file_subs.c revision 1.39
1262395Sbapt/*	$NetBSD: file_subs.c,v 1.39 2004/02/13 00:11:30 matt Exp $	*/
2262395Sbapt
3262395Sbapt/*-
4262395Sbapt * Copyright (c) 1992 Keith Muller.
5262395Sbapt * Copyright (c) 1992, 1993
6262395Sbapt *	The Regents of the University of California.  All rights reserved.
7262395Sbapt *
8262395Sbapt * This code is derived from software contributed to Berkeley by
9262395Sbapt * Keith Muller of the University of California, San Diego.
10262395Sbapt *
11262395Sbapt * Redistribution and use in source and binary forms, with or without
12262395Sbapt * modification, are permitted provided that the following conditions
13262395Sbapt * are met:
14262395Sbapt * 1. Redistributions of source code must retain the above copyright
15262395Sbapt *    notice, this list of conditions and the following disclaimer.
16262395Sbapt * 2. Redistributions in binary form must reproduce the above copyright
17262395Sbapt *    notice, this list of conditions and the following disclaimer in the
18262395Sbapt *    documentation and/or other materials provided with the distribution.
19262395Sbapt * 3. Neither the name of the University nor the names of its contributors
20262395Sbapt *    may be used to endorse or promote products derived from this software
21262395Sbapt *    without specific prior written permission.
22262395Sbapt *
23262395Sbapt * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24262395Sbapt * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25262395Sbapt * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26262395Sbapt * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27262395Sbapt * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28262395Sbapt * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29262395Sbapt * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30262395Sbapt * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31262395Sbapt * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32262395Sbapt * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33262395Sbapt * SUCH DAMAGE.
34262395Sbapt */
35262395Sbapt
36262395Sbapt#if HAVE_NBTOOL_CONFIG_H
37262395Sbapt#include "nbtool_config.h"
38262395Sbapt#endif
39262395Sbapt
40262395Sbapt#include <sys/cdefs.h>
41262395Sbapt#if !defined(lint)
42262395Sbapt#if 0
43262395Sbaptstatic char sccsid[] = "@(#)file_subs.c	8.1 (Berkeley) 5/31/93";
44262395Sbapt#else
45262395Sbapt__RCSID("$NetBSD: file_subs.c,v 1.39 2004/02/13 00:11:30 matt Exp $");
46262395Sbapt#endif
47262395Sbapt#endif /* not lint */
48262395Sbapt
49262395Sbapt#include <sys/types.h>
50262395Sbapt#include <sys/time.h>
51262395Sbapt#include <sys/stat.h>
52262395Sbapt#include <unistd.h>
53262395Sbapt#include <sys/param.h>
54262395Sbapt#include <fcntl.h>
55262395Sbapt#include <string.h>
56262395Sbapt#include <stdio.h>
57262395Sbapt#include <ctype.h>
58262395Sbapt#include <errno.h>
59262395Sbapt#include <sys/uio.h>
60262395Sbapt#include <stdlib.h>
61262395Sbapt#include "pax.h"
62262395Sbapt#include "extern.h"
63262395Sbapt#include "options.h"
64262395Sbapt
65262395Sbaptstatic int
66262395Sbaptmk_link(char *,struct stat *,char *, int);
67262395Sbapt
68262395Sbapt/*
69262395Sbapt * routines that deal with file operations such as: creating, removing;
70262395Sbapt * and setting access modes, uid/gid and times of files
71262395Sbapt */
72262395Sbapt
73262395Sbapt#define FILEBITS		(S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
74262395Sbapt#define SETBITS			(S_ISUID | S_ISGID)
75262395Sbapt#define ABITS			(FILEBITS | SETBITS)
76262395Sbapt
77262395Sbapt/*
78262395Sbapt * file_creat()
79262395Sbapt *	Create and open a file.
80262395Sbapt * Return:
81262395Sbapt *	file descriptor or -1 for failure
82262395Sbapt */
83262395Sbapt
84262395Sbaptint
85262395Sbaptfile_creat(ARCHD *arcn)
86262395Sbapt{
87262395Sbapt	int fd = -1;
88262395Sbapt	int oerrno;
89262395Sbapt
90262395Sbapt	/*
91262395Sbapt	 * Create a temporary file name so that the file doesn't have partial
92262395Sbapt	 * contents while restoring.
93262395Sbapt	 */
94262395Sbapt	arcn->tmp_name = malloc(arcn->nlen + 8);
95262395Sbapt	if (arcn->tmp_name == NULL) {
96262395Sbapt		syswarn(1, errno, "Cannot malloc %d bytes", arcn->nlen + 8);
97262395Sbapt		return(-1);
98262395Sbapt	}
99262395Sbapt	(void)snprintf(arcn->tmp_name, arcn->nlen + 8, "%s.XXXXXX", arcn->name);
100262395Sbapt
101262395Sbapt	fd = mkstemp(arcn->tmp_name);
102262395Sbapt	if (fd >= 0)
103262395Sbapt		return(fd);
104262395Sbapt
105262395Sbapt	for (;;) {
106262395Sbapt		/*
107262395Sbapt		 * try to create the temporary file we use to restore the
108262395Sbapt		 * contents info.  if this fails, keep checking all the nodes
109262395Sbapt		 * in the path until chk_path() finds that it cannot fix
110262395Sbapt		 * anything further.  if that happens we just give up.
