xref: /openbsd/bin/pax/buf_subs.c (revision 9b7c3dbb)
1 /*	$OpenBSD: buf_subs.c,v 1.29 2016/08/26 04:11:16 guenther Exp $	*/
2 /*	$NetBSD: buf_subs.c,v 1.5 1995/03/21 09:07:08 cgd Exp $	*/
3 
4 /*-
5  * Copyright (c) 1992 Keith Muller.
6  * Copyright (c) 1992, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * Keith Muller of the University of California, San Diego.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include <sys/types.h>
38 #include <sys/stat.h>
39 #include <stdio.h>
40 #include <errno.h>
41 #include <unistd.h>
42 #include <stdlib.h>
43 #include <string.h>
44 #include "pax.h"
45 #include "extern.h"
46 
47 /*
48  * routines which implement archive and file buffering
49  */
50 
51 #define MINFBSZ		512		/* default block size for hole detect */
52 #define MAXFLT		10		/* default media read error limit */
53 
54 /*
55  * Need to change bufmem to dynamic allocation when the upper
56  * limit on blocking size is removed (though that will violate pax spec)
57  * MAXBLK define and tests will also need to be updated.
58  */
59 static char bufmem[MAXBLK+BLKMULT];	/* i/o buffer + pushback id space */
60 static char *buf;			/* normal start of i/o buffer */
61 static char *bufend;			/* end or last char in i/o buffer */
62 static char *bufpt;			/* read/write point in i/o buffer */
63 int blksz = MAXBLK;			/* block input/output size in bytes */
64 int wrblksz;				/* user spec output size in bytes */
65 int maxflt = MAXFLT;			/* MAX consecutive media errors */
66 int rdblksz;				/* first read blksize (tapes only) */
67 off_t wrlimit;				/* # of bytes written per archive vol */
68 off_t wrcnt;				/* # of bytes written on current vol */
69 off_t rdcnt;				/* # of bytes read on current vol */
70 
71 /*
72  * wr_start()
73  *	set up the buffering system to operate in a write mode
74  * Return:
75  *	0 if ok, -1 if the user specified write block size violates pax spec
76  */
77 
78 int
79 wr_start(void)
80 {
81 	buf = &(bufmem[BLKMULT]);
82 	/*
83 	 * Check to make sure the write block size meets pax specs. If the user
84 	 * does not specify a blocksize, we use the format default blocksize.
85 	 * We must be picky on writes, so we do not allow the user to create an
86 	 * archive that might be hard to read elsewhere. If all ok, we then
87 	 * open the first archive volume
88 	 */
89 	if (!wrblksz)
90 		wrblksz = frmt->bsz;
91 	if (wrblksz > MAXBLK) {
92 		paxwarn(1, "Write block size of %d too large, maximium is: %d",
93 			wrblksz, MAXBLK);
94 		return(-1);
95 	}
96 	if (wrblksz % BLKMULT) {
97 		paxwarn(1, "Write block size of %d is not a %d byte multiple",
98 		    wrblksz, BLKMULT);
99 		return(-1);
100 	}
101 	if (wrblksz > MAXBLK_POSIX) {
102 		paxwarn(0, "Write block size of %d larger than POSIX max %d, archive may not be portable",
103 			wrblksz, MAXBLK_POSIX);
104 		return(-1);
105 	}
106 
107 	/*
108 	 * we only allow wrblksz to be used with all archive operations
109 	 */
110 	blksz = rdblksz = wrblksz;
111 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
112 		return(-1);
113 	wrcnt = 0;
114 	bufend = buf + wrblksz;
115 	bufpt = buf;
116 	return(0);
117 }
118 
119 /*
120  * rd_start()
121  *	set up buffering system to read an archive
122  * Return:
123  *	0 if ok, -1 otherwise
124  */
125 
126 int
127 rd_start(void)
128 {
129 	/*
130 	 * leave space for the header pushback (see get_arc()). If we are
131 	 * going to append and user specified a write block size, check it
132 	 * right away
133 	 */
134 	buf = &(bufmem[BLKMULT]);
135 	if ((act == APPND) && wrblksz) {
136 		if (wrblksz > MAXBLK) {
137 			paxwarn(1,"Write block size %d too large, maximium is: %d",
138 				wrblksz, MAXBLK);
139 			return(-1);
140 		}
141 		if (wrblksz % BLKMULT) {
