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
38 static char sccsid[] = "@(#)buf_subs.c 8.2 (Berkeley) 4/18/94";
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
44 #include <sys/types.h>
54 * routines which implement archive and file buffering
57 #define MINFBSZ 512 /* default block size for hole detect */
58 #define MAXFLT 10 /* default media read error limit */
61 * Need to change bufmem to dynamic allocation when the upper
62 * limit on blocking size is removed (though that will violate pax spec)
63 * MAXBLK define and tests will also need to be updated.
65 static char bufmem[MAXBLK+BLKMULT]; /* i/o buffer + pushback id space */
66 static char *buf; /* normal start of i/o buffer */
67 static char *bufend; /* end or last char in i/o buffer */
68 static char *bufpt; /* read/write point in i/o buffer */
69 int blksz = MAXBLK; /* block input/output size in bytes */
70 int wrblksz; /* user spec output size in bytes */
71 int maxflt = MAXFLT; /* MAX consecutive media errors */
72 int rdblksz; /* first read blksize (tapes only) */
73 off_t wrlimit; /* # of bytes written per archive vol */
74 off_t wrcnt; /* # of bytes written on current vol */
75 off_t rdcnt; /* # of bytes read on current vol */
79 * set up the buffering system to operate in a write mode
81 * 0 if ok, -1 if the user specified write block size violates pax spec
87 buf = &(bufmem[BLKMULT]);
89 * Check to make sure the write block size meets pax specs. If the user
90 * does not specify a blocksize, we use the format default blocksize.
91 * We must be picky on writes, so we do not allow the user to create an
92 * archive that might be hard to read elsewhere. If all ok, we then
93 * open the first archive volume
97 if (wrblksz > MAXBLK) {
98 paxwarn(1, "Write block size of %d too large, maximum is: %d",
102 if (wrblksz % BLKMULT) {
103 paxwarn(1, "Write block size of %d is not a %d byte multiple",
107 if (wrblksz > MAXBLK_POSIX) {
108 paxwarn(0, "Write block size of %d larger than POSIX max %d, archive may not be portable",
109 wrblksz, MAXBLK_POSIX);
114 * we only allow wrblksz to be used with all archive operations
116 blksz = rdblksz = wrblksz;
117 if ((ar_open(arcname) < 0) && (ar_next() < 0))
120 bufend = buf + wrblksz;
127 * set up buffering system to read an archive
129 * 0 if ok, -1 otherwise
136 * leave space for the header pushback (see get_arc()). If we are
137 * going to append and user specified a write block size, check it
140 buf = &(bufmem[BLKMULT]);
141 if ((act == APPND) && wrblksz) {
142 if (wrblksz > MAXBLK) {
143 paxwarn(1,"Write block size %d too large, maximum is: %d",
147 if (wrblksz % BLKMULT) {
148 paxwarn(1, "Write block size %d is not a %d byte multiple",
157 if ((ar_open(arcname) < 0) && (ar_next() < 0))
159 bufend = buf + rdblksz;
167 * set up buffer system for copying within the file system
173 buf = &(bufmem[BLKMULT]);
174 rdblksz = blksz = MAXBLK;
179 * Set up the buffering system to append new members to an archive that
180 * was just read. The last block(s) of an archive may contain a format
181 * specific trailer. To append a new member, this trailer has to be
182 * removed from the archive. The first byte of the trailer is replaced by
183 * the start of the header of the first file added to the archive. The
184 * format specific end read function tells us how many bytes to move
185 * backwards in the archive to be positioned BEFORE the trailer. Two
186 * different positions have to be adjusted, the O.S. file offset (e.g. the
187 * position of the tape head) and the write point within the data we have
188 * stored in the read (soon to become write) buffer. We may have to move
189 * back several records (the number depends on the size of the archive
190 * record and the size of the format trailer) to read up the record where
191 * the first byte of the trailer is recorded. Trailers may span (and
192 * overlap) record boundaries.
193 * We first calculate which record has the first byte of the trailer. We
194 * move the OS file offset back to the start of this record and read it
195 * up. We set the buffer write pointer to be at this byte (the byte where
196 * the trailer starts). We then move the OS file pointer back to the
197 * start of this record so a flush of this buffer will replace the record
199 * A major problem is rewriting this last record. For archives stored
200 * on disk files, this is trivial. However, many devices are really picky
201 * about the conditions under which they will allow a write to occur.
202 * Often devices restrict the conditions where writes can be made writes,
203 * so it may not be feasible to append archives stored on all types of
206 * 0 for success, -1 for failure
210 appnd_start(off_t skcnt)
216 paxwarn(0, "Cannot append to an archive that may have flaws.");
220 * if the user did not specify a write blocksize, inherit the size used
221 * in the last archive volume read. (If a is set we still use rdblksz
222 * until next volume, cannot shift sizes within a single volume).
