file_subs.c revision 76351
1/*-
2 * Copyright (c) 1992 Keith Muller.
3 * Copyright (c) 1992, 1993
4 *	The Regents of the University of California.  All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * Keith Muller of the University of California, San Diego.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 *    notice, this list of conditions and the following disclaimer in the
16 *    documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 *    must display the following acknowledgement:
19 *	This product includes software developed by the University of
20 *	California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 *    may be used to endorse or promote products derived from this software
23 *    without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 */
37
38#ifndef lint
39#if 0
40static char sccsid[] = "@(#)file_subs.c	8.1 (Berkeley) 5/31/93";
41#endif
42static const char rcsid[] =
43  "$FreeBSD: head/bin/pax/file_subs.c 76351 2001-05-08 06:19:06Z kris $";
44#endif /* not lint */
45
46#include <sys/types.h>
47#include <sys/time.h>
48#include <sys/stat.h>
49#include <unistd.h>
50#include <fcntl.h>
51#include <string.h>
52#include <stdio.h>
53#include <errno.h>
54#include <sys/uio.h>
55#include <stdlib.h>
56#include "pax.h"
57#include "options.h"
58#include "extern.h"
59
60static int
61mk_link __P((register char *,register struct stat *,register char *, int));
62
63/*
64 * routines that deal with file operations such as: creating, removing;
65 * and setting access modes, uid/gid and times of files
66 */
67
68#define FILEBITS		(S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
69#define SETBITS			(S_ISUID | S_ISGID)
70#define ABITS			(FILEBITS | SETBITS)
71
72/*
73 * file_creat()
74 *	Create and open a file.
75 * Return:
76 *	file descriptor or -1 for failure
77 */
78
79#ifdef __STDC__
80int
81file_creat(register ARCHD *arcn)
82#else
83int
84file_creat(arcn)
85	register ARCHD *arcn;
86#endif
87{
88	int fd = -1;
89	mode_t file_mode;
90	int oerrno;
91
92	/*
93	 * assume file doesn't exist, so just try to create it, most times this
94	 * works. We have to take special handling when the file does exist. To
95	 * detect this, we use O_EXCL. For example when trying to create a
96	 * file and a character device or fifo exists with the same name, we
97	 * can accidently open the device by mistake (or block waiting to open)
98	 * If we find that the open has failed, then figure spend the effort to
99	 * figure out why. This strategy was found to have better average
100	 * performance in common use than checking the file (and the path)
101	 * first with lstat.
102	 */
103	file_mode = arcn->sb.st_mode & FILEBITS;
104	if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
105	    file_mode)) >= 0)
106		return(fd);
107
108	/*
109	 * the file seems to exist. First we try to get rid of it (found to be
110	 * the second most common failure when traced). If this fails, only
111	 * then we go to the expense to check and create the path to the file
112	 */
113	if (unlnk_exist(arcn->name, arcn->type) != 0)
114		return(-1);
115
116	for (;;) {
117		/*
118		 * try to open it again, if this fails, check all the nodes in
119		 * the path and give it a final try. if chk_path() finds that
120		 * it cannot fix anything, we will skip the last attempt
121		 */
122		if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
123		    file_mode)) >= 0)
124			break;
125		oerrno = errno;
126		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
127			syswarn(1, oerrno, "Unable to create %s", arcn->name);
128			return(-1);
129		}
130	}
131	return(fd);
132}
133
134/*
135 * file_close()
136 *	Close file descriptor to a file just created by pax. Sets modes,
137 *	ownership and times as required.
138 * Return:
139 *	0 for success, -1 for failure
140 */
141
142#ifdef __STDC__
143void
144file_close(register ARCHD *arcn, int fd)
145#else
146void
147file_close(arcn, fd)
148	register ARCHD *arcn;
149	int fd;
150#endif
151{
152	int res = 0;
153
154	if (fd < 0)
155		return;
156	if (close(fd) < 0)
157		syswarn(0, errno, "Unable to close file descriptor on %s",
158		    arcn->name);
159
160	/*
161	 * set owner/groups first as this may strip off mode bits we want
162	 * then set file permission modes. Then set file access and
163	 * modification times.
