = '\0';
730 
731 		/*
732 		 * if it exists we assume it is a directory, it is not within
733 		 * the spec (at least it seems to read that way) to alter the
734 		 * file system for nodes NOT EXPLICITLY stored on the archive.
735 		 * If that assumption is changed, you would test the node here
736 		 * and figure out how to get rid of it (probably like some
737 		 * recursive unlink()) or fix up the directory permissions if
738 		 * required (do an access()).
739 		 */
740 		if (lstat(name, &sb) == 0) {
741 			*(spt++) = '/';
742 			continue;
743 		}
744 
745 		/*
746 		 * the path fails at this point, see if we can create the
747 		 * needed directory and continue on
748 		 */
749 		if (domkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) == -1) {
750 			*spt = '/';
751 			retval = -1;
752 			break;
753 		}
754 
755 		/*
756 		 * we were able to create the directory. We will tell the
757 		 * caller that we found something to fix, and it is ok to try
758 		 * and create the node again.
759 		 */
760 		retval = 0;
761 		if (pids)
762 			(void)set_ids(name, st_uid, st_gid);
763 
764 		/*
765 		 * make sure the user doesn't have some strange umask that
766 		 * causes this newly created directory to be unusable. We fix
767 		 * the modes and restore them back to the creation default at
768 		 * the end of pax
769 		 */
770 		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
771 		    (lstat(name, &sb) == 0)) {
772 			set_pmode(name, ((sb.st_mode & FILEBITS(0)) |
773 			    S_IRWXU));
774 			add_dir(name, spt - name, &sb, 1);
775 		}
776 		*(spt++) = '/';
777 		continue;
778 	}
779 	/*
780 	 * We perform one final check here, because if someone else
781 	 * created the directory in parallel with us, we might return
782 	 * the wrong error code, even if the directory exists now.
783 	 */
784 	if (retval == -1 && stat(name, &sb) == 0 && S_ISDIR(sb.st_mode))
785 		retval = 0;
786 	return retval;
787 }
788 
789 /*
790  * set_ftime()
791  *	Set the access time and modification time for a named file. If frc
792  *	is non-zero we force these times to be set even if the user did not
793  *	request access and/or modification time preservation (this is also
794  *	used by -t to reset access times).
795  *	When ign is zero, only those times the user has asked for are set, the
796  *	other ones are left alone. We do not assume the un-documented feature
797  *	of many utimes() implementations that consider a 0 time value as a do
798  *	not set request.
799  *
800  *	Unfortunately, there are systems where lutimes() is present but does
801  *	not work on some filesystem types, which cannot be detected at
802  * 	compile time.  This requires passing down symlink knowledge into
803  *	this function to obtain correct operation.  Linux with XFS is one
804  * 	example of such a system.
805  */
806 
807 void
808 set_ftime(char *fnm, time_t mtime, time_t atime, int frc, int slk)
809 {
810 	struct timeval tv[2];
811 	struct stat sb;
812 
813 	tv[0].tv_sec = (long)atime;
814 	tv[0].tv_usec = 0;
815 	tv[1].tv_sec = (long)mtime;
816 	tv[1].tv_usec = 0;
817 	if (!frc && (!patime || !pmtime)) {
818 		/*
819 		 * if we are not forcing, only set those times the user wants
820 		 * set. We get the current values of the times if we need them.
821 		 */
822 		if (lstat(fnm, &sb) == 0) {
823 #if BSD4_4 && !HAVE_NBTOOL_CONFIG_H
824 			if (!patime)
825 				TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
826 			if (!pmtime)
827 				TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
828 #else
829 			if (!patime)
830 				tv[0].tv_sec = sb.st_atime;
831 			if (!pmtime)
832 				tv[1].tv_sec = sb.st_mtime;
833 #endif
834 		} else
835 			syswarn(0, errno, "Cannot obtain file stats %s", fnm);
836 	}
837 
838 	/*
839 	 * set the times
840 	 */
841 #if HAVE_LUTIMES
842 	if (lutimes(fnm, tv) == 0)
843 		return;
844 	if (errno != ENOSYS)	/* XXX linux: lutimes is per-FS */
845 		goto bad;
846 #endif
847 	if (slk)
848 		return;
849 #ifdef __minix
850 	{
851 		struct utimbuf timp;
852 		timp.actime = atime;
853 		timp.modtime = mtime;
854 		if (utime(fnm, &timp) < 0)
855 			goto bad;
856 	}
857 #else
858 	if (utimes(fnm, tv) == -1)
859 		goto bad;
860 #endif
861 	return;
862 bad:
863 	syswarn(1, errno, "Access/modification time set failed on: %s", fnm);
864 }
865 
866 /*
867  * set_ids()
868  *	set the uid and gid of a file system node
869  * Return:
870  *	0 when set, -1 on failure
871  */
872 
873 int
874 set_ids(char *fnm, uid_t uid, gid_t gid)
875 {
876 	if (geteuid() == 0)
877 		if (lchown(fnm, uid, gid)) {
878 			(void)fflush(listf);
879 			syswarn(1, errno, "Cannot set file uid/gid of %s",
880 			    fnm);
881 			return -1;
882 		}
883 	return 0;
884 }
885 
886 /*
887  * set_pmode()
888  *	Set file access mode
889  */
890 
891 void
892 set_pmode(char *fnm, mode_t mode)
893 {
894 	mode &= A_BITS;
895 	if (lchmod(fnm, mode)) {
896 		(void)fflush(listf);
897 		syswarn(1, errno, "Cannot set permissions on %s", fnm);
898 	}
899 	return;
900 }
901 
902 /*
903  * set_chflags()
904  *	Set 4.4BSD file flags
905  */
906 void
907 set_chflags(char *fnm, u_int32_t flags)
908 {
909 
910 #if 0
911 	if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
912 		syswarn(1, errno, "Cannot set file flags on %s", fnm);
913 #endif
914 	return;
915 }
916 
917 /*
918  * file_write()
919  *	Write/copy a file (during copy or archive extract). This routine knows
920  *	how to copy files with lseek holes in it. (Which are read as file
921  *	blocks containing all 0's but do not have any file blocks associated
922  *	with the data). Typical examples of these are files created by dbm
923  *	variants (.pag files). While the file size of these files are huge, the
924  *	actual storage is quite small (the files are sparse). The problem is
925  *	the holes read as all zeros so are probably stored on the archive that
926  *	way (there is no way to determine if the file block is really a hole,
927  *	we only know that a file block of all zero's can be a hole).
