]>
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 1.44 2001/04/24 10:59:12 stelian Exp $";
53 #include <compaterr.h>
63 #include <sys/types.h>
65 #include <linux/types.h>
67 #include <sys/param.h>
68 #include <sys/socket.h>
73 #include <linux/ext2_fs.h>
74 #include <ext2fs/ext2fs.h>
75 #include <bsdcompat.h>
77 #include <sys/vnode.h>
80 #include <ufs/inode.h>
82 #include <ufs/ufs/dinode.h>
83 #include <ufs/ffs/fs.h>
84 #endif /* __linux__ */
86 #include <protocols/dumprestore.h>
90 #endif /* HAVE_ZLIB */
94 int writesize
; /* size of malloc()ed buffer for tape */
95 long lastspclrec
= -1; /* tape block number of last written header */
96 int trecno
= 0; /* next record to write in current block */
97 extern long blocksperfile
; /* number of blocks per output file */
98 long blocksthisvol
; /* number of blocks on current output file */
99 extern int ntrec
; /* blocking factor on tape */
100 extern int cartridge
;
105 long long tapea_bytes
= 0; /* bytes_written at start of current volume */
107 static ssize_t atomic_read
__P((int, void *, size_t));
108 static ssize_t atomic_write
__P((int, const void *, size_t));
109 static void doslave
__P((int, int));
110 static void enslave
__P((void));
111 static void flushtape
__P((void));
112 static void killall
__P((void));
113 static void rollforward
__P((void));
114 static int system_command
__P((const char *, const char *, int));
117 * Concurrent dump mods (Caltech) - disk block reading and tape writing
118 * are exported to several slave processes. While one slave writes the
119 * tape, the others read disk blocks; they pass control of the tape in
120 * a ring via signals. The parent process traverses the filesystem and
121 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
122 * The following structure defines the instruction packets sent to slaves.
130 struct slave_results
{
131 ssize_t unclen
; /* uncompressed length */
132 ssize_t clen
; /* compressed length */
135 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
137 int tapea
; /* header number at start of this chunk */
138 int count
; /* count to next header (used for TS_TAPE */
140 int inode
; /* inode that we are currently dealing with */
141 int fd
; /* FD for this slave */
142 int pid
; /* PID for this slave */
143 int sent
; /* 1 == we've sent this slave requests */
144 int firstrec
; /* record number of this block */
145 char (*tblock
)[TP_BSIZE
]; /* buffer for data blocks */
146 struct req
*req
; /* buffer for requests */
150 char (*nextblock
)[TP_BSIZE
];
152 static time_t tstart_volume
; /* time of volume start */
153 static int tapea_volume
; /* value of spcl.c_tapea at volume start */
155 int master
; /* pid of master, for sending error signals */
156 int tenths
; /* length of tape overhead per block written */
157 static int caught
; /* have we caught the signal to proceed? */
158 static int ready
; /* have we reached the lock point without having */
159 /* received the SIGUSR2 signal from the prev slave? */
160 static sigjmp_buf jmpbuf
; /* where to jump to if we are ready when the */
161 /* SIGUSR2 arrives from the previous slave */
163 static int gtperr
= 0;
169 int pgoff
= getpagesize() - 1;
173 writesize
= ntrec
* TP_BSIZE
;
174 reqsiz
= (ntrec
+ 1) * sizeof(struct req
);
176 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
177 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
178 * repositioning after stopping, i.e, streaming mode, where the gap is
179 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
181 if (blocksperfile
== 0 && !unlimited
)
182 tenths
= (cartridge
? 16 : density
== 625 ? 5 : 8);
188 * Allocate tape buffer contiguous with the array of instruction
189 * packets, so flushtape() can write them together with one write().
190 * Align tape buffer on page boundary to speed up tape write().
