xref: /freebsd/bin/pax/buf_subs.c (revision 8e6b01171e30297084bb0b4457c4183c2746aacc)
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  *	$Id: buf_subs.c,v 1.3 1995/05/30 00:06:53 rgrimes Exp $
38  */
39 
40 #ifndef lint
41 static char sccsid[] = "@(#)buf_subs.c	8.2 (Berkeley) 4/18/94";
42 #endif /* not lint */
43 
44 #include <sys/types.h>
45 #include <sys/time.h>
46 #include <sys/stat.h>
47 #include <sys/param.h>
48 #include <stdio.h>
49 #include <errno.h>
50 #include <unistd.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 #if __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 		warn(1, "Write block size of %d too large, maximium is: %d",
107 			wrblksz, MAXBLK);
108 		return(-1);
109 	}
110 	if (wrblksz % BLKMULT) {
111 		warn(1, "Write block size of %d is not a %d byte multiple",
112 		    wrblksz, BLKMULT);
113 		return(-1);
114 	}
115 
116 	/*
117 	 * we only allow wrblksz to be used with all archive operations
118 	 */
119 	blksz = rdblksz = wrblksz;
120 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
121 		return(-1);
122 	wrcnt = 0;
123 	bufend = buf + wrblksz;
124 	bufpt = buf;
125 	return(0);
126 }
127 
128 /*
129  * rd_start()
130  *	set up buffering system to read an archive
131  * Return:
132  *	0 if ok, -1 otherwise
133  */
134 
135 #if __STDC__
136 int
137 rd_start(void)
138 #else
139 int
140 rd_start()
141 #endif
142 {
143 	/*
144 	 * leave space for the header pushback (see get_arc()). If we are
145 	 * going to append and user specified a write block size, check it
146 	 * right away
147 	 */
148 	buf = &(bufmem[BLKMULT]);
149 	if ((act == APPND) && wrblksz) {
150 		if (wrblksz > MAXBLK) {
151 			warn(1,"Write block size %d too large, maximium is: %d",
152 				wrblksz, MAXBLK);
153 			return(-1);
154 		}
155 		if (wrblksz % BLKMULT) {
156 			warn(1, "Write block size %d is not a %d byte multiple",
157 		    	wrblksz, BLKMULT);
158 			return(-1);
159 		}
160 	}
161 
162 	/*
163 	 * open the archive
164 	 */
165 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
166 		return(-1);
167 	bufend = buf + rdblksz;
168 	bufpt = bufend;
169 	rdcnt = 0;
170 	return(0);
171 }
172 
173 /*
174  * cp_start()
175  *	set up buffer system for copying within the file system
176  */
177 
178 #if __STDC__
179 void
180 cp_start(void)
181 #else
182 void
183 cp_start()
184 #endif
185 {
186 	buf = &(bufmem[BLKMULT]);
187 	rdblksz = blksz = MAXBLK;
188 }
189 
190 /*
191  * appnd_start()
192  *	Set up the buffering system to append new members to an archive that
193  *	was just read. The last block(s) of an archive may contain a format
194  *	specific trailer. To append a new member, this trailer has to be
195  *	removed from the archive. The first byte of the trailer is replaced by
196  *	the start of the header of the first file added to the archive. The
197  *	format specific end read function tells us how many bytes to move
198  *	backwards in the archive to be positioned BEFORE the trailer. Two
199  *	different postions have to be adjusted, the O.S. file offset (e.g. the
200  *	position of the tape head) and the write point within the data we have
201  *	stored in the read (soon to become write) buffer. We may have to move
202  *	back several records (the number depends on the size of the archive
203  *	record and the size of the format trailer) to read up the record where
204  *	the first byte of the trailer is recorded. Trailers may span (and
205  *	overlap) record boundries.
206  *	We first calculate which record has the first byte of the trailer. We
207  *	move the OS file offset back to the start of this record and read it
208  *	up. We set the buffer write pointer to be at this byte (the byte where
209  *	the trailer starts). We then move the OS file pointer back to the
210  *	start of this record so a flush of this buffer will replace the record
211  *	in the archive.
