file_subs.c revision 1.11
1/*	$NetBSD: file_subs.c,v 1.11 1999/10/22 10:38:41 mrg 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. All advertising materials mentioning features or use of this software
20 *    must display the following acknowledgement:
21 *	This product includes software developed by the University of
22 *	California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 *    may be used to endorse or promote products derived from this software
25 *    without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * SUCH DAMAGE.
38 */
39
40#include <sys/cdefs.h>
41#ifndef lint
42#if 0
43static char sccsid[] = "@(#)file_subs.c	8.1 (Berkeley) 5/31/93";
44#else
45__RCSID("$NetBSD: file_subs.c,v 1.11 1999/10/22 10:38:41 mrg 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
64static int
65mk_link __P((char *,struct stat *,char *, int));
66
67/*
68 * routines that deal with file operations such as: creating, removing;
69 * and setting access modes, uid/gid and times of files
70 */
71
72#define FILEBITS		(S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
73#define SETBITS			(S_ISUID | S_ISGID)
74#define ABITS			(FILEBITS | SETBITS)
75
76/*
77 * file_creat()
78 *	Create and open a file.
79 * Return:
80 *	file descriptor or -1 for failure
81 */
82
83#if __STDC__
84int
85file_creat(ARCHD *arcn)
86#else
87int
88file_creat(arcn)
89	ARCHD *arcn;
90#endif
91{
92	int fd = -1;
93	mode_t file_mode;
94	int oerrno;
95
96	/*
97	 * assume file doesn't exist, so just try to create it, most times this
98	 * works. We have to take special handling when the file does exist. To
99	 * detect this, we use O_EXCL. For example when trying to create a
100	 * file and a character device or fifo exists with the same name, we
101	 * can accidently open the device by mistake (or block waiting to open)
102	 * If we find that the open has failed, then figure spend the effore to
103	 * figure out why. This strategy was found to have better average
104	 * performance in common use than checking the file (and the path)
105	 * first with lstat.
106	 */
107	file_mode = arcn->sb.st_mode & FILEBITS;
108	if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
109	    file_mode)) >= 0)
110		return(fd);
111
112	/*
113	 * the file seems to exist. First we try to get rid of it (found to be
114	 * the second most common failure when traced). If this fails, only
115	 * then we go to the expense to check and create the path to the file
116	 */
117	if (unlnk_exist(arcn->name, arcn->type) != 0)
118		return(-1);
119
120	for (;;) {
121		/*
122		 * try to open it again, if this fails, check all the nodes in
123		 * the path and give it a final try. if chk_path() finds that
124		 * it cannot fix anything, we will skip the last attempt
125		 */
126		if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
127		    file_mode)) >= 0)
128			break;
129		oerrno = errno;
130		if (chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
131			syswarn(1, oerrno, "Unable to create %s", arcn->name);
132			return(-1);
133		}
134	}
135	return(fd);
136}
137
138/*
139 * file_close()
140 *	Close file descriptor to a file just created by pax. Sets modes,
141 *	ownership and times as required.
142 * Return:
143 *	0 for success, -1 for failure
144 */
145
146#if __STDC__
147void
148file_close(ARCHD *arcn, int fd)
149#else
150void
151file_close(arcn, fd)
152	ARCHD *arcn;
153	int fd;
154#endif
155{
156	int res = 0;
157
158	if (fd < 0)
159		return;
160	if (close(fd) < 0)
161		syswarn(0, errno, "Unable to close file descriptor on %s",
162		    arcn->name);
163
164	/*
165	 * set owner/groups first as this may strip off mode bits we want
166	 * then set file permission modes. Then set file access and
167	 * modification times.
