2 * Copyright (c) 1992 Keith Muller.
3 * Copyright (c) 1992, 1993
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
7 * Keith Muller of the University of California, San Diego.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)buf_subs.c 8.2 (Berkeley) 4/18/94
34 * $FreeBSD: src/bin/pax/buf_subs.c,v 1.12.2.1 2001/08/01 05:03:11 obrien Exp $
37 #include <sys/param.h>
48 * routines which implement archive and file buffering
51 #define MINFBSZ 512 /* default block size for hole detect */
52 #define MAXFLT 10 /* default media read error limit */
55 * Need to change bufmem to dynamic allocation when the upper
56 * limit on blocking size is removed (though that will violate pax spec)
57 * MAXBLK define and tests will also need to be updated.
59 static char bufmem[MAXBLK+BLKMULT]; /* i/o buffer + pushback id space */
60 static char *buf; /* normal start of i/o buffer */
61 static char *bufend; /* end or last char in i/o buffer */
62 static char *bufpt; /* read/write point in i/o buffer */
63 int blksz = MAXBLK; /* block input/output size in bytes */
64 int wrblksz; /* user spec output size in bytes */
65 int maxflt = MAXFLT; /* MAX consecutive media errors */
66 int rdblksz; /* first read blksize (tapes only) */
67 off_t wrlimit; /* # of bytes written per archive vol */
68 off_t wrcnt; /* # of bytes written on current vol */
69 off_t rdcnt; /* # of bytes read on current vol */
73 * set up the buffering system to operate in a write mode
75 * 0 if ok, -1 if the user specified write block size violates pax spec
81 buf = &(bufmem[BLKMULT]);
83 * Check to make sure the write block size meets pax specs. If the user
84 * does not specify a blocksize, we use the format default blocksize.
85 * We must be picky on writes, so we do not allow the user to create an
86 * archive that might be hard to read elsewhere. If all ok, we then
87 * open the first archive volume
91 if (wrblksz > MAXBLK) {
92 paxwarn(1, "Write block size of %d too large, maximum is: %d",
96 if (wrblksz % BLKMULT) {
97 paxwarn(1, "Write block size of %d is not a %d byte multiple",
101 if (wrblksz > MAXBLK_POSIX) {
102 paxwarn(0, "Write block size of %d larger than POSIX max %d, archive may not be portable",
103 wrblksz, MAXBLK_POSIX);
108 * we only allow wrblksz to be used with all archive operations
110 blksz = rdblksz = wrblksz;
111 if ((ar_open(arcname) < 0) && (ar_next() < 0))
114 bufend = buf + wrblksz;
121 * set up buffering system to read an archive
123 * 0 if ok, -1 otherwise
130 * leave space for the header pushback (see get_arc()). If we are
131 * going to append and user specified a write block size, check it
134 buf = &(bufmem[BLKMULT]);
135 if ((act == APPND) && wrblksz) {
136 if (wrblksz > MAXBLK) {
137 paxwarn(1,"Write block size %d too large, maximum is: %d",
141 if (wrblksz % BLKMULT) {
142 paxwarn(1, "Write block size %d is not a %d byte multiple",
151 if ((ar_open(arcname) < 0) && (ar_next() < 0))
153 bufend = buf + rdblksz;
161 * set up buffer system for copying within the file system
167 buf = &(bufmem[BLKMULT]);
168 rdblksz = blksz = MAXBLK;
173 * Set up the buffering system to append new members to an archive that
174 * was just read. The last block(s) of an archive may contain a format
175 * specific trailer. To append a new member, this trailer has to be
176 * removed from the archive. The first byte of the trailer is replaced by
177 * the start of the header of the first file added to the archive. The
178 * format specific end read function tells us how many bytes to move
179 * backwards in the archive to be positioned BEFORE the trailer. Two
180 * different positions have to be adjusted, the O.S. file offset (e.g. the
181 * position of the tape head) and the write point within the data we have
182 * stored in the read (soon to become write) buffer. We may have to move
183 * back several records (the number depends on the size of the archive
184 * record and the size of the format trailer) to read up the record where
185 * the first byte of the trailer is recorded. Trailers may span (and
186 * overlap) record boundaries.
187 * We first calculate which record has the first byte of the trailer. We
188 * move the OS file offset back to the start of this record and read it
189 * up. We set the buffer write pointer to be at this byte (the byte where
190 * the trailer starts). We then move the OS file pointer back to the
191 * start of this record so a flush of this buffer will replace the record
193 * A major problem is rewriting this last record. For archives stored
194 * on disk files, this is trivial. However, many devices are really picky
195 * about the conditions under which they will allow a write to occur.
