Home | History | Annotate | Line # | Download | only in pax
buf_subs.c revision 1.19
      1 /*	$NetBSD: buf_subs.c,v 1.19 2003/02/09 18:27:10 grant Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1992 Keith Muller.
      5  * Copyright (c) 1992, 1993
      6  *	The Regents of the University of California.  All rights reserved.
      7  *
      8  * This code is derived from software contributed to Berkeley by
      9  * Keith Muller of the University of California, San Diego.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *	This product includes software developed by the University of
     22  *	California, Berkeley and its contributors.
     23  * 4. Neither the name of the University nor the names of its contributors
     24  *    may be used to endorse or promote products derived from this software
     25  *    without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     37  * SUCH DAMAGE.
     38  */
     39 
     40 #include <sys/cdefs.h>
     41 #if defined(__RCSID) && !defined(lint)
     42 #if 0
     43 static char sccsid[] = "@(#)buf_subs.c	8.2 (Berkeley) 4/18/94";
     44 #else
     45 __RCSID("$NetBSD: buf_subs.c,v 1.19 2003/02/09 18:27:10 grant Exp $");
     46 #endif
     47 #endif /* not lint */
     48 
     49 #include <sys/types.h>
     50 #include <sys/time.h>
     51 #include <sys/stat.h>
     52 #include <sys/param.h>
     53 #include <stdio.h>
     54 #include <ctype.h>
     55 #include <errno.h>
     56 #include <unistd.h>
     57 #include <stdlib.h>
     58 #include <string.h>
     59 #include "pax.h"
     60 #include "extern.h"
     61 
     62 /*
     63  * routines which implement archive and file buffering
     64  */
     65 
     66 #define MINFBSZ		512		/* default block size for hole detect */
     67 #define MAXFLT		10		/* default media read error limit */
     68 
     69 /*
     70  * Need to change bufmem to dynamic allocation when the upper
     71  * limit on blocking size is removed (though that will violate pax spec)
     72  * MAXBLK define and tests will also need to be updated.
     73  */
     74 static char bufmem[MAXBLK+BLKMULT];	/* i/o buffer + pushback id space */
     75 static char *buf;			/* normal start of i/o buffer */
     76 static char *bufend;			/* end or last char in i/o buffer */
     77 static char *bufpt;			/* read/write point in i/o buffer */
     78 int blksz = MAXBLK;			/* block input/output size in bytes */
     79 int wrblksz;				/* user spec output size in bytes */
     80 int maxflt = MAXFLT;			/* MAX consecutive media errors */
     81 int rdblksz;				/* first read blksize (tapes only) */
     82 off_t wrlimit;				/* # of bytes written per archive vol */
     83 off_t wrcnt;				/* # of bytes written on current vol */
     84 off_t rdcnt;				/* # of bytes read on current vol */
     85 
     86 /*
     87  * wr_start()
     88  *	set up the buffering system to operate in a write mode
     89  * Return:
     90  *	0 if ok, -1 if the user specified write block size violates pax spec
     91  */
     92 
     93 int
     94 wr_start(void)
     95 {
     96 	buf = &(bufmem[BLKMULT]);
     97 	/*
     98 	 * Check to make sure the write block size meets pax specs. If the user
     99 	 * does not specify a blocksize, we use the format default blocksize.
