buf_subs.c revision 1.8
1/*	$NetBSD: buf_subs.c,v 1.8 1997/07/25 23:53:54 scottr 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[] = "@(#)buf_subs.c	8.2 (Berkeley) 4/18/94";
44#else
45__RCSID("$NetBSD: buf_subs.c,v 1.8 1997/07/25 23:53:54 scottr 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 <sys/param.h>
53#include <stdio.h>
54#include <ctype.h>
55#include <errno.h>
56#include <unistd.h>
57#include <stdlib.h>
58#include <string.h>
59#include "pax.h"
60#include "extern.h"
61
62/*
63 * routines which implement archive and file buffering
64 */
65
66#define MINFBSZ		512		/* default block size for hole detect */
67#define MAXFLT          10              /* default media read error limit */
68
69/*
70 * Need to change bufmem to dynamic allocation when the upper
71 * limit on blocking size is removed (though that will violate pax spec)
72 * MAXBLK define and tests will also need to be updated.
73 */
74static char bufmem[MAXBLK+BLKMULT];	/* i/o buffer + pushback id space */
75static char *buf;			/* normal start of i/o buffer */
76static char *bufend;			/* end or last char in i/o buffer */
77static char *bufpt;			/* read/write point in i/o buffer */
78int blksz = MAXBLK;                    	/* block input/output size in bytes */
79int wrblksz;                      	/* user spec output size in bytes */
80int maxflt = MAXFLT;			/* MAX consecutive media errors */
81int rdblksz;				/* first read blksize (tapes only) */
82off_t wrlimit;				/* # of bytes written per archive vol */
83off_t wrcnt;				/* # of bytes written on current vol */
84off_t rdcnt;				/* # of bytes read on current vol */
85
86/*
87 * wr_start()
88 *	set up the buffering system to operate in a write mode
89 * Return:
90 *	0 if ok, -1 if the user specified write block size violates pax spec
91 */
92
93#if __STDC__
94int
95wr_start(void)
96#else
97int
98wr_start()
99#endif
100{
101	buf = &(bufmem[BLKMULT]);
102	/*
103	 * Check to make sure the write block size meets pax specs. If the user
104	 * does not specify a blocksize, we use the format default blocksize.
105	 * We must be picky on writes, so we do not allow the user to create an
106	 * archive that might be hard to read elsewhere. If all ok, we then
107	 * open the first archive volume
108	 */
109	if (!wrblksz)
110		wrblksz = frmt->bsz;
111	if (wrblksz > MAXBLK) {
112		tty_warn(1, "Write block size of %d too large, maximium is: %d",
113			wrblksz, MAXBLK);
114		return(-1);
115	}
116	if (wrblksz % BLKMULT) {
117		tty_warn(1, "Write block size of %d is not a %d byte multiple",
118		    wrblksz, BLKMULT);
119		return(-1);
120	}
121
122	/*
123	 * we only allow wrblksz to be used with all archive operations
124	 */
125	blksz = rdblksz = wrblksz;
126	if ((ar_open(arcname) < 0) && (ar_next() < 0))
127		return(-1);
128	wrcnt = 0;
129	bufend = buf + wrblksz;
130	bufpt = buf;
131	return(0);
132}
133
134/*
135 * rd_start()
136 *	set up buffering system to read an archive
137 * Return:
138 *	0 if ok, -1 otherwise
139 */
140
141#if __STDC__
142int
143rd_start(void)
144#else
145int
146rd_start()
147#endif
148{
149	/*
150	 * leave space for the header pushback (see get_arc()). If we are
151	 * going to append and user specified a write block size, check it
152	 * right away
153	 */
154	buf = &(bufmem[BLKMULT]);
155	if ((act == APPND) && wrblksz) {
156		if (wrblksz > MAXBLK) {
157			tty_warn(1,
158			    "Write block size %d too large, maximium is: %d",
159			    wrblksz, MAXBLK);
160			return(-1);
161		}
162		if (wrblksz % BLKMULT) {
163			tty_warn(1,
164			    "Write block size %d is not a %d byte multiple",
165			    wrblksz, BLKMULT);
166			return(-1);
167		}
168	}
169
170	/*
171	 * open the archive
172	 */