212  *	A major problem is rewriting this last record. For archives stored
213  *	on disk files, this is trival. However, many devices are really picky
214  *	about the conditions under which they will allow a write to occur.
215  *	Often devices restrict the conditions where writes can be made writes,
216  *	so it may not be feasable to append archives stored on all types of
217  *	devices.
218  * Return:
219  *	0 for success, -1 for failure
220  */
221 
222 #if __STDC__
223 int
224 appnd_start(off_t skcnt)
225 #else
226 int
227 appnd_start(skcnt)
228 	off_t skcnt;
229 #endif
230 {
231 	register int res;
232 	off_t cnt;
233 
234 	if (exit_val != 0) {
235 		warn(0, "Cannot append to an archive that may have flaws.");
236 		return(-1);
237 	}
238 	/*
239 	 * if the user did not specify a write blocksize, inherit the size used
240 	 * in the last archive volume read. (If a is set we still use rdblksz
241 	 * until next volume, cannot shift sizes within a single volume).
242 	 */
243 	if (!wrblksz)
244 		wrblksz = blksz = rdblksz;
245 	else
246 		blksz = rdblksz;
247 
248 	/*
249 	 * make sure that this volume allows appends
250 	 */
251 	if (ar_app_ok() < 0)
252 		return(-1);
253 
254 	/*
255 	 * Calculate bytes to move back and move in front of record where we
256 	 * need to start writing from. Remember we have to add in any padding
257 	 * that might be in the buffer after the trailer in the last block. We
258 	 * travel skcnt + padding ROUNDED UP to blksize.
259 	 */
260 	skcnt += bufend - bufpt;
261 	if ((cnt = (skcnt/blksz) * blksz) < skcnt)
262 		cnt += blksz;
263 	if (ar_rev((off_t)cnt) < 0)
264 		goto out;
265 
266 	/*
267 	 * We may have gone too far if there is valid data in the block we are
268 	 * now in front of, read up the block and position the pointer after
269 	 * the valid data.
270 	 */
271 	if ((cnt -= skcnt) > 0) {
272 		/*
273 		 * watch out for stupid tape drives. ar_rev() will set rdblksz
274 		 * to be real physical blocksize so we must loop until we get
275 		 * the old rdblksz (now in blksz). If ar_rev() fouls up the
276 		 * determination of the physical block size, we will fail.
277 		 */
278 		bufpt = buf;
279 		bufend = buf + blksz;
280 		while (bufpt < bufend) {
281 			if ((res = ar_read(bufpt, rdblksz)) <= 0)
282 				goto out;
283 			bufpt += res;
284 		}
285 		if (ar_rev((off_t)(bufpt - buf)) < 0)
286 			goto out;
287 		bufpt = buf + cnt;
288 		bufend = buf + blksz;
289 	} else {
290 		/*
291 		 * buffer is empty
292 		 */
293 		bufend = buf + blksz;
294 		bufpt = buf;
295 	}
296 	rdblksz = blksz;
297 	rdcnt -= skcnt;
298 	wrcnt = 0;
299 
300 	/*
301 	 * At this point we are ready to write. If the device requires special
302 	 * handling to write at a point were previously recorded data resides,
303 	 * that is handled in ar_set_wr(). From now on we operate under normal
304 	 * ARCHIVE mode (write) conditions
305 	 */
306 	if (ar_set_wr() < 0)
307 		return(-1);
308 	act = ARCHIVE;
309 	return(0);
310 
311     out:
312 	warn(1, "Unable to rewrite archive trailer, cannot append.");
313 	return(-1);
314 }
315 
316 /*
317  * rd_sync()
318  *	A read error occurred on this archive volume. Resync the buffer and
319  *	try to reset the device (if possible) so we can continue to read. Keep
320  *	trying to do this until we get a valid read, or we reach the limit on
321  *	consecutive read faults (at which point we give up). The user can
322  *	adjust the read error limit through a command line option.