164	 */
165	if (pids)
166		res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
167
168	/*
169	 * IMPORTANT SECURITY NOTE:
170	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
171	 * set uid/gid bits
172	 */
173	if (!pmode || res)
174		arcn->sb.st_mode &= ~(SETBITS);
175	if (pmode)
176		set_pmode(arcn->name, arcn->sb.st_mode);
177	if (patime || pmtime)
178		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
179}
180
181/*
182 * lnk_creat()
183 *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
184 *	must exist;
185 * Return:
186 *	0 if ok, -1 otherwise
187 */
188
189#ifdef __STDC__
190int
191lnk_creat(register ARCHD *arcn)
192#else
193int
194lnk_creat(arcn)
195	register ARCHD *arcn;
196#endif
197{
198	struct stat sb;
199
200	/*
201	 * we may be running as root, so we have to be sure that link target
202	 * is not a directory, so we lstat and check
203	 */
204	if (lstat(arcn->ln_name, &sb) < 0) {
205		syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
206		    arcn->name);
207		return(-1);
208	}
209
210	if (S_ISDIR(sb.st_mode)) {
211		paxwarn(1, "A hard link to the directory %s is not allowed",
212		    arcn->ln_name);
213		return(-1);
214	}
215
216	return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
217}
218
219/*
220 * cross_lnk()
221 *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
222 *	with the -l flag. No warning or error if this does not succeed (we will
223 *	then just create the file)
224 * Return:
225 *	1 if copy() should try to create this file node
226 *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
227 */
228
229#ifdef __STDC__
230int
231cross_lnk(register ARCHD *arcn)
232#else
233int
234cross_lnk(arcn)
235	register ARCHD *arcn;
236#endif
237{
238	/*
239	 * try to make a link to original file (-l flag in copy mode). make sure
240	 * we do not try to link to directories in case we are running as root
241	 * (and it might succeed).
242	 */
243	if (arcn->type == PAX_DIR)
244		return(1);
245	return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
246}
247
248/*
249 * chk_same()
250 *	In copy mode if we are not trying to make hard links between the src
251 *	and destinations, make sure we are not going to overwrite ourselves by
252 *	accident. This slows things down a little, but we have to protect all
253 *	those people who make typing errors.
254 * Return:
255 *	1 the target does not exist, go ahead and copy
256 *	0 skip it file exists (-k) or may be the same as source file
257 */
258
259#ifdef __STDC__
260int
261chk_same(register ARCHD *arcn)
262#else
263int
264chk_same(arcn)
265	register ARCHD *arcn;
266#endif
267{
268	struct stat sb;
269
270	/*
271	 * if file does not exist, return. if file exists and -k, skip it
272	 * quietly
273	 */
274	if (lstat(arcn->name, &sb) < 0)
275		return(1);
276	if (kflag)
277		return(0);
278
279	/*
280	 * better make sure the user does not have src == dest by mistake
281	 */
282	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
283		paxwarn(1, "Unable to copy %s, file would overwrite itself",
284		    arcn->name);
285		return(0);
286	}
287	return(1);
288}
289
290/*
291 * mk_link()
292 *	try to make a hard link between two files. if ign set, we do not
293 *	complain.
294 * Return:
295 *	0 if successful (or we are done with this file but no error, such as
296 *	finding the from file exists and the user has set -k).
297 *	1 when ign was set to indicates we could not make the link but we
298 *	should try to copy/extract the file as that might work (and is an
299 *	allowed option). -1 an error occurred.
300 */
301
302#ifdef __STDC__
303static int
304mk_link(register char *to, register struct stat *to_sb, register char *from,
305	int ign)
306#else
307static int
308mk_link(to, to_sb, from, ign)
309	register char *to;
310	register struct stat *to_sb;
311	register char *from;
312	int ign;
313#endif
314{
315	struct stat sb;
316	int oerrno;
317
318	/*
319	 * if from file exists, it has to be unlinked to make the link. If the
320	 * file exists and -k is set, skip it quietly
321	 */
322	if (lstat(from, &sb) == 0) {
323		if (kflag)
324			return(0);
325
326		/*
327		 * make sure it is not the same file, protect the user
328		 */
329		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
330			paxwarn(1, "Unable to link file %s to itself", to);
331			return(-1);;
332		}
333
334		/*
335		 * try to get rid of the file, based on the type
336		 */
337		if (S_ISDIR(sb.st_mode)) {
338			if (rmdir(from) < 0) {
339				syswarn(1, errno, "Unable to remove %s", from);
340				return(-1);
341			}
342		} else if (unlink(from) < 0) {
343			if (!ign) {
344				syswarn(1, errno, "Unable to remove %s", from);
345				return(-1);
346			}
347			return(1);
348		}
349	}
350
351	/*
352	 * from file is gone (or did not exist), try to make the hard link.
