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. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * @(#)buf_subs.c 8.2 (Berkeley) 4/18/94
38 * $FreeBSD: src/bin/pax/buf_subs.c,v 1.12.2.1 2001/08/01 05:03:11 obrien Exp $
39 * $DragonFly: src/bin/pax/buf_subs.c,v 1.5 2004/11/07 20:54:51 eirikn Exp $
42 #include <sys/types.h>
53 * routines which implement archive and file buffering
56 #define MINFBSZ 512 /* default block size for hole detect */
57 #define MAXFLT 10 /* default media read error limit */
60 * Need to change bufmem to dynamic allocation when the upper
61 * limit on blocking size is removed (though that will violate pax spec)
62 * MAXBLK define and tests will also need to be updated.
64 static char bufmem[MAXBLK+BLKMULT]; /* i/o buffer + pushback id space */
65 static char *buf; /* normal start of i/o buffer */
66 static char *bufend; /* end or last char in i/o buffer */
67 static char *bufpt; /* read/write point in i/o buffer */
68 int blksz = MAXBLK; /* block input/output size in bytes */
69 int wrblksz; /* user spec output size in bytes */
70 int maxflt = MAXFLT; /* MAX consecutive media errors */
71 int rdblksz; /* first read blksize (tapes only) */
72 off_t wrlimit; /* # of bytes written per archive vol */
73 off_t wrcnt; /* # of bytes written on current vol */
74 off_t rdcnt; /* # of bytes read on current vol */
78 * set up the buffering system to operate in a write mode
80 * 0 if ok, -1 if the user specified write block size violates pax spec
86 buf = &(bufmem[BLKMULT]);
88 * Check to make sure the write block size meets pax specs. If the user
89 * does not specify a blocksize, we use the format default blocksize.
90 * We must be picky on writes, so we do not allow the user to create an
91 * archive that might be hard to read elsewhere. If all ok, we then
92 * open the first archive volume
96 if (wrblksz > MAXBLK) {
97 paxwarn(1, "Write block size of %d too large, maximum is: %d",
101 if (wrblksz % BLKMULT) {
102 paxwarn(1, "Write block size of %d is not a %d byte multiple",
106 if (wrblksz > MAXBLK_POSIX) {
107 paxwarn(0, "Write block size of %d larger than POSIX max %d, archive may not be portable",
108 wrblksz, MAXBLK_POSIX);
113 * we only allow wrblksz to be used with all archive operations
115 blksz = rdblksz = wrblksz;
116 if ((ar_open(arcname) < 0) && (ar_next() < 0))
119 bufend = buf + wrblksz;
126 * set up buffering system to read an archive
128 * 0 if ok, -1 otherwise
135 * leave space for the header pushback (see get_arc()). If we are
136 * going to append and user specified a write block size, check it
139 buf = &(bufmem[BLKMULT]);
140 if ((act == APPND) && wrblksz) {
141 if (wrblksz > MAXBLK) {
142 paxwarn(1,"Write block size %d too large, maximum is: %d",
146 if (wrblksz % BLKMULT) {
147 paxwarn(1, "Write block size %d is not a %d byte multiple",
156 if ((ar_open(arcname) < 0) && (ar_next() < 0))
158 bufend = buf + rdblksz;
166 * set up buffer system for copying within the file system
172 buf = &(bufmem[BLKMULT]);
173 rdblksz = blksz = MAXBLK;
178 * Set up the buffering system to append new members to an archive that
179 * was just read. The last block(s) of an archive may contain a format
180 * specific trailer. To append a new member, this trailer has to be
181 * removed from the archive. The first byte of the trailer is replaced by
182 * the start of the header of the first file added to the archive. The
183 * format specific end read function tells us how many bytes to move
184 * backwards in the archive to be positioned BEFORE the trailer. Two
185 * different postions have to be adjusted, the O.S. file offset (e.g. the
186 * position of the tape head) and the write point within the data we have
187 * stored in the read (soon to become write) buffer. We may have to move
188 * back several records (the number depends on the size of the archive
189 * record and the size of the format trailer) to read up the record where
190 * the first byte of the trailer is recorded. Trailers may span (and
191 * overlap) record boundries.
192 * We first calculate which record has the first byte of the trailer. We
193 * move the OS file offset back to the start of this record and read it
194 * up. We set the buffer write pointer to be at this byte (the byte where
195 * the trailer starts). We then move the OS file pointer back to the
196 * start of this record so a flush of this buffer will replace the record
198 * A major problem is rewriting this last record. For archives stored
199 * on disk files, this is trival. However, many devices are really picky
200 * about the conditions under which they will allow a write to occur.