196 * Often devices restrict the conditions where writes can be made,
197 * so it may not be feasible to append archives stored on all types of
200 * 0 for success, -1 for failure
204 appnd_start(off_t skcnt)
210 paxwarn(0, "Cannot append to an archive that may have flaws.");
214 * if the user did not specify a write blocksize, inherit the size used
215 * in the last archive volume read. (If a is set we still use rdblksz
216 * until next volume, cannot shift sizes within a single volume).
219 wrblksz = blksz = rdblksz;
224 * make sure that this volume allows appends
230 * Calculate bytes to move back and move in front of record where we
231 * need to start writing from. Remember we have to add in any padding
232 * that might be in the buffer after the trailer in the last block. We
233 * travel skcnt + padding ROUNDED UP to blksize.
235 skcnt += bufend - bufpt;
236 if ((cnt = rounddown(skcnt, blksz)) < skcnt)
238 if (ar_rev((off_t)cnt) < 0)
242 * We may have gone too far if there is valid data in the block we are
243 * now in front of, read up the block and position the pointer after
246 if ((cnt -= skcnt) > 0) {
248 * watch out for stupid tape drives. ar_rev() will set rdblksz
249 * to be real physical blocksize so we must loop until we get
250 * the old rdblksz (now in blksz). If ar_rev() fouls up the
251 * determination of the physical block size, we will fail.
254 bufend = buf + blksz;
255 while (bufpt < bufend) {
256 if ((res = ar_read(bufpt, rdblksz)) <= 0)
260 if (ar_rev((off_t)(bufpt - buf)) < 0)
263 bufend = buf + blksz;
268 bufend = buf + blksz;
276 * At this point we are ready to write. If the device requires special
277 * handling to write at a point were previously recorded data resides,
278 * that is handled in ar_set_wr(). From now on we operate under normal
279 * ARCHIVE mode (write) conditions
287 paxwarn(1, "Unable to rewrite archive trailer, cannot append.");
293 * A read error occurred on this archive volume. Resync the buffer and
294 * try to reset the device (if possible) so we can continue to read. Keep
295 * trying to do this until we get a valid read, or we reach the limit on
296 * consecutive read faults (at which point we give up). The user can
297 * adjust the read error limit through a command line option.
299 * 0 on success, and -1 on failure
309 * if the user says bail out on first fault, we are out of here...
314 paxwarn(1, "Unable to append when there are archive read errors.");
319 * poke at device and try to get past media error
321 if (ar_rdsync() < 0) {
329 if ((res = ar_read(buf, blksz)) > 0) {
331 * All right! got some data, fill that buffer
340 * Oh well, yet another failed read...
341 * if error limit reached, ditch. o.w. poke device to move past
342 * bad media and try again. if media is badly damaged, we ask
343 * the poor (and upset user at this point) for the next archive
344 * volume. remember the goal on reads is to get the most we
345 * can extract out of the archive.
347 if ((maxflt > 0) && (++errcnt > maxflt))
348 paxwarn(0,"Archive read error limit (%d) reached",maxflt);
349 else if (ar_rdsync() == 0)
361 * push the data used during the archive id phase back into the I/O
362 * buffer. This is required as we cannot be sure that the header does NOT
363 * overlap a block boundary (as in the case we are trying to recover a
364 * flawed archived). This was not designed to be used for any other
365 * purpose. (What software engineering, HA!)
366 * WARNING: do not even THINK of pback greater than BLKMULT, unless the
367 * pback space is increased.
371 pback(char *pt, int cnt)
374 memcpy(bufpt, pt, cnt);
380 * skip forward in the archive during a archive read. Used to get quickly
381 * past file data and padding for files the user did NOT select.
383 * 0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
394 * consume what data we have in the buffer. If we have to move forward
395 * whole records, we call the low level skip function to see if we can
396 * move within the archive without doing the expensive reads on data we
401 res = MIN((bufend - bufpt), skcnt);
406 * if skcnt is now 0, then no additional i/o is needed
412 * We have to read more, calculate complete and partial record reads
413 * based on rdblksz. we skip over "cnt" complete records
416 cnt = rounddown(skcnt, rdblksz);
419 * if the skip fails, we will have to resync. ar_fow will tell us
420 * how much it can skip over. We will have to read the rest.
422 if (ar_fow(cnt, &skipped) < 0)
424 res += cnt - skipped;
428 * what is left we have to read (which may be the whole thing if
429 * ar_fow() told us the device can only read to skip records);
432 cnt = bufend - bufpt;
434 * if the read fails, we will have to resync
436 if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
449 * flush out any data (and pad if required) the last block. We always pad
450 * with zero (even though we do not have to). Padding with 0 makes it a
451 * lot easier to recover if the archive is damaged. zero padding SHOULD
452 * BE a requirement....
459 memset(bufpt, 0, bufend - bufpt);
467 * fill the write buffer from data passed to it in a buffer (usually used
468 * by format specific write routines to pass a file header). On failure we
469 * punt. We do not allow the user to continue to write flawed archives.