173	if ((ar_open(arcname) < 0) && (ar_next() < 0))
174		return(-1);
175	bufend = buf + rdblksz;
176	bufpt = bufend;
177	rdcnt = 0;
178	return(0);
179}
180
181/*
182 * cp_start()
183 *	set up buffer system for copying within the file system
184 */
185
186#if __STDC__
187void
188cp_start(void)
189#else
190void
191cp_start()
192#endif
193{
194	buf = &(bufmem[BLKMULT]);
195	rdblksz = blksz = MAXBLK;
196}
197
198/*
199 * appnd_start()
200 *	Set up the buffering system to append new members to an archive that
201 *	was just read. The last block(s) of an archive may contain a format
202 *	specific trailer. To append a new member, this trailer has to be
203 *	removed from the archive. The first byte of the trailer is replaced by
204 *	the start of the header of the first file added to the archive. The
205 *	format specific end read function tells us how many bytes to move
206 *	backwards in the archive to be positioned BEFORE the trailer. Two
207 *	different postions have to be adjusted, the O.S. file offset (e.g. the
208 *	position of the tape head) and the write point within the data we have
209 *	stored in the read (soon to become write) buffer. We may have to move
210 *	back several records (the number depends on the size of the archive
211 *	record and the size of the format trailer) to read up the record where
212 *	the first byte of the trailer is recorded. Trailers may span (and
213 *	overlap) record boundries.
214 *	We first calculate which record has the first byte of the trailer. We
215 *	move the OS file offset back to the start of this record and read it
216 *	up. We set the buffer write pointer to be at this byte (the byte where
217 *	the trailer starts). We then move the OS file pointer back to the
218 *	start of this record so a flush of this buffer will replace the record
219 *	in the archive.
220 *	A major problem is rewriting this last record. For archives stored
221 *	on disk files, this is trival. However, many devices are really picky
222 *	about the conditions under which they will allow a write to occur.
223 *	Often devices restrict the conditions where writes can be made writes,
224 *	so it may not be feasable to append archives stored on all types of
225 *	devices.
226 * Return:
227 *	0 for success, -1 for failure
228 */
229
230#if __STDC__
231int
232appnd_start(off_t skcnt)
233#else
234int
235appnd_start(skcnt)
236	off_t skcnt;
237#endif
238{
239	int res;
240	off_t cnt;
241
242#if __GNUC__	/* XXX work around lame compiler problem (gcc 2.7.2) */
243	(void)&cnt;
244#endif
245	if (exit_val != 0) {
246		tty_warn(0, "Cannot append to an archive that may have flaws.");
247		return(-1);
248	}
249	/*
250	 * if the user did not specify a write blocksize, inherit the size used
251	 * in the last archive volume read. (If a is set we still use rdblksz
252	 * until next volume, cannot shift sizes within a single volume).
253	 */
254	if (!wrblksz)
255		wrblksz = blksz = rdblksz;
256	else
257		blksz = rdblksz;
258
259	/*
260	 * make sure that this volume allows appends
261	 */
262	if (ar_app_ok() < 0)
263		return(-1);
264
265	/*
266	 * Calculate bytes to move back and move in front of record where we
267	 * need to start writing from. Remember we have to add in any padding
268	 * that might be in the buffer after the trailer in the last block. We
269	 * travel skcnt + padding ROUNDED UP to blksize.
270	 */
271	skcnt += bufend - bufpt;
272	if ((cnt = (skcnt/blksz) * blksz) < skcnt)
273		cnt += blksz;
274	if (ar_rev((off_t)cnt) < 0)
275		goto out;
276
277	/*
278	 * We may have gone too far if there is valid data in the block we are
279	 * now in front of, read up the block and position the pointer after
280	 * the valid data.
