xref: /netbsd-src/bin/pax/file_subs.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: file_subs.c,v 1.52 2004/09/22 15:03:18 christos 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 1.52 2004/09/22 15:03:18 christos 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 /*
88  * file_creat()
89  *	Create and open a file.
90  * Return:
91  *	file descriptor or -1 for failure
92  */
93 
94 int
95 file_creat(ARCHD *arcn, int write_to_hardlink)
96 {
97 	int fd = -1;
98 	int oerrno;
99 
100 	/*
101 	 * Some horribly busted tar implementations, have directory nodes
102 	 * that end in a /, but they mark as files. Compensate for that
103 	 * by not creating a directory node at this point, but a file node,
104 	 * and not creating the temp file.
105 	 */
106 	if (arcn->nlen != 0 && arcn->name[arcn->nlen - 1] == '/') {
107 		if (!warn_broken) {
108 			tty_warn(0, "Archive was created with a broken tar;"
109 			    " file `%s' is a directory, but marked as plain.",
110 			    arcn->name);
111 			warn_broken = 1;
112 		}
113 		return -1;
114 	}
115 
116 	/*
117 	 * In "cpio" archives it's usually the last record of a set of
118 	 * hardlinks which includes the contents of the file. We cannot
119 	 * use a tempory file in that case because we couldn't link it
120 	 * with the existing other hardlinks after restoring the contents
121 	 * to it. And it's also useless to create the hardlink under a
122 	 * temporary name because the other hardlinks would have partial
123 	 * contents while restoring.
124 	 */
125 	if (write_to_hardlink)
126 		return (open(arcn->name, O_TRUNC | O_EXCL | O_RDWR, 0));
127 
128 	/*
129 	 * Create a temporary file name so that the file doesn't have partial
130 	 * contents while restoring.
131 	 */
132 	arcn->tmp_name = malloc(arcn->nlen + 8);
133 	if (arcn->tmp_name == NULL) {
134 		syswarn(1, errno, "Cannot malloc %d bytes", arcn->nlen + 8);
135 		return(-1);
136 	}
137 	if (xtmp_name != NULL)
138 		abort();
139 	xtmp_name = arcn->tmp_name;
140 
141 	for (;;) {
142 		/*
143 		 * try to create the temporary file we use to restore the
144 		 * contents info.  if this fails, keep checking all the nodes
145 		 * in the path until chk_path() finds that it cannot fix
146 		 * anything further.  if that happens we just give up.
147 		 */
148 		(void)snprintf(arcn->tmp_name, arcn->nlen + 8, "%s.XXXXXX",
149 		    arcn->name);
150 		fd = mkstemp(arcn->tmp_name);
151 		if (fd >= 0)
152 			break;
153 		oerrno = errno;
154 		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
155 			(void)fflush(listf);
156 			syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name);
157 			xtmp_name = NULL;
158 			free(arcn->tmp_name);
159 			arcn->tmp_name = NULL;
160 			return(-1);
161 		}
162 	}
163 	return(fd);
164 }
165 
166 /*
167  * file_close()
168  *	Close file descriptor to a file just created by pax. Sets modes,
169  *	ownership and times as required.
170  * Return:
171  *	0 for success, -1 for failure
172  */
173 
174 void
175 file_close(ARCHD *arcn, int fd)
176 {
177 	char *tmp_name;
178 	int res;
179 
180 	if (fd < 0)
181 		return;
182 
183 	tmp_name = (arcn->tmp_name != NULL) ? arcn->tmp_name : arcn->name;
184 
185 	if (close(fd) < 0)
186 		syswarn(0, errno, "Cannot close file descriptor on %s",
187 		    tmp_name);
188 
189 	/*
190 	 * set owner/groups first as this may strip off mode bits we want
191 	 * then set file permission modes. Then set file access and
192 	 * modification times.
193 	 */
194 	if (pids)
195 		res = set_ids(tmp_name, arcn->sb.st_uid, arcn->sb.st_gid);
196 	else
197 		res = 0;
198 
199 	/*
200 	 * IMPORTANT SECURITY NOTE:
201 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
202 	 * set uid/gid bits but restore the file modes (since mkstemp doesn't).
