xref: /freebsd/bin/pax/buf_subs.c (revision 71fe318b852b8dfb3e799cb12ef184750f7f8eac)
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. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed by the University of
20  *	California, Berkeley and its contributors.
21  * 4. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  */
37 
38 #ifndef lint
39 #if 0
40 static char sccsid[] = "@(#)buf_subs.c	8.2 (Berkeley) 4/18/94";
41 #endif
42 #endif /* not lint */
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include <sys/types.h>
47 #include <sys/stat.h>
48 #include <errno.h>
49 #include <unistd.h>
50 #include <stdio.h>
51 #include <stdlib.h>
52 #include <string.h>
53 #include "pax.h"
54 #include "extern.h"
55 
56 /*
57  * routines which implement archive and file buffering
58  */
59 
60 #define MINFBSZ		512		/* default block size for hole detect */
61 #define MAXFLT		10		/* default media read error limit */
62 
63 /*
64  * Need to change bufmem to dynamic allocation when the upper
65  * limit on blocking size is removed (though that will violate pax spec)
66  * MAXBLK define and tests will also need to be updated.
67  */
68 static char bufmem[MAXBLK+BLKMULT];	/* i/o buffer + pushback id space */
69 static char *buf;			/* normal start of i/o buffer */
70 static char *bufend;			/* end or last char in i/o buffer */
71 static char *bufpt;			/* read/write point in i/o buffer */
72 int blksz = MAXBLK;			/* block input/output size in bytes */
73 int wrblksz;				/* user spec output size in bytes */
74 int maxflt = MAXFLT;			/* MAX consecutive media errors */
75 int rdblksz;				/* first read blksize (tapes only) */
76 off_t wrlimit;				/* # of bytes written per archive vol */
77 off_t wrcnt;				/* # of bytes written on current vol */
78 off_t rdcnt;				/* # of bytes read on current vol */
79 
80 /*
81  * wr_start()
82  *	set up the buffering system to operate in a write mode
83  * Return:
84  *	0 if ok, -1 if the user specified write block size violates pax spec
85  */
86 
87 int
88 wr_start(void)
89 {
90 	buf = &(bufmem[BLKMULT]);
91 	/*
92 	 * Check to make sure the write block size meets pax specs. If the user
93 	 * does not specify a blocksize, we use the format default blocksize.
94 	 * We must be picky on writes, so we do not allow the user to create an
95 	 * archive that might be hard to read elsewhere. If all ok, we then
96 	 * open the first archive volume
97 	 */
98 	if (!wrblksz)
99 		wrblksz = frmt->bsz;
100 	if (wrblksz > MAXBLK) {
101 		paxwarn(1, "Write block size of %d too large, maximum is: %d",
102 			wrblksz, MAXBLK);
103 		return(-1);
104 	}
105 	if (wrblksz % BLKMULT) {
106 		paxwarn(1, "Write block size of %d is not a %d byte multiple",
107 		    wrblksz, BLKMULT);
108 		return(-1);
109 	}
110 	if (wrblksz > MAXBLK_POSIX) {
111 		paxwarn(0, "Write block size of %d larger than POSIX max %d, archive may not be portable",
112 			wrblksz, MAXBLK_POSIX);
113 		return(-1);
114 	}
115 
116 	/*
117 	 * we only allow wrblksz to be used with all archive operations
118 	 */
119 	blksz = rdblksz = wrblksz;
120 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
121 		return(-1);
122 	wrcnt = 0;
123 	bufend = buf + wrblksz;
124 	bufpt = buf;
125 	return(0);
126 }
127 
128 /*
129  * rd_start()
130  *	set up buffering system to read an archive
131  * Return:
132  *	0 if ok, -1 otherwise
133  */
134 
135 int
136 rd_start(void)
137 {
138 	/*
139 	 * leave space for the header pushback (see get_arc()). If we are
140 	 * going to append and user specified a write block size, check it
141 	 * right away
142 	 */
143 	buf = &(bufmem[BLKMULT]);
144 	if ((act == APPND) && wrblksz) {
145 		if (wrblksz > MAXBLK) {
146 			paxwarn(1,"Write block size %d too large, maximum is: %d",
147 				wrblksz, MAXBLK);
148 			return(-1);
149 		}
150 		if (wrblksz % BLKMULT) {
151 			paxwarn(1, "Write block size %d is not a %d byte multiple",
152 		    	wrblksz, BLKMULT);
153 			return(-1);
