2 * SPDX-License-Identifier: BSD-3-Clause
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 #include <sys/types.h>
46 * routines which implement archive and file buffering
49 #define MINFBSZ 512 /* default block size for hole detect */
50 #define MAXFLT 10 /* default media read error limit */
53 * Need to change bufmem to dynamic allocation when the upper
54 * limit on blocking size is removed (though that will violate pax spec)
55 * MAXBLK define and tests will also need to be updated.
57 static char bufmem[MAXBLK+BLKMULT]; /* i/o buffer + pushback id space */
58 static char *buf; /* normal start of i/o buffer */
59 static char *bufend; /* end or last char in i/o buffer */
60 static char *bufpt; /* read/write point in i/o buffer */
61 int blksz = MAXBLK; /* block input/output size in bytes */
62 int wrblksz; /* user spec output size in bytes */
63 int maxflt = MAXFLT; /* MAX consecutive media errors */
64 int rdblksz; /* first read blksize (tapes only) */
65 off_t wrlimit; /* # of bytes written per archive vol */
66 off_t wrcnt; /* # of bytes written on current vol */
67 off_t rdcnt; /* # of bytes read on current vol */
71 * set up the buffering system to operate in a write mode
73 * 0 if ok, -1 if the user specified write block size violates pax spec
79 buf = &(bufmem[BLKMULT]);
81 * Check to make sure the write block size meets pax specs. If the user
82 * does not specify a blocksize, we use the format default blocksize.
83 * We must be picky on writes, so we do not allow the user to create an
84 * archive that might be hard to read elsewhere. If all ok, we then
85 * open the first archive volume
89 if (wrblksz > MAXBLK) {
90 paxwarn(1, "Write block size of %d too large, maximum is: %d",
94 if (wrblksz % BLKMULT) {
95 paxwarn(1, "Write block size of %d is not a %d byte multiple",
99 if (wrblksz > MAXBLK_POSIX) {
100 paxwarn(0, "Write block size of %d larger than POSIX max %d, archive may not be portable",
101 wrblksz, MAXBLK_POSIX);
106 * we only allow wrblksz to be used with all archive operations
108 blksz = rdblksz = wrblksz;
109 if ((ar_open(arcname) < 0) && (ar_next() < 0))
112 bufend = buf + wrblksz;
119 * set up buffering system to read an archive
121 * 0 if ok, -1 otherwise
128 * leave space for the header pushback (see get_arc()). If we are
129 * going to append and user specified a write block size, check it
132 buf = &(bufmem[BLKMULT]);
133 if ((act == APPND) && wrblksz) {
134 if (wrblksz > MAXBLK) {
135 paxwarn(1,"Write block size %d too large, maximum is: %d",
139 if (wrblksz % BLKMULT) {
140 paxwarn(1, "Write block size %d is not a %d byte multiple",
149 if ((ar_open(arcname) < 0) && (ar_next() < 0))
151 bufend = buf + rdblksz;
159 * set up buffer system for copying within the file system
165 buf = &(bufmem[BLKMULT]);
166 rdblksz = blksz = MAXBLK;
171 * Set up the buffering system to append new members to an archive that
172 * was just read. The last block(s) of an archive may contain a format
173 * specific trailer. To append a new member, this trailer has to be
174 * removed from the archive. The first byte of the trailer is replaced by
175 * the start of the header of the first file added to the archive. The
176 * format specific end read function tells us how many bytes to move
177 * backwards in the archive to be positioned BEFORE the trailer. Two
178 * different positions have to be adjusted, the O.S. file offset (e.g. the
179 * position of the tape head) and the write point within the data we have
180 * stored in the read (soon to become write) buffer. We may have to move
181 * back several records (the number depends on the size of the archive
182 * record and the size of the format trailer) to read up the record where
183 * the first byte of the trailer is recorded. Trailers may span (and
184 * overlap) record boundaries.
185 * We first calculate which record has the first byte of the trailer. We
186 * move the OS file offset back to the start of this record and read it
187 * up. We set the buffer write pointer to be at this byte (the byte where
188 * the trailer starts). We then move the OS file pointer back to the
189 * start of this record so a flush of this buffer will replace the record
191 * A major problem is rewriting this last record. For archives stored
192 * on disk files, this is trivial. However, many devices are really picky
193 * about the conditions under which they will allow a write to occur.
194 * Often devices restrict the conditions where writes can be made,
195 * so it may not be feasible to append archives stored on all types of
198 * 0 for success, -1 for failure
202 appnd_start(off_t skcnt)
208 paxwarn(0, "Cannot append to an archive that may have flaws.");
212 * if the user did not specify a write blocksize, inherit the size used
213 * in the last archive volume read. (If a is set we still use rdblksz
214 * until next volume, cannot shift sizes within a single volume).
