xref: /dragonfly/bin/pax/buf_subs.c (revision 0db87cb7)
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  * @(#)buf_subs.c	8.2 (Berkeley) 4/18/94
34  * $FreeBSD: src/bin/pax/buf_subs.c,v 1.12.2.1 2001/08/01 05:03:11 obrien Exp $
35  * $DragonFly: src/bin/pax/buf_subs.c,v 1.6 2006/09/27 21:58:08 pavalos Exp $
36  */
37 
38 #include <sys/types.h>
39 #include <sys/stat.h>
40 #include <errno.h>
41 #include <unistd.h>
42 #include <stdio.h>
43 #include <stdlib.h>
44 #include <string.h>
45 #include "pax.h"
46 #include "extern.h"
47 
48 /*
49  * routines which implement archive and file buffering
50  */
51 
52 #define MINFBSZ		512		/* default block size for hole detect */
53 #define MAXFLT		10		/* default media read error limit */
54 
55 /*
56  * Need to change bufmem to dynamic allocation when the upper
57  * limit on blocking size is removed (though that will violate pax spec)
58  * MAXBLK define and tests will also need to be updated.
59  */
60 static char bufmem[MAXBLK+BLKMULT];	/* i/o buffer + pushback id space */
61 static char *buf;			/* normal start of i/o buffer */
62 static char *bufend;			/* end or last char in i/o buffer */
63 static char *bufpt;			/* read/write point in i/o buffer */
64 int blksz = MAXBLK;			/* block input/output size in bytes */
65 int wrblksz;				/* user spec output size in bytes */
66 int maxflt = MAXFLT;			/* MAX consecutive media errors */
67 int rdblksz;				/* first read blksize (tapes only) */
68 off_t wrlimit;				/* # of bytes written per archive vol */
69 off_t wrcnt;				/* # of bytes written on current vol */
70 off_t rdcnt;				/* # of bytes read on current vol */
71 
72 /*
73  * wr_start()
74  *	set up the buffering system to operate in a write mode
75  * Return:
76  *	0 if ok, -1 if the user specified write block size violates pax spec
77  */
78 
79 int
80 wr_start(void)
81 {
82 	buf = &(bufmem[BLKMULT]);
83 	/*
84 	 * Check to make sure the write block size meets pax specs. If the user
85 	 * does not specify a blocksize, we use the format default blocksize.
86 	 * We must be picky on writes, so we do not allow the user to create an
87 	 * archive that might be hard to read elsewhere. If all ok, we then
88 	 * open the first archive volume
89 	 */
90 	if (!wrblksz)
91 		wrblksz = frmt->bsz;
92 	if (wrblksz > MAXBLK) {
93 		paxwarn(1, "Write block size of %d too large, maximum is: %d",
94 			wrblksz, MAXBLK);
95 		return(-1);
96 	}
97 	if (wrblksz % BLKMULT) {
98 		paxwarn(1, "Write block size of %d is not a %d byte multiple",
99 		    wrblksz, BLKMULT);
100 		return(-1);
101 	}
102 	if (wrblksz > MAXBLK_POSIX) {
103 		paxwarn(0, "Write block size of %d larger than POSIX max %d, archive may not be portable",
104 			wrblksz, MAXBLK_POSIX);
105 		return(-1);
106 	}
107 
108 	/*
109 	 * we only allow wrblksz to be used with all archive operations
110 	 */
111 	blksz = rdblksz = wrblksz;
112 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
113 		return(-1);
114 	wrcnt = 0;
115 	bufend = buf + wrblksz;
116 	bufpt = buf;
117 	return(0);
118 }
119 
120 /*
121  * rd_start()
122  *	set up buffering system to read an archive
123  * Return:
124  *	0 if ok, -1 otherwise
125  */
126 
127 int
128 rd_start(void)
129 {
130 	/*
131 	 * leave space for the header pushback (see get_arc()). If we are
132 	 * going to append and user specified a write block size, check it
133 	 * right away
134 	 */
135 	buf = &(bufmem[BLKMULT]);
136 	if ((act == APPND) && wrblksz) {
137 		if (wrblksz > MAXBLK) {
