file_subs.c revision 1.6
1/*	$NetBSD: file_subs.c,v 1.6 1997/07/20 20:32:28 christos Exp $	*/
2
3/*-
4 * Copyright (c) 1992 Keith Muller.
5 * Copyright (c) 1992, 1993
6 *	The Regents of the University of California.  All rights reserved.
7 *
8 * This code is derived from software contributed to Berkeley by
9 * Keith Muller of the University of California, San Diego.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 *    notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 *    notice, this list of conditions and the following disclaimer in the
18 *    documentation and/or other materials provided with the distribution.
19 * 3. 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 1.6 1997/07/20 20:32:28 christos Exp $");
46#endif
47#endif /* not lint */
48
49#include <sys/types.h>
50#include <sys/time.h>
51#include <sys/stat.h>
52#include <unistd.h>
53#include <sys/param.h>
54#include <fcntl.h>
55#include <string.h>
56#include <stdio.h>
57#include <ctype.h>
58#include <errno.h>
59#include <sys/uio.h>
60#include <stdlib.h>
61#include "pax.h"
62#include "extern.h"
63
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			if ((res = symlink(arcn->ln_name, arcn->name)) == 0)
440				return(0);
441			break;
442		case PAX_CTG:
443		case PAX_HLK:
444		case PAX_HRG:
445		case PAX_REG:
446		default:
447			/*
448			 * we should never get here
449			 */
450			tty_warn(0, "%s has an unknown file type, skipping",
451				arcn->name);
452			return(-1);
453		}
454
455		/*
456		 * if we were able to create the node break out of the loop,
457		 * otherwise try to unlink the node and try again. if that
458		 * fails check the full path and try a final time.
459		 */
460		if (res == 0)
461			break;
462
463		/*
464		 * we failed to make the node
465		 */
466		oerrno = errno;
467		if ((ign = unlnk_exist(arcn->name, arcn->type)) < 0)
468			return(-1);
469
470		if (++pass <= 1)
471			continue;
472
473		if (chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
474			syswarn(1, oerrno, "Could not create: %s", arcn->name);
475			return(-1);
476		}
477	}
478
479	/*
480	 * we were able to create the node. set uid/gid, modes and times
481	 */
482	if (pids)
483		res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
484	else
485		res = 0;
486
487	/*
488	 * IMPORTANT SECURITY NOTE:
489	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
490	 * set uid/gid bits
491	 */
492	if (!pmode || res)
493		arcn->sb.st_mode &= ~(SETBITS);
494	if (pmode)
495		set_pmode(arcn->name, arcn->sb.st_mode);
496
497	if (arcn->type == PAX_DIR) {
498		/*
499		 * Dirs must be processed again at end of extract to set times
500		 * and modes to agree with those stored in the archive. However
501		 * to allow extract to continue, we may have to also set owner
502		 * rights. This allows nodes in the archive that are children
503		 * of this directory to be extracted without failure. Both time
504		 * and modes will be fixed after the entire archive is read and
505		 * before pax exits.
506		 */
507		if (access(arcn->name, R_OK | W_OK | X_OK) < 0) {
508			if (lstat(arcn->name, &sb) < 0) {
509				syswarn(0, errno,"Could not access %s (stat)",
510				    arcn->name);
511				set_pmode(arcn->name,file_mode | S_IRWXU);
512			} else {
513				/*
514				 * We have to add rights to the dir, so we make
515				 * sure to restore the mode. The mode must be
516				 * restored AS CREATED and not as stored if
517				 * pmode is not set.
518				 */
519				set_pmode(arcn->name,
520				    ((sb.st_mode & FILEBITS) | S_IRWXU));
521				if (!pmode)
522					arcn->sb.st_mode = sb.st_mode;
523			}
524
525			/*
526			 * we have to force the mode to what was set here,
527			 * since we changed it from the default as created.
528			 */
529			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1);
530		} else if (pmode || patime || pmtime)
531			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0);
532	}
533
534	if (patime || pmtime)
535		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
536	return(0);
537}
538
539/*
540 * unlnk_exist()
541 *	Remove node from file system with the specified name. We pass the type
542 *	of the node that is going to replace it. When we try to create a
543 *	directory and find that it already exists, we allow processing to
544 *	continue as proper modes etc will always be set for it later on.
545 * Return:
546 *	0 is ok to proceed, no file with the specified name exists
547 *	-1 we were unable to remove the node, or we should not remove it (-k)
548 *	1 we found a directory and we were going to create a directory.
