1 /* $OpenBSD: buf_subs.c,v 1.21 2005/11/09 19:59:06 otto 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
39 static const char sccsid[] = "@(#)buf_subs.c 8.2 (Berkeley) 4/18/94";
41 static const char rcsid[] = "$OpenBSD: buf_subs.c,v 1.21 2005/11/09 19:59:06 otto Exp $";
45 #include <sys/types.h>
48 #include <sys/param.h>
58 * routines which implement archive and file buffering
61 #define MINFBSZ 512 /* default block size for hole detect */
62 #define MAXFLT 10 /* default media read error limit */
65 * Need to change bufmem to dynamic allocation when the upper
66 * limit on blocking size is removed (though that will violate pax spec)
67 * MAXBLK define and tests will also need to be updated.
69 static char bufmem[MAXBLK+BLKMULT]; /* i/o buffer + pushback id space */
70 static char *buf; /* normal start of i/o buffer */
71 static char *bufend; /* end or last char in i/o buffer */
72 static char *bufpt; /* read/write point in i/o buffer */
73 int blksz = MAXBLK; /* block input/output size in bytes */
74 int wrblksz; /* user spec output size in bytes */
75 int maxflt = MAXFLT; /* MAX consecutive media errors */
76 int rdblksz; /* first read blksize (tapes only) */
77 off_t wrlimit; /* # of bytes written per archive vol */
78 off_t wrcnt; /* # of bytes written on current vol */
79 off_t rdcnt; /* # of bytes read on current vol */
83 * set up the buffering system to operate in a write mode
85 * 0 if ok, -1 if the user specified write block size violates pax spec
91 buf = &(bufmem[BLKMULT]);
93 * Check to make sure the write block size meets pax specs. If the user
94 * does not specify a blocksize, we use the format default blocksize.
95 * We must be picky on writes, so we do not allow the user to create an
96 * archive that might be hard to read elsewhere. If all ok, we then
97 * open the first archive volume
101 if (wrblksz > MAXBLK) {
102 paxwarn(1, "Write block size of %d too large, maximium is: %d",
106 if (wrblksz % BLKMULT) {
107 paxwarn(1, "Write block size of %d is not a %d byte multiple",
111 if (wrblksz > MAXBLK_POSIX) {
112 paxwarn(0, "Write block size of %d larger than POSIX max %d, archive may not be portable",
113 wrblksz, MAXBLK_POSIX);
118 * we only allow wrblksz to be used with all archive operations
120 blksz = rdblksz = wrblksz;
121 if ((ar_open(arcname) < 0) && (ar_next() < 0))
124 bufend = buf + wrblksz;
131 * set up buffering system to read an archive
133 * 0 if ok, -1 otherwise
140 * leave space for the header pushback (see get_arc()). If we are
141 * going to append and user specified a write block size, check it
144 buf = &(bufmem[BLKMULT]);
145 if ((act == APPND) && wrblksz) {
146 if (wrblksz > MAXBLK) {
147 paxwarn(1,"Write block size %d too large, maximium is: %d",
151 if (wrblksz % BLKMULT) {
152 paxwarn(1, "Write block size %d is not a %d byte multiple",
161 if ((ar_open(arcname) < 0) && (ar_next() < 0))
163 bufend = buf + rdblksz;
171 * set up buffer system for copying within the file system
177 buf = &(bufmem[BLKMULT]);
178 rdblksz = blksz = MAXBLK;
183 * Set up the buffering system to append new members to an archive that
184 * was just read. The last block(s) of an archive may contain a format
185 * specific trailer. To append a new member, this trailer has to be
186 * removed from the archive. The first byte of the trailer is replaced by
187 * the start of the header of the first file added to the archive. The
188 * format specific end read function tells us how many bytes to move
189 * backwards in the archive to be positioned BEFORE the trailer. Two
190 * different position have to be adjusted, the O.S. file offset (e.g. the
191 * position of the tape head) and the write point within the data we have
192 * stored in the read (soon to become write) buffer. We may have to move
193 * back several records (the number depends on the size of the archive
194 * record and the size of the format trailer) to read up the record where
195 * the first byte of the trailer is recorded. Trailers may span (and
196 * overlap) record boundaries.