142 			paxwarn(1, "Write block size %d is not a %d byte multiple",
143 			wrblksz, BLKMULT);
144 			return(-1);
145 		}
146 	}
147 
148 	/*
149 	 * open the archive
150 	 */
151 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
152 		return(-1);
153 	bufend = buf + rdblksz;
154 	bufpt = bufend;
155 	rdcnt = 0;
156 	return(0);
157 }
158 
159 /*
160  * cp_start()
161  *	set up buffer system for copying within the file system
162  */
163 
164 void
165 cp_start(void)
166 {
167 	buf = &(bufmem[BLKMULT]);
168 	rdblksz = blksz = MAXBLK;
169 }
170 
171 /*
172  * appnd_start()
173  *	Set up the buffering system to append new members to an archive that
174  *	was just read. The last block(s) of an archive may contain a format
175  *	specific trailer. To append a new member, this trailer has to be
176  *	removed from the archive. The first byte of the trailer is replaced by
177  *	the start of the header of the first file added to the archive. The
178  *	format specific end read function tells us how many bytes to move
179  *	backwards in the archive to be positioned BEFORE the trailer. Two
180  *	different position have to be adjusted, the O.S. file offset (e.g. the
181  *	position of the tape head) and the write point within the data we have
182  *	stored in the read (soon to become write) buffer. We may have to move
183  *	back several records (the number depends on the size of the archive
184  *	record and the size of the format trailer) to read up the record where
185  *	the first byte of the trailer is recorded. Trailers may span (and
186  *	overlap) record boundaries.
187  *	We first calculate which record has the first byte of the trailer. We
188  *	move the OS file offset back to the start of this record and read it
189  *	up. We set the buffer write pointer to be at this byte (the byte where
190  *	the trailer starts). We then move the OS file pointer back to the
191  *	start of this record so a flush of this buffer will replace the record
192  *	in the archive.
193  *	A major problem is rewriting this last record. For archives stored
194  *	on disk files, this is trivial. However, many devices are really picky
195  *	about the conditions under which they will allow a write to occur.
196  *	Often devices restrict the conditions where writes can be made,
197  *	so it may not be feasible to append archives stored on all types of
198  *	devices.
199  * Return:
200  *	0 for success, -1 for failure
201  */
202 
203 int
204 appnd_start(off_t skcnt)
205 {
206 	int res;
207 	off_t cnt;
208 
209 	if (exit_val != 0) {
210 		paxwarn(0, "Cannot append to an archive that may have flaws.");
211 		return(-1);
212 	}
213 	/*
214 	 * if the user did not specify a write blocksize, inherit the size used
215 	 * in the last archive volume read. (If a is set we still use rdblksz
216 	 * until next volume, cannot shift sizes within a single volume).
217 	 */
218 	if (!wrblksz)
219 		wrblksz = blksz = rdblksz;
220 	else
221 		blksz = rdblksz;
222 
223 	/*
224 	 * make sure that this volume allows appends
225 	 */
226 	if (ar_app_ok() < 0)
227 		return(-1);
228 
229 	/*
230 	 * Calculate bytes to move back and move in front of record where we
231 	 * need to start writing from. Remember we have to add in any padding
232 	 * that might be in the buffer after the trailer in the last block. We
233 	 * travel skcnt + padding ROUNDED UP to blksize.
234 	 */
235 	skcnt += bufend - bufpt;
236 	if ((cnt = (skcnt/blksz) * blksz) < skcnt)
237 		cnt += blksz;
238 	if (ar_rev(cnt) < 0)
239 		goto out;
240 
241 	/*
242 	 * We may have gone too far if there is valid data in the block we are
243 	 * now in front of, read up the block and position the pointer after
244 	 * the valid data.
