xref: /freebsd/bin/pax/buf_subs.c (revision d6b92ffa)
1 /*-
2  * Copyright (c) 1992 Keith Muller.
3  * Copyright (c) 1992, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * Keith Muller of the University of California, San Diego.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 #ifndef lint
35 #if 0
36 static char sccsid[] = "@(#)buf_subs.c	8.2 (Berkeley) 4/18/94";
37 #endif
38 #endif /* not lint */
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include <sys/types.h>
43 #include <sys/stat.h>
44 #include <errno.h>
45 #include <unistd.h>
46 #include <stdio.h>
47 #include <string.h>
48 #include "pax.h"
49 #include "extern.h"
50 
51 /*
52  * routines which implement archive and file buffering
53  */
54 
55 #define MINFBSZ		512		/* default block size for hole detect */
56 #define MAXFLT		10		/* default media read error limit */
57 
58 /*
59  * Need to change bufmem to dynamic allocation when the upper
60  * limit on blocking size is removed (though that will violate pax spec)
61  * MAXBLK define and tests will also need to be updated.
62  */
63 static char bufmem[MAXBLK+BLKMULT];	/* i/o buffer + pushback id space */
64 static char *buf;			/* normal start of i/o buffer */
65 static char *bufend;			/* end or last char in i/o buffer */
66 static char *bufpt;			/* read/write point in i/o buffer */
67 int blksz = MAXBLK;			/* block input/output size in bytes */
68 int wrblksz;				/* user spec output size in bytes */
69 int maxflt = MAXFLT;			/* MAX consecutive media errors */
70 int rdblksz;				/* first read blksize (tapes only) */
71 off_t wrlimit;				/* # of bytes written per archive vol */
72 off_t wrcnt;				/* # of bytes written on current vol */
73 off_t rdcnt;				/* # of bytes read on current vol */
74 
75 /*
76  * wr_start()
77  *	set up the buffering system to operate in a write mode
78  * Return:
79  *	0 if ok, -1 if the user specified write block size violates pax spec
80  */
81 
82 int
83 wr_start(void)
84 {
85 	buf = &(bufmem[BLKMULT]);
86 	/*
87 	 * Check to make sure the write block size meets pax specs. If the user
88 	 * does not specify a blocksize, we use the format default blocksize.
89 	 * We must be picky on writes, so we do not allow the user to create an
90 	 * archive that might be hard to read elsewhere. If all ok, we then
91 	 * open the first archive volume
92 	 */
93 	if (!wrblksz)
94 		wrblksz = frmt->bsz;
95 	if (wrblksz > MAXBLK) {
96 		paxwarn(1, "Write block size of %d too large, maximum is: %d",
97 			wrblksz, MAXBLK);
98 		return(-1);
99 	}
100 	if (wrblksz % BLKMULT) {
101 		paxwarn(1, "Write block size of %d is not a %d byte multiple",
102 		    wrblksz, BLKMULT);
103 		return(-1);
104 	}
105 	if (wrblksz > MAXBLK_POSIX) {
106 		paxwarn(0, "Write block size of %d larger than POSIX max %d, archive may not be portable",
107 			wrblksz, MAXBLK_POSIX);
108 		return(-1);
109 	}
110 
111 	/*
112 	 * we only allow wrblksz to be used with all archive operations
113 	 */
114 	blksz = rdblksz = wrblksz;
115 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
116 		return(-1);
117 	wrcnt = 0;
118 	bufend = buf + wrblksz;
119 	bufpt = buf;
120 	return(0);
121 }
122 
123 /*
124  * rd_start()
125  *	set up buffering system to read an archive
126  * Return:
127  *	0 if ok, -1 otherwise
128  */
129 
130 int
131 rd_start(void)
132 {
133 	/*
134 	 * leave space for the header pushback (see get_arc()). If we are
135 	 * going to append and user specified a write block size, check it
136 	 * right away
137 	 */
138 	buf = &(bufmem[BLKMULT]);
139 	if ((act == APPND) && wrblksz) {
140 		if (wrblksz > MAXBLK) {
141 			paxwarn(1,"Write block size %d too large, maximum is: %d",
142 				wrblksz, MAXBLK);
143 			return(-1);
144 		}
145 		if (wrblksz % BLKMULT) {
