xref: /freebsd/bin/pax/buf_subs.c (revision 5521ff5a4d1929056e7ffc982fac3341ca54df7c)
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 static const char rcsid[] =
43   "$FreeBSD$";
44 #endif /* not lint */
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 #ifdef __STDC__
88 int
89 wr_start(void)
90 #else
91 int
92 wr_start()
93 #endif
94 {
95 	buf = &(bufmem[BLKMULT]);
96 	/*
97 	 * Check to make sure the write block size meets pax specs. If the user
98 	 * does not specify a blocksize, we use the format default blocksize.
99 	 * We must be picky on writes, so we do not allow the user to create an
100 	 * archive that might be hard to read elsewhere. If all ok, we then
101 	 * open the first archive volume
102 	 */
103 	if (!wrblksz)
104 		wrblksz = frmt->bsz;
105 	if (wrblksz > MAXBLK) {
106 		paxwarn(1, "Write block size of %d too large, maximum is: %d",
107 			wrblksz, MAXBLK);
108 		return(-1);
109 	}
110 	if (wrblksz % BLKMULT) {
111 		paxwarn(1, "Write block size of %d is not a %d byte multiple",
112 		    wrblksz, BLKMULT);
113 		return(-1);
114 	}
115 	if (wrblksz > MAXBLK_POSIX) {
116 		paxwarn(0, "Write block size of %d larger than POSIX max %d, archive may not be portable",
117 			wrblksz, MAXBLK_POSIX);
118 		return(-1);
119 	}
120 
121 	/*
122 	 * we only allow wrblksz to be used with all archive operations
123 	 */
124 	blksz = rdblksz = wrblksz;
125 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
126 		return(-1);
127 	wrcnt = 0;
128 	bufend = buf + wrblksz;
129 	bufpt = buf;
130 	return(0);
131 }
132 
133 /*
134  * rd_start()
135  *	set up buffering system to read an archive
136  * Return:
137  *	0 if ok, -1 otherwise
138  */
139 
140 #ifdef __STDC__
141 int
142 rd_start(void)
143 #else
144 int
145 rd_start()
146 #endif
147 {
148 	/*
149 	 * leave space for the header pushback (see get_arc()). If we are
150 	 * going to append and user specified a write block size, check it
151 	 * right away
152 	 */
153 	buf = &(bufmem[BLKMULT]);
154 	if ((act == APPND) && wrblksz) {
155 		if (wrblksz > MAXBLK) {
156 			paxwarn(1,"Write block size %d too large, maximum is: %d",
157 				wrblksz, MAXBLK);
158 			return(-1);
159 		}
160 		if (wrblksz % BLKMULT) {
161 			paxwarn(1, "Write block size %d is not a %d byte multiple",
162 		    	wrblksz, BLKMULT);
163 			return(-1);
164 		}
165 	}
166 
167 	/*
168 	 * open the archive
169 	 */
170 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
171 		return(-1);
172 	bufend = buf + rdblksz;
173 	bufpt = bufend;
174 	rdcnt = 0;
175 	return(0);
176 }
177 
178 /*
179  * cp_start()
180  *	set up buffer system for copying within the file system
181  */
182 
183 #ifdef __STDC__
184 void
185 cp_start(void)
186 #else
187 void
188 cp_start()
189 #endif
190 {
191 	buf = &(bufmem[BLKMULT]);
192 	rdblksz = blksz = MAXBLK;
193 }
194 
195 /*
196  * appnd_start()
197  *	Set up the buffering system to append new members to an archive that
198  *	was just read. The last block(s) of an archive may contain a format
199  *	specific trailer. To append a new member, this trailer has to be
200  *	removed from the archive. The first byte of the trailer is replaced by
201  *	the start of the header of the first file added to the archive. The
202  *	format specific end read function tells us how many bytes to move
203  *	backwards in the archive to be positioned BEFORE the trailer. Two
204  *	different postions have to be adjusted, the O.S. file offset (e.g. the
205  *	position of the tape head) and the write point within the data we have
206  *	stored in the read (soon to become write) buffer. We may have to move
207  *	back several records (the number depends on the size of the archive
208  *	record and the size of the format trailer) to read up the record where
209  *	the first byte of the trailer is recorded. Trailers may span (and
210  *	overlap) record boundries.
