]>
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 1.29 2001/02/16 10:40:48 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 (void)snprintf(msgbuf
, sizeof(msgbuf
),
319 "%3.2f%% done at %ld KB/s, finished in %d:%02d\n",
320 (blockswritten
* 100.0) / tapesize
,
321 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
322 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
330 * information message upon receipt of SIGINFO
333 statussig(int notused
)
336 int save_errno
= errno
;
342 (void) time((time_t *) &tnow
);
344 buf
= mktimeest(tnow
);
346 write(STDERR_FILENO
, buf
, strlen(buf
));
357 int siz
= (char *)nextblock
- (char *)slp
->req
;
359 slp
->req
[trecno
].count
= 0; /* Sentinel */
361 if (atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
362 quit("error writing command pipe: %s\n", strerror(errno
));
363 slp
->sent
= 1; /* we sent a request, read the response later */
365 lastfirstrec
= slp
->firstrec
;
367 if (++slp
>= &slaves
[SLAVES
])
370 /* Read results back from next slave */
372 if (atomic_read( slp
->fd
, (char *)&got
, sizeof got
)
374 perror(" DUMP: error reading command pipe in master");
379 /* Check for errors */
383 /* Check for end of tape */
384 if (got
< writesize
) {
385 msg("End of tape detected\n");
388 * Drain the results, don't care what the values were.
389 * If we read them here then trewind won't...
391 for (i
= 0; i
< SLAVES
; i
++) {
392 if (slaves
[i
].sent
) {
393 if (atomic_read( slaves
[i
].fd
,
394 (char *)&got
, sizeof got
)
396 perror(" DUMP: error reading command pipe in master");
410 if (spcl
.c_type
!= TS_END
) {
411 for (i
= 0; i
< spcl
.c_count
; i
++)
412 if (spcl
.c_addr
[i
] != 0)
415 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
416 slp
->tapea
= spcl
.c_tapea
;
417 slp
->firstrec
= lastfirstrec
+ ntrec
;
419 nextblock
= slp
->tblock
;
422 blockswritten
+= ntrec
;
423 blocksthisvol
+= ntrec
;
424 if (!pipeout
&& !unlimited
&& (blocksperfile
?
425 (blocksthisvol
>= blocksperfile
) : (asize
> tsize
))) {
433 * Executes the command in a shell.
434 * Returns -1 if an error occured, the exit status of
435 * the command on success.
437 int system_command(const char *command
, const char *device
, int volnum
) {
439 char commandstr
[4096];
443 perror(" DUMP: unable to fork");
449 #if OLD_STYLE_FSCRIPT
450 snprintf(commandstr
, sizeof(commandstr
), "%s", command
);
452 snprintf(commandstr
, sizeof(commandstr
), "%s %s %d", command
, device
, volnum
);
454 commandstr
[sizeof(commandstr
) - 1] = '\0';
455 execl("/bin/sh", "sh", "-c", commandstr
, NULL
);
456 perror(" DUMP: unable to execute shell");
460 if (waitpid(pid
, &status
, 0) == -1) {
461 if (errno
!= EINTR
) {
462 perror(" DUMP: waitpid error");
466 if (WIFEXITED(status
))
467 return WEXITSTATUS(status
);
480 for (f
= 0; f
< SLAVES
; f
++) {
482 * Drain the results, but unlike EOT we DO (or should) care
483 * what the return values were, since if we detect EOT after
484 * we think we've written the last blocks to the tape anyway,
485 * we have to replay those blocks with rollforward.
487 * fixme: punt for now.
489 if (slaves
[f
].sent
) {
490 if (atomic_read( slaves
[f
].fd
, (char *)&got
, sizeof got
)
492 perror(" DUMP: error reading command pipe in master");
500 if (got
!= writesize
) {
501 msg("EOT detected in last 2 tape records!\n");
502 msg("Use a longer tape, decrease the size estimate\n");
503 quit("or use no size estimate at all.\n");
506 (void) close(slaves
[f
].fd
);
508 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
513 msg("Closing %s\n", tape
);
518 while (rmtopen(tape
, 0) < 0)
525 (void) close(tapefd
);
526 while ((f
= open(tape
, 0)) < 0)
531 if (eot_script
&& spcl
.c_type
!= TS_END
) {
532 msg("Launching %s\n", eot_script
);
533 eot_code
= system_command(eot_script
, tape
, tapeno
);
535 if (eot_code
!= 0 && eot_code
!= 1) {
536 msg("Dump aborted by the end of tape script\n");
548 if (nexttape
|| Mflag
|| (eot_code
== 0) )
551 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
552 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
554 while (!query("Is the new volume mounted and ready to go?"))