217 wrblksz = blksz = rdblksz;
222 * make sure that this volume allows appends
228 * Calculate bytes to move back and move in front of record where we
229 * need to start writing from. Remember we have to add in any padding
230 * that might be in the buffer after the trailer in the last block. We
231 * travel skcnt + padding ROUNDED UP to blksize.
233 skcnt += bufend - bufpt;
234 if ((cnt = (skcnt/blksz) * blksz) < skcnt)
236 if (ar_rev((off_t)cnt) < 0)
240 * We may have gone too far if there is valid data in the block we are
241 * now in front of, read up the block and position the pointer after
244 if ((cnt -= skcnt) > 0) {
246 * watch out for stupid tape drives. ar_rev() will set rdblksz
247 * to be real physical blocksize so we must loop until we get
248 * the old rdblksz (now in blksz). If ar_rev() fouls up the
249 * determination of the physical block size, we will fail.
252 bufend = buf + blksz;
253 while (bufpt < bufend) {
254 if ((res = ar_read(bufpt, rdblksz)) <= 0)
258 if (ar_rev((off_t)(bufpt - buf)) < 0)
261 bufend = buf + blksz;
266 bufend = buf + blksz;
274 * At this point we are ready to write. If the device requires special
275 * handling to write at a point were previously recorded data resides,
276 * that is handled in ar_set_wr(). From now on we operate under normal
277 * ARCHIVE mode (write) conditions
285 paxwarn(1, "Unable to rewrite archive trailer, cannot append.");
291 * A read error occurred on this archive volume. Resync the buffer and
292 * try to reset the device (if possible) so we can continue to read. Keep
293 * trying to do this until we get a valid read, or we reach the limit on
294 * consecutive read faults (at which point we give up). The user can
295 * adjust the read error limit through a command line option.
297 * 0 on success, and -1 on failure
307 * if the user says bail out on first fault, we are out of here...
312 paxwarn(1, "Unable to append when there are archive read errors.");
317 * poke at device and try to get past media error
319 if (ar_rdsync() < 0) {
327 if ((res = ar_read(buf, blksz)) > 0) {
329 * All right! got some data, fill that buffer
338 * Oh well, yet another failed read...
339 * if error limit reached, ditch. o.w. poke device to move past
340 * bad media and try again. if media is badly damaged, we ask
341 * the poor (and upset user at this point) for the next archive
342 * volume. remember the goal on reads is to get the most we
343 * can extract out of the archive.
345 if ((maxflt > 0) && (++errcnt > maxflt))
346 paxwarn(0,"Archive read error limit (%d) reached",maxflt);
347 else if (ar_rdsync() == 0)
359 * push the data used during the archive id phase back into the I/O
360 * buffer. This is required as we cannot be sure that the header does NOT
361 * overlap a block boundary (as in the case we are trying to recover a
362 * flawed archived). This was not designed to be used for any other
363 * purpose. (What software engineering, HA!)
364 * WARNING: do not even THINK of pback greater than BLKMULT, unless the
365 * pback space is increased.
369 pback(char *pt, int cnt)
372 memcpy(bufpt, pt, cnt);
378 * skip forward in the archive during an archive read. Used to get quickly
379 * past file data and padding for files the user did NOT select.
381 * 0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
392 * consume what data we have in the buffer. If we have to move forward
393 * whole records, we call the low level skip function to see if we can
394 * move within the archive without doing the expensive reads on data we
399 res = MIN((bufend - bufpt), skcnt);
404 * if skcnt is now 0, then no additional i/o is needed
410 * We have to read more, calculate complete and partial record reads
411 * based on rdblksz. we skip over "cnt" complete records
414 cnt = (skcnt/rdblksz) * rdblksz;
417 * if the skip fails, we will have to resync. ar_fow will tell us
418 * how much it can skip over. We will have to read the rest.
420 if (ar_fow(cnt, &skipped) < 0)
422 res += cnt - skipped;
426 * what is left we have to read (which may be the whole thing if
427 * ar_fow() told us the device can only read to skip records);
430 cnt = bufend - bufpt;
432 * if the read fails, we will have to resync
434 if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
447 * flush out any data (and pad if required) the last block. We always pad
448 * with zero (even though we do not have to). Padding with 0 makes it a
449 * lot easier to recover if the archive is damaged. zero padding SHOULD
450 * BE a requirement....
457 memset(bufpt, 0, bufend - bufpt);
459 (void)buf_flush(blksz);
465 * fill the write buffer from data passed to it in a buffer (usually used
466 * by format specific write routines to pass a file header). On failure we
467 * punt. We do not allow the user to continue to write flawed archives.