192 for (i
= 0; i
<= SLAVES
; i
++) {
194 malloc((unsigned)(reqsiz
+ writesize
+ pgoff
+ TP_BSIZE
));
197 slaves
[i
].tblock
= (char (*)[TP_BSIZE
])
199 (((long)&buf
[reqsiz
] + pgoff
) &~ pgoff
);
201 (((long)&buf
[ntrec
+ 1] + pgoff
) &~ pgoff
);
203 slaves
[i
].req
= (struct req
*)slaves
[i
].tblock
- ntrec
- 1;
209 nextblock
= slp
->tblock
;
214 writerec(const void *dp
, int isspcl
)
217 slp
->req
[trecno
].dblk
= (daddr_t
)0;
218 slp
->req
[trecno
].count
= 1;
219 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
220 *(union u_spcl
*)(*(nextblock
)) = *(union u_spcl
*)dp
;
223 lastspclrec
= spcl
.c_tapea
;
231 dumpblock(daddr_t blkno
, int size
)
233 int avail
, tpblks
, dblkno
;
235 dblkno
= fsbtodb(sblock
, blkno
);
236 tpblks
= size
>> tp_bshift
;
237 while ((avail
= MIN(tpblks
, ntrec
- trecno
)) > 0) {
238 slp
->req
[trecno
].dblk
= dblkno
;
239 slp
->req
[trecno
].count
= avail
;
241 spcl
.c_tapea
+= avail
;
244 dblkno
+= avail
<< (tp_bshift
- dev_bshift
);
256 msg("write error on %s: %s\n", tape
, strerror(errnum
));
257 quit("Cannot recover\n");
260 msg("write error %d blocks into volume %d: %s\n",
261 blocksthisvol
, tapeno
, strerror(errnum
));
262 broadcast("DUMP WRITE ERROR!\n");
263 if (query("Do you want to rewrite this volume?")) {
264 msg("Closing this volume. Prepare to restart with new media;\n");
265 msg("this dump volume will be rewritten.\n");
271 if (query("Do you want to start the next tape?"))
280 quit("Broken pipe\n");
285 * Update xferrate stats
294 ttaken
= tnow
- tstart_volume
;
295 blocks
= spcl
.c_tapea
- tapea_volume
;
296 msg("Volume %d completed at: %s", tapeno
, ctime(&tnow
));
298 msg("Volume %d %ld tape blocks (%.2fMB)\n", tapeno
,
299 blocks
, ((double)blocks
* TP_BSIZE
/ 1048576));
301 long volkb
= (bytes_written
- tapea_bytes
) / 1024;
302 long txfrate
= volkb
/ ttaken
;
303 msg("Volume %d took %d:%02d:%02d\n", tapeno
,
304 ttaken
/ 3600, (ttaken
% 3600) / 60, ttaken
% 60);
305 msg("Volume %d transfer rate: %ld kB/s\n", tapeno
,
309 double rate
= .0005 + (double) blocks
/ (double) volkb
;
310 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
312 tapeno
, blocks
, volkb
, rate
);
319 mktimeest(time_t tnow
)
321 static char msgbuf
[128];
326 if (blockswritten
< 500)
328 if (blockswritten
> tapesize
)
329 tapesize
= blockswritten
;
330 deltat
= tstart_writing
- tnow
+ (1.0 * (tnow
- tstart_writing
))
331 / blockswritten
* tapesize
;
332 if (tnow
> tstart_volume
)
333 (void)snprintf(msgbuf
, sizeof(msgbuf
),
334 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
335 (blockswritten
* 100.0) / tapesize
,
336 (spcl
.c_tapea
- tapea_volume
) / (tnow
- tstart_volume
),
337 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
339 (void)snprintf(msgbuf
, sizeof(msgbuf
),
340 "%3.2f%% done, finished in %d:%02d\n",
341 (blockswritten
* 100.0) / tapesize
,
342 (int)(deltat
/ 3600), (int)((deltat
% 3600) / 60));
350 * information message upon receipt of SIGINFO
353 statussig(int notused
)
355 int save_errno
= errno
;
358 buf
= mktimeest(time(NULL
));
360 write(STDERR_FILENO
, buf
, strlen(buf
));
370 struct slave_results returned
;
372 int siz
= (char *)nextblock
- (char *)slp
->req
;
374 slp
->req
[trecno
].count
= 0; /* Sentinel */
376 if (atomic_write( slp
->fd
, (char *)slp
->req
, siz
) != siz
)
377 quit("error writing command pipe: %s\n", strerror(errno
));
378 slp
->sent
= 1; /* we sent a request, read the response later */
380 lastfirstrec
= slp
->firstrec
;
382 if (++slp
>= &slaves
[SLAVES
])
385 /* Read results back from next slave */
387 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
388 != sizeof returned
) {
389 perror(" DUMP: error reading command pipe in master");
392 got
= returned
.unclen
;
393 bytes_written
+= returned
.clen
;
394 if (returned
.unclen
== returned
.clen
)
398 /* Check for errors or end of tape */
400 /* Check for errors */
404 msg("End of tape detected\n");
407 * Drain the results, don't care what the values were.
