1 /* $NetBSD: buf_subs.c,v 1.27 2006/02/11 10:43:18 dsl Exp $ */
4 * Copyright (c) 1992 Keith Muller.
5 * Copyright (c) 1992, 1993
6 * The Regents of the University of California. All rights reserved.
8 * This code is derived from software contributed to Berkeley by
9 * Keith Muller of the University of California, San Diego.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
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. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 #if HAVE_NBTOOL_CONFIG_H
37 #include "nbtool_config.h"
40 #include <sys/cdefs.h>
43 static char sccsid
[] = "@(#)buf_subs.c 8.2 (Berkeley) 4/18/94";
45 __RCSID("$NetBSD: buf_subs.c,v 1.27 2006/02/11 10:43:18 dsl Exp $");
49 #include <sys/types.h>
52 #include <sys/param.h>
63 * routines which implement archive and file buffering
66 #define MINFBSZ 512 /* default block size for hole detect */
67 #define MAXFLT 10 /* default media read error limit */
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.
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 */
88 * set up the buffering system to operate in a write mode
90 * 0 if ok, -1 if the user specified write block size violates pax spec
96 buf
= &(bufmem
[BLKMULT
]);
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
106 if (wrblksz
> MAXBLK
) {
107 tty_warn(1, "Write block size of %d too large, maximum is: %d",
111 if (wrblksz
% BLKMULT
) {
112 tty_warn(1, "Write block size of %d is not a %d byte multiple",
118 * we only allow wrblksz to be used with all archive operations
120 blksz
= rdblksz
= wrblksz
;
121 if ((ar_open(arcname
) < 0) && (ar_next() < 0))
124 bufend
= buf
+ wrblksz
;
131 * set up buffering system to read an archive
133 * 0 if ok, -1 otherwise
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
144 buf
= &(bufmem
[BLKMULT
]);
145 if ((act
== APPND
) && wrblksz
) {
146 if (wrblksz
> MAXBLK
) {
148 "Write block size %d too large, maximum is: %d",
152 if (wrblksz
% BLKMULT
) {
154 "Write block size %d is not a %d byte multiple",
163 if ((ar_open(arcname
) < 0) && (ar_next() < 0))
165 bufend
= buf
+ rdblksz
;
173 * set up buffer system for copying within the file system
179 buf
= &(bufmem
[BLKMULT
]);
180 rdblksz
= blksz
= MAXBLK
;
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
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
212 * 0 for success, -1 for failure
216 appnd_start(off_t skcnt
)
222 tty_warn(0, "Cannot append to an archive that may have flaws.");
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).
231 wrblksz
= blksz
= rdblksz
;
236 * make sure that this volume allows appends
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.
247 skcnt
+= bufend
- bufpt
;
248 if ((cnt
= (skcnt
/blksz
) * blksz
) < skcnt
)
250 if (ar_rev((off_t
)cnt
) < 0)
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
258 if ((cnt
-= skcnt
) > 0) {
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.
266 bufend
= buf
+ blksz
;
267 while (bufpt
< bufend
) {
268 if ((res
= ar_read(bufpt
, rdblksz
)) <= 0)
272 if (ar_rev((off_t
)(bufpt
- buf
)) < 0)
275 bufend
= buf
+ blksz
;
280 bufend
= buf
+ blksz
;
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
299 tty_warn(1, "Unable to rewrite archive trailer, cannot append.");
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.
311 * 0 on success, and -1 on failure
321 * if the user says bail out on first fault, we are out of here...
327 "Unable to append when there are archive read errors.");
332 * poke at device and try to get past media error
334 if (ar_rdsync() < 0) {
342 if ((res
= ar_read(buf
, blksz
)) > 0) {
344 * All right! got some data, fill that buffer
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.
360 if ((maxflt
> 0) && (++errcnt
> maxflt
))
362 "Archive read error limit (%d) reached",maxflt
);
363 else if (ar_rdsync() == 0)
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.