201 * Often devices restrict the conditions where writes can be made writes,
202 * so it may not be feasable to append archives stored on all types of
205 * 0 for success, -1 for failure
209 appnd_start(off_t skcnt)
215 paxwarn(0, "Cannot append to an archive that may have flaws.");
219 * if the user did not specify a write blocksize, inherit the size used
220 * in the last archive volume read. (If a is set we still use rdblksz
221 * until next volume, cannot shift sizes within a single volume).
224 wrblksz = blksz = rdblksz;
229 * make sure that this volume allows appends
235 * Calculate bytes to move back and move in front of record where we
236 * need to start writing from. Remember we have to add in any padding
237 * that might be in the buffer after the trailer in the last block. We
238 * travel skcnt + padding ROUNDED UP to blksize.
240 skcnt += bufend - bufpt;
241 if ((cnt = (skcnt/blksz) * blksz) < skcnt)
243 if (ar_rev((off_t)cnt) < 0)
247 * We may have gone too far if there is valid data in the block we are
248 * now in front of, read up the block and position the pointer after
251 if ((cnt -= skcnt) > 0) {
253 * watch out for stupid tape drives. ar_rev() will set rdblksz
254 * to be real physical blocksize so we must loop until we get
255 * the old rdblksz (now in blksz). If ar_rev() fouls up the
256 * determination of the physical block size, we will fail.
259 bufend = buf + blksz;
260 while (bufpt < bufend) {
261 if ((res = ar_read(bufpt, rdblksz)) <= 0)
265 if (ar_rev((off_t)(bufpt - buf)) < 0)
268 bufend = buf + blksz;
273 bufend = buf + blksz;
281 * At this point we are ready to write. If the device requires special
282 * handling to write at a point were previously recorded data resides,
283 * that is handled in ar_set_wr(). From now on we operate under normal
284 * ARCHIVE mode (write) conditions
292 paxwarn(1, "Unable to rewrite archive trailer, cannot append.");
298 * A read error occurred on this archive volume. Resync the buffer and
299 * try to reset the device (if possible) so we can continue to read. Keep
300 * trying to do this until we get a valid read, or we reach the limit on
301 * consecutive read faults (at which point we give up). The user can
302 * adjust the read error limit through a command line option.
304 * 0 on success, and -1 on failure
314 * if the user says bail out on first fault, we are out of here...
319 paxwarn(1, "Unable to append when there are archive read errors.");
324 * poke at device and try to get past media error
326 if (ar_rdsync() < 0) {
334 if ((res = ar_read(buf, blksz)) > 0) {
336 * All right! got some data, fill that buffer
345 * Oh well, yet another failed read...
346 * if error limit reached, ditch. o.w. poke device to move past
347 * bad media and try again. if media is badly damaged, we ask
348 * the poor (and upset user at this point) for the next archive
349 * volume. remember the goal on reads is to get the most we
350 * can extract out of the archive.
352 if ((maxflt > 0) && (++errcnt > maxflt))
353 paxwarn(0,"Archive read error limit (%d) reached",maxflt);
354 else if (ar_rdsync() == 0)
366 * push the data used during the archive id phase back into the I/O
367 * buffer. This is required as we cannot be sure that the header does NOT
368 * overlap a block boundry (as in the case we are trying to recover a
369 * flawed archived). This was not designed to be used for any other
370 * purpose. (What software engineering, HA!)
371 * WARNING: do not even THINK of pback greater than BLKMULT, unless the
372 * pback space is increased.
376 pback(char *pt, int cnt)
379 memcpy(bufpt, pt, cnt);
385 * skip foward in the archive during a archive read. Used to get quickly
386 * past file data and padding for files the user did NOT select.
388 * 0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
399 * consume what data we have in the buffer. If we have to move foward
400 * whole records, we call the low level skip function to see if we can
401 * move within the archive without doing the expensive reads on data we
406 res = MIN((bufend - bufpt), skcnt);
411 * if skcnt is now 0, then no additional i/o is needed
417 * We have to read more, calculate complete and partial record reads
418 * based on rdblksz. we skip over "cnt" complete records
421 cnt = (skcnt/rdblksz) * rdblksz;
424 * if the skip fails, we will have to resync. ar_fow will tell us
425 * how much it can skip over. We will have to read the rest.
427 if (ar_fow(cnt, &skipped) < 0)
429 res += cnt - skipped;
433 * what is left we have to read (which may be the whole thing if
434 * ar_fow() told us the device can only read to skip records);
437 cnt = bufend - bufpt;
439 * if the read fails, we will have to resync
441 if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
454 * flush out any data (and pad if required) the last block. We always pad
455 * with zero (even though we do not have to). Padding with 0 makes it a
456 * lot easier to recover if the archive is damaged. zero paddding SHOULD
457 * BE a requirement....
464 memset(bufpt, 0, bufend - bufpt);
472 * fill the write buffer from data passed to it in a buffer (usually used
473 * by format specific write routines to pass a file header). On failure we
474 * punt. We do not allow the user to continue to write flawed archives.
