]>
git.wh0rd.org - dump.git/blob - dump/tape.c
c3628ccc0b4a83d31d26d24e7e0bcc72d63021dc
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 1.26 2000/12/05 16:31:36 stelian Exp $";
48 #include <sys/types.h>
49 #include <linux/types.h>
51 #include <sys/param.h>
52 #include <sys/socket.h>
56 #include <linux/ext2_fs.h>
57 #include <bsdcompat.h>
60 #include <sys/vnode.h>
63 #include <ufs/inode.h>
65 #include <ufs/ufs/dinode.h>
66 #include <ufs/ffs/fs.h>
68 #endif /* __linux__ */
70 #include <protocols/dumprestore.h>
77 #include <compaterr.h>
87 #include <ext2fs/ext2fs.h>
92 int writesize
; /* size of malloc()ed buffer for tape */
93 long lastspclrec
= -1; /* tape block number of last written header */
94 int trecno
= 0; /* next record to write in current block */
95 extern long blocksperfile
; /* number of blocks per output file */
96 long blocksthisvol
; /* number of blocks on current output file */
97 extern int ntrec
; /* blocking factor on tape */
104 static ssize_t atomic_read
__P((int, void *, size_t));
105 static ssize_t atomic_write
__P((int, const void *, size_t));
106 static void doslave
__P((int, int));
107 static void enslave
__P((void));
108 static void flushtape
__P((void));
109 static void killall
__P((void));
110 static void rollforward
__P((void));
111 static int system_command
__P((const char *, const char *, int));
114 * Concurrent dump mods (Caltech) - disk block reading and tape writing
115 * are exported to several slave processes. While one slave writes the
116 * tape, the others read disk blocks; they pass control of the tape in
117 * a ring via signals. The parent process traverses the filesystem and
118 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
119 * The following structure defines the instruction packets sent to slaves.
127 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
129 int tapea
; /* header number at start of this chunk */
130 int count
; /* count to next header (used for TS_TAPE */
132 int inode
; /* inode that we are currently dealing with */
133 int fd
; /* FD for this slave */
134 int pid
; /* PID for this slave */
135 int sent
; /* 1 == we've sent this slave requests */
136 int firstrec
; /* record number of this block */
137 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
138 struct req
*req
; /* buffer for requests */
142 char (*nextblock
)[TP_BSIZE
];
144 static time_t tstart_volume
; /* time of volume start */
145 static int tapea_volume
; /* value of spcl.c_tapea at volume start */
147 int master
; /* pid of master, for sending error signals */
148 int tenths
; /* length of tape used per block written */
149 static int caught
; /* have we caught the signal to proceed? */
150 static int ready
; /* have we reached the lock point without having */
151 /* received the SIGUSR2 signal from the prev slave? */
152 static sigjmp_buf jmpbuf
; /* where to jump to if we are ready when the */
153 /* SIGUSR2 arrives from the previous slave */
158 int pgoff
= getpagesize() - 1;
162 writesize
= ntrec
* TP_BSIZE
;
163 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
165 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
166 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
167 * repositioning after stopping, i.e, streaming mode, where the gap is
168 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
170 if (blocksperfile
== 0 && !unlimited
)
171 tenths
= writesize
/ density
+
172 (cartridge
? 16 : density
== 625 ? 5 : 8);
174 * Allocate tape buffer contiguous with the array of instruction
175 * packets, so flushtape() can write them together with one write().
176 * Align tape buffer on page boundary to speed up tape write().
178 for (i
= 0; i
<= SLAVES
; i
++) {
180 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
183 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
185 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
187 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
189 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
195 nextblock
= slp
->tblock
;
200 writerec(const void *dp
, int isspcl
)
203 slp
->req
[trecno
].dblk
= (daddr_t
)0;
204 slp
->req
[trecno
].count
= 1;
205 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
206 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
209 lastspclrec
= spcl
.c_tapea
;
217 dumpblock(daddr_t blkno
, int size
)
219 int avail
, tpblks
, dblkno
;
221 dblkno
= fsbtodb(sblock
, blkno
);
222 tpblks
= size
>> tp_bshift
;
223 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
224 slp
->req
[trecno
].dblk
= dblkno
;
225 slp
->req
[trecno
].count
= avail
;
227 spcl
.c_tapea
+= avail
;
230 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
242 msg("write error on %s\n", tape
);
243 quit("Cannot recover\n");
246 msg("write error %d blocks into volume %d\n", blocksthisvol
, tapeno
);
247 broadcast("DUMP WRITE ERROR!\n");
248 if (!query("Do you want to restart?"))