111262395Sbapt		 */
112262395Sbapt		fd = mkstemp(arcn->tmp_name);
113262395Sbapt		if (fd >= 0)
114262395Sbapt			break;
115262395Sbapt		oerrno = errno;
116262395Sbapt		if (nodirs || chk_path(arcn->tmp_name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
117262395Sbapt			(void)fflush(listf);
118262395Sbapt			syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name);
119262395Sbapt			free(arcn->tmp_name);
120262395Sbapt			arcn->tmp_name = NULL;
121262395Sbapt			return(-1);
122262395Sbapt		}
123262395Sbapt	}
124262395Sbapt	return(fd);
125262395Sbapt}
126262395Sbapt
127262395Sbapt/*
128262395Sbapt * file_close()
129262395Sbapt *	Close file descriptor to a file just created by pax. Sets modes,
130262395Sbapt *	ownership and times as required.
131262395Sbapt * Return:
132262395Sbapt *	0 for success, -1 for failure
133262395Sbapt */
134262395Sbapt
135262395Sbaptvoid
136262395Sbaptfile_close(ARCHD *arcn, int fd)
137262395Sbapt{
138262395Sbapt	int res = 0;
139262395Sbapt
140262395Sbapt	if (fd < 0)
141262395Sbapt		return;
142262395Sbapt	if (close(fd) < 0)
143262395Sbapt		syswarn(0, errno, "Cannot close file descriptor on %s",
144262395Sbapt		    arcn->tmp_name);
145262395Sbapt
146262395Sbapt	/*
147262395Sbapt	 * set owner/groups first as this may strip off mode bits we want
148262395Sbapt	 * then set file permission modes. Then set file access and
149262395Sbapt	 * modification times.
150262395Sbapt	 */
151262395Sbapt	if (pids)
152262395Sbapt		res = set_ids(arcn->tmp_name, arcn->sb.st_uid, arcn->sb.st_gid);
153262395Sbapt
154262395Sbapt	/*
155262395Sbapt	 * IMPORTANT SECURITY NOTE:
156262395Sbapt	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
157262395Sbapt	 * set uid/gid bits but restore the file modes (since mkstemp doesn't).
158262395Sbapt	 */
159262395Sbapt	if (!pmode || res)
160262395Sbapt		arcn->sb.st_mode &= ~(SETBITS);
161262395Sbapt	if (pmode)
162262395Sbapt		set_pmode(arcn->tmp_name, arcn->sb.st_mode);
163262395Sbapt	else
164262395Sbapt		set_pmode(arcn->tmp_name, arcn->sb.st_mode & FILEBITS);
165262395Sbapt	if (patime || pmtime)
166262395Sbapt		set_ftime(arcn->tmp_name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
167262395Sbapt#if HAVE_STRUCT_STAT_ST_FLAGS
168262395Sbapt	if (pfflags && arcn->type != PAX_SLK)
169262395Sbapt		set_chflags(arcn->tmp_name, arcn->sb.st_flags);
170262395Sbapt#endif
171262395Sbapt
172262395Sbapt	/*
173262395Sbapt	 * Finally, now the temp file is fully instantiated rename it to
174262395Sbapt	 * the desired file name.
175262395Sbapt	 */
176262395Sbapt	if (rename(arcn->tmp_name, arcn->name) < 0) {
177262395Sbapt		syswarn(0, errno, "Cannot rename %s to %s",
178262395Sbapt		    arcn->tmp_name, arcn->name);
179262395Sbapt		(void)unlink(arcn->tmp_name);
180262395Sbapt	}
181262395Sbapt
182262395Sbapt	free(arcn->tmp_name);
183262395Sbapt	arcn->tmp_name = NULL;
184262395Sbapt}
185262395Sbapt
186262395Sbapt/*
187262395Sbapt * lnk_creat()
188262395Sbapt *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
189262395Sbapt *	must exist;
190262395Sbapt * Return:
191262395Sbapt *	0 if ok, -1 otherwise
192262395Sbapt */
193262395Sbapt
194262395Sbaptint
195262395Sbaptlnk_creat(ARCHD *arcn)
196262395Sbapt{
197262395Sbapt	struct stat sb;
198262395Sbapt
199262395Sbapt	/*
200262395Sbapt	 * we may be running as root, so we have to be sure that link target
201262395Sbapt	 * is not a directory, so we lstat and check
202262395Sbapt	 */
203262395Sbapt	if (lstat(arcn->ln_name, &sb) < 0) {
204262395Sbapt		syswarn(1, errno, "Cannot link to %s from %s", arcn->ln_name,
205262395Sbapt		    arcn->name);
206262395Sbapt		return(-1);
207262395Sbapt	}
208262395Sbapt
209262395Sbapt	if (S_ISDIR(sb.st_mode)) {
210262395Sbapt		tty_warn(1, "A hard link to the directory %s is not allowed",
211262395Sbapt		    arcn->ln_name);
212262395Sbapt		return(-1);
213262395Sbapt	}
214262395Sbapt
215262395Sbapt	return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
216262395Sbapt}
217262395Sbapt
218262395Sbapt/*
219262395Sbapt * cross_lnk()
220262395Sbapt *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
221262395Sbapt *	with the -l flag. No warning or error if this does not succeed (we will
222262395Sbapt *	then just create the file)
223262395Sbapt * Return:
224262395Sbapt *	1 if copy() should try to create this file node
225262395Sbapt *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
226262395Sbapt */
227262395Sbapt
228262395Sbaptint
229262395Sbaptcross_lnk(ARCHD *arcn)
230262395Sbapt{
231262395Sbapt	/*
232262395Sbapt	 * try to make a link to original file (-l flag in copy mode). make
233262395Sbapt	 * sure we do not try to link to directories in case we are running as
234262395Sbapt	 * root (and it might succeed).