353	 * if it fails, check the path and try it again (if chk_path() says to
354	 * try again)
355	 */
356	for (;;) {
357		if (link(to, from) == 0)
358			break;
359		oerrno = errno;
360		if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
361			continue;
362		if (!ign) {
363			syswarn(1, oerrno, "Could not link to %s from %s", to,
364			    from);
365			return(-1);
366		}
367		return(1);
368	}
369
370	/*
371	 * all right the link was made
372	 */
373	return(0);
374}
375
376/*
377 * node_creat()
378 *	create an entry in the file system (other than a file or hard link).
379 *	If successful, sets uid/gid modes and times as required.
380 * Return:
381 *	0 if ok, -1 otherwise
382 */
383
384#ifdef __STDC__
385int
386node_creat(register ARCHD *arcn)
387#else
388int
389node_creat(arcn)
390	register ARCHD *arcn;
391#endif
392{
393	register int res;
394	register int ign = 0;
395	register int oerrno;
396	register int pass = 0;
397	mode_t file_mode;
398	struct stat sb;
399
400	/*
401	 * create node based on type, if that fails try to unlink the node and
402	 * try again. finally check the path and try again. As noted in the
403	 * file and link creation routines, this method seems to exhibit the
404	 * best performance in general use workloads.
405	 */
406	file_mode = arcn->sb.st_mode & FILEBITS;
407
408	for (;;) {
409		switch(arcn->type) {
410		case PAX_DIR:
411			res = mkdir(arcn->name, file_mode);
412			if (ign)
413				res = 0;
414			break;
415		case PAX_CHR:
416			file_mode |= S_IFCHR;
417			res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
418			break;
419		case PAX_BLK:
420			file_mode |= S_IFBLK;
421			res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
422			break;
423		case PAX_FIF:
424			res = mkfifo(arcn->name, file_mode);
425			break;
426		case PAX_SCK:
427			/*
428			 * Skip sockets, operation has no meaning under BSD
429			 */
430			paxwarn(0,
431			    "%s skipped. Sockets cannot be copied or extracted",
432			    arcn->name);
433			return(-1);
434		case PAX_SLK:
435			res = symlink(arcn->ln_name, arcn->name);
436			break;
437		case PAX_CTG:
438		case PAX_HLK:
439		case PAX_HRG:
440		case PAX_REG:
441		default:
442			/*
443			 * we should never get here
444			 */
445			paxwarn(0, "%s has an unknown file type, skipping",
446				arcn->name);
447			return(-1);
448		}
449
450		/*
451		 * if we were able to create the node break out of the loop,
452		 * otherwise try to unlink the node and try again. if that
453		 * fails check the full path and try a final time.
454		 */
455		if (res == 0)
456			break;
457
458		/*
459		 * we failed to make the node
460		 */
461		oerrno = errno;
462		if ((ign = unlnk_exist(arcn->name, arcn->type)) < 0)
463			return(-1);
464
465		if (++pass <= 1)
466			continue;
467
468		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
469			syswarn(1, oerrno, "Could not create: %s", arcn->name);
470			return(-1);
471		}
472	}
473
474	/*
475	 * we were able to create the node. set uid/gid, modes and times
476	 */
477	if (pids)
478		res = ((arcn->type == PAX_SLK) ?
479		    set_lids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid) :
480		    set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid));
481	else
482		res = 0;
483
484	/*
485	 * symlinks are done now.
486	 */
487	if (arcn->type == PAX_SLK)
488		return(0);
489
490	/*
491	 * IMPORTANT SECURITY NOTE:
492	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
493	 * set uid/gid bits
494	 */
495	if (!pmode || res)
496		arcn->sb.st_mode &= ~(SETBITS);
497	if (pmode)
498		set_pmode(arcn->name, arcn->sb.st_mode);
499
500	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
501		/*
502		 * Dirs must be processed again at end of extract to set times
503		 * and modes to agree with those stored in the archive. However
504		 * to allow extract to continue, we may have to also set owner
505		 * rights. This allows nodes in the archive that are children
506		 * of this directory to be extracted without failure. Both time
507		 * and modes will be fixed after the entire archive is read and
508		 * before pax exits.
