buf_subs.c revision 1.22
1/*	$NetBSD: buf_subs.c,v 1.22 2003/10/13 07:41:22 agc 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. Neither the name of the University nor the names of its contributors
20 *    may be used to endorse or promote products derived from this software
21 *    without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 */
35
36#include <sys/cdefs.h>
37#if defined(__RCSID) && !defined(lint)
38#if 0
39static char sccsid[] = "@(#)buf_subs.c	8.2 (Berkeley) 4/18/94";
40#else
41__RCSID("$NetBSD: buf_subs.c,v 1.22 2003/10/13 07:41:22 agc Exp $");
42#endif
43#endif /* not lint */
44
45#include <sys/types.h>
46#include <sys/time.h>
47#include <sys/stat.h>
48#include <sys/param.h>
49#include <stdio.h>
50#include <ctype.h>
51#include <errno.h>
52#include <unistd.h>
53#include <stdlib.h>
54#include <string.h>
55#include "pax.h"
56#include "extern.h"
57
58/*
59 * routines which implement archive and file buffering
60 */
61
62#define MINFBSZ		512		/* default block size for hole detect */
63#define MAXFLT		10		/* default media read error limit */
64
65/*
66 * Need to change bufmem to dynamic allocation when the upper
67 * limit on blocking size is removed (though that will violate pax spec)
68 * MAXBLK define and tests will also need to be updated.
69 */
70static char bufmem[MAXBLK+BLKMULT];	/* i/o buffer + pushback id space */
71static char *buf;			/* normal start of i/o buffer */
72static char *bufend;			/* end or last char in i/o buffer */
73static char *bufpt;			/* read/write point in i/o buffer */
74int blksz = MAXBLK;			/* block input/output size in bytes */
75int wrblksz;				/* user spec output size in bytes */
76int maxflt = MAXFLT;			/* MAX consecutive media errors */
77int rdblksz;				/* first read blksize (tapes only) */
78off_t wrlimit;				/* # of bytes written per archive vol */
79off_t wrcnt;				/* # of bytes written on current vol */
80off_t rdcnt;				/* # of bytes read on current vol */
81
82/*
83 * wr_start()
84 *	set up the buffering system to operate in a write mode
85 * Return:
86 *	0 if ok, -1 if the user specified write block size violates pax spec
87 */
88
89int
90wr_start(void)
91{
92	buf = &(bufmem[BLKMULT]);
93	/*
94	 * Check to make sure the write block size meets pax specs. If the user
95	 * does not specify a blocksize, we use the format default blocksize.
96	 * We must be picky on writes, so we do not allow the user to create an
97	 * archive that might be hard to read elsewhere. If all ok, we then
98	 * open the first archive volume
99	 */
100	if (!wrblksz)
101		wrblksz = frmt->bsz;
102	if (wrblksz > MAXBLK) {
103		tty_warn(1, "Write block size of %d too large, maximum is: %d",
104			wrblksz, MAXBLK);
105		return(-1);
106	}
107	if (wrblksz % BLKMULT) {
108		tty_warn(1, "Write block size of %d is not a %d byte multiple",
109		    wrblksz, BLKMULT);
110		return(-1);
111	}
112
113	/*
114	 * we only allow wrblksz to be used with all archive operations
115	 */
116	blksz = rdblksz = wrblksz;
117	if ((ar_open(arcname) < 0) && (ar_next() < 0))
118		return(-1);
119	wrcnt = 0;
120	bufend = buf + wrblksz;
121	bufpt = buf;
122	return(0);
123}
124
125/*
126 * rd_start()
127 *	set up buffering system to read an archive
128 * Return:
129 *	0 if ok, -1 otherwise
130 */
131
132int
133rd_start(void)
134{
135	/*
136	 * leave space for the header pushback (see get_arc()). If we are
137	 * going to append and user specified a write block size, check it
138	 * right away
139	 */
140	buf = &(bufmem[BLKMULT]);
141	if ((act == APPND) && wrblksz) {
142		if (wrblksz > MAXBLK) {
143			tty_warn(1,
144			    "Write block size %d too large, maximum is: %d",
145			    wrblksz, MAXBLK);
146			return(-1);