    100 	 * We must be picky on writes, so we do not allow the user to create an
    101 	 * archive that might be hard to read elsewhere. If all ok, we then
    102 	 * open the first archive volume
    103 	 */
    104 	if (!wrblksz)
    105 		wrblksz = frmt->bsz;
    106 	if (wrblksz > MAXBLK) {
    107 		tty_warn(1, "Write block size of %d too large, maximum is: %d",
    108 			wrblksz, MAXBLK);
    109 		return(-1);
    110 	}
    111 	if (wrblksz % BLKMULT) {
    112 		tty_warn(1, "Write block size of %d is not a %d byte multiple",
    113 		    wrblksz, BLKMULT);
    114 		return(-1);
    115 	}
    116 
    117 	/*
    118 	 * we only allow wrblksz to be used with all archive operations
    119 	 */
    120 	blksz = rdblksz = wrblksz;
    121 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
    122 		return(-1);
    123 	wrcnt = 0;
    124 	bufend = buf + wrblksz;
    125 	bufpt = buf;
    126 	return(0);
    127 }
    128 
    129 /*
    130  * rd_start()
    131  *	set up buffering system to read an archive
    132  * Return:
    133  *	0 if ok, -1 otherwise
    134  */
    135 
    136 int
    137 rd_start(void)
    138 {
    139 	/*
    140 	 * leave space for the header pushback (see get_arc()). If we are
    141 	 * going to append and user specified a write block size, check it
    142 	 * right away
    143 	 */
    144 	buf = &(bufmem[BLKMULT]);
    145 	if ((act == APPND) && wrblksz) {
    146 		if (wrblksz > MAXBLK) {
    147 			tty_warn(1,
    148 			    "Write block size %d too large, maximum is: %d",
    149 			    wrblksz, MAXBLK);
    150 			return(-1);
    151 		}
    152 		if (wrblksz % BLKMULT) {
    153 			tty_warn(1,
    154 			    "Write block size %d is not a %d byte multiple",
    155 			    wrblksz, BLKMULT);
    156 			return(-1);
    157 		}
    158 	}
    159 
    160 	/*
    161 	 * open the archive
    162 	 */
    163 	if ((ar_open(arcname) < 0) && (ar_next() < 0))
    164 		return(-1);
    165 	bufend = buf + rdblksz;
    166 	bufpt = bufend;
    167 	rdcnt = 0;
    168 	return(0);
    169 }
    170 
    171 /*
    172  * cp_start()
    173  *	set up buffer system for copying within the file system
    174  */
    175 
    176 void
    177 cp_start(void)
    178 {
    179 	buf = &(bufmem[BLKMULT]);
    180 	rdblksz = blksz = MAXBLK;
    181 }
    182 
    183 /*
    184  * appnd_start()
    185  *	Set up the buffering system to append new members to an archive that
    186  *	was just read. The last block(s) of an archive may contain a format
    187  *	specific trailer. To append a new member, this trailer has to be
    188  *	removed from the archive. The first byte of the trailer is replaced by
    189  *	the start of the header of the first file added to the archive. The
    190  *	format specific end read function tells us how many bytes to move
    191  *	backwards in the archive to be positioned BEFORE the trailer. Two
    192  *	different positions have to be adjusted, the O.S. file offset (e.g. the
    193  *	position of the tape head) and the write point within the data we have
    194  *	stored in the read (soon to become write) buffer. We may have to move
    195  *	back several records (the number depends on the size of the archive
    196  *	record and the size of the format trailer) to read up the record where
    197  *	the first byte of the trailer is recorded. Trailers may span (and
    198  *	overlap) record boundaries.
    199  *	We first calculate which record has the first byte of the trailer. We
    200  *	move the OS file offset back to the start of this record and read it
    201  *	up. We set the buffer write pointer to be at this byte (the byte where
    202  *	the trailer starts). We then move the OS file pointer back to the
    203  *	start of this record so a flush of this buffer will replace the record
    204  *	in the archive.
    205  *	A major problem is rewriting this last record. For archives stored
    206  *	on disk files, this is trivial. However, many devices are really picky
    207  *	about the conditions under which they will allow a write to occur.
    208  *	Often devices restrict the conditions where writes can be made,
    209  *	so it may not be feasable to append archives stored on all types of
    210  *	devices.
    211  * Return:
    212  *	0 for success, -1 for failure
    213  */
    214 
    215 int
    216 appnd_start(off_t skcnt)
    217 {
    218 	int res;
    219 	off_t cnt;
    220 
    221 	if (exit_val != 0) {
    222 		tty_warn(0, "Cannot append to an archive that may have flaws.");
    223 		return(-1);
    224 	}
    225 	/*
    226 	 * if the user did not specify a write blocksize, inherit the size used
    227 	 * in the last archive volume read. (If a is set we still use rdblksz
    228 	 * until next volume, cannot shift sizes within a single volume).
    229 	 */
    230 	if (!wrblksz)
    231 		wrblksz = blksz = rdblksz;
    232 	else
    233 		blksz = rdblksz;
    234 
    235 	/*
    236 	 * make sure that this volume allows appends
    237 	 */
    238 	if (ar_app_ok() < 0)
    239 		return(-1);
    240 
    241 	/*
    242 	 * Calculate bytes to move back and move in front of record where we
    243 	 * need to start writing from. Remember we have to add in any padding
    244 	 * that might be in the buffer after the trailer in the last block. We
    245 	 * travel skcnt + padding ROUNDED UP to blksize.
