file_subs.c revision 28904
1228763Smm/*-
2228753Smm * Copyright (c) 1992 Keith Muller.
3228753Smm * Copyright (c) 1992, 1993
4228753Smm *	The Regents of the University of California.  All rights reserved.
5228753Smm *
6228753Smm * This code is derived from software contributed to Berkeley by
7228753Smm * Keith Muller of the University of California, San Diego.
8228753Smm *
9228753Smm * Redistribution and use in source and binary forms, with or without
10228753Smm * modification, are permitted provided that the following conditions
11228753Smm * are met:
12228753Smm * 1. Redistributions of source code must retain the above copyright
13228753Smm *    notice, this list of conditions and the following disclaimer.
14228753Smm * 2. Redistributions in binary form must reproduce the above copyright
15228753Smm *    notice, this list of conditions and the following disclaimer in the
16228753Smm *    documentation and/or other materials provided with the distribution.
17228753Smm * 3. All advertising materials mentioning features or use of this software
18228753Smm *    must display the following acknowledgement:
19228753Smm *	This product includes software developed by the University of
20228753Smm *	California, Berkeley and its contributors.
21228753Smm * 4. Neither the name of the University nor the names of its contributors
22228753Smm *    may be used to endorse or promote products derived from this software
23228753Smm *    without specific prior written permission.
24228753Smm *
25228753Smm * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26228753Smm * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27228753Smm * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28228753Smm * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29228753Smm * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30228753Smm * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31228753Smm * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32228753Smm * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33228753Smm * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34228753Smm * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35228753Smm * SUCH DAMAGE.
36228753Smm *
37228753Smm *	$Id: file_subs.c,v 1.7 1997/02/22 14:04:26 peter Exp $
38228753Smm */
39228753Smm
40#ifndef lint
41static char const sccsid[] = "@(#)file_subs.c	8.1 (Berkeley) 5/31/93";
42#endif /* not lint */
43
44#include <sys/types.h>
45#include <sys/time.h>
46#include <sys/stat.h>
47#include <unistd.h>
48#include <sys/param.h>
49#include <fcntl.h>
50#include <string.h>
51#include <stdio.h>
52#include <errno.h>
53#include <sys/uio.h>
54#include <stdlib.h>
55#include "pax.h"
56#include "extern.h"
57
58static int
59mk_link __P((register char *,register struct stat *,register char *, int));
60
61/*
62 * routines that deal with file operations such as: creating, removing;
63 * and setting access modes, uid/gid and times of files
64 */
65
66#define FILEBITS		(S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
67#define SETBITS			(S_ISUID | S_ISGID)
68#define ABITS			(FILEBITS | SETBITS)
69
70/*
71 * file_creat()
72 *	Create and open a file.
73 * Return:
74 *	file descriptor or -1 for failure
75 */
76
77#if __STDC__
78int
79file_creat(register ARCHD *arcn)
80#else
81int
82file_creat(arcn)
83	register ARCHD *arcn;
84#endif
85{
86	int fd = -1;
87	mode_t file_mode;
88	int oerrno;
89
90	/*
91	 * assume file doesn't exist, so just try to create it, most times this
92	 * works. We have to take special handling when the file does exist. To
93	 * detect this, we use O_EXCL. For example when trying to create a
94	 * file and a character device or fifo exists with the same name, we
95	 * can accidently open the device by mistake (or block waiting to open)
96	 * If we find that the open has failed, then figure spend the effore to
97	 * figure out why. This strategy was found to have better average
98	 * performance in common use than checking the file (and the path)
99	 * first with lstat.
100	 */
101	file_mode = arcn->sb.st_mode & FILEBITS;
102	if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
103	    file_mode)) >= 0)
104		return(fd);
105
106	/*
107	 * the file seems to exist. First we try to get rid of it (found to be
108	 * the second most common failure when traced). If this fails, only
109	 * then we go to the expense to check and create the path to the file
110	 */
111	if (unlnk_exist(arcn->name, arcn->type) != 0)
112		return(-1);
113
114	for (;;) {
115		/*
116		 * try to open it again, if this fails, check all the nodes in
117		 * the path and give it a final try. if chk_path() finds that
118		 * it cannot fix anything, we will skip the last attempt
119		 */
120		if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
121		    file_mode)) >= 0)
122			break;
123		oerrno = errno;
124		if (chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
125			sys_warn(1, oerrno, "Unable to create %s", arcn->name);
126			return(-1);
127		}
128	}
129	return(fd);
130}
131
132/*
133 * file_close()
134 *	Close file descriptor to a file just created by pax. Sets modes,
135 *	ownership and times as required.