168	 */
169	if (pids)
170		res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
171
172	/*
173	 * IMPORTANT SECURITY NOTE:
174	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
175	 * set uid/gid bits
176	 */
177	if (!pmode || res)
178		arcn->sb.st_mode &= ~(SETBITS);
179	if (pmode)
180		set_pmode(arcn->name, arcn->sb.st_mode);
181	if (patime || pmtime)
182		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
183}
184
185/*
186 * lnk_creat()
187 *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
188 *	must exist;
189 * Return:
190 *	0 if ok, -1 otherwise
191 */
192
193#if __STDC__
194int
195lnk_creat(ARCHD *arcn)
196#else
197int
198lnk_creat(arcn)
199	ARCHD *arcn;
200#endif
201{
202	struct stat sb;
203
204	/*
205	 * we may be running as root, so we have to be sure that link target
206	 * is not a directory, so we lstat and check
207	 */
208	if (lstat(arcn->ln_name, &sb) < 0) {
209		syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
210		    arcn->name);
211		return(-1);
212	}
213
214	if (S_ISDIR(sb.st_mode)) {
215		tty_warn(1, "A hard link to the directory %s is not allowed",
216		    arcn->ln_name);
217		return(-1);
218	}
219
220	return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
221}
222
223/*
224 * cross_lnk()
225 *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
226 *	with the -l flag. No warning or error if this does not succeed (we will
227 *	then just create the file)
228 * Return:
229 *	1 if copy() should try to create this file node
230 *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
231 */
232
233#if __STDC__
234int
235cross_lnk(ARCHD *arcn)
236#else
237int
238cross_lnk(arcn)
239	ARCHD *arcn;
240#endif
241{
242	/*
243	 * try to make a link to orginal file (-l flag in copy mode). make sure
244	 * we do not try to link to directories in case we are running as root
245	 * (and it might succeed).
246	 */
247	if (arcn->type == PAX_DIR)
248		return(1);
249	return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
250}
251
252/*
253 * chk_same()
254 *	In copy mode if we are not trying to make hard links between the src
255 *	and destinations, make sure we are not going to overwrite ourselves by
256 *	accident. This slows things down a little, but we have to protect all
257 *	those people who make typing errors.
258 * Return:
259 *	1 the target does not exist, go ahead and copy
260 *	0 skip it file exists (-k) or may be the same as source file
261 */
262
263#if __STDC__
264int
265chk_same(ARCHD *arcn)
266#else
267int
268chk_same(arcn)
269	ARCHD *arcn;
270#endif
271{
272	struct stat sb;
273
274	/*
275	 * if file does not exist, return. if file exists and -k, skip it
276	 * quietly
277	 */
278	if (lstat(arcn->name, &sb) < 0)
279		return(1);
280	if (kflag)
281		return(0);
282
283	/*
284	 * better make sure the user does not have src == dest by mistake
285	 */
286	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
287		tty_warn(1, "Unable to copy %s, file would overwrite itself",
288		    arcn->name);
289		return(0);
290	}
291	return(1);
292}
293
294/*
295 * mk_link()
296 *	try to make a hard link between two files. if ign set, we do not
297 *	complain.
298 * Return:
299 *	0 if successful (or we are done with this file but no error, such as
300 *	finding the from file exists and the user has set -k).
301 *	1 when ign was set to indicates we could not make the link but we
302 *	should try to copy/extract the file as that might work (and is an
303 *	allowed option). -1 an error occurred.
304 */
305
306#if __STDC__
307static int
308mk_link(char *to, struct stat *to_sb, char *from,
309	int ign)
310#else
311static int
312mk_link(to, to_sb, from, ign)
313	char *to;
314	struct stat *to_sb;
315	char *from;
316	int ign;
317#endif
318{
319	struct stat sb;
320	int oerrno;
321
322	/*
323	 * if from file exists, it has to be unlinked to make the link. If the
324	 * file exists and -k is set, skip it quietly
325	 */
326	if (lstat(from, &sb) == 0) {
327		if (kflag)
328			return(0);
329
330		/*
331		 * make sure it is not the same file, protect the user
332		 */
333		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
334			tty_warn(1, "Unable to link file %s to itself", to);
335			return(-1);;
336		}
337
338		/*
339		 * try to get rid of the file, based on the type
340		 */
341		if (S_ISDIR(sb.st_mode)) {
342			if (rmdir(from) < 0) {
343				syswarn(1, errno, "Unable to remove %s", from);
344				return(-1);
345			}
346		} else if (unlink(from) < 0) {
347			if (!ign) {
348				syswarn(1, errno, "Unable to remove %s", from);
349				return(-1);
350			}
351			return(1);
352		}
353	}
354
355	/*
356	 * from file is gone (or did not exist), try to make the hard link.