281	 */
282	if ((cnt -= skcnt) > 0) {
283		/*
284		 * watch out for stupid tape drives. ar_rev() will set rdblksz
285		 * to be real physical blocksize so we must loop until we get
286		 * the old rdblksz (now in blksz). If ar_rev() fouls up the
287		 * determination of the physical block size, we will fail.
288		 */
289		bufpt = buf;
290		bufend = buf + blksz;
291		while (bufpt < bufend) {
292			if ((res = ar_read(bufpt, rdblksz)) <= 0)
293				goto out;
294			bufpt += res;
295		}
296		if (ar_rev((off_t)(bufpt - buf)) < 0)
297			goto out;
298		bufpt = buf + cnt;
299		bufend = buf + blksz;
300	} else {
301		/*
302		 * buffer is empty
303		 */
304		bufend = buf + blksz;
305		bufpt = buf;
306	}
307	rdblksz = blksz;
308	rdcnt -= skcnt;
309	wrcnt = 0;
310
311	/*
312	 * At this point we are ready to write. If the device requires special
313	 * handling to write at a point were previously recorded data resides,
314	 * that is handled in ar_set_wr(). From now on we operate under normal
315	 * ARCHIVE mode (write) conditions
316	 */
317	if (ar_set_wr() < 0)
318		return(-1);
319	act = ARCHIVE;
320	return(0);
321
322    out:
323	tty_warn(1, "Unable to rewrite archive trailer, cannot append.");
324	return(-1);
325}
326
327/*
328 * rd_sync()
329 *	A read error occurred on this archive volume. Resync the buffer and
330 *	try to reset the device (if possible) so we can continue to read. Keep
331 *	trying to do this until we get a valid read, or we reach the limit on
332 *	consecutive read faults (at which point we give up). The user can
333 *	adjust the read error limit through a command line option.
334 * Returns:
335 *	0 on success, and -1 on failure
336 */
337
338#if __STDC__
339int
340rd_sync(void)
341#else
342int
343rd_sync()
344#endif
345{
346	int errcnt = 0;
347	int res;
348
349	/*
350	 * if the user says bail out on first fault, we are out of here...
351	 */
352	if (maxflt == 0)
353		return(-1);
354	if (act == APPND) {
355		tty_warn(1,
356		    "Unable to append when there are archive read errors.");
357		return(-1);
358	}
359
360	/*
361	 * poke at device and try to get past media error
362	 */
363	if (ar_rdsync() < 0) {
364		if (ar_next() < 0)
365			return(-1);
366		else
367			rdcnt = 0;
368	}
369
370	for (;;) {
371		if ((res = ar_read(buf, blksz)) > 0) {
372			/*
373			 * All right! got some data, fill that buffer
374			 */
375			bufpt = buf;
376			bufend = buf + res;
377			rdcnt += res;
378			return(0);
379		}
380
381		/*
382		 * Oh well, yet another failed read...
383		 * if error limit reached, ditch. o.w. poke device to move past
384		 * bad media and try again. if media is badly damaged, we ask
385		 * the poor (and upset user at this point) for the next archive
386		 * volume. remember the goal on reads is to get the most we
387		 * can extract out of the archive.
388		 */
389		if ((maxflt > 0) && (++errcnt > maxflt))
390			tty_warn(0,
391			    "Archive read error limit (%d) reached",maxflt);
392		else if (ar_rdsync() == 0)
393			continue;
394		if (ar_next() < 0)
395			break;
396		rdcnt = 0;
397		errcnt = 0;
398	}
399	return(-1);
400}
401
402/*
403 * pback()
404 *	push the data used during the archive id phase back into the I/O
405 *	buffer. This is required as we cannot be sure that the header does NOT
406 *	overlap a block boundry (as in the case we are trying to recover a
407 *	flawed archived). This was not designed to be used for any other
408 *	purpose. (What software engineering, HA!)
409 *	WARNING: do not even THINK of pback greater than BLKMULT, unless the
410 *	pback space is increased.
