]>
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>, 1994-1997
5 * Stelian Pop <pop@cybercable.fr>, 1999
10 * Copyright (c) 1980, 1991, 1993
11 * The Regents of the University of California. All rights reserved.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 * must display the following acknowledgement:
23 * This product includes software developed by the University of
24 * California, Berkeley and its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 * may be used to endorse or promote products derived from this software
27 * without specific prior written permission.
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
44 static char sccsid
[] = "@(#)tape.c 8.4 (Berkeley) 5/1/95";
46 static const char rcsid
[] =
47 "$Id: tape.c,v 1.2 1999/10/11 12:53:22 stelian Exp $";
51 #include <sys/types.h>
52 #include <linux/types.h>
54 #include <sys/param.h>
55 #include <sys/socket.h>
59 #include <linux/ext2_fs.h>
60 #include <bsdcompat.h>
63 #include <sys/vnode.h>
66 #include <ufs/inode.h>
68 #include <ufs/ufs/dinode.h>
69 #include <ufs/ffs/fs.h>
71 #endif /* __linux__ */
73 #include <protocols/dumprestore.h>
89 #include <ext2fs/ext2fs.h>
93 int writesize
; /* size of malloc()ed buffer for tape */
94 long lastspclrec
= -1; /* tape block number of last written header */
95 int trecno
= 0; /* next record to write in current block */
96 extern long blocksperfile
; /* number of blocks per output file */
97 long blocksthisvol
; /* number of blocks on current output file */
98 extern int ntrec
; /* blocking factor on tape */
103 static int atomic
__P((ssize_t (*)(), int, char *, int));
104 static void doslave
__P((int, int));
105 static void enslave
__P((void));
106 static void flushtape
__P((void));
107 static void killall
__P((void));
108 static void rollforward
__P((void));
111 * Concurrent dump mods (Caltech) - disk block reading and tape writing
112 * are exported to several slave processes. While one slave writes the
113 * tape, the others read disk blocks; they pass control of the tape in
114 * a ring via signals. The parent process traverses the filesystem and
115 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
116 * The following structure defines the instruction packets sent to slaves.
124 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
126 int tapea
; /* header number at start of this chunk */
127 int count
; /* count to next header (used for TS_TAPE */
129 int inode
; /* inode that we are currently dealing with */
130 int fd
; /* FD for this slave */
131 int pid
; /* PID for this slave */
132 int sent
; /* 1 == we've sent this slave requests */
133 int firstrec
; /* record number of this block */
134 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
135 struct req
*req
; /* buffer for requests */
139 char (*nextblock
)[TP_BSIZE
];
141 int master
; /* pid of master, for sending error signals */
142 int tenths
; /* length of tape used per block written */
143 static int caught
; /* have we caught the signal to proceed? */
144 static int ready
; /* have we reached the lock point without having */
145 /* received the SIGUSR2 signal from the prev slave? */
146 static jmp_buf jmpbuf
; /* where to jump to if we are ready when the */
147 /* SIGUSR2 arrives from the previous slave */
152 int pgoff
= getpagesize() - 1;
156 writesize
= ntrec
* TP_BSIZE
;
157 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
159 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
160 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
161 * repositioning after stopping, i.e, streaming mode, where the gap is
162 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
164 if (blocksperfile
== 0 && !unlimited
)
165 tenths
= writesize
/ density
+
166 (cartridge
? 16 : density
== 625 ? 5 : 8);
168 * Allocate tape buffer contiguous with the array of instruction
169 * packets, so flushtape() can write them together with one write().
170 * Align tape buffer on page boundary to speed up tape write().
172 for (i
= 0; i
<= SLAVES
; i
++) {
174 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
177 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
179 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
181 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
183 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
189 nextblock
= slp
->tblock
;
199 slp
->req
[trecno
].dblk
= (daddr_t
)0;
200 slp
->req
[trecno
].count
= 1;
201 *(union u_spcl
*)(*(nextblock
)++) = *(union u_spcl
*)dp
;
203 lastspclrec
= spcl
.c_tapea
;
211 dumpblock(blkno
, size
)
215 int avail
, tpblks
, dblkno
;
217 dblkno
= fsbtodb(sblock
, blkno
);
218 tpblks
= size
>> tp_bshift
;
219 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
220 slp
->req
[trecno
].dblk
= dblkno
;
221 slp
->req
[trecno
].count
= avail
;
223 spcl
.c_tapea
+= avail
;
226 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
239 msg("write error on %s\n", tape
);
240 quit("Cannot recover\n");
243 msg("write error %d blocks into volume %d\n", blocksthisvol
, tapeno
);
244 broadcast("DUMP WRITE ERROR!\n");
245 if (!query("Do you want to restart?"))