235262395Sbapt	 */
236262395Sbapt	if (arcn->type == PAX_DIR)
237262395Sbapt		return(1);
238262395Sbapt	return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
239262395Sbapt}
240262395Sbapt
241262395Sbapt/*
242262395Sbapt * chk_same()
243262395Sbapt *	In copy mode if we are not trying to make hard links between the src
244262395Sbapt *	and destinations, make sure we are not going to overwrite ourselves by
245262395Sbapt *	accident. This slows things down a little, but we have to protect all
246262395Sbapt *	those people who make typing errors.
247262395Sbapt * Return:
248262395Sbapt *	1 the target does not exist, go ahead and copy
249262395Sbapt *	0 skip it file exists (-k) or may be the same as source file
250262395Sbapt */
251262395Sbapt
252262395Sbaptint
253262395Sbaptchk_same(ARCHD *arcn)
254262395Sbapt{
255262395Sbapt	struct stat sb;
256262395Sbapt
257262395Sbapt	/*
258262395Sbapt	 * if file does not exist, return. if file exists and -k, skip it
259262395Sbapt	 * quietly
260262395Sbapt	 */
261262395Sbapt	if (lstat(arcn->name, &sb) < 0)
262262395Sbapt		return(1);
263262395Sbapt	if (kflag)
264262395Sbapt		return(0);
265262395Sbapt
266262395Sbapt	/*
267262395Sbapt	 * better make sure the user does not have src == dest by mistake
268262395Sbapt	 */
269262395Sbapt	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
270262395Sbapt		tty_warn(1, "Unable to copy %s, file would overwrite itself",
271262395Sbapt		    arcn->name);
272262395Sbapt		return(0);
273262395Sbapt	}
274262395Sbapt	return(1);
275262395Sbapt}
276262395Sbapt
277262395Sbapt/*
278262395Sbapt * mk_link()
279262395Sbapt *	try to make a hard link between two files. if ign set, we do not
280262395Sbapt *	complain.
281262395Sbapt * Return:
282262395Sbapt *	0 if successful (or we are done with this file but no error, such as
283262395Sbapt *	finding the from file exists and the user has set -k).
284262395Sbapt *	1 when ign was set to indicates we could not make the link but we
285262395Sbapt *	should try to copy/extract the file as that might work (and is an
286262395Sbapt *	allowed option). -1 an error occurred.
287262395Sbapt */
288262395Sbapt
289262395Sbaptstatic int
290262395Sbaptmk_link(char *to, struct stat *to_sb, char *from, int ign)
291262395Sbapt{
292262395Sbapt	struct stat sb;
293262395Sbapt	int oerrno;
294262395Sbapt
295262395Sbapt	/*
296262395Sbapt	 * if from file exists, it has to be unlinked to make the link. If the
297262395Sbapt	 * file exists and -k is set, skip it quietly
298262395Sbapt	 */
299262395Sbapt	if (lstat(from, &sb) == 0) {
300262395Sbapt		if (kflag)
301262395Sbapt			return(0);
302262395Sbapt
303262395Sbapt		/*
304262395Sbapt		 * make sure it is not the same file, protect the user
305262395Sbapt		 */
306262395Sbapt		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
307262395Sbapt			tty_warn(1, "Cannot link file %s to itself", to);
308262395Sbapt			return(-1);
309262395Sbapt		}
310262395Sbapt
311262395Sbapt		/*
312262395Sbapt		 * try to get rid of the file, based on the type
313262395Sbapt		 */
314262395Sbapt		if (S_ISDIR(sb.st_mode)) {
315262395Sbapt			if (rmdir(from) < 0) {
316262395Sbapt				syswarn(1, errno, "Cannot remove %s", from);
317262395Sbapt				return(-1);
318262395Sbapt			}
319262395Sbapt		} else if (unlink(from) < 0) {
320262395Sbapt			if (!ign) {
321262395Sbapt				syswarn(1, errno, "Cannot remove %s", from);
322262395Sbapt				return(-1);
323262395Sbapt			}
324262395Sbapt			return(1);
325262395Sbapt		}
326262395Sbapt	}
327262395Sbapt
328262395Sbapt	/*
329262395Sbapt	 * from file is gone (or did not exist), try to make the hard link.
330262395Sbapt	 * if it fails, check the path and try it again (if chk_path() says to
331262395Sbapt	 * try again)
332262395Sbapt	 */
333262395Sbapt	for (;;) {
334262395Sbapt		if (link(to, from) == 0)
335262395Sbapt			break;
336262395Sbapt		oerrno = errno;
337262395Sbapt		if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
338262395Sbapt			continue;
339262395Sbapt		if (!ign) {
340262395Sbapt			syswarn(1, oerrno, "Cannot link to %s from %s", to,
341262395Sbapt			    from);
342262395Sbapt			return(-1);
343262395Sbapt		}
344262395Sbapt		return(1);
345262395Sbapt	}
346262395Sbapt
347262395Sbapt	/*
348262395Sbapt	 * all right the link was made
349262395Sbapt	 */
350262395Sbapt	return(0);
351262395Sbapt}
352262395Sbapt
353262395Sbapt/*
354262395Sbapt * node_creat()
355262395Sbapt *	create an entry in the file system (other than a file or hard link).
356262395Sbapt *	If successful, sets uid/gid modes and times as required.
