]>
git.wh0rd.org - dump.git/blob - dump/tape.c
2 * Ported to Linux's Second Extended File System as part of the
3 * dump and restore backup suit
4 * Remy Card <card@Linux.EU.Org>, 1994, 1995, 1996
9 * Copyright (c) 1980, 1991, 1993
10 * The Regents of the University of California. All rights reserved.
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. All advertising materials mentioning features or use of this software
21 * must display the following acknowledgement:
22 * This product includes software developed by the University of
23 * California, Berkeley and its contributors.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
42 static char sccsid
[] = "@(#)tape.c 8.4 (Berkeley) 5/1/95";
45 #include <sys/param.h>
46 #include <sys/socket.h>
50 #include <linux/ext2_fs.h>
51 #include <bsdcompat.h>
54 #include <sys/vnode.h>
57 #include <ufs/inode.h>
59 #include <ufs/ufs/dinode.h>
60 #include <ufs/ffs/fs.h>
62 #endif /* __linux__ */
64 #include <protocols/dumprestore.h>
80 #include <ext2fs/ext2fs.h>
83 #include "pathnames.h"
85 int writesize
; /* size of malloc()ed buffer for tape */
86 long lastspclrec
= -1; /* tape block number of last written header */
87 int trecno
= 0; /* next record to write in current block */
88 extern long blocksperfile
; /* number of blocks per output file */
89 long blocksthisvol
; /* number of blocks on current output file */
90 extern int ntrec
; /* blocking factor on tape */
95 static int atomic
__P((int (*)(), int, char *, int));
96 static void doslave
__P((int, int));
97 static void enslave
__P((void));
98 static void flushtape
__P((void));
99 static void killall
__P((void));
100 static void rollforward
__P((void));
103 * Concurrent dump mods (Caltech) - disk block reading and tape writing
104 * are exported to several slave processes. While one slave writes the
105 * tape, the others read disk blocks; they pass control of the tape in
106 * a ring via signals. The parent process traverses the filesystem and
107 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
108 * The following structure defines the instruction packets sent to slaves.
116 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
118 int tapea
; /* header number at start of this chunk */
119 int count
; /* count to next header (used for TS_TAPE */
121 int inode
; /* inode that we are currently dealing with */
122 int fd
; /* FD for this slave */
123 int pid
; /* PID for this slave */
124 int sent
; /* 1 == we've sent this slave requests */
125 int firstrec
; /* record number of this block */
126 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
127 struct req
*req
; /* buffer for requests */
131 char (*nextblock
)[TP_BSIZE
];
133 int master
; /* pid of master, for sending error signals */
134 int tenths
; /* length of tape used per block written */
135 static int caught
; /* have we caught the signal to proceed? */
136 static int ready
; /* have we reached the lock point without having */
137 /* received the SIGUSR2 signal from the prev slave? */
138 static jmp_buf jmpbuf
; /* where to jump to if we are ready when the */
139 /* SIGUSR2 arrives from the previous slave */
144 int pgoff
= getpagesize() - 1;
148 writesize
= ntrec
* TP_BSIZE
;
149 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
151 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
152 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
153 * repositioning after stopping, i.e, streaming mode, where the gap is
154 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
156 if (blocksperfile
== 0)
157 tenths
= writesize
/ density
+
158 (cartridge
? 16 : density
== 625 ? 5 : 8);
160 * Allocate tape buffer contiguous with the array of instruction
161 * packets, so flushtape() can write them together with one write().
162 * Align tape buffer on page boundary to speed up tape write().
164 for (i
= 0; i
<= SLAVES
; i
++) {
166 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
169 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
171 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
173 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
175 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
181 nextblock
= slp
->tblock
;
191 slp
->req
[trecno
].dblk
= (daddr_t
)0;
192 slp
->req
[trecno
].count
= 1;
193 *(union u_spcl
*)(*(nextblock
)++) = *(union u_spcl
*)dp
;
195 lastspclrec
= spcl
.c_tapea
;
203 dumpblock(blkno
, size
)
207 int avail
, tpblks
, dblkno
;
209 dblkno
= fsbtodb(sblock
, blkno
);
210 tpblks
= size
>> tp_bshift
;
211 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
212 slp
->req
[trecno
].dblk
= dblkno
;
213 slp
->req
[trecno
].count
= avail
;
215 spcl
.c_tapea
+= avail
;
218 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
231 msg("write error on %s\n", tape
);
232 quit("Cannot recover\n");
235 msg("write error %d blocks into volume %d\n", blocksthisvol
, tapeno
);
236 broadcast("DUMP WRITE ERROR!\n");
237 if (!query("Do you want to restart?"))