147		}
148		if (wrblksz % BLKMULT) {
149			tty_warn(1,
150			    "Write block size %d is not a %d byte multiple",
151			    wrblksz, BLKMULT);
152			return(-1);
153		}
154	}
155
156	/*
157	 * open the archive
158	 */
159	if ((ar_open(arcname) < 0) && (ar_next() < 0))
160		return(-1);
161	bufend = buf + rdblksz;
162	bufpt = bufend;
163	rdcnt = 0;
164	return(0);
165}
166
167/*
168 * cp_start()
169 *	set up buffer system for copying within the file system
170 */
171
172void
173cp_start(void)
174{
175	buf = &(bufmem[BLKMULT]);
176	rdblksz = blksz = MAXBLK;
177}
178
179/*
180 * appnd_start()
181 *	Set up the buffering system to append new members to an archive that
182 *	was just read. The last block(s) of an archive may contain a format
183 *	specific trailer. To append a new member, this trailer has to be
184 *	removed from the archive. The first byte of the trailer is replaced by
185 *	the start of the header of the first file added to the archive. The
186 *	format specific end read function tells us how many bytes to move
187 *	backwards in the archive to be positioned BEFORE the trailer. Two
188 *	different positions have to be adjusted, the O.S. file offset (e.g. the
189 *	position of the tape head) and the write point within the data we have
190 *	stored in the read (soon to become write) buffer. We may have to move
191 *	back several records (the number depends on the size of the archive
192 *	record and the size of the format trailer) to read up the record where
193 *	the first byte of the trailer is recorded. Trailers may span (and
194 *	overlap) record boundaries.
195 *	We first calculate which record has the first byte of the trailer. We
196 *	move the OS file offset back to the start of this record and read it
197 *	up. We set the buffer write pointer to be at this byte (the byte where
198 *	the trailer starts). We then move the OS file pointer back to the
199 *	start of this record so a flush of this buffer will replace the record
200 *	in the archive.
201 *	A major problem is rewriting this last record. For archives stored
202 *	on disk files, this is trivial. However, many devices are really picky
203 *	about the conditions under which they will allow a write to occur.
204 *	Often devices restrict the conditions where writes can be made,
205 *	so it may not be feasable to append archives stored on all types of
206 *	devices.
207 * Return:
208 *	0 for success, -1 for failure
209 */
210
211int
212appnd_start(off_t skcnt)
213{
214	int res;
215	off_t cnt;
216
217	if (exit_val != 0) {
218		tty_warn(0, "Cannot append to an archive that may have flaws.");
219		return(-1);
220	}
221	/*
222	 * if the user did not specify a write blocksize, inherit the size used
223	 * in the last archive volume read. (If a is set we still use rdblksz
224	 * until next volume, cannot shift sizes within a single volume).
225	 */
226	if (!wrblksz)
227		wrblksz = blksz = rdblksz;
228	else
229		blksz = rdblksz;
230
231	/*
232	 * make sure that this volume allows appends
233	 */
234	if (ar_app_ok() < 0)
235		return(-1);
236
237	/*
238	 * Calculate bytes to move back and move in front of record where we
239	 * need to start writing from. Remember we have to add in any padding
240	 * that might be in the buffer after the trailer in the last block. We
241	 * travel skcnt + padding ROUNDED UP to blksize.
242	 */
243	skcnt += bufend - bufpt;
244	if ((cnt = (skcnt/blksz) * blksz) < skcnt)
245		cnt += blksz;
246	if (ar_rev((off_t)cnt) < 0)
247		goto out;
248
249	/*
250	 * We may have gone too far if there is valid data in the block we are
251	 * now in front of, read up the block and position the pointer after
252	 * the valid data.