357	 * if it fails, check the path and try it again (if chk_path() says to
358	 * try again)
359	 */
360	for (;;) {
361		if (link(to, from) == 0)
362			break;
363		oerrno = errno;
364		if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
365			continue;
366		if (!ign) {
367			syswarn(1, oerrno, "Could not link to %s from %s", to,
368			    from);
369			return(-1);
370		}
371		return(1);
372	}
373
374	/*
375	 * all right the link was made
376	 */
377	return(0);
378}
379
380/*
381 * node_creat()
382 *	create an entry in the file system (other than a file or hard link).
383 *	If successful, sets uid/gid modes and times as required.
384 * Return:
385 *	0 if ok, -1 otherwise
386 */
387
388#if __STDC__
389int
390node_creat(ARCHD *arcn)
391#else
392int
393node_creat(arcn)
394	ARCHD *arcn;
395#endif
396{
397	int res;
398	int ign = 0;
399	int oerrno;
400	int pass = 0;
401	mode_t file_mode;
402	struct stat sb;
403
404	/*
405	 * create node based on type, if that fails try to unlink the node and
406	 * try again. finally check the path and try again. As noted in the
407	 * file and link creation routines, this method seems to exhibit the
408	 * best performance in general use workloads.
409	 */
410	file_mode = arcn->sb.st_mode & FILEBITS;
411
412	for (;;) {
413		switch(arcn->type) {
414		case PAX_DIR:
415			res = mkdir(arcn->name, file_mode);
416			if (ign)
417				res = 0;
418			break;
419		case PAX_CHR:
420			file_mode |= S_IFCHR;
421			res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
422			break;
423		case PAX_BLK:
424			file_mode |= S_IFBLK;
425			res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
426			break;
427		case PAX_FIF:
428			res = mkfifo(arcn->name, file_mode);
429			break;
430		case PAX_SCK:
431			/*
432			 * Skip sockets, operation has no meaning under BSD
433			 */
434			tty_warn(0,
435			    "%s skipped. Sockets cannot be copied or extracted",
436			    arcn->name);
437			return(-1);
438		case PAX_SLK:
439			res = symlink(arcn->ln_name, arcn->name);
440			break;
441		case PAX_CTG:
442		case PAX_HLK:
443		case PAX_HRG:
444		case PAX_REG:
445		default:
446			/*
447			 * we should never get here
448			 */
449			tty_warn(0, "%s has an unknown file type, skipping",
450				arcn->name);
451			return(-1);
452		}
453
454		/*
455		 * if we were able to create the node break out of the loop,
456		 * otherwise try to unlink the node and try again. if that
457		 * fails check the full path and try a final time.
458		 */
459		if (res == 0)
460			break;
461
462		/*
463		 * we failed to make the node
464		 */
465		oerrno = errno;
466		if ((ign = unlnk_exist(arcn->name, arcn->type)) < 0)
467			return(-1);
468
469		if (++pass <= 1)
470			continue;
471
472		if (chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
473			syswarn(1, oerrno, "Could not create: %s", arcn->name);
474			return(-1);
475		}
476	}
477
478	/*
479	 * we were able to create the node. set uid/gid, modes and times
480	 */
481	if (pids)
482		res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
483	else
484		res = 0;
485
486	/*
487	 * IMPORTANT SECURITY NOTE:
488	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
489	 * set uid/gid bits
490	 */
491	if (!pmode || res)
492		arcn->sb.st_mode &= ~(SETBITS);
493	if (pmode)
494		set_pmode(arcn->name, arcn->sb.st_mode);
495
496	if (arcn->type == PAX_DIR) {
497		/*
498		 * Dirs must be processed again at end of extract to set times
499		 * and modes to agree with those stored in the archive. However
500		 * to allow extract to continue, we may have to also set owner
501		 * rights. This allows nodes in the archive that are children
502		 * of this directory to be extracted without failure. Both time
503		 * and modes will be fixed after the entire archive is read and
504		 * before pax exits.