203 	 */
204 	if (!pmode || res)
205 		arcn->sb.st_mode &= ~SETBITS(0);
206 	if (pmode)
207 		set_pmode(tmp_name, arcn->sb.st_mode);
208 	else
209 		set_pmode(tmp_name, arcn->sb.st_mode & FILEBITS(0));
210 	if (patime || pmtime)
211 		set_ftime(tmp_name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
212 
213 	/* Did we write directly to the target file? */
214 	if (arcn->tmp_name == NULL)
215 		return;
216 
217 	/*
218 	 * Finally, now the temp file is fully instantiated rename it to
219 	 * the desired file name.
220 	 */
221 	if (rename(tmp_name, arcn->name) < 0) {
222 		syswarn(0, errno, "Cannot rename %s to %s",
223 		    tmp_name, arcn->name);
224 		(void)unlink(tmp_name);
225 	}
226 
227 #if HAVE_STRUCT_STAT_ST_FLAGS
228 	if (pfflags && arcn->type != PAX_SLK)
229 		set_chflags(arcn->name, arcn->sb.st_flags);
230 #endif
231 
232 	free(arcn->tmp_name);
233 	arcn->tmp_name = NULL;
234 	xtmp_name = NULL;
235 }
236 
237 /*
238  * lnk_creat()
239  *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
240  *	must exist;
241  * Return:
242  *	0 if ok, -1 otherwise
243  */
244 
245 int
246 lnk_creat(ARCHD *arcn, int *payload)
247 {
248 	struct stat sb;
249 
250 	/*
251 	 * Check if this hardlink carries the "payload". In "cpio" archives
252 	 * it's usually the last record of a set of hardlinks which includes
253 	 * the contents of the file.
254 	 *
255 	 */
256 	*payload = S_ISREG(arcn->sb.st_mode) &&
257 	    (arcn->sb.st_size > 0) && (arcn->sb.st_size <= arcn->skip);
258 
259 	/*
260 	 * We may be running as root, so we have to be sure that link target
261 	 * is not a directory, so we lstat and check
262 	 */
263 	if (lstat(arcn->ln_name, &sb) < 0) {
264 		syswarn(1, errno, "Cannot link to %s from %s", arcn->ln_name,
265 		    arcn->name);
266 		return(-1);
267 	}
268 
269 	if (S_ISDIR(sb.st_mode)) {
270 		tty_warn(1, "A hard link to the directory %s is not allowed",
271 		    arcn->ln_name);
272 		return(-1);
273 	}
274 
275 	return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
276 }
277 
278 /*
279  * cross_lnk()
280  *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
281  *	with the -l flag. No warning or error if this does not succeed (we will
282  *	then just create the file)
283  * Return:
284  *	1 if copy() should try to create this file node
285  *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
286  */
287 
288 int
289 cross_lnk(ARCHD *arcn)
290 {
291 	/*
292 	 * try to make a link to original file (-l flag in copy mode). make
293 	 * sure we do not try to link to directories in case we are running as
294 	 * root (and it might succeed).
295 	 */
296 	if (arcn->type == PAX_DIR)
297 		return(1);
298 	return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
299 }
300 
301 /*
302  * chk_same()
303  *	In copy mode if we are not trying to make hard links between the src
304  *	and destinations, make sure we are not going to overwrite ourselves by
305  *	accident. This slows things down a little, but we have to protect all
306  *	those people who make typing errors.
307  * Return:
308  *	1 the target does not exist, go ahead and copy
309  *	0 skip it file exists (-k) or may be the same as source file
310  */
311 
312 int
313 chk_same(ARCHD *arcn)
314 {
315 	struct stat sb;
316 
317 	/*
318 	 * if file does not exist, return. if file exists and -k, skip it
319 	 * quietly
320 	 */
321 	if (lstat(arcn->name, &sb) < 0)
322 		return(1);
323 	if (kflag)
324 		return(0);
325 
326 	/*
327 	 * better make sure the user does not have src == dest by mistake
328 	 */
329 	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
330 		tty_warn(1, "Unable to copy %s, file would overwrite itself",
331 		    arcn->name);
332 		return(0);
333 	}
334 	return(1);
335 }
336 
337 /*
338  * mk_link()
339  *	try to make a hard link between two files. if ign set, we do not
340  *	complain.