928  *	At this writing, no major archive format knows how to archive files
929  *	with holes. However, on extraction (or during copy, -rw) we have to
930  *	deal with these files. Without detecting the holes, the files can
931  *	consume a lot of file space if just written to disk. This replacement
932  *	for write when passed the basic allocation size of a file system block,
933  *	uses lseek whenever it detects the input data is all 0 within that
934  *	file block. In more detail, the strategy is as follows:
935  *	While the input is all zero keep doing an lseek. Keep track of when we
936  *	pass over file block boundaries. Only write when we hit a non zero
937  *	input. once we have written a file block, we continue to write it to
938  *	the end (we stop looking at the input). When we reach the start of the
939  *	next file block, start checking for zero blocks again. Working on file
940  *	block boundaries significantly reduces the overhead when copying files
941  *	that are NOT very sparse. This overhead (when compared to a write) is
942  *	almost below the measurement resolution on many systems. Without it,
943  *	files with holes cannot be safely copied. It does has a side effect as
944  *	it can put holes into files that did not have them before, but that is
945  *	not a problem since the file contents are unchanged (in fact it saves
946  *	file space). (Except on paging files for diskless clients. But since we
947  *	cannot determine one of those file from here, we ignore them). If this
948  *	ever ends up on a system where CTG files are supported and the holes
949  *	are not desired, just do a conditional test in those routines that
950  *	call file_write() and have it call write() instead. BEFORE CLOSING THE
951  *	FILE, make sure to call file_flush() when the last write finishes with
952  *	an empty block. A lot of file systems will not create an lseek hole at
953  *	the end. In this case we drop a single 0 at the end to force the
954  *	trailing 0's in the file.
955  *	---Parameters---
956  *	rem: how many bytes left in this file system block
957  *	isempt: have we written to the file block yet (is it empty)
958  *	sz: basic file block allocation size
959  *	cnt: number of bytes on this write
960  *	str: buffer to write
961  * Return:
962  *	number of bytes written, -1 on write (or lseek) error.
963  */
964 
965 int
966 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
967 	char *name)
968 {
969 	char *pt;
970 	char *end;
971 	int wcnt;
972 	char *st = str;
973 	char **strp;
974 	size_t *lenp;
975 
976 	/*
977 	 * while we have data to process
978 	 */
979 	while (cnt) {
980 		if (!*rem) {
981 			/*
982 			 * We are now at the start of file system block again
983 			 * (or what we think one is...). start looking for
984 			 * empty blocks again
985 			 */
986 			*isempt = 1;
987 			*rem = sz;
988 		}
989 
990 		/*
991 		 * only examine up to the end of the current file block or
992 		 * remaining characters to write, whatever is smaller
993 		 */
994 		wcnt = MIN(cnt, *rem);
995 		cnt -= wcnt;
996 		*rem -= wcnt;
997 		if (*isempt) {
998 			/*
999 			 * have not written to this block yet, so we keep
1000 			 * looking for zero's
1001 			 */
1002 			pt = st;
1003 			end = st + wcnt;
1004 
1005 			/*
1006 			 * look for a zero filled buffer
1007 			 */
1008 			while ((pt < end) && (*pt == '\0'))
1009 				++pt;
1010 
1011 			if (pt == end) {
1012 				/*
1013 				 * skip, buf is empty so far
1014 				 */
1015 				if (fd > -1 &&
1016 				    lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
1017 					syswarn(1, errno, "File seek on %s",
1018 					    name);
1019 					return -1;
1020 				}
1021 				st = pt;
1022 				continue;
1023 			}
1024 			/*
1025 			 * drat, the buf is not zero filled
1026 			 */
1027 			*isempt = 0;
1028 		}
1029 
1030 		/*
1031 		 * have non-zero data in this file system block, have to write
1032 		 */
1033 		switch (fd) {
1034 		case -PAX_GLF:
1035 			strp = &gnu_name_string;
1036 			lenp = &gnu_name_length;
