xref: /netbsd-src/bin/pax/file_subs.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
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 char *xtmp_name;
66 
67 static int
68 mk_link(char *,struct stat *,char *, int);
69 
70 static int warn_broken;
71 
72 /*
73  * routines that deal with file operations such as: creating, removing;
74  * and setting access modes, uid/gid and times of files
75  */
76 #define SET_BITS		(S_ISUID | S_ISGID)
77 #define FILE_BITS		(S_IRWXU | S_IRWXG | S_IRWXO)
78 #define A_BITS			(FILE_BITS | SET_BITS | S_ISVTX)
79 
80 /*
81  * The S_ISVTX (sticky bit) can be set by non-superuser on directories
82  * but not other kinds of files.
83  */
84 #define FILEBITS(dir)		((dir) ? (FILE_BITS | S_ISVTX) : FILE_BITS)
85 #define SETBITS(dir)		((dir) ? SET_BITS : (SET_BITS | S_ISVTX))
86 
87 static mode_t
88 apply_umask(mode_t mode)
89 {
90 	static mode_t cached_umask;
91 	static int cached_umask_valid;
92 
93 	if (!cached_umask_valid) {
94 		cached_umask = umask(0);
95 		umask(cached_umask);
96 		cached_umask_valid = 1;
97 	}
98 
99 	return mode & ~cached_umask;
100 }
101 
102 /*
103  * file_creat()
104  *	Create and open a file.
105  * Return:
106  *	file descriptor or -1 for failure
107  */
108 
109 int
110 file_creat(ARCHD *arcn, int write_to_hardlink)
111 {
112 	int fd = -1;
113 	int oerrno;
114 
115 	/*
116 	 * Some horribly busted tar implementations, have directory nodes
117 	 * that end in a /, but they mark as files. Compensate for that
118 	 * by not creating a directory node at this point, but a file node,
119 	 * and not creating the temp file.
120 	 */
121 	if (arcn->nlen != 0 && arcn->name[arcn->nlen - 1] == '/') {
122 		if (!warn_broken) {
123 			tty_warn(0, "Archive was created with a broken tar;"
124 			    " file `%s' is a directory, but marked as plain.",
125 			    arcn->name);
126 			warn_broken = 1;
127 		}
128 		return -1;
129 	}
130 
131 	/*
132 	 * In "cpio" archives it's usually the last record of a set of
133 	 * hardlinks which includes the contents of the file. We cannot
134 	 * use a tempory file in that case because we couldn't link it
135 	 * with the existing other hardlinks after restoring the contents
136 	 * to it. And it's also useless to create the hardlink under a
137 	 * temporary name because the other hardlinks would have partial
138 	 * contents while restoring.
139 	 */
140 	if (write_to_hardlink)
141 		return (open(arcn->name, O_TRUNC | O_EXCL | O_RDWR, 0));
142 
143 	/*
144 	 * Create a temporary file name so that the file doesn't have partial
145 	 * contents while restoring.
146 	 */
147 	arcn->tmp_name = malloc(arcn->nlen + 8);
148 	if (arcn->tmp_name == NULL) {
149 		syswarn(1, errno, "Cannot malloc %d bytes", arcn->nlen + 8);
150 		return -1;
151 	}
152 	if (xtmp_name != NULL)
153 		abort();
154 	xtmp_name = arcn->tmp_name;
155 
156 	for (;;) {
157 		/*
158 		 * try to create the temporary file we use to restore the
159 		 * contents info.  if this fails, keep checking all the nodes
160 		 * in the path until chk_path() finds that it cannot fix
161 		 * anything further.  if that happens we just give up.
162 		 */
163 		(void)snprintf(arcn->tmp_name, arcn->nlen + 8, "%s.XXXXXX",
164 		    arcn->name);
165 		fd = mkstemp(arcn->tmp_name);
166 		if (fd >= 0)
167 			break;
168 		oerrno = errno;
169 		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
170 			(void)fflush(listf);
171 			syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name);
172 			xtmp_name = NULL;
173 			free(arcn->tmp_name);
174 			arcn->tmp_name = NULL;
175 			return -1;
176 		}
177 	}
178 	return fd;
179 }
180 
181 /*
182  * file_close()
183  *	Close file descriptor to a file just created by pax. Sets modes,
184  *	ownership and times as required.