505		 */
506		if (access(arcn->name, R_OK | W_OK | X_OK) < 0) {
507			if (lstat(arcn->name, &sb) < 0) {
508				syswarn(0, errno,"Could not access %s (stat)",
509				    arcn->name);
510				set_pmode(arcn->name,file_mode | S_IRWXU);
511			} else {
512				/*
513				 * We have to add rights to the dir, so we make
514				 * sure to restore the mode. The mode must be
515				 * restored AS CREATED and not as stored if
516				 * pmode is not set.
517				 */
518				set_pmode(arcn->name,
519				    ((sb.st_mode & FILEBITS) | S_IRWXU));
520				if (!pmode)
521					arcn->sb.st_mode = sb.st_mode;
522			}
523
524			/*
525			 * we have to force the mode to what was set here,
526			 * since we changed it from the default as created.
527			 */
528			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1);
529		} else if (pmode || patime || pmtime)
530			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0);
531	}
532
533	if (patime || pmtime)
534		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
535	return(0);
536}
537
538/*
539 * unlnk_exist()
540 *	Remove node from file system with the specified name. We pass the type
541 *	of the node that is going to replace it. When we try to create a
542 *	directory and find that it already exists, we allow processing to
543 *	continue as proper modes etc will always be set for it later on.
544 * Return:
545 *	0 is ok to proceed, no file with the specified name exists
546 *	-1 we were unable to remove the node, or we should not remove it (-k)
547 *	1 we found a directory and we were going to create a directory.
548 */
549
550#if __STDC__
551int
552unlnk_exist(char *name, int type)
553#else
554int
555unlnk_exist(name, type)
556	char *name;
557	int type;
558#endif
559{
560	struct stat sb;
561
562	/*
563	 * the file does not exist, or -k we are done
564	 */
565	if (lstat(name, &sb) < 0)
566		return(0);
567	if (kflag)
568		return(-1);
569
570	if (S_ISDIR(sb.st_mode)) {
571		/*
572		 * try to remove a directory, if it fails and we were going to
573		 * create a directory anyway, tell the caller (return a 1)
574		 */
575		if (rmdir(name) < 0) {
576			if (type == PAX_DIR)
577				return(1);
578			syswarn(1,errno,"Unable to remove directory %s", name);
579			return(-1);
580		}
581		return(0);
582	}
583
584	/*
585	 * try to get rid of all non-directory type nodes
586	 */
587	if (unlink(name) < 0) {
588		syswarn(1, errno, "Could not unlink %s", name);
589		return(-1);
590	}
591	return(0);
592}
593
594/*
595 * chk_path()
596 *	We were trying to create some kind of node in the file system and it
597 *	failed. chk_path() makes sure the path up to the node exists and is
598 *	writeable. When we have to create a directory that is missing along the
599 *	path somewhere, the directory we create will be set to the same
600 *	uid/gid as the file has (when uid and gid are being preserved).
601 *	NOTE: this routine is a real performance loss. It is only used as a
602 *	last resort when trying to create entries in the file system.
603 * Return:
604 *	-1 when it could find nothing it is allowed to fix.
605 *	0 otherwise
606 */
607
608#if __STDC__
609int
610chk_path( char *name, uid_t st_uid, gid_t st_gid)
611#else
612int
613chk_path(name, st_uid, st_gid)
614	char *name;
615	uid_t st_uid;
616	gid_t st_gid;
617#endif
618{
619	char *spt = name;
620	struct stat sb;
621	int retval = -1;
622
623	/*
624	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
625	 */
626	if (*spt == '/')
627		++spt;
628
629	for(;;) {
630		/*
631		 * work foward from the first / and check each part of the path
632		 */
633		spt = strchr(spt, '/');
634		if (spt == NULL)
635			break;
636		*spt = '\0';
637
638		/*
639		 * if it exists we assume it is a directory, it is not within
640		 * the spec (at least it seems to read that way) to alter the
641		 * file system for nodes NOT EXPLICITLY stored on the archive.
