]>
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@noos.fr>, 1999-2000
6 * Stelian Pop <pop@noos.fr> - AlcĂ´ve <www.alcove.fr>, 2000
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
43 static const char rcsid
[] =
44 "$Id: tape.c,v 1.30 2001/02/16 13:38:47 stelian Exp $";
49 #include <sys/types.h>
50 #include <linux/types.h>
52 #include <sys/param.h>
53 #include <sys/socket.h>
57 #include <linux/ext2_fs.h>
58 #include <bsdcompat.h>
61 #include <sys/vnode.h>
64 #include <ufs/inode.h>
66 #include <ufs/ufs/dinode.h>
67 #include <ufs/ffs/fs.h>
69 #endif /* __linux__ */
71 #include <protocols/dumprestore.h>
78 #include <compaterr.h>
88 #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 */
105 static ssize_t atomic_read
__P((int, void *, size_t));
106 static ssize_t atomic_write
__P((int, const void *, size_t));
107 static void doslave
__P((int, int));
108 static void enslave
__P((void));
109 static void flushtape
__P((void));
110 static void killall
__P((void));
111 static void rollforward
__P((void));
112 static int system_command
__P((const char *, const char *, int));
115 * Concurrent dump mods (Caltech) - disk block reading and tape writing
116 * are exported to several slave processes. While one slave writes the
117 * tape, the others read disk blocks; they pass control of the tape in
118 * a ring via signals. The parent process traverses the filesystem and
119 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
120 * The following structure defines the instruction packets sent to slaves.
128 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
130 int tapea
; /* header number at start of this chunk */
131 int count
; /* count to next header (used for TS_TAPE */
133 int inode
; /* inode that we are currently dealing with */
134 int fd
; /* FD for this slave */
135 int pid
; /* PID for this slave */
136 int sent
; /* 1 == we've sent this slave requests */
137 int firstrec
; /* record number of this block */
138 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
139 struct req
*req
; /* buffer for requests */
143 char (*nextblock
)[TP_BSIZE
];
145 static time_t tstart_volume
; /* time of volume start */
146 static int tapea_volume
; /* value of spcl.c_tapea at volume start */
148 int master
; /* pid of master, for sending error signals */
149 int tenths
; /* length of tape used per block written */
150 static int caught
; /* have we caught the signal to proceed? */
151 static int ready
; /* have we reached the lock point without having */
152 /* received the SIGUSR2 signal from the prev slave? */
153 static sigjmp_buf jmpbuf
; /* where to jump to if we are ready when the */
154 /* SIGUSR2 arrives from the previous slave */
159 int pgoff
= getpagesize() - 1;
163 writesize
= ntrec
* TP_BSIZE
;
164 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
166 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
167 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
168 * repositioning after stopping, i.e, streaming mode, where the gap is
169 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
171 if (blocksperfile
== 0 && !unlimited
)
172 tenths
= writesize
/ density
+
173 (cartridge
? 16 : density
== 625 ? 5 : 8);
175 * Allocate tape buffer contiguous with the array of instruction
176 * packets, so flushtape() can write them together with one write().
177 * Align tape buffer on page boundary to speed up tape write().
179 for (i
= 0; i
<= SLAVES
; i
++) {
181 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
184 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
186 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
188 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
190 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
196 nextblock
= slp
->tblock
;
201 writerec(const void *dp
, int isspcl
)
204 slp
->req
[trecno
].dblk
= (daddr_t
)0;
205 slp
->req
[trecno
].count
= 1;
206 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
207 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
210 lastspclrec
= spcl
.c_tapea
;
218 dumpblock(daddr_t blkno
, int size
)
220 int avail
, tpblks
, dblkno
;
222 dblkno
= fsbtodb(sblock
, blkno
);
223 tpblks
= size
>> tp_bshift
;
224 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
225 slp
->req
[trecno
].dblk
= dblkno
;
226 slp
->req
[trecno
].count
= avail
;
228 spcl
.c_tapea
+= avail
;
231 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
243 msg("write error on %s: %s\n", tape
, strerror(errnum
));
244 quit("Cannot recover\n");
247 msg("write error %d blocks into volume %d: %s\n", blocksthisvol
, tapeno
, strerror(errnum
));
248 broadcast("DUMP WRITE ERROR!\n");
249 if (query("Do you want to rewrite this volume?")) {
250 msg("Closing this volume. Prepare to restart with new media;\n");
251 msg("this dump volume will be rewritten.\n");
257 if (query("Do you want to start the next tape?"))