385 pback(char *pt
, int cnt
)
388 memcpy(bufpt
, pt
, cnt
);
394 * skip forward in the archive during an archive read. Used to get quickly
395 * past file data and padding for files the user did NOT select.
397 * 0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
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
415 res
= MIN((bufend
- bufpt
), skcnt
);
420 * if skcnt is now 0, then no additional i/o is needed
426 * We have to read more, calculate complete and partial record reads
427 * based on rdblksz. we skip over "cnt" complete records
430 cnt
= (skcnt
/rdblksz
) * rdblksz
;
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.
436 if (ar_fow(cnt
, &skipped
) < 0)
438 res
+= cnt
- skipped
;
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);
446 cnt
= bufend
- bufpt
;
448 * if the read fails, we will have to resync
450 if ((cnt
<= 0) && ((cnt
= buf_fill()) < 0))
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....
473 memset(bufpt
, 0, bufend
- bufpt
);
475 (void)buf_flush(blksz
);
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
487 * 0 if buffer was filled ok, -1 o.w. (buffer flush failure)
491 wr_rdbuf(char *out
, int outcnt
)
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
500 cnt
= bufend
- bufpt
;
501 if ((cnt
<= 0) && ((cnt
= buf_flush(blksz
)) < 0))
504 * only move what we have space for
506 cnt
= MIN(cnt
, outcnt
);
507 memcpy(bufpt
, out
, cnt
);
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.
522 * number of bytes copied to the buffer, 0 indicates EOF on archive volume,
527 rd_wrbuf(char *in
, int cpcnt
)
534 * loop until we fill the buffer with the requested number of bytes
537 cnt
= bufend
- bufpt
;
538 if ((cnt
<= 0) && ((cnt
= buf_fill()) <= 0)) {
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
545 if ((res
= cpcnt
- incnt
) > 0)
551 * calculate how much data to copy based on whats left and
554 cnt
= MIN(cnt
, incnt
);
555 memcpy(in
, bufpt
, cnt
);
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)
571 * 0 if ok, -1 if there was a buf_flush failure
580 * loop while there is more padding to add
583 cnt
= bufend
- bufpt
;
584 if ((cnt
<= 0) && ((cnt
= buf_flush(blksz
)) < 0))
586 cnt
= MIN(cnt
, skcnt
);
587 memset(bufpt
, 0, cnt
);
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.
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.
613 wr_rdfile(ARCHD
*arcn
, int ifd
, off_t
*left
)
617 off_t size
= arcn
->sb
.st_size
;
618 struct stat origsb
, sb
;
621 * by default, remember the previously obtained stat information
622 * (in arcn->sb) for comparing the mtime after reading.
623 * if Mflag is set, use the actual mtime instead.
626 if (Mflag
&& (fstat(ifd
, &origsb
) < 0))
627 syswarn(1, errno
, "Failed stat on %s", arcn
->org_name
);
630 * while there are more bytes to write
633 cnt
= bufend
- bufpt
;
634 if ((cnt
<= 0) && ((cnt
= buf_flush(blksz
)) < 0)) {
638 cnt
= MIN(cnt
, size
);
639 if ((res
= read_with_restart(ifd
, bufpt
, cnt
)) <= 0)
646 * better check the file did not change during this operation
647 * or the file read failed.
650 syswarn(1, errno
, "Read fault on %s", arcn
->org_name
);
652 tty_warn(1, "File changed size during read %s", arcn
->org_name
);
653 else if (fstat(ifd
, &sb
) < 0)
654 syswarn(1, errno
, "Failed stat on %s", arcn
->org_name
);
655 else if (origsb
.st_mtime
!= sb
.st_mtime
)
656 tty_warn(1, "File %s was modified during copy to archive",
664 * extract the contents of a file from the archive. If we are unable to
665 * extract the entire file (due to failure to write the file) we return
666 * the numbers of bytes we did NOT process. This way the caller knows how
667 * many bytes to skip past to find the next archive header. If the failure
668 * was due to an archive read, we will catch that when we try to skip. If
669 * the format supplies a file data crc value, we calculate the actual crc
670 * so that it can be compared to the value stored in the header
672 * We call a special function to write the file. This function attempts to
673 * restore file holes (blocks of zeros) into the file. When files are
674 * sparse this saves space, and is a LOT faster. For non sparse files
675 * the performance hit is small. As of this writing, no archive supports
676 * information on where the file holes are.