357262395Sbapt * Return:
358262395Sbapt *	0 if ok, -1 otherwise
359262395Sbapt */
360262395Sbapt
361262395Sbaptint
362262395Sbaptnode_creat(ARCHD *arcn)
363262395Sbapt{
364262395Sbapt	int res;
365262395Sbapt	int ign = 0;
366262395Sbapt	int oerrno;
367262395Sbapt	int pass = 0;
368262395Sbapt	mode_t file_mode;
369262395Sbapt	struct stat sb;
370262395Sbapt	char target[MAXPATHLEN];
371262395Sbapt	char *nm = arcn->name;
372262395Sbapt	int len;
373262395Sbapt
374262395Sbapt	/*
375262395Sbapt	 * create node based on type, if that fails try to unlink the node and
376262395Sbapt	 * try again. finally check the path and try again. As noted in the
377262395Sbapt	 * file and link creation routines, this method seems to exhibit the
378262395Sbapt	 * best performance in general use workloads.
379262395Sbapt	 */
380262395Sbapt	file_mode = arcn->sb.st_mode & FILEBITS;
381262395Sbapt
382262395Sbapt	for (;;) {
383262395Sbapt		switch(arcn->type) {
384262395Sbapt		case PAX_DIR:
385262395Sbapt			/*
386262395Sbapt			 * If -h (or -L) was given in tar-mode, follow the
387262395Sbapt			 * potential symlink chain before trying to create the
388262395Sbapt			 * directory.
389262395Sbapt			 */
390262395Sbapt			if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
391262395Sbapt				while (lstat(nm, &sb) == 0 &&
392262395Sbapt				    S_ISLNK(sb.st_mode)) {
393262395Sbapt					len = readlink(nm, target,
394262395Sbapt					    sizeof target - 1);
395262395Sbapt					if (len == -1) {
396262395Sbapt						syswarn(0, errno,
397262395Sbapt						   "cannot follow symlink %s in chain for %s",
398262395Sbapt						    nm, arcn->name);
399262395Sbapt						res = -1;
400262395Sbapt						goto badlink;
401262395Sbapt					}
402262395Sbapt					target[len] = '\0';
403262395Sbapt					nm = target;
404262395Sbapt				}
405262395Sbapt			}
406262395Sbapt			res = mkdir(nm, file_mode);
407262395Sbapt
408262395Sbaptbadlink:
409262395Sbapt			if (ign)
410262395Sbapt				res = 0;
411262395Sbapt			break;
412262395Sbapt		case PAX_CHR:
413262395Sbapt			file_mode |= S_IFCHR;
414262395Sbapt			res = mknod(nm, file_mode, arcn->sb.st_rdev);
415262395Sbapt			break;
416262395Sbapt		case PAX_BLK:
417262395Sbapt			file_mode |= S_IFBLK;
418262395Sbapt			res = mknod(nm, file_mode, arcn->sb.st_rdev);
419262395Sbapt			break;
420262395Sbapt		case PAX_FIF:
421262395Sbapt			res = mkfifo(nm, file_mode);
422262395Sbapt			break;
423262395Sbapt		case PAX_SCK:
424262395Sbapt			/*
425262395Sbapt			 * Skip sockets, operation has no meaning under BSD
426262395Sbapt			 */
427262395Sbapt			tty_warn(0,
428262395Sbapt			    "%s skipped. Sockets cannot be copied or extracted",
429262395Sbapt			    nm);
430262395Sbapt			return(-1);
431262395Sbapt		case PAX_SLK:
432262395Sbapt			res = symlink(arcn->ln_name, nm);
433262395Sbapt			break;
434262395Sbapt		case PAX_CTG:
435262395Sbapt		case PAX_HLK:
436262395Sbapt		case PAX_HRG:
437262395Sbapt		case PAX_REG:
438262395Sbapt		default:
439262395Sbapt			/*
440262395Sbapt			 * we should never get here
441262395Sbapt			 */
442262395Sbapt			tty_warn(0, "%s has an unknown file type, skipping",
443262395Sbapt			    nm);
444262395Sbapt			return(-1);
445262395Sbapt		}
446262395Sbapt
447262395Sbapt		/*
448262395Sbapt		 * if we were able to create the node break out of the loop,
449262395Sbapt		 * otherwise try to unlink the node and try again. if that
450262395Sbapt		 * fails check the full path and try a final time.
451262395Sbapt		 */
452262395Sbapt		if (res == 0)
453262395Sbapt			break;
454262395Sbapt
455262395Sbapt		/*
456262395Sbapt		 * we failed to make the node
457262395Sbapt		 */
458262395Sbapt		oerrno = errno;
459262395Sbapt		if ((ign = unlnk_exist(nm, arcn->type)) < 0)
460262395Sbapt			return(-1);
461262395Sbapt
462262395Sbapt		if (++pass <= 1)
463262395Sbapt			continue;
464262395Sbapt
465262395Sbapt		if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
466262395Sbapt			syswarn(1, oerrno, "Cannot create %s", nm);
467262395Sbapt			return(-1);
468262395Sbapt		}
469262395Sbapt	}
470262395Sbapt
471262395Sbapt	/*
472262395Sbapt	 * we were able to create the node. set uid/gid, modes and times
473262395Sbapt	 */
474262395Sbapt	if (pids)
475262395Sbapt		res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
476262395Sbapt	else
477262395Sbapt		res = 0;
478262395Sbapt
479262395Sbapt	/*
480262395Sbapt	 * IMPORTANT SECURITY NOTE:
481262395Sbapt	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
482262395Sbapt	 * set uid/gid bits
483262395Sbapt	 */
484262395Sbapt	if (!pmode || res)
485262395Sbapt		arcn->sb.st_mode &= ~(SETBITS);
486262395Sbapt	if (pmode)
487262395Sbapt		set_pmode(arcn->name, arcn->sb.st_mode);
488262395Sbapt
489262395Sbapt	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
490262395Sbapt		/*
491262395Sbapt		 * Dirs must be processed again at end of extract to set times
492262395Sbapt		 * and modes to agree with those stored in the archive. However
493262395Sbapt		 * to allow extract to continue, we may have to also set owner
494262395Sbapt		 * rights. This allows nodes in the archive that are children
495262395Sbapt		 * of this directory to be extracted without failure. Both time
496262395Sbapt		 * and modes will be fixed after the entire archive is read and
497262395Sbapt		 * before pax exits.