549 */
550
551#if __STDC__
552int
553unlnk_exist(char *name, int type)
554#else
555int
556unlnk_exist(name, type)
557	char *name;
558	int type;
559#endif
560{
561	struct stat sb;
562
563	/*
564	 * the file does not exist, or -k we are done
565	 */
566	if (lstat(name, &sb) < 0)
567		return(0);
568	if (kflag)
569		return(-1);
570
571	if (S_ISDIR(sb.st_mode)) {
572		/*
573		 * try to remove a directory, if it fails and we were going to
574		 * create a directory anyway, tell the caller (return a 1)
575		 */
576		if (rmdir(name) < 0) {
577			if (type == PAX_DIR)
578				return(1);
579			syswarn(1,errno,"Unable to remove directory %s", name);
580			return(-1);
581		}
582		return(0);
583	}
584
585	/*
586	 * try to get rid of all non-directory type nodes
587	 */
588	if (unlink(name) < 0) {
589		syswarn(1, errno, "Could not unlink %s", name);
590		return(-1);
591	}
592	return(0);
593}
594
595/*
596 * chk_path()
597 *	We were trying to create some kind of node in the file system and it
598 *	failed. chk_path() makes sure the path up to the node exists and is
599 *	writeable. When we have to create a directory that is missing along the
600 *	path somewhere, the directory we create will be set to the same
601 *	uid/gid as the file has (when uid and gid are being preserved).
602 *	NOTE: this routine is a real performance loss. It is only used as a
603 *	last resort when trying to create entries in the file system.
604 * Return:
605 *	-1 when it could find nothing it is allowed to fix.
606 *	0 otherwise
607 */
608
609#if __STDC__
610int
611chk_path( char *name, uid_t st_uid, gid_t st_gid)
612#else
613int
614chk_path(name, st_uid, st_gid)
615	char *name;
616	uid_t st_uid;
617	gid_t st_gid;
618#endif
619{
620	char *spt = name;
621	struct stat sb;
622	int retval = -1;
623
624	/*
625	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
626	 */
627	if (*spt == '/')
628		++spt;
629
630	for(;;) {
631		/*
632		 * work foward from the first / and check each part of the path
633		 */
634		spt = strchr(spt, '/');
635		if (spt == NULL)
636			break;
637		*spt = '\0';
638
639		/*
640		 * if it exists we assume it is a directory, it is not within
641		 * the spec (at least it seems to read that way) to alter the
642		 * file system for nodes NOT EXPLICITLY stored on the archive.
643		 * If that assumption is changed, you would test the node here
644		 * and figure out how to get rid of it (probably like some
645		 * recursive unlink()) or fix up the directory permissions if
646		 * required (do an access()).
647		 */
648		if (lstat(name, &sb) == 0) {
649			*(spt++) = '/';
650			continue;
651		}
652
653		/*
654		 * the path fails at this point, see if we can create the
655		 * needed directory and continue on
656		 */
657		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
658			*spt = '/';
659			retval = -1;
660			break;
661		}
662
663		/*
664		 * we were able to create the directory. We will tell the
665		 * caller that we found something to fix, and it is ok to try
666		 * and create the node again.
667		 */
668		retval = 0;
669		if (pids)
670			(void)set_ids(name, st_uid, st_gid);
671
672		/*
673		 * make sure the user doen't have some strange umask that
674		 * causes this newly created directory to be unusable. We fix
675		 * the modes and restore them back to the creation default at
676		 * the end of pax
677		 */
678		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
679		    (lstat(name, &sb) == 0)) {
680			set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
681			add_dir(name, spt - name, &sb, 1);
682		}
683		*(spt++) = '/';
684		continue;
685	}
686	return(retval);
687}
688
689/*
690 * set_ftime()
691 *	Set the access time and modification time for a named file. If frc is
692 *	non-zero we force these times to be set even if the the user did not
693 *	request access and/or modification time preservation (this is also
694 *	used by -t to reset access times).
695 *	When ign is zero, only those times the user has asked for are set, the
696 *	other ones are left alone. We do not assume the un-documented feature
697 *	of many utimes() implementations that consider a 0 time value as a do
698 *	not set request.
699 */
700
701#if __STDC__
702void
703set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
704#else
705void
706set_ftime(fnm, mtime, atime, frc)
707	char *fnm;
708	time_t mtime;
709	time_t atime;
710	int frc;
711#endif
712{
713	static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
714	struct stat sb;
715
716	tv[0].tv_sec = (long)atime;
717	tv[1].tv_sec = (long)mtime;
718	if (!frc && (!patime || !pmtime)) {
719		/*
720		 * if we are not forcing, only set those times the user wants
721		 * set. We get the current values of the times if we need them.