154 		}
155 	}
156 
157 	/*
158 	 * open the archive
159 	 */
160 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
161 		return(-1);
162 	bufend = buf + rdblksz;
163 	bufpt = bufend;
164 	rdcnt = 0;
165 	return(0);
166 }
167 
168 /*
169  * cp_start()
170  *	set up buffer system for copying within the file system
171  */
172 
173 void
174 cp_start(void)
175 {
176 	buf = &(bufmem[BLKMULT]);
177 	rdblksz = blksz = MAXBLK;
178 }
179 
180 /*
181  * appnd_start()
182  *	Set up the buffering system to append new members to an archive that
183  *	was just read. The last block(s) of an archive may contain a format
184  *	specific trailer. To append a new member, this trailer has to be
185  *	removed from the archive. The first byte of the trailer is replaced by
186  *	the start of the header of the first file added to the archive. The
187  *	format specific end read function tells us how many bytes to move
188  *	backwards in the archive to be positioned BEFORE the trailer. Two
189  *	different postions have to be adjusted, the O.S. file offset (e.g. the
190  *	position of the tape head) and the write point within the data we have
191  *	stored in the read (soon to become write) buffer. We may have to move
192  *	back several records (the number depends on the size of the archive
193  *	record and the size of the format trailer) to read up the record where
194  *	the first byte of the trailer is recorded. Trailers may span (and
195  *	overlap) record boundries.
196  *	We first calculate which record has the first byte of the trailer. We
197  *	move the OS file offset back to the start of this record and read it
198  *	up. We set the buffer write pointer to be at this byte (the byte where
199  *	the trailer starts). We then move the OS file pointer back to the
200  *	start of this record so a flush of this buffer will replace the record
201  *	in the archive.
202  *	A major problem is rewriting this last record. For archives stored
203  *	on disk files, this is trival. However, many devices are really picky
204  *	about the conditions under which they will allow a write to occur.
205  *	Often devices restrict the conditions where writes can be made writes,
206  *	so it may not be feasable to append archives stored on all types of
207  *	devices.
208  * Return:
209  *	0 for success, -1 for failure
210  */
211 
212 int
213 appnd_start(off_t skcnt)
214 {
215 	int res;
216 	off_t cnt;
217 
218 	if (exit_val != 0) {
219 		paxwarn(0, "Cannot append to an archive that may have flaws.");
220 		return(-1);
221 	}
222 	/*
223 	 * if the user did not specify a write blocksize, inherit the size used
224 	 * in the last archive volume read. (If a is set we still use rdblksz
225 	 * until next volume, cannot shift sizes within a single volume).
226 	 */
227 	if (!wrblksz)
228 		wrblksz = blksz = rdblksz;
229 	else
230 		blksz = rdblksz;
231 
232 	/*
233 	 * make sure that this volume allows appends
234 	 */
235 	if (ar_app_ok() < 0)
236 		return(-1);
237 
238 	/*
239 	 * Calculate bytes to move back and move in front of record where we
240 	 * need to start writing from. Remember we have to add in any padding
241 	 * that might be in the buffer after the trailer in the last block. We
242 	 * travel skcnt + padding ROUNDED UP to blksize.
243 	 */
244 	skcnt += bufend - bufpt;
245 	if ((cnt = (skcnt/blksz) * blksz) < skcnt)
246 		cnt += blksz;
247 	if (ar_rev((off_t)cnt) < 0)
248 		goto out;
249 
250 	/*
251 	 * We may have gone too far if there is valid data in the block we are
252 	 * now in front of, read up the block and position the pointer after
253 	 * the valid data.
254 	 */
255 	if ((cnt -= skcnt) > 0) {
256 		/*
257 		 * watch out for stupid tape drives. ar_rev() will set rdblksz
258 		 * to be real physical blocksize so we must loop until we get
259 		 * the old rdblksz (now in blksz). If ar_rev() fouls up the
260 		 * determination of the physical block size, we will fail.