468 * We assume these headers are not very large (the memory copy we use is
471 * 0 if buffer was filled ok, -1 o.w. (buffer flush failure)
475 wr_rdbuf(char *out, int outcnt)
480 * while there is data to copy into the write buffer. when the
481 * write buffer fills, flush it to the archive and continue
484 cnt = bufend - bufpt;
485 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
488 * only move what we have space for
490 cnt = MIN(cnt, outcnt);
491 memcpy(bufpt, out, cnt);
501 * copy from the read buffer into a supplied buffer a specified number of
502 * bytes. If the read buffer is empty fill it and continue to copy.
503 * usually used to obtain a file header for processing by a format
504 * specific read routine.
506 * number of bytes copied to the buffer, 0 indicates EOF on archive volume,
511 rd_wrbuf(char *in, int cpcnt)
518 * loop until we fill the buffer with the requested number of bytes
521 cnt = bufend - bufpt;
522 if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
524 * read error, return what we got (or the error if
525 * no data was copied). The caller must know that an
526 * error occurred and has the best knowledge what to
529 if ((res = cpcnt - incnt) > 0)
535 * calculate how much data to copy based on what's left and
538 cnt = MIN(cnt, incnt);
539 memcpy(in, bufpt, cnt);
549 * skip forward during a write. In other words add padding to the file.
550 * we add zero filled padding as it makes flawed archives much easier to
551 * recover from. the caller tells us how many bytes of padding to add
552 * This routine was not designed to add HUGE amount of padding, just small
553 * amounts (a few 512 byte blocks at most)
555 * 0 if ok, -1 if there was a buf_flush failure
564 * loop while there is more padding to add
567 cnt = bufend - bufpt;
568 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
570 cnt = MIN(cnt, skcnt);
571 memset(bufpt, 0, cnt);
580 * fill write buffer with the contents of a file. We are passed an open
581 * file descriptor to the file and the archive structure that describes the
582 * file we are storing. The variable "left" is modified to contain the
583 * number of bytes of the file we were NOT able to write to the archive.
584 * it is important that we always write EXACTLY the number of bytes that
585 * the format specific write routine told us to. The file can also get
586 * bigger, so reading to the end of file would create an improper archive,
587 * we just detect this case and warn the user. We never create a bad
588 * archive if we can avoid it. Of course trying to archive files that are
589 * active is asking for trouble. It we fail, we pass back how much we
590 * could NOT copy and let the caller deal with it.
592 * 0 ok, -1 if archive write failure. a short read of the file returns a
593 * 0, but "left" is set to be greater than zero.
597 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
601 off_t size = arcn->sb.st_size;
605 * while there are more bytes to write
608 cnt = bufend - bufpt;
609 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
613 cnt = MIN(cnt, size);
614 if ((res = read(ifd, bufpt, cnt)) <= 0)
621 * better check the file did not change during this operation
622 * or the file read failed.
625 syswarn(1, errno, "Read fault on %s", arcn->org_name);
627 paxwarn(1, "File changed size during read %s", arcn->org_name);
628 else if (fstat(ifd, &sb) < 0)
629 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
630 else if (arcn->sb.st_mtime != sb.st_mtime)
631 paxwarn(1, "File %s was modified during copy to archive",
639 * extract the contents of a file from the archive. If we are unable to
640 * extract the entire file (due to failure to write the file) we return
641 * the numbers of bytes we did NOT process. This way the caller knows how
642 * many bytes to skip past to find the next archive header. If the failure
643 * was due to an archive read, we will catch that when we try to skip. If
644 * the format supplies a file data crc value, we calculate the actual crc
645 * so that it can be compared to the value stored in the header
647 * We call a special function to write the file. This function attempts to
648 * restore file holes (blocks of zeros) into the file. When files are
649 * sparse this saves space, and is a LOT faster. For non sparse files
650 * the performance hit is small. As of this writing, no archive supports
651 * information on where the file holes are.
653 * 0 ok, -1 if archive read failure. if we cannot write the entire file,
654 * we return a 0 but "left" is set to be the amount unwritten
658 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
661 off_t size = arcn->sb.st_size;
663 char *fnm = arcn->name;
671 * pass the blocksize of the file being written to the write routine,
672 * if the size is zero, use the default MINFBSZ
674 if (fstat(ofd, &sb) == 0) {
675 if (sb.st_blksize > 0)
676 sz = (int)sb.st_blksize;
678 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
683 * Copy the archive to the file the number of bytes specified. We have
684 * to assume that we want to recover file holes as none of the archive
685 * formats can record the location of file holes.
688 cnt = bufend - bufpt;
690 * if we get a read error, we do not want to skip, as we may
691 * miss a header, so we do not set left, but if we get a write
692 * error, we do want to skip over the unprocessed data.
694 if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
696 cnt = MIN(cnt, size);
697 if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
704 * update the actual crc value
708 crc += *bufpt++ & 0xff;
715 * if the last block has a file hole (all zero), we must make sure this
716 * gets updated in the file. We force the last block of zeros to be
717 * written. just closing with the file offset moved forward may not put
718 * a hole at the end of the file.