146 			paxwarn(1, "Write block size %d is not a %d byte multiple",
147 			wrblksz, BLKMULT);
148 			return(-1);
149 		}
150 	}
151 
152 	/*
153 	 * open the archive
154 	 */
155 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
156 		return(-1);
157 	bufend = buf + rdblksz;
158 	bufpt = bufend;
159 	rdcnt = 0;
160 	return(0);
161 }
162 
163 /*
164  * cp_start()
165  *	set up buffer system for copying within the file system
166  */
167 
168 void
169 cp_start(void)
170 {
171 	buf = &(bufmem[BLKMULT]);
172 	rdblksz = blksz = MAXBLK;
173 }
174 
175 /*
176  * appnd_start()
177  *	Set up the buffering system to append new members to an archive that
178  *	was just read. The last block(s) of an archive may contain a format
179  *	specific trailer. To append a new member, this trailer has to be
180  *	removed from the archive. The first byte of the trailer is replaced by
181  *	the start of the header of the first file added to the archive. The
182  *	format specific end read function tells us how many bytes to move
183  *	backwards in the archive to be positioned BEFORE the trailer. Two
184  *	different positions have to be adjusted, the O.S. file offset (e.g. the
185  *	position of the tape head) and the write point within the data we have
186  *	stored in the read (soon to become write) buffer. We may have to move
187  *	back several records (the number depends on the size of the archive
188  *	record and the size of the format trailer) to read up the record where
189  *	the first byte of the trailer is recorded. Trailers may span (and
190  *	overlap) record boundaries.
191  *	We first calculate which record has the first byte of the trailer. We
192  *	move the OS file offset back to the start of this record and read it
193  *	up. We set the buffer write pointer to be at this byte (the byte where
194  *	the trailer starts). We then move the OS file pointer back to the
195  *	start of this record so a flush of this buffer will replace the record
196  *	in the archive.
197  *	A major problem is rewriting this last record. For archives stored
198  *	on disk files, this is trivial. However, many devices are really picky
199  *	about the conditions under which they will allow a write to occur.
200  *	Often devices restrict the conditions where writes can be made writes,
201  *	so it may not be feasible to append archives stored on all types of
202  *	devices.
203  * Return:
204  *	0 for success, -1 for failure
205  */
206 
207 int
208 appnd_start(off_t skcnt)
209 {
210 	int res;
211 	off_t cnt;
212 
213 	if (exit_val != 0) {
214 		paxwarn(0, "Cannot append to an archive that may have flaws.");
215 		return(-1);
216 	}
217 	/*
218 	 * if the user did not specify a write blocksize, inherit the size used
219 	 * in the last archive volume read. (If a is set we still use rdblksz
220 	 * until next volume, cannot shift sizes within a single volume).
221 	 */
222 	if (!wrblksz)
223 		wrblksz = blksz = rdblksz;
224 	else
225 		blksz = rdblksz;
226 
227 	/*
228 	 * make sure that this volume allows appends
229 	 */
230 	if (ar_app_ok() < 0)
231 		return(-1);
232 
233 	/*
234 	 * Calculate bytes to move back and move in front of record where we
235 	 * need to start writing from. Remember we have to add in any padding
236 	 * that might be in the buffer after the trailer in the last block. We
237 	 * travel skcnt + padding ROUNDED UP to blksize.
238 	 */
239 	skcnt += bufend - bufpt;
240 	if ((cnt = (skcnt/blksz) * blksz) < skcnt)
241 		cnt += blksz;
242 	if (ar_rev((off_t)cnt) < 0)
243 		goto out;
244 
245 	/*
246 	 * We may have gone too far if there is valid data in the block we are
247 	 * now in front of, read up the block and position the pointer after
248 	 * the valid data.