411 */
412
413#if __STDC__
414void
415pback(char *pt, int cnt)
416#else
417void
418pback(pt, cnt)
419	char *pt;
420	int cnt;
421#endif
422{
423	bufpt -= cnt;
424	memcpy(bufpt, pt, cnt);
425	return;
426}
427
428/*
429 * rd_skip()
430 *	skip foward in the archive during a archive read. Used to get quickly
431 *	past file data and padding for files the user did NOT select.
432 * Return:
433 *	0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
434 */
435
436#if __STDC__
437int
438rd_skip(off_t skcnt)
439#else
440int
441rd_skip(skcnt)
442	off_t skcnt;
443#endif
444{
445	off_t res;
446	off_t cnt;
447	off_t skipped = 0;
448
449#if __GNUC__	/* XXX work around lame compiler problem (gcc 2.7.2) */
450	(void)&cnt;
451#endif
452
453	/*
454	 * consume what data we have in the buffer. If we have to move foward
455	 * whole records, we call the low level skip function to see if we can
456	 * move within the archive without doing the expensive reads on data we
457	 * do not want.
458	 */
459	if (skcnt == 0)
460		return(0);
461	res = MIN((bufend - bufpt), skcnt);
462	bufpt += res;
463	skcnt -= res;
464
465	/*
466	 * if skcnt is now 0, then no additional i/o is needed
467	 */
468	if (skcnt == 0)
469		return(0);
470
471	/*
472	 * We have to read more, calculate complete and partial record reads
473	 * based on rdblksz. we skip over "cnt" complete records
474	 */
475	res = skcnt%rdblksz;
476	cnt = (skcnt/rdblksz) * rdblksz;
477
478	/*
479	 * if the skip fails, we will have to resync. ar_fow will tell us
480	 * how much it can skip over. We will have to read the rest.
481	 */
482	if (ar_fow(cnt, &skipped) < 0)
483		return(-1);
484	res += cnt - skipped;
485	rdcnt += skipped;
486
487	/*
488	 * what is left we have to read (which may be the whole thing if
489	 * ar_fow() told us the device can only read to skip records);
490	 */
491	while (res > 0L) {
492		cnt = bufend - bufpt;
493		/*
494		 * if the read fails, we will have to resync
495		 */
496		if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
497			return(-1);
498		if (cnt == 0)
499			return(1);
500		cnt = MIN(cnt, res);
501		bufpt += cnt;
502		res -= cnt;
503	}
504	return(0);
505}
506
507/*
508 * wr_fin()
509 *	flush out any data (and pad if required) the last block. We always pad
510 *	with zero (even though we do not have to). Padding with 0 makes it a
511 *	lot easier to recover if the archive is damaged. zero paddding SHOULD
512 *	BE a requirement....
513 */
514
515#if __STDC__
516void
517wr_fin(void)
518#else
519void
520wr_fin()
521#endif
522{
523	if (bufpt > buf) {
524		memset(bufpt, 0, bufend - bufpt);
525		bufpt = bufend;
526		(void)buf_flush(blksz);
527	}
528}
529
530/*
531 * wr_rdbuf()
532 *	fill the write buffer from data passed to it in a buffer (usually used
533 *	by format specific write routines to pass a file header). On failure we
534 *	punt. We do not allow the user to continue to write flawed archives.
535 *	We assume these headers are not very large (the memory copy we use is
536 *	a bit expensive).
537 * Return:
538 *	0 if buffer was filled ok, -1 o.w. (buffer flush failure)
539 */
540
541#if __STDC__
542int
543wr_rdbuf(char *out, int outcnt)
544#else
545int
546wr_rdbuf(out, outcnt)
547	char *out;
548	int outcnt;
549#endif
550{
551	int cnt;
552
553	/*
554	 * while there is data to copy copy into the write buffer. when the
555	 * write buffer fills, flush it to the archive and continue
556	 */
557	while (outcnt > 0) {
558		cnt = bufend - bufpt;
559		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
560			return(-1);
561		/*
562		 * only move what we have space for
563		 */
564		cnt = MIN(cnt, outcnt);
565		memcpy(bufpt, out, cnt);
566		bufpt += cnt;
567		out += cnt;
568		outcnt -= cnt;
569	}
570	return(0);
571}
572
573/*
574 * rd_wrbuf()
575 *	copy from the read buffer into a supplied buffer a specified number of
576 *	bytes. If the read buffer is empty fill it and continue to copy.