136 * Return:
137 *	0 for success, -1 for failure
138 */
139
140#if __STDC__
141void
142file_close(register ARCHD *arcn, int fd)
143#else
144void
145file_close(arcn, fd)
146	register ARCHD *arcn;
147	int fd;
148#endif
149{
150	int res = 0;
151
152	if (fd < 0)
153		return;
154	if (close(fd) < 0)
155		sys_warn(0, errno, "Unable to close file descriptor on %s",
156		    arcn->name);
157
158	/*
159	 * set owner/groups first as this may strip off mode bits we want
160	 * then set file permission modes. Then set file access and
161	 * modification times.
162	 */
163	if (pids)
164		res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
165
166	/*
167	 * IMPORTANT SECURITY NOTE:
168	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
169	 * set uid/gid bits
170	 */
171	if (!pmode || res)
172		arcn->sb.st_mode &= ~(SETBITS);
173	if (pmode)
174		set_pmode(arcn->name, arcn->sb.st_mode);
175	if (patime || pmtime)
176		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
177}
178
179/*
180 * lnk_creat()
181 *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
182 *	must exist;
183 * Return:
184 *	0 if ok, -1 otherwise
185 */
186
187#if __STDC__
188int
189lnk_creat(register ARCHD *arcn)
190#else
191int
192lnk_creat(arcn)
193	register ARCHD *arcn;
194#endif
195{
196	struct stat sb;
197
198	/*
199	 * we may be running as root, so we have to be sure that link target
200	 * is not a directory, so we lstat and check
201	 */
202	if (lstat(arcn->ln_name, &sb) < 0) {
203		sys_warn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
204		    arcn->name);
205		return(-1);
206	}
207
208	if (S_ISDIR(sb.st_mode)) {
209		pax_warn(1, "A hard link to the directory %s is not allowed",
210		    arcn->ln_name);
211		return(-1);
212	}
213
214	return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
215}
216
217/*
218 * cross_lnk()
219 *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
220 *	with the -l flag. No pax_warning or error if this does not succeed (we will
221 *	then just create the file)
222 * Return:
223 *	1 if copy() should try to create this file node
224 *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
225 */
226
227#if __STDC__
228int
229cross_lnk(register ARCHD *arcn)
230#else
231int
232cross_lnk(arcn)
233	register ARCHD *arcn;
234#endif
235{
236	/*
237	 * try to make a link to orginal file (-l flag in copy mode). make sure
238	 * we do not try to link to directories in case we are running as root
239	 * (and it might succeed).
240	 */
241	if (arcn->type == PAX_DIR)
242		return(1);
243	return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
244}
245
246/*
247 * chk_same()
248 *	In copy mode if we are not trying to make hard links between the src
249 *	and destinations, make sure we are not going to overwrite ourselves by
250 *	accident. This slows things down a little, but we have to protect all
251 *	those people who make typing errors.
252 * Return:
253 *	1 the target does not exist, go ahead and copy
254 *	0 skip it file exists (-k) or may be the same as source file
255 */
256
257#if __STDC__
258int
259chk_same(register ARCHD *arcn)
260#else
261int
262chk_same(arcn)
263	register ARCHD *arcn;
264#endif
265{
266	struct stat sb;
267
268	/*
269	 * if file does not exist, return. if file exists and -k, skip it
270	 * quietly
271	 */
272	if (lstat(arcn->name, &sb) < 0)
273		return(1);
274	if (kflag)
275		return(0);
276
277	/*
278	 * better make sure the user does not have src == dest by mistake
279	 */
280	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
281		pax_warn(1, "Unable to copy %s, file would overwrite itself",
282		    arcn->name);
283		return(0);
284	}
285	return(1);
286}
287
288/*
289 * mk_link()
290 *	try to make a hard link between two files. if ign set, we do not
291 *	complain.
292 * Return:
293 *	0 if successful (or we are done with this file but no error, such as
294 *	finding the from file exists and the user has set -k).
295 *	1 when ign was set to indicates we could not make the link but we
296 *	should try to copy/extract the file as that might work (and is an
297 *	allowed option). -1 an error occurred.