185  * Return:
186  *	0 for success, -1 for failure
187  */
188 
189 void
190 file_close(ARCHD *arcn, int fd)
191 {
192 	char *tmp_name;
193 	int res;
194 
195 	if (fd < 0)
196 		return;
197 
198 	tmp_name = (arcn->tmp_name != NULL) ? arcn->tmp_name : arcn->name;
199 
200 	if (close(fd) < 0)
201 		syswarn(0, errno, "Cannot close file descriptor on %s",
202 		    tmp_name);
203 
204 	/*
205 	 * set owner/groups first as this may strip off mode bits we want
206 	 * then set file permission modes. Then set file access and
207 	 * modification times.
208 	 */
209 	if (pids)
210 		res = set_ids(tmp_name, arcn->sb.st_uid, arcn->sb.st_gid);
211 	else
212 		res = 0;
213 
214 	/*
215 	 * IMPORTANT SECURITY NOTE:
216 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
217 	 * set uid/gid bits but restore the file modes (since mkstemp doesn't).
218 	 */
219 	if (!pmode || res)
220 		arcn->sb.st_mode &= ~SETBITS(0);
221 	if (pmode)
222 		set_pmode(tmp_name, arcn->sb.st_mode);
223 	else
224 		set_pmode(tmp_name,
225 		    apply_umask((arcn->sb.st_mode & FILEBITS(0))));
226 	if (patime || pmtime)
227 		set_ftime(tmp_name, arcn->sb.st_mtime,
228 		    arcn->sb.st_atime, 0, 0);
229 
230 	/* Did we write directly to the target file? */
231 	if (arcn->tmp_name == NULL)
232 		return;
233 
234 	/*
235 	 * Finally, now the temp file is fully instantiated rename it to
236 	 * the desired file name.
237 	 */
238 	if (rename(tmp_name, arcn->name) < 0) {
239 		syswarn(0, errno, "Cannot rename %s to %s",
240 		    tmp_name, arcn->name);
241 		(void)unlink(tmp_name);
242 	}
243 
244 #if HAVE_STRUCT_STAT_ST_FLAGS
245 	if (pfflags && arcn->type != PAX_SLK)
246 		set_chflags(arcn->name, arcn->sb.st_flags);
247 #endif
248 
249 	free(arcn->tmp_name);
250 	arcn->tmp_name = NULL;
251 	xtmp_name = NULL;
252 }
253 
254 /*
255  * lnk_creat()
256  *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
257  *	must exist;
258  * Return:
259  *	0 if ok, -1 otherwise
260  */
261 
262 int
263 lnk_creat(ARCHD *arcn, int *payload)
264 {
265 	struct stat sb;
266 
267 	/*
268 	 * Check if this hardlink carries the "payload". In "cpio" archives
269 	 * it's usually the last record of a set of hardlinks which includes
270 	 * the contents of the file.
271 	 *
272 	 */
273 	*payload = S_ISREG(arcn->sb.st_mode) &&
274 	    (arcn->sb.st_size > 0) && (arcn->sb.st_size <= arcn->skip);
275 
276 	/*
277 	 * We may be running as root, so we have to be sure that link target
278 	 * is not a directory, so we lstat and check. XXX: This is still racy.
279 	 */
280 	if (lstat(arcn->ln_name, &sb) != -1 && S_ISDIR(sb.st_mode)) {
281 		tty_warn(1, "A hard link to the directory %s is not allowed",
282 		    arcn->ln_name);
283 		return -1;
284 	}
285 
286 	return mk_link(arcn->ln_name, &sb, arcn->name, 0);
287 }
288 
289 /*
290  * cross_lnk()
291  *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
292  *	with the -l flag. No warning or error if this does not succeed (we will
293  *	then just create the file)
294  * Return:
295  *	1 if copy() should try to create this file node
296  *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
297  */
298 
299 int
300 cross_lnk(ARCHD *arcn)
301 {
302 	/*
303 	 * try to make a link to original file (-l flag in copy mode). make
304 	 * sure we do not try to link to directories in case we are running as
305 	 * root (and it might succeed).
306 	 */
307 	if (arcn->type == PAX_DIR)
308 		return 1;
309 	return mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1);
310 }
311 
312 /*
313  * chk_same()
314  *	In copy mode if we are not trying to make hard links between the src
315  *	and destinations, make sure we are not going to overwrite ourselves by
316  *	accident. This slows things down a little, but we have to protect all
317  *	those people who make typing errors.