211  *	We first calculate which record has the first byte of the trailer. We
212  *	move the OS file offset back to the start of this record and read it
213  *	up. We set the buffer write pointer to be at this byte (the byte where
214  *	the trailer starts). We then move the OS file pointer back to the
215  *	start of this record so a flush of this buffer will replace the record
216  *	in the archive.
217  *	A major problem is rewriting this last record. For archives stored
218  *	on disk files, this is trival. However, many devices are really picky
219  *	about the conditions under which they will allow a write to occur.
220  *	Often devices restrict the conditions where writes can be made writes,
221  *	so it may not be feasable to append archives stored on all types of
222  *	devices.
223  * Return:
224  *	0 for success, -1 for failure
225  */
226 
227 #ifdef __STDC__
228 int
229 appnd_start(off_t skcnt)
230 #else
231 int
232 appnd_start(skcnt)
233 	off_t skcnt;
234 #endif
235 {
236 	register int res;
237 	off_t cnt;
238 
239 	if (exit_val != 0) {
240 		paxwarn(0, "Cannot append to an archive that may have flaws.");
241 		return(-1);
242 	}
243 	/*
244 	 * if the user did not specify a write blocksize, inherit the size used
245 	 * in the last archive volume read. (If a is set we still use rdblksz
246 	 * until next volume, cannot shift sizes within a single volume).
247 	 */
248 	if (!wrblksz)
249 		wrblksz = blksz = rdblksz;
250 	else
251 		blksz = rdblksz;
252 
253 	/*
254 	 * make sure that this volume allows appends
255 	 */
256 	if (ar_app_ok() < 0)
257 		return(-1);
258 
259 	/*
260 	 * Calculate bytes to move back and move in front of record where we
261 	 * need to start writing from. Remember we have to add in any padding
262 	 * that might be in the buffer after the trailer in the last block. We
263 	 * travel skcnt + padding ROUNDED UP to blksize.
264 	 */
265 	skcnt += bufend - bufpt;
266 	if ((cnt = (skcnt/blksz) * blksz) < skcnt)
267 		cnt += blksz;
268 	if (ar_rev((off_t)cnt) < 0)
269 		goto out;
270 
271 	/*
272 	 * We may have gone too far if there is valid data in the block we are
273 	 * now in front of, read up the block and position the pointer after
274 	 * the valid data.
275 	 */
276 	if ((cnt -= skcnt) > 0) {
277 		/*
278 		 * watch out for stupid tape drives. ar_rev() will set rdblksz
279 		 * to be real physical blocksize so we must loop until we get
280 		 * the old rdblksz (now in blksz). If ar_rev() fouls up the
281 		 * determination of the physical block size, we will fail.
282 		 */
283 		bufpt = buf;
284 		bufend = buf + blksz;
285 		while (bufpt < bufend) {
286 			if ((res = ar_read(bufpt, rdblksz)) <= 0)
287 				goto out;
288 			bufpt += res;
289 		}
290 		if (ar_rev((off_t)(bufpt - buf)) < 0)
291 			goto out;
292 		bufpt = buf + cnt;
293 		bufend = buf + blksz;
294 	} else {
295 		/*
296 		 * buffer is empty
297 		 */
298 		bufend = buf + blksz;
299 		bufpt = buf;
300 	}
301 	rdblksz = blksz;
302 	rdcnt -= skcnt;
303 	wrcnt = 0;
304 
305 	/*
306 	 * At this point we are ready to write. If the device requires special
307 	 * handling to write at a point were previously recorded data resides,
308 	 * that is handled in ar_set_wr(). From now on we operate under normal
309 	 * ARCHIVE mode (write) conditions
310 	 */
311 	if (ar_set_wr() < 0)
312 		return(-1);
313 	act = ARCHIVE;
314 	return(0);
315 
316     out:
317 	paxwarn(1, "Unable to rewrite archive trailer, cannot append.");
318 	return(-1);
319 }
320 
321 /*
322  * rd_sync()
323  *	A read error occurred on this archive volume. Resync the buffer and
324  *	try to reset the device (if possible) so we can continue to read. Keep
325  *	trying to do this until we get a valid read, or we reach the limit on
326  *	consecutive read faults (at which point we give up). The user can
327  *	adjust the read error limit through a command line option.