261 		 */
262 		bufpt = buf;
263 		bufend = buf + blksz;
264 		while (bufpt < bufend) {
265 			if ((res = ar_read(bufpt, rdblksz)) <= 0)
266 				goto out;
267 			bufpt += res;
268 		}
269 		if (ar_rev((off_t)(bufpt - buf)) < 0)
270 			goto out;
271 		bufpt = buf + cnt;
272 		bufend = buf + blksz;
273 	} else {
274 		/*
275 		 * buffer is empty
276 		 */
277 		bufend = buf + blksz;
278 		bufpt = buf;
279 	}
280 	rdblksz = blksz;
281 	rdcnt -= skcnt;
282 	wrcnt = 0;
283 
284 	/*
285 	 * At this point we are ready to write. If the device requires special
286 	 * handling to write at a point were previously recorded data resides,
287 	 * that is handled in ar_set_wr(). From now on we operate under normal
288 	 * ARCHIVE mode (write) conditions
289 	 */
290 	if (ar_set_wr() < 0)
291 		return(-1);
292 	act = ARCHIVE;
293 	return(0);
294 
295     out:
296 	paxwarn(1, "Unable to rewrite archive trailer, cannot append.");
297 	return(-1);
298 }
299 
300 /*
301  * rd_sync()
302  *	A read error occurred on this archive volume. Resync the buffer and
303  *	try to reset the device (if possible) so we can continue to read. Keep
304  *	trying to do this until we get a valid read, or we reach the limit on
305  *	consecutive read faults (at which point we give up). The user can
306  *	adjust the read error limit through a command line option.
307  * Returns:
308  *	0 on success, and -1 on failure
309  */
310 
311 int
312 rd_sync(void)
313 {
314 	int errcnt = 0;
315 	int res;
316 
317 	/*
318 	 * if the user says bail out on first fault, we are out of here...
319 	 */
320 	if (maxflt == 0)
321 		return(-1);
322 	if (act == APPND) {
323 		paxwarn(1, "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. o.w. 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 			paxwarn(0,"Archive read error limit (%d) reached",maxflt);
358 		else if (ar_rdsync() == 0)
359 			continue;
360 		if (ar_next() < 0)
361 			break;
362 		rdcnt = 0;
363 		errcnt = 0;
364 	}
365 	return(-1);
366 }
367 
368 /*
369  * pback()
370  *	push the data used during the archive id phase back into the I/O
371  *	buffer. This is required as we cannot be sure that the header does NOT
372  *	overlap a block boundry (as in the case we are trying to recover a
373  *	flawed archived). This was not designed to be used for any other
374  *	purpose. (What software engineering, HA!)
375  *	WARNING: do not even THINK of pback greater than BLKMULT, unless the
376  *	pback space is increased.
377  */
378 
379 void
380 pback(char *pt, int cnt)
381 {
382 	bufpt -= cnt;
383 	memcpy(bufpt, pt, cnt);
384 	return;
385 }
386 
387 /*
388  * rd_skip()
389  *	skip foward in the archive during a archive read. Used to get quickly
390  *	past file data and padding for files the user did NOT select.
391  * Return:
392  *	0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
393  */
394 
395 int
396 rd_skip(off_t skcnt)
397 {
398 	off_t res;
399 	off_t cnt;
400 	off_t skipped = 0;
401 
402 	/*
403 	 * consume what data we have in the buffer. If we have to move foward
404 	 * whole records, we call the low level skip function to see if we can
405 	 * move within the archive without doing the expensive reads on data we
406 	 * do not want.
407 	 */
408 	if (skcnt == 0)
409 		return(0);
410 	res = MIN((bufend - bufpt), skcnt);
411 	bufpt += res;
412 	skcnt -= res;
413 
414 	/*
415 	 * if skcnt is now 0, then no additional i/o is needed
416 	 */
417 	if (skcnt == 0)
418 		return(0);
419 
420 	/*
421 	 * We have to read more, calculate complete and partial record reads
422 	 * based on rdblksz. we skip over "cnt" complete records
423 	 */
424 	res = skcnt%rdblksz;
425 	cnt = (skcnt/rdblksz) * rdblksz;
426 
427 	/*
428 	 * if the skip fails, we will have to resync. ar_fow will tell us
429 	 * how much it can skip over. We will have to read the rest.