318  * Return:
319  *	1 the target does not exist, go ahead and copy
320  *	0 skip it file exists (-k) or may be the same as source file
321  */
322 
323 int
324 chk_same(ARCHD *arcn)
325 {
326 	struct stat sb;
327 
328 	/*
329 	 * if file does not exist, return. if file exists and -k, skip it
330 	 * quietly
331 	 */
332 	if (lstat(arcn->name, &sb) < 0)
333 		return 1;
334 	if (kflag)
335 		return 0;
336 
337 	/*
338 	 * better make sure the user does not have src == dest by mistake
339 	 */
340 	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
341 		tty_warn(1, "Unable to copy %s, file would overwrite itself",
342 		    arcn->name);
343 		return 0;
344 	}
345 	return 1;
346 }
347 
348 /*
349  * mk_link()
350  *	try to make a hard link between two files. if ign set, we do not
351  *	complain.
352  * Return:
353  *	0 if successful (or we are done with this file but no error, such as
354  *	finding the from file exists and the user has set -k).
355  *	1 when ign was set to indicates we could not make the link but we
356  *	should try to copy/extract the file as that might work (and is an
357  *	allowed option). -1 an error occurred.
358  */
359 
360 static int
361 mk_link(char *to, struct stat *to_sb, char *from, int ign)
362 {
363 	struct stat sb;
364 	int oerrno;
365 
366 	/*
367 	 * if from file exists, it has to be unlinked to make the link. If the
368 	 * file exists and -k is set, skip it quietly
369 	 */
370 	if (lstat(from, &sb) == 0) {
371 		if (kflag)
372 			return 0;
373 
374 		/*
375 		 * make sure it is not the same file, protect the user
376 		 */
377 		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
378 			tty_warn(1, "Cannot link file %s to itself", to);
379 			return -1;
380 		}
381 
382 		/*
383 		 * try to get rid of the file, based on the type
384 		 */
385 		if (S_ISDIR(sb.st_mode) && strcmp(from, ".") != 0) {
386 			if (rmdir(from) < 0) {
387 				syswarn(1, errno, "Cannot remove %s", from);
388 				return -1;
389 			}
390 		} else if (unlink(from) < 0) {
391 			if (!ign) {
392 				syswarn(1, errno, "Cannot remove %s", from);
393 				return -1;
394 			}
395 			return 1;
396 		}
397 	}
398 
399 	/*
400 	 * from file is gone (or did not exist), try to make the hard link.
401 	 * if it fails, check the path and try it again (if chk_path() says to
402 	 * try again)
403 	 */
404 	for (;;) {
405 		if (link(to, from) == 0)
406 			break;
407 		oerrno = errno;
408 		if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
409 			continue;
410 		if (!ign) {
411 			syswarn(1, oerrno, "Cannot link to %s from %s", to,
412 			    from);
413 			return -1;
414 		}
415 		return 1;
416 	}
417 
418 	/*
419 	 * all right the link was made
420 	 */
421 	return 0;
422 }
423 
424 /*
425  * node_creat()
426  *	create an entry in the file system (other than a file or hard link).
427  *	If successful, sets uid/gid modes and times as required.
428  * Return:
429  *	0 if ok, -1 otherwise
430  */
431 
432 int
433 node_creat(ARCHD *arcn)
434 {
435 	int res;
436 	int ign = 0;
437 	int oerrno;
438 	int pass = 0;
439 	mode_t file_mode;
440 	struct stat sb;
441 	char target[MAXPATHLEN];
442 	char *nm = arcn->name;
443 	int len;
444 
445 	/*
446 	 * create node based on type, if that fails try to unlink the node and
447 	 * try again. finally check the path and try again. As noted in the
448 	 * file and link creation routines, this method seems to exhibit the
449 	 * best performance in general use workloads.
450 	 */
451 	file_mode = arcn->sb.st_mode & FILEBITS(arcn->type == PAX_DIR);
452 
453 	for (;;) {
454 		switch (arcn->type) {
455 		case PAX_DIR:
456 			/*
457 			 * If -h (or -L) was given in tar-mode, follow the
458 			 * potential symlink chain before trying to create the
459 			 * directory.