323  * Returns:
324  *	0 on success, and -1 on failure
325  */
326 
327 #if __STDC__
328 int
329 rd_sync(void)
330 #else
331 int
332 rd_sync()
333 #endif
334 {
335 	register int errcnt = 0;
336 	register int res;
337 
338 	/*
339 	 * if the user says bail out on first fault, we are out of here...
340 	 */
341 	if (maxflt == 0)
342 		return(-1);
343 	if (act == APPND) {
344 		warn(1, "Unable to append when there are archive read errors.");
345 		return(-1);
346 	}
347 
348 	/*
349 	 * poke at device and try to get past media error
350 	 */
351 	if (ar_rdsync() < 0) {
352 		if (ar_next() < 0)
353 			return(-1);
354 		else
355 			rdcnt = 0;
356 	}
357 
358 	for (;;) {
359 		if ((res = ar_read(buf, blksz)) > 0) {
360 			/*
361 			 * All right! got some data, fill that buffer
362 			 */
363 			bufpt = buf;
364 			bufend = buf + res;
365 			rdcnt += res;
366 			return(0);
367 		}
368 
369 		/*
370 		 * Oh well, yet another failed read...
371 		 * if error limit reached, ditch. o.w. poke device to move past
372 		 * bad media and try again. if media is badly damaged, we ask
373 		 * the poor (and upset user at this point) for the next archive
374 		 * volume. remember the goal on reads is to get the most we
375 		 * can extract out of the archive.
376 		 */
377 		if ((maxflt > 0) && (++errcnt > maxflt))
378 			warn(0,"Archive read error limit (%d) reached",maxflt);
379 		else if (ar_rdsync() == 0)
380 			continue;
381 		if (ar_next() < 0)
382 			break;
383 		rdcnt = 0;
384 		errcnt = 0;
385 	}
386 	return(-1);
387 }
388 
389 /*
390  * pback()
391  *	push the data used during the archive id phase back into the I/O
392  *	buffer. This is required as we cannot be sure that the header does NOT
393  *	overlap a block boundry (as in the case we are trying to recover a
394  *	flawed archived). This was not designed to be used for any other
395  *	purpose. (What software engineering, HA!)
396  *	WARNING: do not even THINK of pback greater than BLKMULT, unless the
397  *	pback space is increased.
398  */
399 
400 #if __STDC__
401 void
402 pback(char *pt, int cnt)
403 #else
404 void
405 pback(pt, cnt)
406 	char *pt;
407 	int cnt;
408 #endif
409 {
410 	bufpt -= cnt;
411 	bcopy(pt, bufpt, cnt);
412 	return;
413 }
414 
415 /*
416  * rd_skip()
417  *	skip foward in the archive during a archive read. Used to get quickly
418  *	past file data and padding for files the user did NOT select.
419  * Return:
420  *	0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
421  */
422 
423 #if __STDC__
424 int
425 rd_skip(off_t skcnt)
426 #else
427 int
428 rd_skip(skcnt)
429 	off_t skcnt;
430 #endif
431 {
432 	off_t res;
433 	off_t cnt;
434 	off_t skipped = 0;
435 
436 	/*
437 	 * consume what data we have in the buffer. If we have to move foward
438 	 * whole records, we call the low level skip function to see if we can
439 	 * move within the archive without doing the expensive reads on data we
440 	 * do not want.
441 	 */
442 	if (skcnt == 0)
443 		return(0);
444 	res = MIN((bufend - bufpt), skcnt);
445 	bufpt += res;
446 	skcnt -= res;
447 
448 	/*
449 	 * if skcnt is now 0, then no additional i/o is needed
450 	 */
451 	if (skcnt == 0)
452 		return(0);
453 
454 	/*
455 	 * We have to read more, calculate complete and partial record reads
456 	 * based on rdblksz. we skip over "cnt" complete records
457 	 */
458 	res = skcnt%rdblksz;
459 	cnt = (skcnt/rdblksz) * rdblksz;
460 
461 	/*
462 	 * if the skip fails, we will have to resync. ar_fow will tell us
463 	 * how much it can skip over. We will have to read the rest.