247 msg("Closing this volume. Prepare to restart with new media;\n");
248 msg("this dump volume will be rewritten.\n");
260 quit("Broken pipe\n");
269 int siz
= (char *)nextblock
- (char *)slp
->req
;
271 slp
->req
[trecno
].count
= 0; /* Sentinel */
273 if (atomic(write
, slp
->fd
, (char *)slp
->req
, siz
) != siz
)
274 quit("error writing command pipe: %s\n", strerror(errno
));
275 slp
->sent
= 1; /* we sent a request, read the response later */
277 lastfirstrec
= slp
->firstrec
;
279 if (++slp
>= &slaves
[SLAVES
])
282 /* Read results back from next slave */
284 if (atomic(read
, slp
->fd
, (char *)&got
, sizeof got
)
286 perror(" DUMP: error reading command pipe in master");
291 /* Check for end of tape */
292 if (got
< writesize
) {
293 msg("End of tape detected\n");
296 * Drain the results, don't care what the values were.
297 * If we read them here then trewind won't...
299 for (i
= 0; i
< SLAVES
; i
++) {
300 if (slaves
[i
].sent
) {
301 if (atomic(read
, slaves
[i
].fd
,
302 (char *)&got
, sizeof got
)
304 perror(" DUMP: error reading command pipe in master");
318 if (spcl
.c_type
!= TS_END
) {
319 for (i
= 0; i
< spcl
.c_count
; i
++)
320 if (spcl
.c_addr
[i
] != 0)
323 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
324 slp
->tapea
= spcl
.c_tapea
;
325 slp
->firstrec
= lastfirstrec
+ ntrec
;
327 nextblock
= slp
->tblock
;
330 blockswritten
+= ntrec
;
331 blocksthisvol
+= ntrec
;
332 if (!pipeout
&& !unlimited
&& (blocksperfile
?
333 (blocksthisvol
>= blocksperfile
) : (asize
> tsize
))) {
346 for (f
= 0; f
< SLAVES
; f
++) {
348 * Drain the results, but unlike EOT we DO (or should) care
349 * what the return values were, since if we detect EOT after
350 * we think we've written the last blocks to the tape anyway,
351 * we have to replay those blocks with rollforward.
353 * fixme: punt for now.
355 if (slaves
[f
].sent
) {
356 if (atomic(read
, slaves
[f
].fd
, (char *)&got
, sizeof got
)
358 perror(" DUMP: error reading command pipe in master");
362 if (got
!= writesize
) {
363 msg("EOT detected in last 2 tape records!\n");
364 msg("Use a longer tape, decrease the size estimate\n");
365 quit("or use no size estimate at all.\n");
368 (void) close(slaves
[f
].fd
);
370 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
376 msg("Closing %s\n", tape
);
381 while (rmtopen(tape
, 0) < 0)
387 (void) close(tapefd
);
388 while ((f
= open(tape
, 0)) < 0)
396 time_t tstart_changevol
, tend_changevol
;
402 (void)time4(&(tstart_changevol
));
404 (void)time((time_t *)&(tstart_changevol
));
407 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
408 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
410 while (!query("Is the new volume mounted and ready to go?"))
411 if (query("Do you want to abort?")) {
416 (void)time4(&(tend_changevol
));
418 (void)time((time_t *)&(tend_changevol
));
420 if ((tstart_changevol
!= (time_t)-1) && (tend_changevol
!= (time_t)-1))
421 tstart_writing
+= (tend_changevol
- tstart_changevol
);
427 register struct req
*p
, *q
, *prev
;
428 register struct slave
*tslp
;
429 int i
, size
, savedtapea
, got
;
430 union u_spcl
*ntb
, *otb
;
435 tslp
= &slaves
[SLAVES
];
436 ntb
= (union u_spcl
*)tslp
->tblock
[1];
439 * Each of the N slaves should have requests that need to
440 * be replayed on the next tape. Use the extra slave buffers
441 * (slaves[SLAVES]) to construct request lists to be sent to
442 * each slave in turn.
444 for (i
= 0; i
< SLAVES
; i
++) {
446 otb
= (union u_spcl
*)slp
->tblock
;
449 * For each request in the current slave, copy it to tslp.
453 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
456 *ntb
++ = *otb
++; /* copy the datablock also */
461 quit("rollforward: protocol botch");
473 nextblock
= tslp
->tblock
;
474 savedtapea
= spcl
.c_tapea
;
475 spcl
.c_tapea
= slp
->tapea
;
477 spcl
.c_tapea
= savedtapea
;
478 lastspclrec
= savedtapea
- 1;
480 size
= (char *)ntb
- (char *)q
;
481 if (atomic(write
, slp
->fd
, (char *)q
, size
) != size
) {
482 perror(" DUMP: error writing command pipe");
487 lastfirstrec
= slp
->firstrec
;
489 if (++slp
>= &slaves
[SLAVES
])
494 if (prev
->dblk
!= 0) {
496 * If the last one was a disk block, make the
497 * first of this one be the last bit of that disk
500 q
->dblk
= prev
->dblk
+
501 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
502 ntb
= (union u_spcl
*)tslp
->tblock
;
505 * It wasn't a disk block. Copy the data to its
506 * new location in the buffer.
