buf_subs.c revision 1.21
1/*	$NetBSD: buf_subs.c,v 1.21 2003/08/07 09:05:19 agc Exp $	*/
2
3/*-
4 * Copyright (c) 1992, 1993
5 *	The Regents of the University of California.  All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Keith Muller of the University of California, San Diego.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 *    notice, this list of conditions and the following disclaimer in the
17 *    documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 *    may be used to endorse or promote products derived from this software
20 *    without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 */
34
35/*-
36 * Copyright (c) 1992 Keith Muller.
37 *
38 * This code is derived from software contributed to Berkeley by
39 * Keith Muller of the University of California, San Diego.
40 *
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
43 * are met:
44 * 1. Redistributions of source code must retain the above copyright
45 *    notice, this list of conditions and the following disclaimer.
46 * 2. Redistributions in binary form must reproduce the above copyright
47 *    notice, this list of conditions and the following disclaimer in the
48 *    documentation and/or other materials provided with the distribution.
49 * 3. All advertising materials mentioning features or use of this software
50 *    must display the following acknowledgement:
51 *	This product includes software developed by the University of
52 *	California, Berkeley and its contributors.
53 * 4. Neither the name of the University nor the names of its contributors
54 *    may be used to endorse or promote products derived from this software
55 *    without specific prior written permission.
56 *
57 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
58 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
59 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
60 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
61 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
62 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
63 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
64 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
65 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
66 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67 * SUCH DAMAGE.
68 */
69
70#include <sys/cdefs.h>
71#if defined(__RCSID) && !defined(lint)
72#if 0
73static char sccsid[] = "@(#)buf_subs.c	8.2 (Berkeley) 4/18/94";
74#else
75__RCSID("$NetBSD: buf_subs.c,v 1.21 2003/08/07 09:05:19 agc Exp $");
76#endif
77#endif /* not lint */
78
79#include <sys/types.h>
80#include <sys/time.h>
81#include <sys/stat.h>
82#include <sys/param.h>
83#include <stdio.h>
84#include <ctype.h>
85#include <errno.h>
86#include <unistd.h>
87#include <stdlib.h>
88#include <string.h>
89#include "pax.h"
90#include "extern.h"
91
92/*
93 * routines which implement archive and file buffering
94 */
95
96#define MINFBSZ		512		/* default block size for hole detect */
97#define MAXFLT		10		/* default media read error limit */
98
99/*
100 * Need to change bufmem to dynamic allocation when the upper
101 * limit on blocking size is removed (though that will violate pax spec)
102 * MAXBLK define and tests will also need to be updated.
103 */
104static char bufmem[MAXBLK+BLKMULT];	/* i/o buffer + pushback id space */
105static char *buf;			/* normal start of i/o buffer */
106static char *bufend;			/* end or last char in i/o buffer */
107static char *bufpt;			/* read/write point in i/o buffer */
108int blksz = MAXBLK;			/* block input/output size in bytes */
109int wrblksz;				/* user spec output size in bytes */
110int maxflt = MAXFLT;			/* MAX consecutive media errors */
111int rdblksz;				/* first read blksize (tapes only) */
112off_t wrlimit;				/* # of bytes written per archive vol */
113off_t wrcnt;				/* # of bytes written on current vol */
114off_t rdcnt;				/* # of bytes read on current vol */
115
116/*
117 * wr_start()
118 *	set up the buffering system to operate in a write mode
119 * Return:
120 *	0 if ok, -1 if the user specified write block size violates pax spec
121 */
122
123int
124wr_start(void)
125{
126	buf = &(bufmem[BLKMULT]);
127	/*
128	 * Check to make sure the write block size meets pax specs. If the user
129	 * does not specify a blocksize, we use the format default blocksize.