430 	 */
431 	if (ar_fow(cnt, &skipped) < 0)
432 		return(-1);
433 	res += cnt - skipped;
434 	rdcnt += skipped;
435 
436 	/*
437 	 * what is left we have to read (which may be the whole thing if
438 	 * ar_fow() told us the device can only read to skip records);
439 	 */
440 	while (res > 0L) {
441 		cnt = bufend - bufpt;
442 		/*
443 		 * if the read fails, we will have to resync
444 		 */
445 		if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
446 			return(-1);
447 		if (cnt == 0)
448 			return(1);
449 		cnt = MIN(cnt, res);
450 		bufpt += cnt;
451 		res -= cnt;
452 	}
453 	return(0);
454 }
455 
456 /*
457  * wr_fin()
458  *	flush out any data (and pad if required) the last block. We always pad
459  *	with zero (even though we do not have to). Padding with 0 makes it a
460  *	lot easier to recover if the archive is damaged. zero paddding SHOULD
461  *	BE a requirement....
462  */
463 
464 void
465 wr_fin(void)
466 {
467 	if (bufpt > buf) {
468 		memset(bufpt, 0, bufend - bufpt);
469 		bufpt = bufend;
470 		(void)buf_flush(blksz);
471 	}
472 }
473 
474 /*
475  * wr_rdbuf()
476  *	fill the write buffer from data passed to it in a buffer (usually used
477  *	by format specific write routines to pass a file header). On failure we
478  *	punt. We do not allow the user to continue to write flawed archives.
479  *	We assume these headers are not very large (the memory copy we use is
480  *	a bit expensive).
481  * Return:
482  *	0 if buffer was filled ok, -1 o.w. (buffer flush failure)
483  */
484 
485 int
486 wr_rdbuf(char *out, int outcnt)
487 {
488 	int cnt;
489 
490 	/*
491 	 * while there is data to copy copy into the write buffer. when the
492 	 * write buffer fills, flush it to the archive and continue
493 	 */
494 	while (outcnt > 0) {
495 		cnt = bufend - bufpt;
496 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
497 			return(-1);
498 		/*
499 		 * only move what we have space for
500 		 */
501 		cnt = MIN(cnt, outcnt);
502 		memcpy(bufpt, out, cnt);
503 		bufpt += cnt;
504 		out += cnt;
505 		outcnt -= cnt;
506 	}
507 	return(0);
508 }
509 
510 /*
511  * rd_wrbuf()
512  *	copy from the read buffer into a supplied buffer a specified number of
513  *	bytes. If the read buffer is empty fill it and continue to copy.
514  *	usually used to obtain a file header for processing by a format
515  *	specific read routine.
516  * Return
517  *	number of bytes copied to the buffer, 0 indicates EOF on archive volume,
518  *	-1 is a read error
519  */
520 
521 int
522 rd_wrbuf(char *in, int cpcnt)
523 {
524 	int res;
525 	int cnt;
526 	int incnt = cpcnt;
527 
528 	/*
529 	 * loop until we fill the buffer with the requested number of bytes
530 	 */
531 	while (incnt > 0) {
532 		cnt = bufend - bufpt;
533 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
534 			/*
535 			 * read error, return what we got (or the error if
536 			 * no data was copied). The caller must know that an
537 			 * error occured and has the best knowledge what to
538 			 * do with it
539 			 */
540 			if ((res = cpcnt - incnt) > 0)
541 				return(res);
542 			return(cnt);
543 		}
544 
545 		/*
546 		 * calculate how much data to copy based on whats left and
547 		 * state of buffer
548 		 */
549 		cnt = MIN(cnt, incnt);
550 		memcpy(in, bufpt, cnt);
551 		bufpt += cnt;
552 		incnt -= cnt;
553 		in += cnt;
554 	}
555 	return(cpcnt);
556 }
557 
558 /*
559  * wr_skip()
560  *	skip forward during a write. In other words add padding to the file.
561  *	we add zero filled padding as it makes flawed archives much easier to
562  *	recover from. the caller tells us how many bytes of padding to add
563  *	This routine was not designed to add HUGE amount of padding, just small
564  *	amounts (a few 512 byte blocks at most)
565  * Return:
566  *	0 if ok, -1 if there was a buf_flush failure
567  */
568 
569 int
570 wr_skip(off_t skcnt)
571 {
572 	int cnt;
573 
574 	/*
575 	 * loop while there is more padding to add
576 	 */
577 	while (skcnt > 0L) {
578 		cnt = bufend - bufpt;
579 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
580 			return(-1);
581 		cnt = MIN(cnt, skcnt);
582 		memset(bufpt, 0, cnt);
583 		bufpt += cnt;
584 		skcnt -= cnt;
585 	}
586 	return(0);
587 }
588 
589 /*
590  * wr_rdfile()
591  *	fill write buffer with the contents of a file. We are passed an	open
592  *	file descriptor to the file and the archive structure that describes the
593  *	file we are storing. The variable "left" is modified to contain the
594  *	number of bytes of the file we were NOT able to write to the archive.