464 	 */
465 	if (ar_fow(cnt, &skipped) < 0)
466 		return(-1);
467 	res += cnt - skipped;
468 	rdcnt += skipped;
469 
470 	/*
471 	 * what is left we have to read (which may be the whole thing if
472 	 * ar_fow() told us the device can only read to skip records);
473 	 */
474 	while (res > 0L) {
475 		cnt = bufend - bufpt;
476 		/*
477 		 * if the read fails, we will have to resync
478 		 */
479 		if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
480 			return(-1);
481 		if (cnt == 0)
482 			return(1);
483 		cnt = MIN(cnt, res);
484 		bufpt += cnt;
485 		res -= cnt;
486 	}
487 	return(0);
488 }
489 
490 /*
491  * wr_fin()
492  *	flush out any data (and pad if required) the last block. We always pad
493  *	with zero (even though we do not have to). Padding with 0 makes it a
494  *	lot easier to recover if the archive is damaged. zero paddding SHOULD
495  *	BE a requirement....
496  */
497 
498 #if __STDC__
499 void
500 wr_fin(void)
501 #else
502 void
503 wr_fin()
504 #endif
505 {
506 	if (bufpt > buf) {
507 		bzero(bufpt, bufend - bufpt);
508 		bufpt = bufend;
509 		(void)buf_flush(blksz);
510 	}
511 }
512 
513 /*
514  * wr_rdbuf()
515  *	fill the write buffer from data passed to it in a buffer (usually used
516  *	by format specific write routines to pass a file header). On failure we
517  *	punt. We do not allow the user to continue to write flawed archives.
518  *	We assume these headers are not very large (the memory copy we use is
519  *	a bit expensive).
520  * Return:
521  *	0 if buffer was filled ok, -1 o.w. (buffer flush failure)
522  */
523 
524 #if __STDC__
525 int
526 wr_rdbuf(register char *out, register int outcnt)
527 #else
528 int
529 wr_rdbuf(out, outcnt)
530 	register char *out;
531 	register int outcnt;
532 #endif
533 {
534 	register int cnt;
535 
536 	/*
537 	 * while there is data to copy copy into the write buffer. when the
538 	 * write buffer fills, flush it to the archive and continue
539 	 */
540 	while (outcnt > 0) {
541 		cnt = bufend - bufpt;
542 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
543 			return(-1);
544 		/*
545 		 * only move what we have space for
546 		 */
547 		cnt = MIN(cnt, outcnt);
548 		bcopy(out, bufpt, cnt);
549 		bufpt += cnt;
550 		out += cnt;
551 		outcnt -= cnt;
552 	}
553 	return(0);
554 }
555 
556 /*
557  * rd_wrbuf()
558  *	copy from the read buffer into a supplied buffer a specified number of
559  *	bytes. If the read buffer is empty fill it and continue to copy.
560  *	usually used to obtain a file header for processing by a format
561  *	specific read routine.
562  * Return
563  *	number of bytes copied to the buffer, 0 indicates EOF on archive volume,
564  *	-1 is a read error
565  */
566 
567 #if __STDC__
568 int
569 rd_wrbuf(register char *in, register int cpcnt)
570 #else
571 int
572 rd_wrbuf(in, cpcnt)
573 	register char *in;
574 	register int cpcnt;
575 #endif
576 {
577 	register int res;
578 	register int cnt;
579 	register int incnt = cpcnt;
580 
581 	/*
582 	 * loop until we fill the buffer with the requested number of bytes
583 	 */
584 	while (incnt > 0) {
585 		cnt = bufend - bufpt;
586 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
587 			/*
588 			 * read error, return what we got (or the error if
589 			 * no data was copied). The caller must know that an
590 			 * error occured and has the best knowledge what to
591 			 * do with it
592 			 */
593 			if ((res = cpcnt - incnt) > 0)
594 				return(res);
595 			return(cnt);
596 		}
597 
598 		/*
599 		 * calculate how much data to copy based on whats left and
600 		 * state of buffer
601 		 */
602 		cnt = MIN(cnt, incnt);
603 		bcopy(bufpt, in, cnt);
604 		bufpt += cnt;
605 		incnt -= cnt;
606 		in += cnt;
607 	}
608 	return(cpcnt);
609 }
610 
611 /*
612  * wr_skip()
613  *	skip foward during a write. In other words add padding to the file.