    246 	 */
    247 	skcnt += bufend - bufpt;
    248 	if ((cnt = (skcnt/blksz) * blksz) < skcnt)
    249 		cnt += blksz;
    250 	if (ar_rev((off_t)cnt) < 0)
    251 		goto out;
    252 
    253 	/*
    254 	 * We may have gone too far if there is valid data in the block we are
    255 	 * now in front of, read up the block and position the pointer after
    256 	 * the valid data.
    257 	 */
    258 	if ((cnt -= skcnt) > 0) {
    259 		/*
    260 		 * watch out for stupid tape drives. ar_rev() will set rdblksz
    261 		 * to be real physical blocksize so we must loop until we get
    262 		 * the old rdblksz (now in blksz). If ar_rev() fouls up the
    263 		 * determination of the physical block size, we will fail.
    264 		 */
    265 		bufpt = buf;
    266 		bufend = buf + blksz;
    267 		while (bufpt < bufend) {
    268 			if ((res = ar_read(bufpt, rdblksz)) <= 0)
    269 				goto out;
    270 			bufpt += res;
    271 		}
    272 		if (ar_rev((off_t)(bufpt - buf)) < 0)
    273 			goto out;
    274 		bufpt = buf + cnt;
    275 		bufend = buf + blksz;
    276 	} else {
    277 		/*
    278 		 * buffer is empty
    279 		 */
    280 		bufend = buf + blksz;
    281 		bufpt = buf;
    282 	}
    283 	rdblksz = blksz;
    284 	rdcnt -= skcnt;
    285 	wrcnt = 0;
    286 
    287 	/*
    288 	 * At this point we are ready to write. If the device requires special
    289 	 * handling to write at a point were previously recorded data resides,
    290 	 * that is handled in ar_set_wr(). From now on we operate under normal
    291 	 * ARCHIVE mode (write) conditions
    292 	 */
    293 	if (ar_set_wr() < 0)
    294 		return(-1);
    295 	act = ARCHIVE;
    296 	return(0);
    297 
    298     out:
    299 	tty_warn(1, "Unable to rewrite archive trailer, cannot append.");
    300 	return(-1);
    301 }
    302 
    303 /*
    304  * rd_sync()
    305  *	A read error occurred on this archive volume. Resync the buffer and
    306  *	try to reset the device (if possible) so we can continue to read. Keep
    307  *	trying to do this until we get a valid read, or we reach the limit on
    308  *	consecutive read faults (at which point we give up). The user can
    309  *	adjust the read error limit through a command line option.
    310  * Returns:
    311  *	0 on success, and -1 on failure
    312  */
    313 
    314 int
    315 rd_sync(void)
    316 {
    317 	int errcnt = 0;
    318 	int res;
    319 
    320 	/*
    321 	 * if the user says bail out on first fault, we are out of here...
    322 	 */
    323 	if (maxflt == 0)
    324 		return(-1);
    325 	if (act == APPND) {
    326 		tty_warn(1,
    327 		    "Unable to append when there are archive read errors.");
    328 		return(-1);
    329 	}
    330 
    331 	/*
    332 	 * poke at device and try to get past media error
    333 	 */
    334 	if (ar_rdsync() < 0) {
    335 		if (ar_next() < 0)
    336 			return(-1);
    337 		else
    338 			rdcnt = 0;
    339 	}
    340 
    341 	for (;;) {
    342 		if ((res = ar_read(buf, blksz)) > 0) {
    343 			/*
    344 			 * All right! got some data, fill that buffer
    345 			 */
    346 			bufpt = buf;
    347 			bufend = buf + res;
    348 			rdcnt += res;
    349 			return(0);
    350 		}
    351 
    352 		/*
    353 		 * Oh well, yet another failed read...
    354 		 * if error limit reached, ditch. otherwise poke device to move past
    355 		 * bad media and try again. if media is badly damaged, we ask
    356 		 * the poor (and upset user at this point) for the next archive
    357 		 * volume. remember the goal on reads is to get the most we
    358 		 * can extract out of the archive.