266 quit("Broken pipe\n");
271 * Update xferrate stats
284 ttaken
= tnow
- tstart_volume
;
285 blocks
= spcl
.c_tapea
- tapea_volume
;
286 msg("Volume %d completed at: %s", tapeno
,
292 msg("Volume %d: %ld tape blocks (%.2fMB)\n", tapeno
,
293 blocks
, ((double)blocks
* TP_BSIZE
/ 1048576));
295 msg("Volume %d took %d:%02d:%02d\n", tapeno
,
296 ttaken
/ 3600, (ttaken
% 3600) / 60, ttaken
% 60);
297 msg("Volume %d transfer rate: %ld KB/s\n", tapeno
,
299 xferrate
+= blocks
/ ttaken
;
305 mktimeest(time_t tnow
)
307 static char msgbuf
[128];
312 if (blockswritten
< 500)
314 if (blockswritten
> tapesize
)
315 tapesize
= blockswritten
;
316 deltat
= tstart_writing
- tnow
+ (1.0 * (tnow
- tstart_writing
))
317 / blockswritten
* tapesize
;
318 if (tnow
> tstart_volume
)
319 (void)snprintf(msgbuf
, sizeof(msgbuf
),
320 "%3.2f%% done at %ld KB/s, finished in %d:%02d\n",
321 (blockswritten
* 100.0) / tapesize
,
322 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
323 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
325 (void)snprintf(msgbuf
, sizeof(msgbuf
),
326 "%3.2f%% done, finished in %d:%02d\n",
327 (blockswritten
* 100.0) / tapesize
,
328 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
336 * information message upon receipt of SIGINFO
339 statussig(int notused
)
342 int save_errno
= errno
;
348 (void) time((time_t *) &tnow
);
350 buf
= mktimeest(tnow
);
352 write(STDERR_FILENO
, buf
, strlen(buf
));
363 int siz
= (char *)nextblock
- (char *)slp
->req
;
365 slp
->req
[trecno
].count
= 0; /* Sentinel */
367 if (atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
368 quit("error writing command pipe: %s\n", strerror(errno
));
369 slp
->sent
= 1; /* we sent a request, read the response later */
371 lastfirstrec
= slp
->firstrec
;
373 if (++slp
>= &slaves
[SLAVES
])
376 /* Read results back from next slave */
378 if (atomic_read( slp
->fd
, (char *)&got
, sizeof got
)
380 perror(" DUMP: error reading command pipe in master");
385 /* Check for errors */
389 /* Check for end of tape */
390 if (got
< writesize
) {
391 msg("End of tape detected\n");
394 * Drain the results, don't care what the values were.
395 * If we read them here then trewind won't...
397 for (i
= 0; i
< SLAVES
; i
++) {
398 if (slaves
[i
].sent
) {
399 if (atomic_read( slaves
[i
].fd
,
400 (char *)&got
, sizeof got
)
402 perror(" DUMP: error reading command pipe in master");
416 if (spcl
.c_type
!= TS_END
) {
417 for (i
= 0; i
< spcl
.c_count
; i
++)
418 if (spcl
.c_addr
[i
] != 0)
421 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
422 slp
->tapea
= spcl
.c_tapea
;
423 slp
->firstrec
= lastfirstrec
+ ntrec
;
425 nextblock
= slp
->tblock
;
428 blockswritten
+= ntrec
;
429 blocksthisvol
+= ntrec
;
430 if (!pipeout
&& !unlimited
&& (blocksperfile
?
431 (blocksthisvol
>= blocksperfile
) : (asize
> tsize
))) {
439 * Executes the command in a shell.
440 * Returns -1 if an error occured, the exit status of
441 * the command on success.
443 int system_command(const char *command
, const char *device
, int volnum
) {
445 char commandstr
[4096];
449 perror(" DUMP: unable to fork");
455 #if OLD_STYLE_FSCRIPT
456 snprintf(commandstr
, sizeof(commandstr
), "%s", command
);
458 snprintf(commandstr
, sizeof(commandstr
), "%s %s %d", command
, device
, volnum
);
460 commandstr
[sizeof(commandstr
) - 1] = '\0';
461 execl("/bin/sh", "sh", "-c", commandstr
, NULL
);
462 perror(" DUMP: unable to execute shell");
466 if (waitpid(pid
, &status
, 0) == -1) {
467 if (errno
!= EINTR
) {
468 perror(" DUMP: waitpid error");
472 if (WIFEXITED(status
))
473 return WEXITSTATUS(status
);
486 for (f
= 0; f
< SLAVES
; f
++) {
488 * Drain the results, but unlike EOT we DO (or should) care
489 * what the return values were, since if we detect EOT after
490 * we think we've written the last blocks to the tape anyway,
491 * we have to replay those blocks with rollforward.