253	 */
254	if ((cnt -= skcnt) > 0) {
255		/*
256		 * watch out for stupid tape drives. ar_rev() will set rdblksz
257		 * to be real physical blocksize so we must loop until we get
258		 * the old rdblksz (now in blksz). If ar_rev() fouls up the
259		 * determination of the physical block size, we will fail.
260		 */
261		bufpt = buf;
262		bufend = buf + blksz;
263		while (bufpt < bufend) {
264			if ((res = ar_read(bufpt, rdblksz)) <= 0)
265				goto out;
266			bufpt += res;
267		}
268		if (ar_rev((off_t)(bufpt - buf)) < 0)
269			goto out;
270		bufpt = buf + cnt;
271		bufend = buf + blksz;
272	} else {
273		/*
274		 * buffer is empty
275		 */
276		bufend = buf + blksz;
277		bufpt = buf;
278	}
279	rdblksz = blksz;
280	rdcnt -= skcnt;
281	wrcnt = 0;
282
283	/*
284	 * At this point we are ready to write. If the device requires special
285	 * handling to write at a point were previously recorded data resides,
286	 * that is handled in ar_set_wr(). From now on we operate under normal
287	 * ARCHIVE mode (write) conditions
288	 */
289	if (ar_set_wr() < 0)
290		return(-1);
291	act = ARCHIVE;
292	return(0);
293
294    out:
295	tty_warn(1, "Unable to rewrite archive trailer, cannot append.");
296	return(-1);
297}
298
299/*
300 * rd_sync()
301 *	A read error occurred on this archive volume. Resync the buffer and
302 *	try to reset the device (if possible) so we can continue to read. Keep
303 *	trying to do this until we get a valid read, or we reach the limit on
304 *	consecutive read faults (at which point we give up). The user can
305 *	adjust the read error limit through a command line option.
306 * Returns:
307 *	0 on success, and -1 on failure
308 */
309
310int
311rd_sync(void)
312{
313	int errcnt = 0;
314	int res;
315
316	/*
317	 * if the user says bail out on first fault, we are out of here...
318	 */
319	if (maxflt == 0)
320		return(-1);
321	if (act == APPND) {
322		tty_warn(1,
323		    "Unable to append when there are archive read errors.");
324		return(-1);
325	}
326
327	/*
328	 * poke at device and try to get past media error
329	 */
330	if (ar_rdsync() < 0) {
331		if (ar_next() < 0)
332			return(-1);
333		else
334			rdcnt = 0;
335	}
336
337	for (;;) {
338		if ((res = ar_read(buf, blksz)) > 0) {
339			/*
340			 * All right! got some data, fill that buffer
341			 */
342			bufpt = buf;
343			bufend = buf + res;
344			rdcnt += res;
345			return(0);
346		}
347
348		/*
349		 * Oh well, yet another failed read...
350		 * if error limit reached, ditch. otherwise poke device to move past
351		 * bad media and try again. if media is badly damaged, we ask
352		 * the poor (and upset user at this point) for the next archive
353		 * volume. remember the goal on reads is to get the most we
354		 * can extract out of the archive.
355		 */
356		if ((maxflt > 0) && (++errcnt > maxflt))
357			tty_warn(0,
358			    "Archive read error limit (%d) reached",maxflt);
359		else if (ar_rdsync() == 0)
360			continue;
361		if (ar_next() < 0)
362			break;
363		rdcnt = 0;
364		errcnt = 0;
365	}
366	return(-1);
367}
368
369/*
370 * pback()
371 *	push the data used during the archive id phase back into the I/O
372 *	buffer. This is required as we cannot be sure that the header does NOT
373 *	overlap a block boundary (as in the case we are trying to recover a
374 *	flawed archived). This was not designed to be used for any other
375 *	purpose. (What software engineering, HA!)