408 * If we read them here then trewind won't...
410 for (i
= 0; i
< SLAVES
; i
++) {
411 if (slaves
[i
].sent
) {
412 if (atomic_read( slaves
[i
].fd
,
413 (char *)&returned
, sizeof returned
)
414 != sizeof returned
) {
415 perror(" DUMP: error reading command pipe in master");
429 if (spcl
.c_type
!= TS_END
) {
430 for (i
= 0; i
< spcl
.c_count
; i
++)
431 if (spcl
.c_addr
[i
] != 0)
434 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
435 slp
->tapea
= spcl
.c_tapea
;
436 slp
->firstrec
= lastfirstrec
+ ntrec
;
438 nextblock
= slp
->tblock
;
440 asize
+= tenths
+ returned
.clen
/ density
;
441 blockswritten
+= ntrec
;
442 blocksthisvol
+= ntrec
;
443 if (!pipeout
&& !unlimited
&& (blocksperfile
?
444 (blocksthisvol
>= blocksperfile
) : (asize
> tsize
))) {
452 * Executes the command in a shell.
453 * Returns -1 if an error occured, the exit status of
454 * the command on success.
456 int system_command(const char *command
, const char *device
, int volnum
) {
458 char commandstr
[4096];
462 perror(" DUMP: unable to fork");
468 #if OLD_STYLE_FSCRIPT
469 snprintf(commandstr
, sizeof(commandstr
), "%s", command
);
471 snprintf(commandstr
, sizeof(commandstr
), "%s %s %d", command
, device
, volnum
);
473 commandstr
[sizeof(commandstr
) - 1] = '\0';
474 execl("/bin/sh", "sh", "-c", commandstr
, NULL
);
475 perror(" DUMP: unable to execute shell");
479 if (waitpid(pid
, &status
, 0) == -1) {
480 if (errno
!= EINTR
) {
481 perror(" DUMP: waitpid error");
485 if (WIFEXITED(status
))
486 return WEXITSTATUS(status
);
498 struct slave_results returned
;
500 for (f
= 0; f
< SLAVES
; f
++) {
502 * Drain the results, but unlike EOT we DO (or should) care
503 * what the return values were, since if we detect EOT after
504 * we think we've written the last blocks to the tape anyway,
505 * we have to replay those blocks with rollforward.
507 * fixme: punt for now.