    359 		 */
    360 		if ((maxflt > 0) && (++errcnt > maxflt))
    361 			tty_warn(0,
    362 			    "Archive read error limit (%d) reached",maxflt);
    363 		else if (ar_rdsync() == 0)
    364 			continue;
    365 		if (ar_next() < 0)
    366 			break;
    367 		rdcnt = 0;
    368 		errcnt = 0;
    369 	}
    370 	return(-1);
    371 }
    372 
    373 /*
    374  * pback()
    375  *	push the data used during the archive id phase back into the I/O
    376  *	buffer. This is required as we cannot be sure that the header does NOT
    377  *	overlap a block boundary (as in the case we are trying to recover a
    378  *	flawed archived). This was not designed to be used for any other
    379  *	purpose. (What software engineering, HA!)
    380  *	WARNING: do not even THINK of pback greater than BLKMULT, unless the
    381  *	pback space is increased.
    382  */
    383 
    384 void
    385 pback(char *pt, int cnt)
    386 {
    387 	bufpt -= cnt;
    388 	memcpy(bufpt, pt, cnt);
    389 	return;
    390 }
    391 
    392 /*
    393  * rd_skip()
    394  *	skip forward in the archive during a archive read. Used to get quickly
    395  *	past file data and padding for files the user did NOT select.
    396  * Return:
    397  *	0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
    398  */
    399 
    400 int
    401 rd_skip(off_t skcnt)
    402 {
    403 	off_t res;
    404 	off_t cnt;
    405 	off_t skipped = 0;
    406 
    407 	/*
    408 	 * consume what data we have in the buffer. If we have to move forward
    409 	 * whole records, we call the low level skip function to see if we can
    410 	 * move within the archive without doing the expensive reads on data we
    411 	 * do not want.
    412 	 */
    413 	if (skcnt == 0)
    414 		return(0);
    415 	res = MIN((bufend - bufpt), skcnt);
    416 	bufpt += res;
    417 	skcnt -= res;
    418 
    419 	/*
    420 	 * if skcnt is now 0, then no additional i/o is needed
    421 	 */
    422 	if (skcnt == 0)
    423 		return(0);
    424 
    425 	/*
    426 	 * We have to read more, calculate complete and partial record reads
    427 	 * based on rdblksz. we skip over "cnt" complete records
    428 	 */
    429 	res = skcnt%rdblksz;
    430 	cnt = (skcnt/rdblksz) * rdblksz;
    431 
    432 	/*
    433 	 * if the skip fails, we will have to resync. ar_fow will tell us
    434 	 * how much it can skip over. We will have to read the rest.
    435 	 */
    436 	if (ar_fow(cnt, &skipped) < 0)
    437 		return(-1);
    438 	res += cnt - skipped;
    439 	rdcnt += skipped;
    440 
    441 	/*
    442 	 * what is left we have to read (which may be the whole thing if
    443 	 * ar_fow() told us the device can only read to skip records);
    444 	 */
    445 	while (res > 0L) {
    446 		cnt = bufend - bufpt;
    447 		/*
    448 		 * if the read fails, we will have to resync
    449 		 */
    450 		if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
    451 			return(-1);
    452 		if (cnt == 0)
    453 			return(1);
    454 		cnt = MIN(cnt, res);
    455 		bufpt += cnt;
    456 		res -= cnt;
    457 	}
    458 	return(0);
    459 }
    460 
    461 /*
    462  * wr_fin()
    463  *	flush out any data (and pad if required) the last block. We always pad
    464  *	with zero (even though we do not have to). Padding with 0 makes it a
    465  *	lot easier to recover if the archive is damaged. zero paddding SHOULD
    466  *	BE a requirement....
    467  */
    468 
    469 void
    470 wr_fin(void)
    471 {
    472 	if (bufpt > buf) {
    473 		memset(bufpt, 0, bufend - bufpt);
    474 		bufpt = bufend;
    475 		(void)buf_flush(blksz);
    476 	}
    477 }
    478 
    479 /*
    480  * wr_rdbuf()
    481  *	fill the write buffer from data passed to it in a buffer (usually used
    482  *	by format specific write routines to pass a file header). On failure we
    483  *	punt. We do not allow the user to continue to write flawed archives.
    484  *	We assume these headers are not very large (the memory copy we use is
    485  *	a bit expensive).