328  * Returns:
329  *	0 on success, and -1 on failure
330  */
331 
332 #ifdef __STDC__
333 int
334 rd_sync(void)
335 #else
336 int
337 rd_sync()
338 #endif
339 {
340 	register int errcnt = 0;
341 	register int res;
342 
343 	/*
344 	 * if the user says bail out on first fault, we are out of here...
345 	 */
346 	if (maxflt == 0)
347 		return(-1);
348 	if (act == APPND) {
349 		paxwarn(1, "Unable to append when there are archive read errors.");
350 		return(-1);
351 	}
352 
353 	/*
354 	 * poke at device and try to get past media error
355 	 */
356 	if (ar_rdsync() < 0) {
357 		if (ar_next() < 0)
358 			return(-1);
359 		else
360 			rdcnt = 0;
361 	}
362 
363 	for (;;) {
364 		if ((res = ar_read(buf, blksz)) > 0) {
365 			/*
366 			 * All right! got some data, fill that buffer
367 			 */
368 			bufpt = buf;
369 			bufend = buf + res;
370 			rdcnt += res;
371 			return(0);
372 		}
373 
374 		/*
375 		 * Oh well, yet another failed read...
376 		 * if error limit reached, ditch. o.w. poke device to move past
377 		 * bad media and try again. if media is badly damaged, we ask
378 		 * the poor (and upset user at this point) for the next archive
379 		 * volume. remember the goal on reads is to get the most we
380 		 * can extract out of the archive.
381 		 */
382 		if ((maxflt > 0) && (++errcnt > maxflt))
383 			paxwarn(0,"Archive read error limit (%d) reached",maxflt);
384 		else if (ar_rdsync() == 0)
385 			continue;
386 		if (ar_next() < 0)
387 			break;
388 		rdcnt = 0;
389 		errcnt = 0;
390 	}
391 	return(-1);
392 }
393 
394 /*
395  * pback()
396  *	push the data used during the archive id phase back into the I/O
397  *	buffer. This is required as we cannot be sure that the header does NOT
398  *	overlap a block boundry (as in the case we are trying to recover a
399  *	flawed archived). This was not designed to be used for any other
400  *	purpose. (What software engineering, HA!)
401  *	WARNING: do not even THINK of pback greater than BLKMULT, unless the
402  *	pback space is increased.
403  */
404 
405 #ifdef __STDC__
406 void
407 pback(char *pt, int cnt)
408 #else
409 void
410 pback(pt, cnt)
411 	char *pt;
412 	int cnt;
413 #endif
414 {
415 	bufpt -= cnt;
416 	memcpy(bufpt, pt, cnt);
417 	return;
418 }
419 
420 /*
421  * rd_skip()
422  *	skip foward in the archive during a archive read. Used to get quickly
423  *	past file data and padding for files the user did NOT select.
424  * Return:
425  *	0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
426  */
427 
428 #ifdef __STDC__
429 int
430 rd_skip(off_t skcnt)
431 #else
432 int
433 rd_skip(skcnt)
434 	off_t skcnt;
435 #endif
436 {
437 	off_t res;
438 	off_t cnt;
439 	off_t skipped = 0;
440 
441 	/*
442 	 * consume what data we have in the buffer. If we have to move foward
443 	 * whole records, we call the low level skip function to see if we can
444 	 * move within the archive without doing the expensive reads on data we
445 	 * do not want.