1037 			break;
1038 		case -PAX_GLL:
1039 			strp = &gnu_link_string;
1040 			lenp = &gnu_link_length;
1041 			break;
1042 		default:
1043 			strp = NULL;
1044 			lenp = NULL;
1045 			break;
1046 		}
1047 		if (strp) {
1048 			char *nstr = *strp ? realloc(*strp, *lenp + wcnt + 1) :
1049 				malloc(wcnt + 1);
1050 			if (nstr == NULL) {
1051 				tty_warn(1, "Out of memory");
1052 				return -1;
1053 			}
1054 			(void)strlcpy(&nstr[*lenp], st, wcnt + 1);
1055 			*strp = nstr;
1056 			*lenp += wcnt;
1057 		} else if (xwrite(fd, st, wcnt) != wcnt) {
1058 			syswarn(1, errno, "Failed write to file %s", name);
1059 			return -1;
1060 		}
1061 		st += wcnt;
1062 	}
1063 	return st - str;
1064 }
1065 
1066 /*
1067  * file_flush()
1068  *	when the last file block in a file is zero, many file systems will not
1069  *	let us create a hole at the end. To get the last block with zeros, we
1070  *	write the last BYTE with a zero (back up one byte and write a zero).
1071  */
1072 
1073 void
1074 file_flush(int fd, char *fname, int isempt)
1075 {
1076 	static char blnk[] = "\0";
1077 
1078 	/*
1079 	 * silly test, but make sure we are only called when the last block is
1080 	 * filled with all zeros.
1081 	 */
1082 	if (!isempt)
1083 		return;
1084 
1085 	/*
1086 	 * move back one byte and write a zero
1087 	 */
1088 	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1089 		syswarn(1, errno, "Failed seek on file %s", fname);
1090 		return;
1091 	}
1092 
1093 	if (write_with_restart(fd, blnk, 1) < 0)
1094 		syswarn(1, errno, "Failed write to file %s", fname);
1095 	return;
1096 }
1097 
1098 /*
1099  * rdfile_close()
1100  *	close a file we have been reading (to copy or archive). If we have to
1101  *	reset access time (tflag) do so (the times are stored in arcn).
1102  */
1103 
1104 void
1105 rdfile_close(ARCHD *arcn, int *fd)
1106 {
1107 	/*
1108 	 * make sure the file is open
1109 	 */
1110 	if (*fd < 0)
1111 		return;
1112 
1113 	(void)close(*fd);
1114 	*fd = -1;
1115 	if (!tflag)
1116 		return;
1117 
1118 	/*
1119 	 * user wants last access time reset
1120 	 */
1121 	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1, 0);
1122 	return;
1123 }
1124 
1125 /*
1126  * set_crc()
1127  *	read a file to calculate its crc. This is a real drag. Archive formats
1128  *	that have this, end up reading the file twice (we have to write the
1129  *	header WITH the crc before writing the file contents. Oh well...
1130  * Return:
1131  *	0 if was able to calculate the crc, -1 otherwise
1132  */
1133 
1134 int
1135 set_crc(ARCHD *arcn, int fd)
1136 {
1137 	int i;
1138 	int res;
1139 	off_t cpcnt = 0L;
1140 	u_long size;
1141 	unsigned long crc = 0L;
1142 	char tbuf[FILEBLK];
1143 	struct stat sb;
1144 
1145 	if (fd < 0) {
1146 		/*
1147 		 * hmm, no fd, should never happen. well no crc then.
1148 		 */
1149 		arcn->crc = 0L;
1150 		return 0;
1151 	}
1152 
1153 	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1154 		size = (u_long)sizeof(tbuf);
1155 
1156 	/*
1157 	 * read all the bytes we think that there are in the file. If the user
1158 	 * is trying to archive an active file, forget this file.
1159 	 */
1160 	for(;;) {
1161 		if ((res = read(fd, tbuf, size)) <= 0)
1162 			break;
1163 		cpcnt += res;
1164 		for (i = 0; i < res; ++i)
1165 			crc += (tbuf[i] & 0xff);
1166 	}
1167 
1168 	/*
1169 	 * safety check. we want to avoid archiving files that are active as
1170 	 * they can create inconsistent archive copies.
1171 	 */
1172 	if (cpcnt != arcn->sb.st_size)
1173 		tty_warn(1, "File changed size %s", arcn->org_name);
1174 	else if (fstat(fd, &sb) < 0)
1175 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1176 	else if (arcn->sb.st_mtime != sb.st_mtime)
1177 		tty_warn(1, "File %s was modified during read", arcn->org_name);
1178 	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1179 		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1180 	else {
1181 		arcn->crc = crc;
1182 		return 0;
1183 	}
1184 	return -1;
1185 }
1186