577 *	usually used to obtain a file header for processing by a format
578 *	specific read routine.
579 * Return
580 *	number of bytes copied to the buffer, 0 indicates EOF on archive volume,
581 *	-1 is a read error
582 */
583
584#if __STDC__
585int
586rd_wrbuf(char *in, int cpcnt)
587#else
588int
589rd_wrbuf(in, cpcnt)
590	char *in;
591	int cpcnt;
592#endif
593{
594	int res;
595	int cnt;
596	int incnt = cpcnt;
597
598	/*
599	 * loop until we fill the buffer with the requested number of bytes
600	 */
601	while (incnt > 0) {
602		cnt = bufend - bufpt;
603		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
604			/*
605			 * read error, return what we got (or the error if
606			 * no data was copied). The caller must know that an
607			 * error occured and has the best knowledge what to
608			 * do with it
609			 */
610			if ((res = cpcnt - incnt) > 0)
611				return(res);
612			return(cnt);
613		}
614
615		/*
616		 * calculate how much data to copy based on whats left and
617		 * state of buffer
618		 */
619		cnt = MIN(cnt, incnt);
620		memcpy(in, bufpt, cnt);
621		bufpt += cnt;
622		incnt -= cnt;
623		in += cnt;
624	}
625	return(cpcnt);
626}
627
628/*
629 * wr_skip()
630 *	skip foward during a write. In other words add padding to the file.
631 *	we add zero filled padding as it makes flawed archives much easier to
632 *	recover from. the caller tells us how many bytes of padding to add
633 *	This routine was not designed to add HUGE amount of padding, just small
634 *	amounts (a few 512 byte blocks at most)
635 * Return:
636 *	0 if ok, -1 if there was a buf_flush failure
637 */
638
639#if __STDC__
640int
641wr_skip(off_t skcnt)
642#else
643int
644wr_skip(skcnt)
645	off_t skcnt;
646#endif
647{
648	int cnt;
649
650	/*
651	 * loop while there is more padding to add
652	 */
653	while (skcnt > 0L) {
654		cnt = bufend - bufpt;
655		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
656			return(-1);
657		cnt = MIN(cnt, skcnt);
658		memset(bufpt, 0, cnt);
659		bufpt += cnt;
660		skcnt -= cnt;
661	}
662	return(0);
663}
664
665/*
666 * wr_rdfile()
667 *	fill write buffer with the contents of a file. We are passed an	open
668 *	file descriptor to the file an the archive structure that describes the
669 *	file we are storing. The variable "left" is modified to contain the
670 *	number of bytes of the file we were NOT able to write to the archive.
671 *	it is important that we always write EXACTLY the number of bytes that
672 *	the format specific write routine told us to. The file can also get
673 *	bigger, so reading to the end of file would create an improper archive,
674 *	we just detect this case and warn the user. We never create a bad
675 *	archive if we can avoid it. Of course trying to archive files that are
676 *	active is asking for trouble. It we fail, we pass back how much we
677 *	could NOT copy and let the caller deal with it.
678 * Return:
679 *	0 ok, -1 if archive write failure. a short read of the file returns a
680 *	0, but "left" is set to be greater than zero.
681 */
682
683#if __STDC__
684int
685wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
686#else
687int
688wr_rdfile(arcn, ifd, left)
689	ARCHD *arcn;
690	int ifd;
691	off_t *left;
692#endif
693{
694	int cnt;
695	int res = 0;
696	off_t size = arcn->sb.st_size;
697	struct stat sb;
698
699	/*
700	 * while there are more bytes to write
701	 */
702	while (size > 0L) {
703		cnt = bufend - bufpt;
704		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
705			*left = size;
706			return(-1);
707		}
708		cnt = MIN(cnt, size);
709		if ((res = read(ifd, bufpt, cnt)) <= 0)
710			break;
711		size -= res;
712		bufpt += res;
713	}
714
715	/*
716	 * better check the file did not change during this operation
717	 * or the file read failed.