298 */
299
300#if __STDC__
301static int
302mk_link(register char *to, register struct stat *to_sb, register char *from,
303	int ign)
304#else
305static int
306mk_link(to, to_sb, from, ign)
307	register char *to;
308	register struct stat *to_sb;
309	register char *from;
310	int ign;
311#endif
312{
313	struct stat sb;
314	int oerrno;
315
316	/*
317	 * if from file exists, it has to be unlinked to make the link. If the
318	 * file exists and -k is set, skip it quietly
319	 */
320	if (lstat(from, &sb) == 0) {
321		if (kflag)
322			return(0);
323
324		/*
325		 * make sure it is not the same file, protect the user
326		 */
327		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
328			pax_warn(1, "Unable to link file %s to itself", to);
329			return(-1);;
330		}
331
332		/*
333		 * try to get rid of the file, based on the type
334		 */
335		if (S_ISDIR(sb.st_mode)) {
336			if (rmdir(from) < 0) {
337				sys_warn(1, errno, "Unable to remove %s", from);
338				return(-1);
339			}
340		} else if (unlink(from) < 0) {
341			if (!ign) {
342				sys_warn(1, errno, "Unable to remove %s", from);
343				return(-1);
344			}
345			return(1);
346		}
347	}
348
349	/*
350	 * from file is gone (or did not exist), try to make the hard link.
351	 * if it fails, check the path and try it again (if chk_path() says to
352	 * try again)
353	 */
354	for (;;) {
355		if (link(to, from) == 0)
356			break;
357		oerrno = errno;
358		if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
359			continue;
360		if (!ign) {
361			sys_warn(1, oerrno, "Could not link to %s from %s", to,
362			    from);
363			return(-1);
364		}
365		return(1);
366	}
367
368	/*
369	 * all right the link was made
370	 */
371	return(0);
372}
373
374/*
375 * node_creat()
376 *	create an entry in the file system (other than a file or hard link).
377 *	If successful, sets uid/gid modes and times as required.
378 * Return:
379 *	0 if ok, -1 otherwise
380 */
381
382#if __STDC__
383int
384node_creat(register ARCHD *arcn)
385#else
386int
387node_creat(arcn)
388	register ARCHD *arcn;
389#endif
390{
391	register int res;
392	register int ign = 0;
393	register int oerrno;
394	register int pass = 0;
395	mode_t file_mode;
396	struct stat sb;
397
398	/*
399	 * create node based on type, if that fails try to unlink the node and
400	 * try again. finally check the path and try again. As noted in the
401	 * file and link creation routines, this method seems to exhibit the
402	 * best performance in general use workloads.
403	 */
404	file_mode = arcn->sb.st_mode & FILEBITS;
405
406	for (;;) {
407		switch(arcn->type) {
408		case PAX_DIR:
409			res = mkdir(arcn->name, file_mode);
410			if (ign)
411				res = 0;
412			break;
413		case PAX_CHR:
414			file_mode |= S_IFCHR;
415			res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
416			break;
417		case PAX_BLK:
418			file_mode |= S_IFBLK;
419			res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
420			break;
421		case PAX_FIF:
422			res = mkfifo(arcn->name, file_mode);
423			break;
424		case PAX_SCK:
425			/*
426			 * Skip sockets, operation has no meaning under BSD
427			 */
428			pax_warn(0,
429			    "%s skipped. Sockets cannot be copied or extracted",
430			    arcn->name);
431			return(-1);
432		case PAX_SLK:
433			if ((res = symlink(arcn->ln_name, arcn->name)) == 0)
434				return(0);
435			break;
436		case PAX_CTG:
437		case PAX_HLK:
438		case PAX_HRG:
439		case PAX_REG:
440		default:
441			/*
442			 * we should never get here
443			 */
444			pax_warn(0, "%s has an unknown file type, skipping",
445				arcn->name);
446			return(-1);
447		}
448
449		/*
450		 * if we were able to create the node break out of the loop,
451		 * otherwise try to unlink the node and try again. if that
452		 * fails check the full path and try a final time.