    486  * Return:
    487  *	0 if buffer was filled ok, -1 o.w. (buffer flush failure)
    488  */
    489 
    490 int
    491 wr_rdbuf(char *out, int outcnt)
    492 {
    493 	int cnt;
    494 
    495 	/*
    496 	 * while there is data to copy copy into the write buffer. when the
    497 	 * write buffer fills, flush it to the archive and continue
    498 	 */
    499 	while (outcnt > 0) {
    500 		cnt = bufend - bufpt;
    501 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
    502 			return(-1);
    503 		/*
    504 		 * only move what we have space for
    505 		 */
    506 		cnt = MIN(cnt, outcnt);
    507 		memcpy(bufpt, out, cnt);
    508 		bufpt += cnt;
    509 		out += cnt;
    510 		outcnt -= cnt;
    511 	}
    512 	return(0);
    513 }
    514 
    515 /*
    516  * rd_wrbuf()
    517  *	copy from the read buffer into a supplied buffer a specified number of
    518  *	bytes. If the read buffer is empty fill it and continue to copy.
    519  *	usually used to obtain a file header for processing by a format
    520  *	specific read routine.
    521  * Return
    522  *	number of bytes copied to the buffer, 0 indicates EOF on archive volume,
    523  *	-1 is a read error
    524  */
    525 
    526 int
    527 rd_wrbuf(char *in, int cpcnt)
    528 {
    529 	int res;
    530 	int cnt;
    531 	int incnt = cpcnt;
    532 
    533 	/*
    534 	 * loop until we fill the buffer with the requested number of bytes
    535 	 */
    536 	while (incnt > 0) {
    537 		cnt = bufend - bufpt;
    538 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
    539 			/*
    540 			 * read error, return what we got (or the error if
    541 			 * no data was copied). The caller must know that an
    542 			 * error occurred and has the best knowledge what to
    543 			 * do with it
    544 			 */
    545 			if ((res = cpcnt - incnt) > 0)
    546 				return(res);
    547 			return(cnt);
    548 		}
    549 
    550 		/*
    551 		 * calculate how much data to copy based on whats left and
    552 		 * state of buffer
    553 		 */
    554 		cnt = MIN(cnt, incnt);
    555 		memcpy(in, bufpt, cnt);
    556 		bufpt += cnt;
    557 		incnt -= cnt;
    558 		in += cnt;
    559 	}
    560 	return(cpcnt);
    561 }
    562 
    563 /*
    564  * wr_skip()
    565  *	skip forward during a write. In other words add padding to the file.
    566  *	we add zero filled padding as it makes flawed archives much easier to
    567  *	recover from. the caller tells us how many bytes of padding to add
    568  *	This routine was not designed to add HUGE amount of padding, just small
    569  *	amounts (a few 512 byte blocks at most)
    570  * Return:
    571  *	0 if ok, -1 if there was a buf_flush failure
    572  */
    573 
    574 int
    575 wr_skip(off_t skcnt)
    576 {
    577 	int cnt;
    578 
    579 	/*
    580 	 * loop while there is more padding to add
    581 	 */
    582 	while (skcnt > 0L) {
    583 		cnt = bufend - bufpt;
    584 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
    585 			return(-1);
    586 		cnt = MIN(cnt, skcnt);
    587 		memset(bufpt, 0, cnt);
    588 		bufpt += cnt;
    589 		skcnt -= cnt;
    590 	}
    591 	return(0);
    592 }
    593 
    594 /*
    595  * wr_rdfile()
    596  *	fill write buffer with the contents of a file. We are passed an	open
    597  *	file descriptor to the file an the archive structure that describes the
    598  *	file we are storing. The variable "left" is modified to contain the
    599  *	number of bytes of the file we were NOT able to write to the archive.
    600  *	it is important that we always write EXACTLY the number of bytes that
    601  *	the format specific write routine told us to. The file can also get
    602  *	bigger, so reading to the end of file would create an improper archive,
    603  *	we just detect this case and warn the user. We never create a bad
    604  *	archive if we can avoid it. Of course trying to archive files that are
    605  *	active is asking for trouble. It we fail, we pass back how much we
    606  *	could NOT copy and let the caller deal with it.
    607  * Return:
    608  *	0 ok, -1 if archive write failure. a short read of the file returns a
    609  *	0, but "left" is set to be greater than zero.
    610  */
    611 
    612 int
    613 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
    614 {
    615 	int cnt;
    616 	int res = 0;
    617 	off_t size = arcn->sb.st_size;
    618 	struct stat sb;
    619 
    620 	/*
    621 	 * while there are more bytes to write
    622 	 */
    623 	while (size > 0L) {
    624 		cnt = bufend - bufpt;
    625 		if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
    626 			*left = size;
    627 			return(-1);
    628 		}
    629 		cnt = MIN(cnt, size);
    630 		if ((res = read_with_restart(ifd, bufpt, cnt)) <= 0)
    631 			break;
    632 		size -= res;
    633 		bufpt += res;
    634 	}
    635 
    636 	/*
    637 	 * better check the file did not change during this operation
    638 	 * or the file read failed.