470 * We assume these headers are not very large (the memory copy we use is
473 * 0 if buffer was filled ok, -1 o.w. (buffer flush failure)
477 wr_rdbuf(char *out, int outcnt)
482 * while there is data to copy copy into the write buffer. when the
483 * write buffer fills, flush it to the archive and continue
486 cnt = bufend - bufpt;
487 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
490 * only move what we have space for
492 cnt = MIN(cnt, outcnt);
493 memcpy(bufpt, out, cnt);
503 * copy from the read buffer into a supplied buffer a specified number of
504 * bytes. If the read buffer is empty fill it and continue to copy.
505 * usually used to obtain a file header for processing by a format
506 * specific read routine.
508 * number of bytes copied to the buffer, 0 indicates EOF on archive volume,
513 rd_wrbuf(char *in, int cpcnt)
520 * loop until we fill the buffer with the requested number of bytes
523 cnt = bufend - bufpt;
524 if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
526 * read error, return what we got (or the error if
527 * no data was copied). The caller must know that an
528 * error occurred and has the best knowledge what to
531 if ((res = cpcnt - incnt) > 0)
537 * calculate how much data to copy based on whats left and
540 cnt = MIN(cnt, incnt);
541 memcpy(in, bufpt, cnt);
551 * skip forward during a write. In other words add padding to the file.
552 * we add zero filled padding as it makes flawed archives much easier to
553 * recover from. the caller tells us how many bytes of padding to add
554 * This routine was not designed to add HUGE amount of padding, just small
555 * amounts (a few 512 byte blocks at most)
557 * 0 if ok, -1 if there was a buf_flush failure
566 * loop while there is more padding to add
569 cnt = bufend - bufpt;
570 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
572 cnt = MIN(cnt, skcnt);
573 memset(bufpt, 0, cnt);
582 * fill write buffer with the contents of a file. We are passed an open
583 * file descriptor to the file an the archive structure that describes the
584 * file we are storing. The variable "left" is modified to contain the
585 * number of bytes of the file we were NOT able to write to the archive.
586 * it is important that we always write EXACTLY the number of bytes that
587 * the format specific write routine told us to. The file can also get
588 * bigger, so reading to the end of file would create an improper archive,
589 * we just detect this case and warn the user. We never create a bad
590 * archive if we can avoid it. Of course trying to archive files that are
591 * active is asking for trouble. It we fail, we pass back how much we
592 * could NOT copy and let the caller deal with it.
594 * 0 ok, -1 if archive write failure. a short read of the file returns a
595 * 0, but "left" is set to be greater than zero.
599 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
603 off_t size = arcn->sb.st_size;
607 * while there are more bytes to write
610 cnt = bufend - bufpt;
611 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
615 cnt = MIN(cnt, size);
616 if ((res = read(ifd, bufpt, cnt)) <= 0)
623 * better check the file did not change during this operation
624 * or the file read failed.
627 syswarn(1, errno, "Read fault on %s", arcn->org_name);
629 paxwarn(1, "File changed size during read %s", arcn->org_name);
630 else if (fstat(ifd, &sb) < 0)
631 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
632 else if (arcn->sb.st_mtime != sb.st_mtime)
633 paxwarn(1, "File %s was modified during copy to archive",
641 * extract the contents of a file from the archive. If we are unable to
642 * extract the entire file (due to failure to write the file) we return
643 * the numbers of bytes we did NOT process. This way the caller knows how
644 * many bytes to skip past to find the next archive header. If the failure
645 * was due to an archive read, we will catch that when we try to skip. If
646 * the format supplies a file data crc value, we calculate the actual crc
647 * so that it can be compared to the value stored in the header
649 * We call a special function to write the file. This function attempts to
650 * restore file holes (blocks of zeros) into the file. When files are
651 * sparse this saves space, and is a LOT faster. For non sparse files
652 * the performance hit is small. As of this writing, no archive supports
653 * information on where the file holes are.
655 * 0 ok, -1 if archive read failure. if we cannot write the entire file,
656 * we return a 0 but "left" is set to be the amount unwritten
660 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
663 off_t size = arcn->sb.st_size;
665 char *fnm = arcn->name;
673 * pass the blocksize of the file being written to the write routine,
674 * if the size is zero, use the default MINFBSZ
676 if (fstat(ofd, &sb) == 0) {
677 if (sb.st_blksize > 0)
678 sz = (int)sb.st_blksize;
680 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
685 * Copy the archive to the file the number of bytes specified. We have
686 * to assume that we want to recover file holes as none of the archive
687 * formats can record the location of file holes.
690 cnt = bufend - bufpt;
692 * if we get a read error, we do not want to skip, as we may
693 * miss a header, so we do not set left, but if we get a write
694 * error, we do want to skip over the unprocessed data.