453		 */
454		if (res == 0)
455			break;
456
457		/*
458		 * we failed to make the node
459		 */
460		oerrno = errno;
461		if ((ign = unlnk_exist(arcn->name, arcn->type)) < 0)
462			return(-1);
463
464		if (++pass <= 1)
465			continue;
466
467		if (chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
468			sys_warn(1, oerrno, "Could not create: %s", arcn->name);
469			return(-1);
470		}
471	}
472
473	/*
474	 * we were able to create the node. set uid/gid, modes and times
475	 */
476	if (pids)
477		res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
478	else
479		res = 0;
480
481	/*
482	 * IMPORTANT SECURITY NOTE:
483	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
484	 * set uid/gid bits
485	 */
486	if (!pmode || res)
487		arcn->sb.st_mode &= ~(SETBITS);
488	if (pmode)
489		set_pmode(arcn->name, arcn->sb.st_mode);
490
491	if (arcn->type == PAX_DIR) {
492		/*
493		 * Dirs must be processed again at end of extract to set times
494		 * and modes to agree with those stored in the archive. However
495		 * to allow extract to continue, we may have to also set owner
496		 * rights. This allows nodes in the archive that are children
497		 * of this directory to be extracted without failure. Both time
498		 * and modes will be fixed after the entire archive is read and
499		 * before pax exits.
500		 */
501		if (access(arcn->name, R_OK | W_OK | X_OK) < 0) {
502			if (lstat(arcn->name, &sb) < 0) {
503				sys_warn(0, errno,"Could not access %s (stat)",
504				    arcn->name);
505				set_pmode(arcn->name,file_mode | S_IRWXU);
506			} else {
507				/*
508				 * We have to add rights to the dir, so we make
509				 * sure to restore the mode. The mode must be
510				 * restored AS CREATED and not as stored if
511				 * pmode is not set.
512				 */
513				set_pmode(arcn->name,
514				    ((sb.st_mode & FILEBITS) | S_IRWXU));
515				if (!pmode)
516					arcn->sb.st_mode = sb.st_mode;
517			}
518
519			/*
520			 * we have to force the mode to what was set here,
521			 * since we changed it from the default as created.
522			 */
523			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1);
524		} else if (pmode || patime || pmtime)
525			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0);
526	}
527
528	if (patime || pmtime)
529		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
530	return(0);
531}
532
533/*
534 * unlnk_exist()
535 *	Remove node from file system with the specified name. We pass the type
536 *	of the node that is going to replace it. When we try to create a
537 *	directory and find that it already exists, we allow processing to
538 *	continue as proper modes etc will always be set for it later on.
539 * Return:
540 *	0 is ok to proceed, no file with the specified name exists
541 *	-1 we were unable to remove the node, or we should not remove it (-k)
542 *	1 we found a directory and we were going to create a directory.
543 */
544
545#if __STDC__
546int
547unlnk_exist(register char *name, register int type)
548#else
549int
550unlnk_exist(name, type)
551	register char *name;
552	register int type;
553#endif
554{
555	struct stat sb;
556
557	/*
558	 * the file does not exist, or -k we are done
559	 */
560	if (lstat(name, &sb) < 0)
561		return(0);
562	if (kflag)
563		return(-1);
564
565	if (S_ISDIR(sb.st_mode)) {
566		/*
567		 * try to remove a directory, if it fails and we were going to
568		 * create a directory anyway, tell the caller (return a 1)
569		 */
570		if (rmdir(name) < 0) {
571			if (type == PAX_DIR)
572				return(1);
573			sys_warn(1,errno,"Unable to remove directory %s", name);
574			return(-1);
575		}
576		return(0);
577	}
578
579	/*
580	 * try to get rid of all non-directory type nodes
581	 */
582	if (unlink(name) < 0) {
583		sys_warn(1, errno, "Could not unlink %s", name);
584		return(-1);
585	}
586	return(0);
587}
588
589/*
590 * chk_path()
591 *	We were trying to create some kind of node in the file system and it
592 *	failed. chk_path() makes sure the path up to the node exists and is
593 *	writeable. When we have to create a directory that is missing along the
594 *	path somewhere, the directory we create will be set to the same
595 *	uid/gid as the file has (when uid and gid are being preserved).
596 *	NOTE: this routine is a real performance loss. It is only used as a
597 *	last resort when trying to create entries in the file system.
598 * Return:
599 *	-1 when it could find nothing it is allowed to fix.