341  * Return:
342  *	0 if successful (or we are done with this file but no error, such as
343  *	finding the from file exists and the user has set -k).
344  *	1 when ign was set to indicates we could not make the link but we
345  *	should try to copy/extract the file as that might work (and is an
346  *	allowed option). -1 an error occurred.
347  */
348 
349 static int
350 mk_link(char *to, struct stat *to_sb, char *from, int ign)
351 {
352 	struct stat sb;
353 	int oerrno;
354 
355 	/*
356 	 * if from file exists, it has to be unlinked to make the link. If the
357 	 * file exists and -k is set, skip it quietly
358 	 */
359 	if (lstat(from, &sb) == 0) {
360 		if (kflag)
361 			return(0);
362 
363 		/*
364 		 * make sure it is not the same file, protect the user
365 		 */
366 		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
367 			tty_warn(1, "Cannot link file %s to itself", to);
368 			return(-1);
369 		}
370 
371 		/*
372 		 * try to get rid of the file, based on the type
373 		 */
374 		if (S_ISDIR(sb.st_mode) && strcmp(from, ".") != 0) {
375 			if (rmdir(from) < 0) {
376 				syswarn(1, errno, "Cannot remove %s", from);
377 				return(-1);
378 			}
379 		} else if (unlink(from) < 0) {
380 			if (!ign) {
381 				syswarn(1, errno, "Cannot remove %s", from);
382 				return(-1);
383 			}
384 			return(1);
385 		}
386 	}
387 
388 	/*
389 	 * from file is gone (or did not exist), try to make the hard link.
390 	 * if it fails, check the path and try it again (if chk_path() says to
391 	 * try again)
392 	 */
393 	for (;;) {
394 		if (link(to, from) == 0)
395 			break;
396 		oerrno = errno;
397 		if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
398 			continue;
399 		if (!ign) {
400 			syswarn(1, oerrno, "Cannot link to %s from %s", to,
401 			    from);
402 			return(-1);
403 		}
404 		return(1);
405 	}
406 
407 	/*
408 	 * all right the link was made
409 	 */
410 	return(0);
411 }
412 
413 /*
414  * node_creat()
415  *	create an entry in the file system (other than a file or hard link).
416  *	If successful, sets uid/gid modes and times as required.
417  * Return:
418  *	0 if ok, -1 otherwise
419  */
420 
421 int
422 node_creat(ARCHD *arcn)
423 {
424 	int res;
425 	int ign = 0;
426 	int oerrno;
427 	int pass = 0;
428 	mode_t file_mode;
429 	struct stat sb;
430 	char target[MAXPATHLEN];
431 	char *nm = arcn->name;
432 	int len;
433 
434 	/*
435 	 * create node based on type, if that fails try to unlink the node and
436 	 * try again. finally check the path and try again. As noted in the
437 	 * file and link creation routines, this method seems to exhibit the
438 	 * best performance in general use workloads.
439 	 */
440 	file_mode = arcn->sb.st_mode & FILEBITS(arcn->type == PAX_DIR);
441 
442 	for (;;) {
443 		switch (arcn->type) {
444 		case PAX_DIR:
445 			/*
446 			 * If -h (or -L) was given in tar-mode, follow the
447 			 * potential symlink chain before trying to create the
448 			 * directory.
