1 /* $OpenBSD: buf_subs.c,v 1.24 2012/12/04 02:24:45 deraadt Exp $ */
2 /* $NetBSD: buf_subs.c,v 1.5 1995/03/21 09:07:08 cgd Exp $ */
5 * Copyright (c) 1992 Keith Muller.
6 * Copyright (c) 1992, 1993
7 * The Regents of the University of California. All rights reserved.
9 * This code is derived from software contributed to Berkeley by
10 * Keith Muller of the University of California, San Diego.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 #include <sys/types.h>
49 * routines which implement archive and file buffering
52 #define MINFBSZ 512 /* default block size for hole detect */
53 #define MAXFLT 10 /* default media read error limit */
56 * Need to change bufmem to dynamic allocation when the upper
57 * limit on blocking size is removed (though that will violate pax spec)
58 * MAXBLK define and tests will also need to be updated.
60 static char bufmem
[MAXBLK
+BLKMULT
]; /* i/o buffer + pushback id space */
61 static char *buf
; /* normal start of i/o buffer */
62 static char *bufend
; /* end or last char in i/o buffer */
63 static char *bufpt
; /* read/write point in i/o buffer */
64 int blksz
= MAXBLK
; /* block input/output size in bytes */
65 int wrblksz
; /* user spec output size in bytes */
66 int maxflt
= MAXFLT
; /* MAX consecutive media errors */
67 int rdblksz
; /* first read blksize (tapes only) */
68 off_t wrlimit
; /* # of bytes written per archive vol */
69 off_t wrcnt
; /* # of bytes written on current vol */
70 off_t rdcnt
; /* # of bytes read on current vol */
74 * set up the buffering system to operate in a write mode
76 * 0 if ok, -1 if the user specified write block size violates pax spec
82 buf
= &(bufmem
[BLKMULT
]);
84 * Check to make sure the write block size meets pax specs. If the user
85 * does not specify a blocksize, we use the format default blocksize.
86 * We must be picky on writes, so we do not allow the user to create an
87 * archive that might be hard to read elsewhere. If all ok, we then
88 * open the first archive volume
92 if (wrblksz
> MAXBLK
) {
93 paxwarn(1, "Write block size of %d too large, maximium is: %d",
97 if (wrblksz
% BLKMULT
) {
98 paxwarn(1, "Write block size of %d is not a %d byte multiple",
102 if (wrblksz
> MAXBLK_POSIX
) {
103 paxwarn(0, "Write block size of %d larger than POSIX max %d, archive may not be portable",
104 wrblksz
, MAXBLK_POSIX
);
109 * we only allow wrblksz to be used with all archive operations
111 blksz
= rdblksz
= wrblksz
;
112 if ((ar_open(arcname
) < 0) && (ar_next() < 0))
115 bufend
= buf
+ wrblksz
;
122 * set up buffering system to read an archive
124 * 0 if ok, -1 otherwise
131 * leave space for the header pushback (see get_arc()). If we are
132 * going to append and user specified a write block size, check it
135 buf
= &(bufmem
[BLKMULT
]);
136 if ((act
== APPND
) && wrblksz
) {
137 if (wrblksz
> MAXBLK
) {
138 paxwarn(1,"Write block size %d too large, maximium is: %d",
142 if (wrblksz
% BLKMULT
) {
143 paxwarn(1, "Write block size %d is not a %d byte multiple",
152 if ((ar_open(arcname
) < 0) && (ar_next() < 0))
154 bufend
= buf
+ rdblksz
;
162 * set up buffer system for copying within the file system
168 buf
= &(bufmem
[BLKMULT
]);
169 rdblksz
= blksz
= MAXBLK
;
174 * Set up the buffering system to append new members to an archive that
175 * was just read. The last block(s) of an archive may contain a format
176 * specific trailer. To append a new member, this trailer has to be
177 * removed from the archive. The first byte of the trailer is replaced by
178 * the start of the header of the first file added to the archive. The
179 * format specific end read function tells us how many bytes to move
180 * backwards in the archive to be positioned BEFORE the trailer. Two
181 * different position have to be adjusted, the O.S. file offset (e.g. the
182 * position of the tape head) and the write point within the data we have
183 * stored in the read (soon to become write) buffer. We may have to move
184 * back several records (the number depends on the size of the archive
185 * record and the size of the format trailer) to read up the record where
186 * the first byte of the trailer is recorded. Trailers may span (and
187 * overlap) record boundaries.