    639 	 */
    640 	if (res < 0)
    641 		syswarn(1, errno, "Read fault on %s", arcn->org_name);
    642 	else if (size != 0L)
    643 		tty_warn(1, "File changed size during read %s", arcn->org_name);
    644 	else if (fstat(ifd, &sb) < 0)
    645 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
    646 	else if (arcn->sb.st_mtime != sb.st_mtime)
    647 		tty_warn(1, "File %s was modified during copy to archive",
    648 			arcn->org_name);
    649 	*left = size;
    650 	return(0);
    651 }
    652 
    653 /*
    654  * rd_wrfile()
    655  *	extract the contents of a file from the archive. If we are unable to
    656  *	extract the entire file (due to failure to write the file) we return
    657  *	the numbers of bytes we did NOT process. This way the caller knows how
    658  *	many bytes to skip past to find the next archive header. If the failure
    659  *	was due to an archive read, we will catch that when we try to skip. If
    660  *	the format supplies a file data crc value, we calculate the actual crc
    661  *	so that it can be compared to the value stored in the header
    662  * NOTE:
    663  *	We call a special function to write the file. This function attempts to
    664  *	restore file holes (blocks of zeros) into the file. When files are
    665  *	sparse this saves space, and is a LOT faster. For non sparse files
    666  *	the performance hit is small. As of this writing, no archive supports
    667  *	information on where the file holes are.
    668  * Return:
    669  *	0 ok, -1 if archive read failure. if we cannot write the entire file,
    670  *	we return a 0 but "left" is set to be the amount unwritten
    671  */
    672 
    673 int
    674 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
    675 {
    676 	int cnt = 0;
    677 	off_t size = arcn->sb.st_size;
    678 	int res = 0;
    679 	char *fnm = arcn->name;
    680 	int isem = 1;
    681 	int rem;
    682 	int sz = MINFBSZ;
    683 	struct stat sb;
    684 	u_long crc = 0L;
    685 
    686 	/*
    687 	 * pass the blocksize of the file being written to the write routine,
    688 	 * if the size is zero, use the default MINFBSZ
    689 	 */
    690 	if (ofd == -1)
    691 		sz = PAXPATHLEN+1;
    692 	else if (fstat(ofd, &sb) == 0) {
    693 		if (sb.st_blksize > 0)
    694 			sz = (int)sb.st_blksize;
    695 	} else
    696 		syswarn(0, errno,
    697 		    "Unable to obtain block size for file %s", fnm);
    698 	rem = sz;
    699 	*left = 0L;
    700 
    701 	/*
    702 	 * Copy the archive to the file the number of bytes specified. We have
    703 	 * to assume that we want to recover file holes as none of the archive
    704 	 * formats can record the location of file holes.
    705 	 */
    706 	while (size > 0L) {
    707 		cnt = bufend - bufpt;
    708 		/*
    709 		 * if we get a read error, we do not want to skip, as we may
    710 		 * miss a header, so we do not set left, but if we get a write
    711 		 * error, we do want to skip over the unprocessed data.
    712 		 */
    713 		if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
    714 			break;
    715 		cnt = MIN(cnt, size);
    716 		if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
    717 			*left = size;
    718 			break;
    719 		}
    720 
    721 		if (docrc) {
    722 			/*
    723 			 * update the actual crc value
    724 			 */
    725 			cnt = res;
    726 			while (--cnt >= 0)
    727 				crc += *bufpt++ & 0xff;
    728 		} else
    729 			bufpt += res;
    730 		size -= res;
    731 	}
    732 
    733 	/*
    734 	 * if the last block has a file hole (all zero), we must make sure this
    735 	 * gets updated in the file. We force the last block of zeros to be
    736 	 * written. just closing with the file offset moved forward may not put
    737 	 * a hole at the end of the file.