138 			paxwarn(1,"Write block size %d too large, maximum is: %d",
139 				wrblksz, MAXBLK);
140 			return(-1);
141 		}
142 		if (wrblksz % BLKMULT) {
143 			paxwarn(1, "Write block size %d is not a %d byte multiple",
144 		    	wrblksz, BLKMULT);
145 			return(-1);
146 		}
147 	}
148 
149 	/*
150 	 * open the archive
151 	 */
152 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
153 		return(-1);
154 	bufend = buf + rdblksz;
155 	bufpt = bufend;
156 	rdcnt = 0;
157 	return(0);
158 }
159 
160 /*
161  * cp_start()
162  *	set up buffer system for copying within the file system
163  */
164 
165 void
166 cp_start(void)
167 {
168 	buf = &(bufmem[BLKMULT]);
169 	rdblksz = blksz = MAXBLK;
170 }
171 
172 /*
173  * appnd_start()
174  *	Set up the buffering system to append new members to an archive that
175  *	was just read. The last block(s) of an archive may contain a format
176  *	specific trailer. To append a new member, this trailer has to be
177  *	removed from the archive. The first byte of the trailer is replaced by
178  *	the start of the header of the first file added to the archive. The
179  *	format specific end read function tells us how many bytes to move
180  *	backwards in the archive to be positioned BEFORE the trailer. Two
181  *	different positions have to be adjusted, the O.S. file offset (e.g. the
182  *	position of the tape head) and the write point within the data we have
183  *	stored in the read (soon to become write) buffer. We may have to move
184  *	back several records (the number depends on the size of the archive
185  *	record and the size of the format trailer) to read up the record where
186  *	the first byte of the trailer is recorded. Trailers may span (and
187  *	overlap) record boundaries.
188  *	We first calculate which record has the first byte of the trailer. We
189  *	move the OS file offset back to the start of this record and read it
190  *	up. We set the buffer write pointer to be at this byte (the byte where
191  *	the trailer starts). We then move the OS file pointer back to the
192  *	start of this record so a flush of this buffer will replace the record
193  *	in the archive.
194  *	A major problem is rewriting this last record. For archives stored
195  *	on disk files, this is trivial. However, many devices are really picky
196  *	about the conditions under which they will allow a write to occur.
197  *	Often devices restrict the conditions where writes can be made,
198  *	so it may not be feasible to append archives stored on all types of
199  *	devices.
200  * Return:
201  *	0 for success, -1 for failure
202  */
203 
204 int
205 appnd_start(off_t skcnt)
206 {
207 	int res;
208 	off_t cnt;
209 
210 	if (exit_val != 0) {
211 		paxwarn(0, "Cannot append to an archive that may have flaws.");
212 		return(-1);
213 	}
214 	/*
215 	 * if the user did not specify a write blocksize, inherit the size used
216 	 * in the last archive volume read. (If a is set we still use rdblksz
217 	 * until next volume, cannot shift sizes within a single volume).
218 	 */
219 	if (!wrblksz)
220 		wrblksz = blksz = rdblksz;
221 	else
222 		blksz = rdblksz;
223 
224 	/*
225 	 * make sure that this volume allows appends
226 	 */
227 	if (ar_app_ok() < 0)
228 		return(-1);
229 
230 	/*
231 	 * Calculate bytes to move back and move in front of record where we
232 	 * need to start writing from. Remember we have to add in any padding
233 	 * that might be in the buffer after the trailer in the last block. We
234 	 * travel skcnt + padding ROUNDED UP to blksize.