493 * fixme: punt for now.
495 if (slaves
[f
].sent
) {
496 if (atomic_read( slaves
[f
].fd
, (char *)&got
, sizeof got
)
498 perror(" DUMP: error reading command pipe in master");
506 if (got
!= writesize
) {
507 msg("EOT detected in last 2 tape records!\n");
508 msg("Use a longer tape, decrease the size estimate\n");
509 quit("or use no size estimate at all.\n");
512 (void) close(slaves
[f
].fd
);
514 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
519 msg("Closing %s\n", tape
);
524 while (rmtopen(tape
, 0) < 0)
531 (void) close(tapefd
);
532 while ((f
= open(tape
, 0)) < 0)
537 if (eot_script
&& spcl
.c_type
!= TS_END
) {
538 msg("Launching %s\n", eot_script
);
539 eot_code
= system_command(eot_script
, tape
, tapeno
);
541 if (eot_code
!= 0 && eot_code
!= 1) {
542 msg("Dump aborted by the end of tape script\n");
554 if (nexttape
|| Mflag
|| (eot_code
== 0) )
557 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
558 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
560 while (!query("Is the new volume mounted and ready to go?"))
561 if (query("Do you want to abort?")) {
570 register struct req
*p
, *q
, *prev
;
571 register struct slave
*tslp
;
572 int i
, size
, savedtapea
, got
;
573 union u_spcl
*ntb
, *otb
;
578 tslp
= &slaves
[SLAVES
];
579 ntb
= (union u_spcl
*)tslp
->tblock
[1];
582 * Each of the N slaves should have requests that need to
583 * be replayed on the next tape. Use the extra slave buffers
584 * (slaves[SLAVES]) to construct request lists to be sent to
585 * each slave in turn.
587 for (i
= 0; i
< SLAVES
; i
++) {
589 otb
= (union u_spcl
*)slp
->tblock
;
592 * For each request in the current slave, copy it to tslp.
596 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
599 *ntb
++ = *otb
++; /* copy the datablock also */
604 quit("rollforward: protocol botch");
616 nextblock
= tslp
->tblock
;
617 savedtapea
= spcl
.c_tapea
;
618 spcl
.c_tapea
= slp
->tapea
;
620 spcl
.c_tapea
= savedtapea
;
621 lastspclrec
= savedtapea
- 1;
623 size
= (char *)ntb
- (char *)q
;
624 if (atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
625 perror(" DUMP: error writing command pipe");
630 lastfirstrec
= slp
->firstrec
;
632 if (++slp
>= &slaves
[SLAVES
])
637 if (prev
->dblk
!= 0) {
639 * If the last one was a disk block, make the
640 * first of this one be the last bit of that disk
643 q
->dblk
= prev
->dblk
+
644 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
645 ntb
= (union u_spcl
*)tslp
->tblock
;
648 * It wasn't a disk block. Copy the data to its
649 * new location in the buffer.
652 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
653 ntb
= (union u_spcl
*)tslp
->tblock
[1];
657 nextblock
= slp
->tblock
;
660 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
661 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
668 * Clear the first slaves' response. One hopes that it
669 * worked ok, otherwise the tape is much too short!
672 if (atomic_read( slp
->fd
, (char *)&got
, sizeof got
)
674 perror(" DUMP: error reading command pipe in master");
682 if (got
!= writesize
) {
683 quit("EOT detected at start of the tape!\n");
689 if (spcl
.c_type
!= TS_END
) {
690 for (i
= 0; i
< spcl
.c_count
; i
++)
691 if (spcl
.c_addr
[i
] != 0)
695 slp
->firstrec
= lastfirstrec
+ ntrec
;
696 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
699 blockswritten
+= ntrec
;
700 blocksthisvol
+= ntrec
;
705 * We implement taking and restoring checkpoints on the tape level.
706 * When each tape is opened, a new process is created by forking; this
707 * saves all of the necessary context in the parent. The child
708 * continues the dump; the parent waits around, saving the context.
709 * If the child returns X_REWRITE, then it had problems writing that tape;
710 * this causes the parent to fork again, duplicating the context, and
711 * everything continues as if nothing had happened.