249 	 */
250 	if ((cnt -= skcnt) > 0) {
251 		/*
252 		 * watch out for stupid tape drives. ar_rev() will set rdblksz
253 		 * to be real physical blocksize so we must loop until we get
254 		 * the old rdblksz (now in blksz). If ar_rev() fouls up the
255 		 * determination of the physical block size, we will fail.
256 		 */
257 		bufpt = buf;
258 		bufend = buf + blksz;
259 		while (bufpt < bufend) {
260 			if ((res = ar_read(bufpt, rdblksz)) <= 0)
261 				goto out;
262 			bufpt += res;
263 		}
264 		if (ar_rev((off_t)(bufpt - buf)) < 0)
265 			goto out;
266 		bufpt = buf + cnt;
267 		bufend = buf + blksz;
268 	} else {
269 		/*
270 		 * buffer is empty
271 		 */
272 		bufend = buf + blksz;
273 		bufpt = buf;
274 	}
275 	rdblksz = blksz;
276 	rdcnt -= skcnt;
277 	wrcnt = 0;
278 
279 	/*
280 	 * At this point we are ready to write. If the device requires special
281 	 * handling to write at a point were previously recorded data resides,
282 	 * that is handled in ar_set_wr(). From now on we operate under normal
283 	 * ARCHIVE mode (write) conditions
284 	 */
285 	if (ar_set_wr() < 0)
286 		return(-1);
287 	act = ARCHIVE;
288 	return(0);
289 
290     out:
291 	paxwarn(1, "Unable to rewrite archive trailer, cannot append.");
292 	return(-1);
293 }
294 
295 /*
296  * rd_sync()
297  *	A read error occurred on this archive volume. Resync the buffer and
298  *	try to reset the device (if possible) so we can continue to read. Keep
299  *	trying to do this until we get a valid read, or we reach the limit on
300  *	consecutive read faults (at which point we give up). The user can
301  *	adjust the read error limit through a command line option.
302  * Returns:
303  *	0 on success, and -1 on failure
304  */
305 
306 int
307 rd_sync(void)
308 {
309 	int errcnt = 0;
310 	int res;
311 
312 	/*
313 	 * if the user says bail out on first fault, we are out of here...
314 	 */
315 	if (maxflt == 0)
316 		return(-1);
317 	if (act == APPND) {
318 		paxwarn(1, "Unable to append when there are archive read errors.");
319 		return(-1);
320 	}
321 
322 	/*
323 	 * poke at device and try to get past media error
324 	 */
325 	if (ar_rdsync() < 0) {
326 		if (ar_next() < 0)
327 			return(-1);
328 		else
329 			rdcnt = 0;
330 	}
331 
332 	for (;;) {
333 		if ((res = ar_read(buf, blksz)) > 0) {
334 			/*
335 			 * All right! got some data, fill that buffer
336 			 */
337 			bufpt = buf;
338 			bufend = buf + res;
339 			rdcnt += res;
340 			return(0);
341 		}
342 
343 		/*
344 		 * Oh well, yet another failed read...
345 		 * if error limit reached, ditch. o.w. poke device to move past
346 		 * bad media and try again. if media is badly damaged, we ask
347 		 * the poor (and upset user at this point) for the next archive
348 		 * volume. remember the goal on reads is to get the most we
349 		 * can extract out of the archive.
350 		 */
351 		if ((maxflt > 0) && (++errcnt > maxflt))
352 			paxwarn(0,"Archive read error limit (%d) reached",maxflt);
353 		else if (ar_rdsync() == 0)
354 			continue;
355 		if (ar_next() < 0)
356 			break;
357 		rdcnt = 0;
358 		errcnt = 0;
359 	}
360 	return(-1);
361 }
362 
363 /*
364  * pback()
365  *	push the data used during the archive id phase back into the I/O
366  *	buffer. This is required as we cannot be sure that the header does NOT
367  *	overlap a block boundary (as in the case we are trying to recover a
368  *	flawed archived). This was not designed to be used for any other
369  *	purpose. (What software engineering, HA!)