600 *	0 otherwise
601 */
602
603#if __STDC__
604int
605chk_path( register char *name, uid_t st_uid, gid_t st_gid)
606#else
607int
608chk_path(name, st_uid, st_gid)
609	register char *name;
610	uid_t st_uid;
611	gid_t st_gid;
612#endif
613{
614	register char *spt = name;
615	struct stat sb;
616	int retval = -1;
617
618	/*
619	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
620	 */
621	if (*spt == '/')
622		++spt;
623
624	for(;;) {
625		/*
626		 * work foward from the first / and check each part of the path
627		 */
628		spt = strchr(spt, '/');
629		if (spt == NULL)
630			break;
631		*spt = '\0';
632
633		/*
634		 * if it exists we assume it is a directory, it is not within
635		 * the spec (at least it seems to read that way) to alter the
636		 * file system for nodes NOT EXPLICITLY stored on the archive.
637		 * If that assumption is changed, you would test the node here
638		 * and figure out how to get rid of it (probably like some
639		 * recursive unlink()) or fix up the directory permissions if
640		 * required (do an access()).
641		 */
642		if (lstat(name, &sb) == 0) {
643			*(spt++) = '/';
644			continue;
645		}
646
647		/*
648		 * the path fails at this point, see if we can create the
649		 * needed directory and continue on
650		 */
651		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
652			*spt = '/';
653			retval = -1;
654			break;
655		}
656
657		/*
658		 * we were able to create the directory. We will tell the
659		 * caller that we found something to fix, and it is ok to try
660		 * and create the node again.
661		 */
662		retval = 0;
663		if (pids)
664			(void)set_ids(name, st_uid, st_gid);
665
666		/*
667		 * make sure the user doen't have some strange umask that
668		 * causes this newly created directory to be unusable. We fix
669		 * the modes and restore them back to the creation default at
670		 * the end of pax
671		 */
672		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
673		    (lstat(name, &sb) == 0)) {
674			set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
675			add_dir(name, spt - name, &sb, 1);
676		}
677		*(spt++) = '/';
678		continue;
679	}
680	return(retval);
681}
682
683/*
684 * set_ftime()
685 *	Set the access time and modification time for a named file. If frc is
686 *	non-zero we force these times to be set even if the the user did not
687 *	request access and/or modification time preservation (this is also
688 *	used by -t to reset access times).
689 *	When ign is zero, only those times the user has asked for are set, the
690 *	other ones are left alone. We do not assume the un-documented feature
691 *	of many utimes() implementations that consider a 0 time value as a do
692 *	not set request.
693 */
694
695#if __STDC__
696void
697set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
698#else
699void
700set_ftime(fnm, mtime, atime, frc)
701	char *fnm;
702	time_t mtime;
703	time_t atime;
704	int frc;
705#endif
706{
707	static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
708	struct stat sb;
709
710	tv[0].tv_sec = (long)atime;
711	tv[1].tv_sec = (long)mtime;
712	if (!frc && (!patime || !pmtime)) {
713		/*
714		 * if we are not forcing, only set those times the user wants
715		 * set. We get the current values of the times if we need them.
716		 */
717		if (lstat(fnm, &sb) == 0) {
718			if (!patime)
719				tv[0].tv_sec = (long)sb.st_atime;
720			if (!pmtime)
721				tv[1].tv_sec = (long)sb.st_mtime;
722		} else
723			sys_warn(0,errno,"Unable to obtain file stats %s", fnm);
724	}
725
726	/*
727	 * set the times
728	 */
729	if (utimes(fnm, tv) < 0)
730		sys_warn(1, errno, "Access/modification time set failed on: %s",
731		    fnm);
732	return;
733}
734
735/*
736 * set_ids()
737 *	set the uid and gid of a file system node
738 * Return:
739 *	0 when set, -1 on failure
740 */
741
742#if __STDC__
743int
744set_ids(char *fnm, uid_t uid, gid_t gid)
745#else
746int
747set_ids(fnm, uid, gid)
748	char *fnm;
749	uid_t uid;
750	gid_t gid;
751#endif
752{
753	if (chown(fnm, uid, gid) < 0) {
754		sys_warn(1, errno, "Unable to set file uid/gid of %s", fnm);
755		return(-1);
756	}
757	return(0);
758}
759
760/*
761 * set_pmode()
762 *	Set file access mode
763 */
764
765#if __STDC__
766void
767set_pmode(char *fnm, mode_t mode)
768#else
769void
770set_pmode(fnm, mode)
771	char *fnm;
772	mode_t mode;
773#endif
774{
775	mode &= ABITS;
776	if (chmod(fnm, mode) < 0)
777		sys_warn(1, errno, "Could not set permissions on %s", fnm);
778	return;
779}
780
781/*
782 * file_write()
783 *	Write/copy a file (during copy or archive extract). This routine knows
784 *	how to copy files with lseek holes in it. (Which are read as file
785 *	blocks containing all 0's but do not have any file blocks associated
786 *	with the data). Typical examples of these are files created by dbm
787 *	variants (.pag files). While the file size of these files are huge, the
788 *	actual storage is quite small (the files are sparse). The problem is
789 *	the holes read as all zeros so are probably stored on the archive that
790 *	way (there is no way to determine if the file block is really a hole,
791 *	we only know that a file block of all zero's can be a hole).