555 if (query("Do you want to abort?")) {
564 register struct req
*p
, *q
, *prev
;
565 register struct slave
*tslp
;
566 int i
, size
, savedtapea
, got
;
567 union u_spcl
*ntb
, *otb
;
572 tslp
= &slaves
[SLAVES
];
573 ntb
= (union u_spcl
*)tslp
->tblock
[1];
576 * Each of the N slaves should have requests that need to
577 * be replayed on the next tape. Use the extra slave buffers
578 * (slaves[SLAVES]) to construct request lists to be sent to
579 * each slave in turn.
581 for (i
= 0; i
< SLAVES
; i
++) {
583 otb
= (union u_spcl
*)slp
->tblock
;
586 * For each request in the current slave, copy it to tslp.
590 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
593 *ntb
++ = *otb
++; /* copy the datablock also */
598 quit("rollforward: protocol botch");
610 nextblock
= tslp
->tblock
;
611 savedtapea
= spcl
.c_tapea
;
612 spcl
.c_tapea
= slp
->tapea
;
614 spcl
.c_tapea
= savedtapea
;
615 lastspclrec
= savedtapea
- 1;
617 size
= (char *)ntb
- (char *)q
;
618 if (atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
619 perror(" DUMP: error writing command pipe");
624 lastfirstrec
= slp
->firstrec
;
626 if (++slp
>= &slaves
[SLAVES
])
631 if (prev
->dblk
!= 0) {
633 * If the last one was a disk block, make the
634 * first of this one be the last bit of that disk
637 q
->dblk
= prev
->dblk
+
638 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
639 ntb
= (union u_spcl
*)tslp
->tblock
;
642 * It wasn't a disk block. Copy the data to its
643 * new location in the buffer.
646 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
647 ntb
= (union u_spcl
*)tslp
->tblock
[1];
651 nextblock
= slp
->tblock
;
654 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
655 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
662 * Clear the first slaves' response. One hopes that it
663 * worked ok, otherwise the tape is much too short!
666 if (atomic_read( slp
->fd
, (char *)&got
, sizeof got
)
668 perror(" DUMP: error reading command pipe in master");
676 if (got
!= writesize
) {
677 quit("EOT detected at start of the tape!\n");
683 if (spcl
.c_type
!= TS_END
) {
684 for (i
= 0; i
< spcl
.c_count
; i
++)
685 if (spcl
.c_addr
[i
] != 0)
689 slp
->firstrec
= lastfirstrec
+ ntrec
;
690 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
693 blockswritten
+= ntrec
;
694 blocksthisvol
+= ntrec
;
699 * We implement taking and restoring checkpoints on the tape level.
700 * When each tape is opened, a new process is created by forking; this
701 * saves all of the necessary context in the parent. The child
702 * continues the dump; the parent waits around, saving the context.
703 * If the child returns X_REWRITE, then it had problems writing that tape;
704 * this causes the parent to fork again, duplicating the context, and
705 * everything continues as if nothing had happened.
708 startnewtape(int top
)
719 sigaddset(&sigs
, SIGINT
);
720 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
721 #else /* __linux__ */
723 void (*interrupt_save
)();
725 sig_t interrupt_save
;
727 interrupt_save
= signal(SIGINT
, SIG_IGN
);
728 #endif /* __linux__ */
730 parentpid
= getpid();
731 tapea_volume
= spcl
.c_tapea
;
733 (void)time4(&tstart_volume
);
735 (void)time((&tstart_volume
);
740 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
742 (void)signal(SIGINT
, interrupt_save
);
745 * All signals are inherited...
749 msg("Context save fork fails in parent %d\n", parentpid
);
755 * save the context by waiting
756 * until the child doing all of the work returns.