370  *	WARNING: do not even THINK of pback greater than BLKMULT, unless the
371  *	pback space is increased.
372  */
373 
374 void
375 pback(char *pt, int cnt)
376 {
377 	bufpt -= cnt;
378 	memcpy(bufpt, pt, cnt);
379 	return;
380 }
381 
382 /*
383  * rd_skip()
384  *	skip forward in the archive during an archive read. Used to get quickly
385  *	past file data and padding for files the user did NOT select.
386  * Return:
387  *	0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
388  */
389 
390 int
391 rd_skip(off_t skcnt)
392 {
393 	off_t res;
394 	off_t cnt;
395 	off_t skipped = 0;
396 
397 	/*
398 	 * consume what data we have in the buffer. If we have to move forward
399 	 * whole records, we call the low level skip function to see if we can
400 	 * move within the archive without doing the expensive reads on data we
401 	 * do not want.
402 	 */
403 	if (skcnt == 0)
404 		return(0);
405 	res = MIN((bufend - bufpt), skcnt);
406 	bufpt += res;
407 	skcnt -= res;
408 
409 	/*
410 	 * if skcnt is now 0, then no additional i/o is needed
411 	 */
412 	if (skcnt == 0)
413 		return(0);
414 
415 	/*
416 	 * We have to read more, calculate complete and partial record reads
417 	 * based on rdblksz. we skip over "cnt" complete records
418 	 */
419 	res = skcnt%rdblksz;
420 	cnt = (skcnt/rdblksz) * rdblksz;
421 
422 	/*
423 	 * if the skip fails, we will have to resync. ar_fow will tell us
424 	 * how much it can skip over. We will have to read the rest.
425 	 */
426 	if (ar_fow(cnt, &skipped) < 0)
427 		return(-1);
428 	res += cnt - skipped;
429 	rdcnt += skipped;
430 
431 	/*
432 	 * what is left we have to read (which may be the whole thing if
433 	 * ar_fow() told us the device can only read to skip records);
434 	 */
435 	while (res > 0L) {
436 		cnt = bufend - bufpt;
437 		/*
438 		 * if the read fails, we will have to resync
439 		 */
440 		if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
441 			return(-1);
442 		if (cnt == 0)
443 			return(1);
444 		cnt = MIN(cnt, res);
445 		bufpt += cnt;
446 		res -= cnt;
447 	}
448 	return(0);
449 }
450 
451 /*
452  * wr_fin()
453  *	flush out any data (and pad if required) the last block. We always pad
454  *	with zero (even though we do not have to). Padding with 0 makes it a
455  *	lot easier to recover if the archive is damaged. zero padding SHOULD
456  *	BE a requirement....
457  */
458 
459 void
460 wr_fin(void)
461 {
462 	if (bufpt > buf) {
463 		memset(bufpt, 0, bufend - bufpt);
464 		bufpt = bufend;
465 		(void)buf_flush(blksz);
466 	}
467 }
468 
469 /*
470  * wr_rdbuf()
471  *	fill the write buffer from data passed to it in a buffer (usually used
472  *	by format specific write routines to pass a file header). On failure we
473  *	punt. We do not allow the user to continue to write flawed archives.
474  *	We assume these headers are not very large (the memory copy we use is
475  *	a bit expensive).
476  * Return:
477  *	0 if buffer was filled ok, -1 o.w. (buffer flush failure)
478  */
479 
480 int
481 wr_rdbuf(char *out, int outcnt)
482 {
483 	int cnt;
484 
485 	/*
486 	 * while there is data to copy copy into the write buffer. when the
487 	 * write buffer fills, flush it to the archive and continue
488 	 */
489 	while (outcnt > 0) {
490 		cnt = bufend - bufpt;
491 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
492 			return(-1);
493 		/*
494 		 * only move what we have space for
495 		 */
496 		cnt = MIN(cnt, outcnt);
497 		memcpy(bufpt, out, cnt);
498 		bufpt += cnt;
499 		out += cnt;
500 		outcnt -= cnt;
501 	}
502 	return(0);
503 }
504 
505 /*
506  * rd_wrbuf()
507  *	copy from the read buffer into a supplied buffer a specified number of
508  *	bytes. If the read buffer is empty fill it and continue to copy.