792 *	At this writing, no major archive format knows how to archive files
793 *	with holes. However, on extraction (or during copy, -rw) we have to
794 *	deal with these files. Without detecting the holes, the files can
795 *	consume a lot of file space if just written to disk. This replacement
796 *	for write when passed the basic allocation size of a file system block,
797 *	uses lseek whenever it detects the input data is all 0 within that
798 *	file block. In more detail, the strategy is as follows:
799 *	While the input is all zero keep doing an lseek. Keep track of when we
800 *	pass over file block boundries. Only write when we hit a non zero
801 *	input. once we have written a file block, we continue to write it to
802 *	the end (we stop looking at the input). When we reach the start of the
803 *	next file block, start checking for zero blocks again. Working on file
804 *	block boundries significantly reduces the overhead when copying files
805 *	that are NOT very sparse. This overhead (when compared to a write) is
806 *	almost below the measurement resolution on many systems. Without it,
807 *	files with holes cannot be safely copied. It does has a side effect as
808 *	it can put holes into files that did not have them before, but that is
809 *	not a problem since the file contents are unchanged (in fact it saves
810 *	file space). (Except on paging files for diskless clients. But since we
811 *	cannot determine one of those file from here, we ignore them). If this
812 *	ever ends up on a system where CTG files are supported and the holes
813 *	are not desired, just do a conditional test in those routines that
814 *	call file_write() and have it call write() instead. BEFORE CLOSING THE
815 *	FILE, make sure to call file_flush() when the last write finishes with
816 *	an empty block. A lot of file systems will not create an lseek hole at
817 *	the end. In this case we drop a single 0 at the end to force the
818 *	trailing 0's in the file.
819 *	---Parameters---
820 *	rem: how many bytes left in this file system block
821 *	isempt: have we written to the file block yet (is it empty)
822 *	sz: basic file block allocation size
823 *	cnt: number of bytes on this write
824 *	str: buffer to write
825 * Return:
826 *	number of bytes written, -1 on write (or lseek) error.
827 */
828
829#if __STDC__
830int
831file_write(int fd, char *str, register int cnt, int *rem, int *isempt, int sz,
832	char *name)
833#else
834int
835file_write(fd, str, cnt, rem, isempt, sz, name)
836	int fd;
837	char *str;
838	register int cnt;
839	int *rem;
840	int *isempt;
841	int sz;
842	char *name;
843#endif
844{
845	register char *pt;
846	register char *end;
847	register int wcnt;
848	register char *st = str;
849
850	/*
851	 * while we have data to process
852	 */
853	while (cnt) {
854		if (!*rem) {
855			/*
856			 * We are now at the start of file system block again
857			 * (or what we think one is...). start looking for
858			 * empty blocks again
859			 */
860			*isempt = 1;
861			*rem = sz;
862		}
863
864		/*
865		 * only examine up to the end of the current file block or
866		 * remaining characters to write, whatever is smaller
867		 */
868		wcnt = MIN(cnt, *rem);
869		cnt -= wcnt;
870		*rem -= wcnt;
871		if (*isempt) {
872			/*
873			 * have not written to this block yet, so we keep
874			 * looking for zero's
875			 */
876			pt = st;
877			end = st + wcnt;
878
879			/*
880			 * look for a zero filled buffer
881			 */
882			while ((pt < end) && (*pt == '\0'))
883				++pt;
884
885			if (pt == end) {
886				/*
887				 * skip, buf is empty so far
888				 */
889				if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
890					sys_warn(1,errno,"File seek on %s",
891					    name);
892					return(-1);
893				}
894				st = pt;
895				continue;
896			}
897			/*
898			 * drat, the buf is not zero filled
899			 */
900			*isempt = 0;
901		}
902
903		/*
904		 * have non-zero data in this file system block, have to write
905		 */
906		if (write(fd, st, wcnt) != wcnt) {
907			sys_warn(1, errno, "Failed write to file %s", name);
908			return(-1);
909		}
910		st += wcnt;
911	}
912	return(st - str);
913}
914
915/*
916 * file_flush()
917 *	when the last file block in a file is zero, many file systems will not
918 *	let us create a hole at the end. To get the last block with zeros, we
919 *	write the last BYTE with a zero (back up one byte and write a zero).