235 	 */
236 	skcnt += bufend - bufpt;
237 	if ((cnt = (skcnt/blksz) * blksz) < skcnt)
238 		cnt += blksz;
239 	if (ar_rev((off_t)cnt) < 0)
240 		goto out;
241 
242 	/*
243 	 * We may have gone too far if there is valid data in the block we are
244 	 * now in front of, read up the block and position the pointer after
245 	 * the valid data.
246 	 */
247 	if ((cnt -= skcnt) > 0) {
248 		/*
249 		 * watch out for stupid tape drives. ar_rev() will set rdblksz
250 		 * to be real physical blocksize so we must loop until we get
251 		 * the old rdblksz (now in blksz). If ar_rev() fouls up the
252 		 * determination of the physical block size, we will fail.
253 		 */
254 		bufpt = buf;
255 		bufend = buf + blksz;
256 		while (bufpt < bufend) {
257 			if ((res = ar_read(bufpt, rdblksz)) <= 0)
258 				goto out;
259 			bufpt += res;
260 		}
261 		if (ar_rev((off_t)(bufpt - buf)) < 0)
262 			goto out;
263 		bufpt = buf + cnt;
264 		bufend = buf + blksz;
265 	} else {
266 		/*
267 		 * buffer is empty
268 		 */
269 		bufend = buf + blksz;
270 		bufpt = buf;
271 	}
272 	rdblksz = blksz;
273 	rdcnt -= skcnt;
274 	wrcnt = 0;
275 
276 	/*
277 	 * At this point we are ready to write. If the device requires special
278 	 * handling to write at a point were previously recorded data resides,
279 	 * that is handled in ar_set_wr(). From now on we operate under normal
280 	 * ARCHIVE mode (write) conditions
281 	 */
282 	if (ar_set_wr() < 0)
283 		return(-1);
284 	act = ARCHIVE;
285 	return(0);
286 
287     out:
288 	paxwarn(1, "Unable to rewrite archive trailer, cannot append.");
289 	return(-1);
290 }
291 
292 /*
293  * rd_sync()
294  *	A read error occurred on this archive volume. Resync the buffer and
295  *	try to reset the device (if possible) so we can continue to read. Keep
296  *	trying to do this until we get a valid read, or we reach the limit on
297  *	consecutive read faults (at which point we give up). The user can
298  *	adjust the read error limit through a command line option.
299  * Returns:
300  *	0 on success, and -1 on failure
301  */
302 
303 int
304 rd_sync(void)
305 {
306 	int errcnt = 0;
307 	int res;
308 
309 	/*
310 	 * if the user says bail out on first fault, we are out of here...
311 	 */
312 	if (maxflt == 0)
313 		return(-1);
314 	if (act == APPND) {
315 		paxwarn(1, "Unable to append when there are archive read errors.");
316 		return(-1);
317 	}
318 
319 	/*
320 	 * poke at device and try to get past media error
321 	 */
322 	if (ar_rdsync() < 0) {
323 		if (ar_next() < 0)
324 			return(-1);
325 		else
326 			rdcnt = 0;
327 	}
328 
329 	for (;;) {
330 		if ((res = ar_read(buf, blksz)) > 0) {
331 			/*
332 			 * All right! got some data, fill that buffer
333 			 */
334 			bufpt = buf;
335 			bufend = buf + res;
336 			rdcnt += res;
337 			return(0);
338 		}
339 
340 		/*
341 		 * Oh well, yet another failed read...
342 		 * if error limit reached, ditch. o.w. poke device to move past
343 		 * bad media and try again. if media is badly damaged, we ask
344 		 * the poor (and upset user at this point) for the next archive
345 		 * volume. remember the goal on reads is to get the most we
346 		 * can extract out of the archive.