449 			 */
450 			if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
451 				while (lstat(nm, &sb) == 0 &&
452 				    S_ISLNK(sb.st_mode)) {
453 					len = readlink(nm, target,
454 					    sizeof target - 1);
455 					if (len == -1) {
456 						syswarn(0, errno,
457 						   "cannot follow symlink %s "
458 						   "in chain for %s",
459 						    nm, arcn->name);
460 						res = -1;
461 						goto badlink;
462 					}
463 					target[len] = '\0';
464 					nm = target;
465 				}
466 			}
467 			res = mkdir(nm, file_mode);
468 
469 badlink:
470 			if (ign)
471 				res = 0;
472 			break;
473 		case PAX_CHR:
474 			file_mode |= S_IFCHR;
475 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
476 			break;
477 		case PAX_BLK:
478 			file_mode |= S_IFBLK;
479 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
480 			break;
481 		case PAX_FIF:
482 			res = mkfifo(nm, file_mode);
483 			break;
484 		case PAX_SCK:
485 			/*
486 			 * Skip sockets, operation has no meaning under BSD
487 			 */
488 			tty_warn(0,
489 			    "%s skipped. Sockets cannot be copied or extracted",
490 			    nm);
491 			return (-1);
492 		case PAX_SLK:
493 			res = symlink(arcn->ln_name, nm);
494 			break;
495 		case PAX_CTG:
496 		case PAX_HLK:
497 		case PAX_HRG:
498 		case PAX_REG:
499 		default:
500 			/*
501 			 * we should never get here
502 			 */
503 			tty_warn(0, "%s has an unknown file type, skipping",
504 			    nm);
505 			return (-1);
506 		}
507 
508 		/*
509 		 * if we were able to create the node break out of the loop,
510 		 * otherwise try to unlink the node and try again. if that
511 		 * fails check the full path and try a final time.
512 		 */
513 		if (res == 0)
514 			break;
515 
516 		/*
517 		 * we failed to make the node
518 		 */
519 		oerrno = errno;
520 		switch (pass++) {
521 		case 0:
522 			if ((ign = unlnk_exist(nm, arcn->type)) < 0)
523 				return (-1);
524 			continue;
525 
526 		case 1:
527 			if (nodirs ||
528 			    chk_path(nm, arcn->sb.st_uid,
529 			    arcn->sb.st_gid) < 0) {
530 				syswarn(1, oerrno, "Cannot create %s", nm);
531 				return (-1);
532 			}
533 			continue;
534 		}
535 
536 		/*
537 		 * it must be a file that exists but we can't create or
538 		 * remove, but we must avoid the infinite loop.
539 		 */
540 		break;
541 	}
542 
543 	/*
544 	 * we were able to create the node. set uid/gid, modes and times
545 	 */
546 	if (pids)
547 		res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
548 	else
549 		res = 0;
550 
551 	/*
552 	 * IMPORTANT SECURITY NOTE:
553 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
554 	 * set uid/gid bits
555 	 */
556 	if (!pmode || res)
557 		arcn->sb.st_mode &= ~SETBITS(arcn->type == PAX_DIR);
558 	if (pmode)
559 		set_pmode(arcn->name, arcn->sb.st_mode);
560 
561 	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
562 		/*
563 		 * Dirs must be processed again at end of extract to set times
564 		 * and modes to agree with those stored in the archive. However
565 		 * to allow extract to continue, we may have to also set owner
566 		 * rights. This allows nodes in the archive that are children
567 		 * of this directory to be extracted without failure. Both time
568 		 * and modes will be fixed after the entire archive is read and
569 		 * before pax exits.
570 		 */
571 		if (access(nm, R_OK | W_OK | X_OK) < 0) {
572 			if (lstat(nm, &sb) < 0) {
573 				syswarn(0, errno,"Cannot access %s (stat)",
574 				    arcn->name);
575 				set_pmode(nm,file_mode | S_IRWXU);
576 			} else {
577 				/*
578 				 * We have to add rights to the dir, so we make
579 				 * sure to restore the mode. The mode must be
580 				 * restored AS CREATED and not as stored if
581 				 * pmode is not set.
582 				 */
583 				set_pmode(nm, ((sb.st_mode &
584 				    FILEBITS(arcn->type == PAX_DIR)) |
585 				    S_IRWXU));
586 				if (!pmode)
587 					arcn->sb.st_mode = sb.st_mode;
588 			}
589 
590 			/*
591 			 * we have to force the mode to what was set here,
592 			 * since we changed it from the default as created.