722		 */
723		if (lstat(fnm, &sb) == 0) {
724			if (!patime)
725				tv[0].tv_sec = (long)sb.st_atime;
726			if (!pmtime)
727				tv[1].tv_sec = (long)sb.st_mtime;
728		} else
729			syswarn(0,errno,"Unable to obtain file stats %s", fnm);
730	}
731
732	/*
733	 * set the times
734	 */
735	if (utimes(fnm, tv) < 0)
736		syswarn(1, errno, "Access/modification time set failed on: %s",
737		    fnm);
738	return;
739}
740
741/*
742 * set_ids()
743 *	set the uid and gid of a file system node
744 * Return:
745 *	0 when set, -1 on failure
746 */
747
748#if __STDC__
749int
750set_ids(char *fnm, uid_t uid, gid_t gid)
751#else
752int
753set_ids(fnm, uid, gid)
754	char *fnm;
755	uid_t uid;
756	gid_t gid;
757#endif
758{
759	if (chown(fnm, uid, gid) < 0) {
760		syswarn(1, errno, "Unable to set file uid/gid of %s", fnm);
761		return(-1);
762	}
763	return(0);
764}
765
766/*
767 * set_pmode()
768 *	Set file access mode
769 */
770
771#if __STDC__
772void
773set_pmode(char *fnm, mode_t mode)
774#else
775void
776set_pmode(fnm, mode)
777	char *fnm;
778	mode_t mode;
779#endif
780{
781	mode &= ABITS;
782	if (chmod(fnm, mode) < 0)
783		syswarn(1, errno, "Could not set permissions on %s", fnm);
784	return;
785}
786
787/*
788 * file_write()
789 *	Write/copy a file (during copy or archive extract). This routine knows
790 *	how to copy files with lseek holes in it. (Which are read as file
791 *	blocks containing all 0's but do not have any file blocks associated
792 *	with the data). Typical examples of these are files created by dbm
793 *	variants (.pag files). While the file size of these files are huge, the
794 *	actual storage is quite small (the files are sparse). The problem is
795 *	the holes read as all zeros so are probably stored on the archive that
796 *	way (there is no way to determine if the file block is really a hole,
797 *	we only know that a file block of all zero's can be a hole).
798 *	At this writing, no major archive format knows how to archive files
799 *	with holes. However, on extraction (or during copy, -rw) we have to
800 *	deal with these files. Without detecting the holes, the files can
801 *	consume a lot of file space if just written to disk. This replacement
802 *	for write when passed the basic allocation size of a file system block,
803 *	uses lseek whenever it detects the input data is all 0 within that
804 *	file block. In more detail, the strategy is as follows:
805 *	While the input is all zero keep doing an lseek. Keep track of when we
806 *	pass over file block boundries. Only write when we hit a non zero
807 *	input. once we have written a file block, we continue to write it to
808 *	the end (we stop looking at the input). When we reach the start of the
809 *	next file block, start checking for zero blocks again. Working on file
810 *	block boundries significantly reduces the overhead when copying files
811 *	that are NOT very sparse. This overhead (when compared to a write) is
812 *	almost below the measurement resolution on many systems. Without it,
813 *	files with holes cannot be safely copied. It does has a side effect as
814 *	it can put holes into files that did not have them before, but that is
815 *	not a problem since the file contents are unchanged (in fact it saves
816 *	file space). (Except on paging files for diskless clients. But since we
817 *	cannot determine one of those file from here, we ignore them). If this
818 *	ever ends up on a system where CTG files are supported and the holes
819 *	are not desired, just do a conditional test in those routines that
820 *	call file_write() and have it call write() instead. BEFORE CLOSING THE
821 *	FILE, make sure to call file_flush() when the last write finishes with
822 *	an empty block. A lot of file systems will not create an lseek hole at
823 *	the end. In this case we drop a single 0 at the end to force the
824 *	trailing 0's in the file.
825 *	---Parameters---
826 *	rem: how many bytes left in this file system block
827 *	isempt: have we written to the file block yet (is it empty)
828 *	sz: basic file block allocation size
829 *	cnt: number of bytes on this write
830 *	str: buffer to write
831 * Return:
832 *	number of bytes written, -1 on write (or lseek) error.