509 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
510 ntb
= (union u_spcl
*)tslp
->tblock
[1];
514 nextblock
= slp
->tblock
;
517 *(union u_spcl
*)(*(nextblock
)++) = *(union u_spcl
*)tslp
->tblock
;
524 * Clear the first slaves' response. One hopes that it
525 * worked ok, otherwise the tape is much too short!
528 if (atomic(read
, slp
->fd
, (char *)&got
, sizeof got
)
530 perror(" DUMP: error reading command pipe in master");
535 if (got
!= writesize
) {
536 quit("EOT detected at start of the tape!\n");
542 if (spcl
.c_type
!= TS_END
) {
543 for (i
= 0; i
< spcl
.c_count
; i
++)
544 if (spcl
.c_addr
[i
] != 0)
548 slp
->firstrec
= lastfirstrec
+ ntrec
;
549 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
552 blockswritten
+= ntrec
;
553 blocksthisvol
+= ntrec
;
558 * We implement taking and restoring checkpoints on the tape level.
559 * When each tape is opened, a new process is created by forking; this
560 * saves all of the necessary context in the parent. The child
561 * continues the dump; the parent waits around, saving the context.
562 * If the child returns X_REWRITE, then it had problems writing that tape;
563 * this causes the parent to fork again, duplicating the context, and
564 * everything continues as if nothing had happened.
576 void (*interrupt_save
)();
577 #else /* __linux__ */
579 void (*interrupt_save
)();
581 sig_t interrupt_save
;
583 #endif /* __linux__ */
585 interrupt_save
= signal(SIGINT
, SIG_IGN
);
586 parentpid
= getpid();
589 (void)signal(SIGINT
, interrupt_save
);
591 * All signals are inherited...
595 msg("Context save fork fails in parent %d\n", parentpid
);
601 * save the context by waiting
602 * until the child doing all of the work returns.
603 * don't catch the interrupt
605 signal(SIGINT
, SIG_IGN
);
607 msg("Tape: %d; parent process: %d child process %d\n",
608 tapeno
+1, parentpid
, childpid
);
610 while ((waitpid
= wait(&status
)) != childpid
)
611 msg("Parent %d waiting for child %d has another child %d return\n",
612 parentpid
, childpid
, waitpid
);
614 msg("Child %d returns LOB status %o\n",
615 childpid
, status
&0xFF);
617 status
= (status
>> 8) & 0xFF;
621 msg("Child %d finishes X_FINOK\n", childpid
);
624 msg("Child %d finishes X_ABORT\n", childpid
);
627 msg("Child %d finishes X_REWRITE\n", childpid
);
630 msg("Child %d finishes unknown %d\n",
641 goto restore_check_point
;
643 msg("Bad return code from dump: %d\n", status
);
647 } else { /* we are the child; just continue */
649 sleep(4); /* allow time for parent's message to get out */
650 msg("Child on Tape %d has parent %d, my pid = %d\n",
651 tapeno
+1, parentpid
, getpid());
654 * If we have a name like "/dev/rmt0,/dev/rmt1",
655 * use the name before the comma first, and save
656 * the remaining names for subsequent volumes.
658 tapeno
++; /* current tape sequence */
659 if (nexttape
|| strchr(tape
, ',')) {
660 if (nexttape
&& *nexttape
)
662 if ((p
= strchr(tape
, ',')) != NULL
) {
667 msg("Dumping volume %d on %s\n", tapeno
, tape
);
670 while ((tapefd
= (host
? rmtopen(tape
, 2) :
671 pipeout
? 1 : open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
673 while ((tapefd
= (pipeout
? 1 :
674 open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
677 msg("Cannot open output \"%s\".\n", tape
);
678 if (!query("Do you want to retry the open?"))
682 enslave(); /* Share open tape file descriptor with slaves */
687 newtape
++; /* new tape signal */
688 spcl
.c_count
= slp
->count
;
690 * measure firstrec in TP_BSIZE units since restore doesn't
691 * know the correct ntrec value...