593 			 */
594 			add_dir(nm, arcn->nlen, &(arcn->sb), 1);
595 		} else if (pmode || patime || pmtime)
596 			add_dir(nm, arcn->nlen, &(arcn->sb), 0);
597 	}
598 
599 #if HAVE_LUTIMES
600 	if (patime || pmtime)
601 #else
602 	if ((patime || pmtime) && arcn->type != PAX_SLK)
603 #endif
604 		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
605 
606 #if HAVE_STRUCT_STAT_ST_FLAGS
607 	if (pfflags && arcn->type != PAX_SLK)
608 		set_chflags(arcn->name, arcn->sb.st_flags);
609 #endif
610 	return(0);
611 }
612 
613 /*
614  * unlnk_exist()
615  *	Remove node from file system with the specified name. We pass the type
616  *	of the node that is going to replace it. When we try to create a
617  *	directory and find that it already exists, we allow processing to
618  *	continue as proper modes etc will always be set for it later on.
619  * Return:
620  *	0 is ok to proceed, no file with the specified name exists
621  *	-1 we were unable to remove the node, or we should not remove it (-k)
622  *	1 we found a directory and we were going to create a directory.
623  */
624 
625 int
626 unlnk_exist(char *name, int type)
627 {
628 	struct stat sb;
629 
630 	/*
631 	 * the file does not exist, or -k we are done
632 	 */
633 	if (lstat(name, &sb) < 0)
634 		return(0);
635 	if (kflag)
636 		return(-1);
637 
638 	if (S_ISDIR(sb.st_mode)) {
639 		/*
640 		 * try to remove a directory, if it fails and we were going to
641 		 * create a directory anyway, tell the caller (return a 1).
642 		 *
643 		 * don't try to remove the directory if the name is "."
644 		 * otherwise later file/directory creation fails.
645 		 */
646 		if (strcmp(name, ".") == 0)
647 			return(1);
648 		if (rmdir(name) < 0) {
649 			if (type == PAX_DIR)
650 				return(1);
651 			syswarn(1, errno, "Cannot remove directory %s", name);
652 			return(-1);
653 		}
654 		return(0);
655 	}
656 
657 	/*
658 	 * try to get rid of all non-directory type nodes
659 	 */
660 	if (unlink(name) < 0) {
661 		(void)fflush(listf);
662 		syswarn(1, errno, "Cannot unlink %s", name);
663 		return(-1);
664 	}
665 	return(0);
666 }
667 
668 /*
669  * chk_path()
670  *	We were trying to create some kind of node in the file system and it
671  *	failed. chk_path() makes sure the path up to the node exists and is
672  *	writable. When we have to create a directory that is missing along the
673  *	path somewhere, the directory we create will be set to the same
674  *	uid/gid as the file has (when uid and gid are being preserved).
675  *	NOTE: this routine is a real performance loss. It is only used as a
676  *	last resort when trying to create entries in the file system.
677  * Return:
678  *	-1 when it could find nothing it is allowed to fix.
679  *	0 otherwise
680  */
681 
682 int
683 chk_path(char *name, uid_t st_uid, gid_t st_gid)
684 {
685 	char *spt = name;
686 	struct stat sb;
687 	int retval = -1;
688 
689 	/*
690 	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
691 	 */
692 	if (*spt == '/')
693 		++spt;
694 
695 	for(;;) {
696 		/*
697 		 * work forward from the first / and check each part of
698 		 * the path
699 		 */
700 		spt = strchr(spt, '/');
701 		if (spt == NULL)
702 			break;
703 		*spt = '\0';
704 
705 		/*
706 		 * if it exists we assume it is a directory, it is not within
707 		 * the spec (at least it seems to read that way) to alter the
708 		 * file system for nodes NOT EXPLICITLY stored on the archive.
709 		 * If that assumption is changed, you would test the node here
710 		 * and figure out how to get rid of it (probably like some
711 		 * recursive unlink()) or fix up the directory permissions if
712 		 * required (do an access()).