446 	 */
447 	if (skcnt == 0)
448 		return(0);
449 	res = MIN((bufend - bufpt), skcnt);
450 	bufpt += res;
451 	skcnt -= res;
452 
453 	/*
454 	 * if skcnt is now 0, then no additional i/o is needed
455 	 */
456 	if (skcnt == 0)
457 		return(0);
458 
459 	/*
460 	 * We have to read more, calculate complete and partial record reads
461 	 * based on rdblksz. we skip over "cnt" complete records
462 	 */
463 	res = skcnt%rdblksz;
464 	cnt = (skcnt/rdblksz) * rdblksz;
465 
466 	/*
467 	 * if the skip fails, we will have to resync. ar_fow will tell us
468 	 * how much it can skip over. We will have to read the rest.
469 	 */
470 	if (ar_fow(cnt, &skipped) < 0)
471 		return(-1);
472 	res += cnt - skipped;
473 	rdcnt += skipped;
474 
475 	/*
476 	 * what is left we have to read (which may be the whole thing if
477 	 * ar_fow() told us the device can only read to skip records);
478 	 */
479 	while (res > 0L) {
480 		cnt = bufend - bufpt;
481 		/*
482 		 * if the read fails, we will have to resync
483 		 */
484 		if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
485 			return(-1);
486 		if (cnt == 0)
487 			return(1);
488 		cnt = MIN(cnt, res);
489 		bufpt += cnt;
490 		res -= cnt;
491 	}
492 	return(0);
493 }
494 
495 /*
496  * wr_fin()
497  *	flush out any data (and pad if required) the last block. We always pad
498  *	with zero (even though we do not have to). Padding with 0 makes it a
499  *	lot easier to recover if the archive is damaged. zero paddding SHOULD
500  *	BE a requirement....
501  */
502 
503 #ifdef __STDC__
504 void
505 wr_fin(void)
506 #else
507 void
508 wr_fin()
509 #endif
510 {
511 	if (bufpt > buf) {
512 		memset(bufpt, 0, bufend - bufpt);
513 		bufpt = bufend;
514 		(void)buf_flush(blksz);
515 	}
516 }
517 
518 /*
519  * wr_rdbuf()
520  *	fill the write buffer from data passed to it in a buffer (usually used
521  *	by format specific write routines to pass a file header). On failure we
522  *	punt. We do not allow the user to continue to write flawed archives.
523  *	We assume these headers are not very large (the memory copy we use is
524  *	a bit expensive).
525  * Return:
526  *	0 if buffer was filled ok, -1 o.w. (buffer flush failure)
527  */
528 
529 #ifdef __STDC__
530 int
531 wr_rdbuf(register char *out, register int outcnt)
532 #else
533 int
534 wr_rdbuf(out, outcnt)
535 	register char *out;
536 	register int outcnt;
537 #endif
538 {
539 	register int cnt;
540 
541 	/*
542 	 * while there is data to copy copy into the write buffer. when the
543 	 * write buffer fills, flush it to the archive and continue
544 	 */
545 	while (outcnt > 0) {
546 		cnt = bufend - bufpt;
547 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
548 			return(-1);
549 		/*
550 		 * only move what we have space for
551 		 */
552 		cnt = MIN(cnt, outcnt);
553 		memcpy(bufpt, out, cnt);
554 		bufpt += cnt;
555 		out += cnt;
556 		outcnt -= cnt;
557 	}
558 	return(0);
559 }
560 
561 /*
562  * rd_wrbuf()
563  *	copy from the read buffer into a supplied buffer a specified number of
564  *	bytes. If the read buffer is empty fill it and continue to copy.
565  *	usually used to obtain a file header for processing by a format
566  *	specific read routine.