250 msg("Closing this volume. Prepare to restart with new media;\n");
251 msg("this dump volume will be rewritten.\n");
262 quit("Broken pipe\n");
267 * Update xferrate stats
280 ttaken
= tnow
- tstart_volume
;
281 blocks
= spcl
.c_tapea
- tapea_volume
;
282 msg("Volume %d completed at: %s", tapeno
,
289 msg("Volume %d took %d:%02d:%02d\n", tapeno
,
290 ttaken
/ 3600, (ttaken
% 3600) / 60, ttaken
% 60);
291 msg("Volume %d transfer rate: %ld KB/s\n", tapeno
,
293 xferrate
+= blocks
/ ttaken
;
299 mktimeest(time_t tnow
)
301 static char msgbuf
[128];
306 if (blockswritten
< 500)
308 if (blockswritten
> tapesize
)
309 tapesize
= blockswritten
;
310 deltat
= tstart_writing
- tnow
+ (1.0 * (tnow
- tstart_writing
))
311 / blockswritten
* tapesize
;
312 (void)snprintf(msgbuf
, sizeof(msgbuf
),
313 "%3.2f%% done at %ld KB/s, finished in %d:%02d\n",
314 (blockswritten
* 100.0) / tapesize
,
315 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
316 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
324 * information message upon receipt of SIGINFO
327 statussig(int notused
)
330 int save_errno
= errno
;
336 (void) time((time_t *) &tnow
);
338 buf
= mktimeest(tnow
);
340 write(STDERR_FILENO
, buf
, strlen(buf
));
351 int siz
= (char *)nextblock
- (char *)slp
->req
;
353 slp
->req
[trecno
].count
= 0; /* Sentinel */
355 if (atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
356 quit("error writing command pipe: %s\n", strerror(errno
));
357 slp
->sent
= 1; /* we sent a request, read the response later */
359 lastfirstrec
= slp
->firstrec
;
361 if (++slp
>= &slaves
[SLAVES
])
364 /* Read results back from next slave */
366 if (atomic_read( slp
->fd
, (char *)&got
, sizeof got
)
368 perror(" DUMP: error reading command pipe in master");
373 /* Check for end of tape */
374 if (got
< writesize
) {
375 msg("End of tape detected\n");
378 * Drain the results, don't care what the values were.
379 * If we read them here then trewind won't...
381 for (i
= 0; i
< SLAVES
; i
++) {
382 if (slaves
[i
].sent
) {
383 if (atomic_read( slaves
[i
].fd
,
384 (char *)&got
, sizeof got
)
386 perror(" DUMP: error reading command pipe in master");
400 if (spcl
.c_type
!= TS_END
) {
401 for (i
= 0; i
< spcl
.c_count
; i
++)
402 if (spcl
.c_addr
[i
] != 0)
405 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
406 slp
->tapea
= spcl
.c_tapea
;
407 slp
->firstrec
= lastfirstrec
+ ntrec
;
409 nextblock
= slp
->tblock
;
412 blockswritten
+= ntrec
;
413 blocksthisvol
+= ntrec
;
414 if (!pipeout
&& !unlimited
&& (blocksperfile
?
415 (blocksthisvol
>= blocksperfile
) : (asize
> tsize
))) {
423 * Executes the command in a shell.
424 * Returns -1 if an error occured, the exit status of
425 * the command on success.
427 int system_command(const char *command
, const char *device
, int volnum
) {
429 char commandstr
[4096];
433 perror(" DUMP: unable to fork");
439 #if OLD_STYLE_FSCRIPT
440 snprintf(commandstr
, sizeof(commandstr
), "%s", command
);
442 snprintf(commandstr
, sizeof(commandstr
), "%s %s %d", command
, device
, volnum
);
444 commandstr
[sizeof(commandstr
) - 1] = '\0';
445 execl("/bin/sh", "sh", "-c", commandstr
, NULL
);
446 perror(" DUMP: unable to execute shell");
450 if (waitpid(pid
, &status
, 0) == -1) {
451 if (errno
!= EINTR
) {
452 perror(" DUMP: waitpid error");
456 if (WIFEXITED(status
))
457 return WEXITSTATUS(status
);
470 for (f
= 0; f
< SLAVES
; f
++) {
472 * Drain the results, but unlike EOT we DO (or should) care
473 * what the return values were, since if we detect EOT after
474 * we think we've written the last blocks to the tape anyway,
475 * we have to replay those blocks with rollforward.
477 * fixme: punt for now.