718	 */
719	if (res < 0)
720		syswarn(1, errno, "Read fault on %s", arcn->org_name);
721	else if (size != 0L)
722		tty_warn(1, "File changed size during read %s", arcn->org_name);
723	else if (fstat(ifd, &sb) < 0)
724		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
725	else if (arcn->sb.st_mtime != sb.st_mtime)
726		tty_warn(1, "File %s was modified during copy to archive",
727			arcn->org_name);
728	*left = size;
729	return(0);
730}
731
732/*
733 * rd_wrfile()
734 *	extract the contents of a file from the archive. If we are unable to
735 *	extract the entire file (due to failure to write the file) we return
736 *	the numbers of bytes we did NOT process. This way the caller knows how
737 *	many bytes to skip past to find the next archive header. If the failure
738 *	was due to an archive read, we will catch that when we try to skip. If
739 *	the format supplies a file data crc value, we calculate the actual crc
740 *	so that it can be compared to the value stored in the header
741 * NOTE:
742 *	We call a special function to write the file. This function attempts to
743 *	restore file holes (blocks of zeros) into the file. When files are
744 *	sparse this saves space, and is a LOT faster. For non sparse files
745 *	the performance hit is small. As of this writing, no archive supports
746 *	information on where the file holes are.
747 * Return:
748 *	0 ok, -1 if archive read failure. if we cannot write the entire file,
749 *	we return a 0 but "left" is set to be the amount unwritten
750 */
751
752#if __STDC__
753int
754rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
755#else
756int
757rd_wrfile(arcn, ofd, left)
758	ARCHD *arcn;
759	int ofd;
760	off_t *left;
761#endif
762{
763	int cnt = 0;
764	off_t size = arcn->sb.st_size;
765	int res = 0;
766	char *fnm = arcn->name;
767	int isem = 1;
768	int rem;
769	int sz = MINFBSZ;
770 	struct stat sb;
771	u_long crc = 0L;
772
773	/*
774	 * pass the blocksize of the file being written to the write routine,
775	 * if the size is zero, use the default MINFBSZ
776	 */
777        if (fstat(ofd, &sb) == 0) {
778		if (sb.st_blksize > 0)
779			sz = (int)sb.st_blksize;
780        } else
781                syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
782	rem = sz;
783	*left = 0L;
784
785	/*
786	 * Copy the archive to the file the number of bytes specified. We have
787	 * to assume that we want to recover file holes as none of the archive
788	 * formats can record the location of file holes.
789	 */
790	while (size > 0L) {
791		cnt = bufend - bufpt;
792		/*
793		 * if we get a read error, we do not want to skip, as we may
794		 * miss a header, so we do not set left, but if we get a write
795		 * error, we do want to skip over the unprocessed data.
796		 */
797		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
798			break;
799		cnt = MIN(cnt, size);
800		if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
801			*left = size;
802			break;
803		}
804
805		if (docrc) {
806			/*
807			 * update the actual crc value
808			 */
809			cnt = res;
810			while (--cnt >= 0)
811				crc += *bufpt++ & 0xff;
812		} else
813			bufpt += res;
814		size -= res;
815	}
816
817	/*
818	 * if the last block has a file hole (all zero), we must make sure this
819	 * gets updated in the file. We force the last block of zeros to be
820	 * written. just closing with the file offset moved foward may not put
821	 * a hole at the end of the file.