188 * We first calculate which record has the first byte of the trailer. We
189 * move the OS file offset back to the start of this record and read it
190 * up. We set the buffer write pointer to be at this byte (the byte where
191 * the trailer starts). We then move the OS file pointer back to the
192 * start of this record so a flush of this buffer will replace the record
194 * A major problem is rewriting this last record. For archives stored
195 * on disk files, this is trivial. However, many devices are really picky
196 * about the conditions under which they will allow a write to occur.
197 * Often devices restrict the conditions where writes can be made,
198 * so it may not be feasible to append archives stored on all types of
201 * 0 for success, -1 for failure
205 appnd_start(off_t skcnt
)
211 paxwarn(0, "Cannot append to an archive that may have flaws.");
215 * if the user did not specify a write blocksize, inherit the size used
216 * in the last archive volume read. (If a is set we still use rdblksz
217 * until next volume, cannot shift sizes within a single volume).
220 wrblksz
= blksz
= rdblksz
;
225 * make sure that this volume allows appends
231 * Calculate bytes to move back and move in front of record where we
232 * need to start writing from. Remember we have to add in any padding
233 * that might be in the buffer after the trailer in the last block. We
234 * travel skcnt + padding ROUNDED UP to blksize.
236 skcnt
+= bufend
- bufpt
;
237 if ((cnt
= (skcnt
/blksz
) * blksz
) < skcnt
)
239 if (ar_rev((off_t
)cnt
) < 0)
243 * We may have gone too far if there is valid data in the block we are
244 * now in front of, read up the block and position the pointer after
247 if ((cnt
-= skcnt
) > 0) {
249 * watch out for stupid tape drives. ar_rev() will set rdblksz
250 * to be real physical blocksize so we must loop until we get
251 * the old rdblksz (now in blksz). If ar_rev() fouls up the
252 * determination of the physical block size, we will fail.
255 bufend
= buf
+ blksz
;
256 while (bufpt
< bufend
) {
257 if ((res
= ar_read(bufpt
, rdblksz
)) <= 0)
261 if (ar_rev((off_t
)(bufpt
- buf
)) < 0)
264 bufend
= buf
+ blksz
;
269 bufend
= buf
+ blksz
;
277 * At this point we are ready to write. If the device requires special
278 * handling to write at a point were previously recorded data resides,
279 * that is handled in ar_set_wr(). From now on we operate under normal
280 * ARCHIVE mode (write) conditions
288 paxwarn(1, "Unable to rewrite archive trailer, cannot append.");
294 * A read error occurred on this archive volume. Resync the buffer and
295 * try to reset the device (if possible) so we can continue to read. Keep
296 * trying to do this until we get a valid read, or we reach the limit on
297 * consecutive read faults (at which point we give up). The user can
298 * adjust the read error limit through a command line option.
300 * 0 on success, and -1 on failure
310 * if the user says bail out on first fault, we are out of here...
315 paxwarn(1, "Unable to append when there are archive read errors.");
320 * poke at device and try to get past media error
322 if (ar_rdsync() < 0) {
330 if ((res
= ar_read(buf
, blksz
)) > 0) {
332 * All right! got some data, fill that buffer
341 * Oh well, yet another failed read...
342 * if error limit reached, ditch. o.w. poke device to move past
343 * bad media and try again. if media is badly damaged, we ask
344 * the poor (and upset user at this point) for the next archive
345 * volume. remember the goal on reads is to get the most we
346 * can extract out of the archive.
348 if ((maxflt
> 0) && (++errcnt
> maxflt
))
349 paxwarn(0,"Archive read error limit (%d) reached",maxflt
);
350 else if (ar_rdsync() == 0)
362 * push the data used during the archive id phase back into the I/O
363 * buffer. This is required as we cannot be sure that the header does NOT
364 * overlap a block boundary (as in the case we are trying to recover a
365 * flawed archived). This was not designed to be used for any other
366 * purpose. (What software engineering, HA!)
367 * WARNING: do not even THINK of pback greater than BLKMULT, unless the
368 * pback space is increased.
372 pback(char *pt
, int cnt
)
375 memcpy(bufpt
, pt
, cnt
);
381 * skip forward in the archive during a archive read. Used to get quickly
382 * past file data and padding for files the user did NOT select.