720 if (isem && (arcn->sb.st_size > 0L))
721 file_flush(ofd, fnm, isem);
724 * if we failed from archive read, we do not want to skip
726 if ((size > 0L) && (*left == 0L))
730 * some formats record a crc on file data. If so, then we compare the
731 * calculated crc to the crc stored in the archive
733 if (docrc && (size == 0L) && (arcn->crc != crc))
734 paxwarn(1,"Actual crc does not match expected crc %s",arcn->name);
740 * copy the contents of one file to another. used during -rw phase of pax
741 * just as in rd_wrfile() we use a special write function to write the
742 * destination file so we can properly copy files with holes.
746 cp_file(ARCHD *arcn, int fd1, int fd2)
751 char *fnm = arcn->name;
759 * check for holes in the source file. If none, we will use regular
760 * write instead of file write.
762 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
766 * pass the blocksize of the file being written to the write routine,
767 * if the size is zero, use the default MINFBSZ
769 if (fstat(fd2, &sb) == 0) {
770 if (sb.st_blksize > 0)
773 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
777 * read the source file and copy to destination file until EOF
780 if ((cnt = read(fd1, buf, blksz)) <= 0)
783 res = write(fd2, buf, cnt);
785 res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
792 * check to make sure the copy is valid.
795 syswarn(1, errno, "Failed write during copy of %s to %s",
796 arcn->org_name, arcn->name);
797 else if (cpcnt != arcn->sb.st_size)
798 paxwarn(1, "File %s changed size during copy to %s",
799 arcn->org_name, arcn->name);
800 else if (fstat(fd1, &sb) < 0)
801 syswarn(1, errno, "Failed stat of %s", arcn->org_name);
802 else if (arcn->sb.st_mtime != sb.st_mtime)
803 paxwarn(1, "File %s was modified during copy to %s",
804 arcn->org_name, arcn->name);
807 * if the last block has a file hole (all zero), we must make sure this
808 * gets updated in the file. We force the last block of zeros to be
809 * written. just closing with the file offset moved forward may not put
810 * a hole at the end of the file.
812 if (!no_hole && isem && (arcn->sb.st_size > 0L))
813 file_flush(fd2, fnm, isem);
819 * fill the read buffer with the next record (or what we can get) from
820 * the archive volume.
822 * Number of bytes of data in the read buffer, -1 for read error, and
823 * 0 when finished (user specified termination in ar_next()).
837 * try to fill the buffer. on error the next archive volume is
838 * opened and we try again.
840 if ((cnt = ar_read(buf, blksz)) > 0) {
848 * errors require resync, EOF goes to next archive
849 * but in case we have not determined yet the format,
850 * this means that we have a very short file, so we
855 if (frmt == NULL || ar_next() < 0) {
867 * force the write buffer to the archive. We are passed the number of
868 * bytes in the buffer at the point of the flush. When we change archives
869 * the record size might change. (either larger or smaller).
871 * 0 if all is ok, -1 when a write error occurs.
875 buf_flush(int bufcnt)
882 * if we have reached the user specified byte count for each archive
883 * volume, prompt for the next volume. (The non-standard -R flag).
884 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
885 * at least one record. We always round limit UP to next blocksize.
887 if ((wrlimit > 0) && (wrcnt > wrlimit)) {
888 paxwarn(0, "User specified archive volume byte limit reached.");
897 * The new archive volume might have changed the size of the
898 * write blocksize. if so we figure out if we need to write
899 * (one or more times), or if there is now free space left in
900 * the buffer (it is no longer full). bufcnt has the number of
901 * bytes in the buffer, (the blocksize, at the point we were
902 * CALLED). Push has the amount of "extra" data in the buffer
903 * if the block size has shrunk from a volume change.
905 bufend = buf + blksz;
909 push = bufcnt - blksz;
913 * We have enough data to write at least one archive block
917 * write a block and check if it all went out ok
919 cnt = ar_write(buf, blksz);
927 /* we have extra data to push to the front.
928 * check for more than 1 block of push, and if
929 * so we loop back to write again
931 memcpy(buf, bufend, push);
940 } else if (cnt > 0) {
942 * Oh drat we got a partial write!
943 * if format doesn't care about alignment let it go,
944 * we warned the user in ar_write().... but this means
945 * the last record on this volume violates pax spec....
951 memcpy(buf, bufpt, cnt);
953 if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
959 * All done, go to next archive
966 * The new archive volume might also have changed the block
967 * size. if so, figure out if we have too much or too little
968 * data for using the new block size
970 bufend = buf + blksz;
974 push = bufcnt - blksz;
978 * write failed, stop pax. we must not create a bad archive!