713 		 */
714 		if (lstat(name, &sb) == 0) {
715 			*(spt++) = '/';
716 			continue;
717 		}
718 
719 		/*
720 		 * the path fails at this point, see if we can create the
721 		 * needed directory and continue on
722 		 */
723 		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
724 			*spt = '/';
725 			retval = -1;
726 			break;
727 		}
728 
729 		/*
730 		 * we were able to create the directory. We will tell the
731 		 * caller that we found something to fix, and it is ok to try
732 		 * and create the node again.
733 		 */
734 		retval = 0;
735 		if (pids)
736 			(void)set_ids(name, st_uid, st_gid);
737 
738 		/*
739 		 * make sure the user doesn't have some strange umask that
740 		 * causes this newly created directory to be unusable. We fix
741 		 * the modes and restore them back to the creation default at
742 		 * the end of pax
743 		 */
744 		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
745 		    (lstat(name, &sb) == 0)) {
746 			set_pmode(name, ((sb.st_mode & FILEBITS(0)) |
747 			    S_IRWXU));
748 			add_dir(name, spt - name, &sb, 1);
749 		}
750 		*(spt++) = '/';
751 		continue;
752 	}
753 	return(retval);
754 }
755 
756 /*
757  * set_ftime()
758  *	Set the access time and modification time for a named file. If frc
759  *	is non-zero we force these times to be set even if the user did not
760  *	request access and/or modification time preservation (this is also
761  *	used by -t to reset access times).
762  *	When ign is zero, only those times the user has asked for are set, the
763  *	other ones are left alone. We do not assume the un-documented feature
764  *	of many utimes() implementations that consider a 0 time value as a do
765  *	not set request.
766  */
767 
768 void
769 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
770 {
771 	struct timeval tv[2];
772 	struct stat sb;
773 
774 	tv[0].tv_sec = (long)atime;
775 	tv[0].tv_usec = 0;
776 	tv[1].tv_sec = (long)mtime;
777 	tv[1].tv_usec = 0;
778 	if (!frc && (!patime || !pmtime)) {
779 		/*
780 		 * if we are not forcing, only set those times the user wants
781 		 * set. We get the current values of the times if we need them.
782 		 */
783 		if (lstat(fnm, &sb) == 0) {
784 #if BSD4_4 && !HAVE_NBTOOL_CONFIG_H
785 			if (!patime)
786 				TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
787 			if (!pmtime)
788 				TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
789 #else
790 			if (!patime)
791 				tv[0].tv_sec = sb.st_atime;
792 			if (!pmtime)
793 				tv[1].tv_sec = sb.st_mtime;
794 #endif
795 		} else
796 			syswarn(0, errno, "Cannot obtain file stats %s", fnm);
797 	}
798 
799 	/*
800 	 * set the times
801 	 */
802 #if HAVE_LUTIMES
803 	if (lutimes(fnm, tv))
804 #else
805 	if (utimes(fnm, tv))
806 #endif
807 		syswarn(1, errno, "Access/modification time set failed on: %s",
808 		    fnm);
809 	return;
810 }
811 
812 /*
813  * set_ids()
814  *	set the uid and gid of a file system node
815  * Return:
816  *	0 when set, -1 on failure
817  */
818 
819 int
820 set_ids(char *fnm, uid_t uid, gid_t gid)
821 {
822 	if (geteuid() == 0)
823 		if (lchown(fnm, uid, gid)) {
824 			(void)fflush(listf);
825 			syswarn(1, errno, "Cannot set file uid/gid of %s",
826 			    fnm);
827 			return(-1);
828 		}
829 	return(0);
830 }
831 
832 /*
833  * set_pmode()
834  *	Set file access mode
835  */
836 
837 void
838 set_pmode(char *fnm, mode_t mode)
839 {
840 	mode &= A_BITS;
841 	if (lchmod(fnm, mode)) {
842 		(void)fflush(listf);
843 		syswarn(1, errno, "Cannot set permissions on %s", fnm);
844 	}
845 	return;
846 }
847 
848 /*
849  * set_chflags()
850  *	Set 4.4BSD file flags
851  */
852 void
853 set_chflags(char *fnm, u_int32_t flags)
854 {
855 
856 #if 0
857 	if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
858 		syswarn(1, errno, "Cannot set file flags on %s", fnm);
859 #endif
860 	return;
861 }
862 
863 /*
864  * file_write()
865  *	Write/copy a file (during copy or archive extract). This routine knows
866  *	how to copy files with lseek holes in it. (Which are read as file
867  *	blocks containing all 0's but do not have any file blocks associated
868  *	with the data). Typical examples of these are files created by dbm
869  *	variants (.pag files). While the file size of these files are huge, the
870  *	actual storage is quite small (the files are sparse). The problem is
871  *	the holes read as all zeros so are probably stored on the archive that
872  *	way (there is no way to determine if the file block is really a hole,
873  *	we only know that a file block of all zero's can be a hole).