642		 * If that assumption is changed, you would test the node here
643		 * and figure out how to get rid of it (probably like some
644		 * recursive unlink()) or fix up the directory permissions if
645		 * required (do an access()).
646		 */
647		if (lstat(name, &sb) == 0) {
648			*(spt++) = '/';
649			continue;
650		}
651
652		/*
653		 * the path fails at this point, see if we can create the
654		 * needed directory and continue on
655		 */
656		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
657			*spt = '/';
658			retval = -1;
659			break;
660		}
661
662		/*
663		 * we were able to create the directory. We will tell the
664		 * caller that we found something to fix, and it is ok to try
665		 * and create the node again.
666		 */
667		retval = 0;
668		if (pids)
669			(void)set_ids(name, st_uid, st_gid);
670
671		/*
672		 * make sure the user doen't have some strange umask that
673		 * causes this newly created directory to be unusable. We fix
674		 * the modes and restore them back to the creation default at
675		 * the end of pax
676		 */
677		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
678		    (lstat(name, &sb) == 0)) {
679			set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
680			add_dir(name, spt - name, &sb, 1);
681		}
682		*(spt++) = '/';
683		continue;
684	}
685	return(retval);
686}
687
688/*
689 * set_ftime()
690 *	Set the access time and modification time for a named file. If frc is
691 *	non-zero we force these times to be set even if the the user did not
692 *	request access and/or modification time preservation (this is also
693 *	used by -t to reset access times).
694 *	When ign is zero, only those times the user has asked for are set, the
695 *	other ones are left alone. We do not assume the un-documented feature
696 *	of many utimes() implementations that consider a 0 time value as a do
697 *	not set request.
698 */
699
700#if __STDC__
701void
702set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
703#else
704void
705set_ftime(fnm, mtime, atime, frc)
706	char *fnm;
707	time_t mtime;
708	time_t atime;
709	int frc;
710#endif
711{
712	struct timeval tv[2];
713	struct stat sb;
714
715	tv[0].tv_sec = (long)atime;
716	tv[0].tv_usec = 0;
717	tv[1].tv_sec = (long)mtime;
718	tv[1].tv_usec = 0;
719	if (!frc && (!patime || !pmtime)) {
720		/*
721		 * if we are not forcing, only set those times the user wants
722		 * set. We get the current values of the times if we need them.
723		 */
724		if (lstat(fnm, &sb) == 0) {
725			if (!patime)
726				TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
727			if (!pmtime)
728				TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
729		} else
730			syswarn(0,errno,"Unable to obtain file stats %s", fnm);
731	}
732
733	/*
734	 * set the times
735	 */
736	if (lutimes(fnm, tv) < 0)
737		syswarn(1, errno, "Access/modification time set failed on: %s",
738		    fnm);
739	return;
740}
741
742/*
743 * set_ids()
744 *	set the uid and gid of a file system node
745 * Return:
746 *	0 when set, -1 on failure
747 */
748
749#if __STDC__
750int
751set_ids(char *fnm, uid_t uid, gid_t gid)
752#else
753int
754set_ids(fnm, uid, gid)
755	char *fnm;
756	uid_t uid;
757	gid_t gid;
758#endif
759{
760	if (lchown(fnm, uid, gid) < 0) {
761		syswarn(1, errno, "Unable to set file uid/gid of %s", fnm);
762		return(-1);
763	}
764	return(0);
765}
766
767/*
768 * set_pmode()
769 *	Set file access mode
770 */
771
772#if __STDC__
773void
774set_pmode(char *fnm, mode_t mode)
775#else
776void
777set_pmode(fnm, mode)
778	char *fnm;
779	mode_t mode;
780#endif
781{
782	mode &= ABITS;
783	if (lchmod(fnm, mode) < 0)
784		syswarn(1, errno, "Could not set permissions on %s", fnm);
785	return;
786}
787
788/*
789 * file_write()
790 *	Write/copy a file (during copy or archive extract). This routine knows
791 *	how to copy files with lseek holes in it. (Which are read as file
792 *	blocks containing all 0's but do not have any file blocks associated
793 *	with the data). Typical examples of these are files created by dbm
794 *	variants (.pag files). While the file size of these files are huge, the
795 *	actual storage is quite small (the files are sparse). The problem is
796 *	the holes read as all zeros so are probably stored on the archive that
797 *	way (there is no way to determine if the file block is really a hole,
798 *	we only know that a file block of all zero's can be a hole).