239 msg("Closing this volume. Prepare to restart with new media;\n");
240 msg("this dump volume will be rewritten.\n");
252 quit("Broken pipe\n");
261 int siz
= (char *)nextblock
- (char *)slp
->req
;
263 slp
->req
[trecno
].count
= 0; /* Sentinel */
265 if (atomic(write
, slp
->fd
, (char *)slp
->req
, siz
) != siz
)
266 quit("error writing command pipe: %s\n", strerror(errno
));
267 slp
->sent
= 1; /* we sent a request, read the response later */
269 lastfirstrec
= slp
->firstrec
;
271 if (++slp
>= &slaves
[SLAVES
])
274 /* Read results back from next slave */
276 if (atomic(read
, slp
->fd
, (char *)&got
, sizeof got
)
278 perror(" DUMP: error reading command pipe in master");
283 /* Check for end of tape */
284 if (got
< writesize
) {
285 msg("End of tape detected\n");
288 * Drain the results, don't care what the values were.
289 * If we read them here then trewind won't...
291 for (i
= 0; i
< SLAVES
; i
++) {
292 if (slaves
[i
].sent
) {
293 if (atomic(read
, slaves
[i
].fd
,
294 (char *)&got
, sizeof got
)
296 perror(" DUMP: error reading command pipe in master");
310 if (spcl
.c_type
!= TS_END
) {
311 for (i
= 0; i
< spcl
.c_count
; i
++)
312 if (spcl
.c_addr
[i
] != 0)
315 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
316 slp
->tapea
= spcl
.c_tapea
;
317 slp
->firstrec
= lastfirstrec
+ ntrec
;
319 nextblock
= slp
->tblock
;
322 blockswritten
+= ntrec
;
323 blocksthisvol
+= ntrec
;
324 if (!pipeout
&& (blocksperfile
?
325 (blocksthisvol
>= blocksperfile
) : (asize
> tsize
))) {
338 for (f
= 0; f
< SLAVES
; f
++) {
340 * Drain the results, but unlike EOT we DO (or should) care
341 * what the return values were, since if we detect EOT after
342 * we think we've written the last blocks to the tape anyway,
343 * we have to replay those blocks with rollforward.
345 * fixme: punt for now.
347 if (slaves
[f
].sent
) {
348 if (atomic(read
, slaves
[f
].fd
, (char *)&got
, sizeof got
)
350 perror(" DUMP: error reading command pipe in master");
354 if (got
!= writesize
) {
355 msg("EOT detected in last 2 tape records!\n");
356 msg("Use a longer tape, decrease the size estimate\n");
357 quit("or use no size estimate at all.\n");
360 (void) close(slaves
[f
].fd
);
362 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
368 msg("Closing %s\n", tape
);
373 while (rmtopen(tape
, 0) < 0)
379 (void) close(tapefd
);
380 while ((f
= open(tape
, 0)) < 0)
392 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
393 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
395 while (!query("Is the new volume mounted and ready to go?"))
396 if (query("Do you want to abort?")) {
405 register struct req
*p
, *q
, *prev
;
406 register struct slave
*tslp
;
407 int i
, size
, savedtapea
, got
;
408 union u_spcl
*ntb
, *otb
;
409 tslp
= &slaves
[SLAVES
];
410 ntb
= (union u_spcl
*)tslp
->tblock
[1];
413 * Each of the N slaves should have requests that need to
414 * be replayed on the next tape. Use the extra slave buffers
415 * (slaves[SLAVES]) to construct request lists to be sent to
416 * each slave in turn.
418 for (i
= 0; i
< SLAVES
; i
++) {
420 otb
= (union u_spcl
*)slp
->tblock
;
423 * For each request in the current slave, copy it to tslp.
427 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
430 *ntb
++ = *otb
++; /* copy the datablock also */
435 quit("rollforward: protocol botch");
447 nextblock
= tslp
->tblock
;
448 savedtapea
= spcl
.c_tapea
;
449 spcl
.c_tapea
= slp
->tapea
;
451 spcl
.c_tapea
= savedtapea
;
452 lastspclrec
= savedtapea
- 1;
454 size
= (char *)ntb
- (char *)q
;
455 if (atomic(write
, slp
->fd
, (char *)q
, size
) != size
) {
456 perror(" DUMP: error writing command pipe");
460 if (++slp
>= &slaves
[SLAVES
])
465 if (prev
->dblk
!= 0) {
467 * If the last one was a disk block, make the
468 * first of this one be the last bit of that disk
471 q
->dblk
= prev
->dblk
+
472 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
473 ntb
= (union u_spcl
*)tslp
->tblock
;
476 * It wasn't a disk block. Copy the data to its
477 * new location in the buffer.