822	 */
823	if (isem && (arcn->sb.st_size > 0L))
824		file_flush(ofd, fnm, isem);
825
826	/*
827	 * if we failed from archive read, we do not want to skip
828	 */
829	if ((size > 0L) && (*left == 0L))
830		return(-1);
831
832	/*
833	 * some formats record a crc on file data. If so, then we compare the
834	 * calculated crc to the crc stored in the archive
835	 */
836	if (docrc && (size == 0L) && (arcn->crc != crc))
837		tty_warn(1,"Actual crc does not match expected crc %s",
838		    arcn->name);
839	return(0);
840}
841
842/*
843 * cp_file()
844 *	copy the contents of one file to another. used during -rw phase of pax
845 *	just as in rd_wrfile() we use a special write function to write the
846 *	destination file so we can properly copy files with holes.
847 */
848
849#if __STDC__
850void
851cp_file(ARCHD *arcn, int fd1, int fd2)
852#else
853void
854cp_file(arcn, fd1, fd2)
855	ARCHD *arcn;
856	int fd1;
857	int fd2;
858#endif
859{
860	int cnt;
861	off_t cpcnt = 0L;
862	int res = 0;
863	char *fnm = arcn->name;
864	int no_hole = 0;
865	int isem = 1;
866	int rem;
867	int sz = MINFBSZ;
868	struct stat sb;
869
870	/*
871	 * check for holes in the source file. If none, we will use regular
872	 * write instead of file write.
873	 */
874	 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
875		++no_hole;
876
877	/*
878	 * pass the blocksize of the file being written to the write routine,
879	 * if the size is zero, use the default MINFBSZ
880	 */
881        if (fstat(fd2, &sb) == 0) {
882		if (sb.st_blksize > 0)
883			sz = sb.st_blksize;
884        } else
885                syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
886	rem = sz;
887
888	/*
889	 * read the source file and copy to destination file until EOF
890	 */
891	for(;;) {
892		if ((cnt = read(fd1, buf, blksz)) <= 0)
893			break;
894		if (no_hole)
895			res = write(fd2, buf, cnt);
896		else
897			res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
898		if (res != cnt)
899			break;
900		cpcnt += cnt;
901	}
902
903	/*
904	 * check to make sure the copy is valid.
905	 */
906	if (res < 0)
907		syswarn(1, errno, "Failed write during copy of %s to %s",
908			arcn->org_name, arcn->name);
909	else if (cpcnt != arcn->sb.st_size)
910		tty_warn(1, "File %s changed size during copy to %s",
911			arcn->org_name, arcn->name);
912	else if (fstat(fd1, &sb) < 0)
913		syswarn(1, errno, "Failed stat of %s", arcn->org_name);
914	else if (arcn->sb.st_mtime != sb.st_mtime)
915		tty_warn(1, "File %s was modified during copy to %s",
916			arcn->org_name, arcn->name);
917
918	/*
919	 * if the last block has a file hole (all zero), we must make sure this
920	 * gets updated in the file. We force the last block of zeros to be
921	 * written. just closing with the file offset moved foward may not put
922	 * a hole at the end of the file.
923	 */
924	if (!no_hole && isem && (arcn->sb.st_size > 0L))
925		file_flush(fd2, fnm, isem);
926	return;
927}
928
929/*
930 * buf_fill()
931 *	fill the read buffer with the next record (or what we can get) from
932 *	the archive volume.
933 * Return:
934 *	Number of bytes of data in the read buffer, -1 for read error, and
935 *	0 when finished (user specified termination in ar_next()).
936 */
937
938#if __STDC__
939int
940buf_fill(void)
941#else
942int
943buf_fill()
944#endif
945{
946	int cnt;
947	static int fini = 0;
948
949	if (fini)
950		return(0);
951
952	for(;;) {
953		/*
954		 * try to fill the buffer. on error the next archive volume is
955		 * opened and we try again.
956		 */
957		if ((cnt = ar_read(buf, blksz)) > 0) {
958			bufpt = buf;
959			bufend = buf + cnt;
960			rdcnt += cnt;
961			return(cnt);
962		}
963
964		/*
965		 * errors require resync, EOF goes to next archive
966		 */
967		if (cnt < 0)
968			break;
969		if (ar_next() < 0) {
970			fini = 1;
971			return(0);
972		}
973		rdcnt = 0;
974	}
975	exit_val = 1;
976	return(-1);
977}
978
979/*
980 * buf_flush()
981 *	force the write buffer to the archive. We are passed the number of
982 *	bytes in the buffer at the point of the flush. When we change archives
983 *	the record size might change. (either larger or smaller).