714 startnewtape(int top
)
725 sigaddset(&sigs
, SIGINT
);
726 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
727 #else /* __linux__ */
729 void (*interrupt_save
)();
731 sig_t interrupt_save
;
733 interrupt_save
= signal(SIGINT
, SIG_IGN
);
734 #endif /* __linux__ */
736 parentpid
= getpid();
737 tapea_volume
= spcl
.c_tapea
;
739 (void)time4(&tstart_volume
);
741 (void)time((&tstart_volume
);
746 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
748 (void)signal(SIGINT
, interrupt_save
);
751 * All signals are inherited...
755 msg("Context save fork fails in parent %d\n", parentpid
);
761 * save the context by waiting
762 * until the child doing all of the work returns.
763 * don't catch the interrupt
766 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
768 signal(SIGINT
, SIG_IGN
);
771 msg("Tape: %d; parent process: %d child process %d\n",
772 tapeno
+1, parentpid
, childpid
);
774 while ((waitpid
= wait(&status
)) != childpid
)
775 if (waitpid
!= rshpid
)
776 msg("Parent %d waiting for child %d has another child %d return\n",
777 parentpid
, childpid
, waitpid
);
779 msg("Child %d returns LOB status %o\n",
780 childpid
, status
&0xFF);
782 status
= (status
>> 8) & 0xFF;
786 msg("Child %d finishes X_FINOK\n", childpid
);
789 msg("Child %d finishes X_ABORT\n", childpid
);
792 msg("Child %d finishes X_REWRITE\n", childpid
);
795 msg("Child %d finishes unknown %d\n",
806 goto restore_check_point
;
808 msg("Bad return code from dump: %d\n", status
);
812 } else { /* we are the child; just continue */
814 sleep(4); /* allow time for parent's message to get out */
815 msg("Child on Tape %d has parent %d, my pid = %d\n",
816 tapeno
+1, parentpid
, getpid());
819 * If we have a name like "/dev/rmt0,/dev/rmt1",
820 * use the name before the comma first, and save
821 * the remaining names for subsequent volumes.
823 tapeno
++; /* current tape sequence */
825 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
826 tape
[MAXPATHLEN
- 1] = '\0';
827 msg("Dumping volume %d on %s\n", tapeno
, tape
);
829 else if (nexttape
|| strchr(tapeprefix
, ',')) {
830 if (nexttape
&& *nexttape
)
831 tapeprefix
= nexttape
;
832 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
837 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
838 tape
[MAXPATHLEN
- 1] = '\0';
839 msg("Dumping volume %d on %s\n", tapeno
, tape
);
842 while ((tapefd
= (host
? rmtopen(tape
, 2) : pipeout
?
844 open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
846 while ((tapefd
= (pipeout
? fileno(stdout
) :
847 open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
850 msg("Cannot open output \"%s\".\n", tape
);
851 if (!query("Do you want to retry the open?"))
855 enslave(); /* Share open tape file descriptor with slaves */
860 newtape
++; /* new tape signal */
861 spcl
.c_count
= slp
->count
;
863 * measure firstrec in TP_BSIZE units since restore doesn't
864 * know the correct ntrec value...
866 spcl
.c_firstrec
= slp
->firstrec
;
868 spcl
.c_type
= TS_TAPE
;
869 spcl
.c_flags
|= DR_NEWHEADER
;
870 writeheader((ino_t
)slp
->inode
);
871 spcl
.c_flags
&=~ DR_NEWHEADER
;
872 msg("Volume %d started at: %s", tapeno
,
874 ctime4(&tstart_volume
));
876 ctime(&tstart_volume
));
879 msg("Volume %d begins with blocks from inode %d\n",
888 if (master
!= 0 && master
!= getpid())
889 /* Signals master to call dumpabort */
890 (void) kill(master
, SIGTERM
);
893 msg("The ENTIRE dump is aborted.\n");
906 msg("pid = %d exits with status %d\n", getpid(), status
);
912 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
918 siglongjmp(jmpbuf
, 1);
926 #ifdef LINUX_FORK_BUG
934 { struct sigaction sa
;
935 memset(&sa
, 0, sizeof sa
);
936 sigemptyset(&sa
.sa_mask
);
937 sa
.sa_handler
= dumpabort
;
938 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
939 sa
.sa_handler
= sigpipe
;
940 sigaction(SIGPIPE
, &sa
, NULL
);
941 sa
.sa_handler
= proceed
;
942 sa
.sa_flags
= SA_RESTART
;
943 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
946 for (i
= 0; i
< SLAVES
; i
++) {
947 if (i
== slp
- &slaves
[0]) {
953 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
954 (slaves
[i
].pid
= fork()) < 0)
955 quit("too many slaves, %d (recompile smaller): %s\n",
958 slaves
[i
].fd