567  * Return
568  *	number of bytes copied to the buffer, 0 indicates EOF on archive volume,
569  *	-1 is a read error
570  */
571 
572 #ifdef __STDC__
573 int
574 rd_wrbuf(register char *in, register int cpcnt)
575 #else
576 int
577 rd_wrbuf(in, cpcnt)
578 	register char *in;
579 	register int cpcnt;
580 #endif
581 {
582 	register int res;
583 	register int cnt;
584 	register int incnt = cpcnt;
585 
586 	/*
587 	 * loop until we fill the buffer with the requested number of bytes
588 	 */
589 	while (incnt > 0) {
590 		cnt = bufend - bufpt;
591 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
592 			/*
593 			 * read error, return what we got (or the error if
594 			 * no data was copied). The caller must know that an
595 			 * error occured and has the best knowledge what to
596 			 * do with it
597 			 */
598 			if ((res = cpcnt - incnt) > 0)
599 				return(res);
600 			return(cnt);
601 		}
602 
603 		/*
604 		 * calculate how much data to copy based on whats left and
605 		 * state of buffer
606 		 */
607 		cnt = MIN(cnt, incnt);
608 		memcpy(in, bufpt, cnt);
609 		bufpt += cnt;
610 		incnt -= cnt;
611 		in += cnt;
612 	}
613 	return(cpcnt);
614 }
615 
616 /*
617  * wr_skip()
618  *	skip forward during a write. In other words add padding to the file.
619  *	we add zero filled padding as it makes flawed archives much easier to
620  *	recover from. the caller tells us how many bytes of padding to add
621  *	This routine was not designed to add HUGE amount of padding, just small
622  *	amounts (a few 512 byte blocks at most)
623  * Return:
624  *	0 if ok, -1 if there was a buf_flush failure
625  */
626 
627 #ifdef __STDC__
628 int
629 wr_skip(off_t skcnt)
630 #else
631 int
632 wr_skip(skcnt)
633 	off_t skcnt;
634 #endif
635 {
636 	register int cnt;
637 
638 	/*
639 	 * loop while there is more padding to add
640 	 */
641 	while (skcnt > 0L) {
642 		cnt = bufend - bufpt;
643 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
644 			return(-1);
645 		cnt = MIN(cnt, skcnt);
646 		memset(bufpt, 0, cnt);
647 		bufpt += cnt;
648 		skcnt -= cnt;
649 	}
650 	return(0);
651 }
652 
653 /*
654  * wr_rdfile()
655  *	fill write buffer with the contents of a file. We are passed an	open
656  *	file descriptor to the file an the archive structure that describes the
657  *	file we are storing. The variable "left" is modified to contain the
658  *	number of bytes of the file we were NOT able to write to the archive.
659  *	it is important that we always write EXACTLY the number of bytes that
660  *	the format specific write routine told us to. The file can also get
661  *	bigger, so reading to the end of file would create an improper archive,
662  *	we just detect this case and warn the user. We never create a bad
663  *	archive if we can avoid it. Of course trying to archive files that are
664  *	active is asking for trouble. It we fail, we pass back how much we
665  *	could NOT copy and let the caller deal with it.
666  * Return:
667  *	0 ok, -1 if archive write failure. a short read of the file returns a
668  *	0, but "left" is set to be greater than zero.
669  */
670 
671 #ifdef __STDC__
672 int
673 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
674 #else
675 int
676 wr_rdfile(arcn, ifd, left)
677 	ARCHD *arcn;
678 	int ifd;
679 	off_t *left;
680 #endif
681 {
682 	register int cnt;
683 	register int res = 0;
684 	register off_t size = arcn->sb.st_size;
685 	struct stat sb;
686 
687 	/*
688 	 * while there are more bytes to write
689 	 */
690 	while (size > 0L) {
691 		cnt = bufend - bufpt;
692 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
693 			*left = size;
694 			return(-1);
695 		}
696 		cnt = MIN(cnt, size);
697 		if ((res = read(ifd, bufpt, cnt)) <= 0)
698 			break;
699 		size -= res;
700 		bufpt += res;
701 	}
702 
703 	/*
704 	 * better check the file did not change during this operation
705 	 * or the file read failed.