696 if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
698 cnt = MIN(cnt, size);
699 if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
706 * update the actual crc value
710 crc += *bufpt++ & 0xff;
717 * if the last block has a file hole (all zero), we must make sure this
718 * gets updated in the file. We force the last block of zeros to be
719 * written. just closing with the file offset moved forward may not put
720 * a hole at the end of the file.
722 if (isem && (arcn->sb.st_size > 0L))
723 file_flush(ofd, fnm, isem);
726 * if we failed from archive read, we do not want to skip
728 if ((size > 0L) && (*left == 0L))
732 * some formats record a crc on file data. If so, then we compare the
733 * calculated crc to the crc stored in the archive
735 if (docrc && (size == 0L) && (arcn->crc != crc))
736 paxwarn(1,"Actual crc does not match expected crc %s",arcn->name);
742 * copy the contents of one file to another. used during -rw phase of pax
743 * just as in rd_wrfile() we use a special write function to write the
744 * destination file so we can properly copy files with holes.
748 cp_file(ARCHD *arcn, int fd1, int fd2)
753 char *fnm = arcn->name;
761 * check for holes in the source file. If none, we will use regular
762 * write instead of file write.
764 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
768 * pass the blocksize of the file being written to the write routine,
769 * if the size is zero, use the default MINFBSZ
771 if (fstat(fd2, &sb) == 0) {
772 if (sb.st_blksize > 0)
775 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
779 * read the source file and copy to destination file until EOF
782 if ((cnt = read(fd1, buf, blksz)) <= 0)
785 res = write(fd2, buf, cnt);
787 res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
794 * check to make sure the copy is valid.
797 syswarn(1, errno, "Failed write during copy of %s to %s",
798 arcn->org_name, arcn->name);
799 else if (cpcnt != arcn->sb.st_size)
800 paxwarn(1, "File %s changed size during copy to %s",
801 arcn->org_name, arcn->name);
802 else if (fstat(fd1, &sb) < 0)
803 syswarn(1, errno, "Failed stat of %s", arcn->org_name);
804 else if (arcn->sb.st_mtime != sb.st_mtime)
805 paxwarn(1, "File %s was modified during copy to %s",
806 arcn->org_name, arcn->name);
809 * if the last block has a file hole (all zero), we must make sure this
810 * gets updated in the file. We force the last block of zeros to be
811 * written. just closing with the file offset moved forward may not put
812 * a hole at the end of the file.
814 if (!no_hole && isem && (arcn->sb.st_size > 0L))
815 file_flush(fd2, fnm, isem);
821 * fill the read buffer with the next record (or what we can get) from
822 * the archive volume.
824 * Number of bytes of data in the read buffer, -1 for read error, and
825 * 0 when finished (user specified termination in ar_next()).
839 * try to fill the buffer. on error the next archive volume is
840 * opened and we try again.
842 if ((cnt = ar_read(buf, blksz)) > 0) {
850 * errors require resync, EOF goes to next archive
866 * force the write buffer to the archive. We are passed the number of
867 * bytes in the buffer at the point of the flush. When we change archives
868 * the record size might change. (either larger or smaller).
870 * 0 if all is ok, -1 when a write error occurs.
874 buf_flush(int bufcnt)
881 * if we have reached the user specified byte count for each archive
882 * volume, prompt for the next volume. (The non-standard -R flag).
883 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
884 * at least one record. We always round limit UP to next blocksize.
886 if ((wrlimit > 0) && (wrcnt > wrlimit)) {
887 paxwarn(0, "User specified archive volume byte limit reached.");
896 * The new archive volume might have changed the size of the
897 * write blocksize. if so we figure out if we need to write
898 * (one or more times), or if there is now free space left in
899 * the buffer (it is no longer full). bufcnt has the number of
900 * bytes in the buffer, (the blocksize, at the point we were
901 * CALLED). Push has the amount of "extra" data in the buffer
902 * if the block size has shrunk from a volume change.
904 bufend = buf + blksz;
908 push = bufcnt - blksz;
912 * We have enough data to write at least one archive block
916 * write a block and check if it all went out ok
918 cnt = ar_write(buf, blksz);
926 /* we have extra data to push to the front.
927 * check for more than 1 block of push, and if
928 * so we loop back to write again
930 memcpy(buf, bufend, push);
939 } else if (cnt > 0) {
941 * Oh drat we got a partial write!
942 * if format doesnt care about alignment let it go,
943 * we warned the user in ar_write().... but this means
944 * the last record on this volume violates pax spec....
950 memcpy(buf, bufpt, cnt);
952 if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
958 * All done, go to next archive
965 * The new archive volume might also have changed the block
966 * size. if so, figure out if we have too much or too little
967 * data for using the new block size
969 bufend = buf + blksz;
973 push = bufcnt - blksz;
977 * write failed, stop pax. we must not create a bad archive!