130	 * We must be picky on writes, so we do not allow the user to create an
131	 * archive that might be hard to read elsewhere. If all ok, we then
132	 * open the first archive volume
133	 */
134	if (!wrblksz)
135		wrblksz = frmt->bsz;
136	if (wrblksz > MAXBLK) {
137		tty_warn(1, "Write block size of %d too large, maximum is: %d",
138			wrblksz, MAXBLK);
139		return(-1);
140	}
141	if (wrblksz % BLKMULT) {
142		tty_warn(1, "Write block size of %d is not a %d byte multiple",
143		    wrblksz, BLKMULT);
144		return(-1);
145	}
146
147	/*
148	 * we only allow wrblksz to be used with all archive operations
149	 */
150	blksz = rdblksz = wrblksz;
151	if ((ar_open(arcname) < 0) && (ar_next() < 0))
152		return(-1);
153	wrcnt = 0;
154	bufend = buf + wrblksz;
155	bufpt = buf;
156	return(0);
157}
158
159/*
160 * rd_start()
161 *	set up buffering system to read an archive
162 * Return:
163 *	0 if ok, -1 otherwise
164 */
165
166int
167rd_start(void)
168{
169	/*
170	 * leave space for the header pushback (see get_arc()). If we are
171	 * going to append and user specified a write block size, check it
172	 * right away
173	 */
174	buf = &(bufmem[BLKMULT]);
175	if ((act == APPND) && wrblksz) {
176		if (wrblksz > MAXBLK) {
177			tty_warn(1,
178			    "Write block size %d too large, maximum is: %d",
179			    wrblksz, MAXBLK);
180			return(-1);
181		}
182		if (wrblksz % BLKMULT) {
183			tty_warn(1,
184			    "Write block size %d is not a %d byte multiple",
185			    wrblksz, BLKMULT);
186			return(-1);
187		}
188	}
189
190	/*
191	 * open the archive
192	 */
193	if ((ar_open(arcname) < 0) && (ar_next() < 0))
194		return(-1);
195	bufend = buf + rdblksz;
196	bufpt = bufend;
197	rdcnt = 0;
198	return(0);
199}
200
201/*
202 * cp_start()
203 *	set up buffer system for copying within the file system
204 */
205
206void
207cp_start(void)
208{
209	buf = &(bufmem[BLKMULT]);
210	rdblksz = blksz = MAXBLK;
211}
212
213/*
214 * appnd_start()
215 *	Set up the buffering system to append new members to an archive that
216 *	was just read. The last block(s) of an archive may contain a format
217 *	specific trailer. To append a new member, this trailer has to be
218 *	removed from the archive. The first byte of the trailer is replaced by
219 *	the start of the header of the first file added to the archive. The
220 *	format specific end read function tells us how many bytes to move
221 *	backwards in the archive to be positioned BEFORE the trailer. Two
222 *	different positions have to be adjusted, the O.S. file offset (e.g. the
223 *	position of the tape head) and the write point within the data we have
224 *	stored in the read (soon to become write) buffer. We may have to move
225 *	back several records (the number depends on the size of the archive
226 *	record and the size of the format trailer) to read up the record where
227 *	the first byte of the trailer is recorded. Trailers may span (and
228 *	overlap) record boundaries.
229 *	We first calculate which record has the first byte of the trailer. We
230 *	move the OS file offset back to the start of this record and read it
231 *	up. We set the buffer write pointer to be at this byte (the byte where
232 *	the trailer starts). We then move the OS file pointer back to the
233 *	start of this record so a flush of this buffer will replace the record
234 *	in the archive.
235 *	A major problem is rewriting this last record. For archives stored
236 *	on disk files, this is trivial. However, many devices are really picky
237 *	about the conditions under which they will allow a write to occur.
238 *	Often devices restrict the conditions where writes can be made,
239 *	so it may not be feasable to append archives stored on all types of
240 *	devices.
241 * Return:
242 *	0 for success, -1 for failure
243 */
244
245int
246appnd_start(off_t skcnt)
247{
248	int res;
249	off_t cnt;
250
251	if (exit_val != 0) {
252		tty_warn(0, "Cannot append to an archive that may have flaws.");
253		return(-1);
254	}
255	/*
256	 * if the user did not specify a write blocksize, inherit the size used
257	 * in the last archive volume read. (If a is set we still use rdblksz
258	 * until next volume, cannot shift sizes within a single volume).