833 */
834
835#if __STDC__
836int
837file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
838	char *name)
839#else
840int
841file_write(fd, str, cnt, rem, isempt, sz, name)
842	int fd;
843	char *str;
844	int cnt;
845	int *rem;
846	int *isempt;
847	int sz;
848	char *name;
849#endif
850{
851	char *pt;
852	char *end;
853	int wcnt;
854	char *st = str;
855
856	/*
857	 * while we have data to process
858	 */
859	while (cnt) {
860		if (!*rem) {
861			/*
862			 * We are now at the start of file system block again
863			 * (or what we think one is...). start looking for
864			 * empty blocks again
865			 */
866			*isempt = 1;
867			*rem = sz;
868		}
869
870		/*
871		 * only examine up to the end of the current file block or
872		 * remaining characters to write, whatever is smaller
873		 */
874		wcnt = MIN(cnt, *rem);
875		cnt -= wcnt;
876		*rem -= wcnt;
877		if (*isempt) {
878			/*
879			 * have not written to this block yet, so we keep
880			 * looking for zero's
881			 */
882			pt = st;
883			end = st + wcnt;
884
885			/*
886			 * look for a zero filled buffer
887			 */
888			while ((pt < end) && (*pt == '\0'))
889				++pt;
890
891			if (pt == end) {
892				/*
893				 * skip, buf is empty so far
894				 */
895				if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
896					syswarn(1,errno,"File seek on %s",
897					    name);
898					return(-1);
899				}
900				st = pt;
901				continue;
902			}
903			/*
904			 * drat, the buf is not zero filled
905			 */
906			*isempt = 0;
907		}
908
909		/*
910		 * have non-zero data in this file system block, have to write
911		 */
912		if (write(fd, st, wcnt) != wcnt) {
913			syswarn(1, errno, "Failed write to file %s", name);
914			return(-1);
915		}
916		st += wcnt;
917	}
918	return(st - str);
919}
920
921/*
922 * file_flush()
923 *	when the last file block in a file is zero, many file systems will not
924 *	let us create a hole at the end. To get the last block with zeros, we
925 *	write the last BYTE with a zero (back up one byte and write a zero).
926 */
927
928#if __STDC__
929void
930file_flush(int fd, char *fname, int isempt)
931#else
932void
933file_flush(fd, fname, isempt)
934	int fd;
935	char *fname;
936	int isempt;
937#endif
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(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 beed reading (to copy or archive). If we have to
964 *	reset access time (tflag) do so (the times are stored in arcn).
965 */
966
967#if __STDC__
968void
969rdfile_close(ARCHD *arcn, int *fd)
970#else
971void
972rdfile_close(arcn, fd)
973	ARCHD *arcn;
974	int *fd;
975#endif
976{
977	/*
978	 * make sure the file is open
979	 */
980	if (*fd < 0)
981		return;
982
983	(void)close(*fd);
984	*fd = -1;
985	if (!tflag)
986		return;
987
988	/*
989	 * user wants last access time reset
990	 */
991	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
992	return;
993}
994
995/*
996 * set_crc()
997 *	read a file to calculate its crc. This is a real drag. Archive formats
998 *	that have this, end up reading the file twice (we have to write the
999 *	header WITH the crc before writing the file contents. Oh well...
1000 * Return:
1001 *	0 if was able to calculate the crc, -1 otherwise
1002 */
1003
1004#if __STDC__
1005int
1006set_crc(ARCHD *arcn, int fd)
1007#else
1008int
1009set_crc(arcn, fd)
1010	ARCHD *arcn;
1011	int fd;
1012#endif
1013{
1014	int i;
1015	int res;
1016	off_t cpcnt = 0L;
1017	u_long size;
1018	unsigned long crc = 0L;
1019	char tbuf[FILEBLK];
1020	struct stat sb;
1021
1022	if (fd < 0) {
1023		/*
1024		 * hmm, no fd, should never happen. well no crc then.
1025		 */
1026		arcn->crc = 0L;
1027		return(0);
1028	}
1029
1030	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1031		size = (u_long)sizeof(tbuf);
1032
1033	/*
1034	 * read all the bytes we think that there are in the file. If the user
1035	 * is trying to archive an active file, forget this file.
1036	 */
1037	for(;;) {
1038		if ((res = read(fd, tbuf, size)) <= 0)
1039			break;
1040		cpcnt += res;
1041		for (i = 0; i < res; ++i)
1042			crc += (tbuf[i] & 0xff);
1043	}
1044
1045	/*
1046	 * safety check. we want to avoid archiving files that are active as
1047	 * they can create inconsistant archive copies.
1048	 */
1049	if (cpcnt != arcn->sb.st_size)
1050		tty_warn(1, "File changed size %s", arcn->org_name);
1051	else if (fstat(fd, &sb) < 0)
1052		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1053	else if (arcn->sb.st_mtime != sb.st_mtime)
1054		tty_warn(1, "File %s was modified during read", arcn->org_name);
1055	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1056		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1057	else {
1058		arcn->crc = crc;
1059		return(0);
1060	}
1061	return(-1);
1062}
1063