376 *	WARNING: do not even THINK of pback greater than BLKMULT, unless the
377 *	pback space is increased.
378 */
379
380void
381pback(char *pt, int cnt)
382{
383	bufpt -= cnt;
384	memcpy(bufpt, pt, cnt);
385	return;
386}
387
388/*
389 * rd_skip()
390 *	skip forward in the archive during a archive read. Used to get quickly
391 *	past file data and padding for files the user did NOT select.
392 * Return:
393 *	0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
394 */
395
396int
397rd_skip(off_t skcnt)
398{
399	off_t res;
400	off_t cnt;
401	off_t skipped = 0;
402
403	/*
404	 * consume what data we have in the buffer. If we have to move forward
405	 * whole records, we call the low level skip function to see if we can
406	 * move within the archive without doing the expensive reads on data we
407	 * do not want.
408	 */
409	if (skcnt == 0)
410		return(0);
411	res = MIN((bufend - bufpt), skcnt);
412	bufpt += res;
413	skcnt -= res;
414
415	/*
416	 * if skcnt is now 0, then no additional i/o is needed
417	 */
418	if (skcnt == 0)
419		return(0);
420
421	/*
422	 * We have to read more, calculate complete and partial record reads
423	 * based on rdblksz. we skip over "cnt" complete records
424	 */
425	res = skcnt%rdblksz;
426	cnt = (skcnt/rdblksz) * rdblksz;
427
428	/*
429	 * if the skip fails, we will have to resync. ar_fow will tell us
430	 * how much it can skip over. We will have to read the rest.
431	 */
432	if (ar_fow(cnt, &skipped) < 0)
433		return(-1);
434	res += cnt - skipped;
435	rdcnt += skipped;
436
437	/*
438	 * what is left we have to read (which may be the whole thing if
439	 * ar_fow() told us the device can only read to skip records);
440	 */
441	while (res > 0L) {
442		cnt = bufend - bufpt;
443		/*
444		 * if the read fails, we will have to resync
445		 */
446		if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
447			return(-1);
448		if (cnt == 0)
449			return(1);
450		cnt = MIN(cnt, res);
451		bufpt += cnt;
452		res -= cnt;
453	}
454	return(0);
455}
456
457/*
458 * wr_fin()
459 *	flush out any data (and pad if required) the last block. We always pad
460 *	with zero (even though we do not have to). Padding with 0 makes it a
461 *	lot easier to recover if the archive is damaged. zero paddding SHOULD
462 *	BE a requirement....
463 */
464
465void
466wr_fin(void)
467{
468	if (bufpt > buf) {
469		memset(bufpt, 0, bufend - bufpt);
470		bufpt = bufend;
471		(void)buf_flush(blksz);
472	}
473}
474
475/*
476 * wr_rdbuf()
477 *	fill the write buffer from data passed to it in a buffer (usually used
478 *	by format specific write routines to pass a file header). On failure we
479 *	punt. We do not allow the user to continue to write flawed archives.
480 *	We assume these headers are not very large (the memory copy we use is
481 *	a bit expensive).
482 * Return:
483 *	0 if buffer was filled ok, -1 o.w. (buffer flush failure)
484 */
485
486int
487wr_rdbuf(char *out, int outcnt)
488{
489	int cnt;
490
491	/*
492	 * while there is data to copy copy into the write buffer. when the
493	 * write buffer fills, flush it to the archive and continue
494	 */
495	while (outcnt > 0) {
496		cnt = bufend - bufpt;
497		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
498			return(-1);
499		/*
500		 * only move what we have space for
501		 */
502		cnt = MIN(cnt, outcnt);
503		memcpy(bufpt, out, cnt);
504		bufpt += cnt;
505		out += cnt;
506		outcnt -= cnt;
507	}
508	return(0);
509}
510
511/*
512 * rd_wrbuf()
513 *	copy from the read buffer into a supplied buffer a specified number of
514 *	bytes. If the read buffer is empty fill it and continue to copy.