475 * We assume these headers are not very large (the memory copy we use is
478 * 0 if buffer was filled ok, -1 o.w. (buffer flush failure)
482 wr_rdbuf(char *out, int outcnt)
487 * while there is data to copy copy into the write buffer. when the
488 * write buffer fills, flush it to the archive and continue
491 cnt = bufend - bufpt;
492 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
495 * only move what we have space for
497 cnt = MIN(cnt, outcnt);
498 memcpy(bufpt, out, cnt);
508 * copy from the read buffer into a supplied buffer a specified number of
509 * bytes. If the read buffer is empty fill it and continue to copy.
510 * usually used to obtain a file header for processing by a format
511 * specific read routine.
513 * number of bytes copied to the buffer, 0 indicates EOF on archive volume,
518 rd_wrbuf(char *in, int cpcnt)
525 * loop until we fill the buffer with the requested number of bytes
528 cnt = bufend - bufpt;
529 if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
531 * read error, return what we got (or the error if
532 * no data was copied). The caller must know that an
533 * error occured and has the best knowledge what to
536 if ((res = cpcnt - incnt) > 0)
542 * calculate how much data to copy based on whats left and
545 cnt = MIN(cnt, incnt);
546 memcpy(in, bufpt, cnt);
556 * skip forward during a write. In other words add padding to the file.
557 * we add zero filled padding as it makes flawed archives much easier to
558 * recover from. the caller tells us how many bytes of padding to add
559 * This routine was not designed to add HUGE amount of padding, just small
560 * amounts (a few 512 byte blocks at most)
562 * 0 if ok, -1 if there was a buf_flush failure
571 * loop while there is more padding to add
574 cnt = bufend - bufpt;
575 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
577 cnt = MIN(cnt, skcnt);
578 memset(bufpt, 0, cnt);
587 * fill write buffer with the contents of a file. We are passed an open
588 * file descriptor to the file an the archive structure that describes the
589 * file we are storing. The variable "left" is modified to contain the
590 * number of bytes of the file we were NOT able to write to the archive.
591 * it is important that we always write EXACTLY the number of bytes that
592 * the format specific write routine told us to. The file can also get
593 * bigger, so reading to the end of file would create an improper archive,
594 * we just detect this case and warn the user. We never create a bad
595 * archive if we can avoid it. Of course trying to archive files that are
596 * active is asking for trouble. It we fail, we pass back how much we
597 * could NOT copy and let the caller deal with it.
599 * 0 ok, -1 if archive write failure. a short read of the file returns a
600 * 0, but "left" is set to be greater than zero.
604 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
608 off_t size = arcn->sb.st_size;
612 * while there are more bytes to write
615 cnt = bufend - bufpt;
616 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
620 cnt = MIN(cnt, size);
621 if ((res = read(ifd, bufpt, cnt)) <= 0)
628 * better check the file did not change during this operation
629 * or the file read failed.
632 syswarn(1, errno, "Read fault on %s", arcn->org_name);
634 paxwarn(1, "File changed size during read %s", arcn->org_name);
635 else if (fstat(ifd, &sb) < 0)
636 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
637 else if (arcn->sb.st_mtime != sb.st_mtime)
638 paxwarn(1, "File %s was modified during copy to archive",
646 * extract the contents of a file from the archive. If we are unable to
647 * extract the entire file (due to failure to write the file) we return
648 * the numbers of bytes we did NOT process. This way the caller knows how
649 * many bytes to skip past to find the next archive header. If the failure
650 * was due to an archive read, we will catch that when we try to skip. If
651 * the format supplies a file data crc value, we calculate the actual crc
652 * so that it can be compared to the value stored in the header
654 * We call a special function to write the file. This function attempts to
655 * restore file holes (blocks of zeros) into the file. When files are
656 * sparse this saves space, and is a LOT faster. For non sparse files
657 * the performance hit is small. As of this writing, no archive supports
658 * information on where the file holes are.
660 * 0 ok, -1 if archive read failure. if we cannot write the entire file,
661 * we return a 0 but "left" is set to be the amount unwritten
665 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
668 off_t size = arcn->sb.st_size;
670 char *fnm = arcn->name;
678 * pass the blocksize of the file being written to the write routine,
679 * if the size is zero, use the default MINFBSZ
681 if (fstat(ofd, &sb) == 0) {
682 if (sb.st_blksize > 0)
683 sz = (int)sb.st_blksize;
685 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
690 * Copy the archive to the file the number of bytes specified. We have
691 * to assume that we want to recover file holes as none of the archive
692 * formats can record the location of file holes.
695 cnt = bufend - bufpt;
697 * if we get a read error, we do not want to skip, as we may
698 * miss a header, so we do not set left, but if we get a write
699 * error, we do want to skip over the unprocessed data.