225 wrblksz = blksz = rdblksz;
230 * make sure that this volume allows appends
236 * Calculate bytes to move back and move in front of record where we
237 * need to start writing from. Remember we have to add in any padding
238 * that might be in the buffer after the trailer in the last block. We
239 * travel skcnt + padding ROUNDED UP to blksize.
241 skcnt += bufend - bufpt;
242 if ((cnt = (skcnt/blksz) * blksz) < skcnt)
244 if (ar_rev((off_t)cnt) < 0)
248 * We may have gone too far if there is valid data in the block we are
249 * now in front of, read up the block and position the pointer after
252 if ((cnt -= skcnt) > 0) {
254 * watch out for stupid tape drives. ar_rev() will set rdblksz
255 * to be real physical blocksize so we must loop until we get
256 * the old rdblksz (now in blksz). If ar_rev() fouls up the
257 * determination of the physical block size, we will fail.
260 bufend = buf + blksz;
261 while (bufpt < bufend) {
262 if ((res = ar_read(bufpt, rdblksz)) <= 0)
266 if (ar_rev((off_t)(bufpt - buf)) < 0)
269 bufend = buf + blksz;
274 bufend = buf + blksz;
282 * At this point we are ready to write. If the device requires special
283 * handling to write at a point were previously recorded data resides,
284 * that is handled in ar_set_wr(). From now on we operate under normal
285 * ARCHIVE mode (write) conditions
293 paxwarn(1, "Unable to rewrite archive trailer, cannot append.");
299 * A read error occurred on this archive volume. Resync the buffer and
300 * try to reset the device (if possible) so we can continue to read. Keep
301 * trying to do this until we get a valid read, or we reach the limit on
302 * consecutive read faults (at which point we give up). The user can
303 * adjust the read error limit through a command line option.
305 * 0 on success, and -1 on failure
315 * if the user says bail out on first fault, we are out of here...
320 paxwarn(1, "Unable to append when there are archive read errors.");
325 * poke at device and try to get past media error
327 if (ar_rdsync() < 0) {
335 if ((res = ar_read(buf, blksz)) > 0) {
337 * All right! got some data, fill that buffer
346 * Oh well, yet another failed read...
347 * if error limit reached, ditch. o.w. poke device to move past
348 * bad media and try again. if media is badly damaged, we ask
349 * the poor (and upset user at this point) for the next archive
350 * volume. remember the goal on reads is to get the most we
351 * can extract out of the archive.
353 if ((maxflt > 0) && (++errcnt > maxflt))
354 paxwarn(0,"Archive read error limit (%d) reached",maxflt);
355 else if (ar_rdsync() == 0)
367 * push the data used during the archive id phase back into the I/O
368 * buffer. This is required as we cannot be sure that the header does NOT
369 * overlap a block boundary (as in the case we are trying to recover a
370 * flawed archived). This was not designed to be used for any other
371 * purpose. (What software engineering, HA!)
372 * WARNING: do not even THINK of pback greater than BLKMULT, unless the
373 * pback space is increased.
377 pback(char *pt, int cnt)
380 memcpy(bufpt, pt, cnt);
386 * skip forward in the archive during an archive read. Used to get quickly
387 * past file data and padding for files the user did NOT select.
389 * 0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
400 * consume what data we have in the buffer. If we have to move forward
401 * whole records, we call the low level skip function to see if we can
402 * move within the archive without doing the expensive reads on data we
407 res = MIN((bufend - bufpt), skcnt);
412 * if skcnt is now 0, then no additional i/o is needed
418 * We have to read more, calculate complete and partial record reads
419 * based on rdblksz. we skip over "cnt" complete records
422 cnt = (skcnt/rdblksz) * rdblksz;
425 * if the skip fails, we will have to resync. ar_fow will tell us
426 * how much it can skip over. We will have to read the rest.
428 if (ar_fow(cnt, &skipped) < 0)
430 res += cnt - skipped;
434 * what is left we have to read (which may be the whole thing if
435 * ar_fow() told us the device can only read to skip records);
438 cnt = bufend - bufpt;
440 * if the read fails, we will have to resync
442 if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
455 * flush out any data (and pad if required) the last block. We always pad
456 * with zero (even though we do not have to). Padding with 0 makes it a
457 * lot easier to recover if the archive is damaged. zero padding SHOULD
458 * BE a requirement....
465 memset(bufpt, 0, bufend - bufpt);
467 (void)buf_flush(blksz);
473 * fill the write buffer from data passed to it in a buffer (usually used
474 * by format specific write routines to pass a file header). On failure we
475 * punt. We do not allow the user to continue to write flawed archives.