509 if (slaves
[f
].sent
) {
510 if (atomic_read( slaves
[f
].fd
, (char *)&returned
, sizeof returned
)
511 != sizeof returned
) {
512 perror(" DUMP: error reading command pipe in master");
515 got
= returned
.unclen
;
516 bytes_written
+= returned
.clen
;
517 if (returned
.unclen
== returned
.clen
)
525 msg("EOT detected in last 2 tape records!\n");
526 msg("Use a longer tape, decrease the size estimate\n");
527 quit("or use no size estimate at all.\n");
530 (void) close(slaves
[f
].fd
);
532 while (wait((int *)NULL
) >= 0) /* wait for any signals from slaves */
537 msg("Closing %s\n", tape
);
542 while (rmtopen(tape
, 0) < 0)
549 (void) close(tapefd
);
550 while ((f
= open(tape
, 0)) < 0)
555 if (eot_script
&& spcl
.c_type
!= TS_END
) {
556 msg("Launching %s\n", eot_script
);
557 eot_code
= system_command(eot_script
, tape
, tapeno
);
559 if (eot_code
!= 0 && eot_code
!= 1) {
560 msg("Dump aborted by the end of tape script\n");
572 if (nexttape
|| Mflag
|| (eot_code
== 0) )
575 msg("Change Volumes: Mount volume #%d\n", tapeno
+1);
576 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
578 while (!query("Is the new volume mounted and ready to go?"))
579 if (query("Do you want to abort?")) {
588 register struct req
*p
, *q
, *prev
;
589 register struct slave
*tslp
;
590 int i
, size
, savedtapea
, got
;
591 union u_spcl
*ntb
, *otb
;
592 struct slave_results returned
;
597 tslp
= &slaves
[SLAVES
];
598 ntb
= (union u_spcl
*)tslp
->tblock
[1];
601 * Each of the N slaves should have requests that need to
602 * be replayed on the next tape. Use the extra slave buffers
603 * (slaves[SLAVES]) to construct request lists to be sent to
604 * each slave in turn.
606 for (i
= 0; i
< SLAVES
; i
++) {
608 otb
= (union u_spcl
*)slp
->tblock
;
611 * For each request in the current slave, copy it to tslp.
615 for (p
= slp
->req
; p
->count
> 0; p
+= p
->count
) {
618 *ntb
++ = *otb
++; /* copy the datablock also */
623 quit("rollforward: protocol botch");
635 nextblock
= tslp
->tblock
;
636 savedtapea
= spcl
.c_tapea
;
637 spcl
.c_tapea
= slp
->tapea
;
639 spcl
.c_tapea
= savedtapea
;
640 lastspclrec
= savedtapea
- 1;
642 size
= (char *)ntb
- (char *)q
;
643 if (atomic_write( slp
->fd
, (char *)q
, size
) != size
) {
644 perror(" DUMP: error writing command pipe");
649 lastfirstrec
= slp
->firstrec
;
651 if (++slp
>= &slaves
[SLAVES
])
656 if (prev
->dblk
!= 0) {
658 * If the last one was a disk block, make the
659 * first of this one be the last bit of that disk
662 q
->dblk
= prev
->dblk
+
663 prev
->count
* (TP_BSIZE
/ DEV_BSIZE
);
664 ntb
= (union u_spcl
*)tslp
->tblock
;
667 * It wasn't a disk block. Copy the data to its
668 * new location in the buffer.
671 *((union u_spcl
*)tslp
->tblock
) = *ntb
;
672 ntb
= (union u_spcl
*)tslp
->tblock
[1];
676 nextblock
= slp
->tblock
;
679 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
680 *(union u_spcl
*)(*nextblock
) = *(union u_spcl
*)tslp
->tblock
;
687 * Clear the first slaves' response. One hopes that it
688 * worked ok, otherwise the tape is much too short!
691 if (atomic_read( slp
->fd
, (char *)&returned
, sizeof returned
)
692 != sizeof returned
) {
693 perror(" DUMP: error reading command pipe in master");
696 got
= returned
.unclen
;
697 bytes_written
+= returned
.clen
;
698 if (returned
.clen
== returned
.unclen
)
706 quit("EOT detected at start of the tape!\n");
712 if (spcl
.c_type
!= TS_END
) {
713 for (i
= 0; i
< spcl
.c_count
; i
++)
714 if (spcl
.c_addr
[i
] != 0)
718 slp
->firstrec
= lastfirstrec
+ ntrec
;
719 slp
->count
= lastspclrec
+ blks
+ 1 - spcl
.c_tapea
;
721 asize
+= tenths
+ returned
.clen
/ density
;
722 blockswritten
+= ntrec
;
723 blocksthisvol
+= ntrec
;
728 * We implement taking and restoring checkpoints on the tape level.