515 *	usually used to obtain a file header for processing by a format
516 *	specific read routine.
517 * Return
518 *	number of bytes copied to the buffer, 0 indicates EOF on archive volume,
519 *	-1 is a read error
520 */
521
522int
523rd_wrbuf(char *in, int cpcnt)
524{
525	int res;
526	int cnt;
527	int incnt = cpcnt;
528
529	/*
530	 * loop until we fill the buffer with the requested number of bytes
531	 */
532	while (incnt > 0) {
533		cnt = bufend - bufpt;
534		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
535			/*
536			 * read error, return what we got (or the error if
537			 * no data was copied). The caller must know that an
538			 * error occurred and has the best knowledge what to
539			 * do with it
540			 */
541			if ((res = cpcnt - incnt) > 0)
542				return(res);
543			return(cnt);
544		}
545
546		/*
547		 * calculate how much data to copy based on whats left and
548		 * state of buffer
549		 */
550		cnt = MIN(cnt, incnt);
551		memcpy(in, bufpt, cnt);
552		bufpt += cnt;
553		incnt -= cnt;
554		in += cnt;
555	}
556	return(cpcnt);
557}
558
559/*
560 * wr_skip()
561 *	skip forward during a write. In other words add padding to the file.
562 *	we add zero filled padding as it makes flawed archives much easier to
563 *	recover from. the caller tells us how many bytes of padding to add
564 *	This routine was not designed to add HUGE amount of padding, just small
565 *	amounts (a few 512 byte blocks at most)
566 * Return:
567 *	0 if ok, -1 if there was a buf_flush failure
568 */
569
570int
571wr_skip(off_t skcnt)
572{
573	int cnt;
574
575	/*
576	 * loop while there is more padding to add
577	 */
578	while (skcnt > 0L) {
579		cnt = bufend - bufpt;
580		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
581			return(-1);
582		cnt = MIN(cnt, skcnt);
583		memset(bufpt, 0, cnt);
584		bufpt += cnt;
585		skcnt -= cnt;
586	}
587	return(0);
588}
589
590/*
591 * wr_rdfile()
592 *	fill write buffer with the contents of a file. We are passed an	open
593 *	file descriptor to the file an the archive structure that describes the
594 *	file we are storing. The variable "left" is modified to contain the
595 *	number of bytes of the file we were NOT able to write to the archive.
596 *	it is important that we always write EXACTLY the number of bytes that
597 *	the format specific write routine told us to. The file can also get
598 *	bigger, so reading to the end of file would create an improper archive,
599 *	we just detect this case and warn the user. We never create a bad
600 *	archive if we can avoid it. Of course trying to archive files that are
601 *	active is asking for trouble. It we fail, we pass back how much we
602 *	could NOT copy and let the caller deal with it.
603 * Return:
604 *	0 ok, -1 if archive write failure. a short read of the file returns a
605 *	0, but "left" is set to be greater than zero.
606 */
607
608int
609wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
610{
611	int cnt;
612	int res = 0;
613	off_t size = arcn->sb.st_size;
614	struct stat origsb, sb;
615
616	/*
617	 * by default, remember the previously obtained stat information
618	 * (in arcn->sb) for comparing the mtime after reading.
619	 * if Mflag is set, use the actual mtime instead.
620	 */
621	origsb = arcn->sb;
622	if (Mflag && (fstat(ifd, &origsb) < 0))
623		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
624
625	/*
626	 * while there are more bytes to write
627	 */
628	while (size > 0L) {
629		cnt = bufend - bufpt;
630		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
631			*left = size;
632			return(-1);
633		}
634		cnt = MIN(cnt, size);
635		if ((res = read_with_restart(ifd, bufpt, cnt)) <= 0)
636			break;
637		size -= res;
638		bufpt += res;
639	}
640
641	/*
642	 * better check the file did not change during this operation
643	 * or the file read failed.