384 * 0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
395 * consume what data we have in the buffer. If we have to move forward
396 * whole records, we call the low level skip function to see if we can
397 * move within the archive without doing the expensive reads on data we
402 res
= MIN((bufend
- bufpt
), skcnt
);
407 * if skcnt is now 0, then no additional i/o is needed
413 * We have to read more, calculate complete and partial record reads
414 * based on rdblksz. we skip over "cnt" complete records
417 cnt
= (skcnt
/rdblksz
) * rdblksz
;
420 * if the skip fails, we will have to resync. ar_fow will tell us
421 * how much it can skip over. We will have to read the rest.
423 if (ar_fow(cnt
, &skipped
) < 0)
425 res
+= cnt
- skipped
;
429 * what is left we have to read (which may be the whole thing if
430 * ar_fow() told us the device can only read to skip records);
433 cnt
= bufend
- bufpt
;
435 * if the read fails, we will have to resync
437 if ((cnt
<= 0) && ((cnt
= buf_fill()) < 0))
450 * flush out any data (and pad if required) the last block. We always pad
451 * with zero (even though we do not have to). Padding with 0 makes it a
452 * lot easier to recover if the archive is damaged. zero padding SHOULD
453 * BE a requirement....
460 memset(bufpt
, 0, bufend
- bufpt
);
462 (void)buf_flush(blksz
);
468 * fill the write buffer from data passed to it in a buffer (usually used
469 * by format specific write routines to pass a file header). On failure we
470 * punt. We do not allow the user to continue to write flawed archives.
471 * We assume these headers are not very large (the memory copy we use is
474 * 0 if buffer was filled ok, -1 o.w. (buffer flush failure)
478 wr_rdbuf(char *out
, int outcnt
)
483 * while there is data to copy copy into the write buffer. when the
484 * write buffer fills, flush it to the archive and continue
487 cnt
= bufend
- bufpt
;
488 if ((cnt
<= 0) && ((cnt
= buf_flush(blksz
)) < 0))
491 * only move what we have space for
493 cnt
= MIN(cnt
, outcnt
);
494 memcpy(bufpt
, out
, cnt
);
504 * copy from the read buffer into a supplied buffer a specified number of
505 * bytes. If the read buffer is empty fill it and continue to copy.
506 * usually used to obtain a file header for processing by a format
507 * specific read routine.
509 * number of bytes copied to the buffer, 0 indicates EOF on archive volume,
514 rd_wrbuf(char *in
, int cpcnt
)
521 * loop until we fill the buffer with the requested number of bytes
524 cnt
= bufend
- bufpt
;
525 if ((cnt
<= 0) && ((cnt
= buf_fill()) <= 0)) {
527 * read error, return what we got (or the error if
528 * no data was copied). The caller must know that an
529 * error occurred and has the best knowledge what to
532 if ((res
= cpcnt
- incnt
) > 0)
538 * calculate how much data to copy based on whats left and
541 cnt
= MIN(cnt
, incnt
);
542 memcpy(in
, bufpt
, cnt
);
552 * skip forward during a write. In other words add padding to the file.
553 * we add zero filled padding as it makes flawed archives much easier to
554 * recover from. the caller tells us how many bytes of padding to add
555 * This routine was not designed to add HUGE amount of padding, just small
556 * amounts (a few 512 byte blocks at most)
558 * 0 if ok, -1 if there was a buf_flush failure
567 * loop while there is more padding to add
570 cnt
= bufend
- bufpt
;
571 if ((cnt
<= 0) && ((cnt
= buf_flush(blksz
)) < 0))
573 cnt
= MIN(cnt
, skcnt
);
574 memset(bufpt
, 0, cnt
);
583 * fill write buffer with the contents of a file. We are passed an open
584 * file descriptor to the file an the archive structure that describes the
585 * file we are storing. The variable "left" is modified to contain the
586 * number of bytes of the file we were NOT able to write to the archive.
587 * it is important that we always write EXACTLY the number of bytes that
588 * the format specific write routine told us to. The file can also get
589 * bigger, so reading to the end of file would create an improper archive,
590 * we just detect this case and warn the user. We never create a bad
591 * archive if we can avoid it. Of course trying to archive files that are
592 * active is asking for trouble. It we fail, we pass back how much we
593 * could NOT copy and let the caller deal with it.