614  *	we add zero filled padding as it makes flawed archives much easier to
615  *	recover from. the caller tells us how many bytes of padding to add
616  *	This routine was not designed to add HUGE amount of padding, just small
617  *	amounts (a few 512 byte blocks at most)
618  * Return:
619  *	0 if ok, -1 if there was a buf_flush failure
620  */
621 
622 #if __STDC__
623 int
624 wr_skip(off_t skcnt)
625 #else
626 int
627 wr_skip(skcnt)
628 	off_t skcnt;
629 #endif
630 {
631 	register int cnt;
632 
633 	/*
634 	 * loop while there is more padding to add
635 	 */
636 	while (skcnt > 0L) {
637 		cnt = bufend - bufpt;
638 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
639 			return(-1);
640 		cnt = MIN(cnt, skcnt);
641 		bzero(bufpt, cnt);
642 		bufpt += cnt;
643 		skcnt -= cnt;
644 	}
645 	return(0);
646 }
647 
648 /*
649  * wr_rdfile()
650  *	fill write buffer with the contents of a file. We are passed an	open
651  *	file descriptor to the file an the archive structure that describes the
652  *	file we are storing. The variable "left" is modified to contain the
653  *	number of bytes of the file we were NOT able to write to the archive.
654  *	it is important that we always write EXACTLY the number of bytes that
655  *	the format specific write routine told us to. The file can also get
656  *	bigger, so reading to the end of file would create an improper archive,
657  *	we just detect this case and warn the user. We never create a bad
658  *	archive if we can avoid it. Of course trying to archive files that are
659  *	active is asking for trouble. It we fail, we pass back how much we
660  *	could NOT copy and let the caller deal with it.
661  * Return:
662  *	0 ok, -1 if archive write failure. a short read of the file returns a
663  *	0, but "left" is set to be greater than zero.
664  */
665 
666 #if __STDC__
667 int
668 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
669 #else
670 int
671 wr_rdfile(arcn, ifd, left)
672 	ARCHD *arcn;
673 	int ifd;
674 	off_t *left;
675 #endif
676 {
677 	register int cnt;
678 	register int res = 0;
679 	register off_t size = arcn->sb.st_size;
680 	struct stat sb;
681 
682 	/*
683 	 * while there are more bytes to write
684 	 */
685 	while (size > 0L) {
686 		cnt = bufend - bufpt;
687 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
688 			*left = size;
689 			return(-1);
690 		}
691 		cnt = MIN(cnt, size);
692 		if ((res = read(ifd, bufpt, cnt)) <= 0)
693 			break;
694 		size -= res;
695 		bufpt += res;
696 	}
697 
698 	/*
699 	 * better check the file did not change during this operation
700 	 * or the file read failed.