347 		 */
348 		if ((maxflt > 0) && (++errcnt > maxflt))
349 			paxwarn(0,"Archive read error limit (%d) reached",maxflt);
350 		else if (ar_rdsync() == 0)
351 			continue;
352 		if (ar_next() < 0)
353 			break;
354 		rdcnt = 0;
355 		errcnt = 0;
356 	}
357 	return(-1);
358 }
359 
360 /*
361  * pback()
362  *	push the data used during the archive id phase back into the I/O
363  *	buffer. This is required as we cannot be sure that the header does NOT
364  *	overlap a block boundary (as in the case we are trying to recover a
365  *	flawed archived). This was not designed to be used for any other
366  *	purpose. (What software engineering, HA!)
367  *	WARNING: do not even THINK of pback greater than BLKMULT, unless the
368  *	pback space is increased.
369  */
370 
371 void
372 pback(char *pt, int cnt)
373 {
374 	bufpt -= cnt;
375 	memcpy(bufpt, pt, cnt);
376 	return;
377 }
378 
379 /*
380  * rd_skip()
381  *	skip forward in the archive during a archive read. Used to get quickly
382  *	past file data and padding for files the user did NOT select.
383  * Return:
384  *	0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
385  */
386 
387 int
388 rd_skip(off_t skcnt)
389 {
390 	off_t res;
391 	off_t cnt;
392 	off_t skipped = 0;
393 
394 	/*
395 	 * consume what data we have in the buffer. If we have to move forward
396 	 * whole records, we call the low level skip function to see if we can
397 	 * move within the archive without doing the expensive reads on data we
398 	 * do not want.
399 	 */
400 	if (skcnt == 0)
401 		return(0);
402 	res = MIN((bufend - bufpt), skcnt);
403 	bufpt += res;
404 	skcnt -= res;
405 
406 	/*
407 	 * if skcnt is now 0, then no additional i/o is needed
408 	 */
409 	if (skcnt == 0)
410 		return(0);
411 
412 	/*
413 	 * We have to read more, calculate complete and partial record reads
414 	 * based on rdblksz. we skip over "cnt" complete records
415 	 */
416 	res = skcnt%rdblksz;
417 	cnt = (skcnt/rdblksz) * rdblksz;
418 
419 	/*
420 	 * if the skip fails, we will have to resync. ar_fow will tell us
421 	 * how much it can skip over. We will have to read the rest.
422 	 */
423 	if (ar_fow(cnt, &skipped) < 0)
424 		return(-1);
425 	res += cnt - skipped;
426 	rdcnt += skipped;
427 
428 	/*
429 	 * what is left we have to read (which may be the whole thing if
430 	 * ar_fow() told us the device can only read to skip records);
431 	 */
432 	while (res > 0L) {
433 		cnt = bufend - bufpt;
434 		/*
435 		 * if the read fails, we will have to resync
436 		 */
437 		if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
438 			return(-1);
439 		if (cnt == 0)
440 			return(1);
441 		cnt = MIN(cnt, res);
442 		bufpt += cnt;
443 		res -= cnt;
444 	}
445 	return(0);
446 }
447 
448 /*
449  * wr_fin()
450  *	flush out any data (and pad if required) the last block. We always pad
451  *	with zero (even though we do not have to). Padding with 0 makes it a
452  *	lot easier to recover if the archive is damaged. zero padding SHOULD
453  *	BE a requirement....
454  */
455 
456 void
457 wr_fin(void)
458 {
459 	if (bufpt > buf) {
460 		memset(bufpt, 0, bufend - bufpt);
461 		bufpt = bufend;
462 		buf_flush(blksz);
463 	}
464 }
465 
466 /*
467  * wr_rdbuf()
468  *	fill the write buffer from data passed to it in a buffer (usually used
469  *	by format specific write routines to pass a file header). On failure we
470  *	punt. We do not allow the user to continue to write flawed archives.
471  *	We assume these headers are not very large (the memory copy we use is
472  *	a bit expensive).