476 * We assume these headers are not very large (the memory copy we use is
479 * 0 if buffer was filled ok, -1 o.w. (buffer flush failure)
483 wr_rdbuf(char *out, int outcnt)
488 * while there is data to copy into the write buffer. when the
489 * write buffer fills, flush it to the archive and continue
492 cnt = bufend - bufpt;
493 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
496 * only move what we have space for
498 cnt = MIN(cnt, outcnt);
499 memcpy(bufpt, out, cnt);
509 * copy from the read buffer into a supplied buffer a specified number of
510 * bytes. If the read buffer is empty fill it and continue to copy.
511 * usually used to obtain a file header for processing by a format
512 * specific read routine.
514 * number of bytes copied to the buffer, 0 indicates EOF on archive volume,
519 rd_wrbuf(char *in, int cpcnt)
526 * loop until we fill the buffer with the requested number of bytes
529 cnt = bufend - bufpt;
530 if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
532 * read error, return what we got (or the error if
533 * no data was copied). The caller must know that an
534 * error occurred and has the best knowledge what to
537 if ((res = cpcnt - incnt) > 0)
543 * calculate how much data to copy based on whats left and
546 cnt = MIN(cnt, incnt);
547 memcpy(in, bufpt, cnt);
557 * skip forward during a write. In other words add padding to the file.
558 * we add zero filled padding as it makes flawed archives much easier to
559 * recover from. the caller tells us how many bytes of padding to add
560 * This routine was not designed to add HUGE amount of padding, just small
561 * amounts (a few 512 byte blocks at most)
563 * 0 if ok, -1 if there was a buf_flush failure
572 * loop while there is more padding to add
575 cnt = bufend - bufpt;
576 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
578 cnt = MIN(cnt, skcnt);
579 memset(bufpt, 0, cnt);
588 * fill write buffer with the contents of a file. We are passed an open
589 * file descriptor to the file and the archive structure that describes the
590 * file we are storing. The variable "left" is modified to contain the
591 * number of bytes of the file we were NOT able to write to the archive.
592 * it is important that we always write EXACTLY the number of bytes that
593 * the format specific write routine told us to. The file can also get
594 * bigger, so reading to the end of file would create an improper archive,
595 * we just detect this case and warn the user. We never create a bad
596 * archive if we can avoid it. Of course trying to archive files that are
597 * active is asking for trouble. It we fail, we pass back how much we
598 * could NOT copy and let the caller deal with it.
600 * 0 ok, -1 if archive write failure. a short read of the file returns a
601 * 0, but "left" is set to be greater than zero.
605 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
609 off_t size = arcn->sb.st_size;
613 * while there are more bytes to write
616 cnt = bufend - bufpt;
617 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
621 cnt = MIN(cnt, size);
622 if ((res = read(ifd, bufpt, cnt)) <= 0)
629 * better check the file did not change during this operation
630 * or the file read failed.
633 syswarn(1, errno, "Read fault on %s", arcn->org_name);
635 paxwarn(1, "File changed size during read %s", arcn->org_name);
636 else if (fstat(ifd, &sb) < 0)
637 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
638 else if (arcn->sb.st_mtime != sb.st_mtime)
639 paxwarn(1, "File %s was modified during copy to archive",
647 * extract the contents of a file from the archive. If we are unable to
648 * extract the entire file (due to failure to write the file) we return
649 * the numbers of bytes we did NOT process. This way the caller knows how
650 * many bytes to skip past to find the next archive header. If the failure
651 * was due to an archive read, we will catch that when we try to skip. If
652 * the format supplies a file data crc value, we calculate the actual crc
653 * so that it can be compared to the value stored in the header
655 * We call a special function to write the file. This function attempts to
656 * restore file holes (blocks of zeros) into the file. When files are
657 * sparse this saves space, and is a LOT faster. For non sparse files
658 * the performance hit is small. As of this writing, no archive supports
659 * information on where the file holes are.
661 * 0 ok, -1 if archive read failure. if we cannot write the entire file,
662 * we return a 0 but "left" is set to be the amount unwritten
666 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
669 off_t size = arcn->sb.st_size;
671 char *fnm = arcn->name;
679 * pass the blocksize of the file being written to the write routine,
680 * if the size is zero, use the default MINFBSZ
682 if (fstat(ofd, &sb) == 0) {
683 if (sb.st_blksize > 0)
684 sz = (int)sb.st_blksize;
686 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
691 * Copy the archive to the file the number of bytes specified. We have
692 * to assume that we want to recover file holes as none of the archive
693 * formats can record the location of file holes.
696 cnt = bufend - bufpt;
698 * if we get a read error, we do not want to skip, as we may
699 * miss a header, so we do not set left, but if we get a write
700 * error, we do want to skip over the unprocessed data.
702 if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
704 cnt = MIN(cnt, size);
705 if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
712 * update the actual crc value
716 crc += *bufpt++ & 0xff;
723 * if the last block has a file hole (all zero), we must make sure this
724 * gets updated in the file. We force the last block of zeros to be
725 * written. just closing with the file offset moved forward may not put
726 * a hole at the end of the file.