678 * 0 ok, -1 if archive read failure. if we cannot write the entire file,
679 * we return a 0 but "left" is set to be the amount unwritten
683 rd_wrfile(ARCHD
*arcn
, int ofd
, off_t
*left
)
686 off_t size
= arcn
->sb
.st_size
;
688 char *fnm
= arcn
->name
;
696 * pass the blocksize of the file being written to the write routine,
697 * if the size is zero, use the default MINFBSZ
701 else if (fstat(ofd
, &sb
) == 0) {
702 if (sb
.st_blksize
> 0)
703 sz
= (int)sb
.st_blksize
;
706 "Unable to obtain block size for file %s", fnm
);
711 * Copy the archive to the file the number of bytes specified. We have
712 * to assume that we want to recover file holes as none of the archive
713 * formats can record the location of file holes.
716 cnt
= bufend
- bufpt
;
718 * if we get a read error, we do not want to skip, as we may
719 * miss a header, so we do not set left, but if we get a write
720 * error, we do want to skip over the unprocessed data.
722 if ((cnt
<= 0) && ((cnt
= buf_fill()) <= 0))
724 cnt
= MIN(cnt
, size
);
725 if ((res
= file_write(ofd
,bufpt
,cnt
,&rem
,&isem
,sz
,fnm
)) <= 0) {
732 * update the actual crc value
736 crc
+= *bufpt
++ & 0xff;
743 * if the last block has a file hole (all zero), we must make sure this
744 * gets updated in the file. We force the last block of zeros to be
745 * written. just closing with the file offset moved forward may not put
746 * a hole at the end of the file.
748 if (ofd
>= 0 && isem
&& (arcn
->sb
.st_size
> 0L))
749 file_flush(ofd
, fnm
, isem
);
752 * if we failed from archive read, we do not want to skip
754 if ((size
> 0L) && (*left
== 0L))
758 * some formats record a crc on file data. If so, then we compare the
759 * calculated crc to the crc stored in the archive
761 if (docrc
&& (size
== 0L) && (arcn
->crc
!= crc
))
762 tty_warn(1,"Actual crc does not match expected crc %s",
769 * copy the contents of one file to another. used during -rw phase of pax
770 * just as in rd_wrfile() we use a special write function to write the
771 * destination file so we can properly copy files with holes.
775 cp_file(ARCHD
*arcn
, int fd1
, int fd2
)
780 char *fnm
= arcn
->name
;
785 struct stat sb
, origsb
;
788 * check for holes in the source file. If none, we will use regular
789 * write instead of file write.
791 if (((off_t
)(arcn
->sb
.st_blocks
* BLKMULT
)) >= arcn
->sb
.st_size
)
795 * by default, remember the previously obtained stat information
796 * (in arcn->sb) for comparing the mtime after reading.
797 * if Mflag is set, use the actual mtime instead.
800 if (Mflag
&& (fstat(fd1
, &origsb
) < 0))
801 syswarn(1, errno
, "Failed stat on %s", arcn
->org_name
);
804 * pass the blocksize of the file being written to the write routine,
805 * if the size is zero, use the default MINFBSZ
807 if (fstat(fd2
, &sb
) == 0) {
808 if (sb
.st_blksize
> 0)
812 "Unable to obtain block size for file %s", fnm
);
816 * read the source file and copy to destination file until EOF
819 if ((cnt
= read_with_restart(fd1
, buf
, blksz
)) <= 0)
822 res
= xwrite(fd2
, buf
, cnt
);
824 res
= file_write(fd2
, buf
, cnt
, &rem
, &isem
, sz
, fnm
);
831 * check to make sure the copy is valid.