595 * 0 ok, -1 if archive write failure. a short read of the file returns a
596 * 0, but "left" is set to be greater than zero.
600 wr_rdfile(ARCHD
*arcn
, int ifd
, off_t
*left
)
604 off_t size
= arcn
->sb
.st_size
;
608 * while there are more bytes to write
611 cnt
= bufend
- bufpt
;
612 if ((cnt
<= 0) && ((cnt
= buf_flush(blksz
)) < 0)) {
616 cnt
= MIN(cnt
, size
);
617 if ((res
= read(ifd
, bufpt
, cnt
)) <= 0)
624 * better check the file did not change during this operation
625 * or the file read failed.
628 syswarn(1, errno
, "Read fault on %s", arcn
->org_name
);
630 paxwarn(1, "File changed size during read %s", arcn
->org_name
);
631 else if (fstat(ifd
, &sb
) < 0)
632 syswarn(1, errno
, "Failed stat on %s", arcn
->org_name
);
633 else if (arcn
->sb
.st_mtime
!= sb
.st_mtime
)
634 paxwarn(1, "File %s was modified during copy to archive",
642 * extract the contents of a file from the archive. If we are unable to
643 * extract the entire file (due to failure to write the file) we return
644 * the numbers of bytes we did NOT process. This way the caller knows how
645 * many bytes to skip past to find the next archive header. If the failure
646 * was due to an archive read, we will catch that when we try to skip. If
647 * the format supplies a file data crc value, we calculate the actual crc
648 * so that it can be compared to the value stored in the header
650 * We call a special function to write the file. This function attempts to
651 * restore file holes (blocks of zeros) into the file. When files are
652 * sparse this saves space, and is a LOT faster. For non sparse files
653 * the performance hit is small. As of this writing, no archive supports
654 * information on where the file holes are.
656 * 0 ok, -1 if archive read failure. if we cannot write the entire file,
657 * we return a 0 but "left" is set to be the amount unwritten
661 rd_wrfile(ARCHD
*arcn
, int ofd
, off_t
*left
)
664 off_t size
= arcn
->sb
.st_size
;
666 char *fnm
= arcn
->name
;
674 * pass the blocksize of the file being written to the write routine,
675 * if the size is zero, use the default MINFBSZ
678 sz
= PAXPATHLEN
+ 1; /* GNU tar long link/file */
679 else if (fstat(ofd
, &sb
) == 0) {
680 if (sb
.st_blksize
> 0)
681 sz
= (int)sb
.st_blksize
;
683 syswarn(0,errno
,"Unable to obtain block size for file %s",fnm
);
688 * Copy the archive to the file the number of bytes specified. We have
689 * to assume that we want to recover file holes as none of the archive
690 * formats can record the location of file holes.
693 cnt
= bufend
- bufpt
;
695 * if we get a read error, we do not want to skip, as we may
696 * miss a header, so we do not set left, but if we get a write
697 * error, we do want to skip over the unprocessed data.
699 if ((cnt
<= 0) && ((cnt
= buf_fill()) <= 0))
701 cnt
= MIN(cnt
, size
);
702 if ((res
= file_write(ofd
,bufpt
,cnt
,&rem
,&isem
,sz
,fnm
)) <= 0) {
709 * update the actual crc value
713 crc
+= *bufpt
++ & 0xff;
720 * if the last block has a file hole (all zero), we must make sure this
721 * gets updated in the file. We force the last block of zeros to be
722 * written. just closing with the file offset moved forward may not put
723 * a hole at the end of the file.
725 if (isem
&& (arcn
->sb
.st_size
> 0L))
726 file_flush(ofd
, fnm
, isem
);
729 * if we failed from archive read, we do not want to skip
731 if ((size
> 0L) && (*left
== 0L))
735 * some formats record a crc on file data. If so, then we compare the
736 * calculated crc to the crc stored in the archive
738 if (docrc
&& (size
== 0L) && (arcn
->crc
!= crc
))
739 paxwarn(1,"Actual crc does not match expected crc %s",arcn
->name
);
745 * copy the contents of one file to another. used during -rw phase of pax
746 * just as in rd_wrfile() we use a special write function to write the
747 * destination file so we can properly copy files with holes.
751 cp_file(ARCHD
*arcn
, int fd1
, int fd2
)
756 char *fnm
= arcn
->name
;
764 * check for holes in the source file. If none, we will use regular
765 * write instead of file write.