245 	 */
246 	if ((cnt -= skcnt) > 0) {
247 		/*
248 		 * watch out for stupid tape drives. ar_rev() will set rdblksz
249 		 * to be real physical blocksize so we must loop until we get
250 		 * the old rdblksz (now in blksz). If ar_rev() fouls up the
251 		 * determination of the physical block size, we will fail.
252 		 */
253 		bufpt = buf;
254 		bufend = buf + blksz;
255 		while (bufpt < bufend) {
256 			if ((res = ar_read(bufpt, rdblksz)) <= 0)
257 				goto out;
258 			bufpt += res;
259 		}
260 		if (ar_rev(bufpt - buf) < 0)
261 			goto out;
262 		bufpt = buf + cnt;
263 		bufend = buf + blksz;
264 	} else {
265 		/*
266 		 * buffer is empty
267 		 */
268 		bufend = buf + blksz;
269 		bufpt = buf;
270 	}
271 	rdblksz = blksz;
272 	rdcnt -= skcnt;
273 	wrcnt = 0;
274 
275 	/*
276 	 * At this point we are ready to write. If the device requires special
277 	 * handling to write at a point were previously recorded data resides,
278 	 * that is handled in ar_set_wr(). From now on we operate under normal
279 	 * ARCHIVE mode (write) conditions
280 	 */
281 	if (ar_set_wr() < 0)
282 		return(-1);
283 	act = ARCHIVE;
284 	return(0);
285 
286     out:
287 	paxwarn(1, "Unable to rewrite archive trailer, cannot append.");
288 	return(-1);
289 }
290 
291 /*
292  * rd_sync()
293  *	A read error occurred on this archive volume. Resync the buffer and
294  *	try to reset the device (if possible) so we can continue to read. Keep
295  *	trying to do this until we get a valid read, or we reach the limit on
296  *	consecutive read faults (at which point we give up). The user can
297  *	adjust the read error limit through a command line option.
298  * Returns:
299  *	0 on success, and -1 on failure
300  */
301 
302 int
303 rd_sync(void)
304 {
305 	int errcnt = 0;
306 	int res;
307 
308 	/*
309 	 * if the user says bail out on first fault, we are out of here...
310 	 */
311 	if (maxflt == 0)
312 		return(-1);
313 	if (act == APPND) {
314 		paxwarn(1, "Unable to append when there are archive read errors.");
315 		return(-1);
316 	}
317 
318 	/*
319 	 * poke at device and try to get past media error
320 	 */
321 	if (ar_rdsync() < 0) {
322 		if (ar_next() < 0)
323 			return(-1);
324 		else
325 			rdcnt = 0;
326 	}
327 
328 	for (;;) {
329 		if ((res = ar_read(buf, blksz)) > 0) {
330 			/*
331 			 * All right! got some data, fill that buffer
332 			 */
333 			bufpt = buf;
334 			bufend = buf + res;
335 			rdcnt += res;
336 			return(0);
337 		}
338 
339 		/*
340 		 * Oh well, yet another failed read...
341 		 * if error limit reached, ditch. o.w. poke device to move past
342 		 * bad media and try again. if media is badly damaged, we ask
343 		 * the poor (and upset user at this point) for the next archive
344 		 * volume. remember the goal on reads is to get the most we
345 		 * can extract out of the archive.
346 		 */
347 		if ((maxflt > 0) && (++errcnt > maxflt))
348 			paxwarn(0,"Archive read error limit (%d) reached",maxflt);
349 		else if (ar_rdsync() == 0)
350 			continue;
351 		if (ar_next() < 0)
352 			break;
353 		rdcnt = 0;
354 		errcnt = 0;
355 	}
356 	return(-1);
357 }
358 
359 /*
360  * pback()
361  *	push the data used during the archive id phase back into the I/O
362  *	buffer. This is required as we cannot be sure that the header does NOT
363  *	overlap a block boundary (as in the case we are trying to recover a
364  *	flawed archived). This was not designed to be used for any other
365  *	purpose. (What software engineering, HA!)