509		 */
510		if (access(arcn->name, R_OK | W_OK | X_OK) < 0) {
511			if (lstat(arcn->name, &sb) < 0) {
512				syswarn(0, errno,"Could not access %s (stat)",
513				    arcn->name);
514				set_pmode(arcn->name,file_mode | S_IRWXU);
515			} else {
516				/*
517				 * We have to add rights to the dir, so we make
518				 * sure to restore the mode. The mode must be
519				 * restored AS CREATED and not as stored if
520				 * pmode is not set.
521				 */
522				set_pmode(arcn->name,
523				    ((sb.st_mode & FILEBITS) | S_IRWXU));
524				if (!pmode)
525					arcn->sb.st_mode = sb.st_mode;
526			}
527
528			/*
529			 * we have to force the mode to what was set here,
530			 * since we changed it from the default as created.
531			 */
532			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1);
533		} else if (pmode || patime || pmtime)
534			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0);
535	}
536
537	if (patime || pmtime)
538		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
539	return(0);
540}
541
542/*
543 * unlnk_exist()
544 *	Remove node from file system with the specified name. We pass the type
545 *	of the node that is going to replace it. When we try to create a
546 *	directory and find that it already exists, we allow processing to
547 *	continue as proper modes etc will always be set for it later on.
548 * Return:
549 *	0 is ok to proceed, no file with the specified name exists
550 *	-1 we were unable to remove the node, or we should not remove it (-k)
551 *	1 we found a directory and we were going to create a directory.
552 */
553
554#ifdef __STDC__
555int
556unlnk_exist(register char *name, register int type)
557#else
558int
559unlnk_exist(name, type)
560	register char *name;
561	register int type;
562#endif
563{
564	struct stat sb;
565
566	/*
567	 * the file does not exist, or -k we are done
568	 */
569	if (lstat(name, &sb) < 0)
570		return(0);
571	if (kflag)
572		return(-1);
573
574	if (S_ISDIR(sb.st_mode)) {
575		/*
576		 * try to remove a directory, if it fails and we were going to
577		 * create a directory anyway, tell the caller (return a 1)
578		 */
579		if (rmdir(name) < 0) {
580			if (type == PAX_DIR)
581				return(1);
582			syswarn(1,errno,"Unable to remove directory %s", name);
583			return(-1);
584		}
585		return(0);
586	}
587
588	/*
589	 * try to get rid of all non-directory type nodes
590	 */
591	if (unlink(name) < 0) {
592		syswarn(1, errno, "Could not unlink %s", name);
593		return(-1);
594	}
595	return(0);
596}
597
598/*
599 * chk_path()
600 *	We were trying to create some kind of node in the file system and it
601 *	failed. chk_path() makes sure the path up to the node exists and is
602 *	writeable. When we have to create a directory that is missing along the
603 *	path somewhere, the directory we create will be set to the same
604 *	uid/gid as the file has (when uid and gid are being preserved).
605 *	NOTE: this routine is a real performance loss. It is only used as a
606 *	last resort when trying to create entries in the file system.
607 * Return:
608 *	-1 when it could find nothing it is allowed to fix.
609 *	0 otherwise
610 */
611
612#ifdef __STDC__
613int
614chk_path( register char *name, uid_t st_uid, gid_t st_gid)
615#else
616int
617chk_path(name, st_uid, st_gid)
618	register char *name;
619	uid_t st_uid;
620	gid_t st_gid;
621#endif
622{
623	register char *spt = name;
624	struct stat sb;
625	int retval = -1;
626
627	/*
628	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
629	 */
630	if (*spt == '/')
631		++spt;
632
633	for(;;) {
634		/*
635		 * work foward from the first / and check each part of the path
636		 */
637		spt = strchr(spt, '/');
638		if (spt == NULL)
639			break;
640		*spt = '\0';
641
642		/*
643		 * if it exists we assume it is a directory, it is not within
644		 * the spec (at least it seems to read that way) to alter the
645		 * file system for nodes NOT EXPLICITLY stored on the archive.
646		 * If that assumption is changed, you would test the node here
647		 * and figure out how to get rid of it (probably like some
648		 * recursive unlink()) or fix up the directory permissions if
649		 * required (do an access()).