498262395Sbapt		 */
499262395Sbapt		if (access(nm, R_OK | W_OK | X_OK) < 0) {
500262395Sbapt			if (lstat(nm, &sb) < 0) {
501262395Sbapt				syswarn(0, errno,"Cannot access %s (stat)",
502262395Sbapt				    arcn->name);
503262395Sbapt				set_pmode(nm,file_mode | S_IRWXU);
504262395Sbapt			} else {
505262395Sbapt				/*
506262395Sbapt				 * We have to add rights to the dir, so we make
507262395Sbapt				 * sure to restore the mode. The mode must be
508262395Sbapt				 * restored AS CREATED and not as stored if
509262395Sbapt				 * pmode is not set.
510262395Sbapt				 */
511262395Sbapt				set_pmode(nm,
512262395Sbapt				    ((sb.st_mode & FILEBITS) | S_IRWXU));
513262395Sbapt				if (!pmode)
514262395Sbapt					arcn->sb.st_mode = sb.st_mode;
515262395Sbapt			}
516262395Sbapt
517262395Sbapt			/*
518262395Sbapt			 * we have to force the mode to what was set here,
519262395Sbapt			 * since we changed it from the default as created.
520262395Sbapt			 */
521262395Sbapt			add_dir(nm, arcn->nlen, &(arcn->sb), 1);
522262395Sbapt		} else if (pmode || patime || pmtime)
523262395Sbapt			add_dir(nm, arcn->nlen, &(arcn->sb), 0);
524262395Sbapt	}
525262395Sbapt
526262395Sbapt#if HAVE_LUTIMES
527262395Sbapt	if (patime || pmtime)
528262395Sbapt#else
529262395Sbapt	if ((patime || pmtime) && arcn->type != PAX_SLK)
530262395Sbapt#endif
531262395Sbapt		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
532262395Sbapt
533262395Sbapt#if HAVE_STRUCT_STAT_ST_FLAGS
534262395Sbapt	if (pfflags && arcn->type != PAX_SLK)
535262395Sbapt		set_chflags(arcn->name, arcn->sb.st_flags);
536262395Sbapt#endif
537262395Sbapt	return(0);
538262395Sbapt}
539262395Sbapt
540262395Sbapt/*
541262395Sbapt * unlnk_exist()
542262395Sbapt *	Remove node from file system with the specified name. We pass the type
543262395Sbapt *	of the node that is going to replace it. When we try to create a
544262395Sbapt *	directory and find that it already exists, we allow processing to
545262395Sbapt *	continue as proper modes etc will always be set for it later on.
546262395Sbapt * Return:
547262395Sbapt *	0 is ok to proceed, no file with the specified name exists
548262395Sbapt *	-1 we were unable to remove the node, or we should not remove it (-k)
549262395Sbapt *	1 we found a directory and we were going to create a directory.
550262395Sbapt */
551262395Sbapt
552262395Sbaptint
553262395Sbaptunlnk_exist(char *name, int type)
554262395Sbapt{
555262395Sbapt	struct stat sb;
556262395Sbapt
557262395Sbapt	/*
558262395Sbapt	 * the file does not exist, or -k we are done
559262395Sbapt	 */
560262395Sbapt	if (lstat(name, &sb) < 0)
561262395Sbapt		return(0);
562262395Sbapt	if (kflag)
563262395Sbapt		return(-1);
564262395Sbapt
565262395Sbapt	if (S_ISDIR(sb.st_mode)) {
566262395Sbapt		/*
567262395Sbapt		 * try to remove a directory, if it fails and we were going to
568262395Sbapt		 * create a directory anyway, tell the caller (return a 1)
569262395Sbapt		 */
570262395Sbapt		if (rmdir(name) < 0) {
571262395Sbapt			if (type == PAX_DIR)
572262395Sbapt				return(1);
573262395Sbapt			syswarn(1, errno, "Cannot remove directory %s", name);
574262395Sbapt			return(-1);
575262395Sbapt		}
576262395Sbapt		return(0);
577262395Sbapt	}
578262395Sbapt
579262395Sbapt	/*
580262395Sbapt	 * try to get rid of all non-directory type nodes
581262395Sbapt	 */
582262395Sbapt	if (unlink(name) < 0) {
583262395Sbapt		(void)fflush(listf);
584262395Sbapt		syswarn(1, errno, "Cannot unlink %s", name);
585262395Sbapt		return(-1);
586262395Sbapt	}
587262395Sbapt	return(0);
588262395Sbapt}
589262395Sbapt
590262395Sbapt/*
591262395Sbapt * chk_path()
592262395Sbapt *	We were trying to create some kind of node in the file system and it
593262395Sbapt *	failed. chk_path() makes sure the path up to the node exists and is
594262395Sbapt *	writable. When we have to create a directory that is missing along the
595262395Sbapt *	path somewhere, the directory we create will be set to the same
596262395Sbapt *	uid/gid as the file has (when uid and gid are being preserved).
597262395Sbapt *	NOTE: this routine is a real performance loss. It is only used as a
598262395Sbapt *	last resort when trying to create entries in the file system.
599262395Sbapt * Return:
600262395Sbapt *	-1 when it could find nothing it is allowed to fix.