460 			 */
461 			if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
462 				while (lstat(nm, &sb) == 0 &&
463 				    S_ISLNK(sb.st_mode)) {
464 					len = readlink(nm, target,
465 					    sizeof target - 1);
466 					if (len == -1) {
467 						syswarn(0, errno,
468 						   "cannot follow symlink %s "
469 						   "in chain for %s",
470 						    nm, arcn->name);
471 						res = -1;
472 						goto badlink;
473 					}
474 					target[len] = '\0';
475 					nm = target;
476 				}
477 			}
478 			res = domkdir(nm, file_mode);
479 badlink:
480 			if (ign)
481 				res = 0;
482 			break;
483 		case PAX_CHR:
484 			file_mode |= S_IFCHR;
485 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
486 			break;
487 		case PAX_BLK:
488 			file_mode |= S_IFBLK;
489 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
490 			break;
491 		case PAX_FIF:
492 			res = mkfifo(nm, file_mode);
493 			break;
494 		case PAX_SCK:
495 			/*
496 			 * Skip sockets, operation has no meaning under BSD
497 			 */
498 			tty_warn(0,
499 			    "%s skipped. Sockets cannot be copied or extracted",
500 			    nm);
501 			return (-1);
502 		case PAX_SLK:
503 			res = symlink(arcn->ln_name, nm);
504 			break;
505 		case PAX_CTG:
506 		case PAX_HLK:
507 		case PAX_HRG:
508 		case PAX_REG:
509 		default:
510 			/*
511 			 * we should never get here
512 			 */
513 			tty_warn(0, "%s has an unknown file type, skipping",
514 			    nm);
515 			return (-1);
516 		}
517 
518 		/*
519 		 * if we were able to create the node break out of the loop,
520 		 * otherwise try to unlink the node and try again. if that
521 		 * fails check the full path and try a final time.
522 		 */
523 		if (res == 0)
524 			break;
525 
526 		/*
527 		 * we failed to make the node
528 		 */
529 		oerrno = errno;
530 		switch (pass++) {
531 		case 0:
532 			if ((ign = unlnk_exist(nm, arcn->type)) < 0)
533 				return (-1);
534 			continue;
535 
536 		case 1:
537 			if (nodirs ||
538 			    chk_path(nm, arcn->sb.st_uid,
539 			    arcn->sb.st_gid) < 0) {
540 				syswarn(1, oerrno, "Cannot create %s", nm);
541 				return (-1);
542 			}
543 			continue;
544 		}
545 
546 		/*
547 		 * it must be a file that exists but we can't create or
548 		 * remove, but we must avoid the infinite loop.
549 		 */
550 		break;
551 	}
552 
553 	/*
554 	 * we were able to create the node. set uid/gid, modes and times
555 	 */
556 	if (pids)
557 		res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
558 	else
559 		res = 0;
560 
561 	/*
562 	 * IMPORTANT SECURITY NOTE:
563 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
564 	 * set uid/gid bits
565 	 */
566 	if (!pmode || res)
567 		arcn->sb.st_mode &= ~SETBITS(arcn->type == PAX_DIR);
568 	if (pmode)
569 		set_pmode(arcn->name, arcn->sb.st_mode);
570 
571 	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
572 		/*
573 		 * Dirs must be processed again at end of extract to set times
574 		 * and modes to agree with those stored in the archive. However
575 		 * to allow extract to continue, we may have to also set owner
576 		 * rights. This allows nodes in the archive that are children
577 		 * of this directory to be extracted without failure. Both time
578 		 * and modes will be fixed after the entire archive is read and
579 		 * before pax exits.
580 		 */
581 		if (access(nm, R_OK | W_OK | X_OK) < 0) {
582 			if (lstat(nm, &sb) < 0) {
583 				syswarn(0, errno,"Cannot access %s (stat)",
584 				    arcn->name);
585 				set_pmode(nm,file_mode | S_IRWXU);
586 			} else {
587 				/*
588 				 * We have to add rights to the dir, so we make
589 				 * sure to restore the mode. The mode must be
590 				 * restored AS CREATED and not as stored if
591 				 * pmode is not set.
592 				 */
593 				set_pmode(nm, ((sb.st_mode &
594 				    FILEBITS(arcn->type == PAX_DIR)) |
595 				    S_IRWXU));
596 				if (!pmode)
597 					arcn->sb.st_mode = sb.st_mode;
598 			}
599 
600 			/*
601 			 * we have to force the mode to what was set here,
602 			 * since we changed it from the default as created.
603 			 */
604 			add_dir(nm, arcn->nlen, &(arcn->sb), 1);
605 		} else if (pmode || patime || pmtime)
606 			add_dir(nm, arcn->nlen, &(arcn->sb), 0);
607 	}
608 
609 	if (patime || pmtime)
610 		set_ftime(arcn->name, arcn->sb.st_mtime,
611 		    arcn->sb.st_atime, 0, (arcn->type == PAX_SLK) ? 1 : 0);
612 
613 #if HAVE_STRUCT_STAT_ST_FLAGS
614 	if (pfflags && arcn->type != PAX_SLK)
615 		set_chflags(arcn->name, arcn->sb.st_flags);
616 #endif
617 	return 0;
618 }
619 
620 /*
621  * unlnk_exist()
622  *	Remove node from file system with the specified name. We pass the type
623  *	of the node that is going to replace it. When we try to create a
624  *	directory and find that it already exists, we allow processing to
625  *	continue as proper modes etc will always be set for it later on.