920 */
921
922#if __STDC__
923void
924file_flush(int fd, char *fname, int isempt)
925#else
926void
927file_flush(fd, fname, isempt)
928	int fd;
929	char *fname;
930	int isempt;
931#endif
932{
933	static char blnk[] = "\0";
934
935	/*
936	 * silly test, but make sure we are only called when the last block is
937	 * filled with all zeros.
938	 */
939	if (!isempt)
940		return;
941
942	/*
943	 * move back one byte and write a zero
944	 */
945	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
946		sys_warn(1, errno, "Failed seek on file %s", fname);
947		return;
948	}
949
950	if (write(fd, blnk, 1) < 0)
951		sys_warn(1, errno, "Failed write to file %s", fname);
952	return;
953}
954
955/*
956 * rdfile_close()
957 *	close a file we have beed reading (to copy or archive). If we have to
958 *	reset access time (tflag) do so (the times are stored in arcn).
959 */
960
961#if __STDC__
962void
963rdfile_close(register ARCHD *arcn, register int *fd)
964#else
965void
966rdfile_close(arcn, fd)
967	register ARCHD *arcn;
968	register int *fd;
969#endif
970{
971	/*
972	 * make sure the file is open
973	 */
974	if (*fd < 0)
975		return;
976
977	(void)close(*fd);
978	*fd = -1;
979	if (!tflag)
980		return;
981
982	/*
983	 * user wants last access time reset
984	 */
985	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
986	return;
987}
988
989/*
990 * set_crc()
991 *	read a file to calculate its crc. This is a real drag. Archive formats
992 *	that have this, end up reading the file twice (we have to write the
993 *	header WITH the crc before writing the file contents. Oh well...
994 * Return:
995 *	0 if was able to calculate the crc, -1 otherwise
996 */
997
998#if __STDC__
999int
1000set_crc(register ARCHD *arcn, register int fd)
1001#else
1002int
1003set_crc(arcn, fd)
1004	register ARCHD *arcn;
1005	register int fd;
1006#endif
1007{
1008	register int i;
1009	register int res;
1010	off_t cpcnt = 0L;
1011	u_long size;
1012	unsigned long crc = 0L;
1013	char tbuf[FILEBLK];
1014	struct stat sb;
1015
1016	if (fd < 0) {
1017		/*
1018		 * hmm, no fd, should never happen. well no crc then.
1019		 */
1020		arcn->crc = 0L;
1021		return(0);
1022	}
1023
1024	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1025		size = (u_long)sizeof(tbuf);
1026
1027	/*
1028	 * read all the bytes we think that there are in the file. If the user
1029	 * is trying to archive an active file, forget this file.
1030	 */
1031	for(;;) {
1032		if ((res = read(fd, tbuf, size)) <= 0)
1033			break;
1034		cpcnt += res;
1035		for (i = 0; i < res; ++i)
1036			crc += (tbuf[i] & 0xff);
1037	}
1038
1039	/*
1040	 * safety check. we want to avoid archiving files that are active as
1041	 * they can create inconsistant archive copies.
1042	 */
1043	if (cpcnt != arcn->sb.st_size)
1044		pax_warn(1, "File changed size %s", arcn->org_name);
1045	else if (fstat(fd, &sb) < 0)
1046		sys_warn(1, errno, "Failed stat on %s", arcn->org_name);
1047	else if (arcn->sb.st_mtime != sb.st_mtime)
1048		pax_warn(1, "File %s was modified during read", arcn->org_name);
1049	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1050		sys_warn(1, errno, "File rewind failed on: %s", arcn->org_name);
1051	else {
1052		arcn->crc = crc;
1053		return(0);
1054	}
1055	return(-1);
1056}
1057