601262395Sbapt *	0 otherwise
602262395Sbapt */
603262395Sbapt
604262395Sbaptint
605262395Sbaptchk_path( char *name, uid_t st_uid, gid_t st_gid)
606262395Sbapt{
607262395Sbapt	char *spt = name;
608262395Sbapt	struct stat sb;
609262395Sbapt	int retval = -1;
610262395Sbapt
611262395Sbapt	/*
612262395Sbapt	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
613262395Sbapt	 */
614262395Sbapt	if (*spt == '/')
615262395Sbapt		++spt;
616262395Sbapt
617262395Sbapt	for(;;) {
618262395Sbapt		/*
619262395Sbapt		 * work forward from the first / and check each part of
620262395Sbapt		 * the path
621262395Sbapt		 */
622262395Sbapt		spt = strchr(spt, '/');
623262395Sbapt		if (spt == NULL)
624262395Sbapt			break;
625262395Sbapt		*spt = '\0';
626262395Sbapt
627262395Sbapt		/*
628262395Sbapt		 * if it exists we assume it is a directory, it is not within
629262395Sbapt		 * the spec (at least it seems to read that way) to alter the
630262395Sbapt		 * file system for nodes NOT EXPLICITLY stored on the archive.
631262395Sbapt		 * If that assumption is changed, you would test the node here
632262395Sbapt		 * and figure out how to get rid of it (probably like some
633262395Sbapt		 * recursive unlink()) or fix up the directory permissions if
634262395Sbapt		 * required (do an access()).
635262395Sbapt		 */
636262395Sbapt		if (lstat(name, &sb) == 0) {
637262395Sbapt			*(spt++) = '/';
638262395Sbapt			continue;
639262395Sbapt		}
640262395Sbapt
641262395Sbapt		/*
642262395Sbapt		 * the path fails at this point, see if we can create the
643262395Sbapt		 * needed directory and continue on
644262395Sbapt		 */
645262395Sbapt		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
646262395Sbapt			*spt = '/';
647262395Sbapt			retval = -1;
648262395Sbapt			break;
649262395Sbapt		}
650262395Sbapt
651262395Sbapt		/*
652262395Sbapt		 * we were able to create the directory. We will tell the
653262395Sbapt		 * caller that we found something to fix, and it is ok to try
654262395Sbapt		 * and create the node again.
655262395Sbapt		 */
656262395Sbapt		retval = 0;
657262395Sbapt		if (pids)
658262395Sbapt			(void)set_ids(name, st_uid, st_gid);
659262395Sbapt
660262395Sbapt		/*
661262395Sbapt		 * make sure the user doesn't have some strange umask that
662262395Sbapt		 * causes this newly created directory to be unusable. We fix
663262395Sbapt		 * the modes and restore them back to the creation default at
664262395Sbapt		 * the end of pax
665262395Sbapt		 */
666262395Sbapt		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
667262395Sbapt		    (lstat(name, &sb) == 0)) {
668262395Sbapt			set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
669262395Sbapt			add_dir(name, spt - name, &sb, 1);
670262395Sbapt		}
671262395Sbapt		*(spt++) = '/';
672262395Sbapt		continue;
673262395Sbapt	}
674262395Sbapt	return(retval);
675262395Sbapt}
676262395Sbapt
677262395Sbapt/*
678262395Sbapt * set_ftime()
679262395Sbapt *	Set the access time and modification time for a named file. If frc
680262395Sbapt *	is non-zero we force these times to be set even if the user did not
681262395Sbapt *	request access and/or modification time preservation (this is also
682262395Sbapt *	used by -t to reset access times).
683262395Sbapt *	When ign is zero, only those times the user has asked for are set, the
684262395Sbapt *	other ones are left alone. We do not assume the un-documented feature
685262395Sbapt *	of many utimes() implementations that consider a 0 time value as a do
686262395Sbapt *	not set request.
687262395Sbapt */
688262395Sbapt
689262395Sbaptvoid
690262395Sbaptset_ftime(char *fnm, time_t mtime, time_t atime, int frc)
691262395Sbapt{
692262395Sbapt	struct timeval tv[2];
693262395Sbapt	struct stat sb;
694262395Sbapt
695262395Sbapt	tv[0].tv_sec = (long)atime;
696262395Sbapt	tv[0].tv_usec = 0;
697262395Sbapt	tv[1].tv_sec = (long)mtime;
698262395Sbapt	tv[1].tv_usec = 0;
699262395Sbapt	if (!frc && (!patime || !pmtime)) {
700262395Sbapt		/*
701262395Sbapt		 * if we are not forcing, only set those times the user wants
702262395Sbapt		 * set. We get the current values of the times if we need them.