473  * Return:
474  *	0 if buffer was filled ok, -1 o.w. (buffer flush failure)
475  */
476 
477 int
478 wr_rdbuf(char *out, int outcnt)
479 {
480 	int cnt;
481 
482 	/*
483 	 * while there is data to copy copy into the write buffer. when the
484 	 * write buffer fills, flush it to the archive and continue
485 	 */
486 	while (outcnt > 0) {
487 		cnt = bufend - bufpt;
488 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
489 			return(-1);
490 		/*
491 		 * only move what we have space for
492 		 */
493 		cnt = MIN(cnt, outcnt);
494 		memcpy(bufpt, out, cnt);
495 		bufpt += cnt;
496 		out += cnt;
497 		outcnt -= cnt;
498 	}
499 	return(0);
500 }
501 
502 /*
503  * rd_wrbuf()
504  *	copy from the read buffer into a supplied buffer a specified number of
505  *	bytes. If the read buffer is empty fill it and continue to copy.
506  *	usually used to obtain a file header for processing by a format
507  *	specific read routine.
508  * Return
509  *	number of bytes copied to the buffer, 0 indicates EOF on archive volume,
510  *	-1 is a read error
511  */
512 
513 int
514 rd_wrbuf(char *in, int cpcnt)
515 {
516 	int res;
517 	int cnt;
518 	int incnt = cpcnt;
519 
520 	/*
521 	 * loop until we fill the buffer with the requested number of bytes
522 	 */
523 	while (incnt > 0) {
524 		cnt = bufend - bufpt;
525 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
526 			/*
527 			 * read error, return what we got (or the error if
528 			 * no data was copied). The caller must know that an
529 			 * error occurred and has the best knowledge what to
530 			 * do with it
531 			 */
532 			if ((res = cpcnt - incnt) > 0)
533 				return(res);
534 			return(cnt);
535 		}
536 
537 		/*
538 		 * calculate how much data to copy based on whats left and
539 		 * state of buffer
540 		 */
541 		cnt = MIN(cnt, incnt);
542 		memcpy(in, bufpt, cnt);
543 		bufpt += cnt;
544 		incnt -= cnt;
545 		in += cnt;
546 	}
547 	return(cpcnt);
548 }
549 
550 /*
551  * wr_skip()
552  *	skip forward during a write. In other words add padding to the file.
553  *	we add zero filled padding as it makes flawed archives much easier to
554  *	recover from. the caller tells us how many bytes of padding to add
555  *	This routine was not designed to add HUGE amount of padding, just small
556  *	amounts (a few 512 byte blocks at most)
557  * Return:
558  *	0 if ok, -1 if there was a buf_flush failure
559  */
560 
561 int
562 wr_skip(off_t skcnt)
563 {
564 	int cnt;
565 
566 	/*
567 	 * loop while there is more padding to add
568 	 */
569 	while (skcnt > 0L) {
570 		cnt = bufend - bufpt;
571 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
572 			return(-1);
573 		cnt = MIN(cnt, skcnt);
574 		memset(bufpt, 0, cnt);
575 		bufpt += cnt;
576 		skcnt -= cnt;
577 	}
578 	return(0);
579 }
580 
581 /*
582  * wr_rdfile()
583  *	fill write buffer with the contents of a file. We are passed an	open
584  *	file descriptor to the file an the archive structure that describes the
585  *	file we are storing. The variable "left" is modified to contain the
586  *	number of bytes of the file we were NOT able to write to the archive.
587  *	it is important that we always write EXACTLY the number of bytes that
588  *	the format specific write routine told us to. The file can also get
589  *	bigger, so reading to the end of file would create an improper archive,
590  *	we just detect this case and warn the user. We never create a bad
591  *	archive if we can avoid it. Of course trying to archive files that are
592  *	active is asking for trouble. It we fail, we pass back how much we
593  *	could NOT copy and let the caller deal with it.
594  * Return:
595  *	0 ok, -1 if archive write failure. a short read of the file returns a
596  *	0, but "left" is set to be greater than zero.