366  *	WARNING: do not even THINK of pback greater than BLKMULT, unless the
367  *	pback space is increased.
368  */
369 
370 void
371 pback(char *pt, int cnt)
372 {
373 	bufpt -= cnt;
374 	memcpy(bufpt, pt, cnt);
375 }
376 
377 /*
378  * rd_skip()
379  *	skip forward in the archive during a archive read. Used to get quickly
380  *	past file data and padding for files the user did NOT select.
381  * Return:
382  *	0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
383  */
384 
385 int
386 rd_skip(off_t skcnt)
387 {
388 	off_t res;
389 	off_t cnt;
390 	off_t skipped = 0;
391 
392 	/*
393 	 * consume what data we have in the buffer. If we have to move forward
394 	 * whole records, we call the low level skip function to see if we can
395 	 * move within the archive without doing the expensive reads on data we
396 	 * do not want.
397 	 */
398 	if (skcnt == 0)
399 		return(0);
400 	res = MINIMUM((bufend - bufpt), skcnt);
401 	bufpt += res;
402 	skcnt -= res;
403 
404 	/*
405 	 * if skcnt is now 0, then no additional i/o is needed
406 	 */
407 	if (skcnt == 0)
408 		return(0);
409 
410 	/*
411 	 * We have to read more, calculate complete and partial record reads
412 	 * based on rdblksz. we skip over "cnt" complete records
413 	 */
414 	res = skcnt%rdblksz;
415 	cnt = (skcnt/rdblksz) * rdblksz;
416 
417 	/*
418 	 * if the skip fails, we will have to resync. ar_fow will tell us
419 	 * how much it can skip over. We will have to read the rest.
420 	 */
421 	if (ar_fow(cnt, &skipped) < 0)
422 		return(-1);
423 	res += cnt - skipped;
424 	rdcnt += skipped;
425 
426 	/*
427 	 * what is left we have to read (which may be the whole thing if
428 	 * ar_fow() told us the device can only read to skip records);
429 	 */
430 	while (res > 0) {
431 		cnt = bufend - bufpt;
432 		/*
433 		 * if the read fails, we will have to resync
434 		 */
435 		if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
436 			return(-1);
437 		if (cnt == 0)
438 			return(1);
439 		cnt = MINIMUM(cnt, res);
440 		bufpt += cnt;
441 		res -= cnt;
442 	}
443 	return(0);
444 }
445 
446 /*
447  * wr_fin()
448  *	flush out any data (and pad if required) the last block. We always pad
449  *	with zero (even though we do not have to). Padding with 0 makes it a
450  *	lot easier to recover if the archive is damaged. zero padding SHOULD
451  *	BE a requirement....
452  */
453 
454 void
455 wr_fin(void)
456 {
457 	if (bufpt > buf) {
458 		memset(bufpt, 0, bufend - bufpt);
459 		bufpt = bufend;
460 		(void)buf_flush(blksz);
461 	}
462 }
463 
464 /*
465  * wr_rdbuf()
466  *	fill the write buffer from data passed to it in a buffer (usually used
467  *	by format specific write routines to pass a file header). On failure we
468  *	punt. We do not allow the user to continue to write flawed archives.
469  *	We assume these headers are not very large (the memory copy we use is
470  *	a bit expensive).
471  * Return:
472  *	0 if buffer was filled ok, -1 o.w. (buffer flush failure)
473  */
474 
475 int
476 wr_rdbuf(char *out, int outcnt)
477 {
478 	int cnt;
479 
480 	/*
481 	 * while there is data to copy copy into the write buffer. when the
482 	 * write buffer fills, flush it to the archive and continue
483 	 */
484 	while (outcnt > 0) {
485 		cnt = bufend - bufpt;
486 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
487 			return(-1);
488 		/*
489 		 * only move what we have space for
490 		 */
491 		cnt = MINIMUM(cnt, outcnt);
492 		memcpy(bufpt, out, cnt);
493 		bufpt += cnt;
494 		out += cnt;
495 		outcnt -= cnt;
496 	}
497 	return(0);
498 }
499 
500 /*
501  * rd_wrbuf()
502  *	copy from the read buffer into a supplied buffer a specified number of
503  *	bytes. If the read buffer is empty fill it and continue to copy.