703262395Sbapt		 */
704262395Sbapt		if (lstat(fnm, &sb) == 0) {
705262395Sbapt#ifdef BSD4_4
706262395Sbapt			if (!patime)
707262395Sbapt				TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
708262395Sbapt			if (!pmtime)
709262395Sbapt				TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
710262395Sbapt#else
711262395Sbapt			if (!patime)
712262395Sbapt				tv[0].tv_sec = sb.st_atime;
713262395Sbapt			if (!pmtime)
714262395Sbapt				tv[1].tv_sec = sb.st_mtime;
715262395Sbapt#endif
716262395Sbapt		} else
717262395Sbapt			syswarn(0, errno, "Cannot obtain file stats %s", fnm);
718262395Sbapt	}
719262395Sbapt
720262395Sbapt	/*
721262395Sbapt	 * set the times
722262395Sbapt	 */
723262395Sbapt#if HAVE_LUTIMES
724262395Sbapt	if (lutimes(fnm, tv))
725262395Sbapt#else
726262395Sbapt	if (utimes(fnm, tv))
727262395Sbapt#endif
728262395Sbapt		syswarn(1, errno, "Access/modification time set failed on: %s",
729262395Sbapt		    fnm);
730262395Sbapt	return;
731262395Sbapt}
732262395Sbapt
733262395Sbapt/*
734262395Sbapt * set_ids()
735262395Sbapt *	set the uid and gid of a file system node
736262395Sbapt * Return:
737262395Sbapt *	0 when set, -1 on failure
738262395Sbapt */
739262395Sbapt
740262395Sbaptint
741262395Sbaptset_ids(char *fnm, uid_t uid, gid_t gid)
742262395Sbapt{
743262395Sbapt	if (geteuid() == 0)
744262395Sbapt		if (lchown(fnm, uid, gid)) {
745262395Sbapt			(void)fflush(listf);
746262395Sbapt			syswarn(1, errno, "Cannot set file uid/gid of %s",
747262395Sbapt			    fnm);
748262395Sbapt			return(-1);
749262395Sbapt		}
750262395Sbapt	return(0);
751262395Sbapt}
752262395Sbapt
753262395Sbapt/*
754262395Sbapt * set_pmode()
755262395Sbapt *	Set file access mode
756262395Sbapt */
757262395Sbapt
758262395Sbaptvoid
759262395Sbaptset_pmode(char *fnm, mode_t mode)
760262395Sbapt{
761262395Sbapt	mode &= ABITS;
762262395Sbapt	if (lchmod(fnm, mode)) {
763262395Sbapt		(void)fflush(listf);
764262395Sbapt		syswarn(1, errno, "Cannot set permissions on %s", fnm);
765262395Sbapt	}
766262395Sbapt	return;
767262395Sbapt}
768262395Sbapt
769262395Sbapt/*
770262395Sbapt * set_chflags()
771262395Sbapt *	Set 4.4BSD file flags
772262395Sbapt */
773262395Sbaptvoid
774262395Sbaptset_chflags(char *fnm, u_int32_t flags)
775262395Sbapt{
776262395Sbapt
777262395Sbapt#if 0
778262395Sbapt	if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
779262395Sbapt		syswarn(1, errno, "Cannot set file flags on %s", fnm);
780262395Sbapt#endif
781262395Sbapt	return;
782262395Sbapt}
783262395Sbapt
784262395Sbapt/*
785262395Sbapt * file_write()
786262395Sbapt *	Write/copy a file (during copy or archive extract). This routine knows
787262395Sbapt *	how to copy files with lseek holes in it. (Which are read as file
788262395Sbapt *	blocks containing all 0's but do not have any file blocks associated
789262395Sbapt *	with the data). Typical examples of these are files created by dbm
790262395Sbapt *	variants (.pag files). While the file size of these files are huge, the
791262395Sbapt *	actual storage is quite small (the files are sparse). The problem is
792262395Sbapt *	the holes read as all zeros so are probably stored on the archive that
793262395Sbapt *	way (there is no way to determine if the file block is really a hole,
794262395Sbapt *	we only know that a file block of all zero's can be a hole).
795262395Sbapt *	At this writing, no major archive format knows how to archive files
796262395Sbapt *	with holes. However, on extraction (or during copy, -rw) we have to
797262395Sbapt *	deal with these files. Without detecting the holes, the files can
798262395Sbapt *	consume a lot of file space if just written to disk. This replacement
799262395Sbapt *	for write when passed the basic allocation size of a file system block,
800262395Sbapt *	uses lseek whenever it detects the input data is all 0 within that
801262395Sbapt *	file block. In more detail, the strategy is as follows:
802262395Sbapt *	While the input is all zero keep doing an lseek. Keep track of when we
803262395Sbapt *	pass over file block boundaries. Only write when we hit a non zero
804262395Sbapt *	input. once we have written a file block, we continue to write it to
805262395Sbapt *	the end (we stop looking at the input). When we reach the start of the
806262395Sbapt *	next file block, start checking for zero blocks again. Working on file
807262395Sbapt *	block boundaries significantly reduces the overhead when copying files
808262395Sbapt *	that are NOT very sparse. This overhead (when compared to a write) is
809262395Sbapt *	almost below the measurement resolution on many systems. Without it,
810262395Sbapt *	files with holes cannot be safely copied. It does has a side effect as
811262395Sbapt *	it can put holes into files that did not have them before, but that is
812262395Sbapt *	not a problem since the file contents are unchanged (in fact it saves
813262395Sbapt *	file space). (Except on paging files for diskless clients. But since we
814262395Sbapt *	cannot determine one of those file from here, we ignore them). If this
815262395Sbapt *	ever ends up on a system where CTG files are supported and the holes
816262395Sbapt *	are not desired, just do a conditional test in those routines that
817262395Sbapt *	call file_write() and have it call write() instead. BEFORE CLOSING THE
818262395Sbapt *	FILE, make sure to call file_flush() when the last write finishes with
819262395Sbapt *	an empty block. A lot of file systems will not create an lseek hole at
820262395Sbapt *	the end. In this case we drop a single 0 at the end to force the
821262395Sbapt *	trailing 0's in the file.
822262395Sbapt *	---Parameters---
823262395Sbapt *	rem: how many bytes left in this file system block
824262395Sbapt *	isempt: have we written to the file block yet (is it empty)
825262395Sbapt *	sz: basic file block allocation size
826262395Sbapt *	cnt: number of bytes on this write
827262395Sbapt *	str: buffer to write
828262395Sbapt * Return:
829262395Sbapt *	number of bytes written, -1 on write (or lseek) error.