480 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
481 ntb
= (union u_spcl
*)tslp
->tblock
[1];
485 nextblock
= slp
->tblock
;
491 * Clear the first slaves' response. One hopes that it
492 * worked ok, otherwise the tape is much too short!
495 if (atomic(read
, slp
->fd
, (char *)&got
, sizeof got
)
497 perror(" DUMP: error reading command pipe in master");
502 if (got
!= writesize
) {
503 quit("EOT detected at start of the tape!\n");
509 * We implement taking and restoring checkpoints on the tape level.
510 * When each tape is opened, a new process is created by forking; this
511 * saves all of the necessary context in the parent. The child
512 * continues the dump; the parent waits around, saving the context.
513 * If the child returns X_REWRITE, then it had problems writing that tape;
514 * this causes the parent to fork again, duplicating the context, and
515 * everything continues as if nothing had happened.
527 void (*interrupt_save
)();
528 #else /* __linux__ */
530 void (*interrupt_save
)();
532 sig_t interrupt_save
;
534 #endif /* __linux__ */
536 interrupt_save
= signal(SIGINT
, SIG_IGN
);
537 parentpid
= getpid();
540 (void)signal(SIGINT
, interrupt_save
);
542 * All signals are inherited...
546 msg("Context save fork fails in parent %d\n", parentpid
);
552 * save the context by waiting
553 * until the child doing all of the work returns.
554 * don't catch the interrupt
556 signal(SIGINT
, SIG_IGN
);
558 msg("Tape: %d; parent process: %d child process %d\n",
559 tapeno
+1, parentpid
, childpid
);
561 while ((waitpid
= wait(&status
)) != childpid
)
562 msg("Parent %d waiting for child %d has another child %d return\n",
563 parentpid
, childpid
, waitpid
);
565 msg("Child %d returns LOB status %o\n",
566 childpid
, status
&0xFF);
568 status
= (status
>> 8) & 0xFF;
572 msg("Child %d finishes X_FINOK\n", childpid
);
575 msg("Child %d finishes X_ABORT\n", childpid
);
578 msg("Child %d finishes X_REWRITE\n", childpid
);
581 msg("Child %d finishes unknown %d\n",
592 goto restore_check_point
;
594 msg("Bad return code from dump: %d\n", status
);
598 } else { /* we are the child; just continue */
600 sleep(4); /* allow time for parent's message to get out */
601 msg("Child on Tape %d has parent %d, my pid = %d\n",
602 tapeno
+1, parentpid
, getpid());
605 * If we have a name like "/dev/rmt0,/dev/rmt1",
606 * use the name before the comma first, and save
607 * the remaining names for subsequent volumes.
609 tapeno
++; /* current tape sequence */
610 if (nexttape
|| strchr(tape
, ',')) {
611 if (nexttape
&& *nexttape
)
613 if ((p
= strchr(tape
, ',')) != NULL
) {
618 msg("Dumping volume %d on %s\n", tapeno
, tape
);
621 while ((tapefd
= (host
? rmtopen(tape
, 2) :
622 pipeout
? 1 : open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
624 while ((tapefd
= (pipeout
? 1 :
625 open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
628 msg("Cannot open output \"%s\".\n", tape
);
629 if (!query("Do you want to retry the open?"))
633 enslave(); /* Share open tape file descriptor with slaves */
638 newtape
++; /* new tape signal */
639 spcl
.c_count
= slp
->count
;
641 * measure firstrec in TP_BSIZE units since restore doesn't
642 * know the correct ntrec value...