874  *	At this writing, no major archive format knows how to archive files
875  *	with holes. However, on extraction (or during copy, -rw) we have to
876  *	deal with these files. Without detecting the holes, the files can
877  *	consume a lot of file space if just written to disk. This replacement
878  *	for write when passed the basic allocation size of a file system block,
879  *	uses lseek whenever it detects the input data is all 0 within that
880  *	file block. In more detail, the strategy is as follows:
881  *	While the input is all zero keep doing an lseek. Keep track of when we
882  *	pass over file block boundaries. Only write when we hit a non zero
883  *	input. once we have written a file block, we continue to write it to
884  *	the end (we stop looking at the input). When we reach the start of the
885  *	next file block, start checking for zero blocks again. Working on file
886  *	block boundaries significantly reduces the overhead when copying files
887  *	that are NOT very sparse. This overhead (when compared to a write) is
888  *	almost below the measurement resolution on many systems. Without it,
889  *	files with holes cannot be safely copied. It does has a side effect as
890  *	it can put holes into files that did not have them before, but that is
891  *	not a problem since the file contents are unchanged (in fact it saves
892  *	file space). (Except on paging files for diskless clients. But since we
893  *	cannot determine one of those file from here, we ignore them). If this
894  *	ever ends up on a system where CTG files are supported and the holes
895  *	are not desired, just do a conditional test in those routines that
896  *	call file_write() and have it call write() instead. BEFORE CLOSING THE
897  *	FILE, make sure to call file_flush() when the last write finishes with
898  *	an empty block. A lot of file systems will not create an lseek hole at
899  *	the end. In this case we drop a single 0 at the end to force the
900  *	trailing 0's in the file.
901  *	---Parameters---
902  *	rem: how many bytes left in this file system block
903  *	isempt: have we written to the file block yet (is it empty)
904  *	sz: basic file block allocation size
905  *	cnt: number of bytes on this write
906  *	str: buffer to write
907  * Return:
908  *	number of bytes written, -1 on write (or lseek) error.
909  */
910 
911 int
912 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
913 	char *name)
914 {
915 	char *pt;
916 	char *end;
917 	int wcnt;
918 	char *st = str;
919 	char **strp;
920 	size_t *lenp;
921 
922 	/*
923 	 * while we have data to process
924 	 */
925 	while (cnt) {
926 		if (!*rem) {
927 			/*
928 			 * We are now at the start of file system block again
929 			 * (or what we think one is...). start looking for
930 			 * empty blocks again
931 			 */
932 			*isempt = 1;
933 			*rem = sz;
934 		}
935 
936 		/*
937 		 * only examine up to the end of the current file block or
938 		 * remaining characters to write, whatever is smaller
939 		 */
940 		wcnt = MIN(cnt, *rem);
941 		cnt -= wcnt;
942 		*rem -= wcnt;
943 		if (*isempt) {
944 			/*
945 			 * have not written to this block yet, so we keep
946 			 * looking for zero's
947 			 */
948 			pt = st;
949 			end = st + wcnt;
950 
951 			/*
952 			 * look for a zero filled buffer
953 			 */
954 			while ((pt < end) && (*pt == '\0'))
955 				++pt;
956 
957 			if (pt == end) {
958 				/*
959 				 * skip, buf is empty so far
960 				 */
961 				if (fd > -1 &&
962 				    lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