197 * We first calculate which record has the first byte of the trailer. We
198 * move the OS file offset back to the start of this record and read it
199 * up. We set the buffer write pointer to be at this byte (the byte where
200 * the trailer starts). We then move the OS file pointer back to the
201 * start of this record so a flush of this buffer will replace the record
203 * A major problem is rewriting this last record. For archives stored
204 * on disk files, this is trivial. However, many devices are really picky
205 * about the conditions under which they will allow a write to occur.
206 * Often devices restrict the conditions where writes can be made,
207 * so it may not be feasible to append archives stored on all types of
210 * 0 for success, -1 for failure
214 appnd_start(off_t skcnt)
220 paxwarn(0, "Cannot append to an archive that may have flaws.");
224 * if the user did not specify a write blocksize, inherit the size used
225 * in the last archive volume read. (If a is set we still use rdblksz
226 * until next volume, cannot shift sizes within a single volume).
229 wrblksz = blksz = rdblksz;
234 * make sure that this volume allows appends
240 * Calculate bytes to move back and move in front of record where we
241 * need to start writing from. Remember we have to add in any padding
242 * that might be in the buffer after the trailer in the last block. We
243 * travel skcnt + padding ROUNDED UP to blksize.
245 skcnt += bufend - bufpt;
246 if ((cnt = (skcnt/blksz) * blksz) < skcnt)
248 if (ar_rev((off_t)cnt) < 0)
252 * We may have gone too far if there is valid data in the block we are
253 * now in front of, read up the block and position the pointer after
256 if ((cnt -= skcnt) > 0) {
258 * watch out for stupid tape drives. ar_rev() will set rdblksz
259 * to be real physical blocksize so we must loop until we get
260 * the old rdblksz (now in blksz). If ar_rev() fouls up the
261 * determination of the physical block size, we will fail.
264 bufend = buf + blksz;
265 while (bufpt < bufend) {
266 if ((res = ar_read(bufpt, rdblksz)) <= 0)
270 if (ar_rev((off_t)(bufpt - buf)) < 0)
273 bufend = buf + blksz;
278 bufend = buf + blksz;
286 * At this point we are ready to write. If the device requires special
287 * handling to write at a point were previously recorded data resides,
288 * that is handled in ar_set_wr(). From now on we operate under normal
289 * ARCHIVE mode (write) conditions
297 paxwarn(1, "Unable to rewrite archive trailer, cannot append.");
303 * A read error occurred on this archive volume. Resync the buffer and
304 * try to reset the device (if possible) so we can continue to read. Keep
305 * trying to do this until we get a valid read, or we reach the limit on
306 * consecutive read faults (at which point we give up). The user can
307 * adjust the read error limit through a command line option.
309 * 0 on success, and -1 on failure
319 * if the user says bail out on first fault, we are out of here...
324 paxwarn(1, "Unable to append when there are archive read errors.");
329 * poke at device and try to get past media error
331 if (ar_rdsync() < 0) {
339 if ((res = ar_read(buf, blksz)) > 0) {
341 * All right! got some data, fill that buffer
350 * Oh well, yet another failed read...
351 * if error limit reached, ditch. o.w. poke device to move past
352 * bad media and try again. if media is badly damaged, we ask
353 * the poor (and upset user at this point) for the next archive
354 * volume. remember the goal on reads is to get the most we
355 * can extract out of the archive.
357 if ((maxflt > 0) && (++errcnt > maxflt))
358 paxwarn(0,"Archive read error limit (%d) reached",maxflt);
359 else if (ar_rdsync() == 0)
371 * push the data used during the archive id phase back into the I/O
372 * buffer. This is required as we cannot be sure that the header does NOT
373 * overlap a block boundary (as in the case we are trying to recover a
374 * flawed archived). This was not designed to be used for any other
375 * purpose. (What software engineering, HA!)
376 * WARNING: do not even THINK of pback greater than BLKMULT, unless the
377 * pback space is increased.