701 	 */
702 	if (res < 0)
703 		syswarn(1, errno, "Read fault on %s", arcn->org_name);
704 	else if (size != 0L)
705 		warn(1, "File changed size during read %s", arcn->org_name);
706 	else if (fstat(ifd, &sb) < 0)
707 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
708 	else if (arcn->sb.st_mtime != sb.st_mtime)
709 		warn(1, "File %s was modified during copy to archive",
710 			arcn->org_name);
711 	*left = size;
712 	return(0);
713 }
714 
715 /*
716  * rd_wrfile()
717  *	extract the contents of a file from the archive. If we are unable to
718  *	extract the entire file (due to failure to write the file) we return
719  *	the numbers of bytes we did NOT process. This way the caller knows how
720  *	many bytes to skip past to find the next archive header. If the failure
721  *	was due to an archive read, we will catch that when we try to skip. If
722  *	the format supplies a file data crc value, we calculate the actual crc
723  *	so that it can be compared to the value stored in the header
724  * NOTE:
725  *	We call a special function to write the file. This function attempts to
726  *	restore file holes (blocks of zeros) into the file. When files are
727  *	sparse this saves space, and is a LOT faster. For non sparse files
728  *	the performance hit is small. As of this writing, no archive supports
729  *	information on where the file holes are.
730  * Return:
731  *	0 ok, -1 if archive read failure. if we cannot write the entire file,
732  *	we return a 0 but "left" is set to be the amount unwritten
733  */
734 
735 #if __STDC__
736 int
737 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
738 #else
739 int
740 rd_wrfile(arcn, ofd, left)
741 	ARCHD *arcn;
742 	int ofd;
743 	off_t *left;
744 #endif
745 {
746 	register int cnt = 0;
747 	register off_t size = arcn->sb.st_size;
748 	register int res = 0;
749 	register char *fnm = arcn->name;
750 	int isem = 1;
751 	int rem;
752 	int sz = MINFBSZ;
753  	struct stat sb;
754 	u_long crc = 0L;
755 
756 	/*
757 	 * pass the blocksize of the file being written to the write routine,
758 	 * if the size is zero, use the default MINFBSZ
759 	 */
760         if (fstat(ofd, &sb) == 0) {
761 		if (sb.st_blksize > 0)
762 			sz = (int)sb.st_blksize;
763         } else
764                 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
765 	rem = sz;
766 	*left = 0L;
767 
768 	/*
769 	 * Copy the archive to the file the number of bytes specified. We have
770 	 * to assume that we want to recover file holes as none of the archive
771 	 * formats can record the location of file holes.
772 	 */
773 	while (size > 0L) {
774 		cnt = bufend - bufpt;
775 		/*
776 		 * if we get a read error, we do not want to skip, as we may
777 		 * miss a header, so we do not set left, but if we get a write
778 		 * error, we do want to skip over the unprocessed data.
779 		 */
780 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
781 			break;
782 		cnt = MIN(cnt, size);
783 		if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
784 			*left = size;
785 			break;
786 		}
787 
788 		if (docrc) {
789 			/*
790 			 * update the actual crc value
791 			 */
792 			cnt = res;
793 			while (--cnt >= 0)
794 				crc += *bufpt++ & 0xff;
795 		} else
796 			bufpt += res;
797 		size -= res;
798 	}
799 
800 	/*
801 	 * if the last block has a file hole (all zero), we must make sure this
802 	 * gets updated in the file. We force the last block of zeros to be
803 	 * written. just closing with the file offset moved foward may not put
804 	 * a hole at the end of the file.
805 	 */
806 	if (isem && (arcn->sb.st_size > 0L))
807 		file_flush(ofd, fnm, isem);
808 
809 	/*
810 	 * if we failed from archive read, we do not want to skip
811 	 */
812 	if ((size > 0L) && (*left == 0L))
813 		return(-1);
814 
815 	/*
816 	 * some formats record a crc on file data. If so, then we compare the
817 	 * calculated crc to the crc stored in the archive
818 	 */
819 	if (docrc && (size == 0L) && (arcn->crc != crc))
820 		warn(1,"Actual crc does not match expected crc %s",arcn->name);
821 	return(0);
822 }
823 
824 /*
825  * cp_file()
826  *	copy the contents of one file to another. used during -rw phase of pax
827  *	just as in rd_wrfile() we use a special write function to write the
828  *	destination file so we can properly copy files with holes.