728 if (isem && (arcn->sb.st_size > 0L))
729 file_flush(ofd, fnm, isem);
732 * if we failed from archive read, we do not want to skip
734 if ((size > 0L) && (*left == 0L))
738 * some formats record a crc on file data. If so, then we compare the
739 * calculated crc to the crc stored in the archive
741 if (docrc && (size == 0L) && (arcn->crc != crc))
742 paxwarn(1,"Actual crc does not match expected crc %s",arcn->name);
748 * copy the contents of one file to another. used during -rw phase of pax
749 * just as in rd_wrfile() we use a special write function to write the
750 * destination file so we can properly copy files with holes.
754 cp_file(ARCHD *arcn, int fd1, int fd2)
759 char *fnm = arcn->name;
767 * check for holes in the source file. If none, we will use regular
768 * write instead of file write.
770 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
774 * pass the blocksize of the file being written to the write routine,
775 * if the size is zero, use the default MINFBSZ
777 if (fstat(fd2, &sb) == 0) {
778 if (sb.st_blksize > 0)
781 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
785 * read the source file and copy to destination file until EOF
788 if ((cnt = read(fd1, buf, blksz)) <= 0)
791 res = write(fd2, buf, cnt);
793 res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
800 * check to make sure the copy is valid.
803 syswarn(1, errno, "Failed write during copy of %s to %s",
804 arcn->org_name, arcn->name);
805 else if (cpcnt != arcn->sb.st_size)
806 paxwarn(1, "File %s changed size during copy to %s",
807 arcn->org_name, arcn->name);
808 else if (fstat(fd1, &sb) < 0)
809 syswarn(1, errno, "Failed stat of %s", arcn->org_name);
810 else if (arcn->sb.st_mtime != sb.st_mtime)
811 paxwarn(1, "File %s was modified during copy to %s",
812 arcn->org_name, arcn->name);
815 * if the last block has a file hole (all zero), we must make sure this
816 * gets updated in the file. We force the last block of zeros to be
817 * written. just closing with the file offset moved forward may not put
818 * a hole at the end of the file.
820 if (!no_hole && isem && (arcn->sb.st_size > 0L))
821 file_flush(fd2, fnm, isem);
827 * fill the read buffer with the next record (or what we can get) from
828 * the archive volume.
830 * Number of bytes of data in the read buffer, -1 for read error, and
831 * 0 when finished (user specified termination in ar_next()).
845 * try to fill the buffer. on error the next archive volume is
846 * opened and we try again.
848 if ((cnt = ar_read(buf, blksz)) > 0) {
856 * errors require resync, EOF goes to next archive
857 * but in case we have not determined yet the format,
858 * this means that we have a very short file, so we
863 if (frmt == NULL || ar_next() < 0) {
875 * force the write buffer to the archive. We are passed the number of
876 * bytes in the buffer at the point of the flush. When we change archives
877 * the record size might change. (either larger or smaller).
879 * 0 if all is ok, -1 when a write error occurs.
883 buf_flush(int bufcnt)
890 * if we have reached the user specified byte count for each archive
891 * volume, prompt for the next volume. (The non-standard -R flag).
892 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
893 * at least one record. We always round limit UP to next blocksize.
895 if ((wrlimit > 0) && (wrcnt > wrlimit)) {
896 paxwarn(0, "User specified archive volume byte limit reached.");
905 * The new archive volume might have changed the size of the
906 * write blocksize. if so we figure out if we need to write
907 * (one or more times), or if there is now free space left in
908 * the buffer (it is no longer full). bufcnt has the number of
909 * bytes in the buffer, (the blocksize, at the point we were
910 * CALLED). Push has the amount of "extra" data in the buffer
911 * if the block size has shrunk from a volume change.
913 bufend = buf + blksz;
917 push = bufcnt - blksz;
921 * We have enough data to write at least one archive block
925 * write a block and check if it all went out ok
927 cnt = ar_write(buf, blksz);
935 /* we have extra data to push to the front.
936 * check for more than 1 block of push, and if
937 * so we loop back to write again
939 memcpy(buf, bufend, push);
948 } else if (cnt > 0) {
950 * Oh drat we got a partial write!
951 * if format doesn't care about alignment let it go,
952 * we warned the user in ar_write().... but this means
953 * the last record on this volume violates pax spec....
959 memcpy(buf, bufpt, cnt);
961 if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
967 * All done, go to next archive
974 * The new archive volume might also have changed the block
975 * size. if so, figure out if we have too much or too little
976 * data for using the new block size
978 bufend = buf + blksz;
982 push = bufcnt - blksz;
986 * write failed, stop pax. we must not create a bad archive!