626  * Return:
627  *	0 is ok to proceed, no file with the specified name exists
628  *	-1 we were unable to remove the node, or we should not remove it (-k)
629  *	1 we found a directory and we were going to create a directory.
630  */
631 
632 int
633 unlnk_exist(char *name, int type)
634 {
635 	struct stat sb;
636 
637 	/*
638 	 * the file does not exist, or -k we are done
639 	 */
640 	if (lstat(name, &sb) < 0)
641 		return 0;
642 	if (kflag)
643 		return -1;
644 
645 	if (S_ISDIR(sb.st_mode)) {
646 		/*
647 		 * try to remove a directory, if it fails and we were going to
648 		 * create a directory anyway, tell the caller (return a 1).
649 		 *
650 		 * don't try to remove the directory if the name is "."
651 		 * otherwise later file/directory creation fails.
652 		 */
653 		if (strcmp(name, ".") == 0)
654 			return 1;
655 		if (rmdir(name) < 0) {
656 			if (type == PAX_DIR)
657 				return 1;
658 			syswarn(1, errno, "Cannot remove directory %s", name);
659 			return -1;
660 		}
661 		return 0;
662 	}
663 
664 	/*
665 	 * try to get rid of all non-directory type nodes
666 	 */
667 	if (unlink(name) < 0) {
668 		(void)fflush(listf);
669 		syswarn(1, errno, "Cannot unlink %s", name);
670 		return -1;
671 	}
672 	return 0;
673 }
674 
675 /*
676  * chk_path()
677  *	We were trying to create some kind of node in the file system and it
678  *	failed. chk_path() makes sure the path up to the node exists and is
679  *	writable. When we have to create a directory that is missing along the
680  *	path somewhere, the directory we create will be set to the same
681  *	uid/gid as the file has (when uid and gid are being preserved).
682  *	NOTE: this routine is a real performance loss. It is only used as a
683  *	last resort when trying to create entries in the file system.
684  * Return:
685  *	-1 when it could find nothing it is allowed to fix.
686  *	0 otherwise
687  */
688 
689 int
690 chk_path(char *name, uid_t st_uid, gid_t st_gid)
691 {
692 	char *spt = name;
693 	struct stat sb;
694 	int retval = -1;
695 
696 	/*
697 	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
698 	 */
699 	if (*spt == '/')
700 		++spt;
701 
702 	for(;;) {
703 		/*
704 		 * work forward from the first / and check each part of
705 		 * the path
706 		 */
707 		spt = strchr(spt, '/');
708 		if (spt == NULL)
709 			break;
710 		*spt = '\0';
711 
712 		/*
713 		 * if it exists we assume it is a directory, it is not within
714 		 * the spec (at least it seems to read that way) to alter the
715 		 * file system for nodes NOT EXPLICITLY stored on the archive.
716 		 * If that assumption is changed, you would test the node here
717 		 * and figure out how to get rid of it (probably like some
718 		 * recursive unlink()) or fix up the directory permissions if
719 		 * required (do an access()).
720 		 */
721 		if (lstat(name, &sb) == 0) {
722 			*(spt++) = '/';
723 			continue;
724 		}
725 
726 		/*
727 		 * the path fails at this point, see if we can create the
728 		 * needed directory and continue on
729 		 */
730 		if (domkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) == -1) {
731 			*spt = '/';
732 			retval = -1;
733 			break;
734 		}
735 
736 		/*
737 		 * we were able to create the directory. We will tell the
738 		 * caller that we found something to fix, and it is ok to try
739 		 * and create the node again.
740 		 */
741 		retval = 0;
742 		if (pids)
743 			(void)set_ids(name, st_uid, st_gid);
744 
745 		/*
746 		 * make sure the user doesn't have some strange umask that
747 		 * causes this newly created directory to be unusable. We fix
748 		 * the modes and restore them back to the creation default at
749 		 * the end of pax
750 		 */
751 		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
752 		    (lstat(name, &sb) == 0)) {
753 			set_pmode(name, ((sb.st_mode & FILEBITS(0)) |
754 			    S_IRWXU));
755 			add_dir(name, spt - name, &sb, 1);
756 		}
757 		*(spt++) = '/';
758 		continue;
759 	}
760 	/*
761 	 * We perform one final check here, because if someone else
762 	 * created the directory in parallel with us, we might return
763 	 * the wrong error code, even if the directory exists now.