597  */
598 
599 int
600 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
601 {
602 	int cnt;
603 	int res = 0;
604 	off_t size = arcn->sb.st_size;
605 	struct stat sb;
606 
607 	/*
608 	 * while there are more bytes to write
609 	 */
610 	while (size > 0L) {
611 		cnt = bufend - bufpt;
612 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
613 			*left = size;
614 			return(-1);
615 		}
616 		cnt = MIN(cnt, size);
617 		if ((res = read(ifd, bufpt, cnt)) <= 0)
618 			break;
619 		size -= res;
620 		bufpt += res;
621 	}
622 
623 	/*
624 	 * better check the file did not change during this operation
625 	 * or the file read failed.
626 	 */
627 	if (res < 0)
628 		syswarn(1, errno, "Read fault on %s", arcn->org_name);
629 	else if (size != 0L)
630 		paxwarn(1, "File changed size during read %s", arcn->org_name);
631 	else if (fstat(ifd, &sb) < 0)
632 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
633 	else if (arcn->sb.st_mtime != sb.st_mtime)
634 		paxwarn(1, "File %s was modified during copy to archive",
635 			arcn->org_name);
636 	*left = size;
637 	return(0);
638 }
639 
640 /*
641  * rd_wrfile()
642  *	extract the contents of a file from the archive. If we are unable to
643  *	extract the entire file (due to failure to write the file) we return
644  *	the numbers of bytes we did NOT process. This way the caller knows how
645  *	many bytes to skip past to find the next archive header. If the failure
646  *	was due to an archive read, we will catch that when we try to skip. If
647  *	the format supplies a file data crc value, we calculate the actual crc
648  *	so that it can be compared to the value stored in the header
649  * NOTE:
650  *	We call a special function to write the file. This function attempts to
651  *	restore file holes (blocks of zeros) into the file. When files are
652  *	sparse this saves space, and is a LOT faster. For non sparse files
653  *	the performance hit is small. As of this writing, no archive supports
654  *	information on where the file holes are.
655  * Return:
656  *	0 ok, -1 if archive read failure. if we cannot write the entire file,
657  *	we return a 0 but "left" is set to be the amount unwritten
658  */
659 
660 int
661 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
662 {
663 	int cnt = 0;
664 	off_t size = arcn->sb.st_size;
665 	int res = 0;
666 	char *fnm = arcn->name;
667 	int isem = 1;
668 	int rem;
669 	int sz = MINFBSZ;
670  	struct stat sb;
671 	u_long crc = 0L;
672 
673 	/*
674 	 * pass the blocksize of the file being written to the write routine,
675 	 * if the size is zero, use the default MINFBSZ
676 	 */
677 	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 = 0L;
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 > 0L) {
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 = MIN(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 > 0L))
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 > 0L) && (*left == 0L))
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 == 0L) && (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 = 0L;
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 (arcn->sb.st_mtime != sb.st_mtime)
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 > 0L))
816 		file_flush(fd2, fnm, isem);
817 	return;
818 }
819 
820 /*
821  * buf_fill()
822  *	fill the read buffer with the next record (or what we can get) from
823  *	the archive volume.
824  * Return:
825  *	Number of bytes of data in the read buffer, -1 for read error, and
826  *	0 when finished (user specified termination in ar_next()).
827  */
828 
829 int
830 buf_fill(void)
831 {
832 	int cnt;
833 	static int fini = 0;
834 
835 	if (fini)
836 		return(0);
837 
838 	for(;;) {
839 		/*
840 		 * try to fill the buffer. on error the next archive volume is
841 		 * opened and we try again.
842 		 */
843 		if ((cnt = ar_read(buf, blksz)) > 0) {
844 			bufpt = buf;
845 			bufend = buf + cnt;
846 			rdcnt += cnt;
847 			return(cnt);
848 		}
849 
850 		/*
851 		 * errors require resync, EOF goes to next archive
852 		 */
853 		if (cnt < 0)
854 			break;
855 		if (ar_next() < 0) {
856 			fini = 1;
857 			return(0);
858 		}
859 		rdcnt = 0;
860 	}
861 	exit_val = 1;
862 	return(-1);
863 }
864 
865 /*
866  * buf_flush()
867  *	force the write buffer to the archive. We are passed the number of
868  *	bytes in the buffer at the point of the flush. When we change archives
869  *	the record size might change. (either larger or smaller).