799 *	At this writing, no major archive format knows how to archive files
800 *	with holes. However, on extraction (or during copy, -rw) we have to
801 *	deal with these files. Without detecting the holes, the files can
802 *	consume a lot of file space if just written to disk. This replacement
803 *	for write when passed the basic allocation size of a file system block,
804 *	uses lseek whenever it detects the input data is all 0 within that
805 *	file block. In more detail, the strategy is as follows:
806 *	While the input is all zero keep doing an lseek. Keep track of when we
807 *	pass over file block boundries. Only write when we hit a non zero
808 *	input. once we have written a file block, we continue to write it to
809 *	the end (we stop looking at the input). When we reach the start of the
810 *	next file block, start checking for zero blocks again. Working on file
811 *	block boundries significantly reduces the overhead when copying files
812 *	that are NOT very sparse. This overhead (when compared to a write) is
813 *	almost below the measurement resolution on many systems. Without it,
814 *	files with holes cannot be safely copied. It does has a side effect as
815 *	it can put holes into files that did not have them before, but that is
816 *	not a problem since the file contents are unchanged (in fact it saves
817 *	file space). (Except on paging files for diskless clients. But since we
818 *	cannot determine one of those file from here, we ignore them). If this
819 *	ever ends up on a system where CTG files are supported and the holes
820 *	are not desired, just do a conditional test in those routines that
821 *	call file_write() and have it call write() instead. BEFORE CLOSING THE
822 *	FILE, make sure to call file_flush() when the last write finishes with
823 *	an empty block. A lot of file systems will not create an lseek hole at
824 *	the end. In this case we drop a single 0 at the end to force the
825 *	trailing 0's in the file.
826 *	---Parameters---
827 *	rem: how many bytes left in this file system block
828 *	isempt: have we written to the file block yet (is it empty)
829 *	sz: basic file block allocation size
830 *	cnt: number of bytes on this write
831 *	str: buffer to write
832 * Return:
833 *	number of bytes written, -1 on write (or lseek) error.
834 */
835
836#if __STDC__
837int
838file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
839	char *name)
840#else
841int
842file_write(fd, str, cnt, rem, isempt, sz, name)
843	int fd;
844	char *str;
845	int cnt;
846	int *rem;
847	int *isempt;
848	int sz;
849	char *name;
850#endif
851{
852	char *pt;
853	char *end;
854	int wcnt;
855	char *st = str;
856
857	/*
858	 * while we have data to process
859	 */
860	while (cnt) {
861		if (!*rem) {
862			/*
863			 * We are now at the start of file system block again
864			 * (or what we think one is...). start looking for
865			 * empty blocks again
866			 */
867			*isempt = 1;
868			*rem = sz;
869		}
870
871		/*
872		 * only examine up to the end of the current file block or
873		 * remaining characters to write, whatever is smaller
874		 */
875		wcnt = MIN(cnt, *rem);
876		cnt -= wcnt;
877		*rem -= wcnt;
878		if (*isempt) {
879			/*
880			 * have not written to this block yet, so we keep
881			 * looking for zero's
882			 */
883			pt = st;
884			end = st + wcnt;
885
886			/*
887			 * look for a zero filled buffer
888			 */
889			while ((pt < end) && (*pt == '\0'))
890				++pt;
891
892			if (pt == end) {
893				/*
894				 * skip, buf is empty so far
895				 */
896				if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
897					syswarn(1,errno,"File seek on %s",
898					    name);
899					return(-1);
900				}
901				st = pt;
902				continue;
903			}
904			/*
905			 * drat, the buf is not zero filled
906			 */
907			*isempt = 0;
908		}
909
910		/*
911		 * have non-zero data in this file system block, have to write
912		 */
913		if (write(fd, st, wcnt) != wcnt) {
914			syswarn(1, errno, "Failed write to file %s", name);
915			return(-1);
916		}
917		st += wcnt;
918	}
919	return(st - str);
920}
921
922/*
923 * file_flush()
924 *	when the last file block in a file is zero, many file systems will not
925 *	let us create a hole at the end. To get the last block with zeros, we
926 *	write the last BYTE with a zero (back up one byte and write a zero).