764 	 */
765 	if (retval == -1 && stat(name, &sb) == 0 && S_ISDIR(sb.st_mode))
766 		retval = 0;
767 	return retval;
768 }
769 
770 /*
771  * set_ftime()
772  *	Set the access time and modification time for a named file. If frc
773  *	is non-zero we force these times to be set even if the user did not
774  *	request access and/or modification time preservation (this is also
775  *	used by -t to reset access times).
776  *	When ign is zero, only those times the user has asked for are set, the
777  *	other ones are left alone. We do not assume the un-documented feature
778  *	of many utimes() implementations that consider a 0 time value as a do
779  *	not set request.
780  *
781  *	Unfortunately, there are systems where lutimes() is present but does
782  *	not work on some filesystem types, which cannot be detected at
783  * 	compile time.  This requires passing down symlink knowledge into
784  *	this function to obtain correct operation.  Linux with XFS is one
785  * 	example of such a system.
786  */
787 
788 void
789 set_ftime(char *fnm, time_t mtime, time_t atime, int frc, int slk)
790 {
791 	struct timeval tv[2];
792 	struct stat sb;
793 
794 	tv[0].tv_sec = (long)atime;
795 	tv[0].tv_usec = 0;
796 	tv[1].tv_sec = (long)mtime;
797 	tv[1].tv_usec = 0;
798 	if (!frc && (!patime || !pmtime)) {
799 		/*
800 		 * if we are not forcing, only set those times the user wants
801 		 * set. We get the current values of the times if we need them.
802 		 */
803 		if (lstat(fnm, &sb) == 0) {
804 #if BSD4_4 && !HAVE_NBTOOL_CONFIG_H
805 			if (!patime)
806 				TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
807 			if (!pmtime)
808 				TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
809 #else
810 			if (!patime)
811 				tv[0].tv_sec = sb.st_atime;
812 			if (!pmtime)
813 				tv[1].tv_sec = sb.st_mtime;
814 #endif
815 		} else
816 			syswarn(0, errno, "Cannot obtain file stats %s", fnm);
817 	}
818 
819 	/*
820 	 * set the times
821 	 */
822 #if HAVE_LUTIMES
823 	if (lutimes(fnm, tv) == 0)
824 		return;
825 	if (errno != ENOSYS)	/* XXX linux: lutimes is per-FS */
826 		goto bad;
827 #endif
828 	if (slk)
829 		return;
830 	if (utimes(fnm, tv) == -1)
831 		goto bad;
832 	return;
833 bad:
834 	syswarn(1, errno, "Access/modification time set failed on: %s", fnm);
835 }
836 
837 /*
838  * set_ids()
839  *	set the uid and gid of a file system node
840  * Return:
841  *	0 when set, -1 on failure
842  */
843 
844 int
845 set_ids(char *fnm, uid_t uid, gid_t gid)
846 {
847 	if (geteuid() == 0)
848 		if (lchown(fnm, uid, gid)) {
849 			(void)fflush(listf);
850 			syswarn(1, errno, "Cannot set file uid/gid of %s",
851 			    fnm);
852 			return -1;
853 		}
854 	return 0;
855 }
856 
857 /*
858  * set_pmode()
859  *	Set file access mode
860  */
861 
862 void
863 set_pmode(char *fnm, mode_t mode)
864 {
865 	mode &= A_BITS;
866 	if (lchmod(fnm, mode)) {
867 		(void)fflush(listf);
868 		syswarn(1, errno, "Cannot set permissions on %s", fnm);
869 	}
870 	return;
871 }
872 
873 /*
874  * set_chflags()
875  *	Set 4.4BSD file flags
876  */
877 void
878 set_chflags(char *fnm, u_int32_t flags)
879 {
880 
881 #if 0
882 	if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
883 		syswarn(1, errno, "Cannot set file flags on %s", fnm);
884 #endif
885 	return;
886 }
887 
888 /*
889  * file_write()
890  *	Write/copy a file (during copy or archive extract). This routine knows
891  *	how to copy files with lseek holes in it. (Which are read as file
892  *	blocks containing all 0's but do not have any file blocks associated
893  *	with the data). Typical examples of these are files created by dbm
894  *	variants (.pag files). While the file size of these files are huge, the
895  *	actual storage is quite small (the files are sparse). The problem is
896  *	the holes read as all zeros so are probably stored on the archive that
897  *	way (there is no way to determine if the file block is really a hole,
898  *	we only know that a file block of all zero's can be a hole).