595  *	it is important that we always write EXACTLY the number of bytes that
596  *	the format specific write routine told us to. The file can also get
597  *	bigger, so reading to the end of file would create an improper archive,
598  *	we just detect this case and warn the user. We never create a bad
599  *	archive if we can avoid it. Of course trying to archive files that are
600  *	active is asking for trouble. It we fail, we pass back how much we
601  *	could NOT copy and let the caller deal with it.
602  * Return:
603  *	0 ok, -1 if archive write failure. a short read of the file returns a
604  *	0, but "left" is set to be greater than zero.
605  */
606 
607 int
608 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
609 {
610 	int cnt;
611 	int res = 0;
612 	off_t size = arcn->sb.st_size;
613 	struct stat sb;
614 
615 	/*
616 	 * while there are more bytes to write
617 	 */
618 	while (size > 0L) {
619 		cnt = bufend - bufpt;
620 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
621 			*left = size;
622 			return(-1);
623 		}
624 		cnt = MIN(cnt, size);
625 		if ((res = read(ifd, bufpt, cnt)) <= 0)
626 			break;
627 		size -= res;
628 		bufpt += res;
629 	}
630 
631 	/*
632 	 * better check the file did not change during this operation
633 	 * or the file read failed.
634 	 */
635 	if (res < 0)
636 		syswarn(1, errno, "Read fault on %s", arcn->org_name);
637 	else if (size != 0L)
638 		paxwarn(1, "File changed size during read %s", arcn->org_name);
639 	else if (fstat(ifd, &sb) < 0)
640 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
641 	else if (arcn->sb.st_mtime != sb.st_mtime)
642 		paxwarn(1, "File %s was modified during copy to archive",
643 			arcn->org_name);
644 	*left = size;
645 	return(0);
646 }
647 
648 /*
649  * rd_wrfile()
650  *	extract the contents of a file from the archive. If we are unable to
651  *	extract the entire file (due to failure to write the file) we return
652  *	the numbers of bytes we did NOT process. This way the caller knows how
653  *	many bytes to skip past to find the next archive header. If the failure
654  *	was due to an archive read, we will catch that when we try to skip. If
655  *	the format supplies a file data crc value, we calculate the actual crc
656  *	so that it can be compared to the value stored in the header
657  * NOTE:
658  *	We call a special function to write the file. This function attempts to
659  *	restore file holes (blocks of zeros) into the file. When files are
660  *	sparse this saves space, and is a LOT faster. For non sparse files
661  *	the performance hit is small. As of this writing, no archive supports
662  *	information on where the file holes are.
663  * Return:
664  *	0 ok, -1 if archive read failure. if we cannot write the entire file,
665  *	we return a 0 but "left" is set to be the amount unwritten
666  */
667 
668 int
669 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
670 {
671 	int cnt = 0;
672 	off_t size = arcn->sb.st_size;
673 	int res = 0;
674 	char *fnm = arcn->name;
675 	int isem = 1;
676 	int rem;
677 	int sz = MINFBSZ;
678  	struct stat sb;
679 	u_long crc = 0L;
680 
681 	/*
682 	 * pass the blocksize of the file being written to the write routine,
683 	 * if the size is zero, use the default MINFBSZ
684 	 */
685 	if (fstat(ofd, &sb) == 0) {
686 		if (sb.st_blksize > 0)
687 			sz = (int)sb.st_blksize;
688 	} else
689 		syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
690 	rem = sz;
691 	*left = 0L;
692 
693 	/*
694 	 * Copy the archive to the file the number of bytes specified. We have
695 	 * to assume that we want to recover file holes as none of the archive
696 	 * formats can record the location of file holes.
697 	 */
698 	while (size > 0L) {
699 		cnt = bufend - bufpt;
700 		/*
701 		 * if we get a read error, we do not want to skip, as we may
702 		 * miss a header, so we do not set left, but if we get a write
703 		 * error, we do want to skip over the unprocessed data.