479 if (slaves
[f
].sent
) {
480 if (atomic_read( slaves
[f
].fd
, (char *)&got
, sizeof got
)
482 perror(" DUMP: error reading command pipe in master");
486 if (got
!= writesize
) {
487 msg("EOT detected in last 2 tape records!\n");
488 msg("Use a longer tape, decrease the size estimate\n");
489 quit("or use no size estimate at all.\n");
492 (void) close(slaves
[f
].fd
);
494 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
499 msg("Closing %s\n", tape
);
504 while (rmtopen(tape
, 0) < 0)
511 (void) close(tapefd
);
512 while ((f
= open(tape
, 0)) < 0)
517 if (eot_script
&& spcl
.c_type
!= TS_END
) {
518 msg("Launching %s\n", eot_script
);
519 eot_code
= system_command(eot_script
, tape
, tapeno
);
521 if (eot_code
!= 0 && eot_code
!= 1) {
522 msg("Dump aborted by the end of tape script\n");
534 if (nexttape
|| Mflag
|| (eot_code
== 0) )
537 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
538 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
540 while (!query("Is the new volume mounted and ready to go?"))
541 if (query("Do you want to abort?")) {
550 register struct req
*p
, *q
, *prev
;
551 register struct slave
*tslp
;
552 int i
, size
, savedtapea
, got
;
553 union u_spcl
*ntb
, *otb
;
558 tslp
= &slaves
[SLAVES
];
559 ntb
= (union u_spcl
*)tslp
->tblock
[1];
562 * Each of the N slaves should have requests that need to
563 * be replayed on the next tape. Use the extra slave buffers
564 * (slaves[SLAVES]) to construct request lists to be sent to
565 * each slave in turn.
567 for (i
= 0; i
< SLAVES
; i
++) {
569 otb
= (union u_spcl
*)slp
->tblock
;
572 * For each request in the current slave, copy it to tslp.
576 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
579 *ntb
++ = *otb
++; /* copy the datablock also */
584 quit("rollforward: protocol botch");
596 nextblock
= tslp
->tblock
;
597 savedtapea
= spcl
.c_tapea
;
598 spcl
.c_tapea
= slp
->tapea
;
600 spcl
.c_tapea
= savedtapea
;
601 lastspclrec
= savedtapea
- 1;
603 size
= (char *)ntb
- (char *)q
;
604 if (atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
605 perror(" DUMP: error writing command pipe");
610 lastfirstrec
= slp
->firstrec
;
612 if (++slp
>= &slaves
[SLAVES
])
617 if (prev
->dblk
!= 0) {
619 * If the last one was a disk block, make the
620 * first of this one be the last bit of that disk
623 q
->dblk
= prev
->dblk
+
624 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
625 ntb
= (union u_spcl
*)tslp
->tblock
;
628 * It wasn't a disk block. Copy the data to its
629 * new location in the buffer.
632 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
633 ntb
= (union u_spcl
*)tslp
->tblock
[1];
637 nextblock
= slp
->tblock
;
640 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
641 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
648 * Clear the first slaves' response. One hopes that it
649 * worked ok, otherwise the tape is much too short!
652 if (atomic_read( slp
->fd
, (char *)&got
, sizeof got
)
654 perror(" DUMP: error reading command pipe in master");
659 if (got
!= writesize
) {
660 quit("EOT detected at start of the tape!\n");
666 if (spcl
.c_type
!= TS_END
) {
667 for (i
= 0; i
< spcl
.c_count
; i
++)
668 if (spcl
.c_addr
[i
] != 0)
672 slp
->firstrec
= lastfirstrec
+ ntrec
;
673 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
676 blockswritten
+= ntrec
;
677 blocksthisvol
+= ntrec
;
682 * We implement taking and restoring checkpoints on the tape level.
683 * When each tape is opened, a new process is created by forking; this
684 * saves all of the necessary context in the parent. The child
685 * continues the dump; the parent waits around, saving the context.
686 * If the child returns X_REWRITE, then it had problems writing that tape;
687 * this causes the parent to fork again, duplicating the context, and
688 * everything continues as if nothing had happened.
691 startnewtape(int top
)
702 sigaddset(&sigs
, SIGINT
);
703 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
704 #else /* __linux__ */
706 void (*interrupt_save
)();
708 sig_t interrupt_save
;
710 interrupt_save
= signal(SIGINT
, SIG_IGN
);
711 #endif /* __linux__ */
713 parentpid
= getpid();
714 tapea_volume
= spcl
.c_tapea
;
716 (void)time4(&tstart_volume
);
718 (void)time((&tstart_volume
);
723 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
725 (void)signal(SIGINT
, interrupt_save
);
728 * All signals are inherited...