899  *	At this writing, no major archive format knows how to archive files
900  *	with holes. However, on extraction (or during copy, -rw) we have to
901  *	deal with these files. Without detecting the holes, the files can
902  *	consume a lot of file space if just written to disk. This replacement
903  *	for write when passed the basic allocation size of a file system block,
904  *	uses lseek whenever it detects the input data is all 0 within that
905  *	file block. In more detail, the strategy is as follows:
906  *	While the input is all zero keep doing an lseek. Keep track of when we
907  *	pass over file block boundaries. Only write when we hit a non zero
908  *	input. once we have written a file block, we continue to write it to
909  *	the end (we stop looking at the input). When we reach the start of the
910  *	next file block, start checking for zero blocks again. Working on file
911  *	block boundaries significantly reduces the overhead when copying files
912  *	that are NOT very sparse. This overhead (when compared to a write) is
913  *	almost below the measurement resolution on many systems. Without it,
914  *	files with holes cannot be safely copied. It does has a side effect as
915  *	it can put holes into files that did not have them before, but that is
916  *	not a problem since the file contents are unchanged (in fact it saves
917  *	file space). (Except on paging files for diskless clients. But since we
918  *	cannot determine one of those file from here, we ignore them). If this
919  *	ever ends up on a system where CTG files are supported and the holes
920  *	are not desired, just do a conditional test in those routines that
921  *	call file_write() and have it call write() instead. BEFORE CLOSING THE
922  *	FILE, make sure to call file_flush() when the last write finishes with
923  *	an empty block. A lot of file systems will not create an lseek hole at
924  *	the end. In this case we drop a single 0 at the end to force the
925  *	trailing 0's in the file.
926  *	---Parameters---
927  *	rem: how many bytes left in this file system block
928  *	isempt: have we written to the file block yet (is it empty)
929  *	sz: basic file block allocation size
930  *	cnt: number of bytes on this write
931  *	str: buffer to write
932  * Return:
933  *	number of bytes written, -1 on write (or lseek) error.
934  */
935 
936 int
937 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
938 	char *name)
939 {
940 	char *pt;
941 	char *end;
942 	int wcnt;
943 	char *st = str;
944 	char **strp;
945 	size_t *lenp;
946 
947 	/*
948 	 * while we have data to process
949 	 */
950 	while (cnt) {
951 		if (!*rem) {
952 			/*
953 			 * We are now at the start of file system block again
954 			 * (or what we think one is...). start looking for
955 			 * empty blocks again
956 			 */
957 			*isempt = 1;
958 			*rem = sz;
959 		}
960 
961 		/*
962 		 * only examine up to the end of the current file block or
963 		 * remaining characters to write, whatever is smaller
964 		 */
965 		wcnt = MIN(cnt, *rem);
966 		cnt -= wcnt;
967 		*rem -= wcnt;
968 		if (*isempt) {
969 			/*
970 			 * have not written to this block yet, so we keep
971 			 * looking for zero's
972 			 */
973 			pt = st;
974 			end = st + wcnt;
975 
976 			/*
977 			 * look for a zero filled buffer
978 			 */
979 			while ((pt < end) && (*pt == '\0'))
980 				++pt;
981 
982 			if (pt == end) {
983 				/*
984 				 * skip, buf is empty so far
985 				 */
986 				if (fd > -1 &&
987 				    lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
988 					syswarn(1, errno, "File seek on %s",
989 					    name);
990 					return -1;
991 				}
992 				st = pt;
993 				continue;
994 			}
995 			/*
996 			 * drat, the buf is not zero filled
997 			 */
998 			*isempt = 0;
999 		}
1000 
1001 		/*
1002 		 * have non-zero data in this file system block, have to write
1003 		 */
1004 		switch (fd) {
1005 		case -PAX_GLF:
1006 			strp = &gnu_name_string;
1007 			lenp = &gnu_name_length;
1008 			break;
1009 		case -PAX_GLL:
1010 			strp = &gnu_link_string;
1011 			lenp = &gnu_link_length;
1012 			break;
1013 		default:
1014 			strp = NULL;
1015 			lenp = NULL;
1016 			break;
1017 		}
1018 		if (strp) {
1019 			char *nstr = *strp ? realloc(*strp, *lenp + wcnt + 1) :
1020 				malloc(wcnt + 1);
1021 			if (nstr == NULL) {
1022 				tty_warn(1, "Out of memory");
1023 				return -1;
1024 			}
1025 			(void)strlcpy(&nstr[*lenp], st, wcnt + 1);
1026 			*strp = nstr;
1027 			*lenp += wcnt;
1028 		} else if (xwrite(fd, st, wcnt) != wcnt) {
1029 			syswarn(1, errno, "Failed write to file %s", name);
1030 			return -1;
1031 		}
1032 		st += wcnt;
1033 	}
1034 	return st - str;
1035 }
1036 
1037 /*
1038  * file_flush()
1039  *	when the last file block in a file is zero, many file systems will not
1040  *	let us create a hole at the end. To get the last block with zeros, we
1041  *	write the last BYTE with a zero (back up one byte and write a zero).