830 */
831
832int
833file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
834	char *name)
835{
836	char *pt;
837	char *end;
838	int wcnt;
839	char *st = str;
840	char **strp;
841
842	/*
843	 * while we have data to process
844	 */
845	while (cnt) {
846		if (!*rem) {
847			/*
848			 * We are now at the start of file system block again
849			 * (or what we think one is...). start looking for
850			 * empty blocks again
851			 */
852			*isempt = 1;
853			*rem = sz;
854		}
855
856		/*
857		 * only examine up to the end of the current file block or
858		 * remaining characters to write, whatever is smaller
859		 */
860		wcnt = MIN(cnt, *rem);
861		cnt -= wcnt;
862		*rem -= wcnt;
863		if (*isempt) {
864			/*
865			 * have not written to this block yet, so we keep
866			 * looking for zero's
867			 */
868			pt = st;
869			end = st + wcnt;
870
871			/*
872			 * look for a zero filled buffer
873			 */
874			while ((pt < end) && (*pt == '\0'))
875				++pt;
876
877			if (pt == end) {
878				/*
879				 * skip, buf is empty so far
880				 */
881				if (fd > -1 &&
882				    lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
883					syswarn(1, errno, "File seek on %s",
884					    name);
885					return(-1);
886				}
887				st = pt;
888				continue;
889			}
890			/*
891			 * drat, the buf is not zero filled
892			 */
893			*isempt = 0;
894		}
895
896		/*
897		 * have non-zero data in this file system block, have to write
898		 */
899		switch (fd) {
900		case -1:
901			strp = &gnu_name_string;
902			break;
903		case -2:
904			strp = &gnu_link_string;
905			break;
906		default:
907			strp = NULL;
908			break;
909		}
910		if (strp) {
911			if (*strp)
912				err(1, "WARNING! Major Internal Error! GNU hack Failing!");
913			*strp = malloc(wcnt + 1);
914			if (*strp == NULL) {
915				tty_warn(1, "Out of memory");
916				return(-1);
917			}
918			strlcpy(*strp, st, wcnt);
919			break;
920		} else if (xwrite(fd, st, wcnt) != wcnt) {
921			syswarn(1, errno, "Failed write to file %s", name);
922			return(-1);
923		}
924		st += wcnt;
925	}
926	return(st - str);
927}
928
929/*
930 * file_flush()
931 *	when the last file block in a file is zero, many file systems will not
932 *	let us create a hole at the end. To get the last block with zeros, we
933 *	write the last BYTE with a zero (back up one byte and write a zero).
934 */
935
936void
937file_flush(int fd, char *fname, int isempt)
938{
939	static char blnk[] = "\0";
940
941	/*
942	 * silly test, but make sure we are only called when the last block is
943	 * filled with all zeros.
944	 */
945	if (!isempt)
946		return;
947
948	/*
949	 * move back one byte and write a zero
950	 */
951	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
952		syswarn(1, errno, "Failed seek on file %s", fname);
953		return;
954	}
955
956	if (write_with_restart(fd, blnk, 1) < 0)
957		syswarn(1, errno, "Failed write to file %s", fname);
958	return;
959}
960
961/*
962 * rdfile_close()
963 *	close a file we have been reading (to copy or archive). If we have to
964 *	reset access time (tflag) do so (the times are stored in arcn).
965 */
966
967void
968rdfile_close(ARCHD *arcn, int *fd)
969{
970	/*
971	 * make sure the file is open
972	 */
973	if (*fd < 0)
974		return;
975
976	(void)close(*fd);
977	*fd = -1;
978	if (!tflag)
979		return;
980
981	/*
982	 * user wants last access time reset
983	 */
984	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
985	return;
986}
987
988/*
989 * set_crc()
990 *	read a file to calculate its crc. This is a real drag. Archive formats
991 *	that have this, end up reading the file twice (we have to write the
992 *	header WITH the crc before writing the file contents. Oh well...
993 * Return:
994 *	0 if was able to calculate the crc, -1 otherwise
995 */
996
997int
998set_crc(ARCHD *arcn, int fd)
999{
1000	int i;
1001	int res;
1002	off_t cpcnt = 0L;
1003	u_long size;
1004	unsigned long crc = 0L;
1005	char tbuf[FILEBLK];
1006	struct stat sb;
1007
1008	if (fd < 0) {
1009		/*
1010		 * hmm, no fd, should never happen. well no crc then.
1011		 */
1012		arcn->crc = 0L;
1013		return(0);
1014	}
1015
1016	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1017		size = (u_long)sizeof(tbuf);
1018
1019	/*
1020	 * read all the bytes we think that there are in the file. If the user
1021	 * is trying to archive an active file, forget this file.
1022	 */
1023	for(;;) {
1024		if ((res = read(fd, tbuf, size)) <= 0)
1025			break;
1026		cpcnt += res;
1027		for (i = 0; i < res; ++i)
1028			crc += (tbuf[i] & 0xff);
1029	}
1030
1031	/*
1032	 * safety check. we want to avoid archiving files that are active as
1033	 * they can create inconsistant archive copies.
1034	 */
1035	if (cpcnt != arcn->sb.st_size)
1036		tty_warn(1, "File changed size %s", arcn->org_name);
1037	else if (fstat(fd, &sb) < 0)
1038		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1039	else if (arcn->sb.st_mtime != sb.st_mtime)
1040		tty_warn(1, "File %s was modified during read", arcn->org_name);
1041	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1042		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1043	else {
1044		arcn->crc = crc;
1045		return(0);
1046	}
1047	return(-1);
1048}
1049