706 	 */
707 	if (res < 0)
708 		syswarn(1, errno, "Read fault on %s", arcn->org_name);
709 	else if (size != 0L)
710 		paxwarn(1, "File changed size during read %s", arcn->org_name);
711 	else if (fstat(ifd, &sb) < 0)
712 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
713 	else if (arcn->sb.st_mtime != sb.st_mtime)
714 		paxwarn(1, "File %s was modified during copy to archive",
715 			arcn->org_name);
716 	*left = size;
717 	return(0);
718 }
719 
720 /*
721  * rd_wrfile()
722  *	extract the contents of a file from the archive. If we are unable to
723  *	extract the entire file (due to failure to write the file) we return
724  *	the numbers of bytes we did NOT process. This way the caller knows how
725  *	many bytes to skip past to find the next archive header. If the failure
726  *	was due to an archive read, we will catch that when we try to skip. If
727  *	the format supplies a file data crc value, we calculate the actual crc
728  *	so that it can be compared to the value stored in the header
729  * NOTE:
730  *	We call a special function to write the file. This function attempts to
731  *	restore file holes (blocks of zeros) into the file. When files are
732  *	sparse this saves space, and is a LOT faster. For non sparse files
733  *	the performance hit is small. As of this writing, no archive supports
734  *	information on where the file holes are.
735  * Return:
736  *	0 ok, -1 if archive read failure. if we cannot write the entire file,
737  *	we return a 0 but "left" is set to be the amount unwritten
738  */
739 
740 #ifdef __STDC__
741 int
742 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
743 #else
744 int
745 rd_wrfile(arcn, ofd, left)
746 	ARCHD *arcn;
747 	int ofd;
748 	off_t *left;
749 #endif
750 {
751 	register int cnt = 0;
752 	register off_t size = arcn->sb.st_size;
753 	register int res = 0;
754 	register char *fnm = arcn->name;
755 	int isem = 1;
756 	int rem;
757 	int sz = MINFBSZ;
758  	struct stat sb;
759 	u_long crc = 0L;
760 
761 	/*
762 	 * pass the blocksize of the file being written to the write routine,
763 	 * if the size is zero, use the default MINFBSZ
764 	 */
765 	if (fstat(ofd, &sb) == 0) {
766 		if (sb.st_blksize > 0)
767 			sz = (int)sb.st_blksize;
768 	} else
769 		syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
770 	rem = sz;
771 	*left = 0L;
772 
773 	/*
774 	 * Copy the archive to the file the number of bytes specified. We have
775 	 * to assume that we want to recover file holes as none of the archive
776 	 * formats can record the location of file holes.
777 	 */
778 	while (size > 0L) {
779 		cnt = bufend - bufpt;
780 		/*
781 		 * if we get a read error, we do not want to skip, as we may
782 		 * miss a header, so we do not set left, but if we get a write
783 		 * error, we do want to skip over the unprocessed data.
784 		 */
785 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
786 			break;
787 		cnt = MIN(cnt, size);
788 		if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
789 			*left = size;
790 			break;
791 		}
792 
793 		if (docrc) {
794 			/*
795 			 * update the actual crc value
796 			 */
797 			cnt = res;
798 			while (--cnt >= 0)
799 				crc += *bufpt++ & 0xff;
800 		} else
801 			bufpt += res;
802 		size -= res;
803 	}
804 
805 	/*
806 	 * if the last block has a file hole (all zero), we must make sure this
807 	 * gets updated in the file. We force the last block of zeros to be
808 	 * written. just closing with the file offset moved forward may not put
809 	 * a hole at the end of the file.
810 	 */
811 	if (isem && (arcn->sb.st_size > 0L))
812 		file_flush(ofd, fnm, isem);
813 
814 	/*
815 	 * if we failed from archive read, we do not want to skip
816 	 */
817 	if ((size > 0L) && (*left == 0L))
818 		return(-1);
819 
820 	/*
821 	 * some formats record a crc on file data. If so, then we compare the
822 	 * calculated crc to the crc stored in the archive
823 	 */
824 	if (docrc && (size == 0L) && (arcn->crc != crc))
825 		paxwarn(1,"Actual crc does not match expected crc %s",arcn->name);
826 	return(0);
827 }
828 
829 /*
830  * cp_file()
831  *	copy the contents of one file to another. used during -rw phase of pax
832  *	just as in rd_wrfile() we use a special write function to write the
833  *	destination file so we can properly copy files with holes.