259	 */
260	if (!wrblksz)
261		wrblksz = blksz = rdblksz;
262	else
263		blksz = rdblksz;
264
265	/*
266	 * make sure that this volume allows appends
267	 */
268	if (ar_app_ok() < 0)
269		return(-1);
270
271	/*
272	 * Calculate bytes to move back and move in front of record where we
273	 * need to start writing from. Remember we have to add in any padding
274	 * that might be in the buffer after the trailer in the last block. We
275	 * travel skcnt + padding ROUNDED UP to blksize.
276	 */
277	skcnt += bufend - bufpt;
278	if ((cnt = (skcnt/blksz) * blksz) < skcnt)
279		cnt += blksz;
280	if (ar_rev((off_t)cnt) < 0)
281		goto out;
282
283	/*
284	 * We may have gone too far if there is valid data in the block we are
285	 * now in front of, read up the block and position the pointer after
286	 * the valid data.
287	 */
288	if ((cnt -= skcnt) > 0) {
289		/*
290		 * watch out for stupid tape drives. ar_rev() will set rdblksz
291		 * to be real physical blocksize so we must loop until we get
292		 * the old rdblksz (now in blksz). If ar_rev() fouls up the
293		 * determination of the physical block size, we will fail.
294		 */
295		bufpt = buf;
296		bufend = buf + blksz;
297		while (bufpt < bufend) {
298			if ((res = ar_read(bufpt, rdblksz)) <= 0)
299				goto out;
300			bufpt += res;
301		}
302		if (ar_rev((off_t)(bufpt - buf)) < 0)
303			goto out;
304		bufpt = buf + cnt;
305		bufend = buf + blksz;
306	} else {
307		/*
308		 * buffer is empty
309		 */
310		bufend = buf + blksz;
311		bufpt = buf;
312	}
313	rdblksz = blksz;
314	rdcnt -= skcnt;
315	wrcnt = 0;
316
317	/*
318	 * At this point we are ready to write. If the device requires special
319	 * handling to write at a point were previously recorded data resides,
320	 * that is handled in ar_set_wr(). From now on we operate under normal
321	 * ARCHIVE mode (write) conditions
322	 */
323	if (ar_set_wr() < 0)
324		return(-1);
325	act = ARCHIVE;
326	return(0);
327
328    out:
329	tty_warn(1, "Unable to rewrite archive trailer, cannot append.");
330	return(-1);
331}
332
333/*
334 * rd_sync()
335 *	A read error occurred on this archive volume. Resync the buffer and
336 *	try to reset the device (if possible) so we can continue to read. Keep
337 *	trying to do this until we get a valid read, or we reach the limit on
338 *	consecutive read faults (at which point we give up). The user can
339 *	adjust the read error limit through a command line option.
340 * Returns:
341 *	0 on success, and -1 on failure
342 */
343
344int
345rd_sync(void)
346{
347	int errcnt = 0;
348	int res;
349
350	/*
351	 * if the user says bail out on first fault, we are out of here...
352	 */
353	if (maxflt == 0)
354		return(-1);
355	if (act == APPND) {
356		tty_warn(1,
357		    "Unable to append when there are archive read errors.");
358		return(-1);
359	}
360
361	/*
362	 * poke at device and try to get past media error
363	 */
364	if (ar_rdsync() < 0) {
365		if (ar_next() < 0)
366			return(-1);
367		else
368			rdcnt = 0;
369	}
370
371	for (;;) {
372		if ((res = ar_read(buf, blksz)) > 0) {
373			/*
374			 * All right! got some data, fill that buffer
375			 */
376			bufpt = buf;
377			bufend = buf + res;
378			rdcnt += res;
379			return(0);
380		}
381
382		/*
383		 * Oh well, yet another failed read...
384		 * if error limit reached, ditch. otherwise poke device to move past
385		 * bad media and try again. if media is badly damaged, we ask
386		 * the poor (and upset user at this point) for the next archive
387		 * volume. remember the goal on reads is to get the most we
388		 * can extract out of the archive.