381 pback(char *pt, int cnt)
384 memcpy(bufpt, pt, cnt);
390 * skip forward in the archive during a archive read. Used to get quickly
391 * past file data and padding for files the user did NOT select.
393 * 0 if ok, -1 failure, and 1 when EOF on the archive volume was detected.
404 * consume what data we have in the buffer. If we have to move forward
405 * whole records, we call the low level skip function to see if we can
406 * move within the archive without doing the expensive reads on data we
411 res = MIN((bufend - bufpt), skcnt);
416 * if skcnt is now 0, then no additional i/o is needed
422 * We have to read more, calculate complete and partial record reads
423 * based on rdblksz. we skip over "cnt" complete records
426 cnt = (skcnt/rdblksz) * rdblksz;
429 * if the skip fails, we will have to resync. ar_fow will tell us
430 * how much it can skip over. We will have to read the rest.
432 if (ar_fow(cnt, &skipped) < 0)
434 res += cnt - skipped;
438 * what is left we have to read (which may be the whole thing if
439 * ar_fow() told us the device can only read to skip records);
442 cnt = bufend - bufpt;
444 * if the read fails, we will have to resync
446 if ((cnt <= 0) && ((cnt = buf_fill()) < 0))
459 * flush out any data (and pad if required) the last block. We always pad
460 * with zero (even though we do not have to). Padding with 0 makes it a
461 * lot easier to recover if the archive is damaged. zero padding SHOULD
462 * BE a requirement....
469 memset(bufpt, 0, bufend - bufpt);
471 (void)buf_flush(blksz);
477 * fill the write buffer from data passed to it in a buffer (usually used
478 * by format specific write routines to pass a file header). On failure we
479 * punt. We do not allow the user to continue to write flawed archives.
480 * We assume these headers are not very large (the memory copy we use is
483 * 0 if buffer was filled ok, -1 o.w. (buffer flush failure)
487 wr_rdbuf(char *out, int outcnt)
492 * while there is data to copy copy into the write buffer. when the
493 * write buffer fills, flush it to the archive and continue
496 cnt = bufend - bufpt;
497 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
500 * only move what we have space for
502 cnt = MIN(cnt, outcnt);
503 memcpy(bufpt, out, cnt);
513 * copy from the read buffer into a supplied buffer a specified number of
514 * bytes. If the read buffer is empty fill it and continue to copy.
515 * usually used to obtain a file header for processing by a format
516 * specific read routine.
518 * number of bytes copied to the buffer, 0 indicates EOF on archive volume,
523 rd_wrbuf(char *in, int cpcnt)
530 * loop until we fill the buffer with the requested number of bytes
533 cnt = bufend - bufpt;
534 if ((cnt <= 0) && ((cnt = buf_fill()) <= 0)) {
536 * read error, return what we got (or the error if
537 * no data was copied). The caller must know that an
538 * error occurred and has the best knowledge what to
541 if ((res = cpcnt - incnt) > 0)
547 * calculate how much data to copy based on whats left and
550 cnt = MIN(cnt, incnt);
551 memcpy(in, bufpt, cnt);
561 * skip forward during a write. In other words add padding to the file.
562 * we add zero filled padding as it makes flawed archives much easier to
563 * recover from. the caller tells us how many bytes of padding to add
564 * This routine was not designed to add HUGE amount of padding, just small
565 * amounts (a few 512 byte blocks at most)
567 * 0 if ok, -1 if there was a buf_flush failure
576 * loop while there is more padding to add
579 cnt = bufend - bufpt;
580 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0))
582 cnt = MIN(cnt, skcnt);
583 memset(bufpt, 0, cnt);
592 * fill write buffer with the contents of a file. We are passed an open
593 * file descriptor to the file an the archive structure that describes the
594 * file we are storing. The variable "left" is modified to contain the
595 * number of bytes of the file we were NOT able to write to the archive.
596 * it is important that we always write EXACTLY the number of bytes that
597 * the format specific write routine told us to. The file can also get
598 * bigger, so reading to the end of file would create an improper archive,
599 * we just detect this case and warn the user. We never create a bad
600 * archive if we can avoid it. Of course trying to archive files that are
601 * active is asking for trouble. It we fail, we pass back how much we
602 * could NOT copy and let the caller deal with it.