984 * Return:
985 *	0 if all is ok, -1 when a write error occurs.
986 */
987
988#if __STDC__
989int
990buf_flush(int bufcnt)
991#else
992int
993buf_flush(bufcnt)
994	int bufcnt;
995#endif
996{
997	int cnt;
998	int push = 0;
999	int totcnt = 0;
1000
1001	/*
1002	 * if we have reached the user specified byte count for each archive
1003	 * volume, prompt for the next volume. (The non-standrad -R flag).
1004	 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
1005	 * at least one record. We always round limit UP to next blocksize.
1006	 */
1007	if ((wrlimit > 0) && (wrcnt > wrlimit)) {
1008		tty_warn(0,
1009		    "User specified archive volume byte limit reached.");
1010		if (ar_next() < 0) {
1011			wrcnt = 0;
1012			exit_val = 1;
1013			return(-1);
1014		}
1015		wrcnt = 0;
1016
1017		/*
1018		 * The new archive volume might have changed the size of the
1019		 * write blocksize. if so we figure out if we need to write
1020		 * (one or more times), or if there is now free space left in
1021		 * the buffer (it is no longer full). bufcnt has the number of
1022		 * bytes in the buffer, (the blocksize, at the point we were
1023		 * CALLED). Push has the amount of "extra" data in the buffer
1024		 * if the block size has shrunk from a volume change.
1025		 */
1026		bufend = buf + blksz;
1027		if (blksz > bufcnt)
1028			return(0);
1029		if (blksz < bufcnt)
1030			push = bufcnt - blksz;
1031	}
1032
1033	/*
1034	 * We have enough data to write at least one archive block
1035	 */
1036	for (;;) {
1037		/*
1038		 * write a block and check if it all went out ok
1039		 */
1040		cnt = ar_write(buf, blksz);
1041		if (cnt == blksz) {
1042			/*
1043			 * the write went ok
1044			 */
1045			wrcnt += cnt;
1046			totcnt += cnt;
1047			if (push > 0) {
1048				/* we have extra data to push to the front.
1049				 * check for more than 1 block of push, and if
1050				 * so we loop back to write again
1051				 */
1052				memcpy(buf, bufend, push);
1053				bufpt = buf + push;
1054				if (push >= blksz) {
1055					push -= blksz;
1056					continue;
1057				}
1058			} else
1059				bufpt = buf;
1060			return(totcnt);
1061		} else if (cnt > 0) {
1062			/*
1063			 * Oh drat we got a partial write!
1064			 * if format doesnt care about alignment let it go,
1065			 * we warned the user in ar_write().... but this means
1066			 * the last record on this volume violates pax spec....
1067			 */
1068			totcnt += cnt;
1069			wrcnt += cnt;
1070			bufpt = buf + cnt;
1071			cnt = bufcnt - cnt;
1072			memcpy(buf, bufpt, cnt);
1073			bufpt = buf + cnt;
1074			if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
1075				return(totcnt);
1076			break;
1077		}
1078
1079		/*
1080		 * All done, go to next archive
1081		 */
1082		wrcnt = 0;
1083		if (ar_next() < 0)
1084			break;
1085
1086		/*
1087		 * The new archive volume might also have changed the block
1088		 * size. if so, figure out if we have too much or too little
1089		 * data for using the new block size
1090		 */
1091		bufend = buf + blksz;
1092		if (blksz > bufcnt)
1093			return(0);
1094		if (blksz < bufcnt)
1095			push = bufcnt - blksz;
1096	}
1097
1098	/*
1099	 * write failed, stop pax. we must not create a bad archive!
1100	 */
1101	exit_val = 1;
1102	return(-1);
1103}
1104