509  *	usually used to obtain a file header for processing by a format
510  *	specific read routine.
511  * Return
512  *	number of bytes copied to the buffer, 0 indicates EOF on archive volume,
513  *	-1 is a read error
514  */
515 
516 int
517 rd_wrbuf(char *in, int cpcnt)
518 {
519 	int res;
520 	int cnt;
521 	int incnt = cpcnt;
522 
523 	/*
524 	 * loop until we fill the buffer with the requested number of bytes
525 	 */
526 	while (incnt > 0) {
527 		cnt = bufend - bufpt;
528 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
529 			/*
530 			 * read error, return what we got (or the error if
531 			 * no data was copied). The caller must know that an
532 			 * error occurred and has the best knowledge what to
533 			 * do with it
534 			 */
535 			if ((res = cpcnt - incnt) > 0)
536 				return(res);
537 			return(cnt);
538 		}
539 
540 		/*
541 		 * calculate how much data to copy based on whats left and
542 		 * state of buffer
543 		 */
544 		cnt = MIN(cnt, incnt);
545 		memcpy(in, bufpt, cnt);
546 		bufpt += cnt;
547 		incnt -= cnt;
548 		in += cnt;
549 	}
550 	return(cpcnt);
551 }
552 
553 /*
554  * wr_skip()
555  *	skip forward during a write. In other words add padding to the file.
556  *	we add zero filled padding as it makes flawed archives much easier to
557  *	recover from. the caller tells us how many bytes of padding to add
558  *	This routine was not designed to add HUGE amount of padding, just small
559  *	amounts (a few 512 byte blocks at most)
560  * Return:
561  *	0 if ok, -1 if there was a buf_flush failure
562  */
563 
564 int
565 wr_skip(off_t skcnt)
566 {
567 	int cnt;
568 
569 	/*
570 	 * loop while there is more padding to add
571 	 */
572 	while (skcnt > 0L) {
573 		cnt = bufend - bufpt;
574 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
575 			return(-1);
576 		cnt = MIN(cnt, skcnt);
577 		memset(bufpt, 0, cnt);
578 		bufpt += cnt;
579 		skcnt -= cnt;
580 	}
581 	return(0);
582 }
583 
584 /*
585  * wr_rdfile()
586  *	fill write buffer with the contents of a file. We are passed an open
587  *	file descriptor to the file and the archive structure that describes the
588  *	file we are storing. The variable "left" is modified to contain the
589  *	number of bytes of the file we were NOT able to write to the archive.
590  *	it is important that we always write EXACTLY the number of bytes that
591  *	the format specific write routine told us to. The file can also get
592  *	bigger, so reading to the end of file would create an improper archive,
593  *	we just detect this case and warn the user. We never create a bad
594  *	archive if we can avoid it. Of course trying to archive files that are
595  *	active is asking for trouble. It we fail, we pass back how much we
596  *	could NOT copy and let the caller deal with it.
597  * Return:
598  *	0 ok, -1 if archive write failure. a short read of the file returns a
599  *	0, but "left" is set to be greater than zero.
600  */
601 
602 int
603 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
604 {
605 	int cnt;
606 	int res = 0;
607 	off_t size = arcn->sb.st_size;
608 	struct stat sb;
609 
610 	/*
611 	 * while there are more bytes to write
612 	 */
613 	while (size > 0L) {
614 		cnt = bufend - bufpt;
615 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
616 			*left = size;
617 			return(-1);
618 		}
619 		cnt = MIN(cnt, size);
620 		if ((res = read(ifd, bufpt, cnt)) <= 0)
621 			break;
622 		size -= res;
623 		bufpt += res;
624 	}
625 
626 	/*
627 	 * better check the file did not change during this operation
628 	 * or the file read failed.