504  *	usually used to obtain a file header for processing by a format
505  *	specific read routine.
506  * Return
507  *	number of bytes copied to the buffer, 0 indicates EOF on archive volume,
508  *	-1 is a read error
509  */
510 
511 int
512 rd_wrbuf(char *in, int cpcnt)
513 {
514 	int res;
515 	int cnt;
516 	int incnt = cpcnt;
517 
518 	/*
519 	 * loop until we fill the buffer with the requested number of bytes
520 	 */
521 	while (incnt > 0) {
522 		cnt = bufend - bufpt;
523 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
524 			/*
525 			 * read error, return what we got (or the error if
526 			 * no data was copied). The caller must know that an
527 			 * error occurred and has the best knowledge what to
528 			 * do with it
529 			 */
530 			if ((res = cpcnt - incnt) > 0)
531 				return(res);
532 			return(cnt);
533 		}
534 
535 		/*
536 		 * calculate how much data to copy based on whats left and
537 		 * state of buffer
538 		 */
539 		cnt = MINIMUM(cnt, incnt);
540 		memcpy(in, bufpt, cnt);
541 		bufpt += cnt;
542 		incnt -= cnt;
543 		in += cnt;
544 	}
545 	return(cpcnt);
546 }
547 
548 /*
549  * wr_skip()
550  *	skip forward during a write. In other words add padding to the file.
551  *	we add zero filled padding as it makes flawed archives much easier to
552  *	recover from. the caller tells us how many bytes of padding to add
553  *	This routine was not designed to add HUGE amount of padding, just small
554  *	amounts (a few 512 byte blocks at most)
555  * Return:
556  *	0 if ok, -1 if there was a buf_flush failure
557  */
558 
559 int
560 wr_skip(off_t skcnt)
561 {
562 	int cnt;
563 
564 	/*
565 	 * loop while there is more padding to add
566 	 */
567 	while (skcnt > 0) {
568 		cnt = bufend - bufpt;
569 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
570 			return(-1);
571 		cnt = MINIMUM(cnt, skcnt);
572 		memset(bufpt, 0, cnt);
573 		bufpt += cnt;
574 		skcnt -= cnt;
575 	}
576 	return(0);
577 }
578 
579 /*
580  * wr_rdfile()
581  *	fill write buffer with the contents of a file. We are passed an	open
582  *	file descriptor to the file an the archive structure that describes the
583  *	file we are storing. The variable "left" is modified to contain the
584  *	number of bytes of the file we were NOT able to write to the archive.
585  *	it is important that we always write EXACTLY the number of bytes that
586  *	the format specific write routine told us to. The file can also get
587  *	bigger, so reading to the end of file would create an improper archive,
588  *	we just detect this case and warn the user. We never create a bad
589  *	archive if we can avoid it. Of course trying to archive files that are
590  *	active is asking for trouble. It we fail, we pass back how much we
591  *	could NOT copy and let the caller deal with it.
592  * Return:
593  *	0 ok, -1 if archive write failure. a short read of the file returns a
594  *	0, but "left" is set to be greater than zero.
595  */
596 
597 int
598 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
599 {
600 	int cnt;
601 	int res = 0;
602 	off_t size = arcn->sb.st_size;
603 	struct stat sb;
604 
605 	/*
606 	 * while there are more bytes to write
607 	 */
608 	while (size > 0) {
609 		cnt = bufend - bufpt;
610 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
611 			*left = size;
612 			return(-1);
613 		}
614 		cnt = MINIMUM(cnt, size);
615 		if ((res = read(ifd, bufpt, cnt)) <= 0)
616 			break;
617 		size -= res;
618 		bufpt += res;
619 	}
620 
621 	/*
622 	 * better check the file did not change during this operation
623 	 * or the file read failed.