704 		 */
705 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
706 			break;
707 		cnt = MIN(cnt, size);
708 		if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
709 			*left = size;
710 			break;
711 		}
712 
713 		if (docrc) {
714 			/*
715 			 * update the actual crc value
716 			 */
717 			cnt = res;
718 			while (--cnt >= 0)
719 				crc += *bufpt++ & 0xff;
720 		} else
721 			bufpt += res;
722 		size -= res;
723 	}
724 
725 	/*
726 	 * if the last block has a file hole (all zero), we must make sure this
727 	 * gets updated in the file. We force the last block of zeros to be
728 	 * written. just closing with the file offset moved forward may not put
729 	 * a hole at the end of the file.
730 	 */
731 	if (isem && (arcn->sb.st_size > 0L))
732 		file_flush(ofd, fnm, isem);
733 
734 	/*
735 	 * if we failed from archive read, we do not want to skip
736 	 */
737 	if ((size > 0L) && (*left == 0L))
738 		return(-1);
739 
740 	/*
741 	 * some formats record a crc on file data. If so, then we compare the
742 	 * calculated crc to the crc stored in the archive
743 	 */
744 	if (docrc && (size == 0L) && (arcn->crc != crc))
745 		paxwarn(1,"Actual crc does not match expected crc %s",arcn->name);
746 	return(0);
747 }
748 
749 /*
750  * cp_file()
751  *	copy the contents of one file to another. used during -rw phase of pax
752  *	just as in rd_wrfile() we use a special write function to write the
753  *	destination file so we can properly copy files with holes.
754  */
755 
756 void
757 cp_file(ARCHD *arcn, int fd1, int fd2)
758 {
759 	int cnt;
760 	off_t cpcnt = 0L;
761 	int res = 0;
762 	char *fnm = arcn->name;
763 	int no_hole = 0;
764 	int isem = 1;
765 	int rem;
766 	int sz = MINFBSZ;
767 	struct stat sb;
768 
769 	/*
770 	 * check for holes in the source file. If none, we will use regular
771 	 * write instead of file write.
772 	 */
773 	 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
774 		++no_hole;
775 
776 	/*
777 	 * pass the blocksize of the file being written to the write routine,
778 	 * if the size is zero, use the default MINFBSZ
779 	 */
780 	if (fstat(fd2, &sb) == 0) {
781 		if (sb.st_blksize > 0)
782 			sz = sb.st_blksize;
783 	} else
784 		syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
785 	rem = sz;
786 
787 	/*
788 	 * read the source file and copy to destination file until EOF
789 	 */
790 	for(;;) {
791 		if ((cnt = read(fd1, buf, blksz)) <= 0)
792 			break;
793 		if (no_hole)
794 			res = write(fd2, buf, cnt);
795 		else
796 			res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
797 		if (res != cnt)
798 			break;
799 		cpcnt += cnt;
800 	}
801 
802 	/*
803 	 * check to make sure the copy is valid.
804 	 */
805 	if (res < 0)
806 		syswarn(1, errno, "Failed write during copy of %s to %s",
807 			arcn->org_name, arcn->name);
808 	else if (cpcnt != arcn->sb.st_size)
809 		paxwarn(1, "File %s changed size during copy to %s",
810 			arcn->org_name, arcn->name);
811 	else if (fstat(fd1, &sb) < 0)
812 		syswarn(1, errno, "Failed stat of %s", arcn->org_name);
813 	else if (arcn->sb.st_mtime != sb.st_mtime)
814 		paxwarn(1, "File %s was modified during copy to %s",
815 			arcn->org_name, arcn->name);
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 forward may not put
821 	 * a hole at the end of the file.
822 	 */
823 	if (!no_hole && isem && (arcn->sb.st_size > 0L))
824 		file_flush(fd2, fnm, isem);
825 	return;
826 }
827 
828 /*
829  * buf_fill()
830  *	fill the read buffer with the next record (or what we can get) from
831  *	the archive volume.
832  * Return:
833  *	Number of bytes of data in the read buffer, -1 for read error, and
834  *	0 when finished (user specified termination in ar_next()).
835  */
836 
837 int
838 buf_fill(void)
839 {
840 	int cnt;
841 	static int fini = 0;
842 
843 	if (fini)
844 		return(0);
845 
846 	for(;;) {
847 		/*
848 		 * try to fill the buffer. on error the next archive volume is
849 		 * opened and we try again.