701 if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
703 cnt = MIN(cnt, size);
704 if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
711 * update the actual crc value
715 crc += *bufpt++ & 0xff;
722 * if the last block has a file hole (all zero), we must make sure this
723 * gets updated in the file. We force the last block of zeros to be
724 * written. just closing with the file offset moved forward may not put
725 * a hole at the end of the file.
727 if (isem && (arcn->sb.st_size > 0L))
728 file_flush(ofd, fnm, isem);
731 * if we failed from archive read, we do not want to skip
733 if ((size > 0L) && (*left == 0L))
737 * some formats record a crc on file data. If so, then we compare the
738 * calculated crc to the crc stored in the archive
740 if (docrc && (size == 0L) && (arcn->crc != crc))
741 paxwarn(1,"Actual crc does not match expected crc %s",arcn->name);
747 * copy the contents of one file to another. used during -rw phase of pax
748 * just as in rd_wrfile() we use a special write function to write the
749 * destination file so we can properly copy files with holes.
753 cp_file(ARCHD *arcn, int fd1, int fd2)
758 char *fnm = arcn->name;
766 * check for holes in the source file. If none, we will use regular
767 * write instead of file write.
769 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
773 * pass the blocksize of the file being written to the write routine,
774 * if the size is zero, use the default MINFBSZ
776 if (fstat(fd2, &sb) == 0) {
777 if (sb.st_blksize > 0)
780 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
784 * read the source file and copy to destination file until EOF
787 if ((cnt = read(fd1, buf, blksz)) <= 0)
790 res = write(fd2, buf, cnt);
792 res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
799 * check to make sure the copy is valid.
802 syswarn(1, errno, "Failed write during copy of %s to %s",
803 arcn->org_name, arcn->name);
804 else if (cpcnt != arcn->sb.st_size)
805 paxwarn(1, "File %s changed size during copy to %s",
806 arcn->org_name, arcn->name);
807 else if (fstat(fd1, &sb) < 0)
808 syswarn(1, errno, "Failed stat of %s", arcn->org_name);
809 else if (arcn->sb.st_mtime != sb.st_mtime)
810 paxwarn(1, "File %s was modified during copy to %s",
811 arcn->org_name, arcn->name);
814 * if the last block has a file hole (all zero), we must make sure this
815 * gets updated in the file. We force the last block of zeros to be
816 * written. just closing with the file offset moved forward may not put
817 * a hole at the end of the file.
819 if (!no_hole && isem && (arcn->sb.st_size > 0L))
820 file_flush(fd2, fnm, isem);
826 * fill the read buffer with the next record (or what we can get) from
827 * the archive volume.
829 * Number of bytes of data in the read buffer, -1 for read error, and
830 * 0 when finished (user specified termination in ar_next()).
844 * try to fill the buffer. on error the next archive volume is
845 * opened and we try again.
847 if ((cnt = ar_read(buf, blksz)) > 0) {
855 * errors require resync, EOF goes to next archive
871 * force the write buffer to the archive. We are passed the number of
872 * bytes in the buffer at the point of the flush. When we change archives
873 * the record size might change. (either larger or smaller).
875 * 0 if all is ok, -1 when a write error occurs.
879 buf_flush(int bufcnt)
886 * if we have reached the user specified byte count for each archive
887 * volume, prompt for the next volume. (The non-standrad -R flag).
888 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
889 * at least one record. We always round limit UP to next blocksize.
891 if ((wrlimit > 0) && (wrcnt > wrlimit)) {
892 paxwarn(0, "User specified archive volume byte limit reached.");
901 * The new archive volume might have changed the size of the
902 * write blocksize. if so we figure out if we need to write
903 * (one or more times), or if there is now free space left in
904 * the buffer (it is no longer full). bufcnt has the number of
905 * bytes in the buffer, (the blocksize, at the point we were
906 * CALLED). Push has the amount of "extra" data in the buffer
907 * if the block size has shrunk from a volume change.
909 bufend = buf + blksz;
913 push = bufcnt - blksz;
917 * We have enough data to write at least one archive block
921 * write a block and check if it all went out ok
923 cnt = ar_write(buf, blksz);
931 /* we have extra data to push to the front.
932 * check for more than 1 block of push, and if
933 * so we loop back to write again
935 memcpy(buf, bufend, push);
944 } else if (cnt > 0) {
946 * Oh drat we got a partial write!
947 * if format doesnt care about alignment let it go,
948 * we warned the user in ar_write().... but this means
949 * the last record on this volume violates pax spec....
955 memcpy(buf, bufpt, cnt);
957 if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
963 * All done, go to next archive
970 * The new archive volume might also have changed the block
971 * size. if so, figure out if we have too much or too little
972 * data for using the new block size
974 bufend = buf + blksz;
978 push = bufcnt - blksz;
982 * write failed, stop pax. we must not create a bad archive!