624 	 */
625 	if (res < 0)
626 		syswarn(1, errno, "Read fault on %s", arcn->org_name);
627 	else if (size != 0)
628 		paxwarn(1, "File changed size during read %s", arcn->org_name);
629 	else if (fstat(ifd, &sb) < 0)
630 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
631 	else if (timespeccmp(&arcn->sb.st_mtim, &sb.st_mtim, !=))
632 		paxwarn(1, "File %s was modified during copy to archive",
633 			arcn->org_name);
634 	*left = size;
635 	return(0);
636 }
637 
638 /*
639  * rd_wrfile()
640  *	extract the contents of a file from the archive. If we are unable to
641  *	extract the entire file (due to failure to write the file) we return
642  *	the numbers of bytes we did NOT process. This way the caller knows how
643  *	many bytes to skip past to find the next archive header. If the failure
644  *	was due to an archive read, we will catch that when we try to skip. If
645  *	the format supplies a file data crc value, we calculate the actual crc
646  *	so that it can be compared to the value stored in the header
647  * NOTE:
648  *	We call a special function to write the file. This function attempts to
649  *	restore file holes (blocks of zeros) into the file. When files are
650  *	sparse this saves space, and is a LOT faster. For non sparse files
651  *	the performance hit is small. As of this writing, no archive supports
652  *	information on where the file holes are.
653  * Return:
654  *	0 ok, -1 if archive read failure. if we cannot write the entire file,
655  *	we return a 0 but "left" is set to be the amount unwritten
656  */
657 
658 int
659 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
660 {
661 	int cnt = 0;
662 	off_t size = arcn->sb.st_size;
663 	int res = 0;
664 	char *fnm = arcn->name;
665 	int isem = 1;
666 	int rem;
667 	int sz = MINFBSZ;
668 	struct stat sb;
669 	u_int32_t crc = 0;
670 
671 	/*
672 	 * pass the blocksize of the file being written to the write routine,
673 	 * if the size is zero, use the default MINFBSZ
674 	 */
675 	if (ofd < 0)
676 		sz = PAXPATHLEN + 1;		/* GNU tar long link/file */
677 	else if (fstat(ofd, &sb) == 0) {
678 		if (sb.st_blksize > 0)
679 			sz = (int)sb.st_blksize;
680 	} else
681 		syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
682 	rem = sz;
683 	*left = 0;
684 
685 	/*
686 	 * Copy the archive to the file the number of bytes specified. We have
687 	 * to assume that we want to recover file holes as none of the archive
688 	 * formats can record the location of file holes.
689 	 */
690 	while (size > 0) {
691 		cnt = bufend - bufpt;
692 		/*
693 		 * if we get a read error, we do not want to skip, as we may
694 		 * miss a header, so we do not set left, but if we get a write
695 		 * error, we do want to skip over the unprocessed data.
696 		 */
697 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
698 			break;
699 		cnt = MINIMUM(cnt, size);
700 		if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
701 			*left = size;
702 			break;
703 		}
704 
705 		if (docrc) {
706 			/*
707 			 * update the actual crc value
708 			 */
709 			cnt = res;
710 			while (--cnt >= 0)
711 				crc += *bufpt++ & 0xff;
712 		} else
713 			bufpt += res;
714 		size -= res;
715 	}
716 
717 	/*
718 	 * if the last block has a file hole (all zero), we must make sure this
719 	 * gets updated in the file. We force the last block of zeros to be
720 	 * written. just closing with the file offset moved forward may not put
721 	 * a hole at the end of the file.