729 * When each tape is opened, a new process is created by forking; this
730 * saves all of the necessary context in the parent. The child
731 * continues the dump; the parent waits around, saving the context.
732 * If the child returns X_REWRITE, then it had problems writing that tape;
733 * this causes the parent to fork again, duplicating the context, and
734 * everything continues as if nothing had happened.
737 startnewtape(int top
)
748 sigaddset(&sigs
, SIGINT
);
749 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
750 #else /* __linux__ */
752 void (*interrupt_save
)();
754 sig_t interrupt_save
;
756 interrupt_save
= signal(SIGINT
, SIG_IGN
);
757 #endif /* __linux__ */
759 parentpid
= getpid();
760 tapea_volume
= spcl
.c_tapea
;
761 tapea_bytes
= bytes_written
;
762 tstart_volume
= time(NULL
);
766 sigprocmask(SIG_UNBLOCK
, &sigs
, NULL
);
768 (void)signal(SIGINT
, interrupt_save
);
771 * All signals are inherited...
775 msg("Context save fork fails in parent %d\n", parentpid
);
781 * save the context by waiting
782 * until the child doing all of the work returns.
783 * don't catch the interrupt
786 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
788 signal(SIGINT
, SIG_IGN
);
791 msg("Tape: %d; parent process: %d child process %d\n",
792 tapeno
+1, parentpid
, childpid
);
794 while ((waitpid
= wait(&status
)) != childpid
)
795 if (waitpid
!= rshpid
)
796 msg("Parent %d waiting for child %d has another child %d return\n",
797 parentpid
, childpid
, waitpid
);
799 msg("Child %d returns LOB status %o\n",
800 childpid
, status
&0xFF);
802 status
= (status
>> 8) & 0xFF;
806 msg("Child %d finishes X_FINOK\n", childpid
);
809 msg("Child %d finishes X_ABORT\n", childpid
);
812 msg("Child %d finishes X_REWRITE\n", childpid
);
815 msg("Child %d finishes unknown %d\n",
826 goto restore_check_point
;
828 msg("Bad return code from dump: %d\n", status
);
832 } else { /* we are the child; just continue */
834 sleep(4); /* allow time for parent's message to get out */
835 msg("Child on Tape %d has parent %d, my pid = %d\n",
836 tapeno
+1, parentpid
, getpid());
839 * If we have a name like "/dev/rmt0,/dev/rmt1",
840 * use the name before the comma first, and save
841 * the remaining names for subsequent volumes.
843 tapeno
++; /* current tape sequence */
845 snprintf(tape
, MAXPATHLEN
, "%s%03d", tapeprefix
, tapeno
);
846 tape
[MAXPATHLEN
- 1] = '\0';
847 msg("Dumping volume %d on %s\n", tapeno
, tape
);
849 else if (nexttape
|| strchr(tapeprefix
, ',')) {
850 if (nexttape
&& *nexttape
)
851 tapeprefix
= nexttape
;
852 if ((p
= strchr(tapeprefix
, ',')) != NULL
) {
857 strncpy(tape
, tapeprefix
, MAXPATHLEN
);
858 tape
[MAXPATHLEN
- 1] = '\0';
859 msg("Dumping volume %d on %s\n", tapeno
, tape
);
862 while ((tapefd
= (host
? rmtopen(tape
, 2) : pipeout
?