757 * don't catch the interrupt
760 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
762 signal(SIGINT
, SIG_IGN
);
765 msg("Tape: %d; parent process: %d child process %d\n",
766 tapeno
+1, parentpid
, childpid
);
768 while ((waitpid
= wait(&status
)) != childpid
)
769 if (waitpid
!= rshpid
)
770 msg("Parent %d waiting for child %d has another child %d return\n",
771 parentpid
, childpid
, waitpid
);
773 msg("Child %d returns LOB status %o\n",
774 childpid
, status
&0xFF);
776 status
= (status
>> 8) & 0xFF;
780 msg("Child %d finishes X_FINOK\n", childpid
);
783 msg("Child %d finishes X_ABORT\n", childpid
);
786 msg("Child %d finishes X_REWRITE\n", childpid
);
789 msg("Child %d finishes unknown %d\n",
800 goto restore_check_point
;
802 msg("Bad return code from dump: %d\n", status
);
806 } else { /* we are the child; just continue */
808 sleep(4); /* allow time for parent's message to get out */
809 msg("Child on Tape %d has parent %d, my pid = %d\n",
810 tapeno
+1, parentpid
, getpid());
813 * If we have a name like "/dev/rmt0,/dev/rmt1",
814 * use the name before the comma first, and save
815 * the remaining names for subsequent volumes.
817 tapeno
++; /* current tape sequence */
819 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
820 tape
[MAXPATHLEN
- 1] = '\0';
821 msg("Dumping volume %d on %s\n", tapeno
, tape
);
823 else if (nexttape
|| strchr(tapeprefix
, ',')) {
824 if (nexttape
&& *nexttape
)
825 tapeprefix
= nexttape
;
826 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
831 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
832 tape
[MAXPATHLEN
- 1] = '\0';
833 msg("Dumping volume %d on %s\n", tapeno
, tape
);
836 while ((tapefd
= (host
? rmtopen(tape
, 2) : pipeout
?
838 open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
840 while ((tapefd
= (pipeout
? fileno(stdout
) :
841 open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
844 msg("Cannot open output \"%s\".\n", tape
);
845 if (!query("Do you want to retry the open?"))
849 enslave(); /* Share open tape file descriptor with slaves */
854 newtape
++; /* new tape signal */
855 spcl
.c_count
= slp
->count
;
857 * measure firstrec in TP_BSIZE units since restore doesn't
858 * know the correct ntrec value...
860 spcl
.c_firstrec
= slp
->firstrec
;
862 spcl
.c_type
= TS_TAPE
;
863 spcl
.c_flags
|= DR_NEWHEADER
;
864 writeheader((ino_t
)slp
->inode
);
865 spcl
.c_flags
&=~ DR_NEWHEADER
;
866 msg("Volume %d started at: %s", tapeno
,
868 ctime4(&tstart_volume
));
870 ctime(&tstart_volume
));
873 msg("Volume %d begins with blocks from inode %d\n",
882 if (master
!= 0 && master
!= getpid())
883 /* Signals master to call dumpabort */
884 (void) kill(master
, SIGTERM
);
887 msg("The ENTIRE dump is aborted.\n");
900 msg("pid = %d exits with status %d\n", getpid(), status
);
906 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
912 siglongjmp(jmpbuf
, 1);
920 #ifdef LINUX_FORK_BUG
928 { struct sigaction sa
;
929 memset(&sa
, 0, sizeof sa
);
930 sigemptyset(&sa
.sa_mask
);
931 sa
.sa_handler
= dumpabort
;
932 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
933 sa
.sa_handler
= sigpipe
;
934 sigaction(SIGPIPE
, &sa
, NULL
);
935 sa
.sa_handler
= proceed
;
936 sa
.sa_flags
= SA_RESTART
;
937 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
940 for (i
= 0; i
< SLAVES
; i
++) {
941 if (i
== slp
- &slaves
[0]) {
947 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
948 (slaves
[i
].pid
= fork()) < 0)
949 quit("too many slaves, %d (recompile smaller): %s\n",
952 slaves
[i
].fd
= cmd
[1];
954 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
956 for (j
= 0; j
<= i
; j
++)
957 (void) close(slaves
[j
].fd
);
959 sigaddset(&sigs
, SIGINT
); /* Master handles this */
961 sigaddset(&sigs
, SIGINFO
);
963 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
965 #ifdef LINUX_FORK_BUG
966 if (atomic_write( cmd
[0], (char *) &i
, sizeof i
)
968 quit("master/slave protocol botched 3\n");
975 #ifdef LINUX_FORK_BUG
977 * Wait for all slaves to _actually_ start to circumvent a bug in
978 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
979 * returned from fork() causes a SEGV in the child process
981 for (i
= 0; i
< SLAVES
; i
++)
982 if (atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
983 quit("master/slave protocol botched 4\n");
986 for (i
= 0; i
< SLAVES
; i
++)
987 (void) atomic_write( slaves
[i
].fd
,
988 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
989 sizeof slaves
[0].pid
);
999 for (i
= 0; i
< SLAVES
; i
++)
1000 if (slaves
[i
].pid
> 0) {
1001 (void) kill(slaves
[i
].pid
, SIGKILL
);
1007 * Synchronization - each process has a lockfile, and shares file
1008 * descriptors to the following process's lockfile. When our write
1009 * completes, we release our lock on the following process's lock-
1010 * file, allowing the following process to lock it and proceed. We
1011 * get the lock back for the next cycle by swapping descriptors.