693 spcl
.c_firstrec
= slp
->firstrec
;
695 spcl
.c_type
= TS_TAPE
;
696 spcl
.c_flags
|= DR_NEWHEADER
;
697 writeheader((ino_t
)slp
->inode
);
698 spcl
.c_flags
&=~ DR_NEWHEADER
;
700 msg("Volume %d begins with blocks from inode %d\n",
710 if (master
!= 0 && master
!= getpid())
711 /* Signals master to call dumpabort */
712 (void) kill(master
, SIGTERM
);
715 msg("The ENTIRE dump is aborted.\n");
729 msg("pid = %d exits with status %d\n", getpid(), status
);
735 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
751 #ifdef LINUX_FORK_BUG
759 signal(SIGTERM
, dumpabort
); /* Slave sends SIGTERM on dumpabort() */
760 signal(SIGPIPE
, sigpipe
);
761 signal(SIGUSR1
, tperror
); /* Slave sends SIGUSR1 on tape errors */
762 signal(SIGUSR2
, proceed
); /* Slave sends SIGUSR2 to next slave */
764 for (i
= 0; i
< SLAVES
; i
++) {
765 if (i
== slp
- &slaves
[0]) {
771 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
772 (slaves
[i
].pid
= fork()) < 0)
773 quit("too many slaves, %d (recompile smaller): %s\n",
776 slaves
[i
].fd
= cmd
[1];
778 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
779 for (j
= 0; j
<= i
; j
++)
780 (void) close(slaves
[j
].fd
);
781 signal(SIGINT
, SIG_IGN
); /* Master handles this */
782 #ifdef LINUX_FORK_BUG
783 if (atomic(write
, cmd
[0], (char *) &i
, sizeof i
)
785 quit("master/slave protocol botched 3\n");
792 #ifdef LINUX_FORK_BUG
794 * Wait for all slaves to _actually_ start to circumvent a bug in
795 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
796 * returned from fork() causes a SEGV in the child process
798 for (i
= 0; i
< SLAVES
; i
++)
799 if (atomic(read
, slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
800 quit("master/slave protocol botched 4\n");
803 for (i
= 0; i
< SLAVES
; i
++)
804 (void) atomic(write
, slaves
[i
].fd
,
805 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
806 sizeof slaves
[0].pid
);
816 for (i
= 0; i
< SLAVES
; i
++)
817 if (slaves
[i
].pid
> 0) {
818 (void) kill(slaves
[i
].pid
, SIGKILL
);
824 * Synchronization - each process has a lockfile, and shares file
825 * descriptors to the following process's lockfile. When our write
826 * completes, we release our lock on the following process's lock-
827 * file, allowing the following process to lock it and proceed. We
828 * get the lock back for the next cycle by swapping descriptors.
831 doslave(cmd
, slave_number
)
836 int nextslave
, size
, wrote
, eot_count
;
842 * Need our own seek pointer.
844 (void) close(diskfd
);
845 if ((diskfd
= open(disk
, O_RDONLY
)) < 0)
846 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
849 retval
= ext2fs_open(disk
, 0, 0, 0, unix_io_manager
, &fs
);
851 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
852 #endif /* __linux__ */
855 * Need the pid of the next slave in the loop...
857 if ((nread
= atomic(read
, cmd
, (char *)&nextslave
, sizeof nextslave
))
858 != sizeof nextslave
) {
859 quit("master/slave protocol botched - didn't get pid of next slave.\n");
863 * Get list of blocks to dump, read the blocks into tape buffer
865 while ((nread
= atomic(read
, cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
866 register struct req
*p
= slp
->req
;
868 for (trecno
= 0; trecno
< ntrec
;
869 trecno
+= p
->count
, p
+= p
->count
) {
871 bread(p
->dblk
, slp
->tblock
[trecno
],
872 p
->count
* TP_BSIZE
);
874 if (p
->count
!= 1 || atomic(read
, cmd
,
875 (char *)slp
->tblock
[trecno
],
876 TP_BSIZE
) != TP_BSIZE
)
877 quit("master/slave protocol botched.\n");
880 if (setjmp(jmpbuf
) == 0) {
888 /* Try to write the data... */
893 while (eot_count
< 10 && size
< writesize
) {
896 wrote
= rmtwrite(slp
->tblock
[0]+size
,
900 wrote
= write(tapefd
, slp
->tblock
[0]+size
,
903 printf("slave %d wrote %d\n", slave_number
, wrote
);
913 if (size
!= writesize
)
914 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
915 slave_number
, size
, writesize
);
919 * Handle ENOSPC as an EOT condition.
921 if (wrote
< 0 && errno
== ENOSPC
) {
930 (void) kill(master
, SIGUSR1
);
935 * pass size of write back to master
938 (void) atomic(write
, cmd
, (char *)&size
, sizeof size
);
942 * If partial write, don't want next slave to go.
943 * Also jolts him awake.
945 (void) kill(nextslave
, SIGUSR2
);
948 quit("error reading command pipe: %s\n", strerror(errno
));
952 * Since a read from a pipe may not return all we asked for,
953 * or a write may not write all we ask if we get a signal,
954 * loop until the count is satisfied (or error).
957 atomic(func
, fd
, buf
, count
)
963 int got
, need
= count
;
965 while ((got
= (*func
)(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
967 return (got
< 0 ? got
: count
- need
);