963 					syswarn(1, errno, "File seek on %s",
964 					    name);
965 					return(-1);
966 				}
967 				st = pt;
968 				continue;
969 			}
970 			/*
971 			 * drat, the buf is not zero filled
972 			 */
973 			*isempt = 0;
974 		}
975 
976 		/*
977 		 * have non-zero data in this file system block, have to write
978 		 */
979 		switch (fd) {
980 		case -1:
981 			strp = &gnu_name_string;
982 			lenp = &gnu_name_length;
983 			break;
984 		case -2:
985 			strp = &gnu_link_string;
986 			lenp = &gnu_link_length;
987 			break;
988 		default:
989 			strp = NULL;
990 			lenp = NULL;
991 			break;
992 		}
993 		if (strp) {
994 			char *nstr = *strp ? realloc(*strp, *lenp + wcnt + 1) :
995 				malloc(wcnt + 1);
996 			if (nstr == NULL) {
997 				tty_warn(1, "Out of memory");
998 				return(-1);
999 			}
1000 			(void)strlcpy(&nstr[*lenp], st, wcnt + 1);
1001 			*strp = nstr;
1002 			*lenp += wcnt;
1003 		} else if (xwrite(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_with_restart(fd, blnk, 1) < 0)
1040 		syswarn(1, errno, "Failed write to file %s", fname);
1041 	return;
1042 }
1043 
1044 /*
1045  * rdfile_close()
1046  *	close a file we have been reading (to copy or archive). If we have to
1047  *	reset access time (tflag) do so (the times are stored in arcn).
1048  */
1049 
1050 void
1051 rdfile_close(ARCHD *arcn, int *fd)
1052 {
1053 	/*
1054 	 * make sure the file is open
1055 	 */
1056 	if (*fd < 0)
1057 		return;
1058 
1059 	(void)close(*fd);
1060 	*fd = -1;
1061 	if (!tflag)
1062 		return;
1063 
1064 	/*
1065 	 * user wants last access time reset
1066 	 */
1067 	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
1068 	return;
1069 }
1070 
1071 /*
1072  * set_crc()
1073  *	read a file to calculate its crc. This is a real drag. Archive formats
1074  *	that have this, end up reading the file twice (we have to write the
1075  *	header WITH the crc before writing the file contents. Oh well...
1076  * Return:
1077  *	0 if was able to calculate the crc, -1 otherwise
1078  */
1079 
1080 int
1081 set_crc(ARCHD *arcn, int fd)
1082 {
1083 	int i;
1084 	int res;
1085 	off_t cpcnt = 0L;
1086 	u_long size;
1087 	unsigned long crc = 0L;
1088 	char tbuf[FILEBLK];
1089 	struct stat sb;
1090 
1091 	if (fd < 0) {
1092 		/*
1093 		 * hmm, no fd, should never happen. well no crc then.
1094 		 */
1095 		arcn->crc = 0L;
1096 		return(0);
1097 	}
1098 
1099 	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1100 		size = (u_long)sizeof(tbuf);
1101 
1102 	/*
1103 	 * read all the bytes we think that there are in the file. If the user
1104 	 * is trying to archive an active file, forget this file.
1105 	 */
1106 	for(;;) {
1107 		if ((res = read(fd, tbuf, size)) <= 0)
1108 			break;
1109 		cpcnt += res;
1110 		for (i = 0; i < res; ++i)
1111 			crc += (tbuf[i] & 0xff);
1112 	}
1113 
1114 	/*
1115 	 * safety check. we want to avoid archiving files that are active as
1116 	 * they can create inconsistant archive copies.
1117 	 */
1118 	if (cpcnt != arcn->sb.st_size)
1119 		tty_warn(1, "File changed size %s", arcn->org_name);
1120 	else if (fstat(fd, &sb) < 0)
1121 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1122 	else if (arcn->sb.st_mtime != sb.st_mtime)
1123 		tty_warn(1, "File %s was modified during read", arcn->org_name);
1124 	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1125 		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1126 	else {
1127 		arcn->crc = crc;
1128 		return(0);
1129 	}
1130 	return(-1);
1131 }
1132