1014 doslave(int cmd
, int slave_number
)
1017 int nextslave
, size
, eot_count
;
1018 volatile int wrote
= 0;
1024 * Need our own seek pointer.
1026 (void) close(diskfd
);
1027 if ((diskfd
= open(disk
, O_RDONLY
)) < 0)
1028 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1031 retval
= dump_fs_open(disk
, &fs
);
1033 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1034 #endif /* __linux__ */
1037 * Need the pid of the next slave in the loop...
1039 if ((nread
= atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1040 != sizeof nextslave
) {
1041 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1045 * Get list of blocks to dump, read the blocks into tape buffer
1047 while ((nread
= atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1048 register struct req
*p
= slp
->req
;
1050 for (trecno
= 0; trecno
< ntrec
;
1051 trecno
+= p
->count
, p
+= p
->count
) {
1053 bread(p
->dblk
, slp
->tblock
[trecno
],
1054 p
->count
* TP_BSIZE
);
1056 if (p
->count
!= 1 || atomic_read( cmd
,
1057 (char *)slp
->tblock
[trecno
],
1058 TP_BSIZE
) != TP_BSIZE
)
1059 quit("master/slave protocol botched.\n");
1062 if (sigsetjmp(jmpbuf
, 1) == 0) {
1070 /* Try to write the data... */
1075 while (eot_count
< 10 && size
< writesize
) {
1078 wrote
= rmtwrite(slp
->tblock
[0]+size
,
1082 wrote
= write(tapefd
, slp
->tblock
[0]+size
,
1085 printf("slave %d wrote %d\n", slave_number
, wrote
);
1095 if (size
!= writesize
)
1096 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1097 slave_number
, size
, writesize
);
1101 * Handle ENOSPC as an EOT condition.
1103 if (wrote
< 0 && errno
== ENOSPC
) {
1112 * pass errno back to master for special handling
1118 * pass size of write back to master
1119 * (for EOT handling)
1121 (void) atomic_write( cmd
, (char *)&size
, sizeof size
);
1124 * If partial write, don't want next slave to go.
1125 * Also jolts him awake.
1127 (void) kill(nextslave
, SIGUSR2
);
1130 quit("error reading command pipe: %s\n", strerror(errno
));
1134 * Since a read from a pipe may not return all we asked for,
1135 * or a write may not write all we ask if we get a signal,
1136 * loop until the count is satisfied (or error).
1139 atomic_read(int fd
, void *buf
, size_t count
)
1141 int got
, need
= count
;
1144 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1146 } while (got
== -1 && errno
== EINTR
);
1147 return (got
< 0 ? got
: count
- need
);
1151 * Since a read from a pipe may not return all we asked for,
1152 * or a write may not write all we ask if we get a signal,
1153 * loop until the count is satisfied (or error).
1156 atomic_write(int fd
, const void *buf
, size_t count
)
1158 int got
, need
= count
;
1161 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1163 } while (got
== -1 && errno
== EINTR
);
1164 return (got
< 0 ? got
: count
- need
);