650		 */
651		if (lstat(name, &sb) == 0) {
652			*(spt++) = '/';
653			continue;
654		}
655
656		/*
657		 * the path fails at this point, see if we can create the
658		 * needed directory and continue on
659		 */
660		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
661			*spt = '/';
662			retval = -1;
663			break;
664		}
665
666		/*
667		 * we were able to create the directory. We will tell the
668		 * caller that we found something to fix, and it is ok to try
669		 * and create the node again.
670		 */
671		retval = 0;
672		if (pids)
673			(void)set_ids(name, st_uid, st_gid);
674
675		/*
676		 * make sure the user doen't have some strange umask that
677		 * causes this newly created directory to be unusable. We fix
678		 * the modes and restore them back to the creation default at
679		 * the end of pax
680		 */
681		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
682		    (lstat(name, &sb) == 0)) {
683			set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
684			add_dir(name, spt - name, &sb, 1);
685		}
686		*(spt++) = '/';
687		continue;
688	}
689	return(retval);
690}
691
692/*
693 * set_ftime()
694 *	Set the access time and modification time for a named file. If frc is
695 *	non-zero we force these times to be set even if the user did not
696 *	request access and/or modification time preservation (this is also
697 *	used by -t to reset access times).
698 *	When ign is zero, only those times the user has asked for are set, the
699 *	other ones are left alone. We do not assume the un-documented feature
700 *	of many utimes() implementations that consider a 0 time value as a do
701 *	not set request.
702 */
703
704#ifdef __STDC__
705void
706set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
707#else
708void
709set_ftime(fnm, mtime, atime, frc)
710	char *fnm;
711	time_t mtime;
712	time_t atime;
713	int frc;
714#endif
715{
716	static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
717	struct stat sb;
718
719	tv[0].tv_sec = (long)atime;
720	tv[1].tv_sec = (long)mtime;
721	if (!frc && (!patime || !pmtime)) {
722		/*
723		 * if we are not forcing, only set those times the user wants
724		 * set. We get the current values of the times if we need them.
725		 */
726		if (lstat(fnm, &sb) == 0) {
727			if (!patime)
728				tv[0].tv_sec = (long)sb.st_atime;
729			if (!pmtime)
730				tv[1].tv_sec = (long)sb.st_mtime;
731		} else
732			syswarn(0,errno,"Unable to obtain file stats %s", fnm);
733	}
734
735	/*
736	 * set the times
737	 */
738	if (utimes(fnm, tv) < 0)
739		syswarn(1, errno, "Access/modification time set failed on: %s",
740		    fnm);
741	return;
742}
743
744/*
745 * set_ids()
746 *	set the uid and gid of a file system node
747 * Return:
748 *	0 when set, -1 on failure
749 */
750
751#ifdef __STDC__
752int
753set_ids(char *fnm, uid_t uid, gid_t gid)
754#else
755int
756set_ids(fnm, uid, gid)
757	char *fnm;
758	uid_t uid;
759	gid_t gid;
760#endif
761{
762	if (chown(fnm, uid, gid) < 0) {
763		/*
764		 * ignore EPERM unless in verbose mode or being run by root.
765		 * if running as pax, POSIX requires a warning.
766		 */
767		if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
768		    geteuid() == 0)
769			syswarn(1, errno, "Unable to set file uid/gid of %s",
770			    fnm);
771		return(-1);
772	}
773	return(0);
774}
775
776/*
777 * set_lids()
778 *	set the uid and gid of a file system node
779 * Return:
780 *	0 when set, -1 on failure
781 */
782
783#ifdef __STDC__
784int
785set_lids(char *fnm, uid_t uid, gid_t gid)
786#else
787int
788set_lids(fnm, uid, gid)
789	char *fnm;
790	uid_t uid;
791	gid_t gid;
792#endif
793{
794	if (lchown(fnm, uid, gid) < 0) {
795		/*
796		 * ignore EPERM unless in verbose mode or being run by root.
797		 * if running as pax, POSIX requires a warning.