864 open(tape
, O_WRONLY
|O_CREAT
, 0666))) < 0)
866 while ((tapefd
= (pipeout
? fileno(stdout
) :
867 open(tape
, O_RDWR
|O_CREAT
, 0666))) < 0)
870 msg("Cannot open output \"%s\".\n", tape
);
871 if (!query("Do you want to retry the open?"))
875 enslave(); /* Share open tape file descriptor with slaves */
880 newtape
++; /* new tape signal */
881 spcl
.c_count
= slp
->count
;
883 * measure firstrec in TP_BSIZE units since restore doesn't
884 * know the correct ntrec value...
886 spcl
.c_firstrec
= slp
->firstrec
;
888 spcl
.c_type
= TS_TAPE
;
889 spcl
.c_flags
|= DR_NEWHEADER
;
891 spcl
.c_flags
|= DR_COMPRESSED
;
892 writeheader((dump_ino_t
)slp
->inode
);
893 spcl
.c_flags
&=~ DR_NEWHEADER
;
894 msg("Volume %d started with block %ld at: %s", tapeno
,
895 spcl
.c_tapea
, ctime(&tstart_volume
));
897 msg("Volume %d begins with blocks from inode %d\n",
906 if (master
!= 0 && master
!= getpid())
907 /* Signals master to call dumpabort */
908 (void) kill(master
, SIGTERM
);
911 msg("The ENTIRE dump is aborted.\n");
924 msg("pid = %d exits with status %d\n", getpid(), status
);
930 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
936 siglongjmp(jmpbuf
, 1);
944 #ifdef LINUX_FORK_BUG
952 { struct sigaction sa
;
953 memset(&sa
, 0, sizeof sa
);
954 sigemptyset(&sa
.sa_mask
);
955 sa
.sa_handler
= dumpabort
;
956 sigaction(SIGTERM
, &sa
, NULL
); /* Slave sends SIGTERM on dumpabort() */
957 sa
.sa_handler
= sigpipe
;
958 sigaction(SIGPIPE
, &sa
, NULL
);
959 sa
.sa_handler
= proceed
;
960 sa
.sa_flags
= SA_RESTART
;
961 sigaction(SIGUSR2
, &sa
, NULL
); /* Slave sends SIGUSR2 to next slave */
964 for (i
= 0; i
< SLAVES
; i
++) {
965 if (i
== slp
- &slaves
[0]) {
971 if (socketpair(AF_UNIX
, SOCK_STREAM
, 0, cmd
) < 0 ||
972 (slaves
[i
].pid
= fork()) < 0)
973 quit("too many slaves, %d (recompile smaller): %s\n",
976 slaves
[i
].fd
= cmd
[1];
978 if (slaves
[i
].pid
== 0) { /* Slave starts up here */
980 for (j
= 0; j
<= i
; j
++)
981 (void) close(slaves
[j
].fd
);
983 sigaddset(&sigs
, SIGINT
); /* Master handles this */
985 sigaddset(&sigs
, SIGINFO
);
987 sigprocmask(SIG_BLOCK
, &sigs
, NULL
);
989 #ifdef LINUX_FORK_BUG
990 if (atomic_write( cmd
[0], (char *) &i
, sizeof i
)
992 quit("master/slave protocol botched 3\n");
1001 #ifdef LINUX_FORK_BUG
1003 * Wait for all slaves to _actually_ start to circumvent a bug in
1004 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
1005 * returned from fork() causes a SEGV in the child process
1007 for (i
= 0; i
< SLAVES
; i
++)
1008 if (atomic_read( slaves
[i
].fd
, (char *) &j
, sizeof j
) != sizeof j
)
1009 quit("master/slave protocol botched 4\n");
1012 for (i
= 0; i
< SLAVES
; i
++)
1013 (void) atomic_write( slaves
[i
].fd
,
1014 (char *) &slaves
[(i
+ 1) % SLAVES
].pid
,
1015 sizeof slaves
[0].pid
);
1025 for (i
= 0; i
< SLAVES
; i
++)
1026 if (slaves
[i
].pid
> 0) {
1027 (void) kill(slaves
[i
].pid
, SIGKILL
);
1033 * Synchronization - each process waits for a SIGUSR2 from the
1034 * previous process before writing to the tape, and sends SIGUSR2
1035 * to the next process when the tape write completes. On tape errors
1036 * a SIGUSR1 is sent to the master which then terminates all of the
1040 doslave(int cmd
, int slave_number
)
1043 int nextslave
, size
, eot_count
, bufsize
;
1044 volatile int wrote
= 0;
1047 struct tapebuf
*comp_buf
= NULL
;
1048 int compresult
, do_compress
= 0;
1049 unsigned long worklen
;
1050 #endif /* HAVE_ZLIB */
1051 struct slave_results returns
;
1057 union u_spcl
*uspclptr
;
1058 struct s_spcl
*spclptr
;
1059 #endif /* USE_QFA */
1062 * Need our own seek pointer.