604 * 0 ok, -1 if archive write failure. a short read of the file returns a
605 * 0, but "left" is set to be greater than zero.
609 wr_rdfile(ARCHD *arcn, int ifd, off_t *left)
613 off_t size = arcn->sb.st_size;
617 * while there are more bytes to write
620 cnt = bufend - bufpt;
621 if ((cnt <= 0) && ((cnt = buf_flush(blksz)) < 0)) {
625 cnt = MIN(cnt, size);
626 if ((res = read(ifd, bufpt, cnt)) <= 0)
633 * better check the file did not change during this operation
634 * or the file read failed.
637 syswarn(1, errno, "Read fault on %s", arcn->org_name);
639 paxwarn(1, "File changed size during read %s", arcn->org_name);
640 else if (fstat(ifd, &sb) < 0)
641 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
642 else if (arcn->sb.st_mtime != sb.st_mtime)
643 paxwarn(1, "File %s was modified during copy to archive",
651 * extract the contents of a file from the archive. If we are unable to
652 * extract the entire file (due to failure to write the file) we return
653 * the numbers of bytes we did NOT process. This way the caller knows how
654 * many bytes to skip past to find the next archive header. If the failure
655 * was due to an archive read, we will catch that when we try to skip. If
656 * the format supplies a file data crc value, we calculate the actual crc
657 * so that it can be compared to the value stored in the header
659 * We call a special function to write the file. This function attempts to
660 * restore file holes (blocks of zeros) into the file. When files are
661 * sparse this saves space, and is a LOT faster. For non sparse files
662 * the performance hit is small. As of this writing, no archive supports
663 * information on where the file holes are.
665 * 0 ok, -1 if archive read failure. if we cannot write the entire file,
666 * we return a 0 but "left" is set to be the amount unwritten
670 rd_wrfile(ARCHD *arcn, int ofd, off_t *left)
673 off_t size = arcn->sb.st_size;
675 char *fnm = arcn->name;
683 * pass the blocksize of the file being written to the write routine,
684 * if the size is zero, use the default MINFBSZ
687 sz = PAXPATHLEN + 1; /* GNU tar long link/file */
688 else if (fstat(ofd, &sb) == 0) {
689 if (sb.st_blksize > 0)
690 sz = (int)sb.st_blksize;
692 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
697 * Copy the archive to the file the number of bytes specified. We have
698 * to assume that we want to recover file holes as none of the archive
699 * formats can record the location of file holes.
702 cnt = bufend - bufpt;
704 * if we get a read error, we do not want to skip, as we may
705 * miss a header, so we do not set left, but if we get a write
706 * error, we do want to skip over the unprocessed data.
708 if ((cnt <= 0) && ((cnt = buf_fill()) <= 0))
710 cnt = MIN(cnt, size);
711 if ((res = file_write(ofd,bufpt,cnt,&rem,&isem,sz,fnm)) <= 0) {
718 * update the actual crc value
722 crc += *bufpt++ & 0xff;
729 * if the last block has a file hole (all zero), we must make sure this
730 * gets updated in the file. We force the last block of zeros to be
731 * written. just closing with the file offset moved forward may not put
732 * a hole at the end of the file.
734 if (isem && (arcn->sb.st_size > 0L))
735 file_flush(ofd, fnm, isem);
738 * if we failed from archive read, we do not want to skip
740 if ((size > 0L) && (*left == 0L))
744 * some formats record a crc on file data. If so, then we compare the
745 * calculated crc to the crc stored in the archive
747 if (docrc && (size == 0L) && (arcn->crc != crc))
748 paxwarn(1,"Actual crc does not match expected crc %s",arcn->name);
754 * copy the contents of one file to another. used during -rw phase of pax
755 * just as in rd_wrfile() we use a special write function to write the
756 * destination file so we can properly copy files with holes.
760 cp_file(ARCHD *arcn, int fd1, int fd2)
765 char *fnm = arcn->name;
773 * check for holes in the source file. If none, we will use regular
774 * write instead of file write.