829  */
830 
831 #if __STDC__
832 void
833 cp_file(ARCHD *arcn, int fd1, int fd2)
834 #else
835 void
836 cp_file(arcn, fd1, fd2)
837 	ARCHD *arcn;
838 	int fd1;
839 	int fd2;
840 #endif
841 {
842 	register int cnt;
843 	register off_t cpcnt = 0L;
844 	register int res = 0;
845 	register char *fnm = arcn->name;
846 	register int no_hole = 0;
847 	int isem = 1;
848 	int rem;
849 	int sz = MINFBSZ;
850 	struct stat sb;
851 
852 	/*
853 	 * check for holes in the source file. If none, we will use regular
854 	 * write instead of file write.
855 	 */
856 	 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
857 		++no_hole;
858 
859 	/*
860 	 * pass the blocksize of the file being written to the write routine,
861 	 * if the size is zero, use the default MINFBSZ
862 	 */
863         if (fstat(fd2, &sb) == 0) {
864 		if (sb.st_blksize > 0)
865 			sz = sb.st_blksize;
866         } else
867                 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
868 	rem = sz;
869 
870 	/*
871 	 * read the source file and copy to destination file until EOF
872 	 */
873 	for(;;) {
874 		if ((cnt = read(fd1, buf, blksz)) <= 0)
875 			break;
876 		if (no_hole)
877 			res = write(fd2, buf, cnt);
878 		else
879 			res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
880 		if (res != cnt)
881 			break;
882 		cpcnt += cnt;
883 	}
884 
885 	/*
886 	 * check to make sure the copy is valid.
887 	 */
888 	if (res < 0)
889 		syswarn(1, errno, "Failed write during copy of %s to %s",
890 			arcn->org_name, arcn->name);
891 	else if (cpcnt != arcn->sb.st_size)
892 		warn(1, "File %s changed size during copy to %s",
893 			arcn->org_name, arcn->name);
894 	else if (fstat(fd1, &sb) < 0)
895 		syswarn(1, errno, "Failed stat of %s", arcn->org_name);
896 	else if (arcn->sb.st_mtime != sb.st_mtime)
897 		warn(1, "File %s was modified during copy to %s",
898 			arcn->org_name, arcn->name);
899 
900 	/*
901 	 * if the last block has a file hole (all zero), we must make sure this
902 	 * gets updated in the file. We force the last block of zeros to be
903 	 * written. just closing with the file offset moved foward may not put
904 	 * a hole at the end of the file.
905 	 */
906 	if (!no_hole && isem && (arcn->sb.st_size > 0L))
907 		file_flush(fd2, fnm, isem);
908 	return;
909 }
910 
911 /*
912  * buf_fill()
913  *	fill the read buffer with the next record (or what we can get) from
914  *	the archive volume.
915  * Return:
916  *	Number of bytes of data in the read buffer, -1 for read error, and
917  *	0 when finished (user specified termination in ar_next()).
918  */
919 
920 #if __STDC__
921 int
922 buf_fill(void)
923 #else
924 int
925 buf_fill()
926 #endif
927 {
928 	register int cnt;
929 	static int fini = 0;
930 
931 	if (fini)
932 		return(0);
933 
934 	for(;;) {
935 		/*
936 		 * try to fill the buffer. on error the next archive volume is
937 		 * opened and we try again.
938 		 */
939 		if ((cnt = ar_read(buf, blksz)) > 0) {
940 			bufpt = buf;
941 			bufend = buf + cnt;
942 			rdcnt += cnt;
943 			return(cnt);
944 		}
945 
946 		/*
947 		 * errors require resync, EOF goes to next archive
948 		 */
949 		if (cnt < 0)
950 			break;
951 		if (ar_next() < 0) {
952 			fini = 1;
953 			return(0);
954 		}
955 		rdcnt = 0;
956 	}
957 	exit_val = 1;
958 	return(-1);
959 }
960 
961 /*
962  * buf_flush()
963  *	force the write buffer to the archive. We are passed the number of
964  *	bytes in the buffer at the point of the flush. When we change archives
965  *	the record size might change. (either larger or smaller).