1064 (void) close(diskfd
);
1065 if ((diskfd
= open(disk
, O_RDONLY
)) < 0)
1066 quit("slave couldn't reopen disk: %s\n", strerror(errno
));
1069 retval
= dump_fs_open(disk
, &fs
);
1071 quit("slave couldn't reopen disk: %s\n", error_message(retval
));
1072 #endif /* __linux__ */
1075 * Need the pid of the next slave in the loop...
1077 if ((nread
= atomic_read( cmd
, (char *)&nextslave
, sizeof nextslave
))
1078 != sizeof nextslave
) {
1079 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1083 /* if we're doing a compressed dump, allocate the compress buffer */
1085 comp_buf
= malloc(sizeof(struct tapebuf
) + TP_BSIZE
+ writesize
);
1086 if (comp_buf
== NULL
)
1087 quit("couldn't allocate a compress buffer.\n");
1088 comp_buf
->flags
= 0;
1090 #endif /* HAVE_ZLIB */
1093 * Get list of blocks to dump, read the blocks into tape buffer
1095 while ((nread
= atomic_read( cmd
, (char *)slp
->req
, reqsiz
)) == reqsiz
) {
1096 register struct req
*p
= slp
->req
;
1098 for (trecno
= 0; trecno
< ntrec
;
1099 trecno
+= p
->count
, p
+= p
->count
) {
1100 if (p
->dblk
) { /* read a disk block */
1101 bread(p
->dblk
, slp
->tblock
[trecno
],
1102 p
->count
* TP_BSIZE
);
1103 } else { /* read record from pipe */
1104 if (p
->count
!= 1 || atomic_read( cmd
,
1105 (char *)slp
->tblock
[trecno
],
1106 TP_BSIZE
) != TP_BSIZE
)
1107 quit("master/slave protocol botched.\n");
1111 /* Try to write the data... */
1115 buffer
= (char *) slp
->tblock
[0]; /* set write pointer */
1116 bufsize
= writesize
; /* length to write */
1117 returns
.clen
= returns
.unclen
= bufsize
;
1121 * When writing a compressed dump, each block is
1122 * written from struct tapebuf with an 4 byte prefix
1123 * followed by the data. This can be less than
1124 * writesize. Restore, on a short read, can compare the
1125 * length read to the compressed length in the header
1126 * to verify that the read was good. Blocks which don't
1127 * compress well are written uncompressed.
1128 * The first block written by each slave is not compressed.