389		 */
390		if ((maxflt > 0) && (++errcnt > maxflt))
391			tty_warn(0,
392			    "Archive read error limit (%d) reached",maxflt);
393		else if (ar_rdsync() == 0)
394			continue;
395		if (ar_next() < 0)
396			break;
397		rdcnt = 0;
398		errcnt = 0;
399	}
400	return(-1);
401}
402
403/*
404 * pback()
405 *	push the data used during the archive id phase back into the I/O
406 *	buffer. This is required as we cannot be sure that the header does NOT
407 *	overlap a block boundary (as in the case we are trying to recover a
408 *	flawed archived). This was not designed to be used for any other
409 *	purpose. (What software engineering, HA!)
410 *	WARNING: do not even THINK of pback greater than BLKMULT, unless the
411 *	pback space is increased.
412 */
413
414void
415pback(char *pt, int cnt)
416{
417	bufpt -= cnt;
418	memcpy(bufpt, pt, cnt);
419	return;
420}
421
422/*
423 * rd_skip()
424 *	skip forward in the archive during a archive read. Used to get quickly
425 *	past file data and padding for files the user did NOT select.
426 * Return:
427 *	0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
428 */
429
430int
431rd_skip(off_t skcnt)
432{
433	off_t res;
434	off_t cnt;
435	off_t skipped = 0;
436
437	/*
438	 * consume what data we have in the buffer. If we have to move forward
439	 * whole records, we call the low level skip function to see if we can
440	 * move within the archive without doing the expensive reads on data we
441	 * do not want.
442	 */
443	if (skcnt == 0)
444		return(0);
445	res = MIN((bufend - bufpt), skcnt);
446	bufpt += res;
447	skcnt -= res;
448
449	/*
450	 * if skcnt is now 0, then no additional i/o is needed
451	 */
452	if (skcnt == 0)
453		return(0);
454
455	/*
456	 * We have to read more, calculate complete and partial record reads
457	 * based on rdblksz. we skip over "cnt" complete records
458	 */
459	res = skcnt%rdblksz;
460	cnt = (skcnt/rdblksz) * rdblksz;
461
462	/*
463	 * if the skip fails, we will have to resync. ar_fow will tell us
464	 * how much it can skip over. We will have to read the rest.
465	 */
466	if (ar_fow(cnt, &skipped) < 0)
467		return(-1);
468	res += cnt - skipped;
469	rdcnt += skipped;
470
471	/*
472	 * what is left we have to read (which may be the whole thing if
473	 * ar_fow() told us the device can only read to skip records);
474	 */
475	while (res > 0L) {
476		cnt = bufend - bufpt;
477		/*
478		 * if the read fails, we will have to resync
479		 */
480		if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
481			return(-1);
482		if (cnt == 0)
483			return(1);
484		cnt = MIN(cnt, res);
485		bufpt += cnt;
486		res -= cnt;
487	}
488	return(0);
489}
490
491/*
492 * wr_fin()
493 *	flush out any data (and pad if required) the last block. We always pad
494 *	with zero (even though we do not have to). Padding with 0 makes it a
495 *	lot easier to recover if the archive is damaged. zero paddding SHOULD
496 *	BE a requirement....
497 */
498
499void
500wr_fin(void)
501{
502	if (bufpt > buf) {
503		memset(bufpt, 0, bufend - bufpt);
504		bufpt = bufend;
505		(void)buf_flush(blksz);
506	}
507}
508
509/*
510 * wr_rdbuf()
511 *	fill the write buffer from data passed to it in a buffer (usually used
512 *	by format specific write routines to pass a file header). On failure we
513 *	punt. We do not allow the user to continue to write flawed archives.
514 *	We assume these headers are not very large (the memory copy we use is
515 *	a bit expensive).