850 		 */
851 		if ((cnt = ar_read(buf, blksz)) > 0) {
852 			bufpt = buf;
853 			bufend = buf + cnt;
854 			rdcnt += cnt;
855 			return(cnt);
856 		}
857 
858 		/*
859 		 * errors require resync, EOF goes to next archive
860 		 */
861 		if (cnt < 0)
862 			break;
863 		if (ar_next() < 0) {
864 			fini = 1;
865 			return(0);
866 		}
867 		rdcnt = 0;
868 	}
869 	exit_val = 1;
870 	return(-1);
871 }
872 
873 /*
874  * buf_flush()
875  *	force the write buffer to the archive. We are passed the number of
876  *	bytes in the buffer at the point of the flush. When we change archives
877  *	the record size might change. (either larger or smaller).
878  * Return:
879  *	0 if all is ok, -1 when a write error occurs.
880  */
881 
882 int
883 buf_flush(int bufcnt)
884 {
885 	int cnt;
886 	int push = 0;
887 	int totcnt = 0;
888 
889 	/*
890 	 * if we have reached the user specified byte count for each archive
891 	 * volume, prompt for the next volume. (The non-standrad -R flag).
892 	 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
893 	 * at least one record. We always round limit UP to next blocksize.
894 	 */
895 	if ((wrlimit > 0) && (wrcnt > wrlimit)) {
896 		paxwarn(0, "User specified archive volume byte limit reached.");
897 		if (ar_next() < 0) {
898 			wrcnt = 0;
899 			exit_val = 1;
900 			return(-1);
901 		}
902 		wrcnt = 0;
903 
904 		/*
905 		 * The new archive volume might have changed the size of the
906 		 * write blocksize. if so we figure out if we need to write
907 		 * (one or more times), or if there is now free space left in
908 		 * the buffer (it is no longer full). bufcnt has the number of
909 		 * bytes in the buffer, (the blocksize, at the point we were
910 		 * CALLED). Push has the amount of "extra" data in the buffer
911 		 * if the block size has shrunk from a volume change.
912 		 */
913 		bufend = buf + blksz;
914 		if (blksz > bufcnt)
915 			return(0);
916 		if (blksz < bufcnt)
917 			push = bufcnt - blksz;
918 	}
919 
920 	/*
921 	 * We have enough data to write at least one archive block
922 	 */
923 	for (;;) {
924 		/*
925 		 * write a block and check if it all went out ok
926 		 */
927 		cnt = ar_write(buf, blksz);
928 		if (cnt == blksz) {
929 			/*
930 			 * the write went ok
931 			 */
932 			wrcnt += cnt;
933 			totcnt += cnt;
934 			if (push > 0) {
935 				/* we have extra data to push to the front.
936 				 * check for more than 1 block of push, and if
937 				 * so we loop back to write again
938 				 */
939 				memcpy(buf, bufend, push);
940 				bufpt = buf + push;
941 				if (push >= blksz) {
942 					push -= blksz;
943 					continue;
944 				}
945 			} else
946 				bufpt = buf;
947 			return(totcnt);
948 		} else if (cnt > 0) {
949 			/*
950 			 * Oh drat we got a partial write!
951 			 * if format doesnt care about alignment let it go,
952 			 * we warned the user in ar_write().... but this means
953 			 * the last record on this volume violates pax spec....
954 			 */
955 			totcnt += cnt;
956 			wrcnt += cnt;
957 			bufpt = buf + cnt;
958 			cnt = bufcnt - cnt;
959 			memcpy(buf, bufpt, cnt);
960 			bufpt = buf + cnt;
961 			if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
962 				return(totcnt);
963 			break;
964 		}
965 
966 		/*
967 		 * All done, go to next archive
968 		 */
969 		wrcnt = 0;
970 		if (ar_next() < 0)
971 			break;
972 
973 		/*
974 		 * The new archive volume might also have changed the block
975 		 * size. if so, figure out if we have too much or too little
976 		 * data for using the new block size
977 		 */
978 		bufend = buf + blksz;
979 		if (blksz > bufcnt)
980 			return(0);
981 		if (blksz < bufcnt)
982 			push = bufcnt - blksz;
983 	}
984 
985 	/*
986 	 * write failed, stop pax. we must not create a bad archive!
987 	 */
988 	exit_val = 1;
989 	return(-1);
990 }
991