629 	 */
630 	if (res < 0)
631 		syswarn(1, errno, "Read fault on %s", arcn->org_name);
632 	else if (size != 0L)
633 		paxwarn(1, "File changed size during read %s", arcn->org_name);
634 	else if (fstat(ifd, &sb) < 0)
635 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
636 	else if (arcn->sb.st_mtime != sb.st_mtime)
637 		paxwarn(1, "File %s was modified during copy to archive",
638 			arcn->org_name);
639 	*left = size;
640 	return(0);
641 }
642 
643 /*
644  * rd_wrfile()
645  *	extract the contents of a file from the archive. If we are unable to
646  *	extract the entire file (due to failure to write the file) we return
647  *	the numbers of bytes we did NOT process. This way the caller knows how
648  *	many bytes to skip past to find the next archive header. If the failure
649  *	was due to an archive read, we will catch that when we try to skip. If
650  *	the format supplies a file data crc value, we calculate the actual crc
651  *	so that it can be compared to the value stored in the header
652  * NOTE:
653  *	We call a special function to write the file. This function attempts to
654  *	restore file holes (blocks of zeros) into the file. When files are
655  *	sparse this saves space, and is a LOT faster. For non sparse files
656  *	the performance hit is small. As of this writing, no archive supports
657  *	information on where the file holes are.
658  * Return:
659  *	0 ok, -1 if archive read failure. if we cannot write the entire file,
660  *	we return a 0 but "left" is set to be the amount unwritten
661  */
662 
663 int
664 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
665 {
666 	int cnt = 0;
667 	off_t size = arcn->sb.st_size;
668 	int res = 0;
669 	char *fnm = arcn->name;
670 	int isem = 1;
671 	int rem;
672 	int sz = MINFBSZ;
673 	struct stat sb;
674 	u_long crc = 0L;
675 
676 	/*
677 	 * pass the blocksize of the file being written to the write routine,
678 	 * if the size is zero, use the default MINFBSZ
679 	 */
680 	if (fstat(ofd, &sb) == 0) {
681 		if (sb.st_blksize > 0)
682 			sz = (int)sb.st_blksize;
683 	} else
684 		syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
685 	rem = sz;
686 	*left = 0L;
687 
688 	/*
689 	 * Copy the archive to the file the number of bytes specified. We have
690 	 * to assume that we want to recover file holes as none of the archive
691 	 * formats can record the location of file holes.
692 	 */
693 	while (size > 0L) {
694 		cnt = bufend - bufpt;
695 		/*
696 		 * if we get a read error, we do not want to skip, as we may
697 		 * miss a header, so we do not set left, but if we get a write
698 		 * error, we do want to skip over the unprocessed data.
699 		 */
700 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
701 			break;
702 		cnt = MIN(cnt, size);
703 		if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
704 			*left = size;
705 			break;
706 		}
707 
708 		if (docrc) {
709 			/*
710 			 * update the actual crc value
711 			 */
712 			cnt = res;
713 			while (--cnt >= 0)
714 				crc += *bufpt++ & 0xff;
715 		} else
716 			bufpt += res;
717 		size -= res;
718 	}
719 
720 	/*
721 	 * if the last block has a file hole (all zero), we must make sure this
722 	 * gets updated in the file. We force the last block of zeros to be
723 	 * written. just closing with the file offset moved forward may not put
724 	 * a hole at the end of the file.
725 	 */
726 	if (isem && (arcn->sb.st_size > 0L))
727 		file_flush(ofd, fnm, isem);
728 
729 	/*
730 	 * if we failed from archive read, we do not want to skip
731 	 */
732 	if ((size > 0L) && (*left == 0L))
733 		return(-1);
734 
735 	/*
736 	 * some formats record a crc on file data. If so, then we compare the
737 	 * calculated crc to the crc stored in the archive
738 	 */
739 	if (docrc && (size == 0L) && (arcn->crc != crc))
740 		paxwarn(1,"Actual crc does not match expected crc %s",arcn->name);
741 	return(0);
742 }
743 
744 /*
745  * cp_file()
746  *	copy the contents of one file to another. used during -rw phase of pax
747  *	just as in rd_wrfile() we use a special write function to write the
748  *	destination file so we can properly copy files with holes.