516 * Return:
517 *	0 if buffer was filled ok, -1 o.w. (buffer flush failure)
518 */
519
520int
521wr_rdbuf(char *out, int outcnt)
522{
523	int cnt;
524
525	/*
526	 * while there is data to copy copy into the write buffer. when the
527	 * write buffer fills, flush it to the archive and continue
528	 */
529	while (outcnt > 0) {
530		cnt = bufend - bufpt;
531		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
532			return(-1);
533		/*
534		 * only move what we have space for
535		 */
536		cnt = MIN(cnt, outcnt);
537		memcpy(bufpt, out, cnt);
538		bufpt += cnt;
539		out += cnt;
540		outcnt -= cnt;
541	}
542	return(0);
543}
544
545/*
546 * rd_wrbuf()
547 *	copy from the read buffer into a supplied buffer a specified number of
548 *	bytes. If the read buffer is empty fill it and continue to copy.
549 *	usually used to obtain a file header for processing by a format
550 *	specific read routine.
551 * Return
552 *	number of bytes copied to the buffer, 0 indicates EOF on archive volume,
553 *	-1 is a read error
554 */
555
556int
557rd_wrbuf(char *in, int cpcnt)
558{
559	int res;
560	int cnt;
561	int incnt = cpcnt;
562
563	/*
564	 * loop until we fill the buffer with the requested number of bytes
565	 */
566	while (incnt > 0) {
567		cnt = bufend - bufpt;
568		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
569			/*
570			 * read error, return what we got (or the error if
571			 * no data was copied). The caller must know that an
572			 * error occurred and has the best knowledge what to
573			 * do with it
574			 */
575			if ((res = cpcnt - incnt) > 0)
576				return(res);
577			return(cnt);
578		}
579
580		/*
581		 * calculate how much data to copy based on whats left and
582		 * state of buffer
583		 */
584		cnt = MIN(cnt, incnt);
585		memcpy(in, bufpt, cnt);
586		bufpt += cnt;
587		incnt -= cnt;
588		in += cnt;
589	}
590	return(cpcnt);
591}
592
593/*
594 * wr_skip()
595 *	skip forward during a write. In other words add padding to the file.
596 *	we add zero filled padding as it makes flawed archives much easier to
597 *	recover from. the caller tells us how many bytes of padding to add
598 *	This routine was not designed to add HUGE amount of padding, just small
599 *	amounts (a few 512 byte blocks at most)
600 * Return:
601 *	0 if ok, -1 if there was a buf_flush failure
602 */
603
604int
605wr_skip(off_t skcnt)
606{
607	int cnt;
608
609	/*
610	 * loop while there is more padding to add
611	 */
612	while (skcnt > 0L) {
613		cnt = bufend - bufpt;
614		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
615			return(-1);
616		cnt = MIN(cnt, skcnt);
617		memset(bufpt, 0, cnt);
618		bufpt += cnt;
619		skcnt -= cnt;
620	}
621	return(0);
622}
623
624/*
625 * wr_rdfile()
626 *	fill write buffer with the contents of a file. We are passed an	open
627 *	file descriptor to the file an the archive structure that describes the
628 *	file we are storing. The variable "left" is modified to contain the
629 *	number of bytes of the file we were NOT able to write to the archive.
630 *	it is important that we always write EXACTLY the number of bytes that
631 *	the format specific write routine told us to. The file can also get
632 *	bigger, so reading to the end of file would create an improper archive,
633 *	we just detect this case and warn the user. We never create a bad
634 *	archive if we can avoid it. Of course trying to archive files that are
635 *	active is asking for trouble. It we fail, we pass back how much we
636 *	could NOT copy and let the caller deal with it.
637 * Return:
638 *	0 ok, -1 if archive write failure. a short read of the file returns a
639 *	0, but "left" is set to be greater than zero.
640 */
641
642int
643wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
644{
645	int cnt;
646	int res = 0;
647	off_t size = arcn->sb.st_size;
648	struct stat origsb, sb;
649
650	/*
651	 * by default, remember the previously obtained stat information
652	 * (in arcn->sb) for comparing the mtime after reading.
653	 * if Mflag is set, use the actual mtime instead.