    738 	 */
    739 	if (ofd != -1 && isem && (arcn->sb.st_size > 0L))
    740 		file_flush(ofd, fnm, isem);
    741 
    742 	/*
    743 	 * if we failed from archive read, we do not want to skip
    744 	 */
    745 	if ((size > 0L) && (*left == 0L))
    746 		return(-1);
    747 
    748 	/*
    749 	 * some formats record a crc on file data. If so, then we compare the
    750 	 * calculated crc to the crc stored in the archive
    751 	 */
    752 	if (docrc && (size == 0L) && (arcn->crc != crc))
    753 		tty_warn(1,"Actual crc does not match expected crc %s",
    754 		    arcn->name);
    755 	return(0);
    756 }
    757 
    758 /*
    759  * cp_file()
    760  *	copy the contents of one file to another. used during -rw phase of pax
    761  *	just as in rd_wrfile() we use a special write function to write the
    762  *	destination file so we can properly copy files with holes.
    763  */
    764 
    765 void
    766 cp_file(ARCHD *arcn, int fd1, int fd2)
    767 {
    768 	int cnt;
    769 	off_t cpcnt = 0L;
    770 	int res = 0;
    771 	char *fnm = arcn->name;
    772 	int no_hole = 0;
    773 	int isem = 1;
    774 	int rem;
    775 	int sz = MINFBSZ;
    776 	struct stat sb;
    777 
    778 	/*
    779 	 * check for holes in the source file. If none, we will use regular
    780 	 * write instead of file write.
    781 	 */
    782 	 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
    783 		++no_hole;
    784 
    785 	/*
    786 	 * pass the blocksize of the file being written to the write routine,
    787 	 * if the size is zero, use the default MINFBSZ
    788 	 */
    789 	if (fstat(fd2, &sb) == 0) {
    790 		if (sb.st_blksize > 0)
    791 			sz = sb.st_blksize;
    792 	} else
    793 		syswarn(0, errno,
    794 		    "Unable to obtain block size for file %s", fnm);
    795 	rem = sz;
    796 
    797 	/*
    798 	 * read the source file and copy to destination file until EOF
    799 	 */
    800 	for(;;) {
    801 		if ((cnt = read_with_restart(fd1, buf, blksz)) <= 0)
    802 			break;
    803 		if (no_hole)
    804 			res = xwrite(fd2, buf, cnt);
    805 		else
    806 			res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
    807 		if (res != cnt)
    808 			break;
    809 		cpcnt += cnt;
    810 	}
    811 
    812 	/*
    813 	 * check to make sure the copy is valid.
    814 	 */
    815 	if (res < 0)
    816 		syswarn(1, errno, "Failed write during copy of %s to %s",
    817 			arcn->org_name, arcn->name);
    818 	else if (cpcnt != arcn->sb.st_size)
    819 		tty_warn(1, "File %s changed size during copy to %s",
    820 			arcn->org_name, arcn->name);
    821 	else if (fstat(fd1, &sb) < 0)
    822 		syswarn(1, errno, "Failed stat of %s", arcn->org_name);
    823 	else if (arcn->sb.st_mtime != sb.st_mtime)
    824 		tty_warn(1, "File %s was modified during copy to %s",
    825 			arcn->org_name, arcn->name);
    826 
    827 	/*
    828 	 * if the last block has a file hole (all zero), we must make sure this
    829 	 * gets updated in the file. We force the last block of zeros to be
    830 	 * written. just closing with the file offset moved forward may not put
    831 	 * a hole at the end of the file.
    832 	 */
    833 	if (!no_hole && isem && (arcn->sb.st_size > 0L))
    834 		file_flush(fd2, fnm, isem);
    835 	return;
    836 }
    837 
    838 /*
    839  * buf_fill()
    840  *	fill the read buffer with the next record (or what we can get) from
    841  *	the archive volume.
    842  * Return:
    843  *	Number of bytes of data in the read buffer, -1 for read error, and
    844  *	0 when finished (user specified termination in ar_next()).
    845  */
    846 
    847 int
    848 buf_fill(void)
    849 {
    850 	int cnt;
    851 	static int fini = 0;
    852 
    853 	if (fini)
    854 		return(0);
    855 
    856 	for(;;) {
    857 		/*
    858 		 * try to fill the buffer. on error the next archive volume is
    859 		 * opened and we try again.