644	 */
645	if (res < 0)
646		syswarn(1, errno, "Read fault on %s", arcn->org_name);
647	else if (size != 0L)
648		tty_warn(1, "File changed size during read %s", arcn->org_name);
649	else if (fstat(ifd, &sb) < 0)
650		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
651	else if (origsb.st_mtime != sb.st_mtime)
652		tty_warn(1, "File %s was modified during copy to archive",
653			arcn->org_name);
654	*left = size;
655	return(0);
656}
657
658/*
659 * rd_wrfile()
660 *	extract the contents of a file from the archive. If we are unable to
661 *	extract the entire file (due to failure to write the file) we return
662 *	the numbers of bytes we did NOT process. This way the caller knows how
663 *	many bytes to skip past to find the next archive header. If the failure
664 *	was due to an archive read, we will catch that when we try to skip. If
665 *	the format supplies a file data crc value, we calculate the actual crc
666 *	so that it can be compared to the value stored in the header
667 * NOTE:
668 *	We call a special function to write the file. This function attempts to
669 *	restore file holes (blocks of zeros) into the file. When files are
670 *	sparse this saves space, and is a LOT faster. For non sparse files
671 *	the performance hit is small. As of this writing, no archive supports
672 *	information on where the file holes are.
673 * Return:
674 *	0 ok, -1 if archive read failure. if we cannot write the entire file,
675 *	we return a 0 but "left" is set to be the amount unwritten
676 */
677
678int
679rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
680{
681	int cnt = 0;
682	off_t size = arcn->sb.st_size;
683	int res = 0;
684	char *fnm = arcn->name;
685	int isem = 1;
686	int rem;
687	int sz = MINFBSZ;
688	struct stat sb;
689	u_long crc = 0L;
690
691	/*
692	 * pass the blocksize of the file being written to the write routine,
693	 * if the size is zero, use the default MINFBSZ
694	 */
695	if (ofd == -1)
696		sz = PAXPATHLEN+1;
697	else if (fstat(ofd, &sb) == 0) {
698		if (sb.st_blksize > 0)
699			sz = (int)sb.st_blksize;
700	} else
701		syswarn(0, errno,
702		    "Unable to obtain block size for file %s", fnm);
703	rem = sz;
704	*left = 0L;
705
706	/*
707	 * Copy the archive to the file the number of bytes specified. We have
708	 * to assume that we want to recover file holes as none of the archive
709	 * formats can record the location of file holes.
710	 */
711	while (size > 0L) {
712		cnt = bufend - bufpt;
713		/*
714		 * if we get a read error, we do not want to skip, as we may
715		 * miss a header, so we do not set left, but if we get a write
716		 * error, we do want to skip over the unprocessed data.
717		 */
718		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
719			break;
720		cnt = MIN(cnt, size);
721		if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
722			*left = size;
723			break;
724		}
725
726		if (docrc) {
727			/*
728			 * update the actual crc value
729			 */
730			cnt = res;
731			while (--cnt >= 0)
732				crc += *bufpt++ & 0xff;
733		} else
734			bufpt += res;
735		size -= res;
736	}
737
738	/*
739	 * if the last block has a file hole (all zero), we must make sure this
740	 * gets updated in the file. We force the last block of zeros to be
741	 * written. just closing with the file offset moved forward may not put
742	 * a hole at the end of the file.