834  */
835 
836 #ifdef __STDC__
837 void
838 cp_file(ARCHD *arcn, int fd1, int fd2)
839 #else
840 void
841 cp_file(arcn, fd1, fd2)
842 	ARCHD *arcn;
843 	int fd1;
844 	int fd2;
845 #endif
846 {
847 	register int cnt;
848 	register off_t cpcnt = 0L;
849 	register int res = 0;
850 	register char *fnm = arcn->name;
851 	register int no_hole = 0;
852 	int isem = 1;
853 	int rem;
854 	int sz = MINFBSZ;
855 	struct stat sb;
856 
857 	/*
858 	 * check for holes in the source file. If none, we will use regular
859 	 * write instead of file write.
860 	 */
861 	 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
862 		++no_hole;
863 
864 	/*
865 	 * pass the blocksize of the file being written to the write routine,
866 	 * if the size is zero, use the default MINFBSZ
867 	 */
868 	if (fstat(fd2, &sb) == 0) {
869 		if (sb.st_blksize > 0)
870 			sz = sb.st_blksize;
871 	} else
872 		syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
873 	rem = sz;
874 
875 	/*
876 	 * read the source file and copy to destination file until EOF
877 	 */
878 	for(;;) {
879 		if ((cnt = read(fd1, buf, blksz)) <= 0)
880 			break;
881 		if (no_hole)
882 			res = write(fd2, buf, cnt);
883 		else
884 			res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
885 		if (res != cnt)
886 			break;
887 		cpcnt += cnt;
888 	}
889 
890 	/*
891 	 * check to make sure the copy is valid.
892 	 */
893 	if (res < 0)
894 		syswarn(1, errno, "Failed write during copy of %s to %s",
895 			arcn->org_name, arcn->name);
896 	else if (cpcnt != arcn->sb.st_size)
897 		paxwarn(1, "File %s changed size during copy to %s",
898 			arcn->org_name, arcn->name);
899 	else if (fstat(fd1, &sb) < 0)
900 		syswarn(1, errno, "Failed stat of %s", arcn->org_name);
901 	else if (arcn->sb.st_mtime != sb.st_mtime)
902 		paxwarn(1, "File %s was modified during copy to %s",
903 			arcn->org_name, arcn->name);
904 
905 	/*
906 	 * if the last block has a file hole (all zero), we must make sure this
907 	 * gets updated in the file. We force the last block of zeros to be
908 	 * written. just closing with the file offset moved forward may not put
909 	 * a hole at the end of the file.
910 	 */
911 	if (!no_hole && isem && (arcn->sb.st_size > 0L))
912 		file_flush(fd2, fnm, isem);
913 	return;
914 }
915 
916 /*
917  * buf_fill()
918  *	fill the read buffer with the next record (or what we can get) from
919  *	the archive volume.
920  * Return:
921  *	Number of bytes of data in the read buffer, -1 for read error, and
922  *	0 when finished (user specified termination in ar_next()).
923  */
924 
925 #ifdef __STDC__
926 int
927 buf_fill(void)
928 #else
929 int
930 buf_fill()
931 #endif
932 {
933 	register int cnt;
934 	static int fini = 0;
935 
936 	if (fini)
937 		return(0);
938 
939 	for(;;) {
940 		/*
941 		 * try to fill the buffer. on error the next archive volume is
942 		 * opened and we try again.
943 		 */
944 		if ((cnt = ar_read(buf, blksz)) > 0) {
945 			bufpt = buf;
946 			bufend = buf + cnt;
947 			rdcnt += cnt;
948 			return(cnt);
949 		}
950 
951 		/*
952 		 * errors require resync, EOF goes to next archive
953 		 */
954 		if (cnt < 0)
955 			break;
956 		if (ar_next() < 0) {
957 			fini = 1;
958 			return(0);
959 		}
960 		rdcnt = 0;
961 	}
962 	exit_val = 1;
963 	return(-1);
964 }
965 
966 /*
967  * buf_flush()
968  *	force the write buffer to the archive. We are passed the number of
969  *	bytes in the buffer at the point of the flush. When we change archives
970  *	the record size might change. (either larger or smaller).