966  * Return:
967  *	0 if all is ok, -1 when a write error occurs.
968  */
969 
970 #if __STDC__
971 int
972 buf_flush(register int bufcnt)
973 #else
974 int
975 buf_flush(bufcnt)
976 	register int bufcnt;
977 #endif
978 {
979 	register int cnt;
980 	register int push = 0;
981 	register int totcnt = 0;
982 
983 	/*
984 	 * if we have reached the user specified byte count for each archive
985 	 * volume, prompt for the next volume. (The non-standrad -R flag).
986 	 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
987 	 * at least one record. We always round limit UP to next blocksize.
988 	 */
989 	if ((wrlimit > 0) && (wrcnt > wrlimit)) {
990 		warn(0, "User specified archive volume byte limit reached.");
991 		if (ar_next() < 0) {
992 			wrcnt = 0;
993 			exit_val = 1;
994 			return(-1);
995 		}
996 		wrcnt = 0;
997 
998 		/*
999 		 * The new archive volume might have changed the size of the
1000 		 * write blocksize. if so we figure out if we need to write
1001 		 * (one or more times), or if there is now free space left in
1002 		 * the buffer (it is no longer full). bufcnt has the number of
1003 		 * bytes in the buffer, (the blocksize, at the point we were
1004 		 * CALLED). Push has the amount of "extra" data in the buffer
1005 		 * if the block size has shrunk from a volume change.
1006 		 */
1007 		bufend = buf + blksz;
1008 		if (blksz > bufcnt)
1009 			return(0);
1010 		if (blksz < bufcnt)
1011 			push = bufcnt - blksz;
1012 	}
1013 
1014 	/*
1015 	 * We have enough data to write at least one archive block
1016 	 */
1017 	for (;;) {
1018 		/*
1019 		 * write a block and check if it all went out ok
1020 		 */
1021 		cnt = ar_write(buf, blksz);
1022 		if (cnt == blksz) {
1023 			/*
1024 			 * the write went ok
1025 			 */
1026 			wrcnt += cnt;
1027 			totcnt += cnt;
1028 			if (push > 0) {
1029 				/* we have extra data to push to the front.
1030 				 * check for more than 1 block of push, and if
1031 				 * so we loop back to write again
1032 				 */
1033 				bcopy(bufend, buf, push);
1034 				bufpt = buf + push;
1035 				if (push >= blksz) {
1036 					push -= blksz;
1037 					continue;
1038 				}
1039 			} else
1040 				bufpt = buf;
1041 			return(totcnt);
1042 		} else if (cnt > 0) {
1043 			/*
1044 			 * Oh drat we got a partial write!
1045 			 * if format doesnt care about alignment let it go,
1046 			 * we warned the user in ar_write().... but this means
1047 			 * the last record on this volume violates pax spec....
1048 			 */
1049 			totcnt += cnt;
1050 			wrcnt += cnt;
1051 			bufpt = buf + cnt;
1052 			cnt = bufcnt - cnt;
1053 			bcopy(bufpt, buf, cnt);
1054 			bufpt = buf + cnt;
1055 			if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
1056 				return(totcnt);
1057 			break;
1058 		}
1059 
1060 		/*
1061 		 * All done, go to next archive
1062 		 */
1063 		wrcnt = 0;
1064 		if (ar_next() < 0)
1065 			break;
1066 
1067 		/*
1068 		 * The new archive volume might also have changed the block
1069 		 * size. if so, figure out if we have too much or too little
1070 		 * data for using the new block size
1071 		 */
1072 		bufend = buf + blksz;
1073 		if (blksz > bufcnt)
1074 			return(0);
1075 		if (blksz < bufcnt)
1076 			push = bufcnt - blksz;
1077 	}
1078 
1079 	/*
1080 	 * write failed, stop pax. we must not create a bad archive!
1081 	 */
1082 	exit_val = 1;
1083 	return(-1);
1084 }
1085