743	 */
744	if (ofd != -1 && isem && (arcn->sb.st_size > 0L))
745		file_flush(ofd, fnm, isem);
746
747	/*
748	 * if we failed from archive read, we do not want to skip
749	 */
750	if ((size > 0L) && (*left == 0L))
751		return(-1);
752
753	/*
754	 * some formats record a crc on file data. If so, then we compare the
755	 * calculated crc to the crc stored in the archive
756	 */
757	if (docrc && (size == 0L) && (arcn->crc != crc))
758		tty_warn(1,"Actual crc does not match expected crc %s",
759		    arcn->name);
760	return(0);
761}
762
763/*
764 * cp_file()
765 *	copy the contents of one file to another. used during -rw phase of pax
766 *	just as in rd_wrfile() we use a special write function to write the
767 *	destination file so we can properly copy files with holes.
768 */
769
770void
771cp_file(ARCHD *arcn, int fd1, int fd2)
772{
773	int cnt;
774	off_t cpcnt = 0L;
775	int res = 0;
776	char *fnm = arcn->name;
777	int no_hole = 0;
778	int isem = 1;
779	int rem;
780	int sz = MINFBSZ;
781	struct stat sb, origsb;
782
783	/*
784	 * check for holes in the source file. If none, we will use regular
785	 * write instead of file write.
786	 */
787	 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
788		++no_hole;
789
790	/*
791	 * by default, remember the previously obtained stat information
792	 * (in arcn->sb) for comparing the mtime after reading.
793	 * if Mflag is set, use the actual mtime instead.
794	 */
795	origsb = arcn->sb;
796	if (Mflag && (fstat(fd1, &origsb) < 0))
797		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
798
799	/*
800	 * pass the blocksize of the file being written to the write routine,
801	 * if the size is zero, use the default MINFBSZ
802	 */
803	if (fstat(fd2, &sb) == 0) {
804		if (sb.st_blksize > 0)
805			sz = sb.st_blksize;
806	} else
807		syswarn(0, errno,
808		    "Unable to obtain block size for file %s", fnm);
809	rem = sz;
810
811	/*
812	 * read the source file and copy to destination file until EOF
813	 */
814	for(;;) {
815		if ((cnt = read_with_restart(fd1, buf, blksz)) <= 0)
816			break;
817		if (no_hole)
818			res = xwrite(fd2, buf, cnt);
819		else
820			res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
821		if (res != cnt)
822			break;
823		cpcnt += cnt;
824	}
825
826	/*
827	 * check to make sure the copy is valid.
828	 */
829	if (res < 0)
830		syswarn(1, errno, "Failed write during copy of %s to %s",
831			arcn->org_name, arcn->name);
832	else if (cpcnt != arcn->sb.st_size)
833		tty_warn(1, "File %s changed size during copy to %s",
834			arcn->org_name, arcn->name);
835	else if (fstat(fd1, &sb) < 0)
836		syswarn(1, errno, "Failed stat of %s", arcn->org_name);
837	else if (origsb.st_mtime != sb.st_mtime)
838		tty_warn(1, "File %s was modified during copy to %s",
839			arcn->org_name, arcn->name);
840
841	/*
842	 * if the last block has a file hole (all zero), we must make sure this
843	 * gets updated in the file. We force the last block of zeros to be
844	 * written. just closing with the file offset moved forward may not put
845	 * a hole at the end of the file.
846	 */
847	if (!no_hole && isem && (arcn->sb.st_size > 0L))
848		file_flush(fd2, fnm, isem);
849	return;
850}
851
852/*
853 * buf_fill()
854 *	fill the read buffer with the next record (or what we can get) from
855 *	the archive volume.
856 * Return:
857 *	Number of bytes of data in the read buffer, -1 for read error, and
858 *	0 when finished (user specified termination in ar_next()).
859 */
860
861int
862buf_fill(void)
863{
864	int cnt;
865	static int fini = 0;
866
867	if (fini)
868		return(0);
869
870	for(;;) {
871		/*
872		 * try to fill the buffer. on error the next archive volume is
873		 * opened and we try again.