1042  */
1043 
1044 void
1045 file_flush(int fd, char *fname, int isempt)
1046 {
1047 	static char blnk[] = "\0";
1048 
1049 	/*
1050 	 * silly test, but make sure we are only called when the last block is
1051 	 * filled with all zeros.
1052 	 */
1053 	if (!isempt)
1054 		return;
1055 
1056 	/*
1057 	 * move back one byte and write a zero
1058 	 */
1059 	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1060 		syswarn(1, errno, "Failed seek on file %s", fname);
1061 		return;
1062 	}
1063 
1064 	if (write_with_restart(fd, blnk, 1) < 0)
1065 		syswarn(1, errno, "Failed write to file %s", fname);
1066 	return;
1067 }
1068 
1069 /*
1070  * rdfile_close()
1071  *	close a file we have been reading (to copy or archive). If we have to
1072  *	reset access time (tflag) do so (the times are stored in arcn).
1073  */
1074 
1075 void
1076 rdfile_close(ARCHD *arcn, int *fd)
1077 {
1078 	/*
1079 	 * make sure the file is open
1080 	 */
1081 	if (*fd < 0)
1082 		return;
1083 
1084 	(void)close(*fd);
1085 	*fd = -1;
1086 	if (!tflag)
1087 		return;
1088 
1089 	/*
1090 	 * user wants last access time reset
1091 	 */
1092 	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1, 0);
1093 	return;
1094 }
1095 
1096 /*
1097  * set_crc()
1098  *	read a file to calculate its crc. This is a real drag. Archive formats
1099  *	that have this, end up reading the file twice (we have to write the
1100  *	header WITH the crc before writing the file contents. Oh well...
1101  * Return:
1102  *	0 if was able to calculate the crc, -1 otherwise
1103  */
1104 
1105 int
1106 set_crc(ARCHD *arcn, int fd)
1107 {
1108 	int i;
1109 	int res;
1110 	off_t cpcnt = 0L;
1111 	u_long size;
1112 	unsigned long crc = 0L;
1113 	char tbuf[FILEBLK];
1114 	struct stat sb;
1115 
1116 	if (fd < 0) {
1117 		/*
1118 		 * hmm, no fd, should never happen. well no crc then.
1119 		 */
1120 		arcn->crc = 0L;
1121 		return 0;
1122 	}
1123 
1124 	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1125 		size = (u_long)sizeof(tbuf);
1126 
1127 	/*
1128 	 * read all the bytes we think that there are in the file. If the user
1129 	 * is trying to archive an active file, forget this file.
1130 	 */
1131 	for(;;) {
1132 		if ((res = read(fd, tbuf, size)) <= 0)
1133 			break;
1134 		cpcnt += res;
1135 		for (i = 0; i < res; ++i)
1136 			crc += (tbuf[i] & 0xff);
1137 	}
1138 
1139 	/*
1140 	 * safety check. we want to avoid archiving files that are active as
1141 	 * they can create inconsistent archive copies.
1142 	 */
1143 	if (cpcnt != arcn->sb.st_size)
1144 		tty_warn(1, "File changed size %s", arcn->org_name);
1145 	else if (fstat(fd, &sb) < 0)
1146 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1147 	else if (arcn->sb.st_mtime != sb.st_mtime)
1148 		tty_warn(1, "File %s was modified during read", arcn->org_name);
1149 	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1150 		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1151 	else {
1152 		arcn->crc = crc;
1153 		return 0;
1154 	}
1155 	return -1;
1156 }
1157