722 	 */
723 	if (isem && (arcn->sb.st_size > 0))
724 		file_flush(ofd, fnm, isem);
725 
726 	/*
727 	 * if we failed from archive read, we do not want to skip
728 	 */
729 	if ((size > 0) && (*left == 0))
730 		return(-1);
731 
732 	/*
733 	 * some formats record a crc on file data. If so, then we compare the
734 	 * calculated crc to the crc stored in the archive
735 	 */
736 	if (docrc && (size == 0) && (arcn->crc != crc))
737 		paxwarn(1,"Actual crc does not match expected crc %s",arcn->name);
738 	return(0);
739 }
740 
741 /*
742  * cp_file()
743  *	copy the contents of one file to another. used during -rw phase of pax
744  *	just as in rd_wrfile() we use a special write function to write the
745  *	destination file so we can properly copy files with holes.
746  */
747 
748 void
749 cp_file(ARCHD *arcn, int fd1, int fd2)
750 {
751 	int cnt;
752 	off_t cpcnt = 0;
753 	int res = 0;
754 	char *fnm = arcn->name;
755 	int no_hole = 0;
756 	int isem = 1;
757 	int rem;
758 	int sz = MINFBSZ;
759 	struct stat sb;
760 
761 	/*
762 	 * check for holes in the source file. If none, we will use regular
763 	 * write instead of file write.
764 	 */
765 	 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
766 		++no_hole;
767 
768 	/*
769 	 * pass the blocksize of the file being written to the write routine,
770 	 * if the size is zero, use the default MINFBSZ
771 	 */
772 	if (fstat(fd2, &sb) == 0) {
773 		if (sb.st_blksize > 0)
774 			sz = sb.st_blksize;
775 	} else
776 		syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
777 	rem = sz;
778 
779 	/*
780 	 * read the source file and copy to destination file until EOF
781 	 */
782 	for (;;) {
783 		if ((cnt = read(fd1, buf, blksz)) <= 0)
784 			break;
785 		if (no_hole)
786 			res = write(fd2, buf, cnt);
787 		else
788 			res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
789 		if (res != cnt)
790 			break;
791 		cpcnt += cnt;
792 	}
793 
794 	/*
795 	 * check to make sure the copy is valid.
796 	 */
797 	if (res < 0)
798 		syswarn(1, errno, "Failed write during copy of %s to %s",
799 			arcn->org_name, arcn->name);
800 	else if (cpcnt != arcn->sb.st_size)
801 		paxwarn(1, "File %s changed size during copy to %s",
802 			arcn->org_name, arcn->name);
803 	else if (fstat(fd1, &sb) < 0)
804 		syswarn(1, errno, "Failed stat of %s", arcn->org_name);
805 	else if (timespeccmp(&arcn->sb.st_mtim, &sb.st_mtim, !=))
806 		paxwarn(1, "File %s was modified during copy to %s",
807 			arcn->org_name, arcn->name);
808 
809 	/*
810 	 * if the last block has a file hole (all zero), we must make sure this
811 	 * gets updated in the file. We force the last block of zeros to be
812 	 * written. just closing with the file offset moved forward may not put
813 	 * a hole at the end of the file.
814 	 */
815 	if (!no_hole && isem && (arcn->sb.st_size > 0))
816 		file_flush(fd2, fnm, isem);
817 }
818 
819 /*
820  * buf_fill()
821  *	fill the read buffer with the next record (or what we can get) from
822  *	the archive volume.
823  * Return:
824  *	Number of bytes of data in the read buffer, -1 for read error, and
825  *	0 when finished (user specified termination in ar_next()).
826  */
827 
828 int
829 buf_fill(void)
830 {
831 	int cnt;
832 	static int fini = 0;
833 
834 	if (fini)
835 		return(0);
836 
837 	for (;;) {
838 		/*
839 		 * try to fill the buffer. on error the next archive volume is
840 		 * opened and we try again.