870  * Return:
871  *	0 if all is ok, -1 when a write error occurs.
872  */
873 
874 int
875 buf_flush(int bufcnt)
876 {
877 	int cnt;
878 	int push = 0;
879 	int totcnt = 0;
880 
881 	/*
882 	 * if we have reached the user specified byte count for each archive
883 	 * volume, prompt for the next volume. (The non-standard -R flag).
884 	 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
885 	 * at least one record. We always round limit UP to next blocksize.
886 	 */
887 	if ((wrlimit > 0) && (wrcnt > wrlimit)) {
888 		paxwarn(0, "User specified archive volume byte limit reached.");
889 		if (ar_next() < 0) {
890 			wrcnt = 0;
891 			exit_val = 1;
892 			return(-1);
893 		}
894 		wrcnt = 0;
895 
896 		/*
897 		 * The new archive volume might have changed the size of the
898 		 * write blocksize. if so we figure out if we need to write
899 		 * (one or more times), or if there is now free space left in
900 		 * the buffer (it is no longer full). bufcnt has the number of
901 		 * bytes in the buffer, (the blocksize, at the point we were
902 		 * CALLED). Push has the amount of "extra" data in the buffer
903 		 * if the block size has shrunk from a volume change.
904 		 */
905 		bufend = buf + blksz;
906 		if (blksz > bufcnt)
907 			return(0);
908 		if (blksz < bufcnt)
909 			push = bufcnt - blksz;
910 	}
911 
912 	/*
913 	 * We have enough data to write at least one archive block
914 	 */
915 	for (;;) {
916 		/*
917 		 * write a block and check if it all went out ok
918 		 */
919 		cnt = ar_write(buf, blksz);
920 		if (cnt == blksz) {
921 			/*
922 			 * the write went ok
923 			 */
924 			wrcnt += cnt;
925 			totcnt += cnt;
926 			if (push > 0) {
927 				/* we have extra data to push to the front.
928 				 * check for more than 1 block of push, and if
929 				 * so we loop back to write again
930 				 */
931 				memcpy(buf, bufend, push);
932 				bufpt = buf + push;
933 				if (push >= blksz) {
934 					push -= blksz;
935 					continue;
936 				}
937 			} else
938 				bufpt = buf;
939 			return(totcnt);
940 		} else if (cnt > 0) {
941 			/*
942 			 * Oh drat we got a partial write!
943 			 * if format doesnt care about alignment let it go,
944 			 * we warned the user in ar_write().... but this means
945 			 * the last record on this volume violates pax spec....
946 			 */
947 			totcnt += cnt;
948 			wrcnt += cnt;
949 			bufpt = buf + cnt;
950 			cnt = bufcnt - cnt;
951 			memcpy(buf, bufpt, cnt);
952 			bufpt = buf + cnt;
953 			if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
954 				return(totcnt);
955 			break;
956 		}
957 
958 		/*
959 		 * All done, go to next archive
960 		 */
961 		wrcnt = 0;
962 		if (ar_next() < 0)
963 			break;
964 
965 		/*
966 		 * The new archive volume might also have changed the block
967 		 * size. if so, figure out if we have too much or too little
968 		 * data for using the new block size
969 		 */
970 		bufend = buf + blksz;
971 		if (blksz > bufcnt)
972 			return(0);
973 		if (blksz < bufcnt)
974 			push = bufcnt - blksz;
975 	}
976 
977 	/*
978 	 * write failed, stop pax. we must not create a bad archive!
979 	 */
980 	exit_val = 1;
981 	return(-1);
982 }
983