= cmd
[1];
960 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
962 for (j
= 0; j
<= i
; j
++)
963 (void) close(slaves
[j
].fd
);
965 sigaddset(&sigs
, SIGINT
); /* Master handles this */
967 sigaddset(&sigs
, SIGINFO
);
969 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
971 #ifdef LINUX_FORK_BUG
972 if (atomic_write( cmd
[0], (char *) &i
, sizeof i
)
974 quit("master/slave protocol botched 3\n");
981 #ifdef LINUX_FORK_BUG
983 * Wait for all slaves to _actually_ start to circumvent a bug in
984 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
985 * returned from fork() causes a SEGV in the child process
987 for (i
= 0; i
< SLAVES
; i
++)
988 if (atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
989 quit("master/slave protocol botched 4\n");
992 for (i
= 0; i
< SLAVES
; i
++)
993 (void) atomic_write( slaves
[i
].fd
,
994 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
995 sizeof slaves
[0].pid
);
1005 for (i
= 0; i
< SLAVES
; i
++)
1006 if (slaves
[i
].pid
> 0) {
1007 (void) kill(slaves
[i
].pid
, SIGKILL
);
1013 * Synchronization - each process has a lockfile, and shares file
1014 * descriptors to the following process's lockfile. When our write
1015 * completes, we release our lock on the following process's lock-
1016 * file, allowing the following process to lock it and proceed. We
1017 * get the lock back for the next cycle by swapping descriptors.
1020 doslave(int cmd
, int slave_number
)
1023 int nextslave
, size
, eot_count
;
1024 volatile int wrote
= 0;
1030 * Need our own seek pointer.
1032 (void) close(diskfd
);
1033 if ((diskfd
= open(disk
, O_RDONLY
)) < 0)
1034 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1037 retval
= dump_fs_open(disk
, &fs
);
1039 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1040 #endif /* __linux__ */
1043 * Need the pid of the next slave in the loop...
1045 if ((nread
= atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1046 != sizeof nextslave
) {
1047 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1051 * Get list of blocks to dump, read the blocks into tape buffer
1053 while ((nread
= atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1054 register struct req
*p
= slp
->req
;
1056 for (trecno
= 0; trecno
< ntrec
;
1057 trecno
+= p
->count
, p
+= p
->count
) {
1059 bread(p
->dblk
, slp
->tblock
[trecno
],
1060 p
->count
* TP_BSIZE
);
1062 if (p
->count
!= 1 || atomic_read( cmd
,
1063 (char *)slp
->tblock
[trecno
],
1064 TP_BSIZE
) != TP_BSIZE
)
1065 quit("master/slave protocol botched.\n");
1068 if (sigsetjmp(jmpbuf
, 1) == 0) {
1076 /* Try to write the data... */
1081 while (eot_count
< 10 && size
< writesize
) {
1084 wrote
= rmtwrite(slp
->tblock
[0]+size
,
1088 wrote
= write(tapefd
, slp
->tblock
[0]+size
,
1091 printf("slave %d wrote %d\n", slave_number
, wrote
);
1101 if (size
!= writesize
)
1102 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1103 slave_number
, size
, writesize
);
1107 * Handle ENOSPC as an EOT condition.
1109 if (wrote
< 0 && errno
== ENOSPC
) {
1118 * pass errno back to master for special handling
1124 * pass size of write back to master
1125 * (for EOT handling)
1127 (void) atomic_write( cmd
, (char *)&size
, sizeof size
);
1130 * If partial write, don't want next slave to go.
1131 * Also jolts him awake.
1133 (void) kill(nextslave
, SIGUSR2
);
1136 quit("error reading command pipe: %s\n", strerror(errno
));
1140 * Since a read from a pipe may not return all we asked for,
1141 * or a write may not write all we ask if we get a signal,
1142 * loop until the count is satisfied (or error).
1145 atomic_read(int fd
, void *buf
, size_t count
)
1147 int got
, need
= count
;
1150 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1152 } while (got
== -1 && errno
== EINTR
);
1153 return (got
< 0 ? got
: count
- need
);
1157 * Since a read from a pipe may not return all we asked for,
1158 * or a write may not write all we ask if we get a signal,
1159 * loop until the count is satisfied (or error).
1162 atomic_write(int fd
, const void *buf
, size_t count
)
1164 int got
, need
= count
;
1167 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1169 } while (got
== -1 && errno
== EINTR
);
1170 return (got
< 0 ? got
: count
- need
);