1132 comp_buf
->length
= bufsize
;
1133 worklen
= TP_BSIZE
+ writesize
;
1134 compresult
= Z_DATA_ERROR
;
1136 compresult
= compress2(comp_buf
->buf
, &worklen
,
1137 (char *)slp
->tblock
[0], writesize
, compressed
);
1138 if (compresult
== Z_OK
&& worklen
<= writesize
-32) {
1139 /* write the compressed buffer */
1140 comp_buf
->length
= worklen
;
1141 comp_buf
->compressed
= 1;
1142 buffer
= (char *) comp_buf
;
1143 returns
.clen
= bufsize
= worklen
+ sizeof(struct tapebuf
);
1146 /* write the data uncompressed */
1147 comp_buf
->length
= writesize
;
1148 comp_buf
->compressed
= 0;
1149 buffer
= (char *) comp_buf
;
1150 returns
.clen
= bufsize
= writesize
+ sizeof(struct tapebuf
);
1151 returns
.unclen
= returns
.clen
;
1152 memcpy(comp_buf
->buf
, (char *)slp
->tblock
[0], writesize
);
1155 /* compress the remaining blocks */
1156 do_compress
= compressed
;
1157 #endif /* HAVE_ZLIB */
1159 if (sigsetjmp(jmpbuf
, 1) == 0) {
1168 if (gTapeposfd
>= 0) {
1169 uspclptr
= (union u_spcl
*)&slp
->tblock
[0];
1170 spclptr
= &uspclptr
->s_spcl
;
1171 if ((spclptr
->c_magic
== NFS_MAGIC
) &&
1172 (spclptr
->c_type
== TS_INODE
)) {
1173 /* if an error occured previously don't
1176 if ((gtperr
= GetTapePos(&curtapepos
)) == 0) {
1178 msg("inode %ld at tapepos %ld\n", spclptr
->c_inumber
, curtapepos
);
1180 sprintf(gTps
, "%ld\t%d\t%ld\n", (unsigned long)spclptr
->c_inumber
, tapeno
, curtapepos
);
1181 if (write(gTapeposfd
, gTps
, strlen(gTps
)) != strlen(gTps
)) {
1182 warn("error writing tapepos file.\n");
1188 #endif /* USE_QFA */
1190 while (eot_count
< 10 && size
< bufsize
) {
1193 wrote
= rmtwrite(buffer
+ size
, bufsize
- size
);
1196 wrote
= write(tapefd
, buffer
+ size
, bufsize
- size
);
1198 printf("slave %d wrote %d\n", slave_number
, wrote
);
1208 if (size
!= bufsize
)
1209 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1210 slave_number
, size
, bufsize
);
1214 * Handle ENOSPC as an EOT condition.
1216 if (wrote
< 0 && errno
== ENOSPC
) {
1222 returns
.clen
= returns
.unclen
= 0;
1225 * pass errno back to master for special handling
1228 returns
.unclen
= -errno
;
1231 * pass size of data and size of write back to master
1232 * (for EOT handling)
1234 (void) atomic_write( cmd
, (char *)&returns
, sizeof returns
);
1237 * Signal the next slave to go.
1239 (void) kill(nextslave
, SIGUSR2
);
1242 quit("error reading command pipe: %s\n", strerror(errno
));
1246 * Since a read from a pipe may not return all we asked for,
1247 * or a write may not write all we ask if we get a signal,
1248 * loop until the count is satisfied (or error).
1251 atomic_read(int fd
, void *buf
, size_t count
)
1253 int got
, need
= count
;
1256 while ((got
= read(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1258 } while (got
== -1 && errno
== EINTR
);
1259 return (got
< 0 ? got
: count
- need
);
1263 * Since a read from a pipe may not return all we asked for,
1264 * or a write may not write all we ask if we get a signal,
1265 * loop until the count is satisfied (or error).
1268 atomic_write(int fd
, const void *buf
, size_t count
)
1270 int got
, need
= count
;
1273 while ((got
= write(fd
, buf
, need
)) > 0 && (need
-= got
) > 0)
1275 } while (got
== -1 && errno
== EINTR
);
1276 return (got
< 0 ? got
: count
- need
);
1282 * read the current tape position
1285 GetTapePos(long *pos
)
1290 if (ioctl(tapefd
, MTIOCPOS
, pos
) == -1) {
1292 msg("[%ld] error: %d (getting tapepos: %ld)\n", getpid(),
1298 #endif /* USE_QFA */