654	 */
655	origsb = arcn->sb;
656	if (Mflag && (fstat(ifd, &origsb) < 0))
657		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
658
659	/*
660	 * while there are more bytes to write
661	 */
662	while (size > 0L) {
663		cnt = bufend - bufpt;
664		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
665			*left = size;
666			return(-1);
667		}
668		cnt = MIN(cnt, size);
669		if ((res = read_with_restart(ifd, bufpt, cnt)) <= 0)
670			break;
671		size -= res;
672		bufpt += res;
673	}
674
675	/*
676	 * better check the file did not change during this operation
677	 * or the file read failed.
678	 */
679	if (res < 0)
680		syswarn(1, errno, "Read fault on %s", arcn->org_name);
681	else if (size != 0L)
682		tty_warn(1, "File changed size during read %s", arcn->org_name);
683	else if (fstat(ifd, &sb) < 0)
684		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
685	else if (origsb.st_mtime != sb.st_mtime)
686		tty_warn(1, "File %s was modified during copy to archive",
687			arcn->org_name);
688	*left = size;
689	return(0);
690}
691
692/*
693 * rd_wrfile()
694 *	extract the contents of a file from the archive. If we are unable to
695 *	extract the entire file (due to failure to write the file) we return
696 *	the numbers of bytes we did NOT process. This way the caller knows how
697 *	many bytes to skip past to find the next archive header. If the failure
698 *	was due to an archive read, we will catch that when we try to skip. If
699 *	the format supplies a file data crc value, we calculate the actual crc
700 *	so that it can be compared to the value stored in the header
701 * NOTE:
702 *	We call a special function to write the file. This function attempts to
703 *	restore file holes (blocks of zeros) into the file. When files are
704 *	sparse this saves space, and is a LOT faster. For non sparse files
705 *	the performance hit is small. As of this writing, no archive supports
706 *	information on where the file holes are.
707 * Return:
708 *	0 ok, -1 if archive read failure. if we cannot write the entire file,
709 *	we return a 0 but "left" is set to be the amount unwritten
710 */
711
712int
713rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
714{
715	int cnt = 0;
716	off_t size = arcn->sb.st_size;
717	int res = 0;
718	char *fnm = arcn->name;
719	int isem = 1;
720	int rem;
721	int sz = MINFBSZ;
722	struct stat sb;
723	u_long crc = 0L;
724
725	/*
726	 * pass the blocksize of the file being written to the write routine,
727	 * if the size is zero, use the default MINFBSZ
728	 */
729	if (ofd == -1)
730		sz = PAXPATHLEN+1;
731	else if (fstat(ofd, &sb) == 0) {
732		if (sb.st_blksize > 0)
733			sz = (int)sb.st_blksize;
734	} else
735		syswarn(0, errno,
736		    "Unable to obtain block size for file %s", fnm);
737	rem = sz;
738	*left = 0L;
739
740	/*
741	 * Copy the archive to the file the number of bytes specified. We have
742	 * to assume that we want to recover file holes as none of the archive
743	 * formats can record the location of file holes.
744	 */
745	while (size > 0L) {
746		cnt = bufend - bufpt;
747		/*
748		 * if we get a read error, we do not want to skip, as we may
749		 * miss a header, so we do not set left, but if we get a write
750		 * error, we do want to skip over the unprocessed data.
751		 */
752		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
753			break;
754		cnt = MIN(cnt, size);
755		if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
756			*left = size;
757			break;
758		}
759
760		if (docrc) {
761			/*
762			 * update the actual crc value
763			 */
764			cnt = res;
765			while (--cnt >= 0)
766				crc += *bufpt++ & 0xff;
767		} else
768			bufpt += res;
769		size -= res;
770	}
771
772	/*
773	 * if the last block has a file hole (all zero), we must make sure this
774	 * gets updated in the file. We force the last block of zeros to be
775	 * written. just closing with the file offset moved forward may not put
776	 * a hole at the end of the file.