732 msg("Context save fork fails in parent %d\n", parentpid
);
738 * save the context by waiting
739 * until the child doing all of the work returns.
740 * don't catch the interrupt
743 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
745 signal(SIGINT
, SIG_IGN
);
748 msg("Tape: %d; parent process: %d child process %d\n",
749 tapeno
+1, parentpid
, childpid
);
751 while ((waitpid
= wait(&status
)) != childpid
)
752 if (waitpid
!= rshpid
)
753 msg("Parent %d waiting for child %d has another child %d return\n",
754 parentpid
, childpid
, waitpid
);
756 msg("Child %d returns LOB status %o\n",
757 childpid
, status
&0xFF);
759 status
= (status
>> 8) & 0xFF;
763 msg("Child %d finishes X_FINOK\n", childpid
);
766 msg("Child %d finishes X_ABORT\n", childpid
);
769 msg("Child %d finishes X_REWRITE\n", childpid
);
772 msg("Child %d finishes unknown %d\n",
783 goto restore_check_point
;
785 msg("Bad return code from dump: %d\n", status
);
789 } else { /* we are the child; just continue */
791 sleep(4); /* allow time for parent's message to get out */
792 msg("Child on Tape %d has parent %d, my pid = %d\n",
793 tapeno
+1, parentpid
, getpid());
796 * If we have a name like "/dev/rmt0,/dev/rmt1",
797 * use the name before the comma first, and save
798 * the remaining names for subsequent volumes.
800 tapeno
++; /* current tape sequence */
802 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
803 tape
[MAXPATHLEN
- 1] = '\0';
804 msg("Dumping volume %d on %s\n", tapeno
, tape
);
806 else if (nexttape
|| strchr(tapeprefix
, ',')) {
807 if (nexttape
&& *nexttape
)
808 tapeprefix
= nexttape
;
809 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
814 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
815 tape
[MAXPATHLEN
- 1] = '\0';
816 msg("Dumping volume %d on %s\n", tapeno
, tape
);
819 while ((tapefd
= (host
? rmtopen(tape
, 2) : pipeout
?
821 open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
823 while ((tapefd
= (pipeout
? fileno(stdout
) :
824 open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
827 msg("Cannot open output \"%s\".\n", tape
);
828 if (!query("Do you want to retry the open?"))
832 enslave(); /* Share open tape file descriptor with slaves */
837 newtape
++; /* new tape signal */
838 spcl
.c_count
= slp
->count
;
840 * measure firstrec in TP_BSIZE units since restore doesn't
841 * know the correct ntrec value...
843 spcl
.c_firstrec
= slp
->firstrec
;
845 spcl
.c_type
= TS_TAPE
;
846 spcl
.c_flags
|= DR_NEWHEADER
;
847 writeheader((ino_t
)slp
->inode
);
848 spcl
.c_flags
&=~ DR_NEWHEADER
;
849 msg("Volume %d started at: %s", tapeno
,
851 ctime4(&tstart_volume
));
853 ctime(&tstart_volume
));
856 msg("Volume %d begins with blocks from inode %d\n",
865 if (master
!= 0 && master
!= getpid())
866 /* Signals master to call dumpabort */
867 (void) kill(master
, SIGTERM
);
870 msg("The ENTIRE dump is aborted.\n");
883 msg("pid = %d exits with status %d\n", getpid(), status
);
889 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
895 siglongjmp(jmpbuf
, 1);
903 #ifdef LINUX_FORK_BUG
911 { struct sigaction sa
;
912 memset(&sa
, 0, sizeof sa
);
913 sigemptyset(&sa
.sa_mask
);
914 sa
.sa_handler
= dumpabort
;
915 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
916 sa
.sa_handler
= sigpipe
;
917 sigaction(SIGPIPE
, &sa
, NULL
);
918 sa
.sa_handler
= tperror
;
919 sigaction(SIGUSR1
, &sa
, NULL
); /* Slave sends SIGUSR1 on tape errors */
920 sa
.sa_handler
= proceed
;
921 sa
.sa_flags
= SA_RESTART
;
922 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
925 for (i
= 0; i
< SLAVES
; i
++) {
926 if (i
== slp
- &slaves
[0]) {
932 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
933 (slaves
[i
].pid
= fork()) < 0)
934 quit("too many slaves, %d (recompile smaller): %s\n",
937 slaves
[i
].fd
= cmd
[1];
939 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
941 for (j
= 0; j
<= i
; j
++)
942 (void) close(slaves
[j
].fd
);
944 sigaddset(&sigs
, SIGINT
); /* Master handles this */
946 sigaddset(&sigs
, SIGINFO
);