971  * Return:
972  *	0 if all is ok, -1 when a write error occurs.
973  */
974 
975 #ifdef __STDC__
976 int
977 buf_flush(register int bufcnt)
978 #else
979 int
980 buf_flush(bufcnt)
981 	register int bufcnt;
982 #endif
983 {
984 	register int cnt;
985 	register int push = 0;
986 	register int totcnt = 0;
987 
988 	/*
989 	 * if we have reached the user specified byte count for each archive
990 	 * volume, prompt for the next volume. (The non-standrad -R flag).
991 	 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
992 	 * at least one record. We always round limit UP to next blocksize.
993 	 */
994 	if ((wrlimit > 0) && (wrcnt > wrlimit)) {
995 		paxwarn(0, "User specified archive volume byte limit reached.");
996 		if (ar_next() < 0) {
997 			wrcnt = 0;
998 			exit_val = 1;
999 			return(-1);
1000 		}
1001 		wrcnt = 0;
1002 
1003 		/*
1004 		 * The new archive volume might have changed the size of the
1005 		 * write blocksize. if so we figure out if we need to write
1006 		 * (one or more times), or if there is now free space left in
1007 		 * the buffer (it is no longer full). bufcnt has the number of
1008 		 * bytes in the buffer, (the blocksize, at the point we were
1009 		 * CALLED). Push has the amount of "extra" data in the buffer
1010 		 * if the block size has shrunk from a volume change.
1011 		 */
1012 		bufend = buf + blksz;
1013 		if (blksz > bufcnt)
1014 			return(0);
1015 		if (blksz < bufcnt)
1016 			push = bufcnt - blksz;
1017 	}
1018 
1019 	/*
1020 	 * We have enough data to write at least one archive block
1021 	 */
1022 	for (;;) {
1023 		/*
1024 		 * write a block and check if it all went out ok
1025 		 */
1026 		cnt = ar_write(buf, blksz);
1027 		if (cnt == blksz) {
1028 			/*
1029 			 * the write went ok
1030 			 */
1031 			wrcnt += cnt;
1032 			totcnt += cnt;
1033 			if (push > 0) {
1034 				/* we have extra data to push to the front.
1035 				 * check for more than 1 block of push, and if
1036 				 * so we loop back to write again
1037 				 */
1038 				memcpy(buf, bufend, push);
1039 				bufpt = buf + push;
1040 				if (push >= blksz) {
1041 					push -= blksz;
1042 					continue;
1043 				}
1044 			} else
1045 				bufpt = buf;
1046 			return(totcnt);
1047 		} else if (cnt > 0) {
1048 			/*
1049 			 * Oh drat we got a partial write!
1050 			 * if format doesnt care about alignment let it go,
1051 			 * we warned the user in ar_write().... but this means
1052 			 * the last record on this volume violates pax spec....
1053 			 */
1054 			totcnt += cnt;
1055 			wrcnt += cnt;
1056 			bufpt = buf + cnt;
1057 			cnt = bufcnt - cnt;
1058 			memcpy(buf, bufpt, cnt);
1059 			bufpt = buf + cnt;
1060 			if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
1061 				return(totcnt);
1062 			break;
1063 		}
1064 
1065 		/*
1066 		 * All done, go to next archive
1067 		 */
1068 		wrcnt = 0;
1069 		if (ar_next() < 0)
1070 			break;
1071 
1072 		/*
1073 		 * The new archive volume might also have changed the block
1074 		 * size. if so, figure out if we have too much or too little
1075 		 * data for using the new block size
1076 		 */
1077 		bufend = buf + blksz;
1078 		if (blksz > bufcnt)
1079 			return(0);
1080 		if (blksz < bufcnt)
1081 			push = bufcnt - blksz;
1082 	}
1083 
1084 	/*
1085 	 * write failed, stop pax. we must not create a bad archive!
1086 	 */
1087 	exit_val = 1;
1088 	return(-1);
1089 }
1090