767 if (((off_t
)(arcn
->sb
.st_blocks
* BLKMULT
)) >= arcn
->sb
.st_size
)
771 * pass the blocksize of the file being written to the write routine,
772 * if the size is zero, use the default MINFBSZ
774 if (fstat(fd2
, &sb
) == 0) {
775 if (sb
.st_blksize
> 0)
778 syswarn(0,errno
,"Unable to obtain block size for file %s",fnm
);
782 * read the source file and copy to destination file until EOF
785 if ((cnt
= read(fd1
, buf
, blksz
)) <= 0)
788 res
= write(fd2
, buf
, cnt
);
790 res
= file_write(fd2
, buf
, cnt
, &rem
, &isem
, sz
, fnm
);
797 * check to make sure the copy is valid.
800 syswarn(1, errno
, "Failed write during copy of %s to %s",
801 arcn
->org_name
, arcn
->name
);
802 else if (cpcnt
!= arcn
->sb
.st_size
)
803 paxwarn(1, "File %s changed size during copy to %s",
804 arcn
->org_name
, arcn
->name
);
805 else if (fstat(fd1
, &sb
) < 0)
806 syswarn(1, errno
, "Failed stat of %s", arcn
->org_name
);
807 else if (arcn
->sb
.st_mtime
!= sb
.st_mtime
)
808 paxwarn(1, "File %s was modified during copy to %s",
809 arcn
->org_name
, arcn
->name
);
812 * if the last block has a file hole (all zero), we must make sure this
813 * gets updated in the file. We force the last block of zeros to be
814 * written. just closing with the file offset moved forward may not put
815 * a hole at the end of the file.
817 if (!no_hole
&& isem
&& (arcn
->sb
.st_size
> 0L))
818 file_flush(fd2
, fnm
, isem
);
824 * fill the read buffer with the next record (or what we can get) from
825 * the archive volume.
827 * Number of bytes of data in the read buffer, -1 for read error, and
828 * 0 when finished (user specified termination in ar_next()).
842 * try to fill the buffer. on error the next archive volume is
843 * opened and we try again.
845 if ((cnt
= ar_read(buf
, blksz
)) > 0) {
853 * errors require resync, EOF goes to next archive
869 * force the write buffer to the archive. We are passed the number of
870 * bytes in the buffer at the point of the flush. When we change archives
871 * the record size might change. (either larger or smaller).
873 * 0 if all is ok, -1 when a write error occurs.
877 buf_flush(int bufcnt
)
884 * if we have reached the user specified byte count for each archive
885 * volume, prompt for the next volume. (The non-standard -R flag).
886 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
887 * at least one record. We always round limit UP to next blocksize.
889 if ((wrlimit
> 0) && (wrcnt
> wrlimit
)) {
890 paxwarn(0, "User specified archive volume byte limit reached.");
899 * The new archive volume might have changed the size of the
900 * write blocksize. if so we figure out if we need to write
901 * (one or more times), or if there is now free space left in
902 * the buffer (it is no longer full). bufcnt has the number of
903 * bytes in the buffer, (the blocksize, at the point we were
904 * CALLED). Push has the amount of "extra" data in the buffer
905 * if the block size has shrunk from a volume change.
907 bufend
= buf
+ blksz
;
911 push
= bufcnt
- blksz
;
915 * We have enough data to write at least one archive block
919 * write a block and check if it all went out ok
921 cnt
= ar_write(buf
, blksz
);
929 /* we have extra data to push to the front.
930 * check for more than 1 block of push, and if
931 * so we loop back to write again
933 memcpy(buf
, bufend
, push
);
942 } else if (cnt
> 0) {
944 * Oh drat we got a partial write!
945 * if format does not care about alignment let it go,
946 * we warned the user in ar_write().... but this means
947 * the last record on this volume violates pax spec....
953 memcpy(buf
, bufpt
, cnt
);
955 if (!frmt
->blkalgn
|| ((cnt
% frmt
->blkalgn
) == 0))
961 * All done, go to next archive
968 * The new archive volume might also have changed the block
969 * size. if so, figure out if we have too much or too little
970 * data for using the new block size
972 bufend
= buf
+ blksz
;
976 push
= bufcnt
- blksz
;
980 * write failed, stop pax. we must not create a bad archive!