777	 */
778	if (ofd != -1 && isem && (arcn->sb.st_size > 0L))
779		file_flush(ofd, fnm, isem);
780
781	/*
782	 * if we failed from archive read, we do not want to skip
783	 */
784	if ((size > 0L) && (*left == 0L))
785		return(-1);
786
787	/*
788	 * some formats record a crc on file data. If so, then we compare the
789	 * calculated crc to the crc stored in the archive
790	 */
791	if (docrc && (size == 0L) && (arcn->crc != crc))
792		tty_warn(1,"Actual crc does not match expected crc %s",
793		    arcn->name);
794	return(0);
795}
796
797/*
798 * cp_file()
799 *	copy the contents of one file to another. used during -rw phase of pax
800 *	just as in rd_wrfile() we use a special write function to write the
801 *	destination file so we can properly copy files with holes.
802 */
803
804void
805cp_file(ARCHD *arcn, int fd1, int fd2)
806{
807	int cnt;
808	off_t cpcnt = 0L;
809	int res = 0;
810	char *fnm = arcn->name;
811	int no_hole = 0;
812	int isem = 1;
813	int rem;
814	int sz = MINFBSZ;
815	struct stat sb, origsb;
816
817	/*
818	 * check for holes in the source file. If none, we will use regular
819	 * write instead of file write.
820	 */
821	 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
822		++no_hole;
823
824	/*
825	 * by default, remember the previously obtained stat information
826	 * (in arcn->sb) for comparing the mtime after reading.
827	 * if Mflag is set, use the actual mtime instead.
828	 */
829	origsb = arcn->sb;
830	if (Mflag && (fstat(fd1, &origsb) < 0))
831		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
832
833	/*
834	 * pass the blocksize of the file being written to the write routine,
835	 * if the size is zero, use the default MINFBSZ
836	 */
837	if (fstat(fd2, &sb) == 0) {
838		if (sb.st_blksize > 0)
839			sz = sb.st_blksize;
840	} else
841		syswarn(0, errno,
842		    "Unable to obtain block size for file %s", fnm);
843	rem = sz;
844
845	/*
846	 * read the source file and copy to destination file until EOF
847	 */
848	for(;;) {
849		if ((cnt = read_with_restart(fd1, buf, blksz)) <= 0)
850			break;
851		if (no_hole)
852			res = xwrite(fd2, buf, cnt);
853		else
854			res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
855		if (res != cnt)
856			break;
857		cpcnt += cnt;
858	}
859
860	/*
861	 * check to make sure the copy is valid.
862	 */
863	if (res < 0)
864		syswarn(1, errno, "Failed write during copy of %s to %s",
865			arcn->org_name, arcn->name);
866	else if (cpcnt != arcn->sb.st_size)
867		tty_warn(1, "File %s changed size during copy to %s",
868			arcn->org_name, arcn->name);
869	else if (fstat(fd1, &sb) < 0)
870		syswarn(1, errno, "Failed stat of %s", arcn->org_name);
871	else if (origsb.st_mtime != sb.st_mtime)
872		tty_warn(1, "File %s was modified during copy to %s",
873			arcn->org_name, arcn->name);
874
875	/*
876	 * if the last block has a file hole (all zero), we must make sure this
877	 * gets updated in the file. We force the last block of zeros to be
878	 * written. just closing with the file offset moved forward may not put
879	 * a hole at the end of the file.
880	 */
881	if (!no_hole && isem && (arcn->sb.st_size > 0L))
882		file_flush(fd2, fnm, isem);
883	return;
884}
885
886/*
887 * buf_fill()
888 *	fill the read buffer with the next record (or what we can get) from
889 *	the archive volume.
890 * Return:
891 *	Number of bytes of data in the read buffer, -1 for read error, and
892 *	0 when finished (user specified termination in ar_next()).
893 */
894
895int
896buf_fill(void)
897{
898	int cnt;
899	static int fini = 0;
900
901	if (fini)
902		return(0);
903
904	for(;;) {
905		/*
906		 * try to fill the buffer. on error the next archive volume is
907		 * opened and we try again.