776 if (((off_t)(arcn->sb.st_blocks * BLKMULT)) >= arcn->sb.st_size)
780 * pass the blocksize of the file being written to the write routine,
781 * if the size is zero, use the default MINFBSZ
783 if (fstat(fd2, &sb) == 0) {
784 if (sb.st_blksize > 0)
787 syswarn(0,errno,"Unable to obtain block size for file %s",fnm);
791 * read the source file and copy to destination file until EOF
794 if ((cnt = read(fd1, buf, blksz)) <= 0)
797 res = write(fd2, buf, cnt);
799 res = file_write(fd2, buf, cnt, &rem, &isem, sz, fnm);
806 * check to make sure the copy is valid.
809 syswarn(1, errno, "Failed write during copy of %s to %s",
810 arcn->org_name, arcn->name);
811 else if (cpcnt != arcn->sb.st_size)
812 paxwarn(1, "File %s changed size during copy to %s",
813 arcn->org_name, arcn->name);
814 else if (fstat(fd1, &sb) < 0)
815 syswarn(1, errno, "Failed stat of %s", arcn->org_name);
816 else if (arcn->sb.st_mtime != sb.st_mtime)
817 paxwarn(1, "File %s was modified during copy to %s",
818 arcn->org_name, arcn->name);
821 * if the last block has a file hole (all zero), we must make sure this
822 * gets updated in the file. We force the last block of zeros to be
823 * written. just closing with the file offset moved forward may not put
824 * a hole at the end of the file.
826 if (!no_hole && isem && (arcn->sb.st_size > 0L))
827 file_flush(fd2, fnm, isem);
833 * fill the read buffer with the next record (or what we can get) from
834 * the archive volume.
836 * Number of bytes of data in the read buffer, -1 for read error, and
837 * 0 when finished (user specified termination in ar_next()).
851 * try to fill the buffer. on error the next archive volume is
852 * opened and we try again.
854 if ((cnt = ar_read(buf, blksz)) > 0) {
862 * errors require resync, EOF goes to next archive
878 * force the write buffer to the archive. We are passed the number of
879 * bytes in the buffer at the point of the flush. When we change archives
880 * the record size might change. (either larger or smaller).
882 * 0 if all is ok, -1 when a write error occurs.
886 buf_flush(int bufcnt)
893 * if we have reached the user specified byte count for each archive
894 * volume, prompt for the next volume. (The non-standard -R flag).
895 * NOTE: If the wrlimit is smaller than wrcnt, we will always write
896 * at least one record. We always round limit UP to next blocksize.
898 if ((wrlimit > 0) && (wrcnt > wrlimit)) {
899 paxwarn(0, "User specified archive volume byte limit reached.");
908 * The new archive volume might have changed the size of the
909 * write blocksize. if so we figure out if we need to write
910 * (one or more times), or if there is now free space left in
911 * the buffer (it is no longer full). bufcnt has the number of
912 * bytes in the buffer, (the blocksize, at the point we were
913 * CALLED). Push has the amount of "extra" data in the buffer
914 * if the block size has shrunk from a volume change.
916 bufend = buf + blksz;
920 push = bufcnt - blksz;
924 * We have enough data to write at least one archive block
928 * write a block and check if it all went out ok
930 cnt = ar_write(buf, blksz);
938 /* we have extra data to push to the front.
939 * check for more than 1 block of push, and if
940 * so we loop back to write again
942 memcpy(buf, bufend, push);
951 } else if (cnt > 0) {
953 * Oh drat we got a partial write!
954 * if format doesnt care about alignment let it go,
955 * we warned the user in ar_write().... but this means
956 * the last record on this volume violates pax spec....
962 memcpy(buf, bufpt, cnt);
964 if (!frmt->blkalgn || ((cnt % frmt->blkalgn) == 0))
970 * All done, go to next archive
977 * The new archive volume might also have changed the block
978 * size. if so, figure out if we have too much or too little
979 * data for using the new block size
981 bufend = buf + blksz;
985 push = bufcnt - blksz;
989 * write failed, stop pax. we must not create a bad archive!