798		 */
799		if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
800		    geteuid() == 0)
801			syswarn(1, errno, "Unable to set file uid/gid of %s",
802			    fnm);
803		return(-1);
804	}
805	return(0);
806}
807
808/*
809 * set_pmode()
810 *	Set file access mode
811 */
812
813#ifdef __STDC__
814void
815set_pmode(char *fnm, mode_t mode)
816#else
817void
818set_pmode(fnm, mode)
819	char *fnm;
820	mode_t mode;
821#endif
822{
823	mode &= ABITS;
824	if (chmod(fnm, mode) < 0)
825		syswarn(1, errno, "Could not set permissions on %s", fnm);
826	return;
827}
828
829/*
830 * file_write()
831 *	Write/copy a file (during copy or archive extract). This routine knows
832 *	how to copy files with lseek holes in it. (Which are read as file
833 *	blocks containing all 0's but do not have any file blocks associated
834 *	with the data). Typical examples of these are files created by dbm
835 *	variants (.pag files). While the file size of these files are huge, the
836 *	actual storage is quite small (the files are sparse). The problem is
837 *	the holes read as all zeros so are probably stored on the archive that
838 *	way (there is no way to determine if the file block is really a hole,
839 *	we only know that a file block of all zero's can be a hole).
840 *	At this writing, no major archive format knows how to archive files
841 *	with holes. However, on extraction (or during copy, -rw) we have to
842 *	deal with these files. Without detecting the holes, the files can
843 *	consume a lot of file space if just written to disk. This replacement
844 *	for write when passed the basic allocation size of a file system block,
845 *	uses lseek whenever it detects the input data is all 0 within that
846 *	file block. In more detail, the strategy is as follows:
847 *	While the input is all zero keep doing an lseek. Keep track of when we
848 *	pass over file block boundries. Only write when we hit a non zero
849 *	input. once we have written a file block, we continue to write it to
850 *	the end (we stop looking at the input). When we reach the start of the
851 *	next file block, start checking for zero blocks again. Working on file
852 *	block boundries significantly reduces the overhead when copying files
853 *	that are NOT very sparse. This overhead (when compared to a write) is
854 *	almost below the measurement resolution on many systems. Without it,
855 *	files with holes cannot be safely copied. It does has a side effect as
856 *	it can put holes into files that did not have them before, but that is
857 *	not a problem since the file contents are unchanged (in fact it saves
858 *	file space). (Except on paging files for diskless clients. But since we
859 *	cannot determine one of those file from here, we ignore them). If this
860 *	ever ends up on a system where CTG files are supported and the holes
861 *	are not desired, just do a conditional test in those routines that
862 *	call file_write() and have it call write() instead. BEFORE CLOSING THE
863 *	FILE, make sure to call file_flush() when the last write finishes with
864 *	an empty block. A lot of file systems will not create an lseek hole at
865 *	the end. In this case we drop a single 0 at the end to force the
866 *	trailing 0's in the file.
867 *	---Parameters---
868 *	rem: how many bytes left in this file system block
869 *	isempt: have we written to the file block yet (is it empty)
870 *	sz: basic file block allocation size
871 *	cnt: number of bytes on this write
872 *	str: buffer to write
873 * Return:
874 *	number of bytes written, -1 on write (or lseek) error.
875 */
876
877#ifdef __STDC__
878int
879file_write(int fd, char *str, register int cnt, int *rem, int *isempt, int sz,
880	char *name)
881#else
882int
883file_write(fd, str, cnt, rem, isempt, sz, name)
884	int fd;
885	char *str;
886	register int cnt;
887	int *rem;
888	int *isempt;
889	int sz;
890	char *name;
891#endif
892{
893	register char *pt;
894	register char *end;
895	register int wcnt;
896	register char *st = str;
897
898	/*
899	 * while we have data to process
900	 */
901	while (cnt) {
902		if (!*rem) {
903			/*
904			 * We are now at the start of file system block again
905			 * (or what we think one is...). start looking for
906			 * empty blocks again
907			 */
908			*isempt = 1;
909			*rem = sz;
910		}
911
912		/*
913		 * only examine up to the end of the current file block or
914		 * remaining characters to write, whatever is smaller
915		 */
916		wcnt = MIN(cnt, *rem);
917		cnt -= wcnt;
918		*rem -= wcnt;
919		if (*isempt) {
920			/*
921			 * have not written to this block yet, so we keep
922			 * looking for zero's
923			 */
924			pt = st;
925			end = st + wcnt;
926
927			/*
928			 * look for a zero filled buffer
929			 */
930			while ((pt < end) && (*pt == '\0'))
931				++pt;
932
933			if (pt == end) {
934				/*
935				 * skip, buf is empty so far
936				 */
937				if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
938					syswarn(1,errno,"File seek on %s",
939					    name);
940					return(-1);
941				}
942				st = pt;
943				continue;
944			}
945			/*
946			 * drat, the buf is not zero filled
947			 */
948			*isempt = 0;
949		}
950
951		/*
952		 * have non-zero data in this file system block, have to write
953		 */
954		if (write(fd, st, wcnt) != wcnt) {
955			syswarn(1, errno, "Failed write to file %s", name);
956			return(-1);
957		}
958		st += wcnt;
959	}
960	return(st - str);
961}
962
963/*
964 * file_flush()
965 *	when the last file block in a file is zero, many file systems will not
966 *	let us create a hole at the end. To get the last block with zeros, we
967 *	write the last BYTE with a zero (back up one byte and write a zero).