749  */
750 
751 void
752 cp_file(ARCHD *arcn, int fd1, int fd2)
753 {
754 	int cnt;
755 	off_t cpcnt = 0L;
756 	int res = 0;
757 	char *fnm = arcn->name;
758 	int no_hole = 0;
759 	int isem = 1;
760 	int rem;
761 	int sz = MINFBSZ;
762 	struct stat sb;
763 
764 	/*
765 	 * check for holes in the source file. If none, we will use regular
766 	 * write instead of file write.
767 	 */
768 	 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
769 		++no_hole;
770 
771 	/*
772 	 * pass the blocksize of the file being written to the write routine,
773 	 * if the size is zero, use the default MINFBSZ
774 	 */
775 	if (fstat(fd2, &sb) == 0) {
776 		if (sb.st_blksize > 0)
777 			sz = sb.st_blksize;
778 	} else
779 		syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
780 	rem = sz;
781 
782 	/*
783 	 * read the source file and copy to destination file until EOF
784 	 */
785 	for(;;) {
786 		if ((cnt = read(fd1, buf, blksz)) <= 0)
787 			break;
788 		if (no_hole)
789 			res = write(fd2, buf, cnt);
790 		else
791 			res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
792 		if (res != cnt)
793 			break;
794 		cpcnt += cnt;
795 	}
796 
797 	/*
798 	 * check to make sure the copy is valid.
799 	 */
800 	if (res < 0)
801 		syswarn(1, errno, "Failed write during copy of %s to %s",
802 			arcn->org_name, arcn->name);
803 	else if (cpcnt != arcn->sb.st_size)
804 		paxwarn(1, "File %s changed size during copy to %s",
805 			arcn->org_name, arcn->name);
806 	else if (fstat(fd1, &sb) < 0)
807 		syswarn(1, errno, "Failed stat of %s", arcn->org_name);
808 	else if (arcn->sb.st_mtime != sb.st_mtime)
809 		paxwarn(1, "File %s was modified during copy to %s",
810 			arcn->org_name, arcn->name);
811 
812 	/*
813 	 * if the last block has a file hole (all zero), we must make sure this
814 	 * gets updated in the file. We force the last block of zeros to be
815 	 * written. just closing with the file offset moved forward may not put
816 	 * a hole at the end of the file.
817 	 */
818 	if (!no_hole && isem && (arcn->sb.st_size > 0L))
819 		file_flush(fd2, fnm, isem);
820 	return;
821 }
822 
823 /*
824  * buf_fill()
825  *	fill the read buffer with the next record (or what we can get) from
826  *	the archive volume.
827  * Return:
828  *	Number of bytes of data in the read buffer, -1 for read error, and
829  *	0 when finished (user specified termination in ar_next()).
830  */
831 
832 int
833 buf_fill(void)
834 {
835 	int cnt;
836 	static int fini = 0;
837 
838 	if (fini)
839 		return(0);
840 
841 	for(;;) {
842 		/*
843 		 * try to fill the buffer. on error the next archive volume is
844 		 * opened and we try again.
845 		 */
846 		if ((cnt = ar_read(buf, blksz)) > 0) {
847 			bufpt = buf;
848 			bufend = buf + cnt;
849 			rdcnt += cnt;
850 			return(cnt);
851 		}
852 
853 		/*
854 		 * errors require resync, EOF goes to next archive
855 		 * but in case we have not determined yet the format,
856 		 * this means that we have a very short file, so we
857 		 * are done again.