908		 */
909		if ((cnt = ar_read(buf, blksz)) > 0) {
910			bufpt = buf;
911			bufend = buf + cnt;
912			rdcnt += cnt;
913			return(cnt);
914		}
915
916		/*
917		 * errors require resync, EOF goes to next archive
918		 */
919		if (cnt < 0)
920			break;
921		if (ar_next() < 0) {
922			fini = 1;
923			return(0);
924		}
925		rdcnt = 0;
926	}
927	exit_val = 1;
928	return(-1);
929}
930
931/*
932 * buf_flush()
933 *	force the write buffer to the archive. We are passed the number of
934 *	bytes in the buffer at the point of the flush. When we change archives
935 *	the record size might change. (either larger or smaller).
936 * Return:
937 *	0 if all is ok, -1 when a write error occurs.
938 */
939
940int
941buf_flush(int bufcnt)
942{
943	int cnt;
944	int push = 0;
945	int totcnt = 0;
946
947	/*
948	 * if we have reached the user specified byte count for each archive
949	 * volume, prompt for the next volume. (The non-standard -R flag).
950	 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
951	 * at least one record. We always round limit UP to next blocksize.
952	 */
953	if ((wrlimit > 0) && (wrcnt > wrlimit)) {
954		tty_warn(0,
955		    "User specified archive volume byte limit reached.");
956		if (ar_next() < 0) {
957			wrcnt = 0;
958			exit_val = 1;
959			return(-1);
960		}
961		wrcnt = 0;
962
963		/*
964		 * The new archive volume might have changed the size of the
965		 * write blocksize. if so we figure out if we need to write
966		 * (one or more times), or if there is now free space left in
967		 * the buffer (it is no longer full). bufcnt has the number of
968		 * bytes in the buffer, (the blocksize, at the point we were
969		 * CALLED). Push has the amount of "extra" data in the buffer
970		 * if the block size has shrunk from a volume change.
971		 */
972		bufend = buf + blksz;
973		if (blksz > bufcnt)
974			return(0);
975		if (blksz < bufcnt)
976			push = bufcnt - blksz;
977	}
978
979	/*
980	 * We have enough data to write at least one archive block
981	 */
982	for (;;) {
983		/*
984		 * write a block and check if it all went out ok
985		 */
986		cnt = ar_write(buf, blksz);
987		if (cnt == blksz) {
988			/*
989			 * the write went ok
990			 */
991			wrcnt += cnt;
992			totcnt += cnt;
993			if (push > 0) {
994				/* we have extra data to push to the front.
995				 * check for more than 1 block of push, and if
996				 * so we loop back to write again
997				 */
998				memcpy(buf, bufend, push);
999				bufpt = buf + push;
1000				if (push >= blksz) {
1001					push -= blksz;
1002					continue;
1003				}
1004			} else
1005				bufpt = buf;
1006			return(totcnt);
1007		} else if (cnt > 0) {
1008			/*
1009			 * Oh drat we got a partial write!
1010			 * if format doesnt care about alignment let it go,
1011			 * we warned the user in ar_write().... but this means
1012			 * the last record on this volume violates pax spec....
1013			 */
1014			totcnt += cnt;
1015			wrcnt += cnt;
1016			bufpt = buf + cnt;
1017			cnt = bufcnt - cnt;
1018			memcpy(buf, bufpt, cnt);
1019			bufpt = buf + cnt;
1020			if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
1021				return(totcnt);
1022			break;
1023		}
1024
1025		/*
1026		 * All done, go to next archive
1027		 */
1028		wrcnt = 0;
1029		if (ar_next() < 0)
1030			break;
1031
1032		/*
1033		 * The new archive volume might also have changed the block
1034		 * size. if so, figure out if we have too much or too little
1035		 * data for using the new block size
1036		 */
1037		bufend = buf + blksz;
1038		if (blksz > bufcnt)
1039			return(0);
1040		if (blksz < bufcnt)
1041			push = bufcnt - blksz;
1042	}
1043
1044	/*
1045	 * write failed, stop pax. we must not create a bad archive!
1046	 */
1047	exit_val = 1;
1048	return(-1);
1049}
1050