841 		 */
842 		if ((cnt = ar_read(buf, blksz)) > 0) {
843 			bufpt = buf;
844 			bufend = buf + cnt;
845 			rdcnt += cnt;
846 			return(cnt);
847 		}
848 
849 		/*
850 		 * errors require resync, EOF goes to next archive
851 		 */
852 		if (cnt < 0)
853 			break;
854 		if (ar_next() < 0) {
855 			fini = 1;
856 			return(0);
857 		}
858 		rdcnt = 0;
859 	}
860 	exit_val = 1;
861 	return(-1);
862 }
863 
864 /*
865  * buf_flush()
866  *	force the write buffer to the archive. We are passed the number of
867  *	bytes in the buffer at the point of the flush. When we change archives
868  *	the record size might change. (either larger or smaller).
869  * Return:
870  *	0 if all is ok, -1 when a write error occurs.
871  */
872 
873 int
874 buf_flush(int bufcnt)
875 {
876 	int cnt;
877 	int push = 0;
878 	int totcnt = 0;
879 
880 	/*
881 	 * if we have reached the user specified byte count for each archive
882 	 * volume, prompt for the next volume. (The non-standard -R flag).
883 	 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
884 	 * at least one record. We always round limit UP to next blocksize.
885 	 */
886 	if ((wrlimit > 0) && (wrcnt > wrlimit)) {
887 		paxwarn(0, "User specified archive volume byte limit reached.");
888 		if (ar_next() < 0) {
889 			wrcnt = 0;
890 			exit_val = 1;
891 			return(-1);
892 		}
893 		wrcnt = 0;
894 
895 		/*
896 		 * The new archive volume might have changed the size of the
897 		 * write blocksize. if so we figure out if we need to write
898 		 * (one or more times), or if there is now free space left in
899 		 * the buffer (it is no longer full). bufcnt has the number of
900 		 * bytes in the buffer, (the blocksize, at the point we were
901 		 * CALLED). Push has the amount of "extra" data in the buffer
902 		 * if the block size has shrunk from a volume change.
903 		 */
904 		bufend = buf + blksz;
905 		if (blksz > bufcnt)
906 			return(0);
907 		if (blksz < bufcnt)
908 			push = bufcnt - blksz;
909 	}
910 
911 	/*
912 	 * We have enough data to write at least one archive block
913 	 */
914 	for (;;) {
915 		/*
916 		 * write a block and check if it all went out ok
917 		 */
918 		cnt = ar_write(buf, blksz);
919 		if (cnt == blksz) {
920 			/*
921 			 * the write went ok
922 			 */
923 			wrcnt += cnt;
924 			totcnt += cnt;
925 			if (push > 0) {
926 				/* we have extra data to push to the front.
927 				 * check for more than 1 block of push, and if
928 				 * so we loop back to write again
929 				 */
930 				memcpy(buf, bufend, push);
931 				bufpt = buf + push;
932 				if (push >= blksz) {
933 					push -= blksz;
934 					continue;
935 				}
936 			} else
937 				bufpt = buf;
938 			return(totcnt);
939 		} else if (cnt > 0) {
940 			/*
941 			 * Oh drat we got a partial write!
942 			 * if format does not care about alignment let it go,
943 			 * we warned the user in ar_write().... but this means
944 			 * the last record on this volume violates pax spec....
945 			 */
946 			totcnt += cnt;
947 			wrcnt += cnt;
948 			bufpt = buf + cnt;
949 			cnt = bufcnt - cnt;
950 			memcpy(buf, bufpt, cnt);
951 			bufpt = buf + cnt;
952 			if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
953 				return(totcnt);
954 			break;
955 		}
956 
957 		/*
958 		 * All done, go to next archive
959 		 */
960 		wrcnt = 0;
961 		if (ar_next() < 0)
962 			break;
963 
964 		/*
965 		 * The new archive volume might also have changed the block
966 		 * size. if so, figure out if we have too much or too little
967 		 * data for using the new block size
968 		 */
969 		bufend = buf + blksz;
970 		if (blksz > bufcnt)
971 			return(0);
972 		if (blksz < bufcnt)
973 			push = bufcnt - blksz;
974 	}
975 
976 	/*
977 	 * write failed, stop pax. we must not create a bad archive!
978 	 */
979 	exit_val = 1;
980 	return(-1);
981 }
982