644 spcl
.c_firstrec
= slp
->firstrec
;
646 spcl
.c_type
= TS_TAPE
;
647 spcl
.c_flags
|= DR_NEWHEADER
;
648 writeheader((ino_t
)slp
->inode
);
649 spcl
.c_flags
&=~ DR_NEWHEADER
;
651 msg("Volume %d begins with blocks from inode %d\n",
661 if (master
!= 0 && master
!= getpid())
662 /* Signals master to call dumpabort */
663 (void) kill(master
, SIGTERM
);
666 msg("The ENTIRE dump is aborted.\n");
680 msg("pid = %d exits with status %d\n", getpid(), status
);
686 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
702 #ifdef LINUX_FORK_BUG
710 signal(SIGTERM
, dumpabort
); /* Slave sends SIGTERM on dumpabort() */
711 signal(SIGPIPE
, sigpipe
);
712 signal(SIGUSR1
, tperror
); /* Slave sends SIGUSR1 on tape errors */
713 signal(SIGUSR2
, proceed
); /* Slave sends SIGUSR2 to next slave */
715 for (i
= 0; i
< SLAVES
; i
++) {
716 if (i
== slp
- &slaves
[0]) {
722 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
723 (slaves
[i
].pid
= fork()) < 0)
724 quit("too many slaves, %d (recompile smaller): %s\n",
727 slaves
[i
].fd
= cmd
[1];
729 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
730 for (j
= 0; j
<= i
; j
++)
731 (void) close(slaves
[j
].fd
);
732 signal(SIGINT
, SIG_IGN
); /* Master handles this */
733 #ifdef LINUX_FORK_BUG
734 if (atomic(write
, cmd
[0], (char *) &i
, sizeof i
)
736 quit("master/slave protocol botched 3\n");
743 #ifdef LINUX_FORK_BUG
745 * Wait for all slaves to _actually_ start to circumvent a bug in
746 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
747 * returned from fork() causes a SEGV in the child process
749 for (i
= 0; i
< SLAVES
; i
++)
750 if (atomic(read
, slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
751 quit("master/slave protocol botched 4\n");
754 /* write pid of next slave to each slave */
755 for (i
= 0; i
< SLAVES
; i
++)
756 (void) atomic(write
, slaves
[i
].fd
,
757 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
758 sizeof slaves
[0].pid
);
768 for (i
= 0; i
< SLAVES
; i
++)
769 if (slaves
[i
].pid
> 0)
770 (void) kill(slaves
[i
].pid
, SIGKILL
);
774 * Synchronization - each process has a lockfile, and shares file
775 * descriptors to the following process's lockfile. When our write
776 * completes, we release our lock on the following process's lock-
777 * file, allowing the following process to lock it and proceed. We
778 * get the lock back for the next cycle by swapping descriptors.
781 doslave(cmd
, slave_number
)
786 int nextslave
, size
, wrote
, eot_count
;
792 msg("slave %d, pid %d\n", slave_number
, getpid());
795 * Need our own seek pointer.
797 (void) close(diskfd
);
798 if ((diskfd
= open(disk
, O_RDONLY
)) < 0)
799 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
802 retval
= ext2fs_open(disk
, 0, 0, 0, unix_io_manager
, &fs
);
804 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
805 #endif /* __linux__ */
808 * Need the pid of the next slave in the loop...
810 if ((nread
= atomic(read
, cmd
, (char *)&nextslave
, sizeof nextslave
))
811 != sizeof nextslave
) {
812 quit("master/slave protocol botched - didn't get pid of next slave.\n");
816 * Get list of blocks to dump, read the blocks into tape buffer
818 while ((nread
= atomic(read
, cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
819 register struct req
*p
= slp
->req
;
821 for (trecno
= 0; trecno
< ntrec
;
822 trecno
+= p
->count
, p
+= p
->count
) {
824 bread(p
->dblk
, slp
->tblock
[trecno
],
825 p
->count
* TP_BSIZE
);
827 if (p
->count
!= 1 || atomic(read
, cmd
,
828 (char *)slp
->tblock
[trecno
],
829 TP_BSIZE
) != TP_BSIZE
)
830 quit("master/slave protocol botched.\n");
833 if (setjmp(jmpbuf
) == 0) {
841 /* Try to write the data... */
845 while (eot_count
< 10 && size
< writesize
) {
848 wrote
= rmtwrite(slp
->tblock
[0]+size
,
852 wrote
= write(tapefd
, slp
->tblock
[0]+size
,
855 msg("slave %d wrote %d\n", slave_number
, wrote
);
865 if (size
!= writesize
)
866 msg("slave %d only wrote %d out of %d bytes and gave up.\n",
867 slave_number
, size
, writesize
);
874 * fixme: Pyramids running OSx return ENOSPC
875 * at EOT on 1/2 inch drives.
878 (void) kill(master
, SIGUSR1
);
883 * pass size of write back to master
886 (void) atomic(write
, cmd
, (char *)&size
, sizeof size
);
890 * If partial write, don't want next slave to go.
891 * Also jolts him awake.
893 (void) kill(nextslave
, SIGUSR2
);
896 quit("error reading command pipe: %s\n", strerror(errno
));
900 * Since a read from a pipe may not return all we asked for,
901 * or a write may not write all we ask if we get a signal,
902 * loop until the count is satisfied (or error).
905 atomic(func
, fd
, buf
, count
)
910 int got
, need
= count
;
912 while ((got
= (*func
)(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
914 return (got
< 0 ? got
: count
- need
);