874		 */
875		if ((cnt = ar_read(buf, blksz)) > 0) {
876			bufpt = buf;
877			bufend = buf + cnt;
878			rdcnt += cnt;
879			return(cnt);
880		}
881
882		/*
883		 * errors require resync, EOF goes to next archive
884		 */
885		if (cnt < 0)
886			break;
887		if (ar_next() < 0) {
888			fini = 1;
889			return(0);
890		}
891		rdcnt = 0;
892	}
893	exit_val = 1;
894	return(-1);
895}
896
897/*
898 * buf_flush()
899 *	force the write buffer to the archive. We are passed the number of
900 *	bytes in the buffer at the point of the flush. When we change archives
901 *	the record size might change. (either larger or smaller).
902 * Return:
903 *	0 if all is ok, -1 when a write error occurs.
904 */
905
906int
907buf_flush(int bufcnt)
908{
909	int cnt;
910	int push = 0;
911	int totcnt = 0;
912
913	/*
914	 * if we have reached the user specified byte count for each archive
915	 * volume, prompt for the next volume. (The non-standard -R flag).
916	 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
917	 * at least one record. We always round limit UP to next blocksize.
918	 */
919	if ((wrlimit > 0) && (wrcnt > wrlimit)) {
920		tty_warn(0,
921		    "User specified archive volume byte limit reached.");
922		if (ar_next() < 0) {
923			wrcnt = 0;
924			exit_val = 1;
925			return(-1);
926		}
927		wrcnt = 0;
928
929		/*
930		 * The new archive volume might have changed the size of the
931		 * write blocksize. if so we figure out if we need to write
932		 * (one or more times), or if there is now free space left in
933		 * the buffer (it is no longer full). bufcnt has the number of
934		 * bytes in the buffer, (the blocksize, at the point we were
935		 * CALLED). Push has the amount of "extra" data in the buffer
936		 * if the block size has shrunk from a volume change.
937		 */
938		bufend = buf + blksz;
939		if (blksz > bufcnt)
940			return(0);
941		if (blksz < bufcnt)
942			push = bufcnt - blksz;
943	}
944
945	/*
946	 * We have enough data to write at least one archive block
947	 */
948	for (;;) {
949		/*
950		 * write a block and check if it all went out ok
951		 */
952		cnt = ar_write(buf, blksz);
953		if (cnt == blksz) {
954			/*
955			 * the write went ok
956			 */
957			wrcnt += cnt;
958			totcnt += cnt;
959			if (push > 0) {
960				/* we have extra data to push to the front.
961				 * check for more than 1 block of push, and if
962				 * so we loop back to write again
963				 */
964				memcpy(buf, bufend, push);
965				bufpt = buf + push;
966				if (push >= blksz) {
967					push -= blksz;
968					continue;
969				}
970			} else
971				bufpt = buf;
972			return(totcnt);
973		} else if (cnt > 0) {
974			/*
975			 * Oh drat we got a partial write!
976			 * if format doesnt care about alignment let it go,
977			 * we warned the user in ar_write().... but this means
978			 * the last record on this volume violates pax spec....
979			 */
980			totcnt += cnt;
981			wrcnt += cnt;
982			bufpt = buf + cnt;
983			cnt = bufcnt - cnt;
984			memcpy(buf, bufpt, cnt);
985			bufpt = buf + cnt;
986			if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
987				return(totcnt);
988			break;
989		}
990
991		/*
992		 * All done, go to next archive
993		 */
994		wrcnt = 0;
995		if (ar_next() < 0)
996			break;
997
998		/*
999		 * The new archive volume might also have changed the block
1000		 * size. if so, figure out if we have too much or too little
1001		 * data for using the new block size
1002		 */
1003		bufend = buf + blksz;
1004		if (blksz > bufcnt)
1005			return(0);
1006		if (blksz < bufcnt)
1007			push = bufcnt - blksz;
1008	}
1009
1010	/*
1011	 * write failed, stop pax. we must not create a bad archive!
1012	 */
1013	exit_val = 1;
1014	return(-1);
1015}
1016