858 		 */
859 		if (cnt < 0)
860 			break;
861 		if (frmt == NULL || ar_next() < 0) {
862 			fini = 1;
863 			return(0);
864 		}
865 		rdcnt = 0;
866 	}
867 	exit_val = 1;
868 	return(-1);
869 }
870 
871 /*
872  * buf_flush()
873  *	force the write buffer to the archive. We are passed the number of
874  *	bytes in the buffer at the point of the flush. When we change archives
875  *	the record size might change. (either larger or smaller).
876  * Return:
877  *	0 if all is ok, -1 when a write error occurs.
878  */
879 
880 int
881 buf_flush(int bufcnt)
882 {
883 	int cnt;
884 	int push = 0;
885 	int totcnt = 0;
886 
887 	/*
888 	 * if we have reached the user specified byte count for each archive
889 	 * volume, prompt for the next volume. (The non-standard -R flag).
890 	 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
891 	 * at least one record. We always round limit UP to next blocksize.
892 	 */
893 	if ((wrlimit > 0) && (wrcnt > wrlimit)) {
894 		paxwarn(0, "User specified archive volume byte limit reached.");
895 		if (ar_next() < 0) {
896 			wrcnt = 0;
897 			exit_val = 1;
898 			return(-1);
899 		}
900 		wrcnt = 0;
901 
902 		/*
903 		 * The new archive volume might have changed the size of the
904 		 * write blocksize. if so we figure out if we need to write
905 		 * (one or more times), or if there is now free space left in
906 		 * the buffer (it is no longer full). bufcnt has the number of
907 		 * bytes in the buffer, (the blocksize, at the point we were
908 		 * CALLED). Push has the amount of "extra" data in the buffer
909 		 * if the block size has shrunk from a volume change.
910 		 */
911 		bufend = buf + blksz;
912 		if (blksz > bufcnt)
913 			return(0);
914 		if (blksz < bufcnt)
915 			push = bufcnt - blksz;
916 	}
917 
918 	/*
919 	 * We have enough data to write at least one archive block
920 	 */
921 	for (;;) {
922 		/*
923 		 * write a block and check if it all went out ok
924 		 */
925 		cnt = ar_write(buf, blksz);
926 		if (cnt == blksz) {
927 			/*
928 			 * the write went ok
929 			 */
930 			wrcnt += cnt;
931 			totcnt += cnt;
932 			if (push > 0) {
933 				/* we have extra data to push to the front.
934 				 * check for more than 1 block of push, and if
935 				 * so we loop back to write again
936 				 */
937 				memcpy(buf, bufend, push);
938 				bufpt = buf + push;
939 				if (push >= blksz) {
940 					push -= blksz;
941 					continue;
942 				}
943 			} else
944 				bufpt = buf;
945 			return(totcnt);
946 		} else if (cnt > 0) {
947 			/*
948 			 * Oh drat we got a partial write!
949 			 * if format doesn't care about alignment let it go,
950 			 * we warned the user in ar_write().... but this means
951 			 * the last record on this volume violates pax spec....
952 			 */
953 			totcnt += cnt;
954 			wrcnt += cnt;
955 			bufpt = buf + cnt;
956 			cnt = bufcnt - cnt;
957 			memcpy(buf, bufpt, cnt);
958 			bufpt = buf + cnt;
959 			if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
960 				return(totcnt);
961 			break;
962 		}
963 
964 		/*
965 		 * All done, go to next archive
966 		 */
967 		wrcnt = 0;
968 		if (ar_next() < 0)
969 			break;
970 
971 		/*
972 		 * The new archive volume might also have changed the block
973 		 * size. if so, figure out if we have too much or too little
974 		 * data for using the new block size
975 		 */
976 		bufend = buf + blksz;
977 		if (blksz > bufcnt)
978 			return(0);
979 		if (blksz < bufcnt)
980 			push = bufcnt - blksz;
981 	}
982 
983 	/*
984 	 * write failed, stop pax. we must not create a bad archive!
985 	 */
986 	exit_val = 1;
987 	return(-1);
988 }
989