968 */
969
970#ifdef __STDC__
971void
972file_flush(int fd, char *fname, int isempt)
973#else
974void
975file_flush(fd, fname, isempt)
976	int fd;
977	char *fname;
978	int isempt;
979#endif
980{
981	static char blnk[] = "\0";
982
983	/*
984	 * silly test, but make sure we are only called when the last block is
985	 * filled with all zeros.
986	 */
987	if (!isempt)
988		return;
989
990	/*
991	 * move back one byte and write a zero
992	 */
993	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
994		syswarn(1, errno, "Failed seek on file %s", fname);
995		return;
996	}
997
998	if (write(fd, blnk, 1) < 0)
999		syswarn(1, errno, "Failed write to file %s", fname);
1000	return;
1001}
1002
1003/*
1004 * rdfile_close()
1005 *	close a file we have beed reading (to copy or archive). If we have to
1006 *	reset access time (tflag) do so (the times are stored in arcn).
1007 */
1008
1009#ifdef __STDC__
1010void
1011rdfile_close(register ARCHD *arcn, register int *fd)
1012#else
1013void
1014rdfile_close(arcn, fd)
1015	register ARCHD *arcn;
1016	register int *fd;
1017#endif
1018{
1019	/*
1020	 * make sure the file is open
1021	 */
1022	if (*fd < 0)
1023		return;
1024
1025	(void)close(*fd);
1026	*fd = -1;
1027	if (!tflag)
1028		return;
1029
1030	/*
1031	 * user wants last access time reset
1032	 */
1033	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
1034	return;
1035}
1036
1037/*
1038 * set_crc()
1039 *	read a file to calculate its crc. This is a real drag. Archive formats
1040 *	that have this, end up reading the file twice (we have to write the
1041 *	header WITH the crc before writing the file contents. Oh well...
1042 * Return:
1043 *	0 if was able to calculate the crc, -1 otherwise
1044 */
1045
1046#ifdef __STDC__
1047int
1048set_crc(register ARCHD *arcn, register int fd)
1049#else
1050int
1051set_crc(arcn, fd)
1052	register ARCHD *arcn;
1053	register int fd;
1054#endif
1055{
1056	register int i;
1057	register int res;
1058	off_t cpcnt = 0L;
1059	u_long size;
1060	unsigned long crc = 0L;
1061	char tbuf[FILEBLK];
1062	struct stat sb;
1063
1064	if (fd < 0) {
1065		/*
1066		 * hmm, no fd, should never happen. well no crc then.
1067		 */
1068		arcn->crc = 0L;
1069		return(0);
1070	}
1071
1072	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1073		size = (u_long)sizeof(tbuf);
1074
1075	/*
1076	 * read all the bytes we think that there are in the file. If the user
1077	 * is trying to archive an active file, forget this file.
1078	 */
1079	for(;;) {
1080		if ((res = read(fd, tbuf, size)) <= 0)
1081			break;
1082		cpcnt += res;
1083		for (i = 0; i < res; ++i)
1084			crc += (tbuf[i] & 0xff);
1085	}
1086
1087	/*
1088	 * safety check. we want to avoid archiving files that are active as
1089	 * they can create inconsistant archive copies.
1090	 */
1091	if (cpcnt != arcn->sb.st_size)
1092		paxwarn(1, "File changed size %s", arcn->org_name);
1093	else if (fstat(fd, &sb) < 0)
1094		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1095	else if (arcn->sb.st_mtime != sb.st_mtime)
1096		paxwarn(1, "File %s was modified during read", arcn->org_name);
1097	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1098		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1099	else {
1100		arcn->crc = crc;
1101		return(0);
1102	}
1103	return(-1);
1104}
1105