948 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
950 #ifdef LINUX_FORK_BUG
951 if (atomic_write( cmd
[0], (char *) &i
, sizeof i
)
953 quit("master/slave protocol botched 3\n");
960 #ifdef LINUX_FORK_BUG
962 * Wait for all slaves to _actually_ start to circumvent a bug in
963 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
964 * returned from fork() causes a SEGV in the child process
966 for (i
= 0; i
< SLAVES
; i
++)
967 if (atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
968 quit("master/slave protocol botched 4\n");
971 for (i
= 0; i
< SLAVES
; i
++)
972 (void) atomic_write( slaves
[i
].fd
,
973 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
974 sizeof slaves
[0].pid
);
984 for (i
= 0; i
< SLAVES
; i
++)
985 if (slaves
[i
].pid
> 0) {
986 (void) kill(slaves
[i
].pid
, SIGKILL
);
992 * Synchronization - each process has a lockfile, and shares file
993 * descriptors to the following process's lockfile. When our write
994 * completes, we release our lock on the following process's lock-
995 * file, allowing the following process to lock it and proceed. We
996 * get the lock back for the next cycle by swapping descriptors.
999 doslave(int cmd
, int slave_number
)
1002 int nextslave
, size
, eot_count
;
1003 volatile int wrote
= 0;
1010 * Need our own seek pointer.
1012 (void) close(diskfd
);
1013 if ((diskfd
= open(disk
, O_RDONLY
)) < 0)
1014 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1017 retval
= dump_fs_open(disk
, &fs
);
1019 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1020 #endif /* __linux__ */
1023 * Need the pid of the next slave in the loop...
1025 if ((nread
= atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1026 != sizeof nextslave
) {
1027 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1031 * Get list of blocks to dump, read the blocks into tape buffer
1033 while ((nread
= atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1034 register struct req
*p
= slp
->req
;
1036 for (trecno
= 0; trecno
< ntrec
;
1037 trecno
+= p
->count
, p
+= p
->count
) {
1039 bread(p
->dblk
, slp
->tblock
[trecno
],
1040 p
->count
* TP_BSIZE
);
1042 if (p
->count
!= 1 || atomic_read( cmd
,
1043 (char *)slp
->tblock
[trecno
],
1044 TP_BSIZE
) != TP_BSIZE
)
1045 quit("master/slave protocol botched.\n");
1048 if (sigsetjmp(jmpbuf
, 1) == 0) {
1056 /* Try to write the data... */
1061 while (eot_count
< 10 && size
< writesize
) {
1064 wrote
= rmtwrite(slp
->tblock
[0]+size
,
1068 wrote
= write(tapefd
, slp
->tblock
[0]+size
,
1071 printf("slave %d wrote %d\n", slave_number
, wrote
);
1081 if (size
!= writesize
)
1082 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1083 slave_number
, size
, writesize
);
1087 * Handle ENOSPC as an EOT condition.
1089 if (wrote
< 0 && errno
== ENOSPC
) {
1098 (void) kill(master
, SIGUSR1
);
1099 sigemptyset(&sigset
);
1100 sigaddset(&sigset
, SIGINT
);
1102 sigsuspend(&sigset
);
1105 * pass size of write back to master
1106 * (for EOT handling)
1108 (void) atomic_write( cmd
, (char *)&size
, sizeof size
);
1112 * If partial write, don't want next slave to go.
1113 * Also jolts him awake.
1115 (void) kill(nextslave
, SIGUSR2
);
1118 quit("error reading command pipe: %s\n", strerror(errno
));
1122 * Since a read from a pipe may not return all we asked for,
1123 * or a write may not write all we ask if we get a signal,
1124 * loop until the count is satisfied (or error).
1127 atomic_read(int fd
, void *buf
, size_t count
)
1129 int got
, need
= count
;
1132 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1134 } while (got
== -1 && errno
== EINTR
);
1135 return (got
< 0 ? got
: count
- need
);
1139 * Since a read from a pipe may not return all we asked for,
1140 * or a write may not write all we ask if we get a signal,
1141 * loop until the count is satisfied (or error).
1144 atomic_write(int fd
, const void *buf
, size_t count
)
1146 int got
, need
= count
;
1149 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1151 } while (got
== -1 && errno
== EINTR
);
1152 return (got
< 0 ? got
: count
- need
);