927 */
928
929#if __STDC__
930void
931file_flush(int fd, char *fname, int isempt)
932#else
933void
934file_flush(fd, fname, isempt)
935	int fd;
936	char *fname;
937	int isempt;
938#endif
939{
940	static char blnk[] = "\0";
941
942	/*
943	 * silly test, but make sure we are only called when the last block is
944	 * filled with all zeros.
945	 */
946	if (!isempt)
947		return;
948
949	/*
950	 * move back one byte and write a zero
951	 */
952	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
953		syswarn(1, errno, "Failed seek on file %s", fname);
954		return;
955	}
956
957	if (write(fd, blnk, 1) < 0)
958		syswarn(1, errno, "Failed write to file %s", fname);
959	return;
960}
961
962/*
963 * rdfile_close()
964 *	close a file we have been reading (to copy or archive). If we have to
965 *	reset access time (tflag) do so (the times are stored in arcn).
966 */
967
968#if __STDC__
969void
970rdfile_close(ARCHD *arcn, int *fd)
971#else
972void
973rdfile_close(arcn, fd)
974	ARCHD *arcn;
975	int *fd;
976#endif
977{
978	/*
979	 * make sure the file is open
980	 */
981	if (*fd < 0)
982		return;
983
984	(void)close(*fd);
985	*fd = -1;
986	if (!tflag)
987		return;
988
989	/*
990	 * user wants last access time reset
991	 */
992	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
993	return;
994}
995
996/*
997 * set_crc()
998 *	read a file to calculate its crc. This is a real drag. Archive formats
999 *	that have this, end up reading the file twice (we have to write the
1000 *	header WITH the crc before writing the file contents. Oh well...
1001 * Return:
1002 *	0 if was able to calculate the crc, -1 otherwise
1003 */
1004
1005#if __STDC__
1006int
1007set_crc(ARCHD *arcn, int fd)
1008#else
1009int
1010set_crc(arcn, fd)
1011	ARCHD *arcn;
1012	int fd;
1013#endif
1014{
1015	int i;
1016	int res;
1017	off_t cpcnt = 0L;
1018	u_long size;
1019	unsigned long crc = 0L;
1020	char tbuf[FILEBLK];
1021	struct stat sb;
1022
1023	if (fd < 0) {
1024		/*
1025		 * hmm, no fd, should never happen. well no crc then.
1026		 */
1027		arcn->crc = 0L;
1028		return(0);
1029	}
1030
1031	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1032		size = (u_long)sizeof(tbuf);
1033
1034	/*
1035	 * read all the bytes we think that there are in the file. If the user
1036	 * is trying to archive an active file, forget this file.
1037	 */
1038	for(;;) {
1039		if ((res = read(fd, tbuf, size)) <= 0)
1040			break;
1041		cpcnt += res;
1042		for (i = 0; i < res; ++i)
1043			crc += (tbuf[i] & 0xff);
1044	}
1045
1046	/*
1047	 * safety check. we want to avoid archiving files that are active as
1048	 * they can create inconsistant archive copies.
1049	 */
1050	if (cpcnt != arcn->sb.st_size)
1051		tty_warn(1, "File changed size %s", arcn->org_name);
1052	else if (fstat(fd, &sb) < 0)
1053		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1054	else if (arcn->sb.st_mtime != sb.st_mtime)
1055		tty_warn(1, "File %s was modified during read", arcn->org_name);
1056	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1057		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1058	else {
1059		arcn->crc = crc;
1060		return(0);
1061	}
1062	return(-1);
1063}
1064