    860 		 */
    861 		if ((cnt = ar_read(buf, blksz)) > 0) {
    862 			bufpt = buf;
    863 			bufend = buf + cnt;
    864 			rdcnt += cnt;
    865 			return(cnt);
    866 		}
    867 
    868 		/*
    869 		 * errors require resync, EOF goes to next archive
    870 		 */
    871 		if (cnt < 0)
    872 			break;
    873 		if (ar_next() < 0) {
    874 			fini = 1;
    875 			return(0);
    876 		}
    877 		rdcnt = 0;
    878 	}
    879 	exit_val = 1;
    880 	return(-1);
    881 }
    882 
    883 /*
    884  * buf_flush()
    885  *	force the write buffer to the archive. We are passed the number of
    886  *	bytes in the buffer at the point of the flush. When we change archives
    887  *	the record size might change. (either larger or smaller).
    888  * Return:
    889  *	0 if all is ok, -1 when a write error occurs.
    890  */
    891 
    892 int
    893 buf_flush(int bufcnt)
    894 {
    895 	int cnt;
    896 	int push = 0;
    897 	int totcnt = 0;
    898 
    899 	/*
    900 	 * if we have reached the user specified byte count for each archive
    901 	 * volume, prompt for the next volume. (The non-standard -R flag).
    902 	 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
    903 	 * at least one record. We always round limit UP to next blocksize.
    904 	 */
    905 	if ((wrlimit > 0) && (wrcnt > wrlimit)) {
    906 		tty_warn(0,
    907 		    "User specified archive volume byte limit reached.");
    908 		if (ar_next() < 0) {
    909 			wrcnt = 0;
    910 			exit_val = 1;
    911 			return(-1);
    912 		}
    913 		wrcnt = 0;
    914 
    915 		/*
    916 		 * The new archive volume might have changed the size of the
    917 		 * write blocksize. if so we figure out if we need to write
    918 		 * (one or more times), or if there is now free space left in
    919 		 * the buffer (it is no longer full). bufcnt has the number of
    920 		 * bytes in the buffer, (the blocksize, at the point we were
    921 		 * CALLED). Push has the amount of "extra" data in the buffer
    922 		 * if the block size has shrunk from a volume change.
    923 		 */
    924 		bufend = buf + blksz;
    925 		if (blksz > bufcnt)
    926 			return(0);
    927 		if (blksz < bufcnt)
    928 			push = bufcnt - blksz;
    929 	}
    930 
    931 	/*
    932 	 * We have enough data to write at least one archive block
    933 	 */
    934 	for (;;) {
    935 		/*
    936 		 * write a block and check if it all went out ok
    937 		 */
    938 		cnt = ar_write(buf, blksz);
    939 		if (cnt == blksz) {
    940 			/*
    941 			 * the write went ok
    942 			 */
    943 			wrcnt += cnt;
    944 			totcnt += cnt;
    945 			if (push > 0) {
    946 				/* we have extra data to push to the front.
    947 				 * check for more than 1 block of push, and if
    948 				 * so we loop back to write again
    949 				 */
    950 				memcpy(buf, bufend, push);
    951 				bufpt = buf + push;
    952 				if (push >= blksz) {
    953 					push -= blksz;
    954 					continue;
    955 				}
    956 			} else
    957 				bufpt = buf;
    958 			return(totcnt);
    959 		} else if (cnt > 0) {
    960 			/*
    961 			 * Oh drat we got a partial write!
    962 			 * if format doesnt care about alignment let it go,
    963 			 * we warned the user in ar_write().... but this means
    964 			 * the last record on this volume violates pax spec....
    965 			 */
    966 			totcnt += cnt;
    967 			wrcnt += cnt;
    968 			bufpt = buf + cnt;
    969 			cnt = bufcnt - cnt;
    970 			memcpy(buf, bufpt, cnt);
    971 			bufpt = buf + cnt;
    972 			if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
    973 				return(totcnt);
    974 			break;
    975 		}
    976 
    977 		/*
    978 		 * All done, go to next archive
    979 		 */
    980 		wrcnt = 0;
    981 		if (ar_next() < 0)
    982 			break;
    983 
    984 		/*
    985 		 * The new archive volume might also have changed the block
    986 		 * size. if so, figure out if we have too much or too little
    987 		 * data for using the new block size
    988 		 */
    989 		bufend = buf + blksz;
    990 		if (blksz > bufcnt)
    991 			return(0);
    992 		if (blksz < bufcnt)
    993 			push = bufcnt - blksz;
    994 	}
    995 
    996 	/*
    997 	 * write failed, stop pax. we must not create a bad archive!
    998 	 */
    999 	exit_val = 1;
   1000 	return(-1);
   1001 }
   1002