834 syswarn(1, errno
, "Failed write during copy of %s to %s",
835 arcn
->org_name
, arcn
->name
);
836 else if (cpcnt
!= arcn
->sb
.st_size
)
837 tty_warn(1, "File %s changed size during copy to %s",
838 arcn
->org_name
, arcn
->name
);
839 else if (fstat(fd1
, &sb
) < 0)
840 syswarn(1, errno
, "Failed stat of %s", arcn
->org_name
);
841 else if (origsb
.st_mtime
!= sb
.st_mtime
)
842 tty_warn(1, "File %s was modified during copy to %s",
843 arcn
->org_name
, arcn
->name
);
846 * if the last block has a file hole (all zero), we must make sure this
847 * gets updated in the file. We force the last block of zeros to be
848 * written. just closing with the file offset moved forward may not put
849 * a hole at the end of the file.
851 if (!no_hole
&& isem
&& (arcn
->sb
.st_size
> 0L))
852 file_flush(fd2
, fnm
, isem
);
858 * fill the read buffer with the next record (or what we can get) from
859 * the archive volume.
861 * Number of bytes of data in the read buffer, -1 for read error, and
862 * 0 when finished (user specified termination in ar_next()).
876 * try to fill the buffer. on error the next archive volume is
877 * opened and we try again.
879 if ((cnt
= ar_read(buf
, blksz
)) > 0) {
887 * errors require resync, EOF goes to next archive
888 * but in case we have not determined yet the format,
889 * this means that we have a very short file, so we
894 if (frmt
== NULL
|| ar_next() < 0) {
906 * force the write buffer to the archive. We are passed the number of
907 * bytes in the buffer at the point of the flush. When we change archives
908 * the record size might change. (either larger or smaller).
910 * 0 if all is ok, -1 when a write error occurs.
914 buf_flush(int bufcnt
)
921 * if we have reached the user specified byte count for each archive
922 * volume, prompt for the next volume. (The non-standard -R flag).
923 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
924 * at least one record. We always round limit UP to next blocksize.
926 if ((wrlimit
> 0) && (wrcnt
> wrlimit
)) {
928 "User specified archive volume byte limit reached.");
937 * The new archive volume might have changed the size of the
938 * write blocksize. if so we figure out if we need to write
939 * (one or more times), or if there is now free space left in
940 * the buffer (it is no longer full). bufcnt has the number of
941 * bytes in the buffer, (the blocksize, at the point we were
942 * CALLED). Push has the amount of "extra" data in the buffer
943 * if the block size has shrunk from a volume change.
945 bufend
= buf
+ blksz
;
949 push
= bufcnt
- blksz
;
953 * We have enough data to write at least one archive block
957 * write a block and check if it all went out ok
959 cnt
= ar_write(buf
, blksz
);
967 /* we have extra data to push to the front.
968 * check for more than 1 block of push, and if
969 * so we loop back to write again
971 memcpy(buf
, bufend
, push
);
980 } else if (cnt
> 0) {
982 * Oh drat we got a partial write!
983 * if format doesnt care about alignment let it go,
984 * we warned the user in ar_write().... but this means
985 * the last record on this volume violates pax spec....
991 memcpy(buf
, bufpt
, cnt
);
993 if (!frmt
->blkalgn
|| ((cnt
% frmt
->blkalgn
) == 0))
999 * All done, go to next archive
1006 * The new archive volume might also have changed the block
1007 * size. if so, figure out if we have too much or too little
1008 * data for using the new block size
1010 bufend
= buf
+ blksz
;
1014 push
= bufcnt
- blksz
;
1018 * write failed, stop pax. we must not create a bad archive!